Successful Use of Propensity Score Methods for HTA in Germany: A Near-Impossible Task?

OBJECTIVES: To date, German HTA bodies G-BA and IQWiG rarely consider non-randomized evidence. However, marketing authorization, especially for orphan drugs and ATMPs, often rely on single-arm trials or immature data that lack the necessary comparative evidence required for HTA assessments.

Under special circumstances, the G-BA can request a routine practice data collection (AbD) to gather real-world data for both the intervention and comparator. In these studies, propensity score (PS) methods are preferred for adjusting the data to account for confounding factors.

We aim to evaluate the viability of PS methods facing the stringent requirements of HTA bodies in Germany and explore data properties under which an additional benefit can be derived according to IQWiG methodology.

METHODS: To assess the viability of this approach in demonstrating an additional benefit, we utilized simulated data that reflect typical scenarios expected in AbDs, assuming the presence of an underlying additional benefit. Since the G-BA and IQWiG do not provide specific details about PS method implementation, we examined commonly used approaches for matching and weighting as outlined in available AbD protocols.

RESULTS: We found that PS methods were addressing the issue of unbalanced confounders between treatment groups in this context. However, implementing the prescribed approach was challenging due to stringent criteria for deriving an additional benefit. The multitude of confounders, coupled with the relatively small sample size, hampered accurate PS estimation through logistic regression, especially when confounders were correlated. Even with large sample sizes, demonstrating an underlying additional benefit often remained elusive.

CONCLUSIONS: The increase in sample size needed to account for the multitude of confounders has not been adequately considered thus far, and realistically achieving it would be challenging, particularly for rare diseases. In summary, our analysis raises doubts about the suitability of the current analyses required for AbDs in determining the additional benefit.