[Paper Review] A varying-coefficient model for characterizing duration-driven heterogeneity in flood-related health impacts

Sarika Aggarwal, Phillip B. Nicol+2 more|arXiv (Cornell University)|Jan 26, 2026

Agricultural risk and resilience0 citations

TL;DR

The paper develops an exposure duration varying-coefficient model (EDVCM) to estimate how flood exposure effects on hospitalizations vary by flood duration and exposure day, using a Bayesian Gaussian-process–regularized framework within a self-matched area-level design, and applies it to nationwide Medicare data (2000–2016).

ABSTRACT

Previous work revealed associations between flood exposure and adverse health outcomes during and in the aftermath of flood events. Floods are highly heterogeneous events, largely owing to vast differences in flood durations, i.e., flash-floods versus slow-moving floods. However, little to no work has incorporated exposure duration into the modeling of flood-related health impacts or has investigated duration-driven effect heterogeneity. To address this gap, we propose an exposure duration varying coefficient modeling (EDVCM) framework for estimating exposure day-specific health effects of consecutive-day environmental exposures that vary in duration. We develop the EDVCM within an area-level self-matched study design to eliminate time-invariant confounding followed by conditional Poisson regression modeling for exposure effect estimation and adjustment of time-varying confounders. Using a Bayesian framework, we introduce duration- and exposure day-specific exposure coefficients within the conditional Poisson model and assign them a two-dimensional Gaussian process prior to allow for sharing of information across both duration and exposure day. This approach enables highly-resolved insights into duration-driven effect heterogeneity while ensuring model stability through information sharing. Through simulations, we demonstrate that the EDVCM out-performs conventional approaches in terms of both effect estimation and uncertainty quantification. We apply the EDVCM to nationwide, multi-decade Medicare claims data linked with high-resolution flood exposure measures to investigate duration-driven heterogeneity in flood effects on musculoskeletal system disease hospitalizations.

Motivation & Objective

• Motivate and address duration-driven heterogeneity in flood-related health impacts.
• Develop a varying-coefficient modeling framework that jointly models duration and exposure day effects.
• Incorporate information sharing across duration and day via Gaussian process priors for stable estimation.
• Implement within an area-level self-matched design to control time-invariant confounding.
• Apply the method to Medicare data linked with high-resolution flood exposure to characterize heterogeneous effects.

Proposed method

• Formulate a Poisson regression for county-day hospitalizations with stratum-specific intercepts conditioned out to obtain a multinomial likelihood.
• Model exposure effects with coefficients beta(d,t) for duration d and day t, and lag effects theta(d,l) for lag l, where d and t (and l) are integers.
• Impose a two-dimensional Gaussian process prior on the vector of beta coefficients with a separable product kernel: Cov(beta(d,t), beta(d',t')) = sigma_beta^2 exp(-|d-d'|/phi) exp(-|t-t'|/tau).
• Impose a similar GP prior for lag coefficients theta(d,l) with kernels across duration and lag: Cov(theta(d,l), theta(d',l')) = sigma_theta^2 exp(-|d-d'|/gamma) exp(-|l-l'|/eta).
• Use weakly informative priors and Hamiltonian Monte Carlo to sample from the posterior; report posterior means and 95% credible intervals.
• Provide a cumulative rate ratio measure by averaging exponentiated betas (with covariate adjustment as appropriate) for summarizing flood duration effects.

Figure 1 : County-level flood characteristics in the contiguous United States from 2000–2016 for the data application. Panel A shows the number of flood events by county and Panel B shows the average flood duration (in days) by county. State borders are shown in black while county borders are shown

Experimental results

Research questions

• RQ1 Do flood effects on hospitalization vary with the length of the flood event (duration)?
• RQ2 Does the duration of flood exposure alter the timing of the most vulnerable window (critical- window) after exposure?
• RQ3 How do post-flood lag effects interact with flood duration in shaping health outcomes?
• RQ4 Can information be shared across durations and days to stabilize estimation when data are sparse for long durations?
• RQ5 How does the EDVCM perform relative to conventional methods in estimating duration-day coefficients and their uncertainty?

Key findings

• For musculoskeletal hospitalizations, floods of duration 6–9 days showed larger adverse effects, with the largest effect on the last day of a 6-day flood (RR = 1.28, 95% CI 1.25–1.32).
• Protective effects were observed on certain days for some shorter durations (e.g., day 5 of a 5-day flood RR = 0.79, 95% CI 0.77–0.81).
• Adverse effects were concentrated in the latter days of each duration, indicating a window of vulnerability.
• Longer flood durations tended to have higher peak effect estimates than shorter durations (e.g., 7-day flood RR = 1.21, 95% CI 1.17–1.25 vs 2-day flood RR = 1.02, 95% CI 1.00–1.05).
• The main results are summarized in Table 1: cumulative rate ratios by duration, with null value 1 across durations.

Figure 2 : Ground truth $\beta_{d,t}$ duration-day coefficients in the smooth and noisy scenarios.