Long‐acting muscarinic antagonist (LAMA) plus long‐acting beta‐agonist (LABA) versus LABA plus inhaled corticosteroid (ICS) for stable chronic obstructive pulmonary disease

Summary of findings 1. Long‐acting muscarinic antagonist (LAMA) plus long‐acting beta‐agonist (LABA) versus LABA plus inhaled corticosteroid (ICS) for stable chronic obstructive pulmonary disease (COPD)

LAMA plus LABA versus LABA plus ICS for stable COPD
Population: people with stable COPD Setting: outpatient. Studies were conducted in > 50 countries including low‐, medium‐, and high‐income countries from all continents Intervention: LAMA+LABA Comparison: LABA+ICS
Outcomes	*Anticipated absolute effects (95% CI)**		Relative effects (95% CI)	Number of participants (studies)	Certainty of the evidence (GRADE)	Comments
Outcomes	LABA+ICS	LAMA+LABA	Relative effects (95% CI)	Number of participants (studies)	Certainty of the evidence (GRADE)	Comments
Exacerbations (number of people experiencing ≥ 1 exacerbations) Follow‐up: 6 to 52 weeks	291 per 1000	272 per 1000 (243 to 303)	OR 0.91 (0.78 to 1.06)	20,960 (13 RCTs)	⊕⊕⊕⊝ Moderate^a‐e	Low OR means favourable outcome
Serious adverse events (number of people experiencing ≥ 1 SAEs) Follow‐up: 6 to 52 weeks	148 per 1000	150 per 1000 (136 to 166)	OR 1.02 (0.91 to 1.15)	23,183 (18 RCTs)	⊕⊕⊕⊕ High^b,c,d,e	Low OR means favourable outcome. Herth 2020 was discarded as no events were reported in either arm.
Quality of life as measured by SGRQ total score change from baseline (MD) Follow‐up: 12 to 52 weeks Scale 0 to 100	‐	‐	MD ‐0.57 (1.36 lower to 0.21 higher)	14,437 (9 RCTs)	⊕⊕⊕⊝ Moderate^a‐e	Low MD means favourable outcome
Trough FEV₁ change from baseline Follow‐up: 6 to 52 weeks	‐	‐	MD 0.07 L (0.05 to 0.08)	14,681 (12 RCTs)	⊕⊕⊕⊝ Moderate^a,b,d,e	High MD means favourable outcome
Pneumonia Follow‐up: 12 to 52 weeks	45 per 1000	28 per 1000 (24 to 33)	OR 0.61 (0.52 to 0.72)	21,829 (14 RCTs)	⊕⊕⊕⊕ High^b,d,e	Low OR means favourable outcome
All‐cause death Follow‐up: 6 to 52 weeks	10 per 1000	14 per 1000 (11 to 18)	OR 1.35 (1.05 to 1.75)	21,510 (15 RCTs)	⊕⊕⊕⊝ Moderate^b,d,e,f	Low OR means favourable outcome
SGRQ total score change from baseline (≥ 4 points, MCID) Follow‐up: 26 to 52 weeks	373 per 1000	387 per 1000 (349 to 427)	OR 1.06 (0.90 to 1.25)	13,614 (4 RCTs)	⊕⊕⊕⊝ Moderate^a‐e	High OR means favourable outcome
The absolute risk* (and its 95% CI) of LAMA+LABA group is based on the assumed risk in the LABA+ICS group and the OR of the intervention (and its 95% CI). CI: confidence interval; COPD: chronic obstructive pulmonary disease; FEV₁: forced expiratory volume in one second; ICS: inhaled corticosteroid; LABA: long‐acting beta‐agonist; LAMA: long‐acting muscarinic antagonist; MCID: minimal clinically important difference; MD: mean difference; OR: odds ratio; RCT: randomised controlled trial; SGRQ: St. George's Respiratory Questionnaire.
GRADE Working Group grades of evidence High certainty: we are very confident that the true effect lies close to that of the estimate of effect. Moderate certainty: we are moderately confident in the effect estimate: the true effect is likely to be close to the estimate of effect but may be substantially different. Low certainty: our confidence in the effect estimate is limited: the true effect may be substantially different from the estimate of effect. Very low certainty: we have very little confidence in the effect estimate: the true effect is likely to be substantially different from the estimate of effect.
^aThere was a considerable heterogeneity, I² > 50%. ^bStudies varied in their inclusion criteria around recent history of exacerbation, baseline percentage predicted FEV₁ and exclusion of asthma. No indirectness downgrades, but impacts interpretation and applicability of results (see Discussion). ^cConfidence interval includes benefit of either treatment but lies within predefined threshold for clinical importance (0.7 to 1.5 for OR outcomes; 4‐point MCID for SGRQ) ‐ no downgrade for imprecision ^dNo downgrades for publication bias. Funnel plots examined for primary outcomes (exacerbations, serious adverse events, SGRQ, FEV₁) do not show obvious asymmetry (see Figure 3; Figure 3; Figure 4; Figure 6). Nearly all studies included in pneumonia and all‐cause death analyses. Only 4 studies included in SGRQ 4+ change analysis but not specified in most studies. ^eNo downgrades for risk of bias. Across outcomes, all or almost all studies were at high risk of 'other' bias due to conflicts of interest, and studies contributing between 18.5% and 35.9% of the analysis weight were rated high risk of bias in at least one other domain. We did not prespecify a threshold or specific domains for downgrading, but considered it insufficient to downgrade if studies contributing more than half the analysis weight were at low or unclear risk of bias in all but the 'other' domain. ^fThere was imprecision due to a low number of events.

Background

Description of the condition

Chronic obstructive pulmonary disease (COPD) is characterised by bronchial obstruction, systemic inflammation, and comorbidities. It is the third leading cause of death worldwide, with more than 3.23 million people dying due to COPD each year (WHO 2019). In addition to active tobacco smoking, indoor biomass smoke, outdoor air pollution, and occupational exposure also play a role in the development of COPD. The most common symptoms of COPD – shortness of breath on exertion and cough – are present for a prolonged period and typically worsen over time (GOLD 2023).

Since the late 1960s, the definition of COPD has repeatedly been modified. Early definitions of COPD included chronic bronchitis, which is clinically characterised by chronic cough, and emphysema, which is pathologically defined by damaged sacs or alveoli in the lungs (Burrows 1966). In 2001, the Global Initiative for Chronic Obstructive Lung Disease (GOLD) released its first report, Global Strategy for the Diagnosis, Management, and Prevention of COPD (Pauwels 2001), which supported the definition of COPD, indicating that the disorder is recognised primarily by chronic obstruction of lung airflow (Pauwels 2001).

If COPD is properly diagnosed and managed, symptoms can be ameliorated. Smoking cessation, smoke‐free cooking, vaccination, and an active lifestyle are the first steps in COPD management, and daily pharmacological treatment is required for most people with symptomatic stable COPD (GOLD 2023).

Description of the intervention

While asymptomatic individuals with mild airflow limitations can be treated with on‐demand short‐acting bronchodilators, key medications for symptomatic COPD management consist of three classes of inhaled medication: long‐acting beta‐agonists (LABAs), long‐acting muscarinic antagonists (LAMAs), and inhaled corticosteroids (ICSs) (GOLD 2023). If the disease cannot be adequately controlled with LAMA or LABA monotherapy, administration of two or more medications from different classes may prove beneficial. When two classes of medication are required, a LAMA plus a LABA (LAMA+LABA) or a LABA plus an ICS (LABA+ICS) are often selected because these combinations can be administered via a single medication device (Frampton 2014; Malerba 2014; Nannini 2013; Schachter 2013), which is most beneficial for improving patient adherence (Horita 2015a).

How the intervention might work

Currently, there are no medications that can cure COPD. Thus, the practical goal of COPD treatment is to control the symptoms, reduce the frequency of exacerbations, and improve exercise tolerance and quality of life. COPD treatment usually consists of smoking cessation, vaccination, inhaled bronchodilators, ICSs, oral medication, long‐term oxygen therapy, and pulmonary rehabilitation (GOLD 2023). According to the GOLD approach, people are classified into three categories depending on the degree of symptoms and the risk of exacerbations (GOLD 2023). Medications belonging to a specific class are recommended based on the following criteria:

Category A (low symptoms plus few exacerbations): bronchodilator (short‐ or long‐acting); consider switching to another depending on response;
Category B (high symptoms plus few exacerbations): LAMA+LABA;
Category E (high symptoms plus many exacerbations): LAMA+LABA; LAMA+LABA+ICS is considered if blood eosinophil count is high.

Long‐acting muscarinic antagonists

LAMAs dilate the airway by selectively blocking acetylcholine M₃ receptors and inhibiting bronchoconstriction (Alagha 2014). Since the early 2000s, LAMAs, especially tiotropium, have been regarded as the first‐choice medication for treating COPD. LAMAs confer anti‐inflammatory effects and, more importantly, anti‐airway remodelling effects (Tashkin 2004).

Long‐acting beta‐agonists

LABAs widen the airway by relaxing the airway muscles. Studies have suggested that LABAs may also provide anti‐inflammatory and protective effects against bronchoconstrictive substances. Regular use of a short‐acting beta‐agonist that works quickly and lasts for four to six hours is not currently recommended for people with asthma or COPD. A LABA that lasts for approximately 12 to 24 hours is considered a maintenance medication (Anderson 2014; Tashkin 2004).

Inhaled corticosteroids

ICSs reduce airway inflammation. Although ICSs are indicated for bronchial asthma, in which eosinophils play a key role, they are not as effective when neutrophils are observed in the airways of people with COPD (Barnes 2010; Hanania 2008; Suissa 2009). The previous GOLD report recommended that ICSs be prescribed in combination with LABA for people with COPD with severe airflow limitation or with a high risk of exacerbations (GOLD 2016). Studies have suggested that LABA+ICS may be highly effective for people with a high sputum/blood eosinophil count (Pascoe 2015).

Why it is important to do this review

The previous GOLD report recommended the first‐line use of ICSs only for people with severe to very severe airflow limitation and two or more exacerbations per year, with one or more hospitalisations for exacerbations (GOLD 2016). That report suggested that ICSs reduce the risk of exacerbations (GOLD 2016). Nonetheless, prescription rates for ICSs and combined LABA+ICS agents are high (Drivenes 2014; White 2013). This is probably because many randomised controlled trials (RCTs) have supported the hypothesis that the salmeterol (LABA) plus fluticasone propionate (ICS) combination – the oldest combination treatment – can improve quality of life, especially for people with dyspnoea, and can also decrease acute exacerbations of COPD and reduce yearly declines in pulmonary function (GOLD 2016). Blood and sputum eosinophil counts can serve as predictive biomarkers for differentiating between people with COPD who will derive the greatest benefit from ICS administration and people who will not benefit from an ICS (Pascoe 2015). The GOLD 2023 report now recommends LAMA+LABA+ICS only for people with category E COPD who have a high eosinophil count, and recommends that ICSs should not be over‐prescribed in COPD cases.

Understanding which type of combination treatment (LAMA+LABA or LABA+ICS) is most beneficial for people with COPD, regardless of eosinophil count, is important, and this systematic review contributes to answering this question.

Objectives

To compare the benefits and harms of LAMA+LABA versus LABA+ICS for treatment of people with stable COPD.

Methods

Criteria for considering studies for this review

Types of studies

We planned to include individually‐randomised and cluster‐randomised controlled trials and cross‐over trials, but not quasi‐RCTs. We included studies reported as full text, those published as abstract only, and unpublished data. When we could not obtain sufficient data from published articles, we contacted authors and sponsors, and accessed trial registration websites. We included open‐label, single‐blinded, and double‐blinded studies. The minimum accepted trial duration was one month.

Types of participants

We included adults with a diagnosis of COPD according to GOLD guidelines (GOLD 2023). We did not set specific exclusion criteria involving comorbidities. We planned to exclude original studies focusing on asthma‐COPD overlap syndrome (ACOS).

Types of interventions

We included trials comparing LAMA+LABA versus LABA+ICS. We permitted treatments administered via a single combined device or via two separate devices. We excluded trials of short‐acting bronchodilators (e.g. ipratropium). We included co‐interventions when they were not part of the randomly assigned treatment.

Types of outcome measures

Primary outcomes

Exacerbations (participants with one or more)
Serious adverse events (participants with one or more)
Quality of life, as measured by the St. George's Respiratory Questionnaire (SGRQ) total score change from baseline
Trough forced expiratory volume in one second (FEV₁) change from baseline

Secondary outcomes

Pneumonia (participants with one or more occurrences); assessed based on chest x‐ray
All‐cause death
SGRQ total score change from baseline (4 points or greater)
Hospitalisations for COPD exacerbations (participants with one or more occurrences)

Search methods for identification of studies

Electronic searches

We identified trials from the Cochrane Airways Group Specialised Register (CAGR), which is maintained by the Information Specialist for the Group. The Register contains trial reports identified through systematic searches of bibliographic databases, including the Cochrane Central Register of Controlled Trials (CENTRAL), MEDLINE, Embase, the Cumulative Index to Nursing and Allied Health Literature (CINAHL), the Allied and Complementary Medicine Database (AMED), and PsycINFO, and through handsearching of respiratory journals and conference abstracts (see Appendix 1 for details). We searched all records in the CAGR from the date it was last searched for the previous version of this review (2 February 2016) to 10 September 2022, using the search strategy provided in Appendix 2.

We conducted a search of ClinicalTrials.gov (www.ClinicalTrials.gov) and the World Health Organization (WHO) Clinical Trials Search Portal (www.who.int/ictrp/en/), from the date they were last searched (4 June 2016) to 10 September 2022 (Appendix 3; Appendix 4).

We searched all databases from their inception, and we imposed no restrictions on language of publication.

Searching other resources

We checked reference lists of all primary studies and review articles for additional references, and we searched relevant manufacturers' websites for trial information. We searched for errata or retractions from included studies published as full text on PubMed (www.ncbi.nlm.nih.gov/pubmed). Handsearches were done up to 1 November 2021.

Data collection and analysis

Selection of studies

We used Cochrane’s Screen4Me workflow to help assess the search results. Screen4Me comprises three components: (1) known assessments – a service that matches records in the search results to records that have already been screened in Cochrane Crowd and labelled as an RCT or 'Not an RCT'; (2) the RCT classifier – a machine learning model that distinguishes RCTs from non‐RCTs; and (3) if appropriate, Cochrane Crowd – Cochrane’s citizen science platform where the Crowd helps to identify and describe health evidence. For more information about Screen4Me and the evaluations that have been done, please visit the Screen4Me webpage on the Cochrane Information Specialists' portal: community.cochrane.org/organizational-info/resources/resources-groups/information-specialists-portal.

Two review authors (NF and NH) independently screened the titles and abstracts of all studies identified by the search for possible inclusion, and coded studies as 'retrieve' (eligible or potentially eligible/unclear) or 'do not retrieve'. We retrieved the full‐text publications. Two review authors (NF and NH) independently screened the full texts to identify studies for inclusion and recorded reasons for exclusion of ineligible studies. We resolved disagreements through discussion, or, when required, we consulted a third review author (TK). We identified and excluded duplicates and collated multiple reports of the same study, so that each study, rather than each report, was the unit of interest in the review. We recorded the selection process in sufficient detail to complete a PRISMA (Preferred Reporting Items for Systematic Reviews and Meta‐Analyses) flow diagram and a Characteristics of excluded studies table (Moher 2009).

Data extraction and management

We used a data collection form that had been piloted on at least one study in the review to document study characteristics and outcome data. Two review authors (NF and NH) extracted the following study characteristics from the included studies.

Methods: study design, duration of study follow‐up and 'run‐in' period, number of study centres and countries, and study start date.
Participants: number, mean and standard deviation (SD) age, gender, mean and SD of baseline FEV₁ key inclusion criteria, number of participants randomised and completed, and follow‐up duration.
Interventions: intervention, comparison, and dosage of the intervention.
Outcomes: primary outcomes specified and collected and time points reported.
Notes: funding for trial and notable conflicts of interest (COI) of trial authors, trial registration, and other information if necessary.

Two review authors (NF and NH) independently extracted outcome data from the included studies. We noted in the Characteristics of included studies table if outcome data were not reported in a useable way. We resolved disagreements by consensus or by consultation with a third review author (TK). One review author (NF) transferred data into Review Manager 5 (Review Manager 2014). We double‐checked that data were entered correctly by comparing data presented in the systematic review versus data provided in study reports. A second review author (NH) spot‐checked study characteristics for accuracy against the trial report.

Assessment of risk of bias in included studies

Two review authors (NF and NH) independently assessed risk of bias for each study using the criteria outlined in the Cochrane Handbook for Systematic Reviews of Interventions (Higgins 2011). We resolved disagreements by discussion or by consultation with another review author (EO). We assessed risk of bias according to 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.

We graded each potential source of bias as high, low, or unclear and provided an explanation from the study report together with a justification for our judgement in the risk of bias table. We summarised risk of bias judgements across studies for each of the domains listed. We classified outcomes as subjective and objective to consider different risks of performance and selection bias for different blinding methods. All‐cause death is the only outcome of interest to this review that may be considered truly objective and the SGRQ is the only subjective, self‐reported measure. The other outcomes (exacerbations, serious adverse events, FEV₁ and pneumonia) might be considered semi‐objective: they involve a degree of patient reporting, behaviour, or clinician judgement that could introduce bias (intentionally or otherwise) if the randomised treatment had been inadvertently detected through side effects or other means. For example, this could happen in the grading of an adverse event as serious or non‐serious, effort given when measuring FEV₁, or the interpretation of a chest X‐ray to determine pneumonia. When we requested information on risk of bias related to unpublished data or corresponded with a trialist, we noted this in the risk of bias table.

When considering treatment effects, we took into account risk of bias for studies that contributed to that outcome.

Assessment of bias in conducting the systematic review

We conducted the review according to the previously published protocol (Horita 2016).

Measures of treatment effect

We analysed dichotomous data as odds ratios (ORs), and continuous data as MDs with 95% confidence intervals (CIs). We entered data presented as a scale with a consistent direction of effect (i.e. data in the LAMA+LABA arm minus data in the LABA+ICS arm). Although there is no universal rule to interpret the magnitude of the therapeutic effect from ORs, we believe that an OR greater than 1.5 and an OR of less than 0.7 mean that there is a considerable chance that the outcome is clinically important.

We undertook meta‐analyses only when this was meaningful (i.e. if treatments, participants, and the underlying clinical question were similar enough for pooling to make sense).

We had planned to describe skewed data using medians and interquartile ranges; however, we found no report describing skewed data.

According to the original protocol, when multiple trial arms were reported in a single trial, we included only the relevant arms.

Unit of analysis issues

We analysed the number of participants, not the number of events, as the unit of analysis for dichotomous data (i.e. participants with one or more events). For continuous data, we used MDs.

Dealing with missing data

We tried to contact investigators, study sponsors, and registration websites to verify key study characteristics and to obtain missing numerical outcome data when possible (e.g. when a study was only reported in an abstract format). When this was not possible, and when missing data were thought to introduce serious bias, we explored the impact of including such studies in the overall assessment of results by conducting a sensitivity analysis.

Assessment of heterogeneity

We used the I² statistic to measure heterogeneity amongst the trials in each analysis: 0% to 40%: might not be important; 30% to 60%: might represent moderate heterogeneity; 50% to 90%: might represent substantial heterogeneity; 75% to 100%: might show considerable heterogeneity (Higgins 2011). When we identified considerable heterogeneity, we reported this and explored possible causes by performing a prespecified subgroup analysis.

Assessment of reporting biases

We created and examined a funnel plot to explore possible small‐study and publication biases for the primary outcomes.

Data synthesis

We used a random‐effects model and performed a sensitivity analysis by using a fixed‐effect model (see Sensitivity analysis).

Subgroup analysis and investigation of heterogeneity

We planned the following subgroup analyses for all primary and secondary outcomes:

LAMA+LABA: 'combined indacaterol + glycopyrronium bromide (IND/GLY)' versus 'combined umeclidinium + vilanterol (UMEC/VI)' versus 'other LAMA/LABA inhalers'.
COPD severity: 'including only mild or moderate (or both) (% predicted FEV₁ 50% or greater)' versus 'including severe and/or very severe (% predicted FEV₁ less than 50%)' versus 'including both categories.'

We were unable to perform the COPD severity subgroup analysis because separate data for participants with different severities were not reported.

We used the I² test to detect heterogeneity, as discussed in Higgins 2003. We used the formal test for subgroup differences provided in Review Manager 5 (Review Manager 2014).

Sensitivity analysis

We conducted sensitivity analyses using a fixed‐effect model for the primary outcomes and reported any important differences compared with results using the random‐effects model.

Summary of findings and assessment of the certainty of the evidence

We created a summary of findings table to present the following outcomes:

exacerbations (participants with one or more);
serious adverse events (participants with one or more);
quality of life, as measured by the St. George's Respiratory Questionnaire (SGRQ) total score change from baseline;
trough FEV₁ change from baseline;
pneumonia (participants with one or more occurrences);
all‐cause death;
SGRQ total score change from baseline (4 points or greater).

We used the five GRADE considerations (study limitations, consistency of effect, imprecision, indirectness, and publication bias) to assess the certainty of a body of evidence (very low, low, moderate, and high certainty of evidence) as it related to studies that contributed data to meta‐analyses for prespecified outcomes (Guyatt 2008). We used the methods and recommendations described in Section 8.5 and Chapter 12 of the Cochrane Handbook for Systematic Reviews of Interventions (Higgins 2011), with GRADEpro software (GRADEpro). We justified all decisions to downgrade or upgrade the certainty of the evidence by using footnotes, and we provided comments to aid readers' understanding of the review when necessary.

We did not prespecify specific domains or percentages for downgrading due to risk of bias, but considered it reasonable to downgrade if studies contributing more than half the analysis weight were at high risk of bias in the 'other bias' domain and at least one other domain. For inconsistency, we downgraded if I² was greater than 50%. For imprecision, we based downgrade decisions on prespecified thresholds (clinical importance 0.7 to 1.5 for outcomes as odds ratios) or established minimal clinically important differences (MCIDs) (4 points for SGRQ).

Results

Description of studies

See Included studies (Table 1), Excluded studies, and Ongoing studies.

Table 1. Summary of characteristics of included studies

Study	LAMA+LABA	LABA+ICS	Key inclusion criteria	Follow‐up duration (weeks)	Mean/median age (years)	Number randomised
Beeh 2016	Tiotropium/olodaterol (2.5/5 μg) or tiotropium/olodaterol (5/5 μg)	Salmeterol/fluticasone (50/250 μg) twice daily or salmeterol/fluticasone (50/500 μg) twice daily	%pred FEV₁ 30% to 80% Ex(‐)	6 × 4 time periods (cross‐over)	64	229
Donohue 2015a	Umeclidinium/vilanterol (62.5/25 μg)	Salmeterol/fluticasone (50/250 μg) twice daily	%pred FEV₁ 30% to 70%, mMRC ≥ 2, Ex(‐)	12	63	707
Donohue 2015b	Umeclidinium/vilanterol (62.5/25 μg)	Salmeterol/fluticasone (50/250 μg) twice daily	%pred FEV₁ 30% to 70%, mMRC ≥ 2, Ex(‐)	12	64	700
Ferguson 2018	Glycopyrronium/formoterol fumarate (18/9.6 μg)	Formoterol/budesonide/ fumarate (160/4.8 μg) twice daily	CAT ≥ 10, %pred FEV₁ 25% to 80%, not required to have exacerbation	24	65	943
Frith 2018	Glycopyrronium/indacaterol (50/110 μg)	Salmeterol/fluticasone (50/500 μg) twice daily	CAT ≥ 10, %pred FEV1 30% to 80%, Ex(+)	12	65	502
Herth 2020	Tiotropium/olodaterol (5/5 μg)	Salmeterol/fluticasone (50/500 μg)	%pred FEV₁ < 70%, Ex(‐)	6	62	76
Hoshino 2015	Tiotropium/indacaterol (18/150 μg)	Salmeterol/fluticasone (50/250 μg) twice daily	%pred FEV₁ 30% to 80%, Ex(‐)	16	71	46
Lipson 2018	Umeclidinium/vilanterol (62.5/25 μg)	Vilanterol/fluticasone furoate (25/100 μg)	CAT ≥ 10, %pred FEV₁ < 80%, Ex(+)	52	65	6204
Magnussen 2012	Tiotropium/salmeterol (18/50 μg) twice daily	Salmeterol/fluticasone (50/500 μg) twice daily	%pred FEV₁ ≤ 65%, Ex(‐)	8 x 2 time periods (cross‐over)	61	344
Mostafa 2021	Tiotropium (18 μg) once daily/formoterol (9 μg) twice daily	Formoterol/budesonide (160/4.5 μg) twice daily	%pred FEV₁ 30% to 80%, Ex(‐)	12	64	40
NCT03240575	Tiotropium (5μg)/olodaterol (5μg) once daily	Salmeterol/fluticasone (50μg/250μg) twice daily	%pred FEV₁ 30% to 80%, Ex(‐)	12	64	302
Rabe 2008	Tiotropium/formoterol (18/24 μg) twice daily	Salmeterol/fluticasone (50/500 μg) twice daily	%pred FEV₁ ≤ 65%, Ex(‐)	6	62	605
Rabe 2020	Glycopyrronium/formoterol (9/4.8 µg) twice daily.	Formoterol/budesonide (4.8/160 µg) twice daily.	%pred FEV₁ 25‐65%, CAT ≥ 10, Ex(+)	52	65	4294
Singh 2015	Umeclidinium/vilanterol (62.5/25 μg)	Salmeterol/fluticasone (50/500 μg) twice daily	%pred FEV₁ 30% to 70%, mMRC ≥ 2, Ex(‐)	12	62	717
Vogelmeier 2013	Indacaterol/glycopyrronium bromide (110/50 μg)	Salmeterol/fluticasone (50/500 μg) twice daily	%pred FEV₁ 30% to 80%, Ex(‐)	26	63	523
Vogelmeier 2016	Aclidinium/formoterol (400/12 μg) twice daily	Salmeterol/fluticasone (50/500 μg) twice daily	%pred FEV₁ < 80%, CAT ≥ 10, Ex(‐)	24	63	933
Vogelmeier 2017	Glycopyrronium/Indacaterol (50/110 μg)	any	%pred FEV₁ 50% ‐80%, mMRC ≥ 1, Ex(+)	12	65	1083
Wedzicha 2016	Indacaterol/glycopyrronium bromide (110/50 μg)	Salmeterol/fluticasone (50/500 μg) twice daily	%pred FEV₁ 25% to 60%, mMRC ≥ 2, Ex(+)	52	65	3362
Zhong 2015	Indacaterol/glycopyrronium bromide (110/50 μg)	Salmeterol/fluticasone (50/500 μg) twice daily	%pred FEV₁ 30% to 80%, mMRC ≥ 2, Ex(‐)	26	65	744

%pred FEV₁: % predicted forced expiratory volume in one second; CAT: chronic obstructive pulmonary disease assessment test; Ex(‐): without recent exacerbation; Ex(+): with recent exacerbation; LABA: long‐acting beta‐agonist; LAMA: long‐acting muscarinic antagonist; mMRC: modified Medical Research Council dyspnoea scale

Results of the search

We identified a total of 2685 records from update searches of electronic databases conducted on 1 November 2021 and 10 September 2022, including ClinicalTrials.gov and the WHO Clinical Trials Search Portal. We found an additional 20 records through manual searches. After removing 1248 duplicate records, there were 1458 records eligible for screening. The Screen4Me workflow removed 376 records, and we excluded 1046 records after viewing titles and abstracts alone. We excluded a further 21 after viewing full texts, which comprised 17 new excluded studies and four records identified as duplicates of records that had already been included. The remaining 15 records met the inclusion criteria for the review, corresponding to eight new included studies and one ongoing study.

Thus, we have included a total of 19 studies in the quantitative synthesis in the review (11 studies in the previous version of the review plus eight new included studies; see Figure 1).

Figure 1

PRISMA study flow diagram

Included studies

The 19 studies included 22,354 participants. Seventeen studies used a parallel‐group design and two used a cross‐over design (Magnussen 2012; Vogelmeier 2017), with an 84‐day washout period between treatments. All studies delivered treatment in a double‐blind manner, except two open‐label studies (Hoshino 2015; Vogelmeier 2017).

The number of participants included in each study ranged from 40 to 6204, with a median of 700 participants per study. Thirteen studies included participants with moderate to severe COPD without recent exacerbation. Four trials that included only participants with a recent exacerbation accounted for 65% of the total participants. In each study, between 54% and 91% (median 70%) of participants were males. The mean age of the participants in each study ranged from 61 to 71 years (median 64 years). The percent predicted (%pred) FEV₁ in each study ranged from 43% to 64%, with a median of 51.5%.

Treatment

Treatment duration ranged from six to 52 weeks. Of the LABA+ICS treatments used in these studies, 14 studies used combined salmeterol/fluticasone propionate (Beeh 2016; Donohue 2015a; Donohue 2015b; Frith 2018; Herth 2020; Hoshino 2015; Magnussen 2012; NCT03240575; Rabe 2008; Singh 2015; Vogelmeier 2013; Vogelmeier 2016; Wedzicha 2016; Zhong 2015), three studies used combined formoterol/budesonide (Ferguson 2018; Mostafa 2021; Rabe 2020), and one study used combined vilanterol/fluticasone furoate (Lipson 2018). One study permitted any type or dose of LABA+ICS (Vogelmeier 2017). Of the administered LAMA+LABA treatments, five studies used glycopyrronium/indacaterol (Frith 2018; Vogelmeier 2013; Vogelmeier 2017; Wedzicha 2016; Zhong 2015), four studies used umeclidinium/vilanterol (Donohue 2015a; Donohue 2015b; Lipson 2018; Singh 2015), two studies used tiotropium/formoterol (Rabe 2008; Mostafa 2021), three studies used tiotropium/olodaterol (Beeh 2016; Herth 2020; NCT03240575), two studies used glycopyrronium/formoterol (Ferguson 2018; Rabe 2020), one study used tiotropium/indacaterol (Hoshino 2015), one study used tiotropium/salmeterol (Magnussen 2012), and one study used aclidinium/formoterol (Vogelmeier 2016).

Outcomes

The reporting of outcomes across studies, organised by the type of LAMA+LABA combination investigated, is shown in Table 2. Thirteen studies reported exacerbations in a way that could be brought together in a meta‐analysis, 17 reported serious adverse events, 13 studies reported FEV₁, 14 reported pneumonia, and 14 reported all‐cause death. Quality of life, as measured by the St. George's Respiratory Questionnaire (SGRQ), was less consistently reported, with nine studies reporting mean total scores and only four reporting the number of participants with a 4‐point or higher improvement on the scale. None of the included studies measured hospitalisations for COPD exacerbations.

See: Summary of findings 1 Long‐acting muscarinic antagonist (LAMA) plus long‐acting beta‐agonist (LABA) versus LABA plus inhaled corticosteroid (ICS) for stable chronic obstructive pulmonary disease (COPD)

Table 2. ORBIT matrix of outcome reporting across included studies

	Objective or semi‐objective outcomes (investigator‐assessed)					Subjective outcomes (patient‐reported)
	Exacerbations	SAEs	Trough FEV₁	Pneumonia	All‐cause death	SGRQ (mean)	SGRQ ≥ 4 unit benefit
Indacaterol/glycopyrronium studies
Frith 2018	X	X	X	X	X
Vogelmeier 2013	X	X	X	X	X	X	X
Vogelmeier 2017	X	X	X	X	X
Wedzicha 2016	X	X	X	X	X	X	X
Zhong 2015	X	X		X	X
Umeclidinium/vilanterol studies
Donohue 2015a	X	X	X	X	X	X
Donohue 2015b	X	X	X	X	X	X
Lipson 2018	X	X	X	X	X	X	X
Singh 2015	X	X	X	X	X	X
Other LAMA/LABA inhaler studies
Beeh 2016		X	X	X	X
Ferguson 2018		X	X	X	X	X
Herth 2020			X
Hoshino 2015						X
Magnussen 2012	X	X
Mostafa 2021		X			X
NCT03240575		X	X	X	X
Rabe 2008	X	X
Rabe 2020	X	X		X	X	X	X
Vogelmeier 2016	X	X		X

Outcomes shown are those included in the summary of findings table. Outcomes are classed as objective or subjective for the purposes of risk of bias assessment to consider differences in performance and detection bias.

FEV₁: forced expiratory volume in one second; ORBIT: outcome reporting bias in trials; SAEs: serious adverse events; SGRQ: St George’s Respiratory Questionnaire

Excluded studies

We excluded 24 studies after viewing full texts (see Characteristics of excluded studies and Figure 1), of which 17 were new excluded studies in this update. Overall, 11 were not relevant because they did not compare LAMA+LABA versus LABA+ICS (Bruhn 2003; Calverley 2007; Knobil 2004a; Knobil 2004b; NCT00120978; Sciurba 2004; NCT03504527; NCT04320342; NCT04923347; NCT05097014; Papi 2018), three studies were not eligible because they were systematic reviews (Aziz 2018; Oba 2016; Pavord 2016), four studies were not eligible because they did not report an RCT (Anzueto 2017; Mahler 2016; NCT03376295; NCT04138758), two studies were not eligible because they reported cost‐effectiveness analyses (Price 2014; Skoupa 2018), two studies were letters or commentaries (Michael 2016; Singh 2019), one study was not eligible because it reported only patients with eosinophilic inflammation (UMIN000024905), and one study was not eligible because the data were only analysed for asthma‐COPD overlap (EUCTR2015‐002046‐31‐ES).

Ongoing studies

We found one ongoing study awaiting results (EUCTR2016‐004473‐41‐GB). The study, which started in 2016, is a trial comparing tiotropium/olodaterol with vilanterol/fluticasone furoate, sponsored by Mundipharma Research Limited (see Characteristics of ongoing studies). The primary end point is change from baseline in eosinophil count and one of the secondary end points is change from baseline in pre‐dose FEV₁.

Risk of bias in included studies

Included studies had generally low risk of bias for random sequence generation, allocation concealment, blinding of participants and personnel, blinding of outcome assessment, incomplete outcome data, and selective outcome reporting. The potential for performance and detection bias for different types of outcome (e.g. subjective, objective) is discussed under the Blinding heading. See Figure 2.

Figure 2

Risk of bias summary: included studies had generally low risk of bias for random sequence generation, allocation concealment, blinding of participants and personnel, blinding of outcome assessment, incomplete outcome data, and selective outcome reporting.

Allocation

Nine studies reported methods of random sequence generation that suggested a low risk of bias (Donohue 2015a; Donohue 2015b; Frith 2018; Magnussen 2012; NCT03240575; Singh 2015; Vogelmeier 2013; Wedzicha 2016; Zhong 2015). Ten studies had an unclear risk of bias in this domain because no details were reported (Beeh 2016; Ferguson 2018; Herth 2020; Hoshino 2015; Lipson 2018; Mostafa 2021; Rabe 2008; Rabe 2020 Vogelmeier 2016; Vogelmeier 2017). We rated 10 studies at low risk of bias relating to allocation concealment (Donohue 2015a; Donohue 2015b; Frith 2018; Magnussen 2012; Mostafa 2021; NCT03240575; Singh 2015; Vogelmeier 2013; Wedzicha 2016; Zhong 2015), two at high risk (Ferguson 2018; Herth 2020), and seven as unclear risk of bias because detailed methods were not provided (Beeh 2016; Hoshino 2015; Lipson 2018; Rabe 2008; Rabe 2020; Vogelmeier 2016; Vogelmeier 2017).

Blinding

We considered the impact of performance and detection bias on subjective and objective outcomes depending on the blinding methods used within studies. The reporting of objective, semi‐objective, and subjective outcomes across studies is shown in Table 2. Though bias is still possible in double‐blind studies for subjective and semi‐objective outcomes where blinding is broken unintentionally, there was no evidence that this had happened. Therefore, we rated all 17 double‐blind studies as low risk of performance and detection bias (Beeh 2016; Donohue 2015a; Donohue 2015b; Frith 2018; Ferguson 2018; Herth 2020; Lipson 2018; Magnussen 2012; Mostafa 2021; NCT03240575; Rabe 2008; Rabe 2020; Singh 2015; Vogelmeier 2013; Vogelmeier 2016; Wedzicha 2016;Zhong 2015).

Two studies adopted neither double‐ nor single‐blinding methods and we rated these high risk of bias for both blinding domains, because there is a risk of bias for all outcomes except all‐cause death (Hoshino 2015; Vogelmeier 2017).

Incomplete outcome data

Our prespecified criteria for high attrition bias was a dropout rate of more than 20% of randomised participants. Two trials had a dropout rate of more than 20% (Mostafa 2021; Rabe 2020), and we therefore rated them high risk of bias. We considered all other studies to be at low risk of attrition bias.

Selective reporting

Our criteria for rating a study as having a risk of selective reporting bias was if it was a non‐registered trial or if it considerably deviated from the registered protocol concerning outcome reporting. Two trials had a high risk of selective reporting bias due to non‐registration (Hoshino 2015; Mostafa 2021).

Other potential sources of bias

Seventeen out of 19 trials were sponsored by pharmaceutical companies (Beeh 2016; Donohue 2015a; Donohue 2015b; Frith 2018; Ferguson 2018; Herth 2020; Lipson 2018; Magnussen 2012; NCT03240575; Rabe 2008; Rabe 2020; Singh 2015; Vogelmeier 2013; Vogelmeier 2016; Vogelmeier 2017; Wedzicha 2016; Zhong 2015). Although pharmaceutical company sponsorship does not automatically signify a high risk of bias, scrutiny of disclosure statements revealed that all 17 studies had authors that were employed by the company, held shares, and/or received grants or fees for the work. In all cases, these authors were involved explicitly in the design of the study, its conduct (including data analysis), or interpretation and write up, which warranted high risk of bias judgements.

We found no other sources of bias apart from conflicts of interest.

Effects of interventions

See summary of findings Table 1.

Primary outcomes

Exacerbations (participants with one or more event)

Thirteen studies with 20,960 participants evaluated exacerbations over six to 52 weeks of observation. Participants randomised to LAMA+LABA had similar odds of experiencing one or more exacerbations to those randomised to LABA+ICS (OR 0.91, 95% CI 0.78 to 1.06; P = 0.22, I² = 56%; moderate‐certainty evidence; Analysis 1.1). Our confidence in the result was reduced from high to moderate due to considerable heterogeneity.

The test for subgroup differences indicated a difference between LAMA+LABA subgroups, with the largest benefit of LAMA+LABA versus LABA+ICS seen in the glycopyrronium/indacaterol studies (OR 0.72, 95% CI 0.63 to 0.83; P = 0.02, I² = 0%) and the pooled effect for the umeclidinium/vilanterol subgroup laying in the opposite direction (OR 1.20, 95% CI 1.03 to 1.40; P < 0.001, I² = 0%; test for subgroup differences P < 0.001). However, results are observational and should be interpreted with caution. The overall pooled result was similar to the random‐effects analysis when a fixed‐effect model was tested (OR 0.94, 95% CI 0.87 to 1.01; P = 0.08, I² = 56%; Analysis 1.8).

Serious adverse events (participants with one or more event)

Eighteen studies with 23,183 participants evaluated serious adverse events over six to 52 weeks of observation. We discarded data from one study because no serious adverse events were reported in either arm (Herth 2020). Participants randomised to LAMA+LABA had similar odds of having a serious adverse event as those randomised to LABA+ICS (OR 1.02, 95% CI 0.91 to 1.15; I² = 20%; high‐certainty evidence; Analysis 1.2; test for subgroup differences P = 0.25). We did not downgrade the evidence in any of the GRADE domains and therefore have high confidence in the result.

Results were similar to the random‐effects analysis when a fixed‐effect model was tested (OR 1.03, 95% CI 0.95 to 1.12; I² = 20%; Analysis 1.9).

Quality of life, as measured by the St. George's Respiratory Questionnaire (SGRQ) total score change from baseline

Nine studies with 14,437 participants assessed mean change in SGRQ scores over 12 to 52 weeks of observation. The benefit of LAMA+LABA versus LAMA+ICS was neither statistically nor clinically important (MD ‐0.57, 95% CI ‐1.36 to 0.21; I² = 78%; moderate‐certainty evidence; Analysis 1.3; test for subgroup differences P < 0.004). Our confidence in the result was reduced from high to moderate due to considerable heterogeneity.

The test for subgroup differences indicated a difference between LAMA+LABA subgroups, with the largest benefit of LAMA+LABA versus LABA+ICS seen in the indacaterol/glycopyrronium studies (MD ‐1.29, 95% CI ‐2.08 to ‐0.50; P = 0.001, I² = 0%). The sensitivity analysis using a fixed‐effect model brought the point estimate from lying in favour of LAMA+LABA to zero (i.e. no effect) (MD 0.00, 95% CI ‐0.02 to 0.02; I² = 78%; Analysis 1.10).

Trough FEV₁ mean change

In total, 12 studies with 14,681 participants reported change in trough FEV₁ over six to 52 weeks. Compared to LABA+ICS, there was a significant increase in the trough FEV₁ change from baseline with LAMA+LABA (MD 0.07 L, 95% CI 0.05 to 0.08; I² = 73%; moderate‐certainty evidence; Analysis 1.4). This difference was less than the minimal clinically important difference (Donohue 2005). Our confidence in the result was reduced from high to moderate due to considerable heterogeneity.

In the LAMA+LABA subgroup analysis, each subgroup was consistently associated with an increase in trough FEV₁ change from baseline (test for subgroup differences P = 0.93). Results from the sensitivity analysis using a fixed‐effect model were in line with the main random‐effects analysis (MD 0.06, 95% CI 0.05 to 0.07; I² = 73%; Analysis 1.11).

Secondary outcomes

Pneumonia (participants with one or more event)

Fourteen studies with 21,829 participants evaluated pneumonia over 12 to 52 weeks of observation. Compared to LABA+ICS, there was a large reduction in the number of participants experiencing one or more episodes of pneumonia with LAMA+LABA (OR 0.61, 95% CI 0.52 to 0.72; I² = 0%; high‐certainty evidence; Analysis 1.5; test for subgroup differences P = 0.90). We did not downgrade the evidence in any of the GRADE domains and therefore have high confidence in the result. Although it would be possible to calculate an absolute risk reduction, we decided not to do so, as the absolute effect size is highly dependent on the study duration.

All‐cause death

Fifteen studies with 21,510 participants evaluated all‐cause death at six to 52 weeks of observation. There was an increased risk of all‐cause death with LAMA+LABA compared with LABA+ICS (OR 1.35, 95% CI 1.05 to 1.75; I² = 0%; moderate‐certainty evidence; Analysis 1.6). Our confidence in the result was reduced from high to moderate due to imprecision in the effect.

Point estimates and confidence intervals varied across subgroups containing studies of different LAMA+LABA combinations, but the test for subgroup differences indicated the differences were not statistically significant (test for subgroup differences P = 0.51).

SGRQ improvement of 4 points or greater

Four studies with 13,614 participants evaluated the SGRQ total score change from baseline (≥ 4 points) at 26 to 52 weeks of observation. There was little to no significant change in the SGRQ total score (4 points or greater) between LAMA+LABA and LABA+ICS (OR 1.06, 95% CI 0.90 to 1.25; I² = 77%; moderate‐certainty evidence; Analysis 1.7; test for subgroup differences P = 0.002). Our confidence in the result was reduced from high to moderate due to considerable heterogeneity.

Hospitalisations for COPD exacerbations

None of the included studies reported this outcome.

Discussion

Summary of main results

We conducted a systematic review and meta‐analysis to compare the efficacy and safety of LAMA+LABA and LABA+ICS therapy in people with stable COPD. The review includes 19 studies with 22,354 participants, including 11 studies from the previous version of the review. Most studies were parallel, double‐blind RCTs. We noted selection, attrition, and selective reporting bias in a small number of studies and there is a risk of bias from involvement of authors employed by the pharmaceutical sponsors in 17 of the 19 studies.

Compared to LABA+ICS, the review found there to be a greater improvement in FEV₁ in the LAMA+LABA group (moderate‐certainty evidence), lower odds of pneumonia (high‐certainty evidence), and increased mortality (moderate‐certainty evidence). This update did not find a difference in exacerbations between LAMA+LABA and LABA+ICS (moderate‐certainty evidence), which was reported in the previous review (Horita 2017). The odds of having a serious adverse event are likely to be similar for LAMA+LABA and LABA+ICS (high‐certainty evidence), as were quality of life scores as measured by the St George's Respiratory Questionnaire (moderate‐certainty evidence).

Overall completeness and applicability of evidence

Most of the studies included in this analysis recruited people with moderate to severe COPD, according to the GOLD reports. Therefore, attention should be paid when applying our results to people with mild and very severe COPD.

Despite the apparent benefits of LAMA+LABA over LABA+ICS for trough FEV₁ and reduced risk of pneumonia, the findings suggest little difference for exacerbations, serious adverse events, and quality of life as measured by the SGRQ. Although exacerbations and pneumonia should share risk factors, a paradox can occur in COPD treated with ICS where the risk of exacerbations and mortality reduces despite an increased risk of pneumonia. The apparent conflicting results of this review may partially relate to this relationship, and should also be interpreted with caution due to the limited follow‐up and number of observed events. It should also be noted that studies recruited outpatient populations and cases of pneumonia may have been mild. Variation in study results also reduced our confidence in some of these findings, and in some cases, such as the exacerbations analysis where Lipson 2018 reported a significant benefit in favour of LABA+ICS, outlier studies impacted on the overall findings.

There was also variation across the included studies in their exclusion of people with a history of asthma, and the reporting of blood eosinophil counts that may be indicative of ACOS was patchy. Given that ICSs are effective in preventing exacerbations in asthma (Reddel 2021), results may have been skewed in favour of the LABA+ICS group because people with ACOS or a high blood eosinophil sensitivity to ICS may have fared better than those receiving LAMA+LABA.

Studies further varied in their inclusion criteria around recent history of exacerbation, which is a known risk factor for COPD mortality (Esteban 2018). The variation may have impacted on analyses where studies that enroled participants with a history of exacerbations carried a large proportion of the weight within an analysis (e.g. Rabe 2020 in the all‐cause death analysis). Moreover, the % predicted FEV₁ of participants in Rabe 2020 was as low as 25% to 65%, suggesting that participants' condition in this trial was poorer than those in other RCTs.

Limitations of the evidence include the lack of long‐term follow‐up data and the possibility that discontinuation of therapy may have contributed to early events such as withdrawal and death. For example, in Wedzicha 2016, all participants were treated with tiotropium during a one‐month run‐in period. Exacerbations in the LAMA+LABA groups might have occurred because participants who had been using ICSs prior to the trial had to discontinue them. As such, in studies where a recent exacerbation was not permitted, removal of these participants may have led to a biased population. In Lipson 2018, the participants who had been receiving ICSs before enroling in the trial were partly randomised to the LAMA+LABA group. These participants were actually stepping down in their treatment; this could lead to COPD exacerbations (Suissa 2018).

Certainty of the evidence

GRADE ratings of certainty in the evidence along with reasons for downgrading, or choosing not to downgrade, are summarised in summary of findings Table 1. We downgraded most outcomes once to moderate certainty, meaning the true effect is likely to be close to the estimate of effect but may be substantially different. The pooled effects for serious adverse events and pneumonia were both rated high certainty, meaning the true effect lies close to that of the estimate of effect. The most common reason for downgrading was considerable heterogeneity (exacerbations, SGRQ mean change, FEV₁ mean change and SGRQ improvement of 4 or more points), which may reflect the clinical heterogeneity between studies, discussed above. Only the all‐cause death analysis was downgraded due to imprecision, because there was a very small number of events in the analysis, leading to wide confidence intervals. The confidence intervals around the pooled result for several other outcomes include a potential benefit in either direction (exacerbations, serious adverse events, SGRQ mean change, SGRQ improvement of 4 or more points), but they lay within the threshold for clinical importance that was predefined in the protocol for this review, so we chose not to downgrade (0.7 to 1.5 for OR outcomes; 4‐point MCID for SGRQ).

We did not downgrade any outcomes due to indirectness of the contributing studies to the review question. However, variation in study inclusion criteria around recent history of exacerbation, baseline percentage predicted FEV₁, and exclusion of asthma are noteworthy, and affect how the pooled results can be interpreted and applied to different subpopulations of people with COPD. Similarly, we did not downgrade any outcomes for suspected publication bias, and there was no obvious asymmetry on funnel plots for the primary outcomes.

The potential impact of risk of bias across analyses is difficult to quantify, and we did not prespecify a threshold for downgrading the evidence due to risk of bias in certain domains or in studies contributing a certain percentage of the analysis weight. Amongst the 19 included studies, only two with a few participants did not have any commercial sponsorship (Hoshino 2015; Mostafa 2021). In the remaining 17 studies, study authors involved in key aspects of the study were employed by, held shares for, or received fees from the sponsoring manufacturer (Table 3). Otherwise, the included studies were mostly well‐designed and conducted, with adequate sample sizes. Therefore, across outcomes, all or almost all studies are at high risk of 'other' bias due to conflicts of interest, and studies contributing between 18.5% and 35.9% of the analysis weight are rated high risk of bias in at least one other domain. We set a post hoc rule for downgrading, and considered it insufficient to downgrade if studies contributing more than half the analysis weight were at low or unclear risk of bias in all but the 'other' domain, meaning no outcomes were downgraded for this reason. There is a need for large, well‐conducted trials funded by national funding bodies to overcome the bias inherent in evidence funded by pharmaceutical companies.

Table 3. Sponsor list for chronic obstructive pulmonary disease studies

Sponsor	Record count	% of 1723
GlaxoSmithKline	134	7.78
Novartis	128	7.43
AstraZeneca	122	7.08
Boehringer Ingelheim	113	6.56
Pfizer	84	4.88
Nycomed	49	2.84
GSK	45	2.61
Chiesi	41	2.38
Almirall	36	2.09
Merck	30	1.74

Web of Science Core Collection, advanced search for "TI=(COPD) AND TS=(inhal*)" without any restriction hit 1723 reports as of 13 June 2016. "Results analysis" > "Source Titles" output the table above.

Potential biases in the review process

Although we tried to extract data on exacerbations of any severity, some trials counted only moderate to severe exacerbations. In addition, some outcomes, such as exacerbation and adverse effects, were dependent on the threshold determined by the researchers of the original articles. Confidence intervals for the two cross‐over studies might be conservative in our analysis (Magnussen 2012; Vogelmeier 2017), due to the way we included the data in the analysis to avoid overestimating precision.

Agreements and disagreements with other studies or reviews

The previous version of this review, which included 11 studies with 9839 participants (Horita 2017), concluded there was no difference between the LAMA+LABA and LABA+ICS groups for all‐cause death (OR 1.01, 95% CI 0.61 to 1.67), whereas this update found a probable benefit of LABA+ICS over LAMA+LABA. As discussed above, this may be partially explained by differences in the presence and reporting of ACOS within the newer studies that carry a large proportion of the analysis weight.

A recent network meta‐analysis which included a LAMA+LABA versus LABA+ICS comparison looked at many of the same outcomes to this review in studies of low‐ and high‐risk COPD populations (Oba 2018). The review included many of the same studies (nine studies), and many others for comparisons not of interest to this review, but excluded those with a cross‐over design and those shorter than 12 weeks. Oba and colleagues also presented data separately for high‐ and low‐risk populations and looked at FEV₁ and the SGRQ at three, six, and 12 months. Results for the high‐risk population are largely in line with this review: the finding relating to exacerbations was similar to ours (OR 0.89, 95% CI 0.78 to 1.04) and quality of life, as measured by the SGRQ (MD ‐1.47, 95% CI ‐3.74 to 0.45). It is noteworthy that there was no significant difference in all‐cause mortality (OR 1.15, 95% CI 0.70 to 1.95). A similarly large effect was seen for pneumonia (OR 0.61, 95% CI 0.34 to 1.01) and trough FEV₁ (MD 0.05, 95% CI 0.03 to 0.07) in favour of LAMA+LABA, and effects were generally smaller or inconclusive in the low‐risk population studies (Oba 2018). A further review of incident pneumonia and mortality in people with COPD showed no difference between ICS and non‐ICS arms in pneumonia‐related mortality despite a significantly increased risk of pneumonia (Festic 2015).

Figure 1

PRISMA study flow diagram

Figure 2

Analysis 1.1

Comparison 1: Long‐acting muscarinic antagonist (LAMA) plus long‐acting beta‐agonist (LABA) versus LABA plus inhaled corticosteroid (ICS), Outcome 1: Exacerbations

Analysis 1.2

Comparison 1: Long‐acting muscarinic antagonist (LAMA) plus long‐acting beta‐agonist (LABA) versus LABA plus inhaled corticosteroid (ICS), Outcome 2: Serious adverse events

Analysis 1.3

Comparison 1: Long‐acting muscarinic antagonist (LAMA) plus long‐acting beta‐agonist (LABA) versus LABA plus inhaled corticosteroid (ICS), Outcome 3: Quality of life as measured by the St George's Respiratory Questionnaire (SGRQ) total score change from baseline

Analysis 1.4

Comparison 1: Long‐acting muscarinic antagonist (LAMA) plus long‐acting beta‐agonist (LABA) versus LABA plus inhaled corticosteroid (ICS), Outcome 4: Trough FEV₁ mean change (litres)

Analysis 1.5

Comparison 1: Long‐acting muscarinic antagonist (LAMA) plus long‐acting beta‐agonist (LABA) versus LABA plus inhaled corticosteroid (ICS), Outcome 5: Pneumonia

Analysis 1.6

Comparison 1: Long‐acting muscarinic antagonist (LAMA) plus long‐acting beta‐agonist (LABA) versus LABA plus inhaled corticosteroid (ICS), Outcome 6: All‐cause death

Analysis 1.7

Comparison 1: Long‐acting muscarinic antagonist (LAMA) plus long‐acting beta‐agonist (LABA) versus LABA plus inhaled corticosteroid (ICS), Outcome 7: SGRQ improvement ≥ 4 points

Analysis 1.8

Comparison 1: Long‐acting muscarinic antagonist (LAMA) plus long‐acting beta‐agonist (LABA) versus LABA plus inhaled corticosteroid (ICS), Outcome 8: Exacerbations (fixed‐effect sensitivity)

Analysis 1.9

Comparison 1: Long‐acting muscarinic antagonist (LAMA) plus long‐acting beta‐agonist (LABA) versus LABA plus inhaled corticosteroid (ICS), Outcome 9: Serious adverse events (fixed‐effect sensitivity)

Analysis 1.10

Comparison 1: Long‐acting muscarinic antagonist (LAMA) plus long‐acting beta‐agonist (LABA) versus LABA plus inhaled corticosteroid (ICS), Outcome 10: SGRQ mean change (fixed‐effect sensitivity)

Analysis 1.11

Comparison 1: Long‐acting muscarinic antagonist (LAMA) plus long‐acting beta‐agonist (LABA) versus LABA plus inhaled corticosteroid (ICS), Outcome 11: Trough FEV₁ mean change (fixed‐effect sensitivity)

LAMA plus LABA versus LABA plus ICS for stable COPD
Population: people with stable COPD Setting: outpatient. Studies were conducted in > 50 countries including low‐, medium‐, and high‐income countries from all continents Intervention: LAMA+LABA Comparison: LABA+ICS
Outcomes	*Anticipated absolute effects (95% CI)**		Relative effects (95% CI)	Number of participants (studies)	Certainty of the evidence (GRADE)	Comments
Outcomes	LABA+ICS	LAMA+LABA	Relative effects (95% CI)	Number of participants (studies)	Certainty of the evidence (GRADE)	Comments
Exacerbations (number of people experiencing ≥ 1 exacerbations) Follow‐up: 6 to 52 weeks	291 per 1000	272 per 1000 (243 to 303)	OR 0.91 (0.78 to 1.06)	20,960 (13 RCTs)	⊕⊕⊕⊝ Moderate^a‐e	Low OR means favourable outcome
Serious adverse events (number of people experiencing ≥ 1 SAEs) Follow‐up: 6 to 52 weeks	148 per 1000	150 per 1000 (136 to 166)	OR 1.02 (0.91 to 1.15)	23,183 (18 RCTs)	⊕⊕⊕⊕ High^b,c,d,e	Low OR means favourable outcome. Herth 2020 was discarded as no events were reported in either arm.
Quality of life as measured by SGRQ total score change from baseline (MD) Follow‐up: 12 to 52 weeks Scale 0 to 100	‐	‐	MD ‐0.57 (1.36 lower to 0.21 higher)	14,437 (9 RCTs)	⊕⊕⊕⊝ Moderate^a‐e	Low MD means favourable outcome
Trough FEV₁ change from baseline Follow‐up: 6 to 52 weeks	‐	‐	MD 0.07 L (0.05 to 0.08)	14,681 (12 RCTs)	⊕⊕⊕⊝ Moderate^a,b,d,e	High MD means favourable outcome
Pneumonia Follow‐up: 12 to 52 weeks	45 per 1000	28 per 1000 (24 to 33)	OR 0.61 (0.52 to 0.72)	21,829 (14 RCTs)	⊕⊕⊕⊕ High^b,d,e	Low OR means favourable outcome
All‐cause death Follow‐up: 6 to 52 weeks	10 per 1000	14 per 1000 (11 to 18)	OR 1.35 (1.05 to 1.75)	21,510 (15 RCTs)	⊕⊕⊕⊝ Moderate^b,d,e,f	Low OR means favourable outcome
SGRQ total score change from baseline (≥ 4 points, MCID) Follow‐up: 26 to 52 weeks	373 per 1000	387 per 1000 (349 to 427)	OR 1.06 (0.90 to 1.25)	13,614 (4 RCTs)	⊕⊕⊕⊝ Moderate^a‐e	High OR means favourable outcome
The absolute risk* (and its 95% CI) of LAMA+LABA group is based on the assumed risk in the LABA+ICS group and the OR of the intervention (and its 95% CI). CI: confidence interval; COPD: chronic obstructive pulmonary disease; FEV₁: forced expiratory volume in one second; ICS: inhaled corticosteroid; LABA: long‐acting beta‐agonist; LAMA: long‐acting muscarinic antagonist; MCID: minimal clinically important difference; MD: mean difference; OR: odds ratio; RCT: randomised controlled trial; SGRQ: St. George's Respiratory Questionnaire.
GRADE Working Group grades of evidence High certainty: we are very confident that the true effect lies close to that of the estimate of effect. Moderate certainty: we are moderately confident in the effect estimate: the true effect is likely to be close to the estimate of effect but may be substantially different. Low certainty: our confidence in the effect estimate is limited: the true effect may be substantially different from the estimate of effect. Very low certainty: we have very little confidence in the effect estimate: the true effect is likely to be substantially different from the estimate of effect.
^aThere was a considerable heterogeneity, I² > 50%. ^bStudies varied in their inclusion criteria around recent history of exacerbation, baseline percentage predicted FEV₁ and exclusion of asthma. No indirectness downgrades, but impacts interpretation and applicability of results (see Discussion). ^cConfidence interval includes benefit of either treatment but lies within predefined threshold for clinical importance (0.7 to 1.5 for OR outcomes; 4‐point MCID for SGRQ) ‐ no downgrade for imprecision ^dNo downgrades for publication bias. Funnel plots examined for primary outcomes (exacerbations, serious adverse events, SGRQ, FEV₁) do not show obvious asymmetry (see Figure 3; Figure 3; Figure 4; Figure 6). Nearly all studies included in pneumonia and all‐cause death analyses. Only 4 studies included in SGRQ 4+ change analysis but not specified in most studies. ^eNo downgrades for risk of bias. Across outcomes, all or almost all studies were at high risk of 'other' bias due to conflicts of interest, and studies contributing between 18.5% and 35.9% of the analysis weight were rated high risk of bias in at least one other domain. We did not prespecify a threshold or specific domains for downgrading, but considered it insufficient to downgrade if studies contributing more than half the analysis weight were at low or unclear risk of bias in all but the 'other' domain. ^fThere was imprecision due to a low number of events.

Table 1. Summary of characteristics of included studies

Study	LAMA+LABA	LABA+ICS	Key inclusion criteria	Follow‐up duration (weeks)	Mean/median age (years)	Number randomised
Beeh 2016	Tiotropium/olodaterol (2.5/5 μg) or tiotropium/olodaterol (5/5 μg)	Salmeterol/fluticasone (50/250 μg) twice daily or salmeterol/fluticasone (50/500 μg) twice daily	%pred FEV₁ 30% to 80% Ex(‐)	6 × 4 time periods (cross‐over)	64	229
Donohue 2015a	Umeclidinium/vilanterol (62.5/25 μg)	Salmeterol/fluticasone (50/250 μg) twice daily	%pred FEV₁ 30% to 70%, mMRC ≥ 2, Ex(‐)	12	63	707
Donohue 2015b	Umeclidinium/vilanterol (62.5/25 μg)	Salmeterol/fluticasone (50/250 μg) twice daily	%pred FEV₁ 30% to 70%, mMRC ≥ 2, Ex(‐)	12	64	700
Ferguson 2018	Glycopyrronium/formoterol fumarate (18/9.6 μg)	Formoterol/budesonide/ fumarate (160/4.8 μg) twice daily	CAT ≥ 10, %pred FEV₁ 25% to 80%, not required to have exacerbation	24	65	943
Frith 2018	Glycopyrronium/indacaterol (50/110 μg)	Salmeterol/fluticasone (50/500 μg) twice daily	CAT ≥ 10, %pred FEV1 30% to 80%, Ex(+)	12	65	502
Herth 2020	Tiotropium/olodaterol (5/5 μg)	Salmeterol/fluticasone (50/500 μg)	%pred FEV₁ < 70%, Ex(‐)	6	62	76
Hoshino 2015	Tiotropium/indacaterol (18/150 μg)	Salmeterol/fluticasone (50/250 μg) twice daily	%pred FEV₁ 30% to 80%, Ex(‐)	16	71	46
Lipson 2018	Umeclidinium/vilanterol (62.5/25 μg)	Vilanterol/fluticasone furoate (25/100 μg)	CAT ≥ 10, %pred FEV₁ < 80%, Ex(+)	52	65	6204
Magnussen 2012	Tiotropium/salmeterol (18/50 μg) twice daily	Salmeterol/fluticasone (50/500 μg) twice daily	%pred FEV₁ ≤ 65%, Ex(‐)	8 x 2 time periods (cross‐over)	61	344
Mostafa 2021	Tiotropium (18 μg) once daily/formoterol (9 μg) twice daily	Formoterol/budesonide (160/4.5 μg) twice daily	%pred FEV₁ 30% to 80%, Ex(‐)	12	64	40
NCT03240575	Tiotropium (5μg)/olodaterol (5μg) once daily	Salmeterol/fluticasone (50μg/250μg) twice daily	%pred FEV₁ 30% to 80%, Ex(‐)	12	64	302
Rabe 2008	Tiotropium/formoterol (18/24 μg) twice daily	Salmeterol/fluticasone (50/500 μg) twice daily	%pred FEV₁ ≤ 65%, Ex(‐)	6	62	605
Rabe 2020	Glycopyrronium/formoterol (9/4.8 µg) twice daily.	Formoterol/budesonide (4.8/160 µg) twice daily.	%pred FEV₁ 25‐65%, CAT ≥ 10, Ex(+)	52	65	4294
Singh 2015	Umeclidinium/vilanterol (62.5/25 μg)	Salmeterol/fluticasone (50/500 μg) twice daily	%pred FEV₁ 30% to 70%, mMRC ≥ 2, Ex(‐)	12	62	717
Vogelmeier 2013	Indacaterol/glycopyrronium bromide (110/50 μg)	Salmeterol/fluticasone (50/500 μg) twice daily	%pred FEV₁ 30% to 80%, Ex(‐)	26	63	523
Vogelmeier 2016	Aclidinium/formoterol (400/12 μg) twice daily	Salmeterol/fluticasone (50/500 μg) twice daily	%pred FEV₁ < 80%, CAT ≥ 10, Ex(‐)	24	63	933
Vogelmeier 2017	Glycopyrronium/Indacaterol (50/110 μg)	any	%pred FEV₁ 50% ‐80%, mMRC ≥ 1, Ex(+)	12	65	1083
Wedzicha 2016	Indacaterol/glycopyrronium bromide (110/50 μg)	Salmeterol/fluticasone (50/500 μg) twice daily	%pred FEV₁ 25% to 60%, mMRC ≥ 2, Ex(+)	52	65	3362
Zhong 2015	Indacaterol/glycopyrronium bromide (110/50 μg)	Salmeterol/fluticasone (50/500 μg) twice daily	%pred FEV₁ 30% to 80%, mMRC ≥ 2, Ex(‐)	26	65	744
%pred FEV₁: % predicted forced expiratory volume in one second; CAT: chronic obstructive pulmonary disease assessment test; Ex(‐): without recent exacerbation; Ex(+): with recent exacerbation; LABA: long‐acting beta‐agonist; LAMA: long‐acting muscarinic antagonist; mMRC: modified Medical Research Council dyspnoea scale

Table 1. Summary of characteristics of included studies

Table 2. ORBIT matrix of outcome reporting across included studies

	Objective or semi‐objective outcomes (investigator‐assessed)					Subjective outcomes (patient‐reported)
	Exacerbations	SAEs	Trough FEV₁	Pneumonia	All‐cause death	SGRQ (mean)	SGRQ ≥ 4 unit benefit
Indacaterol/glycopyrronium studies
Frith 2018	X	X	X	X	X
Vogelmeier 2013	X	X	X	X	X	X	X
Vogelmeier 2017	X	X	X	X	X
Wedzicha 2016	X	X	X	X	X	X	X
Zhong 2015	X	X		X	X
Umeclidinium/vilanterol studies
Donohue 2015a	X	X	X	X	X	X
Donohue 2015b	X	X	X	X	X	X
Lipson 2018	X	X	X	X	X	X	X
Singh 2015	X	X	X	X	X	X
Other LAMA/LABA inhaler studies
Beeh 2016		X	X	X	X
Ferguson 2018		X	X	X	X	X
Herth 2020			X
Hoshino 2015						X
Magnussen 2012	X	X
Mostafa 2021		X			X
NCT03240575		X	X	X	X
Rabe 2008	X	X
Rabe 2020	X	X		X	X	X	X
Vogelmeier 2016	X	X		X
Outcomes shown are those included in the summary of findings table. Outcomes are classed as objective or subjective for the purposes of risk of bias assessment to consider differences in performance and detection bias. FEV₁: forced expiratory volume in one second; ORBIT: outcome reporting bias in trials; SAEs: serious adverse events; SGRQ: St George’s Respiratory Questionnaire

Table 2. ORBIT matrix of outcome reporting across included studies

Table 3. Sponsor list for chronic obstructive pulmonary disease studies

Sponsor	Record count	% of 1723
GlaxoSmithKline	134	7.78
Novartis	128	7.43
AstraZeneca	122	7.08
Boehringer Ingelheim	113	6.56
Pfizer	84	4.88
Nycomed	49	2.84
GSK	45	2.61
Chiesi	41	2.38
Almirall	36	2.09
Merck	30	1.74
Web of Science Core Collection, advanced search for "TI=(COPD) AND TS=(inhal*)" without any restriction hit 1723 reports as of 13 June 2016. "Results analysis" > "Source Titles" output the table above.

Table 3. Sponsor list for chronic obstructive pulmonary disease studies

Comparison 1. Long‐acting muscarinic antagonist (LAMA) plus long‐acting beta‐agonist (LABA) versus LABA plus inhaled corticosteroid (ICS)

Outcome or subgroup title	No. of studies	No. of participants	Statistical method	Effect size
1.1 Exacerbations Show forest plot	13	20960	Odds Ratio (M‐H, Random, 95% CI)	0.91 [0.78, 1.06]

1.1.1 Indacaterol/glycopyrronium	5	6200	Odds Ratio (M‐H, Random, 95% CI)	0.72 [0.63, 0.83]
1.1.2 Umeclidinium/vilanterol	4	8323	Odds Ratio (M‐H, Random, 95% CI)	1.20 [1.03, 1.40]
1.1.3 Other LAMA/LABA inhalers	4	6437	Odds Ratio (M‐H, Random, 95% CI)	0.97 [0.87, 1.08]
1.2 Serious adverse events Show forest plot	18	23158	Odds Ratio (M‐H, Random, 95% CI)	1.02 [0.91, 1.15]

1.2.1 Indacaterol/glycopyrronium	5	6204	Odds Ratio (M‐H, Random, 95% CI)	0.88 [0.76, 1.03]
1.2.2 Umeclidinium/vilanterol	4	8323	Odds Ratio (M‐H, Random, 95% CI)	1.07 [0.73, 1.55]
1.2.3 Other LAMA/LABA inhalers	9	8631	Odds Ratio (M‐H, Random, 95% CI)	1.04 [0.92, 1.19]
1.3 Quality of life as measured by the St George's Respiratory Questionnaire (SGRQ) total score change from baseline Show forest plot	9	14437	Mean Difference (IV, Random, 95% CI)	‐0.57 [‐1.36, 0.21]

1.3.1 Indacaterol/glycopyrronium	2	3693	Mean Difference (IV, Random, 95% CI)	‐1.29 [‐2.08, ‐0.50]
1.3.2 Umeclidinium/vilanterol	4	6615	Mean Difference (IV, Random, 95% CI)	‐0.00 [‐0.02, 0.02]
1.3.3 Other LAMA/LABA inhalers	3	4129	Mean Difference (IV, Random, 95% CI)	‐1.14 [‐4.06, 1.78]
1.4 Trough FEV₁ mean change (litres) Show forest plot	12	14681	Mean Difference (IV, Random, 95% CI)	0.07 [0.05, 0.08]

1.4.1 Indacaterol/glycopyrronium	4	5843	Mean Difference (IV, Random, 95% CI)	0.07 [0.05, 0.09]
1.4.2 Umeclidinium/vilanterol	4	6669	Mean Difference (IV, Random, 95% CI)	0.08 [0.04, 0.11]
1.4.3 Other LAMA/LABA inhalers	4	2169	Mean Difference (IV, Random, 95% CI)	0.07 [0.02, 0.11]
1.5 Pneumonia Show forest plot	14	21829	Odds Ratio (M‐H, Random, 95% CI)	0.61 [0.52, 0.72]

1.5.1 Indacaterol/glycopyrronium	5	6204	Odds Ratio (M‐H, Random, 95% CI)	0.61 [0.44, 0.85]
1.5.2 Umeclidinium/vilanterol	4	8323	Odds Ratio (M‐H, Random, 95% CI)	0.63 [0.50, 0.80]
1.5.3 Other LAMA/LABA inhalers	5	7302	Odds Ratio (M‐H, Random, 95% CI)	0.58 [0.43, 0.78]
1.6 All‐cause death Show forest plot	15	21510	Odds Ratio (M‐H, Random, 95% CI)	1.35 [1.05, 1.75]

1.6.1 Indacaterol/glycopyrronium	5	6204	Odds Ratio (M‐H, Random, 95% CI)	1.02 [0.59, 1.76]
1.6.2 Umeclidinium/vilanterol	4	8324	Odds Ratio (M‐H, Random, 95% CI)	1.51 [1.00, 2.26]
1.6.3 Other LAMA/LABA inhalers	6	6982	Odds Ratio (M‐H, Random, 95% CI)	1.43 [0.94, 2.17]
1.7 SGRQ improvement ≥ 4 points Show forest plot	4	13614	Odds Ratio (M‐H, Random, 95% CI)	1.06 [0.90, 1.25]

1.7.1 Indacaterol/glycopyrronium	2	3192	Odds Ratio (M‐H, Random, 95% CI)	1.25 [1.09, 1.44]
1.7.2 Umeclidinium/vilanterol	1	6204	Odds Ratio (M‐H, Random, 95% CI)	1.00 [0.89, 1.12]
1.7.3 Other LAMA/LABA inhalers	1	4218	Odds Ratio (M‐H, Random, 95% CI)	0.89 [0.79, 1.01]
1.8 Exacerbations (fixed‐effect sensitivity) Show forest plot	13	20960	Odds Ratio (M‐H, Fixed, 95% CI)	0.94 [0.87, 1.01]

1.8.1 Indacaterol/glycopyrronium	5	6200	Odds Ratio (M‐H, Fixed, 95% CI)	0.72 [0.63, 0.83]
1.8.2 Umeclidinium/vilanterol	4	8323	Odds Ratio (M‐H, Fixed, 95% CI)	1.20 [1.03, 1.40]
1.8.3 Other LAMA/LABA inhalers	4	6437	Odds Ratio (M‐H, Fixed, 95% CI)	0.97 [0.87, 1.08]
1.9 Serious adverse events (fixed‐effect sensitivity) Show forest plot	18	23158	Odds Ratio (M‐H, Fixed, 95% CI)	1.03 [0.95, 1.12]

1.9.1 Indacaterol/glycopyrronium	5	6204	Odds Ratio (M‐H, Fixed, 95% CI)	0.88 [0.75, 1.03]
1.9.2 Umeclidinium/vilanterol	4	8323	Odds Ratio (M‐H, Fixed, 95% CI)	1.13 [0.99, 1.27]
1.9.3 Other LAMA/LABA inhalers	9	8631	Odds Ratio (M‐H, Fixed, 95% CI)	1.05 [0.92, 1.20]
1.10 SGRQ mean change (fixed‐effect sensitivity) Show forest plot	9	14437	Mean Difference (IV, Fixed, 95% CI)	‐0.00 [‐0.02, 0.02]

1.10.1 Indacaterol/glycopyrronium	2	3693	Mean Difference (IV, Fixed, 95% CI)	‐1.29 [‐2.08, ‐0.50]
1.10.2 Umeclidinium/vilanterol	4	6615	Mean Difference (IV, Fixed, 95% CI)	‐0.00 [‐0.02, 0.02]
1.10.3 Other LAMA/LABA inhalers	3	4129	Mean Difference (IV, Fixed, 95% CI)	‐0.23 [‐1.01, 0.54]
1.11 Trough FEV₁ mean change (fixed‐effect sensitivity) Show forest plot	12	14681	Mean Difference (IV, Fixed, 95% CI)	0.06 [0.05, 0.07]

1.11.1 Indacaterol/glycopyrronium	4	5843	Mean Difference (IV, Fixed, 95% CI)	0.07 [0.05, 0.08]
1.11.2 Umeclidinium/vilanterol	4	6669	Mean Difference (IV, Fixed, 95% CI)	0.06 [0.05, 0.07]
1.11.3 Other LAMA/LABA inhalers	4	2169	Mean Difference (IV, Fixed, 95% CI)	0.06 [0.04, 0.07]

Comparison 1. Long‐acting muscarinic antagonist (LAMA) plus long‐acting beta‐agonist (LABA) versus LABA plus inhaled corticosteroid (ICS)