A comprehensive search was conducted across multiple databases, including the Cochrane Central Register of Controlled Trials, Cochrane Database of Systematic Reviews, EMBASE, MEDLINE, Web of Science, and ClinicalTrials.gov, (NIH U.S National Library of Medicine, 2022), to identify relevant RCTs published since inception to March 2024. The search strategy utilized a broad spectrum of medical terminology related to fractures, TZDs, and AHGs, as detailed in Supplementary Appendix 1.

RCTs involving participants aged 18 years or older, which reported fracture outcomes in individuals treated with pioglitazone (TZD), compared to placebo or other AHGs, were eligible for inclusion. Trials were excluded if they did not meet these criteria, specifically if they were not RCTs, did not report fractured outcomes, or included participants younger than 18 years. Additionally, non-peer-reviewed sources or trials lacking robust fracture data were omitted to maintain the quality and relevance of the meta-analysis. Duplicate publications were rigorously screened and removed, with only the most comprehensive and complete version retained for final analysis to avoid any data duplication and redundancy.

The extracted data encompassed a comprehensive range of variables, including the baseline demographic and clinical characteristics of the study population, the specific design and methodology of each trial, the total number of participants enrolled, and detailed information on the backgrounds of both AHG users and non-users. Additionally, data regarding the duration of follow-up for each study were collected, which offered key insight into the long-term outcomes and potential cumulative effects of pioglitazone and other AHG drugs on fracture risk.

Each RCT was assessed for risk of bias, adhering to the Cochrane Quality Assessment Tool (Higgins et al., 2003). This tool evaluates six key domains of bias: selection, detection, performance, attrition, reporting, and other potential biases. Each key domain was rated as “low risk,” “high risk,” or “unclear,”. The unclear rating indicates insufficient information to determine bias. Trials that scored “high risk” in any of the first three domains were classified as providing low-quality evidence.

Additionally, the overall certainty of the evidence was appraised using the Grading of Recommendations Assessment, Development, and Evaluation (GRADE) approach (Guyatt et al., 2008). Evidence quality was considered ‘high’ if multiple trials exhibited a low risk of bias across different domains. However, ratings were downgraded in the presence of concerns related to trial design, inconsistency in findings, indirectness of evidence, imprecision, or potential publication bias. The principal investigator conducted a thorough quality assessment of the included trials, while the other investigators independently addressed any discrepancies to ensure the precision and robustness of the conclusions.

The primary outcome of interest was the incidence of fractures, which were analyzed as safety endpoints. These outcomes were stratified considering the severity of events. The fractures were categorized as either non-serious or serious (those resulting in hospitalization, surgery, or physiotherapy). Another criterion to classify fractures was based on their underlying mechanisms, such as stress-related (e.g., repetitive stress or minimal trauma) versus non-stress-related. Fractures were also categorized by energy level: high-energy fractures resulted from significant trauma (e.g., falls from substantial heights or motor vehicle accidents), while low-energy fractures arose from minimal trauma (e.g., falls from standing height or lower). Additionally, pathological fractures, associated with underlying conditions such as osteoporosis, were distinguished from non-pathological ones.

The timing of fractures relative to TZD exposure (before or after treatment initiation) was analyzed alongside fracture location, including key anatomical sites such as the spine, hip, pelvis, femur, and upper and lower extremities (including wrist and ankle). BMD data were also collected for a more comprehensive assessment of bone health and fracture susceptibility. This detailed framework allowed for the examination of pioglitazone’s impact on fracture risk across different subgroups, such as male and female individuals, thereby providing valuable insights into the gender-specific risks of TZD therapy. All fracture events were classified according to the Medical Dictionary for Regulatory Activities (MedDRA, 2022a).

In trials that simultaneously reported fracture risk by subtype (serious or non-serious adverse events), one of these fractured outcomes data was excluded from the intervention of the interest group, re-reported the data of fracture by subtypes separately, and then compared the intervention of interest to placebo or other AHGs.

The comparator AHGs, either as monotherapy or in combination, were categorized into the following drug classes: metformin, sulfonylureas, TZDs, dipeptidyl peptidase-4 inhibitors, glucagon-like peptide-1 agonists, sodium-glucose co-transporter-2 inhibitors, α-glucosidase inhibitors, and insulin. To avoid unit-of-analysis errors, trials involving multiple doses of TZDs were aggregated to develop a unified pairwise comparator.

In trials with two or more intervention groups, where all participants, including those in control and comparator groups, were exposed to TZDs, a high risk of bias was identified. Although these trials were included in the study to provide a comprehensive evaluation, their fracture outcome data were excluded from the meta-analysis to maintain accuracy.

A meta-analysis was conducted to aggregate pooled data for quantitative evaluation. For determining dichotomous outcomes, relative risks (RRs), 95% confidence intervals (CIs), and weighted mean differences were computed. Continuous variables were assessed by calculating the mean values of BMD levels, sample sizes, and standardized mean differences (SMD). Fracture outcomes were combined using the Mantel–Haenszel test, incorporating a continuity correction of 0.5 for trials reporting zero events in any treatment arm. Trials with zero events in both arms were excluded to prevent data distortion.

The meta-analysis addressed fracture risk, including mechanisms and severity, across various skeletal sites, participant sex, and BMD levels. Several fracture events were analyzed to evaluate the overall safety profile of pioglitazone compared to placebo or other AHGs, considering participants with insulin resistance or type 2 diabetes mellitus (T2DM), and those with or without cardiovascular disease (CVD) risk, including stroke. These analyses accounted for different baseline characteristics across studies.

Both fixed and random-effects models were applied in the analysis. The fixed-effect model assumes a common effect size across studies, applicable when heterogeneity within the meta-analysis is minimal. In contrast, the random-effects model accommodates unexplained heterogeneity by assuming varied effect sizes across studies, with effects related to the intervention. This model, based on inverse variance, adjusted the study weights according to the degree of heterogeneity or variation among different intervention effects (Riley et al., 2011).

Inter-study heterogeneity was evaluated using the Q statistic and quantified by the I ² statistic, where an I ² value ≥75% indicated substantial heterogeneity not attributable to chance (Higgins et al., 2003). Potential reporting bias was examined using Egger’s linear regression (Egger et al., 1997) and Begg’s rank correlation tests (Begg and Mazumdar, 1994), with results visually inspected using funnel plots. These plots depicted the total trial inverse sample size (standard error) against the natural logarithm of the treatment-effect magnitude of RRs. A symmetrical inverted funnel suggested no reporting bias.

All statistical tests were two-sided, with a significant threshold set at a P value of <0.05. The absolute risk, absolute risk difference, and risk rates per 100 patient years were determined. Statistical analyses and graphical outputs were produced using Review Manager version 5.4.1 (MedDRA, 2022b) and GRADEpro GDT software (GRADEpro GDT Cochrane Community, 2022).

All participants included in the meta-analysis were diagnosed with either insulin resistance or T2DM, with 10% being nondiabetic and 90% diabetic. The average age of participants was 72 years, and 58% were male individuals. The mean HbA1c level across the studies was approximately 8.5%. Participants who were administered pioglitazone (or TZDs) had a history of stroke ranging from 2.3% to 18.8%, prior fractures between 3.8% and 15%, and bone disease or low BMD in less than 1% of cases.

Most trials utilize various comparative dosing regimens, either oral or injectable. For TZD, the administered doses ranged from 15 to 45 mg per day. Additional AHG drugs were used in accordance with their approved dosages as per medical guidelines. In the pioglitazone trials, the median follow-up duration varied from 12 to 261 weeks, with three RCTs having follow-up periods of less than 24 weeks, nine RCTs ranging from 24 to 52 weeks, eight RCTs from 52 to 104 weeks, and seven RCTs extending beyond 104 weeks.

The risk of bias assessment in the meta-analysis revealed varying levels of quality across the trials. Out of the trials evaluated, 36 were classified as having a low risk of bias, eight as having an unclear risk, and 34 as having a high risk. Although sufficient descriptive data were available for random sequence generation, 26 trials (26%) used an open-label design with two or more intervention groups, by comparing non-TZD pioglitazone trials with other AHG drugs in both arms. In these trials, participants in both arms were exposed to pioglitazone and other TZD drugs, enabling a direct comparison of pioglitazone’s efficacy and safety against other AHG treatments. However, this design introduced a high risk of performance bias, leading to the exclusion of these trials from the meta-analysis. Additionally, 8% of the trials demonstrated unclear risks of detection and selection bias (Figure 2; Supplementary Figure S1).

Among the 78 trials assessed, 70 reported fracture events, whereas eight trials recorded no fractures in either study arm. The latter were excluded from our meta-analysis. Most trials primarily focus on the change in HbA1c levels from baseline. However, the PROactive (Dormandy et al., 2009) trial was specifically designed to assess major adverse cardiovascular events (MACE), including stroke, and the IRIS (Kernan et al., 2016) trial targeted fatal or non-fatal stroke as its primary outcomes. Despite this, fracture outcomes were consistently recorded across all 78 trials as part of the safety assessment process, categorized as either serious or non-serious adverse events.

Detailed results of trial outcomes are provided in Supplemental Tables: fracture risk categorized by severity and mechanism (Supplementary Figures S2, S3), anatomical site (Supplementary Figures S4–S10), sex differences (Supplementary Figures S11, S12), and BMD levels (Supplementary Figure S13). Out of the 70 trials assessed, a meta-analysis was performed on 44 trials that reported at least one fracture event, after excluding 26 trials identified as having a high risk of performance bias, Table 1. This analysis included comparisons between TZDs and non-TZDs, comprising 12 trials on rosiglitazone versus non-rosiglitazone, and 27 trials on pioglitazone versus non-pioglitazone. Of the pioglitazone trials, 14 compared it to placebo, while 13 compared it to other AHG drugs.

In the analysis of 44 trials involving TZDs (86,659 participants; 1,915 fractures), the relative risk of fractures was found to be 35% higher with TZD use compared to non-TZDs. Specifically, TZD use was found associated with an increased fracture incidence (RR 1.35; 95% CI 1.24–1.48; P < 0.00001), as revealed by both fixed- and random-effects models (Supplementary Figures S3, S4).

Similarly, in the 12 trials evaluating rosiglitazone (13,470 participants; 701 fractures), the relative risk of fractures was a 42% increase with rosiglitazone compared to non-rosiglitazone comparators. Rosiglitazone (TZD) use was associated with a significantly increased incidence of fractures (RR 1.42; 95% CI 1.23–1.64; P < 0.00001), as assessed using fixed- and random-effects models (Figure 3).

In the analysis of 27 trials involving TZD pioglitazone (24,718 participants; 548 fractures), the incidence risk of fractures increased by 19% with pioglitazone compared to non-pioglitazone (RR 1.19; 95% CI 1.01–1.40; P = 0.04), as assessed using both fixed-effect and random-effects models (Figure 4; Supplementary Figures S5–S8).

In 14 trials comparing pioglitazone with placebo (13,451 participants; 449 fractures), pioglitazone use was found associated with a significant increase in fracture incidence (RR 1.21; 95% CI 1.01–1.45; P = 0.04). This increase was observed for both non-serious (RR 1.25; 95% CI 1.03–1.51; P = 0.02) and serious fractures (RR 1.48; 95% CI 1.10–1.98; P = 0.01), particularly low-energy fractures from falls (RR 1.49; 95% CI 1.20–1.87; P = 0.0004). However, no significant increase in the risk of high-energy (RR 1.43; 95% CI 0.93–2.20; P = 0.10), stress-related (RR 1.25; 95% CI 0.49–3.16; P = 0.64), or pathological fractures (RR 0.67; 95% CI 0.24–1.87; P = 0.44) was observed with pioglitazone compared to placebo, Figure 5.

Pioglitazone use was associated with a higher incidence risk of fractures in nondiabetic individuals (RR 0.87; 95% CI 0.43–1.76; P = 0.69), especially those with a history of stroke (RR 1.41; 95% CI 1.09–1.83; P = 0.008). Conversely, pioglitazone did not significantly increase fracture risk in patients with T2DM (RR 1.02; 95% CI 0.08–1.30; P = 0.88) or in those at risk or not at risk for CVD (RR 1.44; 95% CI 0.81–2.57; P = 0.22) compared to placebo.

The study results also elucidated that pioglitazone use led to a significant increase in the risk of fractures in the lower extremities (RR 1.85; 95% CI 1.33–2.56; P = 0.0002) and spine (RR 2.13; 95% CI 1.28–3.55; P = 0.004), with a more pronounced risk observed in female individuals (RR 1.56; 95% CI 1.20–2.02; P = 0.0008) compared to male individuals (RR 1.10; 95% CI 0.84–1.43; P = 0.49). In three trials, use of pioglitazone was found associated with decreased SMD in BMD, with the most significant reductions observed in the lumbar spine (SMD -0.18; 95% CI -0.34 to −0.03; P = 0.02) and hip (SMD -0.53; 95% CI -0.96 to −0.10; P = 0.02), but not in the femoral neck (SMD -0.25; 95% CI -0.61 to 0.11; P = 0.17) compared to placebo.

In 13 trials comparing pioglitazone to other AHGs (11,267 participants; 99 fractures), pioglitazone use could not significantly increase the incidence risk of fractures (RR 1.08; 95% CI 0.73–1.59; P = 0.70). There was no statistically significant difference between the risk of non-serious (RR 1.28; 95% CI 0.81–2.01; P = 0.29) and serious fractures (RR 0.71; 95% CI 0.32–1.55; P = 0.38) or high-energy fractures (RR 0.18; 95% CI 0.01–4.40; P = 0.29) and the use of pioglitazone and other AHGs. The fracture risk did not significantly differ between female (RR 1.55; 95% CI 0.84–2.86; P = 0.16) or male individuals (RR 0.95; 95% CI 0.49–1.82; P = 0.87) when comparing pioglitazone to other AHGs.

Detailed meta-analysis results on fracture risk by follow-up duration (Supplementary Figures S9, S10), clinical trial baseline characteristics (Supplementary Figures S11–S14), skeletal location (Supplementary Figures S15, S16), sex differences (Supplementary Figures S17–S20), severity and mechanism of fractures (Supplementary Figures S21–S24), BMD (Supplementary Figures S25, S26), stratification by individual subclasses of AHGs comparators (Supplementary Figures S27, S28), clinical trial risk of bias (Supplementary Figures S29, S30), and treatment dose of pioglitazone (Supplementary Figures S31, S32) are comprehensively detailed in the Supplementary Appendix 1.

Previous meta-analyses examining pioglitazone’s fracture risk in patients with T2DM have yielded inconsistent findings. Some studies (Zhu et al., 2014; Loke et al., 2009) reported an increased fracture risk in women, with significant reductions in BMD at the lumbar spine, femoral neck, and hip, irrespective of age or cumulative exposure. However, one study (Pavlova et al., 2018) reported no significant differences in fracture risk across genders or exposure durations. These discrepancies underscore the need for a more nuanced investigation. This meta-analysis seeks to bridge these gaps by evaluating pioglitazone’s impact on fracture risk compared to other AHG drugs, analyzing mechanisms, severity, and skeletal site-specific risks while accounting for factors like gender, insulin resistance, T2DM, and CVD.

The included trials offered diverse fracture outcomes but lacked uniformity in reporting key variables, such as stroke status, fracture specifics, and risk factors. For instance, while the PROactive (Dormandy et al., 2009) trial involved T2DM patients post-stroke, it did not provide clear fracture data, limiting its applicability to findings from the IRIS (Kernan et al., 2016) trial. Additionally, inconsistencies in fracture definitions—encompassing broad categories like leg/hip, hip/femur, foot/ankle, or hand/wrist fractures and multiple fractures in a single individual—complicated accurate interpretation. Given that fractures were primarily assessed as safety outcomes, crucial details such as fracture-related disability, mortality, fall incidence, follow-up duration, and baseline bone markers were inconsistently reported, further impeding precise risk assessment.

Current evidence indicates an increased fracture risk with pioglitazone use in non-diabetic post-stroke patients; however, the limited number of studies in this population calls for cautious interpretation of evidence. Expanding research to include larger, more diverse cohorts of non-diabetic post-stroke individuals could enhance the reliability of these findings. Furthermore, it remains unclear whether pioglitazone’s fracture risk stems directly from its effects or from neurological disabilities post-stroke, as trials to date have primarily compared pioglitazone to placebo. It may be noted that conducting cross-sectional and comparative studies involving other AHGs, such as metformin, and larger population are critical to understanding relative fracture risks. To guide clinical decision-making, personalized treatment strategies are essential, balancing pioglitazone’s fracture risk against its cardiovascular benefits, particularly in high-risk groups. Future research should prioritize comprehensive safety evaluations, including standardized fracture incidence reporting, BMD changes, and long-term follow-up, to inform clinical guidelines and optimize the safe use of pioglitazone in diverse patient populations.

Although the use of pioglitazone demonstrates substantial cardiovascular benefits, including reduced stroke and MACEs, its associated fracture risk—particularly in postmenopausal female individuals and those with pre-existing skeletal conditions—remains a significant concern. Current evidence suggests that the increased fracture risk is notable primarily in specific subgroups and may be influenced by the interplay between improved glycemic control and hypoglycemia-related falls. Future comparative studies between pioglitazone and other AHG drugs, alongside comprehensive safety evaluations, are essential to refine risk profiles and optimize treatment strategies. Clinicians should adopt personalized approaches that balance pioglitazone’s cardiovascular advantages with its potential adverse effects on bone health, ensuring informed and effective management of patients with T2DM. The results may guide appropriate treatment decisions regarding the use of pioglitazone in selected individuals and inform future post-stroke trial designs.