Monitorización automatizada en comparación con la atención habitual para la detección temprana de la sepsis en pacientes graves
Resumen
Antecedentes
La sepsis es una afección potencialmente mortal que, por lo general, se diagnostica cuando el paciente tiene una infección presunta o documentada, y cumple dos o más criterios del síndrome de respuesta inflamatoria sistémica (SRIS). La incidencia de la sepsis es mayor entre los pacientes en ámbitos de cuidados intensivos, como la unidad de cuidados intensivos (UCI), que entre los pacientes en otros ámbitos. Si no se trata la sepsis puede empeorar rápidamente; la sepsis grave tiene una tasa de mortalidad del 40% o mayor, según la definición. El reconocimiento de la sepsis puede ser desafiante, ya que por lo general requiere combinar los datos del paciente obtenidos a partir de múltiples fuentes inconexas, e interpretarlos correctamente, lo que puede resultar complejo y llevar mucho tiempo. Los sistemas electrónicos diseñados para conectar las fuentes de información y compilarlas automáticamente, analizan y monitorizan de manera continua la información, además de alertar al personal de asistencia sanitaria cuándo se cumplen umbrales diagnósticos predeterminados, y pueden ser beneficiosos al facilitar el reconocimiento más temprano de la sepsis y el inicio más rápido del tratamiento, como el tratamiento antimicrobiano, la reanimación con líquidos y el uso de fármacos inotrópicos y vasopresores, de ser apropiado. Sin embargo, existe la posibilidad de que los sistemas electrónicos, automatizados no ofrezcan beneficios o incluso que tengan efectos perjudiciales. Lo anterior podría suceder si los sistemas no pueden detectar de modo correcto la sepsis (lo que provoca que el tratamiento no se comience cuando debería o se comience cuando no debería) o si el personal de asistencia sanitaria no puede responder a las alertas suficientemente rápido, o si provocan "fatiga por alarma", sobre todo si las alarmas suenan con frecuencia o dan demasiadas falsas alarmas.
Objetivos
Evaluar si los sistemas automatizados para la detección temprana de la sepsis pueden reducir el tiempo transcurrido hasta la administración del tratamiento (como el inicio de la administración de antibióticos, líquidos, fármacos inotrópicos y vasopresores) y mejorar los resultados clínicos en los pacientes en estado crítico en la UCI.
Métodos de búsqueda
Se hicieron búsquedas en: CENTRAL; MEDLINE; Embase; CINAHL; ISI Web of science; en LILACS, clinicaltrials.gov, y en el World Health Organization trials portal. Se realizaron búsquedas en todas las bases de datos desde su fecha de inicio hasta el 18 de septiembre de 2017, sin restricciones de país o idioma de publicación.
Criterios de selección
Se incluyeron los ensayos controlados aleatorios (ECA) que compararon los sistemas automatizados de monitorización de la sepsis con la atención habitual (como los sistemas basados en papeles) en participantes de cualquier edad ingresados en una unidad de cuidados intensivos o para pacientes en estado crítico por enfermedad grave. Un sistema automatizado se definió como cualquier proceso capaz de revisar los registros o los datos (uno o más sistemas) de los pacientes de forma automática a intervalos, para detectar marcadores o características que indican sepsis. La definición de enfermedad grave incluyó, pero no se limitó a, estado posquirúrgico, traumatismo, accidente cerebrovascular, infarto de miocardio, arritmia, quemaduras y shock hipovolémico o hemorrágico. Se excluyeron los estudios no aleatorios, cuasialeatorios y cruzados. También se excluyeron los estudios que incluyeron pacientes que ya tenían un diagnóstico de sepsis.
Obtención y análisis de los datos
Se utilizaron los procedimientos metodológicos estándar previstos por la Colaboración Cochrane. Los resultados primarios fueron: tiempo transcurrido hasta el inicio del tratamiento antimicrobiano; tiempo transcurrido hasta el inicio de la reanimación con líquidos; y la mortalidad a los 30 días. Los resultados secundarios incluyeron: duración de la estancia en la UCI; fracaso en la detección de la sepsis; y calidad de vida. Se utilizó GRADE para evaluar la calidad de la evidencia para cada resultado.
Resultados principales
Se incluyeron tres ECA en esta revisión. No estuvo claro si los ECA fueron tres estudios separados que incluyeron a 1199 participantes en total, o si fueron informes del mismo estudio que incluyeron menos participantes. Se decidió tratar los estudios por separado, ya que no fue posible establecer contacto con los autores de los estudios para aclarar este aspecto.
Los tres ECA son de muy baja calidad debido a problemas relacionados con los métodos inciertos de asignación al azar, la ocultación de la asignación y la incertidumbre con el tamaño del efecto. Algunos de los estudios se informaron como resúmenes solamente y contenían datos limitados, lo que impide realizar un análisis significativo y la evaluación de los sesgos potenciales.
Los estudios incluyeron a participantes que recibieron monitorización electrónica automatizada durante la estancia hospitalaria. Los participantes se asignaron al azar a un grupo de intervención (alertas automatizadas enviadas por el sistema) o a atención habitual (sin alertas automatizadas enviadas por del sistema).
Hubo evidencia de los tres estudios con respecto al "Tiempo transcurrido hasta el inicio del tratamiento antimicrobiano". No fue posible agrupar los datos, pero el estudio más grande que incluyó 680 participantes informó que el tiempo mediano transcurrido hasta el inicio del tratamiento antimicrobiano en el grupo de intervención fue 5,6 horas (rango intercuartil [RIC]: 2,3 a 19,7) en el grupo de intervención (n = no señalado) y 7,8 horas (RIC: 2,5 a 33,1) en el grupo control (n = no señalado).
Ningún estudio informó el "Tiempo transcurrido hasta el inicio de la reanimación con líquidos" ni el evento adverso "Mortalidad a los 30 días". Sin embargo, se obtuvo evidencia de muy baja calidad de la mortalidad informada en otros puntos temporales. Un estudio que incluyó a 77 participantes informó una mortalidad a los 14 días del 20% en el grupo de intervención y del 21% en el grupo control (no se señalaron el numerador ni el denominador). Un estudio que incluyó a 442 participantes informó que la mortalidad a los 28 días, o al alta fue del 14% en el grupo de intervención y del 10% en el grupo control (no se señalaron el numerador ni el denominador). Los tamaños de la muestra no se informaron de manera adecuada para estos resultados, por lo que no fue posible calcular los intervalos de confianza.
Un estudio que incluyó a 442 participantes proporcionó evidencia de muy baja calidad de la "Duración de la estancia hospitalaria en la UCI". La estancia hospitalaria mediana fue 3,0 días en el grupo de intervención (RIC: 2,0 a 5,0) y 3,0 días (RIC: 2,0 a 4,0 en el grupo control).
Un estudio que incluyó al menos a 442 participantes proporcionó evidencia de muy baja calidad del efecto adverso "Fracaso en la detección de la sepsis". Solo se informaron datos del fracaso en la detección de la sepsis en dos participantes y no estuvo claro en qué grupo/s ocurrió este resultado.
Ningún estudio informó la "calidad de vida".
Conclusiones de los autores
No está claro el efecto de los sistemas automatizados en la monitorización de la sepsis sobre cualquiera de los resultados incluidos en esta revisión. Solo hay disponible evidencia de muy baja calidad de las alertas automatizadas, que es un solo componente de los sistemas de monitorización automatizada. No está claro si estos sistemas pueden reemplazar la revisión regular y cuidadosa de la condición del paciente por personal experimentado de asistencia sanitaria.
PICO
Resumen en términos sencillos
Monitorización automatizada para la detección temprana de la sepsis en los pacientes que reciben atención en unidades de cuidados intensivos
Pregunta de la revisión
¿Los sistemas automatizados para la detección temprana de la sepsis pueden reducir el tiempo hasta el tratamiento y mejorar los resultados en los pacientes que se encuentran en una unidad de cuidados intensivos (UCI), en comparación con la atención habitual?
Antecedentes
La sepsis sucede cuando una persona contrae una infección y su sistema inmunológico reacciona de forma exagerada a esta infección. Si no se controla, la sepsis puede evolucionar rápidamente a shock séptico, que provoca que órganos como el hígado y el corazón dejen de trabajar de forma adecuada. Las personas pueden resultar afectadas por la sepsis en cualquier momento, pero los pacientes en ámbitos de cuidados intensivos tienen mayores probabilidades de resultar afectadas. El shock séptico es mortal en el 20% al 70% de los pacientes ingresados en cuidados intensivos en Europa. No hay pruebas diagnósticas que puedan decir si algún paciente presenta sepsis o no. En cambio, se tienen que examinar los resultados de varias pruebas (como los análisis de sangre) en conjunto con otras informaciones acerca del paciente (como la historia clínica) y las observaciones clínicas (como la frecuencia cardíaca, la temperatura y la presión arterial). Este proceso puede ser lento y complicado de realizar. Es probable que los pacientes que están ingresados en cuidados intensivos estén muy enfermos, y puede ser difícil determinar si los resultados anormales se deben al problema que provocó su ingreso en cuidados intensivos o debido a la sepsis.
Los sistemas de monitorización automatizada son sistemas electrónicos que pueden recopilar y analizar información proveniente de diferentes fuentes y se pueden utilizar para alertar al personal cuando se identifican signos y síntomas de sepsis. Este hecho puede significar que la sepsis se diagnostica en el momento más temprano posible, lo que permite que se comience el tratamiento antes de que ocurra daño a los órganos. Sin embargo, existe la posibilidad de que los sistemas de monitorización automatizada no ayuden, o incluso provoquen efectos perjudiciales. Lo anterior podría suceder si los sistemas no pueden detectar de modo correcto la sepsis (lo que significa que el tratamiento no comienza cuando debería o que comienza cuando no debería), o si el personal de asistencia sanitaria no puede responder a las alertas suficientemente rápido, en especial si los sistemas dan demasiadas falsas alarmas.
Características de los estudios
Se realizó una búsqueda para identificar la evidencia publicada antes de septiembre de 2017. Los estudios fueron elegibles para inclusión si compararon la monitorización automatizada de la sepsis con la atención habitual (como sistemas basados en papeles) en pacientes ingresados en una unidad de cuidados intensivos o para pacientes en estado crítico por enfermedad muy grave. No se incluyeron estudios no aleatorios (estudios donde los participantes no se asignan a los grupos de tratamiento por azar), estudios cuasialeatorios (estudios donde los participantes se asignan a los grupos de tratamiento mediante un método que no es verdaderamente al azar, como fecha de nacimiento o de historia clínica) ni estudios cruzados (donde los participantes primero reciben un tratamiento y luego cambian para recibir el otro tratamiento). También se excluyeron los estudios que incorporaron pacientes con sepsis ya diagnosticada.
Resultados clave
En esta revisión se incluyeron tres ensayos controlados aleatorios (estudios donde los participantes se asignaron a los grupos de tratamiento por azar), con 1199 participantes. En general, cuando los sistemas de monitorización automatizada se compararon con atención habitual no hubo diferencias significativas en el tiempo transcurrido hasta el comienzo del tratamiento antimicrobiano (como los tratamientos antimicrobianos y antimicóticos, evidencia de muy baja calidad), la duración de la estancia hospitalaria en ámbitos de cuidados intensivos (evidencia de muy baja calidad) ni en la mortalidad a los 14 días, a los 28 días o al alta (evidencia de muy baja calidad). Hubo evidencia de muy baja calidad disponible sobre el fracaso en la detección de la sepsis, pero el reporte de los datos no estuvo claro y no permitió analizar este resultado de manera significativa. En los estudios no se informaron otros resultados que se deseaban evaluar, como el tiempo transcurrido hasta el inicio de la reanimación con líquidos (el proceso de aumentar la cantidad de líquidos en el cuerpo), la mortalidad a los 30 días ni la calidad de vida.
Calidad de la evidencia
Los resultados de esta revisión proporcionan evidencia de muy baja calidad, limitada, que impide establecer conclusiones significativas. No están claros los efectos que los sistemas automatizados para la monitorización de la sepsis tienen sobre cualquier resultado incluido en esta revisión y, por lo tanto, no está claro si la monitorización automatizada de la sepsis es beneficiosa. Se necesita evidencia adicional y de alta calidad para ayudar a abordar la pregunta de la revisión.
Authors' conclusions
Summary of findings
| Automated monitoring systems compared to standard care for detecting sepsis | ||||||
| Patient or population: participants of any age admitted to the intensive care or critical care unit for any reason (including, but not limited to postsurgery, trauma, stroke, myocardial infarction, arrhythmia, burns, and hypovolaemic or haemorrhagic shock) Settings: hospitals in USA Intervention: automated monitoring systems (any process capable of screening patient records or data (one or more systems) automatically at intervals for markers or characteristics that are indicative of sepsis) Comparison: standard care such as paper‐based systems | ||||||
| Outcomes | Illustrative comparative risks* (95% CI) | Relative effect | No of participants | Quality of the evidence | Comments | |
| Assumed risk | Corresponding risk | |||||
| Standard care | Automated monitoring | |||||
| Time to initiation of antimicrobial therapy (Time to initiation starts at the time of admission) | 3 studies reported data in relation to this outcome but data could not be pooled. The largest study included 680 participants and reported median time to initiation of first or new antibiotic was 5.6 hours (IQR 2.3 to 19.7) in the intervention group (n = not stated) and 7.8 hours (IQR 2.5 to 33.1) in the control group (n = not stated) | Unclear for this outcome (3 studies containing approximately 1200 participants overall) | Very low1,2 | |||
| Time to initiation of fluid resuscitation (Time to initiation starts at the time of admission) | Not reported | Not reported | Not reported | ‐ | ‐ | None of the included studies reported this outcome |
| 30‐day mortality* | *No studies reported 30‐day mortality. 1 study reported 14‐day mortality and found no significant differences between groups (20% in the intervention, 21% in the control). 1 study reported mortality at 28 days or discharge and found no significant differences between groups (14% in the intervention, 10% in the control). Sample sizes were not reported adequately for these outcomes and so we could not estimate confidence intervals | ‐ | Very low1,2 | |||
| Length of stay in ICU (in days) | Median 3.0 (IQR 2 to 4) days | Median 3.0 (IQR 2 to 5) days | ‐ | 442 (1 study) | Very low1,3 | P = 0.22 |
| Failed detection of sepsis (as reported by studies) | 1 study reported failed detection of sepsis in 2 participants but did not state which group(s) they occurred in. | 560 (1 study) | Very low1,2 | |||
| Quality of life measured at the latest available time point post‐discharge from ICU (preferred measure SF‐36 then EQ‐5D) | Not reported | Not reported | Not reported | ‐ | None of the included studies reported this outcome. | |
| *The basis for the assumed risk (e.g. the median control group risk across studies) is provided in footnotes. The corresponding risk (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 High quality: we are very confident that the true effect lies close to that of the estimate of the effect. | ||||||
| 1Downgraded two levels for risk of bias due to unclear randomization methods, allocation concealment and blinding. | ||||||
Background
Description of the condition
Sepsis is a life‐threatening clinical syndrome. The criteria for the diagnosis of sepsis have evolved over time and are generally defined by international consensus groups (ACCP/SCCM 1992; Levy 2003; Singer 2016). It is usually diagnosed when a patient has a suspected or documented infection, alongside systemic inflammatory response syndrome (SIRS). The criteria for diagnosing SIRS typically include the presence of two or more of the following abnormalities in the absence of other known causes, such as chemotherapy.
-
Temperature greater than 38.3°C (hyperthermia) or less than 36.0°C (hypothermia)
-
Heart rate greater than 90 beats per minute (tachycardia)
-
Breathing greater than 20 breaths per minute (tachypnoea) or arterial carbon dioxide concentration (PaCO²) less than 32 mmHg (hyperventilation)
-
Blood glucose greater than 7.7 mmol/L (hyperglycaemia) in the absence of diabetes mellitus
-
New altered behaviour or mental state
-
White blood cell count greater than 12,000 per microlitre (leukocytosis) or less than 4000 per microlitre (leukopenia) or normal white blood cell count with greater than 10% immature forms.
If left untreated, sepsis can develop into severe sepsis (sepsis with organ dysfunction) or septic shock (severe sepsis with hypotension despite adequate fluid resuscitation). Mortality for this group of patients can be 40% or even higher depending on definitions used (Szakmany 2018). Patients with sepsis often require admission to the intensive care unit (ICU). The incidence of sepsis in people admitted to ICU for other critical illnesses is also high (20% to 70% of people admitted to ICU in Europe, with considerable variance by country, Vincent 2006). Diagnosing sepsis is challenging and time consuming. It often requires the combination of information from several sources to be reviewed (e.g. patient history, laboratory data, and physiological data) at regular intervals (Cohen 2015). Further, although many options are available to guide therapy (Andriolo 2017), and many interventions have been tested (Annane 2015; Borthwick 2017), early detection offers the prospect of a better therapeutic response. In addition, the complexity of diagnosis combined with the degree of illness results in a significant cost for treating sepsis in the ICU. For example, the cost of treating each patient with sepsis in the ICU was recently estimated as approximately EUR 29,000 in the Netherlands (Koster‐Brouwer 2014), or GBP 20,000 in the UK (UK Sepsis Trust 2013).
Description of the intervention
Automated monitoring systems provide a means of monitoring patient data continuously, and can facilitate the assembly of data from unconnected information systems (Hooper 2012). These tools are variously referred to as alert systems, detection systems and monitoring systems (Makam 2015). In essence, the systems process clinical data ‐ that are routinely collected ‐ to identify sepsis according to predetermined diagnostic thresholds, and include an electronic means of alerting staff. Although the algorithms (i.e. criteria) used to identify sepsis vary between the different automated systems (Buck 2014; Nachimuthu 2012), their key feature is an ability to monitor one or more electronic systems (e.g. patient electronic health records) for potential indicators of sepsis. For example, a system may 'listen' for modified SIRS criteria (Hooper 2012), although SIRS criteria have recently been deemed to have inadequate specificity and sensitivity for the detection of sepsis (Singer 2016). Following detection of potential sepsis, the system should provide an automated notification (e.g. via email, phone message or pager) to the relevant physician or nurse, flagging the requirement for clinical evaluation and potential initiation of therapy (Hooper 2012; Koenig 2011). The use of electronic early‐recognition tools has previously been validated in the critical care setting for detection of acute respiratory distress syndrome (ARDS) (Koenig 2011). Potential adverse effects of automated systems might include the failure to detect sepsis and alarm fatigue (i.e. where frequent false alarms cause staff to ignore notification of potential sepsis).
How the intervention might work
Automated detection systems monitor patient data continuously to facilitate the early detection of sepsis in the ICU. The diagnosis of sepsis or septic shock is particularly time‐sensitive, as the length of time until initiation of appropriate antimicrobial therapy or fluid resuscitation is a critical determinant of survival in these patients (Dellinger 2013; Kumar 2006; Rivers 2001; Yealy 2014). Therefore, guidelines recommend early fluid resuscitation of the septic patient within six hours of recognition of sepsis, and administration of broad‐spectrum antibiotics within one hour of the recognition of septic shock or severe sepsis without septic shock (Dellinger 2013). Automated detection systems offer the possibility of monitoring patients in 'real time' (Meurer 2009), and can alert the relevant physicians or nurses (e.g. by email or pager) to the need for timely clinical evaluation and potential initiation of treatment.
Why it is important to do this review
Although the rate of mortality from sepsis has improved (Kaukonen 2014; McPherson 2013), national audits indicate that clinical standards relevant to the management of patients with sepsis are not being met, despite ongoing education programmes (CEM 2012). The UK Parliamentary Ombudsman recently published a detailed report that identified common themes in 10 case studies of patients that died following sepsis (Parliamentary Ombudsman 2013). Failings were identified throughout the care pathway, from carrying out a timely initial assessment and identifying the source of infection, to adequate monitoring and timely initiation of treatment (Parliamentary Ombudsman 2013). Automated monitoring systems for the detection of sepsis may facilitate earlier detection and treatment of sepsis in the ICU, potentially increasing adherence to clinical standards and improving patient outcomes.
Additionally, sepsis is the most expensive condition treated in hospitals, accounting for approximately 5% of total hospitalization costs and an overall annual cost of USD 20.3 billion in the USA (Torio 2011), and more than GBP 2.5 billion in the UK (UK Sepsis Trust 2013). Early detection of sepsis via automated systems and subsequent timely intervention may reduce treatment costs and overall resource use. The UK Sepsis Trust estimates that there are more than 100,000 hospitalizations per year for sepsis, and that achieving 80% delivery of basic standards of care could result in a potential cost saving of GBP 170 million per year, even after allowing for increased survival‐related costs (UK Sepsis Trust 2013).
Finally, it is now recognized that sepsis is associated with significant mortality, long‐term morbidity and a reduction in health‐related quality of life (Winters 2010), thus reinforcing the importance of early effective treatment from both a patient and resource utilization perspective. In summary, there is clear rationale to synthesize the evidence relating to the use of automated systems for the detection of sepsis.
Objectives
To evaluate whether automated systems for the early detection of sepsis can reduce the time to appropriate treatment (such as initiation of antibiotics, fluids, inotropes, and vasopressors) and improve clinical outcomes in critically ill patients in the ICU.
Methods
Criteria for considering studies for this review
Types of studies
We included randomized controlled trials (RCTs) reported as full text, or published as abstract only, and unpublished data. We did not exclude unblinded studies. We excluded cross‐over studies as it would not be feasible to evaluate automated monitoring followed by standard care (or vice‐versa) in the same participant as the detection of sepsis requires treatment. We also excluded quasi‐RCTs (studies using inadequate methods for randomization, such as date of birth of participant or date of ICU admission).
Types of participants
We included participants of any age who were admitted to intensive or critical care units for critical illness (including, but not limited to postsurgery, trauma, stroke, myocardial infarction, arrhythmia, burns, and hypovolaemic or haemorrhagic shock). We excluded participants admitted with confirmed sepsis.
Types of interventions
We included studies that randomized participants to receive monitoring for sepsis using an automated system versus standard care (i.e. systems where paper‐based or other formats of observation charts are reviewed by staff directly). We defined an automated system as any process capable of screening patient records or data (one or more systems) automatically at intervals for markers or characteristics that are indicative of sepsis. The parameters/algorithm used by the system (for example, the thresholds of blood pressure indicative of hypotension or the nature of the biomarkers employed) may vary. However, if the system identifies a potential case of sepsis, it should flag the patient's record and alert the relevant healthcare professional (via email, pager or phone message).
Types of outcome measures
Primary outcomes
-
Time to initiation of antimicrobial therapy* (in minutes)
-
Time to initiation of fluid resuscitation* (in minutes)
-
30‐day mortality
*Time to initiation starts at the time of admission.
Note: studies were not required to distinguish between sepsis that is detected via standard care pathways and sepsis detected via the automated system in the intervention group; if studies employ adequate control groups and sample sizes, and if automated monitoring confers a benefit, a difference between groups should be detectable.
Secondary outcomes
-
Length of stay in ICU (in days)
-
Failed detection of sepsis (during ICU stay), as reported by studies
-
Quality of life measured at the latest available time point post‐discharge from ICU (preferred measure SF‐36 then EQ‐5D)
Search methods for identification of studies
Electronic searches
We identified RCTs through literature searching with systematic and sensitive search strategies as outlined in Chapter 6.4 of the Cochrane Handbook of Systematic reviews of Interventions (Lefebvre 2011). We did not apply restrictions to language or publication status.
We searched the following databases for relevant trials.
-
Cochrane Central Register of Controlled Trials (CENTRAL; 2017, Issue 8) in the Cochrane Library
-
MEDLINE (Ovid SP, 1966 to 18 September 2017)
-
Embase (Ovid SP, 1988 to 18 September 2017)
-
CINAHL (Cumulative Index to Nursing and Allied Health Literature, EBSCO, 1937 to 18 September 2017)
-
Web of science (1900 to 18 September 2017)
-
LILACS (Bireme, 1982 to 18 September 2017)
We developed a subject‐specific search strategy in MEDLINE and used that as the basis for the search strategies in the other databases listed. Where appropriate, the search strategy was expanded with search terms for identifying RCTS. All search strategies can be found in Appendix 1, Appendix 2, Appendix 3, Appendix 4, Appendix 5, and Appendix 6.
We scanned the following trials registries for ongoing and unpublished trials:
-
The World Health Organization International Clinical Trials Registry Platform (www.who.int/ictrp/en/)
-
ClinicalTrials.gov (clinicaltrials.gov)
We developed the search strategy in consultation with Cochrane Dementia's Information Specialist.
Searching other resources
We scanned the reference lists and citations of included studies and any relevant systematic reviews identified for further references to additional studies. When necessary we contacted study authors by email for additional information.
Data collection and analysis
Our methods for data collection and analysis differed from those stated in the published protocol (Evans 2016). The differences and reasons for them are detailed in the section 'Differences between protocol and review'.
Selection of studies
Two review authors (SW, PA) independently screened titles and abstracts arising from the searches, for possible inclusion in the review; we retrieved and assessed the full‐text articles of the potentially relevant studies and two review authors (SW, PA) independently identified: a) studies for inclusion in the review; and b) ineligible studies; recording the reasons for exclusion in the 'Characteristics of excluded studies' table. We planned to resolve disagreements by discussion or, if required, through consultation with a third review author (IK). We identified and excluded duplicate records. We also planned to collate multiple reports of the same study so that the study is the unit of interest. The results of this selection process is summarized in a PRISMA flow diagram (Moher 2009).
Data extraction and management
Two authors (SW, IK) extracted the following information for each study:
-
methods: study design; total duration of study; number of study centres and location; study setting; date of study;
-
participants: number of participants that were:
-
randomly assigned,
-
discontinued the study, and
-
excluded from the analyses after randomization; condition and severity of condition; inclusion and exclusion criteria;
-
-
intervention: intervention, comparator, algorithm/criteria used by the automated system;
-
outcomes: primary and secondary outcomes including details of time points;
-
other information: trial funding and potential conflicts of interest of authors
Another review author (PA) checked data extraction accuracy.
Assessment of risk of bias in included studies
Two review authors (SW, PA) independently assessed study risk of bias according to criteria outlined in the Cochrane Handbook for Systematic Reviews of Interventions (Higgins 2011). We assessed the risk of bias for the following domains:
-
random sequence generation;
-
allocation concealment;
-
blinding of participants and personnel;
-
blinding of outcome assessment;
-
incomplete outcome data;
-
selective outcome reporting;
-
other bias.
For each domain, we graded the risk of bias as high, low or unclear, and provided justification for our judgement in the 'Risk of bias' table.
Measures of treatment effect
We planned to analyse dichotomous data using risk ratios with 95% confidence intervals, and continuous data with mean differences and 95% confidence intervals.
Unit of analysis issues
All studies were randomized by individual, and outcome data were reported for participants.
Dealing with missing data
We contacted study investigators to obtain missing outcome data and to verify important study characteristics, but did not receive any responses.
Assessment of heterogeneity
Insufficient data were available to permit assessment of heterogeneity.
Assessment of reporting biases
We planned to explore small study and publication biases by creating and examining a funnel plot if we were able to pool data from more than 10 trials.
To assess within‐study reporting bias of outcomes, we planned to search for trial protocols matching included studies published after 1 July 2005 in the Clinical Trial Register at the International Clinical Trials Registry Platform of the World Health Organization (www.who.int/ictrp/en/), and Clinicaltrials.gov (clinicaltrials.gov/), for the trial protocols.
Data synthesis
Insufficient data were available to permit meta‐analysis or a meaningful summary of the evidence.
'Summary of findings' table and GRADE
We used the principles of the GRADE system (Guyatt 2008), to assess the quality of the body of published and unpublished evidence associated with the following outcomes in our review: time to initiation of antimicrobial therapy, time to initiation of fluid resuscitation, 30‐day mortality, length of stay in ICU, failed detection of sepsis, and quality of life (postdischarge).
Two authors (SW, PA) independently assessed the quality of the evidence. We used the five GRADE considerations (study limitations, inconsistency, imprecision, indirectness and publication bias) to assess the quality of the body of evidence as it relates to the studies that contribute data to the prespecified outcomes. The GRADE approach appraises the quality of a body of evidence based on the extent to which one can be confident that an estimate of effect or association reflects the item being assessed. The quality of a body of evidence takes into consideration within‐study risk of bias (methodologic quality) (Guyatt 2011a), the directness of the evidence (Guyatt 2011b), heterogeneity of the data (Guyatt 2011c), precision of effect estimates (Guyatt 2011d), and risk of publication bias (Guyatt 2011e). We used methods and recommendations described in Chapter 8 (section 8.5 and 8.7; Higgins 2011), Chapter 11 (Schünemann 2011) and Chapter 13 (section 13.5; Reeves 2011) of the Cochrane Handbook for Systematic Reviews of Interventions, using GRADEpro software (GRADEpro GDT 2015). We justified all decisions to downgrade the quality of studies using footnotes.
Subgroup analysis and investigation of heterogeneity
Insufficient data were available to permit subgroup analysis.
Sensitivity analysis
Insufficient data were available to permit sensitivity analysis.
Results
Description of studies
Results of the search
The search retrieved 3233 results, we selected 10 studies for full text consideration, and included three in this review. We have summarized the selection process in Figure 1.
Study flow diagram
Included studies
We included three studies in this review (total of 1199 participants) (Hooper 2010; Hooper 2011; Hooper 2012). However, two of the studies were abstracts from conferences and contained limited data (Hooper 2010; Hooper 2011). We tried to contact the lead author of the studies to obtain additional information, and to check if the studies were different reports relating to a single study but we were unable to make contact. The three publications quote one grant number in common, which appears to be a programme grant, and have the same first author. However they contain different data and we have treated them as three separate studies for this review.
Study populations
The studies included participants admitted to the medical or surgical ICU but no details on the participants' underlying conditions were provided. Some of the participants were receiving mechanical ventilation.
Settings
The studies were described as being conducted in medical intensive care units (MICU) or in a tertiary care centre. One study stated that it was conducted in the USA but two studies did not provide this information although it can be reasonably assumed that they were conducted in the USA too.
Interventions
The interventions included in this review included computerized automated monitoring systems to monitor and alert one or more of the care team when modified SIRS criteria were met. One study described this as a 'listening application' but none of the studies described how the system worked or what information it monitored or listened to.
All of the included studies assessed the automated alert component of the monitoring system. All participants received automated electronic monitoring during their hospital stay, and were randomized to an intervention group (automated alerts sent from the system to the care team) or to usual care (no automated alerts sent from the system). Only one study explained the process for alerting the care team once modified SIRS criteria were met, where a text message notification was sent to the pagers of the care and admissions teams. It also flagged the patient's name on the primary team physician's electronic patient list, and flagged the patient's medical record so that any physicians taking care of the patient could see the information. Physicians were asked to acknowledge receipt of the notification and indicate if the participant had sepsis. If a physician failed to respond, a reminder was resent after one hour. The system did not give any management recommendations and providers were not instructed to treat alerted participants in a different manner than any other patient. If physicians determined a participant to be septic, further notifications by the system were suspended for seven days. If they determined a participant not to be not septic, further notifications were suspended for two days unless a previously normal white blood cell count or temperature became abnormal.
None of the included studies assessed other components of the monitoring systems, such as the underlying sepsis‐detection algorithm.
Comparators
The comparator included in this review was standard care. Two of the studies stated that the comparator was 'usual care' but did not state what this entailed. One study described participants in the usual‐care group as receiving computerized monitoring, which generated a time stamp when modified SIRS criteria were met but notifications were not relayed to any of the care team.
Funding sources
All studies stated that they received funding, but only provided initials of the funders. It can be reasonably assumed that all three studies were funded by the National Institutes of Health (NIH), and one study also received funding from National Centre for Research Resources/National Intitutes of Health (NCRR/NIH), and National Science Foundation (NSF).
Excluded studies
We excluded seven studies from the review.
Three of the seven studies were excluded because they did not report the results from RCTs (Croft 2014; Karch 2016; Slotman 2000). A further three studies were excluded because the participants were diagnosed with sepsis at enrolment (Semler 2013; Semler 2015; Zhang 2013). One study was excluded because it was not based in the ICU (Sawyer 2011).
Studies awaiting classification
There are no studies awaiting classification.
Ongoing studies
We identified no ongoing studies.
Risk of bias in included studies
Risk of bias summary: review authors' judgements about each risk of bias item for each included study.
Risk of bias graph: review authors' judgements about each risk of bias item presented as percentages across all included studies.
Allocation
All three included studies were described as RCTs and we therefore considered them to be at low risk of selection bias. However, none of the three studies provided details of the randomization procedure and so it is unclear if the methods used influenced results.
Blinding
None of the three studies stated if participants or personnel involved in the study were blinded to study group allocation. Patients receiving care in the ICU would usually be unconscious or very unwell and so study participants are likely to be unaware of allocation. In addition it is unlikely that study participants could influence any of the outcomes considered in this review. Therefore participant blinding, or lack of, is unlikely to have any effect on study outcomes. A lack of study investigator or staff blinding could influence behaviour, such that participants in the standard care group are monitored more closely if staff have a heightened awareness of sepsis, or that participants in the intervention group are monitored less closely if staff feel they can rely on the intervention to alert them to deteriorating patient condition. This may mean that potential differences between groups are reduced such that there is no detectable differences between them.
Incomplete outcome data
Insufficent information was provided to assess this in two studies (Hooper 2010; Hooper 2011). One participant in Hooper 2012 was excluded after randomization as they died before an alert could be generated. Otherwise in this study, all participants appear to have been followed up to death or discharge from hospital.
Selective reporting
We were unable to locate the trial protocols by searching on trials registers as listed in the methods, and so we identified no reporting biases.
Other potential sources of bias
We did not identify any other sources of bias.
Effects of interventions
Primary outcomes
1. Time to initiation of antimicrobial therapy
Three studies (n = 1199) reported median time to initiation of first or new antimicrobial therapy.
Hooper 2012 reported a median time of 6.0 hours (interquartile range (IQR) 2.4 to 18.8) for the intervention group (n = 220) and 6.1 hours (IQR 2.5 to 21.0) for the control group (n = 222). No clear difference between the groups was seen (P = 0.95). Hooper 2011 also reported this outcome. This study included 680 participants but did not state the number of participants allocated to each group. In this study, median time to initiation of first or new antibiotic was 5.6 hours (IQR 2.3 to 19.7) in the intervention group (n = not stated) and 7.8 hours (IQR 2.5 to 33.1) in the control group (n = not stated).
Hooper 2012 also reported subgroup analyses for this outcome. Among only those participants diagnosed with sepsis (n = 61), median time to initiation of antimicrobial therapy was 3.4 hours (IQR 1.7 to 12.3) in the intervention subgroup (n = 28), and 3.5 hours (IQR 1.2 to 13.8) in the control subgroup (n = 33). No clear difference between the groups was seen (P = 0.93). Among only those participants not on antibiotics at the time of enrolment (n = 231), 131 were subsequently administered antibiotics at a median time of 5.2 hours (IQR 2.1 to 13.0) in the intervention group (n = 66), and 5.1 hours (IQR 1.5 to 17.0) in the control group (n = 65). No clear difference was seen.
Although Hooper 2010 did not report this outcome for the study group (n = 77), it did report it for a subgroup of participants who were diagnosed with sepsis and received antibiotics at enrolment (n = 9), reporting that there were no differences between groups. Median time to first or new antibiotic initiation was 12.2 hours (0.96 to 29.0, (IQR as this is not stated in the results)) in the intervention group (n = 4), and 6.2 hours (2.4 to 23.5) in the control group (n = 5). Lack of meaningful differences for this result is likely to be because of a small sample size, but the findings are counter‐intuitive and in the opposite direction to results from the other studies, since the median time to initiation is almost double the time in the intervention group, which received alerts, than in the control group. It is unclear why.
Overall we assessed the evidence for this outcome to be very low (see summary of findings Table for the main comparison).
2. Time to initiation of fluid resuscitation
None of the included studies reported this outcome.
3. 30‐day mortality
None of the included studies specifically reported this outcome (30‐day mortality), but two studies reported mortality over different time frames and involved a total of 519 participants.
Hooper 2010 (n = 77) reported 14‐day mortality, which was 20% in the intervention group (numerator and denominator not stated) and 21% in the control group (numerator and denominator not stated, P = 0.94).
Hooper 2012 (n = 442) also reported mortality, with the methods stating follow‐up to 28 days or hospital discharge, whichever occurred first. Overall there was 14% mortality in the intervention group, and 10% mortality in the control group (numerators not reported, P = 0.29).
Overall, we assessed the evidence for this outcome to be very low (see summary of findings Table for the main comparison).
Secondary outcomes
1. Length of stay in ICU
One study, Hooper 2012 involving 442 participants, reported this outcome. Median length of stay in the ICU was 3.0 days (IQR 2.0 to 5.0) in the intervention group (n = 220) and 3.0 days (IQR 2.0 to 4.0) in the control group (n = 222). No clear difference between groups was seen (P = 0.22).
Overall, we assessed the quality of the evidence for this outcome to be low (see summary of findings Table for the main comparison).
2. Failed detection of sepsis during ICU stay
One study, Hooper 2012, reported this outcome. Although this study states that it involved a total of 442 participants who met modified SIRS criteria and were randomized, it reports that 60 out of 560 participants admitted to the medical ICU did not meet modified SIRS criteria at any point, but determined two participants to be septic during their ICU stay. It is unclear if the 560 participants involved in this outcome were different from the 442 randomized to the study, or why all 560 were not included in the study. Therefore it is also unclear if any of the participants with failed detection of sepsis belonged to one of the study groups.
Overall, we assessed the evidence for this outcome to be very low (see summary of findings Table for the main comparison).
3. Quality of life measured at the latest available time point post‐discharge from ICU
None of the included studies reported this outcome.
Discussion
Summary of main results
We included evidence from three studies involving 1199 participants in this review, although it is unclear if the study populations in the three studies were independent of each other. We did not undertake any meta analysis of the data, and we are confident that our conclusions would not change even if the populations were not independent of each other.
All three studies assessed the alert component of the monitoring system. All three studies reported time to initiation of first or new antimicrobial therapy (n = 1999). There were no meaningful differences between those receiving automated monitoring alerts and those receiving standard care (automated monitoring and no alerts) in any of the three studies (Hooper 2010; Hooper 2011; Hooper 2012). This was also the case in subgroup analyses of 61 people diagnosed with sepsis, and 131 people not on antibiotics at time of enrolment to the study who were subsequently administered antibiotics (Hooper 2012). It was not possible to pool the results due to insufficient data reported in the studies, and lack of similar subgroup analysis between studies.
None of the included studies reported our prespecified outcome '30‐day mortality'. Instead, one study (Hooper 2010) reported 14‐day mortality, and another study (Hooper 2012), reported mortality up to 28 days or discharge from hospital, whichever came soonest. Neither of these studies included information on the number of participants included in each study group, and none of the studies found meaningful differences between those receiving automated monitoring and those receiving standard care.
One study reported length of stay in ICU (Hooper 2012), but did not report the number of participants in each study group. No meaningful differences were reported between people receiving automated monitoring and those receiving standard care.
None of the included studies reported time to fluid resuscitation in minutes, or quality of life after the participant was discharged from the ICU. One study did report failed detection of sepsis (Hooper 2012), but did not report whether the cases occurred in participants receiving automated monitoring, standard care, or were from outside the study population.
Overall completeness and applicability of evidence
All of the included studies assessed the automated alert component of the monitoring system, but none of the studies assessed the whole system or its other components, such as the underlying algorithm
All of the automated monitoring technologies used the same criteria for detecting when a patient met the criteria for alerting staff, and all included patient populations that are likely to be representative of those in the intensive or critical care unit. All of the evidence appears to be generated in the USA, therefore the evidence is likely to be applicable to similar care settings in the USA. However, it is unclear if the evidence would be applicable to similar settings in other countries, where care standards and processes may be different to the USA (such as staffing ratios and standard monitoring practices for example). The evidence can be considered to be incomplete, as included studies often reported relevant outcomes without providing sufficient information to enable analysis, or reporting dissimilar subgroup analyses. Some primary and secondary outcomes that we wanted to include in this review were not reported at all. Therefore we are unable to draw meaningful conclusions about automated sepsis monitoring.
Quality of the evidence
In general, studies reported insufficient information to enable us to assess adequately the quality of the evidence. All of the included studies were considered to be RCTs and so biases due to selection processes are likely to be low. The included studies did not state if participants or investigators were blinded to treatment allocation and so we were uncertain if allocation biases were present. We felt that lack of participant blinding was unlikely to influence study results, but if staff delivering care to patients were not adequately blinded, this could mean that potential differences between groups were reduced such that there were no detectable differences between them. Data relating to attrition were not well reported in the studies so it is unclear if all participants have been followed up to death or discharge from hospital. Attrition in short‐term studies in hospital such as those included in this review should be low, but we were unable to be sure.
Reporting of the measured outcomes was poor. Two studies appeared only as conference abstracts and therefore lacked detailed information about outcomes. This made it hard to assess either the results themselves or their consistency and precision. Overall this reduced our confidence in the body of evidence, particularly as results from some whole studies seemed to be missing.
Potential biases in the review process
We made several review decisions after we had reviewed the study data, mainly because the studies reported insufficient data to enable us to progress with our planned approach (see Differences between protocol and review). This may introduce a bias into the review process in that the outcomes reported in the studies may be subject to outcome reporting bias.
Agreements and disagreements with other studies or reviews
We are not aware of other systematic reviews addressing this question.
Study flow diagram
Risk of bias summary: review authors' judgements about each risk of bias item for each included study.
Risk of bias graph: review authors' judgements about each risk of bias item presented as percentages across all included studies.
| Automated monitoring systems compared to standard care for detecting sepsis | ||||||
| Patient or population: participants of any age admitted to the intensive care or critical care unit for any reason (including, but not limited to postsurgery, trauma, stroke, myocardial infarction, arrhythmia, burns, and hypovolaemic or haemorrhagic shock) Settings: hospitals in USA Intervention: automated monitoring systems (any process capable of screening patient records or data (one or more systems) automatically at intervals for markers or characteristics that are indicative of sepsis) Comparison: standard care such as paper‐based systems | ||||||
| Outcomes | Illustrative comparative risks* (95% CI) | Relative effect | No of participants | Quality of the evidence | Comments | |
| Assumed risk | Corresponding risk | |||||
| Standard care | Automated monitoring | |||||
| Time to initiation of antimicrobial therapy (Time to initiation starts at the time of admission) | 3 studies reported data in relation to this outcome but data could not be pooled. The largest study included 680 participants and reported median time to initiation of first or new antibiotic was 5.6 hours (IQR 2.3 to 19.7) in the intervention group (n = not stated) and 7.8 hours (IQR 2.5 to 33.1) in the control group (n = not stated) | Unclear for this outcome (3 studies containing approximately 1200 participants overall) | Very low1,2 | |||
| Time to initiation of fluid resuscitation (Time to initiation starts at the time of admission) | Not reported | Not reported | Not reported | ‐ | ‐ | None of the included studies reported this outcome |
| 30‐day mortality* | *No studies reported 30‐day mortality. 1 study reported 14‐day mortality and found no significant differences between groups (20% in the intervention, 21% in the control). 1 study reported mortality at 28 days or discharge and found no significant differences between groups (14% in the intervention, 10% in the control). Sample sizes were not reported adequately for these outcomes and so we could not estimate confidence intervals | ‐ | Very low1,2 | |||
| Length of stay in ICU (in days) | Median 3.0 (IQR 2 to 4) days | Median 3.0 (IQR 2 to 5) days | ‐ | 442 (1 study) | Very low1,3 | P = 0.22 |
| Failed detection of sepsis (as reported by studies) | 1 study reported failed detection of sepsis in 2 participants but did not state which group(s) they occurred in. | 560 (1 study) | Very low1,2 | |||
| Quality of life measured at the latest available time point post‐discharge from ICU (preferred measure SF‐36 then EQ‐5D) | Not reported | Not reported | Not reported | ‐ | None of the included studies reported this outcome. | |
| *The basis for the assumed risk (e.g. the median control group risk across studies) is provided in footnotes. The corresponding risk (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 High quality: we are very confident that the true effect lies close to that of the estimate of the effect. | ||||||
| 1Downgraded two levels for risk of bias due to unclear randomization methods, allocation concealment and blinding. | ||||||
