Net effects of correlated determinants of health among U.S. older adults

The health effect produced by a contributing factor alone may be different from the effect when other interacting factors are present. The effects of a set of contributors, estimated by regression coefficients, may be invalid or inappropriate if some of the predictors are correlated or the model has interaction and nonlinear terms. The net effect (NE) of a given contributing factor is the effect produced by the factor alone, not by other factors due to their correlations with the factor of interest. This study estimated NE of selected determinants of health for U.S. older adults.

Participants, aged ≥ 65 years, were from the baseline 2020 Medicare Health Outcome Survey (N = 196,641). We examined the NE of each explanatory variable from five specified categories on four health outcomes: physical component summary scores (PCS), mental component summary scores (MCS), health utilities (SF-6D), and 2-year mortality. The five categories of explanatory variables included: six difficulties in activities of daily living (ADLs), three difficulties in instrumental activities of daily living (IADLs), five geriatric syndromes, five demographics, and fifteen chronic conditions. The NE for a given factor was calculated as the average of the factor's marginal effect over all possible values of other factors.

After adjusting for age, race, and sex, difficulty in walking contributed the greatest decrement in PCS (NE= -7.4) among the 6 ADLs. Other ADL decrements were 4.5-points for bathing, 4.4-points for getting in/out of chairs, 2.7-points for dressing, 1.5-points for using the toilet, and 1.3-points for eating. For MCS, difficulty in eating contributed the greatest decrement (-4.9), followed by bathing (-3.9) and walking (-3.2). For the SF-6D, difficulty in walking had a NE of -0.078 while difficulty in bathing had the greatest increment for 2-year mortality (8.1%). The NEs for the explanatory variables in the other categories are presented.

This study introduced NE as a measure of effect produced by each determinant of health and demonstrated how to estimate the net effect for various types of data. Our results highlight how the net effect varies within a given category, and differs according to physical and mental health and mortality.