EXAMINING MODIFICATION OF THE EFFECT OF TEMPERATURE ON MORTALITY IN MEXICO WITH THE CASE-ONLY DESIGN

Funding: 5 T32 ES07069-22, NIEHS ES00002, and EPAR827353, UK Medical Research Council G9900947 The case-only study design is commonly used in evaluating gene-environment interactions. This approach can also be applied as an alternative to time series regression for evaluating modification of the effects of temperature on mortality. The idea is to evaluate whether, among those who died (cases), temperature is associated with a time-fixed factor of interest as a possible effect modifier. Case-only results tell us whether an interaction between temperature and age exists but cannot estimate the main effect of temperature on mortality. However, these results give insights into who is most vulnerable to the health effects of temperature extremes, insights which can be used to understand biological mechanisms and/or guide preventive actions. The case-only design has been infrequently applied to evaluate temperature and mortality, and some methodological questions on its limitations need further exploration. One key assumption of the standard case-only design is that the effect modifier of interest is not also a modifier of the effect of another exposure (e.g., PM10). Practically, the validity of this assumption can be tested by including interaction terms for the effect modifier with both potential exposures (temperature/age, and PM10/age) together in a model, and evaluating whether the inclusion changes the coefficient of the corresponding term, compared to the one-interaction term model. We tested this asssumption using data from Mexico City and Monterrey, Mexico on apparent temperature, PM10, and mortality, stratified by cause (CVD, respiratory) and age (>65, 15–64, under 15). To apply the case-only approach, we fit polytomous regression models on age group (Y variable) with the generalized logit link function and ran models with temperature/age and PM10/age interactions alone and together. In Mexico City, a PM10/age interaction was present for CVD deaths, not affected by including the temperature/age interaction, but no effect was seen for a temperature/age interaction, and no change of that null effect was observed when a PM10/age interaction was included. For respiratory-cause deaths, the PM10/age interaction decreased by 47% (for >65 compared to under 15) when a temperature/age effect with inclusion of PM10/age interactions. In Monterrey for CVD deaths, neither interaction term was affected by the other in these models. For repiratory deaths, the PM10/age interaction was reduced by 46% for >65 compared to under 15, with inclusion of the temperature/age interaction, but the temperature/age term was not substantially changed in a two interaction term model. Implementation of the case-only approach for evaluating modification of temperature or air pollution effect on daily events (mortality or hospital admissions) should evaluate sensitivity of the results to the inclusion of more than on interaction term.