The delayed association in environment-health studies has been acknowledged in recent years. However, the misspecification of lag dimension in time-series models, particularly distributed lag non-linear models (DLNM), would induce considerable deviation of effect estimate. This study reviewed the existing climate-health English literature with time-series and case-crossover design published during 2000-2019 to summarize the statistical methodologies used and the reported delays of association between meteorological variables and 14 common causes of morbidity and mortality. Generalized linear or additive model was the most widely employed method for regression analysis. Different types of lag design were adopted for infectious disease modeling, including cross-correlation analysis, single lag model, moving average lag model, unconstrained or polynomial distributed lag model, and DLNM, whereas studies on non-communicable diseases predominantly used DLNM to assess the delays of association. For infectious outcomes, the association of daily mean temperature was found to be lagged for one to two weeks for influenza, followed by two to five weeks for diarrhea, and eight to twelve weeks for dengue fever. Meanwhile, the association of both cardiovascular and respiratory diseases with hot temperatures lasted for less than five days, whereas the association of cardiovascular diseases with cold temperatures was observed for ten to twenty days. Additionally, rainfall, as a potential risk factor for infectious diseases, showed a four to eight weeks’ lagged association with diarrheal diseases, while the effect was further delayed to eight to twelve weeks for vector-borne diseases. This is the first systematic review that comprehensively provides epidemiological evidence on the delay of association of common meteorological parameters. Biologically plausible and reasonable definition of effect lag in the modeling process is warranted in further environmental epidemiological studies.
Read full abstract