Deep learning for predicting the occurrence of cardiopulmonary diseases in Nanjing, China.

摘要: The efficiency of disease prevention and medical care service necessitated the prediction incidence. However, predictive accuracy power were largely impeded in a complex system including multiple environmental stressors health outcome which occurrence might be episodic irregular time. In this study, we established four different deep learning (DL) models to capture inherent long-term dependencies sequences potential relationships among constituents by initiating with original input into representation at higher abstract level. We collected 504,555 786,324 hospital outpatient visits grouped categories respiratory (RESD) circulatory (CCD), respectively, Nanjing from 2013 through 2018. matched observations time-series that pose risk cardiopulmonary involved conventional air pollutants concentrations metrological conditions. results showed well-trained network architecture built upon long short-term memory block working day enhancer achieved optimal performance three quantitative statistics, i.e., 0.879 0.902 Nash-Sutcliffe efficiency, 0.921% 0.667% percent bias, 0.347 0.312 root mean square error-standard deviation ratio for RESD CCD visits, respectively. observed non-linear association nitrogen dioxide ambient temperature visits. Furthermore, these two identified as most sensitive variables, exerted synergetic effect outcomes, particular winter season. Our study indicated high-quality surveillance data atmospheric environments could provide novel opportunity anticipating temporal trend outcomes based on DL model.