Analysing the impact of health resource allocation on the ARD burden

We explored the impact of health resource allocation on the ARD burden using a panel data analysis approach. All equations were estimated with a log-linear functional form to enable unit-free comparisons of coefficients. To understand the underlying reasons for burden shifts over time, we used the age-standardised ARD burden rate as the dependent variable. Health expenditures and workforce density were adopted as proxy measures for health resources. Total health expenditure per capita was the key independent variable used to measure health expenditures.^{26 27} The key independent variables used to measure health workforce density were three separate sets of indicators: total health professional density, licensed doctor density and licensed nurse density, all per 1000 population. Health professionals included licensed doctors (clinical, dental, public health and traditional Chinese medicine), licensed nurses, pharmacists, clinical laboratory technicians and radiologists.³⁹ We chose three sets of indicators to measure health workforce density because (1) together they accurately represent the distribution of China’s health personnel resources; (2) they are widely used in published literature, and heterogeneity exists in terms of their impact on health outcomes;^{23 24} and (3) data stratified by province and by urban and rural areas are available for all three indicators. We ran three separate regression models to include the three health workforce density indicators.

We included GDP per capita, sex and education as covariates in the regression models to account for the major socioeconomic determinants of the population health burden. China has made remarkable achievements over the past decade in reducing the health disparities between urban and rural residents, primarily through improving maternal and child health and extending health insurance coverage, among others, for its rural residents.⁴⁰ However, there is still a noticeable urban–rural gap in health development, including access to quality health services, health workforce quantity and quality and health outcomes.^41–43 Therefore, to account for the significant gaps in urban–rural development and health across China, we included the percentage of people residing in urban areas (to measure the process of urbanisation) and the ratio of urban–rural health workforce density in the model as covariates. The model also included time dummies (to control period effects), province fixed effects and an error term (see online supplemental appendix for model details). We further explored the correlation of key variables to assess whether multicollinearity could undermine the robustness of the estimates for specific variables in the regression model (online supplemental tables S3 and S4). Standard errors were clustered at the provincial level. Province was the unit of analysis, and a fixed effects estimator was used to remove the potential endogeneity from time-invariant omitted variables.

Since provincial-level data on health expenditures per capita and health workforce density were only broadly available from 2010 onwards, our panel dataset included data from 2010 to 2016. We performed log-linear interpolation to obtain annual estimates of the ARD rate from 2010 to 2016, as the GBD Study only provides estimates in 5-year intervals.³⁸ The health expenditures per capita in 2010 are missing for seven provinces: Shanghai, Hainan, Sichuan, Tibet, Shaanxi, Qinghai and Ningxia. Only data on Tibet is missing in 2011. Therefore, after testing the robustness of the linear increase assumption, we imputed the missing data, assuming a linear increase in per capita health expenditures between 2010 and 2016. All analyses were performed in Stata V.16.0 (Stata Corp LLP, College Station, Texas, USA).