Intervenciones para prevenir la diabetes de tipo 2 en adultos con trastornos mentales en países de ingresos bajos y medios
Resumen
Antecedentes
La prevalencia de la diabetes de tipo 2 es elevada en las personas con trastornos mentales. Una gran parte de la carga de morbimortalidad recae en las poblaciones de países de ingresos bajos y medios (PIBM).
Objetivos
Evaluar los efectos de las intervenciones farmacológicas, de modificación de la conducta e institucionales versus comparadores activos e inactivos para la prevención o el retraso de la diabetes de tipo 2 en las personas con trastornos mentales en los PIBM.
Métodos de búsqueda
Se realizaron búsquedas en el registro de ensayos controlados del Grupo Cochrane de Trastornos mentales comunes (Cochrane Common Mental Disorders), CENTRAL, MEDLINE, Embase y en otras seis bases de datos, así como en tres registros de ensayos internacionales. También se hicieron búsquedas en las actas de congresos y se comprobaron las listas de referencias de revisiones sistemáticas pertinentes. Las búsquedas están actualizadas hasta el 20 de febrero de 2020.
Criterios de selección
Ensayos controlados aleatorizados (ECA) de intervenciones farmacológicas, de modificación de la conducta o institucionales dirigidas a prevenir o retrasar la diabetes de tipo 2 en adultos con trastornos mentales en los PIBM.
Obtención y análisis de los datos
Parejas de autores de la revisión, que trabajaron de forma independiente, extrajeron los datos y evaluaron el riesgo de sesgo. Para los metanálisis se utilizaron modelos de efectos aleatorios.
Resultados principales
Un ECA hospitalario con 150 participantes (99 participantes con esquizofrenia) abordó el desenlace principal de la revisión de la prevención o el retraso de la aparición de la diabetes de tipo 2. La evidencia de certeza baja de este estudio no mostró una diferencia entre los antipsicóticos atípicos y los típicos en la aparición de diabetes a las seis semanas (razón de riesgos [RR] 0,46; intervalo de confianza [IC] del 95%: 0,03 a 7,05) (en un total de 99 participantes con esquizofrenia, 68 estaban en los grupos de antipsicóticos atípicos y 31 en los típicos; 55 participantes sin enfermedad mental no se consideraron en el análisis).
Otros 29 ECA con 2481 participantes evaluaron uno o más desenlaces secundarios de la revisión. Todos los estudios se realizaron en ámbitos hospitalarios e informaron acerca de intervenciones farmacológicas. Un estudio, que no fue posible incluir en el metanálisis, incluyó una intervención con componentes farmacológicos y de modificación de la conducta. No se identificaron estudios sólidos de intervenciones a nivel institucional.
La evidencia de certeza baja a moderada indica que podría no haber diferencias entre el uso de antipsicóticos atípicos y típicos para los desenlaces de los abandonos de la atención (RR 1,31; IC del 95%: 0,63 a 2,69; dos estudios con 144 participantes), y la glucemia en ayunas (diferencia de medias [DM] 0,05 menor; IC del 95%: 0,10 a 0,00; dos estudios con 211 participantes). Los participantes que reciben antipsicóticos típicos podrían tener un índice de masa corporal (IMC) más bajo en el seguimiento que los participantes que reciben antipsicóticos atípicos (DM 0,57; IC del 95%: 0,33 a 0,81; dos estudios con 141 participantes; certeza de evidencia moderada), y podrían tener niveles de colesterol total más bajos a las ocho semanas después de comenzar el tratamiento (DM 0,35; IC del 95%: 0,27 a 0,43; un estudio con 112 participantes).
Hubo evidencia de certeza moderada que indica que no existen diferencias entre el uso de metformina y placebo en el desenlace de los abandonos de la asistencia (RR 1,22; IC del 95%: 0,09 a 16,35; tres estudios con 158 participantes). Hubo evidencia de certeza moderada a alta de que no hubo diferencias entre la metformina y el placebo en la glucemia en ayunas (datos finales: DM ‐0,35; IC del 95%: ‐0,60 a ‐0,11; cambio respecto a los datos iniciales: DM 0,01; IC del 95%: ‐0,21 a 0,22; cinco estudios con 264 participantes). Hubo evidencia de certeza alta de que el IMC fue menor en los participantes que recibieron metformina en comparación con los que recibieron un placebo (DM ‐1,37; IC del 95%: ‐2,04 a ‐0,70; cinco estudios con 264 participantes; evidencia de certeza alta). No hubo diferencias entre la metformina y el placebo en los desenlaces de perímetro abdominal, presión arterial y niveles de colesterol.
Hubo evidencia de certeza baja de un estudio (48 participantes) que indica que podría no haber diferencias entre el uso de melatonina y placebo en el desenlace de los abandonos de la asistencia (RR 1,00; IC del 95%: 0,38 a 2,66). Es probable que la glucemia en ayunas se reduzca más en los participantes tratados con melatonina, en comparación con el placebo (datos finales: DM ‐0,17; IC del 95%: ‐0,35 a ‐0,01; cambio respecto a los datos iniciales: DM ‐0,24; IC del 95%: ‐0,39 a ‐0,09; tres estudios con 202 participantes, evidencia de certeza moderada). No hubo diferencias entre la melatonina y el placebo en los desenlaces de perímetro abdominal, presión arterial y niveles de colesterol.
La evidencia de certeza muy baja de un estudio (25 participantes) indica que los abandonos podrían ser mayores entre los participantes tratados con un antidepresivo tricíclico (ATC), en comparación con los que recibieron inhibidores selectivos de la recaptación de serotonina (ISRS) (RR 0,34; IC del 95%: 0,11 a 1,01). No se sabe si no hay diferencias en la glucemia en ayunas entre estos grupos (DM ‐0,39; IC del 95%: ‐0,88 a 0,10; tres estudios con 141 participantes; evidencia de certeza moderada). No se sabe si existen diferencias en el IMC y la depresión entre los grupos de antidepresivos ATC y ISRS.
Conclusiones de los autores
Sólo un estudio aportó datos sobre el desenlace principal de interés, proporcionando evidencia de certeza baja de que podría no haber diferencias en el riesgo entre los antipsicóticos atípicos y típicos en el desenlace de desarrollar diabetes de tipo 2. Por lo tanto, no es posible establecer conclusiones sobre la prevención de la diabetes de tipo 2 en personas con trastornos mentales en los PIBM.
Con respecto a los estudios que informaron sobre desenlaces secundarios, hubo evidencia de un riesgo de sesgo en sus resultados. Se necesitan más estudios con participantes de PIBM con trastornos mentales, especialmente sobre intervenciones de modificación de la conducta o institucionales dirigidas a prevenir la diabetes de tipo 2 en estas poblaciones.
PICO
Resumen en términos sencillos
Prevención de la diabetes de tipo 2 en adultos con enfermedades mentales en países de ingresos bajos y medios
¿Cómo se identificó y evaluó la evidencia?
Se buscó en la literatura médica para revisar la evidencia sobre los efectos de las intervenciones farmacológicas (medicamentos), conductuales (modificación de la conducta) y organizativas (prestación de asistencia sanitaria) para la prevención de la diabetes de tipo 2 en las personas con trastornos mentales en los países de ingresos bajos y medios (PIBM). La diabetes de tipo 2 es una enfermedad grave que puede aparecer cuando el cuerpo ya no puede utilizar correctamente una hormona llamada insulina. Hay muchas razones por las que una persona puede desarrollar diabetes de tipo 2, como el sobrepeso, la hipertensión arterial, la falta de ejercicio físico, los antecedentes familiares de la enfermedad y otros posibles factores de riesgo.
Se incluyeron los ensayos controlados aleatorizados (ECA), publicados hasta la fecha de búsqueda, 20 de febrero de 2020.
¿Por qué es esto importante?
Las personas con enfermedades mentales como la esquizofrenia, el trastorno bipolar y el trastorno depresivo mayor tienen más probabilidades de desarrollar diabetes de tipo 2 que la población general. Muchas personas que padecen problemas de salud mental, con un mayor riesgo de desarrollar diabetes, viven en los PIBM. El tratamiento de la diabetes en esta población plantea retos a los sistemas sanitarios. Por lo tanto, prevenir la aparición de la diabetes es importante para las personas con problemas de salud mental y para los sistemas sanitarios de los países de ingresos bajos y medios.
¿Qué se encontró?
En adultos con trastornos de salud mental, sólo se identificó un estudio que evaluó el desenlace principal de esta revisión, la prevención de la diabetes de tipo 2. Este estudio realizado en un hospital con 150 participantes (99 participantes con esquizofrenia) encontró evidencia de certeza baja de que no hay diferencias en el riesgo entre el uso de medicamentos antipsicóticos más antiguos (antipsicóticos típicos) y los más nuevos (antipsicóticos atípicos) para el desenlace de desarrollar diabetes de tipo 2.
Se incluyeron otros 29 estudios con 2481 participantes que evaluaron uno o más desenlaces secundarios. Todos los estudios se realizaron en ámbitos hospitalarios y analizaron intervenciones farmacológicas. Ningún estudio examinó intervenciones institucionales. Sólo un estudio evaluó una intervención dirigida a cambiar el comportamiento de las personas, pero también incluyó una intervención farmacológica.
Para el desenlace de los abandonos del estudio (cuántas personas abandonan un estudio antes de que este termine), no se encontró evidencia de una diferencia cuando los participantes fueron tratados con antipsicóticos atípicos, en comparación con los que fueron tratados con antipsicóticos típicos. Este fue también el caso de los estudios que compararon el tratamiento con metformina (un medicamento utilizado para tratar la diabetes) con el placebo (tratamiento falso), y los que compararon el tratamiento con melatonina (una hormona que regula el sueño) con el placebo. La evidencia de certeza muy baja de un estudio indica que los abandonos podrían ser mayores entre los participantes tratados con un antidepresivo tricíclico, en comparación con los participantes tratados con inhibidores selectivos de la recaptación de serotonina (otro tipo de antidepresivo).
No se encontró evidencia de una diferencia en los niveles de glucemia en ayunas en las comparaciones entre antipsicóticos atípicos y típicos, metformina y placebo, o antidepresivos tricíclicos e inhibidores selectivos de la recaptación de serotonina. Sí se encontró que los niveles de glucosa en sangre en ayunas son probablemente más bajos en los participantes tratados con melatonina, en comparación con los que recibieron placebo.
El índice de masa corporal fue menor en los participantes que recibieron metformina en comparación con los que recibieron un placebo, y en los participantes que recibieron antipsicóticos típicos, en comparación con los que recibieron antipsicóticos atípicos.
La concentración de colesterol fue menor en los participantes que recibieron antipsicóticos típicos, en comparación con los que recibieron antipsicóticos atípicos.
No se encontró evidencia de una diferencia en el perímetro de la cintura ni en la presión arterial en ninguno de los grupos de intervención de los estudios incluidos.
Certeza de la evidencia
El único estudio que evaluó la prevención de la diabetes de tipo 2 proporcionó evidencia de certeza baja. La certeza de la evidencia se redujo porque el estudio era pequeño, y varios aspectos importantes del mismo tenían un alto riesgo de sesgo. Los otros estudios que informaron de desenlaces secundarios generalmente proporcionaron evidencia de certeza moderada a alta para estos desenlaces.
Conclusiones
No se sabe cuál es la mejor manera de prevenir la diabetes de tipo 2 en personas con enfermedades mentales de PIBM. Solo uno de los ensayos incluidos proporcionó evidencia de certeza baja sobre la prevención de la diabetes. Los estudios de investigación futuros deberían centrarse no sólo en las intervenciones farmacológicas, sino también de modificación de la conducta y las intervenciones institucionales, para saber si dichas intervenciones pueden ser eficaces y adecuadas en los contextos de PIBM.
Authors' conclusions
Summary of findings
| Atypical compared with typical antipsychotic medications for preventing type 2 diabetes in adults with mental disorders in low‐ and middle‐income countries | ||||||
| Patient or population: Adults with mental disorders in low‐ and middle‐income countries | ||||||
| Outcomes | Anticipated absolute effects* (95% CI) | Relative effect | № of participants | Certainty of the evidence | Comments | |
|---|---|---|---|---|---|---|
| Risk with typical antipsychotic | Risk with atypical antipsychotic | |||||
| Diabetes (ADA criteria) (6 weeks) | Study population | RR 0.50 | 93 | ⊕⊕⊝⊝ | ||
| 32 per 1,000 | 16 per 1,000 | |||||
| Drop‐outs (6 to 54 weeks) | Study population | RR 1.31 | 144 | ⊕⊕⊝⊝ | ||
| 148 per 1,000 | 194 per 1,000 | |||||
| Fasting blood glucose (6 to 8 weeks) | Mean fasting blood glucose was 4.90 to 4.91 mmol/L (normal level) | MD 0.05 lower (0.10 lower to 0.00 lower) | ‐ | 211 | ⊕⊕⊕⊝ | |
| BMI (8 to 54 weeks) | Mean BMI was 21.2 to 24.6 kg/m2 (healthy weight range) | MD 0.57 higher | ‐ | 141 | ⊕⊕⊕⊝ | |
| *The risk in the intervention group (and its 95% confidence interval) is based on the assumed risk in the comparison group and the relative effect of the intervention (and its 95% CI). | ||||||
| GRADE Working Group grades of evidence | ||||||
| 1One small trial; estimate with wide confidence interval crossing 1. Downgraded two levels for imprecision. | ||||||
| Metformin compared with placebo for preventing type 2 diabetes in adults with mental disorders in low‐ and middle‐income countries | ||||||
| Patient or population: Adults with mental disorders in low‐ and middle‐income countries | ||||||
| Outcomes | Anticipated absolute effects* (95% CI) | Relative effect | № of participants | Certainty of the evidence | Comments | |
|---|---|---|---|---|---|---|
| Risk with placebo | Risk with metformin | |||||
| Diabetes | No studies identified | |||||
| Drop‐outs (12 to 14 weeks) | Study population | RR 1.22 (0.09 to 16.35) | 158 | ⊕⊕⊕⊝ | ||
| 49 per 1,000 | 11 more per 1,000 | |||||
| Fasting blood glucose (12 to 14 weeks) | Mean fasting blood glucose was 4.40 to 4.71 mmol/L | MD 0.35 lower (0.60 lower to 0.11 lower) | ‐ | 173 | ⊕⊕⊕⊝ | |
| Fasting blood glucose (12 to 14 weeks) | ‐ | MD 0.01 higher (0.21 lower to 0.22 higher) | 91 (2 RCTs) | ⊕⊕⊕⊕ | ||
| BMI (12 to 14 weeks) | Mean BMI was 23.5 to 25.3 (normal to slightly overweight) | MD 1.37 lower | ‐ | 264 | ⊕⊕⊕⊕ | |
| *The risk in the intervention group (and its 95% confidence interval) is based on the assumed risk in the comparison group and the relative effect of the intervention (and its 95% CI). | ||||||
| GRADE Working Group grades of evidence | ||||||
| 1Some unclear risk of bias domains and unbalanced drop‐out in one study, but this is not a concern for this outcome. | ||||||
| Melatonin compared with placebo for preventing type 2 diabetes in adults with mental disorders in low‐ and middle‐income countries | ||||||
| Patient or population: Adults with mental disorders in low‐ and middle‐income countries | ||||||
| Outcomes | Anticipated absolute effects* (95% CI) | Relative effect | № of participants | Certainty of the evidence | Comments | |
|---|---|---|---|---|---|---|
| Risk with placebo | Risk with melatonin | |||||
| Diabetes | No studies identified | |||||
| Drop‐outs (8 weeks) | Study population | RR 1.00 | 48 | ⊕⊕⊝⊝ | ||
| 250 per 1,000 | 250 per 1,000 | |||||
| Fasting blood glucose (8 to 12 weeks) (endpoint data) | Mean fasting blood glucose was 4.9 to 5.0 mmol/L (normal level) | MD 0.17 lower (0.35 lower to 0.01 higher) | ‐ | 102 | ⊕⊕⊕⊝ | |
| Fasting blood glucose (8 to 12 weeks) (change from baseline data) | ‐ | MD 0.24 lower (0.39 lower to 0.09 lower) | ‐ | 100 (1 RCT) | ⊕⊕⊕⊝ | |
| BMI (8 to 12 weeks) | Mean BMI was 25.1 to 25.2 (slightly overweight) | MD 0.22 lower | ‐ | 202 | ⊕⊝⊝⊝ | |
| *The risk in the intervention group (and its 95% confidence interval) is based on the assumed risk in the comparison group and the relative effect of the intervention (and its 95% CI). | ||||||
| GRADE Working Group grades of evidence | ||||||
| 1One small trial with an equal number of events in both groups. Downgraded two levels for very serious imprecision; estimate includes benefits and harms. | ||||||
| SSRI antidepressants compared with TCA for preventing type 2 diabetes in adults with mental disorders in low‐ and middle‐income countries | ||||||
| Patient or population: Adults with mental disorders in low‐ and middle‐income countries | ||||||
| Outcomes | Anticipated absolute effects* (95% CI) | Relative effect | № of participants | Certainty of the evidence | Comments | |
|---|---|---|---|---|---|---|
| Risk with TCA | Risk with SSRI | |||||
| Diabetes | No studies identified. | |||||
| Drop‐outs (12 weeks) | Study population | RR 0.34 | 25 | ⊕⊝⊝⊝ | ||
| 636 per 1,000 | 216 per 1,000 | |||||
| Fasting blood glucose (8 to 12 weeks) | Mean fasting blood glucose was 4.4 to 5.1 mmol/L (normal level) | MD 0.39 lower (0.88 lower to 0.10 higher) | ‐ | 141 | ⊕⊕⊕⊝ | |
| BMI (12 weeks) | Mean BMI was 25.2 (slightly overweight) | MD 0.7 higher | ‐ | 18 | ⊕⊝⊝⊝ | |
| *The risk in the intervention group (and its 95% confidence interval) is based on the assumed risk in the comparison group and the relative effect of the intervention (and its 95% CI). | ||||||
| GRADE Working Group grades of evidence | ||||||
| 1One small trial; unclear allocation concealment, no blinding, incomplete outcome data. Downgraded two levels for risk of bias. | ||||||
Background
Mental disorders increasingly add to the global disease burden. They are one of the leading causes of disability worldwide, accounting for almost one‐quarter of all years lived with a disability (Murray 2012; Vos 2012) and result in significantly increased mortality (Correll 2017; Lawrence 2010; Mitchell 2013; Scott 2009). Studies have repeatedly reported a 10 to 20 year mortality gap for people with mental illness, and despite an overall improvement in life expectancy in recent years, the absolute mortality gap between people with and without mental illness is widening (Chesney 2014; Hayes 2017; Olfson 2015; Saha 2007). Studies from low‐ and middle‐Income countries (LMIC) show a similar pattern of increased mortality but with an even greater reduction in life expectancy than in high‐income countries (Dube 1984; Fekadu 2015; Kurihara 2011; Mogga 2006). However, only 0.5% to 2% of national health budgets are allocated for the prevention and treatment of mental disorders in LMICs (Stubbs 2017). Mental illness remains a major health challenge in these countries (Rathod 2017).
A considerable proportion of the increased morbidity and mortality experienced by people with mental disorders is driven by comorbid physical illnesses (Firth 2019; Hayes 2017), not just by the mental illness. The vast majority of deaths (around 80%) are due to preventable physical illnesses, most commonly cardiovascular, metabolic and respiratory diseases, as well as infections (Correll 2017; Crump 2013; Firth 2019; Laursen 2011). Mental and physical disorders have a complex and bidirectional relationship. A higher prevalence of comorbid physical health conditions (e.g. diabetes and cardiovascular disease) and poorer management of these illnesses, contribute to health inequalities in people with mental illness (Vancampfort 2016a; Ward 2015). People with severe mental illness (e.g. schizophrenia and bipolar disorder) have a particularly high risk of developing conditions such as diabetes and cardiovascular disease for reasons associated with the underlying mental disorder. These include health risk behaviours such as physical inactivity, smoking, and poor diet (Vancampfort 2017) and treatments that increase cardio‐metabolic risks and mortality (Liu 2017). Conversely, common mental disorders (e.g. depression and anxiety) are more common in people with these physical health conditions (Das‐Munshi 2007).
Globally, noncommunicable chronic diseases such as diabetes are a major cause of morbidity and mortality (contributing to 60% of all deaths) (Miranda 2008), including in LMICs (Lopez 2006). Diabetes is a serious lifelong condition. It is a major global health challenge, with increasing prevalence worldwide, and showing a particularly sharp rise in prevalence in LMICs (Stubbs 2016). A study using data from nationally representative surveys in 28 LMICs showed an overall diabetes prevalence of 8.8% (95% confidence interval (CI) 8.2% to 9.5%) (Manne‐Goehler 2019). Another study showed that across 29 LMICs the prevalence of diabetes among persons age 25 years or older was 7.5% (95% CI 7.1% to 8%) (Seiglie 2020).
Diabetes is strongly associated with mental illness (Vancampfort 2015a). For example, around 13% of people with severe mental illness (Ward 2015) and 9% of people with major depressive disorder (Vancampfort 2015b) are estimated to have diabetes, compared to an estimated 8.5% of the general population globally (WHO 2016) and compared to 6% in the UK (Reilly 2015). A systematic review and meta‐analysis found that odds of having type 2 diabetes were 1.85 times higher among people with severe mental illness than among matched controls (Vancampfort 2016a). People with schizophrenia and bipolar disorder seem to be particularly at risk of developing type 2 diabetes (Pillinger 2017; Stubbs 2015; Vancampfort 2013; Vancampfort 2015b).
There is also good evidence of an association between diabetes and common mental disorders (Das‐Munshi 2007; Moulton 2015; Vancampfort 2016b). People with diabetes have a two‐ to three‐fold increased prevalence of depressive (Ali 2006; Anderson 2001) and anxiety disorders (Grigsby 2002), although this relationship is likely to be bidirectional (Golden 2008). A systematic review of 48 studies showed that the prevalence of depression among people with diabetes is higher in LMICs than in high‐income countries. Across all studies conducted in LMICs, 35.7% of people with diabetes were found to suffer from depression (Mendenhall 2014).
Multiple complex mechanisms are known to contribute to the association between diabetes mellitus and severe mental illness including genetic, environmental, disease‐specific factors, and treatment‐specific factors (Holt 2015). However, the principles of managing diabetes mellitus in people with severe mental illness are similar to those for the general population and should follow currently established treatment algorithms (Holt 2015). Several interventions have been found to be effective for prevention of type 2 diabetes in the general population (Merlotti 2014; White 2016). Prevention of diabetes in people with mental illness is also important. However, due to a complex combination of psychological, social, and financial barriers, generic interventions to prevent diabetes may not be suitable for people with mental disorders (Chwastiak 2015). Some of the additional barriers faced by people with mental illness, not addressed by generic interventions, include social stigma, poor access to medical care (Bradford 2008), fragmentation and lack of coordination between medical and psychiatric treatment in the healthcare systems of many countries (Druss 2010), and "diagnostic overshadowing," where physical health problems are overlooked by health professional in the presence of mental illness (Liu 2017). These difficulties compound the challenges of managing side effects of psychotropic medication and the higher prevalence of health risk behaviours. This is more challenging in LMICs due to limited resources and facilities (Manne‐Goehler 2019).
Several abnormal clinical and metabolic findings (e.g. hypertension, hyperglycaemia, dyslipidaemia, overweight) are predictive of diabetes and other metabolic syndromes. It is therefore essential to also consider these cardio‐metabolic risk factors in patients in this vulnerable group, monitoring blood pressure, blood glucose level, lipid profile, body mass index (BMI) and waist circumference (De Hert 2009) when considering diabetes prevention.
Description of the condition
Ninety percent of people with diabetes have type 2 diabetes, a metabolic disorder that usually results secondary to insulin resistance (IR). It is commonly seen in individuals with obesity and is associated with disturbances in carbohydrate, fat, and protein metabolism. While pancreatic β‐cells initially respond to IR by increasing insulin secretion, the cells eventually fail to keep up with demand resulting in relative insulin deficiency, consequently leading to hyperglycaemia (elevated levels of plasma glucose) (Weir 2020). Prolonged hyperglycaemia may lead to microvascular complications (Andany 2019) including retinopathy (disease of the retina which results in impairment or loss of vision), nephropathy (renal impairment), neuropathy (an abnormal and usually degenerative state of parts of the nervous system) and macrovascular, including coronary artery, cerebrovascular and peripheral artery complications.
The ‘epidemic’ of diabetes seen over recent decades has been attributed to changes in demographics and lifestyle globally (e.g. increased life expectancy, sedentary behaviours, and consumption of high fat and carbohydrate diets) (Miranda 2008). LMIC populations have experienced especially rapid changes, with which health policy and services have failed to keep pace (Popkin 2002). According to the American Diabetes Association (ADA), the risk of developing diabetes increases with age, obesity, lack of physical activity, dyslipidemia (abnormal amount of lipids in the blood), and hypertension (ADA 2017), all of which have been adversely affected by these changes.
Description of the intervention
Prevention of diabetes includes activities targeted at reducing the frequency or level of causal risk factors for development of diabetes (WHO 1994). Diabetes prevention or delay may be achieved with pharmacological, behaviour change, and organisational interventions. Pharmacological interventions aimed at prevention of diabetes in people with mental disorder include diabetes medication, weight loss medication, a combination of diabetes and weight loss medication, diabetes preventive medication and antipsychotic switching. Behaviour change interventions may target health risk behaviours, and may include patient education programmes, psychological interventions (e.g. cognitive behavioural therapy, counselling or motivational interviewing), self‐monitoring (including telehealth, internet‐based interventions, and other communication technologies) or multicomponent interventions (e.g. self‐management programmes that combine education and behavioural approaches) (Taylor 2017). Organisational interventions may include interventions that aim to improve the delivery of care, such as educating health professionals, care planning, or collaborative models of care (Druss 2010).
It may be that there are particular pharmacological, behaviour change and lifestyle, or organisational interventions that would be more applicable to LMICs as the availability of pharmacological interventions, resources and organisational structures in LMICs are different from those in high‐income countries. For instance, some drugs may not yet be available in LMICs; psychological behaviour change interventions might not be feasible due to lack of trained personnel; or there may not be any collaborative models of care in the health system (Koyanagi 2017). In addition, LMICs are not homogenous and the availability of interventions within healthcare systems may differ among countries due to variability in health care resources and organisational structures (Mate 2013).
How the intervention might work
Pharmacological interventions
There are several modes of action for medication in preventing diabetes. Diabetes medication helps regulate carbohydrate and fat metabolism, by increasing insulin sensitivity and reducing the amount of glucose produced and released by the liver. Weight loss medication or anti‐obesity drugs usually act on the gastrointestinal tract by reducing absorption of dietary fat, stimulate energy expenditure and decrease fat storage, or decrease appetite. Diabetes combination medications allow patients to switch between treatments, depending on clinical response. Switching to or adding an atypical antipsychotic associated with fewer metabolic side effects is hypothesised to alleviate weight gain and metabolic abnormalities caused by commonly used antipsychotics such as olanzapine and clozapine. Other medications may work by enhancing lipid profile and metabolic function and regulating or increasing insulin sensitivity (Taylor 2017).
Behaviour change interventions
These interventions target health risk behaviours using educational, psychological, and behavioural approaches, or combinations of these. For diabetes, there has been a focus on self‐management interventions using behaviour change techniques (McBain 2016), influenced by theories of health behaviour change, including social cognitive theory (Bandura 1986), the theory of reasoned action and planned behaviour (Ajzen 1991), self regulation theory (Leventhal 1984) and the transtheoretical model (Prochaska 1997). All of these theories identify concepts that predict health behaviour (and that may be targeted by behaviour change interventions), with a primary focus on beliefs, attitudes, and expectations. For example, a diabetes self‐management intervention based on social cognitive theory (Bandura 1986) may seek to reduce carbohydrate intake by increasing diet‐related self‐efficacy.
Organisational interventions
Organisational capacity building and training programmes may increase the efficacy and communication skills of mental health or diabetes professionals or other health workers and health services to support prevention of diabetes for people with mental illness (Liu 2017).
Why it is important to do this review
To date, a limited number of non‐Cochrane systematic reviews have investigated the effectiveness of interventions to prevent diabetes for people with mental illness (McGinty 2016; Taylor 2017). These reviews, mostly including data from high‐income countries, have reported that diabetes can be prevented or its onset delayed. A comprehensive review by McGinty and colleagues included 33 studies of interventions for diabetes mellitus in people with severe mental illness. It found no high‐certainty evidence for the effectiveness of any interventions; the best available evidence suggested a potential beneficial effect of metformin on glucose control (McGinty 2016). The review by Taylor and colleagues, which also focused on people with severe mental illness, included 54 randomized controlled trials (RCTs) among which only a few were from LMICs (Brazil, India, China, Iran, Venezuela) (Taylor 2017). The authors found some evidence for the effectiveness of pharmacological and non‐pharmacological interventions in improving glycaemic outcomes, but no subgroup analyses were conducted by country income level.
Other non‐Cochrane reviews have investigated the effect of pharmacological (Maayan 2010; Mizuno 2014), behavioural (Bruins 2014; Caemmerer 2012; Fernández‐San‐Martín 2014), or both pharmacological and behavioural interventions (Faulkner 2007) on glycaemic measurements in people with severe mental illness. They have also reported that these interventions may be effective, but again have focused only on people with severe mental illness or those taking antipsychotic medication, and identified very few studies in LMICs. Moreover, these studies considered glycaemic effects only as a secondary outcome.
It is important to assess the evidence for diabetes prevention in people with mental disorders, specifically for LMICs. Not only is the prevalence of diabetes and comorbid mental illness high in LMICs, but we cannot be certain that interventions shown to be effective in high‐income countries work in the same way in LMICs. Populations may differ in terms of demographics and living conditions, affecting risk of diabetes, and healthcare systems may operate differently, potentially influencing availability, feasibility and effectiveness of interventions. In addition, it is important to consider other cardio‐metabolic risk factor measures, e.g. blood pressure, fasting blood glucose level, serum cholesterol level, BMI and waist circumference for this population.
A review of the effectiveness of interventions designed to prevent diabetes in people with any mental disorder, focused on LMICs, is therefore needed to inform practice and future research for this population.
Objectives
To assess the effects of pharmacological, behaviour change, and organisational interventions versus active and non‐active comparators in the prevention or delay of type 2 diabetes among people with mental illness in LMICs.
Methods
Criteria for considering studies for this review
Types of studies
We included RCTs evaluating any interventions to prevent type 2 diabetes in people with any mental disorder in LMICs. LMICs were defined according to the Development Assistance Committee (DAC) list of all countries and territories eligible to receive official development assistance (ODA) (DAC 2017).
Types of participants
We included studies of adults aged 18 years and over, with any mental disorder and without diabetes. Studies that did not explicitly screen for and exclude diabetes at baseline were not included. Mental illness diagnoses were to be established using World Health Organization (WHO) International Classification of Diseases (ICD) criteria for mental and behavioural disorders (ICD‐10, F20‐29 and F30‐31, F 32.3, F33.3) (WHO 1992) and/or the Diagnostic and Statistical Manual of Mental Disorders (DSM) (DSM‐III, APA 1980; DSM‐III‐R, APA 1987; DSM IV, APA 2000; DSM V, APA 2013) or measures based on these. We defined severe mental illness as schizophrenia or other psychotic disorders, bipolar disorder, and depression with psychotic features. Common mental disorders included depression, generalised anxiety disorder (GAD), panic disorder, phobias, social anxiety disorder, obsessive‐compulsive disorder (OCD), and post‐traumatic stress disorder (PTSD) (NICE 2011). Other mental disorders such as personality disorder and somatoform disorders were also included in this review.
Where study populations were mixed (i.e. including people with and without mental disorder), we included studies only if people with mental disorders constituted the predominant population (more than 50%), or if separate outcome data were provided for participants with mental disorders.
To be consistent with changes in the classification of, and diagnostic criteria for diabetes mellitus over the years, studies had to use (and explicitly state) established standard criteria for the diagnosis of type 2 diabetes, valid at the time of the trial commencing (e.g. ADA 1999; ADA 2008; ADA 2017; WHO 1999; WHO 2006).
Types of interventions
Experimental intervention
The review included any pharmacological, behaviour change (targeting health risk behaviours), or organisational interventions targeting the prevention of diabetes in people with any mental disorder in LMICs.
Pharmacological interventions included diabetes medication (e.g. metformin, pioglitazone); weight loss medication (e.g. amantadine, orlistat, sibutramine); combinations of weight loss and diabetes drugs (e.g. amantadine with metformin and zonisamide; metformin with amantadine and zonisamide; metformin with sibutramine); antipsychotic switching (e.g. changing to aripiprazole, quetiapine, or ziprasidone); drugs that can prevent or improve metabolic side effects during antipsychotic treatments (melatonin); antidepressants (e.g fluoxetine, imipramine) or other medications.
The review included any behaviour change interventions aiming to prevent or delay the onset of diabetes, such as patient education programmes, psychological interventions (e.g. cognitive behavioural therapy or counselling, or motivational interviewing), self‐monitoring (including telehealth, internet‐based interventions and other communication technologies) and multicomponent interventions (e.g. self‐management programmes that combine education and behavioural approaches).
Organisational interventions included those that aim to improve the delivery of care, such as educating health professionals, care planning, or collaborative models of care. Examples include relevant training of any health professionals working with people with mental illness, nonspecific health worker interventions, community mental health teams, mass media‐delivered interventions, family interventions, physical health care monitoring, and statutory mental health services interventions.
Comparator intervention
For pharmacological interventions, comparator interventions included no treatment (including trials employing wait‐list conditions), treatment as usual, drug placebo, or an alternative type of medication for diabetes prevention.
For behaviour change and organisational interventions we included the following comparators: usual care or treatment (including pharmacological treatment), attention or other psychological placebo control, or any alternative behaviour change or organisational intervention (as described above under experimental interventions).
Types of outcome measures
Primary outcomes
Our primary outcome is prevention of diabetes, measured as a difference between study arms in the number of participants who developed type 2 diabetes during the study period. A clinical diagnosis of diabetes may be confirmed in the presence of symptoms by various parameters such as HbA1c, fasting blood sugar, random blood sugar or, in unclear cases, 2‐hour plasma glucose following an oral glucose tolerance test (OGTT). We accepted diagnoses made using any of these parameters using cut‐offs consistent with those current at the time of the study, as described in national and international guidance such as WHO (e.g. WHO 2006), National Institute of Health and Care Excellence (NICE) (e.g. NICE 2015), Diabetes UK (e.g. Diabetes UK 2018), American Diabetes Association (e.g. ADA 2017). Current cut‐offs are as follows: HbA1c ≥ 48 mmol/mol, a fasting blood glucose ≥ 7 mmol/L or a random blood glucose ≥ 11.1 mmol/L; and for OGTT 2‐hour glucose ≥ 11.1 mmol/L (ADA 2017). Conversion to prediabetes was not included as part of this outcome.
As the primary adverse outcome, we report drop‐out from care: the number of participants who dropped out of treatment for any reason after randomization.
Secondary outcomes
-
Fasting blood glucose (mmol/L or mg/dL)
-
BMI
-
Waist circumference (cm or inch)
-
Blood pressure (diastolic and systolic in mmHg)
-
Total cholesterol (mmol/L or mg/dL)
-
Depression and anxiety measured by a validated scale, e.g. Patient Health Questionnaire (PHQ 9) (Kroenke 2001), Generalised Anxiety Disorder assessment (GAD‐7) (Spitzer 2006)
-
Health related quality of life (evaluated with a validated generic or disease‐specific instrument (Wee 2006), e.g. the 36‐Item Short Form Health Survey (SF‐36) (McHorney 1993) or other validated scale). We considered language‐ and culture‐adapted instruments, where these were available
-
All‐cause mortality, defined as death from any cause
Search methods for identification of studies
Electronic searches
We searched the following electronic databases (20 February 2020) using a comprehensive list of keywords and subject headings related to diabetes, mental disorders, LMICs, RCTs and systematic reviews (Appendix 1). The search strategies were informed by the review of Taylor and colleagues (Taylor 2017), the Cochrane highly sensitive search strategies for identifying RCTs (Lefebvre 2011), and the Academic Unit of Health Economics (AUHE) Information Specialist's LMIC geographic strategies (AUHE 2018).
-
CINAHL (EBSCO) (1981 to 20 February 2020)
-
Cochrane Central Register of Controlled Trials (Issue 2, February 2020)
-
Cochrane Database of Systematic Reviews (Issue 2, February 2020)
-
Embase Classic + Embase (Ovid) (1947 to 19 February 2020)
-
Global Health (Ovid) (1910 to week 8, 2020)
-
Literatura Latino‐Americana e do Caribe em Ciências da Saúde (LILACS; Latin American and Caribbean Health Sciences Literature) (all available years)
-
Ovid MEDLINE (1946 onwards), MEDLINE In‐Process & Other Non‐Indexed Citations, MEDLINE Epub Ahead of Print
-
PsycINFO (Ovid) (1806 to February, week 2, 2020)
-
PubMed (US National Library of Medicine) (1946 to 20 February 2020)
-
PakMedNet (medical journals of Pakistan) (all available years)
We did not apply any limits on date, language, or publication status to the searches.
An information specialist with the Cochrane Common Mental Disorders Group ran a broad search of the Cochrane Common Mental Disorders Controlled Trials Register, using only terms for outcomes (Appendix 2).
In keeping with the Methodological Expectations of Cochrane Intervention Reviews (MECIR) conduct standards, we ran a search for retractions and errata once the included studies were selected.
Searching other resources
Grey literature
We searched the following sources of grey literature.
-
Conference Proceedings Citation Index ‐ Science (Clarivate Analytics Web of Science) (1990 to the search date)
-
ProQuest Dissertations & Theses Global
Unpublished studies
We searched the following international trial registries to identify ongoing or unpublished studies (all available years).
-
ISRCTN registry (Springer Nature)
-
ClinicalTrials.gov (US National Institutes of Health)
-
International Clinical Trials Registry Platform (WHO)
Reference lists
We checked the reference lists of relevant systematic reviews to identify additional studies.
Data collection and analysis
Selection of studies
We uploaded citations and available abstracts of the search results into Covidence (Covidence 2017) and screened records for potential eligibility in two stages. The first stage involved screening titles and abstracts to exclude studies that did not meet the inclusion criteria, carried out independently by pairs of review authors (from among EU, NS, MPM, SP, FA, ZAA and RA). We resolved discrepancies through discussion. Where we could not reach agreement, we consulted a third review author (NS). In the second stage, we retrieved the full texts of potentially eligible studies and independently assessed them for eligibility. This was again carried out by two review authors (from among EU, NS, MPM, FA, SP and ZAA). We resolved discrepancies by consulting a third review author (NS). We sought any missing data that could help to assess eligibility by contacting the corresponding authors. We created a PRISMA flow diagram to show the process of trial selection (Liberati 2009). For studies excluded during this stage, we recorded a reason for exclusion. For included studies, we linked multiple reports from the same study.
Data extraction and management
For trials that fulfilled our inclusion criteria, review authors extracted data in duplicate (EU, MPM, SP, NT, ZAA, FA). We resolved any discrepancies by discussion, or, if required, consulted a third review author (NS, RA).
To provide information for assessment of the certainty of the evidence and for evidence synthesis, we extracted the following data, where available.
-
Study population (including participant inclusion and exclusion criteria)
-
Country
-
Setting (primary care, community, secondary care, mental health care)
-
Study design (single or multicentre RCT)
-
Number of intervention groups
-
Intervention:
-
For pharmacological interventions: class of drug, dose, frequency, and duration.
-
For behaviour change and organisational interventions: a description of the intervention (including process, target group, e.g. patients or healthcare professionals, and presence of other concurrent interventions), theory (informing intervention design), target (including strategies, applications, and components), context of intervention (i.e. primary health facility), provider and mode of delivery (phone, face‐to‐face, group, online), intensity (length, frequency, and number of contacts), duration (period of time over which contacts delivered), details about group leader (demographics, training, professional status, etc.).
-
Behaviour change techniques: we planned to categorise interventions and identify behaviour change techniques using the ‘template for intervention description and replication’ (TIDieR) checklist (Hoffmann 2014; Hoffmann 2017).
-
-
Comparison intervention(s).
-
Outcome data and information on measures for our primary and secondary outcomes.
We noted in the 'Characteristics of included studies' table if the study authors did not report outcome data in a usable way. Where included trials reported outcome data in insufficient detail to include in a meta‐analysis, for instance, reporting means without confidence intervals (CIs) or standard deviations (SDs), we contacted the study authors to request more information.
Assessment of risk of bias in included studies
We assessed the risk of bias of included randomized trials using the Cochrane 'Risk of bias' tool (Higgins 2011a). Two review authors (from among EU, MPM, SP, NT, ZAA and FA) independently assessed the following items.
-
Sequence generation (i.e. if allocation sequence was adequately generated)
-
Allocation sequence concealment (i.e. if allocation was adequately concealed)
-
Blinding (i.e. if knowledge of the allocated interventions was adequately prevented during the study)
-
Incomplete outcome data (i.e. if incomplete outcome data was adequately addressed)
-
Selective outcome reporting (i.e. whether reports of the study are free of suggestion of selective outcome reporting)
-
Other potential sources of bias (i.e. whether the study is apparently free of other problems that could lead to a high risk of bias e.g. baseline imbalances, evidence of carry‐over in cross‐over trials, comparability of groups in cluster trials)
We judged each potential source of bias as high, low or unclear and provided a supporting quotation from the study report together with a justification for our judgment in the 'Risk of bias' table. We summarised the risk of bias judgements across different studies for each of the domains listed. Differences in assessment were resolved by discussion or consultation with a third review author (NS, EU). Allocation concealment was used as a marker of trial risk of bias for the purposes of undertaking sensitivity analyses.
Measures of treatment effect
For continuous data, we calculated the mean difference (MD) with 95% CIs. Where trials reported the same outcome using different outcome measures, we used standardised mean difference (SMD). For binary outcomes, we calculated a standard estimation of the risk ratio (RR) with a 95% CI.
Unit of analysis issues
We took into account the level at which randomization occurred, with respect to cross‐over trials, cluster RCTs, and multiple observations for the same outcome.
We planned to reanalyse cluster‐RCTs that had not appropriately adjusted for potential clustering of participants in their analyses by inflating the variance of the intervention effects by the design effect. We would have obtained estimates of the intracluster correlation coefficient (ICC) in order to estimate the design effect, through contact with authors, or by imputing them using either estimates from other included trials that reported ICCs or using external estimates from empirical research (e.g. Bell 2013).
In the case of multiple intervention groups, we analysed each intervention group separately against the control group and the sample size for the control group was divided proportionately across each intervention group. Where results were reported at multiple time points in the studies, we analysed each outcome at predefined periods of follow‐up in separate meta‐analyses. We grouped data by time‐point.
If more than one comparison from the same trial was eligible for inclusion in the same meta‐analysis, we either combined groups to create a single pairwise comparison or appropriately reduced the sample size so that the same participants did not contribute data to the meta‐analysis more than once (i.e. splitting the 'shared' group into two or more groups), although we acknowledge this will not account for correlation arising from the same set of participants being in multiple comparisons (Higgins 2011a) .
Dealing with missing data
We carefully evaluated important numerical data such as screened and randomly assigned participants as well as intention‐to‐treat (ITT), as‐treated and per‐protocol populations. We investigated attrition rates (e.g. dropouts, losses to follow‐up, withdrawals), and critically appraised issues concerning missing data.
We analysed data primarily using the ITT principle. In the protocol, we mentioned that if the included studies did not provide enough detail to allow an ITT analysis, and where included trials did not report means and SDs for outcomes, we planned to request data from study authors. If we did not receive the necessary information from trial authors, we planned to impute these values (Higgins 2011a; Higgins 2011b), and investigate the impact of imputation on meta‐analyses by performing sensitivity analyses. However, in the review, requesting further data from study authors or imputing data were not required.
Assessment of heterogeneity
We assessed clinical heterogeneity through the description of the setting, baseline measures, and the intervention approach used in each study. In the case of obvious clinical heterogeneity, we did not pool the data, and summarised results narratively instead.
We assessed statistical heterogeneity using the Chi2 test and the I2 statistic. The Chi2 test was considered statistically significant if P ≤ 0.10. If heterogeneity existed between studies (I2 ≥ 50%) for the primary outcome, we planned to explore the reasons, following guidance in Chapter 9 of the Cochrane Handbook for Systematic Reviews of Interventions (Deeks 2011). This chapter suggests the following guidance for interpretation of the I2 statistic:
-
0% to 40%: might not be important
-
30% to 60%: may represent moderate heterogeneity
-
50% to 90%: may represent substantial heterogeneity
-
75% to 100%: considerable heterogeneity
When interpreting the I2 statistic, we planned to take into account the magnitude and direction of effects and the strength of evidence for heterogeneity (e.g. the P value from the Chi2 test, or a CI for I2). However, in the review, this was not required.
Assessment of reporting biases
If more than 10 studies that investigated a particular outcome were identified for inclusion in this review, we planned to use funnel plots to assess publication biases. We also planned quantitative analysis of publication bias, using the Egger test.
Data synthesis
We combined data from individual trials in meta‐analysis if the interventions, outcomes, and patient groups were sufficiently similar (determined by consensus among review authors). Data were not pooled for meta‐analysis if we detected a high degree of clinical heterogeneity among the studies. Where data were pooled, we used a random‐effects model.
Subgroup analysis and investigation of heterogeneity
We planned to carry out the following subgroup analyses, based on characteristics of the population or intervention that might influence the primary outcome. However, as we identified only one study that assessed the primary outcome, we conducted no subgroup analyses.
-
Age (65 years and over): this may influence the risk of diabetes and effectiveness of the intervention.
-
Sex: this may influence the risk of diabetes and effectiveness of the intervention.
-
Type of mental disorder (severe mental illness versus other mental disorder): people with severe mental illness have additional risk factors for diabetes e.g. side effects of antipsychotic medication).
-
Prospective identification of diabetes using a robust approach to diagnosis e.g. HbA1c or fasting blood sugar, versus studies using retrospective records, random blood glucose testing, or both.
-
Intervention duration (less than three months versus three months or more): length of the intervention may influence outcomes.
-
Duration of follow‐up (less than three months versus three months or more): this is likely to influence detection of outcomes.
Sensitivity analysis
For outcomes where two or more studies were available to include in a meta‐analysis, we performed sensitivity analyses to explore the influence of the following factors (where applicable) on effect sizes:
-
effect of risk of bias: excluding studies that did not report allocation concealment (we acknowledge that we might have missed some studies where allocation concealment may have been used but not reported);
-
effect of large trials: excluding large trials to establish the extent to which they dominate the results;
-
effect of data imputation: excluding trials where missing data have been imputed.
Unplanned sensitivity analysis: we identified two studies (Wu 2006; Wu 2008b) reporting much lower standard deviations than other studies in our review and their effect estimates were extreme outliers in the meta‐analyses. After contacting authors, we were unable to confirm the validity of these data therefore we investigated the impact of removing these studies from our meta‐analyses.
Summary of findings and assessment of the certainty of the evidence
We prepared 'Summary of findings' tables to summarise key findings of this review. We reported the outcomes (including adverse outcomes) and presented standardised effect size estimates and 95% CIs, using the GRADE approach to assess the overall certainty (quality) of the evidence supporting each outcome. GRADE criteria take into account issues related not only to internal validity (risk of bias, inconsistency, imprecision, publication bias) but also to external validity, such as directness of results. We used GRADEproGDT to create our 'Summary of findings' tables (GRADEpro 2015), and followed standard methods as described in Chapter 11 of the Cochrane Handbook to prepare our 'Summary of findings' tables (Schünemann 2017). For each of our main comparisons, the following outcomes (measured at the latest time point) were included.
-
Diabetes diagnosis
-
Drop‐out from treatment
-
Fasting blood glucose level
-
BMI
-
Health‐related quality of life
-
All‐cause mortality
The definitive list of comparisons to be included in the ‘Summary of findings’ tables was agreed among review authors once the categories of interventions were known, guided by clinical relevance. This is because the range of interventions to be included was broad, and at the protocol stage, we were not certain which interventions would be identified by the review.
We created 'Summary of findings' tables after we entered data into RevMan (Review Manager 2014), had written up our results, and conducted the 'Risk of bias' assessment. However, the 'Summary of findings' tables were created before writing our discussion, abstract, and conclusions, to allow the opportunity to consider the impact of the risk of bias in the studies contributing to each outcome upon the mean treatment effect, and our confidence in these findings.
Results
Description of studies
Results of the search
We conducted initial searches up to April 2019 and updated our searches on 20 February 2020. We identified 7703 unique records, of which we excluded 7598 after screening titles and abstracts. Full‐text reports of 105 references were obtained and screened against our eligibility criteria. We excluded 55 articles (Figure 1). The most common reason for exclusions was that the study was not conducted in a LMIC (n = 48). This was often not apparent from reading the abstracts. Four studies were ongoing and 10 were awaiting classification. In total, we included 30 studies in the qualitative synthesis. Fifteen of these studies contributed data to meta‐analyses. We contacted the authors of 21 articles to clarify details on risk of bias of their studies and received responses from three authors. Full details of the study flow are given in a PRISMA flow diagram in Figure 1.
Study flow diagram.
Included studies
We included 30 studies, only one of which (Saddichha 2008) investigated the primary review question, prevention of type 2 diabetes (Characteristics of included studies).
Design
All included studies were parallel‐group RCTs. Two studies were multi‐centre studies (Baptista 2007; Ou 2013) and one was conducted across multiple countries (Tohen 2011).
Sample sizes
For the primary outcome, one study (Saddichha 2008) included 150 participants (99 participants with schizophrenia). An additional 29 studies assessed one or more of the secondary outcomes of the review (2481 participants), giving a total of 30 RCTs with 2631 participants. The number of total participants in any single study ranged from 18 to 260.
Setting
Eleven studies were conducted in China, eight were from Iran, four were from Venezuela, three were from India, two were from South Africa and one study was from Brazil. For one multi‐country study, only data from China were extracted, as other countries in the study were high‐income countries. Participant recruitment and conduct of all of these studies occurred in hospital settings.
Participants
Mean ages ranged from 25 years (Wu 2008a) to 68 years (Chen 2017). One study included only female participants (Moosa 2003), while all others included both men and women. The majority of studies (n = 24) included participants with a diagnosis of schizophrenia or a related psychotic disorder. Four studies included participants with a diagnosis of depression, and one study included participants with depression and bipolar disorder.
Interventions
Studies used a range of pharmacological interventions. Some of these interventions were aimed at treating mental disorders, but the study investigated their potential beneficial or adverse effects on risk factors for the development of diabetes. These include typical and atypical antipsychotic medication for the treatment of psychotic disorders such as schizophrenia, and antidepressants for the treatment of depressive symptoms and disorders. Other interventions were directly aimed at reducing the risk of diabetes, and increased blood glucose levels with metformin, melatonin, and various other medications and supplements.
We identified one study that included a pharmacological and a behaviour change intervention (Wu 2008b). We found no studies of organisational interventions.
Outcomes
Prevention of diabetes was reported in only one study (Saddichha 2008). However, all 30 included studies reported fasting blood glucose levels as a proxy for risk of developing diabetes, in addition to several of the other planned secondary outcomes.
Length of follow‐up was between four weeks (Agnihotri 2013) and 54 weeks (Emsley 2005), with the majority of studies reporting outcomes up to 12 or 14 weeks.
We combined outcome data from 14 studies in meta‐analyses (Agahi 2017; Baptista 2006; Baptista 2007; Carrizo 2009; Emsley 2005; Ghaeli 2004; Modabbernia 2014; Moosa 2003; Saddichha 2008; Salehi 2009; Wang 2012; Wu 2006; Wu 2008a; Wu 2008b). For other studies, it was not possible to combine outcome data in this way due to the variation in intervention and comparator groups used. These data are summarised narratively instead. Meta‐analysis was performed for the following secondary outcomes: fasting blood glucose, BMI, waist circumference, blood pressure, cholesterol and drop outs. No data were available for depression, anxiety, quality of life, or all‐cause mortality (details are summarised under ‘Effects of interventions').
Excluded studies
We excluded 55 studies. Among the excluded studies, 48 were not conducted in LMICs, three were not RCTs, two studies included interventions that were not aimed at preventing diabetes as a primary or secondary objective, one was on a paediatric population, and one did not have a majority of participants with a mental disorder. The numbers of excluded studies and all reasons for exclusion are shown in Figure 1.
Risk of bias in included studies
Risk of bias of included studies is summarised visually in Figure 2 (by domain) and Figure 3 (by study and by domain).
Risk of bias graph: review authors' judgements about each risk of bias item presented as percentages across all included studies.
Risk of bias summary: review authors' judgements about each risk of bias item for each included study.
Allocation
The risk of selection bias was unclear for 20 studies because these studies did not report the information required to assess the 'random sequence generation' and/or 'allocation concealment' bias domains. Eight studies were judged to be at low risk of selection bias for both domains, while one study was judged to be at high risk of bias because it appeared randomization took place between the treatment arms but not between treatment and control arms (Saddichha 2008).
Blinding
Risk of performance bias due to inadequate blinding of participants and/or personnel was unclear or high for 17 studies and low for the remaining 12 studies.
For 19 studies, the risk of detection bias was unclear or high due to incomplete information reported in the paper, as well as inadequate or lack of blinding of outcome assessors. For 10 studies, the risk of bias was low.
Incomplete outcome data
The risk of attrition bias was high for eight studies. For some studies, there were substantial differences between study arms in drop‐out rates and reasons for drop‐out were not reported. In other studies, participants who stopped their involvement in the study were retrospectively excluded, and/or no baseline data were provided. For another eight studies, the risk of attrition bias was unclear, e.g. because information about reasons for drop‐out was not reported.
Selective reporting
For the majority of studies (23 of 30), the risk of reporting bias was unclear because no link to a protocol or online trial registration was provided, or because the protocol was registered online only retrospectively (Akkasheh 2016; Ghaderi 2019; Wu 2008b; Zhao 2015).
Other potential sources of bias
Other potential sources of bias included involvement of the pharmaceutical industry in the funding with potential interference, writing‐up, and monitoring of the trial, no reporting of baseline participant characteristics, indications of failed randomization with unbalanced groups, adjusting of medication throughout the trial, exclusion of unsuitable or unmotivated participants, and unclear study rationale.
For one study, data presented at baseline and in a table raised questions due to identical estimates and 95% CIs for different groups and time points (Characteristics of included studies). The authors did not respond to our request for information (Wu 2008b).
Effects of interventions
See: Summary of findings 1 Atypical compared with typical antipsychotic medications for preventing type 2 diabetes in adults with mental disorders in low‐ and middle‐income countries; Summary of findings 2 Metformin compared with placebo for preventing type 2 diabetes in adults with mental disorders in low‐ and middle‐income countries; Summary of findings 3 Melatonin compared with placebo for preventing type 2 diabetes in adults with mental disorders in low‐ and middle‐income countries; Summary of findings 4 SSRI antidepressants compared with TCA for preventing type 2 diabetes in adults with mental disorders in low‐ and middle‐income countries
Only one study reported on diabetes prevention (Saddichha 2008). The other 29 included studies assessed only secondary outcomes. All 30 included studies compared a pharmacological intervention with a placebo, behavioural intervention, and/or another pharmacological intervention, for patients with mental disorders.
For analysis of the outcomes, we categorised these 30 studies into seven groups.
-
Atypical versus typical antipsychotic medication (Emsley 2005; Saddichha 2008; Wu 2006)
-
Metformin versus placebo; in all trials both the metformin and placebo arms additionally received antipsychotic medication (Baptista 2006; Baptista 2009; Carrizo 2009; Wang 2012; Wu 2008a; Wu 2008b).
-
Melatonin versus placebo (Agahi 2017; Modabbernia 2014)
-
SSRI versus tricyclic antidepressants (TCA) for the treatment of depressive symptoms (Ghaeli 2004; Moosa 2003; Salehi 2009)
-
Atypical versus atypical antipsychotic medications (Chen 2017; Hu 2013; Li 2009; Ou 2013; Zhang 2012; Zhang 2014)
-
Miscellaneous interventions versus placebo, including supplements and vitamins, antidepressant medications, and antipsychotic medications (Agnihotri 2013; Akkasheh 2016; Baptista 2009; Assunção 2006; Fadai 2014; Ghaderi 2019; Narula 2010; Tohen 2011; Zhao 2015)
-
Behavioural intervention versus placebo and metformin (Wu 2008b)
Atypical antipsychotics included aripiprazole, quetiapine, clozapine, olanzapine, ziprasidone, paliperidone, and risperidone. Typical antipsychotics included haloperidol and sulpride.
Meta‐analyses could only be conducted for the first four categories of studies because studies of atypical versus atypical medications could not be compared directly. Studies of miscellaneous interventions could not be compared due to heterogeneity in the interventions and target populations (patients with depressive disorder or schizophrenia).
Sensitivity analyses could only be performed for low risk of bias studies comparing metformin to placebo because no data were available for other outcomes and comparisons. Forest plots were not constructed due to the small number of studies for each comparison.
1. Atypical versus typical antipsychotics
This comparison includes medication for the treatment of psychotic symptoms in patients with schizophrenia or schizoaffective disorder.
Outcome 1.1 Diabetes prevention
One study of 99 participants with schizophrenia showed there may be no difference in the prevention of diabetes between participants with atypical versus typical antipsychotics six weeks after starting treatment (RR 0.46, 95% CI 0.03 to 7.05; 99 participants; 1 study; low‐certainty evidence) (Analysis 1.1) (among total 99 participants with schizophrenia, 68 were in atypical and 31 were in typical antipsychotics group; 55 participants without mental disorder were not considered in the analysis).
Outcome 1.2 Drop‐outs
Two studies were included, but only one reported any drop‐outs. The analysis shows there may be no difference in drop‐out rates between study arms 54 weeks after baseline (RR 1.31, 95% CI 0.63 to 2.69; 144 participants; 2 studies; low‐certainty evidence) (Analysis 1.2).
Outcome 1.3 Fasting blood glucose
Two studies reported on fasting blood glucose levels. The analysis shows there is probably no difference between patients taking atypical versus typical antipsychotics six and eight weeks after starting treatment (MD ‐0.05, 95% CI ‐0.10 to ‐0.00; 211 participants; 2 studies; moderate‐certainty evidence) (Analysis 1.3).
Unplanned sensitivity analysis
Given our concerns about the data presented by Wu 2006 (very small SDs), we investigated the impact of these data on the outcomes by removing the study from the analysis, in an unplanned sensitivity analysis. Removing this study shows no difference between patients taking atypical versus typical antipsychotics (MD 0.01, 95% CI ‐0.35 to 0.37; 99 participants; 1 study) (Analysis 7.1).
Outcome 1.4 BMI
Participants who received typical antipsychotics probably had a lower BMI at follow‐up than participants who received atypical antipsychotics (MD 0.57, 95% CI 0.33 to 0.81; 141 participants; 2 studies; moderate certainty‐evidence) (Analysis 1.4).
Unplanned sensitivity analysis
Given our concerns about the data presented by Wu 2006 (very small SDs), we investigated the impact of these data on the outcomes by removing the study from the analysis, in an unplanned sensitivity analysis. Removing this study shows no difference between patients taking atypical versus typical antipsychotics (MD ‐1.13, 95% CI ‐5.65 to 3.39; 29 participants; 1 study) (Analysis 7.2).
Outcome 1.5 Total cholesterol
Results indicated that total cholesterol eight weeks after starting treatment was probably lower for participants who received typical rather than atypical antipsychotics (MD 0.35, 95% CI 0.27 to 0.43; 112 participants; 1 study) (Analysis 1.5).
No evidence was available for the outcomes of waist circumference, blood pressure, depression, anxiety, quality of life, and all‐cause mortality.
2. Metformin versus placebo
This comparison includes studies of patients with schizophrenia or related disorders comparing metformin, an anti‐diabetic medication, to a placebo. In all studies, both groups also received antipsychotic medication, usually olanzapine or clozapine (atypical antipsychotics). The dose of metformin varied across different studies between 750 mg to 2250 mg per day. Studies reported outcomes at 12 to 14 weeks follow‐up.
Outcome 2.1 Drop‐outs
We found there may be no difference in drop‐outs between participants receiving metformin and those receiving placebo (RR 1.22, 95% CI 0.09 to 16.35; 158 participants; 3 studies; moderate‐certainty evidence) (Analysis 2.1).
Unplanned sensitivity analysis
Given our concerns about the data presented by Wu 2008b (Risk of bias in included studies), we investigated the impact of these data on the outcomes by removing the study from the analysis, in an unplanned sensitivity analysis. Removing this study mostly affected the precision of the analysis (RR 2.78, 95% CI 0.08 to 95.87; 94 participants; 2 studies) (Analysis 6.1).
Outcome 2.2 Fasting blood glucose
Five studies reported fasting blood glucose (264 participants). Three studies reported endpoint data and two reported change from baseline data. The use of different measures meant we could not combine these estimates.
There was no difference in fasting blood glucose between participants in the metformin and placebo groups for endpoint data (MD ‐0.35, 95% CI ‐0.60 to ‐0.11; 173 participants; 3 studies; moderate‐certainty evidence) and change from baseline data (MD 0.01, 95% CI ‐0.21 to 0.22; 91 participants; 2 studies; high‐certainty evidence) (Analysis 2.2).
Planned sensitivity analysis
A sensitivity analysis of high certainty studies, removing Baptista 2006 from the analysis of endpoint data, did not substantially alter the results (MD‐0.37, 95% CI ‐0.68 to ‐0.07; 136 participants; 2 studies) (Analysis 5.1).
Unplanned sensitivity analysis
We removed Wu 2008b from the analysis because of concerns about the primary data. This study was also an outlier, with lower fasting blood glucose levels in the metformin group than in the placebo group at 12 weeks (weighting 19.9%). Without this study, the effect size estimate for endpoint data changed and the CI narrowed (MD‐0.19, 95% CI ‐0.46 to 0.09; 109 participants; 2 studies) (Analysis 6.2). Heterogeneity as indicated by I2 declined from 89% to 0%.
Outcome 2.3 BMI
Endpoint and change from baseline data were included in one meta‐analysis for this outcome, as all studies used the same measure of BMI. BMI was lower for participants receiving metformin compared with those receiving a placebo (MD ‐1.37, 95% CI ‐2.04 to ‐0.70; 264 participants; 5 studies; high‐certainty evidence) (Analysis 2.3).
Planned sensitivity analysis
Findings were similar in a sensitivity analysis including high certainty studies (MD ‐1.46, 95% CI ‐2.15 to ‐0.77, I2 = 77%; 227 participants; 4 studies) (Analysis 5.2).
Unplanned sensitivity analysis
Removing Wu 2008b from the analysis reduced the magnitude and precision of the effect (MD ‐1.13, 95% CI ‐1.86 to ‐0.40; 200 participants; 4 studies) (Analysis 6.3).
Outcome 2.4 Waist circumference
The estimates for endpoint (MD‐0.30, 95% CI ‐6.26, 5.66; 101 participants; 2 studies) and change from baseline (MD ‐0.93, 95% CI ‐1.21 to ‐0.64; 91 participants; 2 studies) data did not show a difference between metformin compared with placebo for waist circumference.
Unplanned sensitivity analysis
Statistical heterogeneity was high (I2 = 96% and I2 = 89%), because two of the four studies, with the same first author, showed relatively precise estimates that strongly favoured metformin (Wu 2008a; Wu 2008b). The other studies showed a lesser difference or no difference. Removing Wu 2008b from the analysis meant only one study with endpoint data was included for this outcome (Baptista 2006), which did not show a difference between metformin and placebo in terms of waist circumference (MD 3.40, 95% CI ‐1.99 to 8.79; 37 participants; 1 study) (Analysis 6.4).
Outcome 2.5 Systolic blood pressure
Systolic blood pressure was reported by one study (Carrizo 2009) and showed no difference between metformin and placebo (MD ‐2.50, 95% CI ‐9.09 to 4.09; 54 participants; 1 study) (Analysis 2.5).
Outcome 2.6 Diastolic blood pressure
Diastolic blood pressure was reported by one study (Carrizo 2009) and showed no difference between metformin versus placebo group (MD 1.20, 95% CI ‐3.55 to 5.95; 54 participants; 1 study) (Analysis 2.6).
Outcome 2.7 Total cholesterol
Meta‐analysis based on two studies showed no difference between metformin and placebo in total cholesterol (MD ‐13.06, 95% CI ‐35.89 to 9.76; 99 participants; 2 studies) (Analysis 2.7).
No data were available for diabetes prevention, depression, anxiety, quality of life, and mortality.
3. Melatonin versus placebo
Three studies included the hormone melatonin versus a placebo in patients with psychotic symptoms and schizophrenia. The dose of melatonin used was 3 mg per day. Outcomes were reported at 8 to 12 weeks.
Outcome 3.1 Drop‐outs
Based on findings from one study, there may be no difference in drop‐out rates between melatonin and placebo (RR 1.00, 95% CI 0.38 to 2.66; 48 participants; 1 study; low‐certainty evidence) (Analysis 3.1).
Outcome 3.2 Fasting blood glucose
Fasting blood glucose was probably reduced more at endpoint in the melatonin compared with the placebo group when considering endpoint data (MD ‐0.17, 95% CI ‐0.35 to 0.01; 102 participants; 2 studies; moderate‐certainty evidence) and change from baseline data (MD ‐0.24, 95% CI ‐0.39 to ‐0.09); 100 participants; 1 study; moderate‐certainty evidence) (Analysis 3.2).
Outcome 3.3 BMI
Three studies reported BMI, it was uncertain if there was any difference between patients with melatonin versus those in the placebo group (MD ‐0.22, 95% CI ‐2.58 to 2.14; 202 participants; 3 studies; very low‐certainty evidence) (Analysis 3.3). Statistical heterogeneity was high (I2 = 93%) due to one study showing a positive effect and one showing a negative effect of melatonin on BMI, compared with placebo.
Outcome 3.4 Waist circumference
Two studies reported waist circumference. Analyses showed there may be no difference between patients receiving melatonin versus those in the placebo group for endpoint data (MD 0.68, 95% CI 0.47 to 1.83; 36 participants; 1 study). Analyses showed there may be a benefit of placebo versus melatonin for change from baseline data (MD 1.19, 95% CI 0.29 to 2.09; 100 participants; 1 study) (Analysis 3.4).
Outcome 3.5 Systolic blood pressure
There may be no difference in systolic blood pressure between melatonin versus placebo in the combined results of two studies (MD ‐1.31, 95% CI ‐6.46 to 3.84; 134 participants; 2 studies) (Analysis 3.5). However, these results showed a high level of statistical heterogeneity (I2 = 86%). In one study, systolic blood pressure at endpoint was higher in the melatonin compared with the placebo group, and in the other study systolic blood pressure reduced in the melatonin group over the course of the study and increased in the placebo group.
Outcome 3.6 Diastolic blood pressure
Two studies reported on diastolic blood pressure, showing there may be no difference between melatonin and placebo (MD ‐1.05, 95% CI ‐1.60 to ‐0.50; 134 participants; 2 studies) (Analysis 3.6).
Outcome 3.7 Total cholesterol
Results from two studies (134 participants) indicated there may be no difference in total cholesterol between patients receiving melatonin and placebo using endpoint data (MD ‐0.11, 95% CI ‐0.27 to 0.05; 36 participants; 1 study) and change from baseline data (MD 0.02, 95% CI ‐0.19 to 0.23; 100 participants; 1 study) (Analysis 3.7).
No evidence was available for diabetes prevention, depression, anxiety, quality of life, and all‐cause mortality.
4. SSRI versus TCA
This is a comparison of fluoxetine, an SSRI antidepressant, and imipramine, a TCA, with three studies contributing data. The dose varied across different studies between 20 mg to 40 mg per day for fluoxetine and between 50 mg to 200 mg per day for imipramine. Outcomes were reported at 8 to 12 weeks.
Outcome 4.1 Drop‐outs
One study reported a higher percentage of drop‐out in the TCA group compared with the SSRI group (RR 0.34, 95% CI 0.11 to 1.01; 25 participants; 1 study; very low‐certainty evidence) (Analysis 4.1).
Outcome 4.2 Fasting blood glucose
There was probably no difference in fasting blood glucose between participants who received an SSRI and a TCA (MD ‐0.39 lower, 95% CI ‐0.88 to 0.10; 141 participants; 3 studies; moderate‐certainty evidence) (Analysis 4.2).
Outcome 4.3 BMI
Results from one study indicated that there may be no difference in BMI between SSRI and TCA (MD 0.70, 95% CI ‐1.10 to 2.50; 18 participants; 1 study; very low‐certainty evidence) (Analysis 4.3).
Outcome 4.4 Depression
Results from one study showed there may be no difference in depression symptoms between SSRI and TCA (MD 0.30, 95% CI ‐0.59 to 1.19; 18 participants; 1 study) (Analysis 4.4).
No evidence was available for diabetes prevention, waist circumference, blood pressure, cholesterol, anxiety, quality of life, and all‐cause mortality.
5. Atypical versus atypical antipsychotics
Six studies with 594 participants in total compared participants receiving different types of atypical antipsychotic medications. Results for five studies reporting outcome data at endpoint (between six and 52 weeks) are shown in Table 1.
| Reference | Comparison | Follow‐up | Outcome | mean (SD) group 1 | number of drop‐outs group 1 | n/N group1 | mean (SD) group 2 | number of drop‐outs group 2 | n/N group2 |
| Chen 2017 | ziprasidone (n = 19) versus olanzapine (n = 19) | 12 weeks | fasting blood glucose (mmol/L) | 5.59 (1.02) | 19/38 | 7.28 (2.22) | 19/38 | ||
| cholesterol (mmol/L) | 1.28 (0.2) | 2.24 (0.31) | |||||||
| Ou 2013 | ziprasidone (n = 130) versus olanzapine (n = 130) | 6 weeks | drop‐outs | 11 | 130/260 | 19 | 130/260 | ||
| fasting blood glucose (mmol/L) | 4.40 (0.50) | 4.94 (0.50) | |||||||
| cholesterol (mmol/L) | 4.06 (0.74) | 4.51 (0.80) | |||||||
| BMI (kg/m2) | 20.87 (3.34) | 22.28 (2.93) | |||||||
| systolic blood pressure (mm Hg) | 117.43 (13.13) | 117.97 (10.50) | |||||||
| diastolic blood pressure (mm Hg) | 75.93 (7.90) | 76.30 (5.65) | |||||||
| Hu 2013 | paliperidone (n = 33) versus olanzapine (n = 23) | 12 weeks | drop‐outs | 7 | 33/56 | 17 | 23/56 | ||
| fasting blood glucose (mmol/L) | 5.21 (0.6) | 5.19 (0.6) | |||||||
| BMI (kg/m2) | 22.17 (3.31) | 23.17 (4.06) | |||||||
| waist circumference (cm) | 80.3 (12.47) | 82.63 (9.49) | |||||||
| Zhang 2014 | aripiprazole (n = 50) versus olanzapine (n = 50) | 8 weeks | drop‐outs | 5 | 50/100 | 5 | 50/100 | ||
| fasting blood glucose (mmol/L) | 5.49 (1.51) | 5.49 (1.51) | |||||||
| cholesterol (mmol/L) | 4.11 (1.04) | 4.72 (1.12) | |||||||
| Zhang 2012 | aripriprazole (n = 71) versus ziprasidone (n = 69) | 52 weeks | drop‐outs | 19 | 71/140 | 14 | 69/140 | ||
| fasting blood glucose (mmol/L) | 5.2 | 4.6 | |||||||
| cholesterol (mmol/L) | 5.1 | 4.7 | |||||||
| BMI (kg/m2) | 24.5 (5.9 | 20.3 (5.2) | |||||||
| waist circumference (cm) | 71.6 (17.6) | 70.3 (16.7) |
SD: standard deviation; n: number of participants; N: total number of participants; BMI: Body mass index.
Comparisons within studies included ziprasidone, olanzapine, paliperidone, aripiprazole, and ziprasidone. Studies reported on drop‐outs, fasting blood glucose, cholesterol, BMI, waist circumference, and systolic and diastolic blood pressure.
Individual studies generally reported more favourable results in terms of blood glucose level and lipid metabolism (BMI, cholesterol, waist circumference) from ziprasidone than olanzapine or other antipsychotics (Chen 2017; Ou 2013; Zhang 2012).
6. Miscellaneous drugs
Ten studies with 699 participants in total were included under the miscellaneous group. Results of these studies reporting outcome data at endpoint (four to 12 weeks) are shown in Table 2.
| Reference | Comparison | Follow‐up | Outcome | mean (SD) group 1 | number of drop‐outs group 1 | n/N group1 | mean (SD) group 2 | number of drop‐outs group 2 | n/N group 2 | mean (SD) group 3) | number of drop‐outs group 3 | n/N group 3 |
| Agnihotri 2013 | Withania somnifera (n = 12) versus placebo (n = 13) | 4 weeks | fasting blood glucose (mmol/L) | 5.14 ( 0.33) | 12/25 | 5.82 ( 0.46) | 13/25 | |||||
| Akkasheh 2016 | probiotic supplements (n = 20) versus placebo (n = 20) | 8 weeks | drop‐outs | 3 | 20/40 | 2 | 20/40 | |||||
| fasting blood glucose (mmol/L) | 5.54 (0.97) | 4.96 (0.42) | ||||||||||
| cholesterol (mmol/L) | 9.58 (1.88) | 9.98 (1.72) | ||||||||||
| BMI (kg/m2) | 27.5 (5.9) | 26.5 (3.9) | ||||||||||
| Assunção 2006 | nizatidine (n = 27) versus placebo (n = 27) | 12 weeks | fasting blood glucose (mmol/L) | 4.84 (0.81) | 27/54 | 4.70 (0.98) | 27/54 | |||||
| cholesterol (mmol/L) | 11.04 (2.56) | 10.02 (1.89) | ||||||||||
| Baptista 2009 | rosiglitazone (n = 14) versus placebo (n = 15) | 12 weeks | fasting blood glucose (mmol/L) | 4.48 (0.77) | 14/29 | 4.38 (0.48) | 15/29 | |||||
| cholesterol (mmol/L) | 9.99 (1.26) | 11.18 (1.45) | ||||||||||
| BMI (kg/m2) | 26.9 (4.2) | 26.3 (2.9) | ||||||||||
| waist circumference (cm) | 91.5 (11.6) | 90.5 (8.6) | ||||||||||
| Fadai 2014 | saffron aqueous extract (n = 20) versus crocin (n = 20) | 12 weeks | drop‐outs | 2 | 20/61 | 2 | 20/61 | 1 | 21/61 | |||
| fasting blood glucose (mmol/L) | 5.58 (0.39) | 5.44 (0.38) | 6.03 (0.56) | |||||||||
| cholesterol (mmol/L) | 9.36 (1.35) | 9.94 (1.91) | 10.9 (1.91) | |||||||||
| waist circumference (cm) | 92.1 (7.4) | 91.9 (8.6) | 98.4 (8.4) | |||||||||
| blood pressure (mm Hg) | 116.5 (5.6) | 109.7 (6.8) | 115.9 (9) | |||||||||
| Ghaderi 2019 | vitamin D and probiotic supplements (n = 30) versus placebo (n = 30) | 12 weeks | drop‐outs | 4 | 30/60 | 4 | 30/60 | |||||
| fasting blood glucose (mmol/L) | 4.89 (0.62) | 5.17 (0.48) | ||||||||||
| cholesterol (mmol/L) | 8.99 (2.04) | 9.92 (1.99) | ||||||||||
| BMI (kg/m2) | 23.2 (2.7) | 24.5 (3.7) | ||||||||||
| Narula 2010 | topiramate (n = 33) versus placebo (n = 34) | 12 weeks | fasting blood glucose (mmol/L) | 4.35 (0.37) | 33/67 | 4.92 (0.67) | 34/67 | |||||
| BMI (kg/m2) | 20.1 (4) | 22.55 (4.11) | ||||||||||
| diastolic blood pressure (mm Hg) | 77.94 (4.8) | 81.41 (6.2) | ||||||||||
| systolic blood pressure (mm Hg) | 177.88 (7) | 122.5 (7.71) | ||||||||||
| Tohen 2011 | olanzapine (n = 140) versus placebo (n = 70) | 6 weeks | drop‐outs | 35 | 140/210 | 35/70 | 35 | 70/210 | ||||
| fasting blood glucose (mmol/L) | 5.05 (0.53) | 5.03 (0.61) | ||||||||||
| cholesterol (mmol/L) | 4.47 (0.95) | 4.56 (0.94) | ||||||||||
| diastolic blood pressure (mm Hg) | 72.72 (8.09) | 72.17 (7.36) | ||||||||||
| systolic blood pressure (mm Hg) | 109.84 (11.41) | 109.76 (11.59) | ||||||||||
| Zhao 2015 | aripiprazole (n = 56) versus placebo (n = 57) | 6 weeks | drop‐outs | 2 | 56/113 | 4 | 57/113 | |||||
| fasting blood glucose (mmol/L) | 4.75 (0.8) | 4.95 (0.87) | ||||||||||
| cholesterol (mmol/L) | 3.32 (0.92) | 4.54 (1.11) | ||||||||||
| BMI (kg/m2) | 23.27 (2.63) | 24.8 (3.92) | ||||||||||
| Sepehrmanesh 2016 | Vitamin D (n = 20) versus placebo (n = 20) | 8 weeks | drop‐outs | 2 | 20/40 | 2 | 20/40 | |||||
| fasting blood glucose(mmol/L) | 4.69 (0.48) | 5.02 (0.56) | ||||||||||
| cholesterol (mmol/L) | 4.69 (0.48) | 5.02 (0.56) | ||||||||||
| BMI (kg/m2) | 26.0 (5.1) | 27.3 (3.5) | ||||||||||
| Depression score, Beak Depression Inventory (BDI) | 17.2 (10.6) | 25.2 (9.9) |
SD: standard deviation; n: number of participants; N: total number of participants; BMI: Body mass index.
Comparisons within the studies included Withania somnifera (an herbal medicine) versus placebo in participants with schizophrenia (Agnihotri 2013); probiotic versus placebo in participants with depression (Akkasheh 2016); rosiglitazone versus placebo in participants receiving olanzapine (Baptista 2009); nizatidine versus placebo in participants receiving olanzapine (Assunção 2006); saffron versus placebo in participants with schizophrenia, receiving olanzapine (Fadai 2014); vitamin D and probiotic versus placebo in participants with schizophrenia (Ghaderi 2019); topiramate versus placebo in participants with schizophrenia, receiving olanzapine (Narula 2010); olanzapine versus placebo in participants with depression (Tohen 2011); aripiprazole versus placebo in participants with schizophrenia, on risperidone (Zhao 2015); and vitamin D versus placebo in participants with major depressive disorder (Sepehrmanesh 2015).
A statistically significant reduction was observed for fasting blood glucose for Withania somnifera (Agnihotri 2013); however, the full results were not available. Significant decreases in fasting blood glucose level and cholesterol were reported for vitamin D plus probiotic supplements (Ghaderi 2019) and topiramate (Narula 2010) compared with placebo. Other studies did not report any significant differences in other assessed outcomes.
7. Behavioural intervention
One study with 128 participants included four study arms: lifestyle intervention plus medication (metformin); medication (metformin); lifestyle intervention plus placebo; and placebo (Wu 2008b). Medication versus placebo results are included in Analysis 2. At the 12 week endpoint, the mean fasting blood sugar levels were lowest in the lifestyle plus metformin and lifestyle plus placebo groups (88.3, 95% CI 86.5 to 90.1 for both groups), and in the metformin group (84.7, 95% CI 82.9 to 86.5). They were highest in the placebo group (93.7, 95% CI 91.9 to 95.5).
Discussion
Summary of main results
We included 30 RCTs with 2631 participants from hospital settings in LMICs including Brazil, China, India, Iran, South Africa and Venezuela. All studies were of pharmacological interventions; one included a pharmacological and a behavioural intervention (Wu 2008b). We did not identify any studies evaluating organisational interventions. Only one study reported on our primary outcome, prevention or delay of diabetes. All studies reported blood glucose level as a proxy for risk of developing diabetes.
Atypical versus typical antipsychotics
Three studies were included in this group, with 256 participants with schizophrenia. Only one study reported there may be no difference between atypical and typical antipsychotics in prevention of diabetes (99 participants with schizophrenia, low‐certainty evidence). There maybe no difference between atypical and typical antipsychotics in drop‐outs (two studies including 144 participants, low‐certainty evidence), and fasting blood glucose levels (two studies including 211 participants, moderate‐certainty evidence). However, participants receiving typical antipsychotics showed more favourable results for BMI in terms of reduction or smaller increase (two studies including 141 participants, moderate‐certainty evidence) and lower cholesterol level (one study with 112 participants). No data were available for waist circumference, blood pressure, depression, anxiety, quality of life, and all‐cause mortality.
Metformin versus placebo
Five studies were included in this group with 264 participants. There may be no difference in drop‐outs (moderate‐certainty evidence), fasting blood glucose (moderate‐certainty evidence), waist circumference, blood pressure, and cholesterol between participants receiving metformin and placebo. For participants receiving metformin rather than placebo, BMI was lower at the trial endpoint or further reduced (high‐certainty evidence). Evidence for fasting glucose and waist circumference was highly inconsistent and unclear. There were no data available for effectiveness in diabetes prevention, depression, anxiety, quality of life, and all‐cause mortality.
Melatonin versus placebo
In three studies with 202 participants, there may be no differences in drop‐outs (low‐certainty evidence), BMI (very low‐certainty evidence), blood pressure, and cholesterol level, between participants receiving melatonin and placebo. Moderate‐certainty evidence showed that fasting blood glucose was probably lower or reduced more for participants receiving melatonin, compared to those receiving placebo. Two studies reporting on waist circumference found either no difference or a greater reduction for placebo, compared with melatonin. Evidence for BMI and systolic blood pressure was highly inconsistent. No data were available for prevention of diabetes, depression, anxiety, quality of life, and all‐cause mortality.
SSRIs versus TCA
Three studies with 141 participants found there may be no differences between participants receiving SSRI and TCA in fasting blood glucose levels (moderate‐certainty evidence), BMI (very low‐certainty evidence), and depression. Drop‐outs were higher for participants receiving TCA, rather than SSRIs, but the evidence was judged to be of very low certainty. No data were available for prevention of diabetes, waist circumference, blood pressure, cholesterol, anxiety, quality of life, and all‐cause mortality.
Other comparisons
Results from 15 studies could not be included in meta‐analyses. These studies could not be compared directly due to heterogeneity in the intervention groups.
Overall completeness and applicability of evidence
Studies did not explicitly state that the aim of the evaluated intervention was to prevent type 2 diabetes, although many included relevant outcomes such as fasting blood glucose levels. Only one study included diagnosis of diabetes as an outcome (Saddichha 2008). Fasting blood glucose is a short‐term proxy for risk of diabetes, but a raised blood glucose level does not necessarily equate to development of diabetes in the future. Likewise, measures such as total cholesterol level and BMI may indicate a reduced or increased risk of diabetes, but these changes were often measured in a short timeframe and cannot be seen as a reliable indicator of diabetes prevention. The majority of the evidence identified in this review therefore does not necessarily apply to settings in which the aim is to prevent diabetes.
We did not examine pre‐diabetes as an outcome, and we did not examine the subset of study populations at a higher risk of diabetes at baseline.
The evidence we identified was based on studies conducted in six countries (Brazil, China, India, Iran, South Africa and Venezuela) in hospital settings. This may not be representative of the global population of people with mental disorders.
A further limitation is that all studies we identified were conducted in hospital settings. This may not be an appropriate setting to provide or evaluate diabetes prevention interventions for people with mental health conditions, many of whom are unlikely to frequently visit a hospital to receive mental healthcare. Most studies included participants with schizophrenia, while common mental disorders, which are more prevalent worldwide, were less frequently studied. There were also gaps in the evidence for behavioural and organisational interventions, with studies mostly focussed on antipsychotics, antidepressants, and melatonin.
Finally, interventions may be most effective for people who are highly motivated to prevent diabetes, and these may have been more likely to participate in the included studies. For example, one study used willingness to lose weight as a criteria for participation (Baptista 2009). This may mean that the evidence we found overestimates the effectiveness of interventions for diabetes prevention for the general population.
Quality of the evidence
The certainty of the evidence ranged from very low to high. For the primary outcome of diabetes prevention, one study contributed low‐certainty evidence. We downgraded this evidence for imprecision. For other outcomes, evidence was downgraded for imprecision, high risk of bias, and inconsistency of results.
Risk of bias domains were frequently judged ‘unclear’ because the information required was not reported. For example, only three studies provided a link to a prospectively published protocol or trial registration. For some studies, a risk of attrition bias was identified, and not all studies were double‐blinded. Several studies were funded by the pharmaceutical industry, and in some cases it appeared the funder was involved in the research.
We had concerns about the data for one study and could not reach authors to seek clarification (Wu 2008b). We explored the impact of this study in unplanned sensitivity analyses. Comparing metformin to placebo, this study showed a more favourable profile for metformin than other studies for fasting blood glucose, BMI, and waist circumference.
We could not investigate the impact of reporting bias due to the limited number of studies available for each comparison.
Potential biases in the review process
We applied a comprehensive search strategy to identify all available evidence. However, it is possible that relevant studies were not identified. Successful searching and screening of studies relies on the identification of keywords in the titles, and abstracts of manuscripts. Although included studies measured a diabetes‐related outcome such as fasting blood glucose, the studies were rarely described as an evaluation of interventions aimed at preventing type 2 diabetes and only one study reported on diabetes prevention.
We identified an additional study through checking of systematic review reference lists, which was not found through database searches, possibly because the focus of the study was on weight gain rather than prevention of diabetes (Wu 2008b). Although we searched global and regional databases as well as grey literature, we may also have missed literature not published in English, or published in databases or on websites not covered by our search. We had planned to include data from multi‐country studies if authors would supply data for LMICs separately. However, these data were obtained for only one study. For ten studies, data could not be obtained (Characteristics of studies awaiting classification).
During the selection of studies we realised that, instead of assessing a clinical diagnosis of diabetes, studies frequently measured fasting glucose levels. We added this to our outcomes after registering the protocol, because we felt it provided important evidence on the efficacy of included interventions (Differences between protocol and review).
Although we only included RCTs in this review, to increase the internal validity of our findings, some bias inherent to the RCT design is likely to remain. For example, the trials are likely to represent a selective population, which may mean that results cannot easily be translated to the wider population of people with mental health disorders.
Despite contacting authors of primary studies to obtain missing information, we were not successful in obtaining all information. For risk of bias assessments in particular, this meant we frequently had to rate a domain as 'unclear.' This could mean our appraisal of the evidence is skewed towards more positive or more negative ratings. The review was not inclusive to pregnant women with mental health disorders.
Our review grouped all LMICs together, while in reality these countries represent diverse populations and settings which differ in their demographics, wealth levels, healthcare systems, and determinants and prevalence of diabetes in the population.
Agreements and disagreements with other studies or reviews
We are not aware of similar reviews including studies only from LMICs. As with the evidence base identified in our review, evidence from high‐income countries is mostly focussed on weight‐loss outcomes, rather than diabetes prevention.
Atypical versus typical antipsychotics
Our evidence mostly indicated no difference between atypical and typical antipsychotics for markers of diabetes prevention, except for more favourable impacts for typical antipsychotics on BMI and cholesterol levels. The wider literature emphasises the increased risk of diabetes for atypical antipsychotics and focuses on the difference between types of atypical antipsychotics (see 'Other comparisons') rather than comparing atypical to typical antipsychotics.
Metformin versus placebo
While we did not find strong evidence to suggest that metformin could prevent the increased risk of diabetes caused by antipsychotic medications, a clinical trial from Taiwan reported favourable results of metformin compared with a placebo for fasting blood glucose and related measures (Chen 2008).
Melatonin versus placebo
From our findings, we cannot draw conclusions on the potential effects of melatonin on prevention of diabetes among people with severe mental illness. In a narrative review of global literature, the potential for melatonin to be used for diabetes prevention in people with schizophrenia was also said to be unclear (Morera‐Fumero 2013).
SSRIs versus TCA
In our review, evidence comparing SSRI and TCA antidepressants was largely absent or inconclusive. A review of case‐control studies and cohort studies found that the risk of developing diabetes was increased among people who were prescribed an SSRI or a TCA to treat depression, without any indication of a difference in effect estimates between the two (Yoon 2013). A review of randomized and non‐randomized studies concluded that successful treatment of depression, measured by a decrease in the severity of depression symptoms, could improve glycaemic control in patients who have comorbid diabetes (Roopan 2017). The authors suggested that treatment with a TCA would require closer glycaemic monitoring than treatment with an SSRI, but no meta‐analysis was performed to verify this claim.
Other comparisons
A large systematic review of 100 RCTs, including 25,952 participants, reported on metabolic side effects of antipsychotics. Olanzapine and clozapine were found to have the greatest detrimental impact on metabolism, while profiles of aripiprazole, brexpiprazole, cariprazine, lurasidone, and ziprasidone were more benign (Pillinger 2020). Our narrative synthesis of studies suggested more favourable results for blood glucose level and lipid metabolism (BMI, cholesterol, waist circumference) from ziprasidone than olanzapine or other antipsychotics in individual studies.
A systematic review of studies evaluating interventions to improve glycaemic control in adults with severe mental illness found that non‐pharmacological interventions, including behaviour change interventions, were effective in lowering blood glucose (Taylor 2017). A systematic review of four US studies concluded that diabetes education which incorporates elements of diet and exercise could reduce fasting blood glucose (Cimo 2012). We found only one study, which included a behavioural intervention in people with mental illness in an LMIC setting.
Study flow diagram.
Risk of bias graph: review authors' judgements about each risk of bias item presented as percentages across all included studies.
Risk of bias summary: review authors' judgements about each risk of bias item for each included study.
Comparison 1: Atypical versus typical antipsychotic, Outcome 1: diabetes (ADA criteria) (6 weeks)
Comparison 1: Atypical versus typical antipsychotic, Outcome 2: drop‐outs (6‐54 weeks)
Comparison 1: Atypical versus typical antipsychotic, Outcome 3: fasting blood glucose (6‐8 weeks)
Comparison 1: Atypical versus typical antipsychotic, Outcome 4: BMI (8‐54 weeks)
Comparison 1: Atypical versus typical antipsychotic, Outcome 5: total cholesterol (8 weeks)
Comparison 2: Metformin versus placebo, Outcome 1: drop‐outs (12‐14 weeks)
Comparison 2: Metformin versus placebo, Outcome 2: fasting blood glucose (12‐14 weeks)
Comparison 2: Metformin versus placebo, Outcome 3: BMI (12‐14 weeks)
Comparison 2: Metformin versus placebo, Outcome 4: waist circumference (12‐14 weeks)
Comparison 2: Metformin versus placebo, Outcome 5: systolic blood pressure (14 weeks)
Comparison 2: Metformin versus placebo, Outcome 6: diastolic blood pressure (14 weeks)
Comparison 2: Metformin versus placebo, Outcome 7: total cholesterol (12‐14 weeks)
Comparison 3: Melatonin versus placebo, Outcome 1: drop‐outs (8 weeks)
Comparison 3: Melatonin versus placebo, Outcome 2: fasting blood glucose (8‐12 weeks)
Comparison 3: Melatonin versus placebo, Outcome 3: BMI (8‐12 weeks)
Comparison 3: Melatonin versus placebo, Outcome 4: Waist circumference (8 weeks)
Comparison 3: Melatonin versus placebo, Outcome 5: systolic blood pressure (8 weeks)
Comparison 3: Melatonin versus placebo, Outcome 6: diastolic blood pressure (8 weeks)
Comparison 3: Melatonin versus placebo, Outcome 7: total cholesterol (8 weeks)
Comparison 4: SSRI versus TCA, Outcome 1: drop‐outs (12 weeks)
Comparison 4: SSRI versus TCA, Outcome 2: fasting blood glucose (8‐12 weeks)
Comparison 4: SSRI versus TCA, Outcome 3: BMI (12 weeks)
Comparison 4: SSRI versus TCA, Outcome 4: depression (12 weeks)
Comparison 5: SENSITIVITY ‐ high quality ‐ metformin vs placebo, Outcome 1: fasting blood glucose (12‐14 weeks)
Comparison 5: SENSITIVITY ‐ high quality ‐ metformin vs placebo, Outcome 2: BMI (12‐14 weeks)
Comparison 6: SENSITIVITY ‐ Wu2008a ‐ metformin vs placebo, Outcome 1: drop‐outs (12‐14 weeks)
Comparison 6: SENSITIVITY ‐ Wu2008a ‐ metformin vs placebo, Outcome 2: fasting blood glucose (12‐14 weeks)
Comparison 6: SENSITIVITY ‐ Wu2008a ‐ metformin vs placebo, Outcome 3: BMI (12‐14 weeks)
Comparison 6: SENSITIVITY ‐ Wu2008a ‐ metformin vs placebo, Outcome 4: waist circumference (12‐14 weeks)
Comparison 7: SENSITIVITY‐ Wu2006‐ Atypical versus typical antipsychotic, Outcome 1: fasting blood glucose (6‐8 weeks)
Comparison 7: SENSITIVITY‐ Wu2006‐ Atypical versus typical antipsychotic, Outcome 2: BMI (52 weeks)
| Atypical compared with typical antipsychotic medications for preventing type 2 diabetes in adults with mental disorders in low‐ and middle‐income countries | ||||||
| Patient or population: Adults with mental disorders in low‐ and middle‐income countries | ||||||
| Outcomes | Anticipated absolute effects* (95% CI) | Relative effect | № of participants | Certainty of the evidence | Comments | |
|---|---|---|---|---|---|---|
| Risk with typical antipsychotic | Risk with atypical antipsychotic | |||||
| Diabetes (ADA criteria) (6 weeks) | Study population | RR 0.50 | 93 | ⊕⊕⊝⊝ | ||
| 32 per 1,000 | 16 per 1,000 | |||||
| Drop‐outs (6 to 54 weeks) | Study population | RR 1.31 | 144 | ⊕⊕⊝⊝ | ||
| 148 per 1,000 | 194 per 1,000 | |||||
| Fasting blood glucose (6 to 8 weeks) | Mean fasting blood glucose was 4.90 to 4.91 mmol/L (normal level) | MD 0.05 lower (0.10 lower to 0.00 lower) | ‐ | 211 | ⊕⊕⊕⊝ | |
| BMI (8 to 54 weeks) | Mean BMI was 21.2 to 24.6 kg/m2 (healthy weight range) | MD 0.57 higher | ‐ | 141 | ⊕⊕⊕⊝ | |
| *The risk in the intervention group (and its 95% confidence interval) is based on the assumed risk in the comparison group and the relative effect of the intervention (and its 95% CI). | ||||||
| GRADE Working Group grades of evidence | ||||||
| 1One small trial; estimate with wide confidence interval crossing 1. Downgraded two levels for imprecision. | ||||||
| Metformin compared with placebo for preventing type 2 diabetes in adults with mental disorders in low‐ and middle‐income countries | ||||||
| Patient or population: Adults with mental disorders in low‐ and middle‐income countries | ||||||
| Outcomes | Anticipated absolute effects* (95% CI) | Relative effect | № of participants | Certainty of the evidence | Comments | |
|---|---|---|---|---|---|---|
| Risk with placebo | Risk with metformin | |||||
| Diabetes | No studies identified | |||||
| Drop‐outs (12 to 14 weeks) | Study population | RR 1.22 (0.09 to 16.35) | 158 | ⊕⊕⊕⊝ | ||
| 49 per 1,000 | 11 more per 1,000 | |||||
| Fasting blood glucose (12 to 14 weeks) | Mean fasting blood glucose was 4.40 to 4.71 mmol/L | MD 0.35 lower (0.60 lower to 0.11 lower) | ‐ | 173 | ⊕⊕⊕⊝ | |
| Fasting blood glucose (12 to 14 weeks) | ‐ | MD 0.01 higher (0.21 lower to 0.22 higher) | 91 (2 RCTs) | ⊕⊕⊕⊕ | ||
| BMI (12 to 14 weeks) | Mean BMI was 23.5 to 25.3 (normal to slightly overweight) | MD 1.37 lower | ‐ | 264 | ⊕⊕⊕⊕ | |
| *The risk in the intervention group (and its 95% confidence interval) is based on the assumed risk in the comparison group and the relative effect of the intervention (and its 95% CI). | ||||||
| GRADE Working Group grades of evidence | ||||||
| 1Some unclear risk of bias domains and unbalanced drop‐out in one study, but this is not a concern for this outcome. | ||||||
| Melatonin compared with placebo for preventing type 2 diabetes in adults with mental disorders in low‐ and middle‐income countries | ||||||
| Patient or population: Adults with mental disorders in low‐ and middle‐income countries | ||||||
| Outcomes | Anticipated absolute effects* (95% CI) | Relative effect | № of participants | Certainty of the evidence | Comments | |
|---|---|---|---|---|---|---|
| Risk with placebo | Risk with melatonin | |||||
| Diabetes | No studies identified | |||||
| Drop‐outs (8 weeks) | Study population | RR 1.00 | 48 | ⊕⊕⊝⊝ | ||
| 250 per 1,000 | 250 per 1,000 | |||||
| Fasting blood glucose (8 to 12 weeks) (endpoint data) | Mean fasting blood glucose was 4.9 to 5.0 mmol/L (normal level) | MD 0.17 lower (0.35 lower to 0.01 higher) | ‐ | 102 | ⊕⊕⊕⊝ | |
| Fasting blood glucose (8 to 12 weeks) (change from baseline data) | ‐ | MD 0.24 lower (0.39 lower to 0.09 lower) | ‐ | 100 (1 RCT) | ⊕⊕⊕⊝ | |
| BMI (8 to 12 weeks) | Mean BMI was 25.1 to 25.2 (slightly overweight) | MD 0.22 lower | ‐ | 202 | ⊕⊝⊝⊝ | |
| *The risk in the intervention group (and its 95% confidence interval) is based on the assumed risk in the comparison group and the relative effect of the intervention (and its 95% CI). | ||||||
| GRADE Working Group grades of evidence | ||||||
| 1One small trial with an equal number of events in both groups. Downgraded two levels for very serious imprecision; estimate includes benefits and harms. | ||||||
| SSRI antidepressants compared with TCA for preventing type 2 diabetes in adults with mental disorders in low‐ and middle‐income countries | ||||||
| Patient or population: Adults with mental disorders in low‐ and middle‐income countries | ||||||
| Outcomes | Anticipated absolute effects* (95% CI) | Relative effect | № of participants | Certainty of the evidence | Comments | |
|---|---|---|---|---|---|---|
| Risk with TCA | Risk with SSRI | |||||
| Diabetes | No studies identified. | |||||
| Drop‐outs (12 weeks) | Study population | RR 0.34 | 25 | ⊕⊝⊝⊝ | ||
| 636 per 1,000 | 216 per 1,000 | |||||
| Fasting blood glucose (8 to 12 weeks) | Mean fasting blood glucose was 4.4 to 5.1 mmol/L (normal level) | MD 0.39 lower (0.88 lower to 0.10 higher) | ‐ | 141 | ⊕⊕⊕⊝ | |
| BMI (12 weeks) | Mean BMI was 25.2 (slightly overweight) | MD 0.7 higher | ‐ | 18 | ⊕⊝⊝⊝ | |
| *The risk in the intervention group (and its 95% confidence interval) is based on the assumed risk in the comparison group and the relative effect of the intervention (and its 95% CI). | ||||||
| GRADE Working Group grades of evidence | ||||||
| 1One small trial; unclear allocation concealment, no blinding, incomplete outcome data. Downgraded two levels for risk of bias. | ||||||
| Reference | Comparison | Follow‐up | Outcome | mean (SD) group 1 | number of drop‐outs group 1 | n/N group1 | mean (SD) group 2 | number of drop‐outs group 2 | n/N group2 |
| Chen 2017 | ziprasidone (n = 19) versus olanzapine (n = 19) | 12 weeks | fasting blood glucose (mmol/L) | 5.59 (1.02) | 19/38 | 7.28 (2.22) | 19/38 | ||
| cholesterol (mmol/L) | 1.28 (0.2) | 2.24 (0.31) | |||||||
| Ou 2013 | ziprasidone (n = 130) versus olanzapine (n = 130) | 6 weeks | drop‐outs | 11 | 130/260 | 19 | 130/260 | ||
| fasting blood glucose (mmol/L) | 4.40 (0.50) | 4.94 (0.50) | |||||||
| cholesterol (mmol/L) | 4.06 (0.74) | 4.51 (0.80) | |||||||
| BMI (kg/m2) | 20.87 (3.34) | 22.28 (2.93) | |||||||
| systolic blood pressure (mm Hg) | 117.43 (13.13) | 117.97 (10.50) | |||||||
| diastolic blood pressure (mm Hg) | 75.93 (7.90) | 76.30 (5.65) | |||||||
| Hu 2013 | paliperidone (n = 33) versus olanzapine (n = 23) | 12 weeks | drop‐outs | 7 | 33/56 | 17 | 23/56 | ||
| fasting blood glucose (mmol/L) | 5.21 (0.6) | 5.19 (0.6) | |||||||
| BMI (kg/m2) | 22.17 (3.31) | 23.17 (4.06) | |||||||
| waist circumference (cm) | 80.3 (12.47) | 82.63 (9.49) | |||||||
| Zhang 2014 | aripiprazole (n = 50) versus olanzapine (n = 50) | 8 weeks | drop‐outs | 5 | 50/100 | 5 | 50/100 | ||
| fasting blood glucose (mmol/L) | 5.49 (1.51) | 5.49 (1.51) | |||||||
| cholesterol (mmol/L) | 4.11 (1.04) | 4.72 (1.12) | |||||||
| Zhang 2012 | aripriprazole (n = 71) versus ziprasidone (n = 69) | 52 weeks | drop‐outs | 19 | 71/140 | 14 | 69/140 | ||
| fasting blood glucose (mmol/L) | 5.2 | 4.6 | |||||||
| cholesterol (mmol/L) | 5.1 | 4.7 | |||||||
| BMI (kg/m2) | 24.5 (5.9 | 20.3 (5.2) | |||||||
| waist circumference (cm) | 71.6 (17.6) | 70.3 (16.7) | |||||||
| SD: standard deviation; n: number of participants; N: total number of participants; BMI: Body mass index. | |||||||||
| Reference | Comparison | Follow‐up | Outcome | mean (SD) group 1 | number of drop‐outs group 1 | n/N group1 | mean (SD) group 2 | number of drop‐outs group 2 | n/N group 2 | mean (SD) group 3) | number of drop‐outs group 3 | n/N group 3 |
| Agnihotri 2013 | Withania somnifera (n = 12) versus placebo (n = 13) | 4 weeks | fasting blood glucose (mmol/L) | 5.14 ( 0.33) | 12/25 | 5.82 ( 0.46) | 13/25 | |||||
| Akkasheh 2016 | probiotic supplements (n = 20) versus placebo (n = 20) | 8 weeks | drop‐outs | 3 | 20/40 | 2 | 20/40 | |||||
| fasting blood glucose (mmol/L) | 5.54 (0.97) | 4.96 (0.42) | ||||||||||
| cholesterol (mmol/L) | 9.58 (1.88) | 9.98 (1.72) | ||||||||||
| BMI (kg/m2) | 27.5 (5.9) | 26.5 (3.9) | ||||||||||
| Assunção 2006 | nizatidine (n = 27) versus placebo (n = 27) | 12 weeks | fasting blood glucose (mmol/L) | 4.84 (0.81) | 27/54 | 4.70 (0.98) | 27/54 | |||||
| cholesterol (mmol/L) | 11.04 (2.56) | 10.02 (1.89) | ||||||||||
| Baptista 2009 | rosiglitazone (n = 14) versus placebo (n = 15) | 12 weeks | fasting blood glucose (mmol/L) | 4.48 (0.77) | 14/29 | 4.38 (0.48) | 15/29 | |||||
| cholesterol (mmol/L) | 9.99 (1.26) | 11.18 (1.45) | ||||||||||
| BMI (kg/m2) | 26.9 (4.2) | 26.3 (2.9) | ||||||||||
| waist circumference (cm) | 91.5 (11.6) | 90.5 (8.6) | ||||||||||
| Fadai 2014 | saffron aqueous extract (n = 20) versus crocin (n = 20) | 12 weeks | drop‐outs | 2 | 20/61 | 2 | 20/61 | 1 | 21/61 | |||
| fasting blood glucose (mmol/L) | 5.58 (0.39) | 5.44 (0.38) | 6.03 (0.56) | |||||||||
| cholesterol (mmol/L) | 9.36 (1.35) | 9.94 (1.91) | 10.9 (1.91) | |||||||||
| waist circumference (cm) | 92.1 (7.4) | 91.9 (8.6) | 98.4 (8.4) | |||||||||
| blood pressure (mm Hg) | 116.5 (5.6) | 109.7 (6.8) | 115.9 (9) | |||||||||
| Ghaderi 2019 | vitamin D and probiotic supplements (n = 30) versus placebo (n = 30) | 12 weeks | drop‐outs | 4 | 30/60 | 4 | 30/60 | |||||
| fasting blood glucose (mmol/L) | 4.89 (0.62) | 5.17 (0.48) | ||||||||||
| cholesterol (mmol/L) | 8.99 (2.04) | 9.92 (1.99) | ||||||||||
| BMI (kg/m2) | 23.2 (2.7) | 24.5 (3.7) | ||||||||||
| Narula 2010 | topiramate (n = 33) versus placebo (n = 34) | 12 weeks | fasting blood glucose (mmol/L) | 4.35 (0.37) | 33/67 | 4.92 (0.67) | 34/67 | |||||
| BMI (kg/m2) | 20.1 (4) | 22.55 (4.11) | ||||||||||
| diastolic blood pressure (mm Hg) | 77.94 (4.8) | 81.41 (6.2) | ||||||||||
| systolic blood pressure (mm Hg) | 177.88 (7) | 122.5 (7.71) | ||||||||||
| Tohen 2011 | olanzapine (n = 140) versus placebo (n = 70) | 6 weeks | drop‐outs | 35 | 140/210 | 35/70 | 35 | 70/210 | ||||
| fasting blood glucose (mmol/L) | 5.05 (0.53) | 5.03 (0.61) | ||||||||||
| cholesterol (mmol/L) | 4.47 (0.95) | 4.56 (0.94) | ||||||||||
| diastolic blood pressure (mm Hg) | 72.72 (8.09) | 72.17 (7.36) | ||||||||||
| systolic blood pressure (mm Hg) | 109.84 (11.41) | 109.76 (11.59) | ||||||||||
| Zhao 2015 | aripiprazole (n = 56) versus placebo (n = 57) | 6 weeks | drop‐outs | 2 | 56/113 | 4 | 57/113 | |||||
| fasting blood glucose (mmol/L) | 4.75 (0.8) | 4.95 (0.87) | ||||||||||
| cholesterol (mmol/L) | 3.32 (0.92) | 4.54 (1.11) | ||||||||||
| BMI (kg/m2) | 23.27 (2.63) | 24.8 (3.92) | ||||||||||
| Sepehrmanesh 2016 | Vitamin D (n = 20) versus placebo (n = 20) | 8 weeks | drop‐outs | 2 | 20/40 | 2 | 20/40 | |||||
| fasting blood glucose(mmol/L) | 4.69 (0.48) | 5.02 (0.56) | ||||||||||
| cholesterol (mmol/L) | 4.69 (0.48) | 5.02 (0.56) | ||||||||||
| BMI (kg/m2) | 26.0 (5.1) | 27.3 (3.5) | ||||||||||
| Depression score, Beak Depression Inventory (BDI) | 17.2 (10.6) | 25.2 (9.9) | ||||||||||
| SD: standard deviation; n: number of participants; N: total number of participants; BMI: Body mass index. | ||||||||||||
| Outcome or subgroup title | No. of studies | No. of participants | Statistical method | Effect size |
| 1.1 diabetes (ADA criteria) (6 weeks) Show forest plot | 1 | 99 | Risk Ratio (M‐H, Random, 95% CI) | 0.46 [0.03, 7.05] |
| 1.2 drop‐outs (6‐54 weeks) Show forest plot | 2 | 144 | Risk Ratio (IV, Random, 95% CI) | 1.31 [0.63, 2.69] |
| 1.3 fasting blood glucose (6‐8 weeks) Show forest plot | 2 | 211 | Mean Difference (IV, Random, 95% CI) | ‐0.05 [‐0.10, ‐0.00] |
| 1.4 BMI (8‐54 weeks) Show forest plot | 2 | 141 | Mean Difference (IV, Random, 95% CI) | 0.57 [0.33, 0.81] |
| 1.5 total cholesterol (8 weeks) Show forest plot | 1 | 112 | Mean Difference (IV, Random, 95% CI) | 0.35 [0.27, 0.43] |
| Outcome or subgroup title | No. of studies | No. of participants | Statistical method | Effect size |
| 2.1 drop‐outs (12‐14 weeks) Show forest plot | 3 | 158 | Risk Ratio (IV, Random, 95% CI) | 1.22 [0.09, 16.35] |
| 2.2 fasting blood glucose (12‐14 weeks) Show forest plot | 5 | 264 | Mean Difference (IV, Random, 95% CI) | ‐0.19 [‐0.46, 0.08] |
| 2.2.1 Endpoint | 3 | 173 | Mean Difference (IV, Random, 95% CI) | ‐0.35 [‐0.60, ‐0.11] |
| 2.2.2 Change from baseline | 2 | 91 | Mean Difference (IV, Random, 95% CI) | 0.01 [‐0.21, 0.22] |
| 2.3 BMI (12‐14 weeks) Show forest plot | 5 | 264 | Mean Difference (IV, Random, 95% CI) | ‐1.37 [‐2.04, ‐0.70] |
| 2.3.1 Endpoint | 3 | 173 | Mean Difference (IV, Random, 95% CI) | ‐1.75 [‐2.38, ‐1.12] |
| 2.3.2 Change from baseline | 2 | 91 | Mean Difference (IV, Random, 95% CI) | ‐1.20 [‐2.17, ‐0.23] |
| 2.4 waist circumference (12‐14 weeks) Show forest plot | 4 | 192 | Mean Difference (IV, Random, 95% CI) | ‐1.44 [‐2.93, 0.04] |
| 2.4.1 Endpoint | 2 | 101 | Mean Difference (IV, Random, 95% CI) | ‐0.30 [‐6.26, 5.66] |
| 2.4.2 Change from baseline | 2 | 91 | Mean Difference (IV, Random, 95% CI) | ‐0.93 [‐1.21, ‐0.64] |
| 2.5 systolic blood pressure (14 weeks) Show forest plot | 1 | 54 | Mean Difference (IV, Random, 95% CI) | ‐2.50 [‐9.09, 4.09] |
| 2.6 diastolic blood pressure (14 weeks) Show forest plot | 1 | 54 | Mean Difference (IV, Random, 95% CI) | 1.20 [‐3.55, 5.95] |
| 2.7 total cholesterol (12‐14 weeks) Show forest plot | 2 | 109 | Mean Difference (IV, Random, 95% CI) | ‐0.34 [‐0.93, 0.26] |
| Outcome or subgroup title | No. of studies | No. of participants | Statistical method | Effect size |
| 3.1 drop‐outs (8 weeks) Show forest plot | 1 | 48 | Risk Ratio (M‐H, Random, 95% CI) | 1.00 [0.38, 2.66] |
| 3.2 fasting blood glucose (8‐12 weeks) Show forest plot | 3 | 202 | Mean Difference (IV, Random, 95% CI) | ‐0.21 [‐0.32, ‐0.10] |
| 3.2.1 Endpoint | 2 | 102 | Mean Difference (IV, Random, 95% CI) | ‐0.17 [‐0.35, 0.01] |
| 3.2.2 Change from baseline | 1 | 100 | Mean Difference (IV, Random, 95% CI) | ‐0.24 [‐0.39, ‐0.09] |
| 3.3 BMI (8‐12 weeks) Show forest plot | 3 | 202 | Mean Difference (IV, Random, 95% CI) | ‐0.22 [‐2.58, 2.14] |
| 3.3.1 Endpoint | 2 | 102 | Mean Difference (IV, Random, 95% CI) | ‐1.52 [‐2.40, ‐0.64] |
| 3.3.2 Change from baseline | 1 | 100 | Mean Difference (IV, Random, 95% CI) | 1.47 [0.45, 2.49] |
| 3.4 Waist circumference (8 weeks) Show forest plot | 2 | 136 | Mean Difference (IV, Random, 95% CI) | 0.68 [‐0.47, 1.83] |
| 3.4.1 Endpoint | 1 | 36 | Mean Difference (IV, Random, 95% CI) | 0.00 [‐1.26, 1.26] |
| 3.4.2 Change from baseline | 1 | 100 | Mean Difference (IV, Random, 95% CI) | 1.19 [0.29, 2.09] |
| 3.5 systolic blood pressure (8 weeks) Show forest plot | 2 | 136 | Mean Difference (IV, Random, 95% CI) | ‐1.31 [‐6.46, 3.84] |
| 3.5.1 Endpoint | 1 | 36 | Mean Difference (IV, Random, 95% CI) | 1.00 [0.07, 1.93] |
| 3.5.2 Change from baseline | 1 | 100 | Mean Difference (IV, Random, 95% CI) | ‐4.30 [‐8.14, ‐0.46] |
| 3.6 diastolic blood pressure (8 weeks) Show forest plot | 2 | 136 | Mean Difference (IV, Random, 95% CI) | ‐1.05 [‐1.60, ‐0.50] |
| 3.6.1 Endpoint | 1 | 36 | Mean Difference (IV, Random, 95% CI) | ‐1.00 [‐1.56, ‐0.44] |
| 3.6.2 Change from baseline | 1 | 100 | Mean Difference (IV, Random, 95% CI) | ‐2.40 [‐5.33, 0.53] |
| 3.7 total cholesterol (8 weeks) Show forest plot | 2 | 136 | Mean Difference (IV, Random, 95% CI) | ‐0.06 [‐0.19, 0.06] |
| 3.7.1 Endpoint | 1 | 36 | Mean Difference (IV, Random, 95% CI) | ‐0.11 [‐0.27, 0.05] |
| 3.7.2 Change from baseline | 1 | 100 | Mean Difference (IV, Random, 95% CI) | 0.02 [‐0.19, 0.23] |
| Outcome or subgroup title | No. of studies | No. of participants | Statistical method | Effect size |
| 4.1 drop‐outs (12 weeks) Show forest plot | 1 | 25 | Risk Ratio (M‐H, Random, 95% CI) | 0.34 [0.11, 1.01] |
| 4.2 fasting blood glucose (8‐12 weeks) Show forest plot | 3 | 141 | Mean Difference (IV, Random, 95% CI) | ‐0.39 [‐0.88, 0.10] |
| 4.3 BMI (12 weeks) Show forest plot | 1 | 18 | Mean Difference (IV, Random, 95% CI) | 0.70 [‐1.10, 2.50] |
| 4.4 depression (12 weeks) Show forest plot | 1 | 18 | Mean Difference (IV, Random, 95% CI) | 0.30 [‐0.59, 1.19] |
| Outcome or subgroup title | No. of studies | No. of participants | Statistical method | Effect size |
| 5.1 fasting blood glucose (12‐14 weeks) Show forest plot | 2 | 136 | Mean Difference (IV, Random, 95% CI) | ‐0.37 [‐0.68, ‐0.07] |
| 5.2 BMI (12‐14 weeks) Show forest plot | 4 | 227 | Mean Difference (IV, Random, 95% CI) | ‐1.46 [‐2.15, ‐0.77] |
| 5.2.1 Change from baseline | 2 | 91 | Mean Difference (IV, Random, 95% CI) | ‐1.20 [‐2.17, ‐0.23] |
| 5.2.2 Endpoint | 2 | 136 | Mean Difference (IV, Random, 95% CI) | ‐1.89 [‐2.29, ‐1.49] |
| Outcome or subgroup title | No. of studies | No. of participants | Statistical method | Effect size |
| 6.1 drop‐outs (12‐14 weeks) Show forest plot | 2 | 94 | Risk Ratio (IV, Random, 95% CI) | 2.78 [0.08, 95.87] |
| 6.2 fasting blood glucose (12‐14 weeks) Show forest plot | 2 | 109 | Mean Difference (IV, Random, 95% CI) | ‐0.19 [‐0.46, 0.09] |
| 6.2.1 Endpoint | 2 | 109 | Mean Difference (IV, Random, 95% CI) | ‐0.19 [‐0.46, 0.09] |
| 6.3 BMI (12‐14 weeks) Show forest plot | 4 | 200 | Mean Difference (IV, Random, 95% CI) | ‐1.13 [‐1.86, ‐0.40] |
| 6.3.1 Endpoint | 2 | 109 | Mean Difference (IV, Random, 95% CI) | ‐0.83 [‐2.48, 0.82] |
| 6.3.2 Change from baseline | 2 | 91 | Mean Difference (IV, Random, 95% CI) | ‐1.20 [‐2.17, ‐0.23] |
| 6.4 waist circumference (12‐14 weeks) Show forest plot | 1 | 37 | Mean Difference (IV, Random, 95% CI) | 3.40 [‐1.99, 8.79] |
| 6.4.1 Endpoint | 1 | 37 | Mean Difference (IV, Random, 95% CI) | 3.40 [‐1.99, 8.79] |
| Outcome or subgroup title | No. of studies | No. of participants | Statistical method | Effect size |
| 7.1 fasting blood glucose (6‐8 weeks) Show forest plot | 1 | 99 | Mean Difference (IV, Random, 95% CI) | 0.01 [‐0.35, 0.37] |
| 7.2 BMI (52 weeks) Show forest plot | 1 | 29 | Mean Difference (IV, Random, 95% CI) | ‐1.13 [‐5.65, 3.39] |
