A Bayesian framework for cost-effectiveness analysis with time-varying treatment decisions

TL;DR

The paper develops a Bayesian joint modeling framework for cost-effectiveness analysis with time-varying treatment decisions in observational claims data. By modeling continuous-time gap times and encounter-level costs and employing Bayesian g-computation, it enables causal evaluation of dynamic treatment regimes through estimands like the net monetary benefit $igl( ext{NMB}=E[\kappa T(sd)-U(sd)]igr)$ and regime contrasts $igl(\Psi_{sd,sd'}(\kappa)igr)$. Simulation studies show the method robustly recovers causal quantities under censoring and outperforms discrete-time or misspecified models, while an application to high-risk, early-stage endometrial cancer (SEER-Medicare) yields limited evidence of substantial cost-effectiveness differences between EBRT and VBT over three years. The framework thus offers a practically viable approach for policy-relevant CEAs with time-varying decisions, irregular observation times, and administrative censoring.

Abstract

Cost-effectiveness analyses (CEAs) compare the costs and health outcomes of treatment regimes to inform medical decisions. With observational claims data, CEAs must address nonrandom treatment assignment, administrative censoring, and irregularly spaced medical visits that reflect the continuous timing of care and treatment initiation. In high-risk, early-stage endometrial cancer (HR-EC), adjuvant radiation is initiated at patient-specific times following hysterectomy, causing confounding between treatment and outcomes that can evolve with post-surgical recovery and clinical course. Most existing CEA methods use point-treatment or discrete-time models. However, point-treatment approaches break down with time-varying confounding, while discrete-time models bin continuous time, expand the data into a person-period format, and can induce zero-inflation by creating many intervals with no cost-accruing events. We propose a Bayesian framework for CEAs with sequential decision-making that jointly models costs and event times in continuous time, accounts for administrative censoring, and supports dynamic treatment regimes with minimal parametric assumptions. We use Bayesian g-computation to estimate causally interpretable cost-effectiveness measures, including net monetary benefit, and to compare regimes through posterior contrasts. We evaluate the finite-sample performance of the proposed method in simulations across censoring levels and compare it against discrete-time and fully parametric alternatives. We then use SEER-Medicare data to assess the cost-effectiveness of initiating adjuvant radiation therapy within six months following hysterectomy among HR-EC patients.

Figures (6)

• Figure 1: Illustration of cost accrual in continuous time for two patients ($i = 1, 2$) with $J_i = 2$ post-surgery encounters. The left panel shows observed calendar times and associated costs, with nonterminal encounters indicated by circles ($\bullet$, $\bullet$). Patient 1 is observed at $V_{11} = 1$, while Patient 2 is observed at $V_{21} = 2.5$, with costs $Y_{11} = 12.75$ and $Y_{21} = 1.5$. Terminal events occur at death ($\blacksquare$) for Patient 1 at $T_1 = 5.75$ and censoring ($\times$) for Patient 2 at $C_2 = 8$, with costs $Y_{12} = 3.5$ and $Y_{22} = 5.5$. The right panel shows cumulative cost accrual as a step function over time, where solid lines represent accumulated total cost and dashed lines indicate jumps at observed encounters. Gap times between events are $W_{11} = V_{11} - V_{10} = 1$, $W_{12} = T_1 - V_{11} = 4.75$, $W_{21} = V_{21} - V_{20} = 2.5$, and $W_{22} = C_2 - V_{21} = 5.5$.
• Figure 2: Directed acyclic graph for $J = 2$ encounters illustrating time ordering and confounding, where baseline covariates $\bsL_0$ and treatment-readiness indicators $Z_j$ are omitted for clarity.
• Figure 3: Posterior summaries for EBRT versus VBT. Left: posterior mean and 95% credible interval for the NMB $\widehat{\Psi}_{d_1, d_2}(\kappa)$ across WTP thresholds $\kappa$ (NMB in 100,000 USD/year). Center: posterior density of three-year restricted mean survival time (RMST; years). Right: posterior density of three-year mean total cost (MC; 100,000 USD). Dashed vertical lines denote posterior means.
• Figure 4: Posterior predictive cumulative cost trajectories for a representative stage IA patient under EBRT and VBT over three years. Thin gray curves are simulated trajectories from posterior draws; solid curves are posterior means with 95% credible bands.
• Figure 5: Kernel density estimates of log-transformed encounter costs, $\log Y_j$, stratified by event type: non-death encounters $\delta_j \in \qty{0, 1}$ versus death encounters $\delta_j = 2$.