Presentation (CTI)

OBJECTIVES: Real‑world data are increasingly used in HTA to inform treatment effects in routine clinical practice. However, the high dimensionality and complexity of these data, together with simplifying assumptions that are often routinely applied in conventional analyses, can undermine credibility and limit acceptance. Causal machine learning (ML) has emerged as a promising approach to address these challenges. This study evaluates the performance of targeted maximum likelihood estimation with a super learner (TMLE‑SL) against conventional methods.
METHODS: An open access real‑world dataset of patients with diabetes from the laboratory of Medical City Hospital in Iraq was used to simulate assignment of hypothetical glucose‑lowering treatments and corresponding HbA1c outcomes. Simulations incorporated complex, non‑linear relationships among variables, including patient characteristics and laboratory measures. Treatment effects were estimated using TMLE‑SL, linear regression, and inverse probability weighting (IPW). Performance was assessed using mean absolute percentage error (MAPE), confidence interval (CI) coverage, and average CI width.
RESULTS: TMLE‑SL reduced MAPE by 38.3% and 45.1% relative to linear regression and IPW respectively, indicating substantially improved estimation accuracy. TMLE-SL’s 95% CI successfully captured the true treatment effect 94.4% of the time, compared with 59.6% for regression and 92.6% for IPW. Additionally, TMLE‑SL decreased average CI width by 40.6% versus IPW while maintaining appropriate coverage.
CONCLUSIONS: This study demonstrates that TMLE-SL can meaningfully improve the accuracy, precision, and reliability of treatment effect estimation in a complex data setting. These findings highlight the potential of causal ML approaches to address limitations of conventional methods when standard modelling assumptions are challenged. Incorporating such methods into HTA-related analyses could improve confidence in real-world treatment effect estimates and support more robust healthcare decision-making.