A linear approach is the simplest way to model the relationship between an exposure and an outcome. But a linear approach is limited in only capturing the association that is either strictly increasing or strictly decreasing. This might not adequately represent the associations at the extremes of the distribution for an exposure. For example, in Nguyen et al. 2021, we found that modeling a linear association between an exposure, body mass index (BMI) and an outcome, all-cause mortality resulted in a null relationship (i.e. there is no association between BMI and mortality risk.) This is nonsensical as we intuitively understand that participants with the lowest and highest BMI are at increased risk for all-cause mortality. Our intuition was confirmed when we considered non-linear associations between BMI and all-cause mortality. Furthermore, we observed that the prediction performance for the linear model were on par with those of the non-linear models, but this creates challenges in selecting the most appropriate model. Thus, we developed a visualization tool, called the stairway plots, to compare the linear and non-linear shapes of the associations to help select the most appropriate model. In Nguyen et al. 2021, we used the stairway plots to characterize the non-linear associations between physiological indicators and all-cause mortality. These plots enable us to assess the relevance of the clinical thresholds in differentiating patients who are at high vs. low risk for mortality. Now, we are currently using this visualization tool to study the non-linear associations between chemical exposures and cancer-specific mortality to quantify the cancer mortality risk found at human relevant doses. We plan to deploy this tool as an R package to enable the characterization of linear vs non-linear associations between any exposures and any outcomes of interest.

This presentation will be given remotely, with the livestream available for group viewing in 2036 Palmer Commons. There will also be a remote viewing option via Zoom.

URL for remote viewing: https://umich-health.zoom.us/j/94886745590?pwd=LzhLU243K2ZhbXNzU1BJRHQ5V25BZz09