Types of studies

We included randomised controlled trials and excluded quasi‐randomised or confounded studies.

Types of participants

We included studies of people over the age of 18 years with a history of stroke or TIA. We used the definition of TIA as provided in the original publications. We excluded studies of people with diabetes who lacked a clear history of stroke or TIA.

Types of interventions

We included trials comparing PPAR‐γ agonists (including pioglitazone, rosiglitazone, glitazone, troglitazone, netoglitazone, rivoglitazone, ciglitazone, balaglitazone, darglitazone, edaglitazone, englitazone, and lobeglitazone) with placebo, regardless of the length of treatment period and dosage of treatment. We included other concomitant therapies providing they were administered to both the intervention and control groups.

Types of outcome measures

We measured all outcomes at the end of follow‐up.

Electronic searches

We searched the Cochrane Stroke Group Trials Register (1 January 2022) and the following electronic bibliographic databases:

• Cochrane Central Register of Controlled Trials (CENTRAL; 2020, Issue 12) in the Cochrane Library (searched 1 January 2022) (Appendix 1);

• MEDLINE (Ovid) (1949 to 1 January 2022) (Appendix 2);

• Embase (Ovid) (1980 to 1 January 2022) (Appendix 3);

• CINAHL (EBSCO) (Cumulative Index to Nursing and Allied Health Literature; 1982 to 1 January 2022) (Appendix 4);

• AMED (Ovid) (Allied and Complementary Medicine Database; 1985 to 1 January 2022) (Appendix 5).

We developed all the search strategies with the help of the Cochrane Stroke Group Information Specialist.

We also searched the following ongoing trials registers on 1 January 2022:

• US National Institutes of Health Ongoing Trials Register ClinicalTrials.gov (www.clinicaltrials.gov/) (Appendix 6);

• World Health Organization International Clinical Trials Registry Platform (apps.who.int/trialsearch) (Appendix 7).

Searching other resources

We searched the following resources with the Chinese equivalents of keywords (Appendix 8):

• Chinese Clinical Trials Registry (searched on 1 January 2022);

• CBM‐disc (China Biological Medicine Databases) (1979 to 1 January 2022);

• CNKI (China National Knowledge Infrastructure) (1979 to 1 January 2022);

• Chinese MD and DD Dissertations in CNKI (searched on 1 January 2022);

• CACP (Chinese Academic Conference Papers Database) (1998 to 1 January 2022);

• CDDB (Chinese Dissertations Database) (1977 to 1 January 2022);

• Chinese Evidence‐Based Medicine Database (searched on 1 January 2022);

• CMAC (China Medical Academic Conferences) (1994 to 1 January 2022);

• CMCC (Chinese Medical Current Contents) (1994 to 1 January 2022);

• Chinese Science and Technique Journals Database (VIP) (1989 to 31 January 2022);

• Wanfang Data (www.wanfangdata.com/) (1984 to 1 January 2022).

We also:

• used Science Citation Index Cited Reference Search for forward tracking of important articles;

• searched reference lists of relevant reviews and retrieved articles;

• searched relevant conference proceedings, including the 7th European Stroke Organisation Conference (2021) and the 4th to 13th World Stroke Congress (from 2000 to 2021);

• contacted authors where necessary for missing information;

• contacted the manufacturers (Takeda Pharmaceutical Company and GlaxoSmithKline Pharmaceuticals) for updated information.

Selection of studies

Two review authors (JL, LW) independently screened titles and abstracts of the references identified by the search and excluded obviously irrelevant reports. We retrieved the full‐text articles for the remaining references, and the same two review authors independently screened the full‐text articles and identified studies for inclusion, and also identified and recorded reasons for exclusion of the ineligible studies. We resolved any disagreements through discussion or, if required, we consulted an external third party. We collated multiple reports of the same study so that each study, not each reference, was the unit of interest in the review. We recorded the selection process and completed a PRISMA flow diagram (Figure 1).

Figure 1

---

---

Study flow diagram

Data extraction and management

Two review authors (JL, LW) independently extracted data from the published reports onto standardised forms, and cross‐checked the data for accuracy. We used checklists to independently record details including methods of generating the randomisation schedule, method of concealment of allocation, blinding of assessors, intention‐to‐treat analysis, adverse events and dropouts for all reasons, important imbalance in prognostic factors, participants (socio‐demographic and related clinical information), interventions (medications and non‐pharmacological interventions), and outcomes. We resolved disagreements with an external third party.

Assessment of risk of bias in included studies

Two review authors (JL, LW) independently assessed the risk of bias for each study using the criteria outlined in the Cochrane Handbook for Systematic Reviews of Interventions (Higgins 2017). We resolved any disagreements by discussion or by involving an external third party. We assessed the 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 the risk of bias for each domain as high, low, or unclear and provided information from the study report together with a justification for our judgement in the risk of bias tables.

Measures of treatment effect

We expected the randomised controlled trials to measure dichotomous data. We expressed dichotomised data as risk ratios (RRs) with their 95% confidence intervals (CI). If a trial (or group within a trial) reported no adverse events or dropouts, we calculated risk differences (RDs) instead of RRs with 95% CI. We entered and analysed data using Review Manager 5 (RevMan 2020).

Unit of analysis issues

We dealt with any unit of analysis issues according to the Cochrane Handbook for Systematic Reviews of Interventions (Higgins 2022).

Dealing with missing data

We planned to contact the authors of the studies for further details if any data were missing or to establish the characteristics of unpublished trials. According to the intention‐to‐treat principle, all randomised participants should be included. We considered different scenarios (best‐case and worst‐case) to account for missing data.

Assessment of heterogeneity

We used the I² statistic to measure heterogeneity among the trials in each analysis, considering I² values over 50% as suggestive of substantial heterogeneity.

Assessment of reporting biases

We planned to use funnel plots to examine potential publication bias if there was a sufficient number of trials (Egger 1997).

Data synthesis

Where we considered studies to be sufficiently similar, we conducted a meta‐analysis by pooling the appropriate data using Review Manager 5 (RevMan 2020). We expressed dichotomised data as RRs with their 95% CI. If a trial (or group within a trial) reported no adverse events or dropouts, we calculated RDs instead of RRs with 95% CI. We calculated the overall effects using a random‐effects model regardless of the level of heterogeneity. We provided a descriptive summary of the results when substantial heterogeneity between the studies prevented us from combining outcome data.

Subgroup analysis and investigation of heterogeneity

We intended to undertake subgroup analyses according to the age and ethnicity of participants, TIA definition, different PPAR‐γ agonists, and dosage and duration of treatment. We intended to use the Chi² test to examine the significance of differences between subgroups.

Sensitivity analysis

We analysed sensitivity by assessing the robustness of results in fixed‐effect versus random‐effects models. We planned to assess the impact of studies at high risk of bias. We also examined potential sources of methodological heterogeneity.

Summary of findings and assessment of the certainty of the evidence

We assessed the certainty of the evidence using the GRADE approach (Schünemann 2013), in relation to the following criteria.

• Study design

• Risk of bias

• Inconsistency

• Indirectness

• Imprecision

• Other considerations (including publication bias)

We used GRADEpro GDT software to create a summary of findings table for the comparison evaluated in the review (GRADEpro GDT). We included the following outcomes: recurrence of stroke, reported adverse events, serious vascular events, deaths due to vascular events, disability due to vascular events, improvement in quality of life, and insulin sensitivity, and used the table to guide our conclusions.