Background: Maternal mortality and racial disparities, key health indicators and quantitative metrics of high-priority issues in the US, have been drawing sustained attention from current academic research, public health service agencies, and policymakers. Most time series models assume data stationarity for valid estimates and future predictions. Limited attention has been paid to examine data properties before selecting models, and how research findings may affect the policymaking process. Our objective was to conduct a time series analysis for these health indicators to verify data properties, which is crucial for conducting analyses, valid inferences, and forecasting.
Methods: We performed a retrospective time series analysis of aggregate population-level data (1960-2016) from the US National Center for Health Statistics. Time plots were generated to analyze data points reported consistently every two years to identify broad behaviors and trends. The Augmented Dickey-Fuller (ADF) test, differencing, and transformation techniques were applied to verify data properties and to achieve stationarity.
Results: The analysis of data points revealed instability in patterns and drifting trends characteristic of a nonstationary process, which was confirmed by the ADF test. Data were transformed to achieve stationarity. Stationarized data are analyzed using standard statistical models for valid estimates and for predicting future values. Maternal mortality rates rise contingent on permanent effects of specific events or shocks which accumulate and integrate into future values, rather than dissipating over time.
Conclusions: Maternal health research should explicitly address and test time series data to verify data properties to avoid errors in estimates, invalid inferences, and misguiding strategic policy decisions.