Data-Enhancement Strategies in Weather-Related Health Studies

Abstract

Although the relationship between weather and health is widely studied, there are still gaps in this knowledge. The present paper proposes data transformation as a way to address these gaps and discusses four different strategies designed to study particular aspects of a weather-health relationship, including (i) temporally aggregating the series, (ii) decomposing the different time scales of the data by empirical model decomposition, (iii) disaggregating the exposure series by considering the whole daily temperature curve as a single function, and (iv) considering the whole year of data as a single, continuous function. These four strategies allow studying non-conventional aspects of the mortality-temperature relationship by retrieving non-dominant time scale from data and allow to study the impact of the time of occurrence of particular event. A real-world case study of temperature-related cardiovascular mortality in the city of Montreal, Canada illustrates that these strategies can shed new lights on the relationship and outlines their strengths and weaknesses. A cross-validation comparison shows that the flexibility of functional regression used in strategies (iii) and (iv) allows a good fit of temperature-related mortality. These strategies can help understanding more accurately climate-related health.

Keywords: Canada; aggregation; empirical mode decomposition (EMD); environment; epidemiology; functional regression; health; time series; weather.

Figure 1

Estimated overall cumulative relative risk (RR) of temperature for the classical model and the first strategy (AG strategy). Dashed lines represent 95% confidence intervals.

Figure 2

Relative risks (RR) of cardiovascular mortality associated to the temperature intrinsic mode functions (IMF) kept by the Lasso versus to the mean period of the IMF. Blue bars indicate 95% confidence intervals.

Figure 3

Estimated overall relationship between the cardiovascular mortality and temperature across the year. Dashed lines indicate 95% confidence intervals. This overall relationship is obtained by summing the functional coefficient along the lag dimension. Note that the seemingly low values of the relative risk (RR) are explained by its continuous nature (the relationship is spread across the whole curve).

Figure 4

Estimated relationship between cardiovascular mortality count and the previous day temperature. From left to right, hours correspond to midnight of previous day to midnight of current day. Dashed lines indicate 95% confidence intervals. Seemingly low values of relative risk (RR) are due to the spreading of the risk along the whole day.

Figure 5

Cross-validated relative RMSE (rRMSE) along the year for each strategy and the benchmark model. The rRMSE is defined as the square root of the mean square prediction error divided by the mean of the raw response. In this figure, the computed rRMSE is smoothed by locally weighted regression (LOESS). DLNM, distributed lag nonlinear model; AG, aggregation of response; EMDR, EMD-regression; FD, functional with daily curves; FY, functional at the yearly level.

Figure 6

Summary of the cases of interest for each strategy. The abscissa indicates the scale corresponding to the objective of the study and the ordinate to the issues potentially present in the data. DLNM, distributed lag nonlinear model; GAM, generalized additive model; AG, aggregation of response; EMDR, EMD-regression; FD, functional with daily curves; FY, functional at the yearly level.