OBJECTIVES: This study aims to investigate novel post-processing methods for probabilistic sensitivity analysis (PSA) to extract additional insights from the results of a cost-effectiveness model. Specifically, we explore the use of heat map (HM) and density-based spatial clustering of applications with noise (DBSCAN) technique.

METHODS: PSA results of a cohort Markov model for comparison of lifestyle intervention with small molecule in nonalcoholic steatohepatitis were generated and examined using HM and DBSCAN. HM technique graphically represents the concentration of PSA results by selecting dimensions, defining the heat map's resolution and analysing the number of points within each grid cell. The DBSCAN technique utilized unsupervised detection of point aggregation in two or three dimensions identifying self-defined data clusters based on neighbouring point theory. The main points aggregation was determined by setting the initial parameters of the detection range function. Outcomes from both processing methods were compared to commonly used metrics such as mean PSA outcomes and deterministic results.

RESULTS: In the deterministic analysis, the model estimated 0.1516 QALYs gained and $8,200 of cost savings for small molecule vs lifestyle intervention. The incremental mean results from PSA were 0.1563 QALYs and $9,198. The HM showed that the boundaries of the highest aggregate points were 0.1357-0.1641 for incremental QALYs and $10,476-$10,030 for the cost savings. Neither the deterministic results nor the PSA mean did fit into HM's highest aggregation. The DBSCAN captured 97.5% of results in the same cluster, which centre was 0.1564 QALYs and $9,167, consistent with the average PSA results.

CONCLUSIONS: In our model, both tested techniques gave similar results as commonly used mean values thereby confirming these estimates. However, this does not necessarily hold true for all models. Point aggregation density and clustering might be an additional important features to be considered while interpreting the PSA outcomes to better understand and possibly reduced accompanying uncertainty.