Factors Affecting the Treatment Heterogeneity of PPARγ and Pan-PPAR Agonists in Type 2 Diabetes Mellitus: A Systematic Review and Machine Learning-Based Meta-Regression Analysis

TL;DR

This study identifies factors like female sex, younger age, and lower HDL-C levels that predict better treatment response to PPARγ and pan-PPAR agonists in type 2 diabetes.

Contribution

The study introduces a machine learning-based meta-regression approach to identify clinical predictors of treatment response to PPARγ and pan-PPAR agonists.

Findings

Female sex, younger age, and lower HDL-C levels are significantly associated with greater HbA1c reduction.

Lower HDL-C and higher female proportion are linked to greater FPG reduction with PPARγ and pan-PPAR agonists.

Meta-random forest models confirm these variables as top predictors of drug response.

Abstract

Background/Objectives: Significant heterogeneity in the treatment response to peroxisome proliferator-activated receptor γ (PPARγ) agonists exists, and predictive factors for their efficacy remain unclear. We aimed to assess the relationships between routinely available clinical features and the efficacy of PPARγ agonists and pan-PPAR agonists by meta-regression analysis. Methods: We searched PubMed, Embase, Cochrane Library, ClinicalTrials.gov and the WHO International Clinical Trials Registry Platform (ICTRP) and included randomised controlled trials involving type 2 diabetes patients with 12-week or longer treatment durations with PPARγ agonists or pan-PPAR agonists published before 11 November 2023 (PROSPERO registration number: CRD42024578987). We conducted mixed-effect meta-regression analyses between baseline variables and treatment response. Moreover, we developed a machine learning-based meta-forest model and ranked the relative importance of each variable. Results: In 147 studies involving 29,250 participants, PPARγ and pan-PPAR agonists significantly reduced HbA1c (mean difference(MD) = −0.8876 [95% confidence interval (CI): −0.8999, −0.8754]; p < 0.0001, I2 = 96.0%) and FPG = (MD = −1.7900 [95% CI: −1.9137, −1.6663]; p < 0.0001, I2 = 92.0%). Multivariable association analysis suggested that a greater proportion of female participants (β = 0.0066 [95% CI: 0.0012, 0.0121]; p = 0.017), younger age (β = −0.0314 [95% CI: −0.05, −0.0129]; p = 0.0009) and lower HDL-C levels (β = −0.9304 [95% CI: −1.5176, −0.3431]; p = 0.0019) were significantly associated with a greater decrease in HbA1c. A greater proportion of female participants (β = 0.0112 [95% CI: 0.0019, 0.0205]; p = 0.0178) and lower baseline HDL-C levels (β = −1.8722 [95% CI: −2.812, −0.9323]; p < 0.0001) were significantly associated with a greater decrease in FPG. These variables also ranked among the top five most important predictors of drug response in the meta-random forest models. Conclusions: Our study demonstrated that female sex, younger age, and lower HDL-C levels were associated with greater glycaemic lowering effect from PPARγ and pan-PPAR agonists.

Figures