Analytic approaches to assess the impact of local spending on sexually transmitted diseases

Abstract

Objective To compare the estimated associations between annual sexually transmitted diseases (STD) expenditures per capita and STD incidence rates among Florida and Washington local health departments (LHDs) from 2001 to 2017, using two approaches—a longitudinal regression model with lagged STD spending and a regression model with the Arellano–Bond panel estimator. Data sources Secondary data for LHDs were obtained from Florida and Washington state government offices and combined with county sociodemographic and health system data from the federal government. Study design We examined LHDs in Florida and Washington using a longitudinal panel study design to estimate ecological relationships between annual STD expenditures per capita and annual STD incidence rates from 2001 to 2017, with LHDs as the unit of analysis. We compared two regression models: generalized estimating equations (GEE) and the Arellano–Bond panel estimator (an instrumental variable approach). Data collection The secondary data were combined to build a longitudinal panel database for LHDs in Florida and Washington from 2001 to 2017. Principal findings In the GEE model with both states, greater STD spending in a prior year was associated unexpectedly with greater STD incidence rates in succeeding years. The Arellano–Bond models for both states had the expected inverse associations but were not significant. In the Arellano–Bond models for Florida, a $1 increase in STD spending in previous years was followed by decreases in STD incidence rates ranging between 29 and 58 points in succeeding years (0.09 ≥ p ≥ 0.04). Conclusions In longitudinal panel data for LHDs in two states, the Arellano–Bond estimator, or other instrumental variable approach, is preferred over conventional regression models to obtain unbiased estimates of the relationship between annual STD spending rates and annual STD rates. Future studies will require accurate, standardized, and detailed longitudinal data and rigorous analytic approaches, such as those illustrated in our study.

Association between maternal health service utilization and under-five mortality rate in China and its provinces, 1990–2017

“…In multivariable models, we controlled for year, years of schooling per capita, GDP per capita, disposable income per capita, birth rate, total number of health workers and emissions of particulate matter in exhaust gas. To strengthen the longitudinal analysis, we also examined the lagged effects [30][31][32] of prenatal visits and postnatal visits. To test the robustness of the results, in sensitivity analyses we replaced gamma distribution with Poisson distribution, gaussian distribution, and negative binomial distribution for the GEE model analysis, respectively.…”