Presentation (CTI)

OBJECTIVES: Health-Related quality of life (HRQoL) is the primary effectiveness measure reported in cost effectiveness analysis of healthcare interventions. The PROMIS-Preference (PROPr) instrument, adopted by studies in the Justice Community Overdose Innovation Network, measures HRQoL across seven utility domains which are combined to compute a multi-attribute composite utility score (representing overall health-state).
METHODS: This study analyzed multi-attribute HRQoL data from 2,648 individuals with OUD to compare five approaches for estimating multi-attribute PROPr scores: (1) unadjusted summation, (2) post-hoc covariate adjustment, (3) pre-summation domain adjustment, (4) unadjusted probabilistic simulation, and (5) covariate-adjusted probabilistic simulation. For the probabilistic approaches (4 and 5) we modeled each domain’s score using beta distributions to accommodate skewed domain-specific distributions and the bounded utility interval [0,1], with Monte Carlo simulation generating the composite scores.
RESULTS: Results from the pooled, cross-study analysis showed that post-hoc and pre-summation adjustments produced compressed distributions for the PROPr HRQoL estimates with limited variance, while unadjusted probabilistic simulation inflated mean HRQoL scores (likely a result of misspecified domain-level covariate relationships). Covariate-adjusted probabilistic simulation yielded variance that was consistent with empirical domain-level variability, produced approximately normal composite distributions, and preserved population heterogeneity across patient-level covariates.
CONCLUSIONS: These findings highlight that conventional linear regression methods applied to composite PROPr scores may underestimate variability and obscure meaningful differences in criminal legal system populations with OUD, where health and social functioning are highly heterogeneous. We argue that probabilistic simulation with domain-level covariate adjustment better captures distributional characteristics of HRQoL and domain-specific effects, including potential asymmetric covariate relationships across domains, and a more reliable alternative to standard methods for HRQoL estimation and subsequent cost-effectiveness analysis.