Comparing disease specific catastrophic cost estimates using longitudinal and cross-sectional designs: The example of...

There has been an increasing interest in assessing disease-specific catastrophic costs incurred by affected households as part of economic evaluations and to inform joint social/health policies for vulnerable groups. Although the longitudinal study design is the gold standard for estimating disease-specific household costs, many assessments are implemented with a cross-sectional design for pragmatic reasons. We aimed at identifying the potential biases of a cross-sectional design for estimating household cost, using the example of tuberculosis (TB), and exploring optimal approaches for sampling and interpolating cross-sectional cost data to estimate household costs.

Data on patient incurred costs, household income and coping strategies were collected from TB patients in Negros Occidental and Cebu in the Philippines between November 2018 and October 2020. The data collection tools were developed by adapting WHO Tuberculosis Patient Cost Surveys: A Handbook into a longitudinal study design. TB-specific catastrophic cost estimates were compared between longitudinal and simulated cross-sectional designs using different random samples from different times points in treatment (intensive and continuation phases).

A total of 530 adult TB patients were enrolled upon TB diagnosis in this study. Using the longitudinal design, the catastrophic cost estimate for TB-affected households was 69 % using the output approach. The catastrophic cost estimates with the simulated cross-sectional design were affected by the reduction and recovery in household income during the episode of TB care and ranged from 40 to 55 %.

Using longitudinally collected costs incurred by TB-affected households, we illustrated the potential limitations and implications of estimating household costs using a cross-sectional design. Not capturing changes in household income at multiple time points during the episode of the disease and estimating from inappropriate samples may result in biases that underestimates catastrophic cost.