A causal approach to analysis of censored medical costs in the presence of time-varying treatment

TL;DR

This paper introduces a causal inference method using nested g-computation to analyze censored medical costs with time-varying treatments, addressing a gap in existing approaches for policy-relevant cost comparisons.

Contribution

It develops a novel nested g-computation approach that handles time-dependent treatments and confounding in censored cost data, improving causal effect estimation.

Findings

Method is robust to distributional assumption violations.

Provides insights into cost differences across treatment regimes.

Addresses a gap in causal analysis of censored medical costs.

Abstract

There has recently been a growing interest in the development of statistical methods to compare medical costs between treatment groups. When cumulative cost is the outcome of interest, right-censoring poses the challenge of informative missingness due to heterogeneity in the rates of cost accumulation across subjects. Existing approaches seeking to address the challenge of informative cost trajectories typically rely on inverse probability weighting and target a net "intent-to-treat" effect. However, no approaches capable of handling time-dependent treatment and confounding in this setting have been developed to date. A method to estimate the joint causal effect of a treatment regime on cost would be of value to inform public policy when comparing interventions. In this paper, we develop a nested g-computation approach to cost analysis in order to accommodate time-dependent treatment and repeated outcome measures. We demonstrate that our procedure is reasonably robust to departures from its distributional assumptions and can provide unique insights into fundamental differences in average cost across time-dependent treatment regimes.