Improving the transportability of regression calibration under the main/external validation study design

This study aimed to improve the validity of regression calibration, a statistical method used to correct for covariate measurement error in studies that examine the relationship between exposure (e.g., diet) and outcomes (e.g., breast cancer incidence). Specifically, the authors focused on situations where external validation studies (EVS), which provide more accurately measured covariate data, are used alongside main studies (MS). The goal was to ensure that EVS data can be effectively applied to MS despite potential differences in key features between the two populations.

Measurement error is a common problem in epidemiology and leads to biased conclusions, usually underestimating the health effects of exposures. For example, food frequency questionnaires often measure actual dietary intake with moderate to substantial error, impacting studies on nutrition and health. Using EVS to correct for such errors is cost-effective, but certain differences between EVS and MS populations can limit their usefulness. This study addresses a critical gap by proposing a method that improves the applicability (transportability) of EVS data, potentially enhancing the reliability of research findings in fields such as nutrition and environmental health.

The researchers developed a new transportable regression calibration (TRC) method. Unlike traditional approaches, which rely solely on EVS data, TRC combines information from both EVS and MS to estimate the calibration models used for bias correction. They developed the method with theoretical mathematics, tested its behavior in simulations under various conditions (e.g., differing population characteristics), and in a real-world dataset from a large cohort study examining dietary intake and body weight.

The TRC method reduced bias in estimates when compared to traditional regression calibration when key differences existed between the EVS and MS study populations. Simulations demonstrated that TRC consistently provided accurate estimates under various conditions, such as high measurement error and non-normal data distributions. When applied to real-world data, TRC produced results more consistent with established scientific understanding, such as identifying protein intake as inversely associated with body weight, aligning with prior evidence.

The TRC method enables researchers to better utilize external validation studies, improving the accuracy of findings in studies where measurement error is a concern. This is particularly relevant in nutritional and environmental epidemiology, where reliable data is essential for public health recommendations. By addressing transportability issues, TRC expands the applicability of existing datasets, potentially reducing research costs and enhancing study designs.

The authors suggest extending the TRC method to other statistical models, such as generalized linear models and survival analyses, to broaden its application.