Advancement of Statistical Methods to Overcome Evidence Gaps

At ISPOR Europe 2023, experts discussed how advanced statistical methods for the analysis of data to support health technology assessment (HTA) submissions can help overcome gaps in the available evidence, and better adapt the analyses to the target population. Two key methods presented included multi-level network meta regression (ML-NMR)¹ and basket trials², and the existing barriers to their uptake by HTA bodies were discussed.

For the comparison of clinical trial results, robust analyses require the patient populations in compared trials to be similar, or for differences in the patient populations to be adjusted for as part of the analysis. Standard population adjustment methods such as matching adjusted indirect comparisons (MAICs) and simulated treatment comparisons (STCs) facilitate population adjustment for pairwise comparisons of two trials (adjusting the population of interest to the comparator population), but they require individual patient data (IPD) which are not always available. ML-NMR³ extends these methodologies to allow comparison of multiple clinical trials, can be performed without IPD, and can adapt to any population of interest, providing a more robust evidence base that is better adapted to the target population. However, a sufficient sample size is required to allow adjustment for multiple variables, which could limit the use of such methods.

In the context of oncology trials, basket trials were discussed as a route to overcome limited data due to small patient populations. Basket trials include patients with the same mutation but across different tumour sites. Analysis of such trials using Bayesian hierarchical methods was presented⁴, which reduces the need for several separate trials specific to each tumour site, or could be used to augment such separate trials. However, these analyses require IPD, which in some cases may not be available, and require the assumption that treatment effects across tumour sites are similar, which may not always be the case.

There are further practical barriers to the widespread acceptance of both methods: they require more computing power and time, and both involve a steep learning curve for statisticians and HTA bodies alike. Pathways to increase the uptake of complex methods in the face of such challenges include the development of efficient open-source analysis code⁵ and clear HTA guidelines⁶. However, to date, both ML-NMR and basket trials have been accepted in a handful of NICE submissions^7-9, paving the way for future work using these methods. Their potential to help improve the available evidence base, and therefore provide more timely access to treatments for patients is clear, particularly in indications where completing further clinical trials is unfeasible.