Nonlinear effect of urban noise pollution on depression of the elderly in China based on the Bayesian machine learning method

Abstract

With the intensification of urbanisation and population ageing globally, it is crucial to understand the effect of urban noise pollution on the mental health of the ageing population. This research employed the Bayesian machine learning method to investigate the nonlinear effect of urban noise pollution on depression scores in the elderly (DSE) population in China. Based on individual survey data, the China Health and Retirement Longitudinal Study (CHARLS) data from 2011 to 2020, and corresponding temporally monitored annual urban equivalent sound pressure level (ESPL) data in China, propensity score matching (PSM) was used to identify the causal effect of urban ESPL on DSE. The quantile regression model (QRM), the Bayesian additive regression tree model (BARTM), and the Bayesian causal forest method (BCFM) were used to detect the nonlinear influencing mechanism of urban ESPL on DSE. The PSM analysis which can balance confounding influence revealed a significant disparity in depression scores between ageing individuals exposed to higher ESPL and those residing in more tranquil settings, suggesting a definitive causal relationship. Furthermore, the QRM results highlight a significantly increasing effect of ESPL on DSE at higher quantiles (0.4528 and 0.5709 at the 75th and 85th percentiles, respectively). The BARTM identifies a ‘U-shaped’ dose–response curve between ESPL and DSE, with low ESPL as beneficial, but exposure above 53.67 dB(A) representing noise pollution, especially past 55.85 dB(A), significantly increasing DSE. Specifically, when ESPL ranges from 53.67 to 55.68 dB(A) with other variables held constant, DSE increases by an average of 0.28 for every 1 dB(A) rise. Beyond 55.68 dB(A), the average DSE increase is 0.85 for each additional 1 dB(A) in ESPL. A notable spatial heterogeneity was observed in the local effects of annual urban ESPL on DSE. The BCFM results indicate that socio-economic exposure level can modulate the effect of ESPL on DSE.