2020 年 2 巻 論文ID: 2020.001
Interrupted time series analysis (ITSA) is generally performed in evaluating the effect of a health policy.Although segmented regression analysis methods are the standard methods for ITSA and are used to fit linearmodels to the data, these methods are oversimplified for real data. Further, ITSA data are often count time seriesdata that total the number of cases of interest. Although several methods of analysis have been proposed and usedfor an ITSA, methods using a count time series model have merely been discussed and used. Thus, we propose touse a count time series model, the integer-valued generalized autoregressive conditional heteroscedastic(INGARCH) model, as an ITSA to evaluate a health policy. The INGARCH model is a count time series modelwhich can model more complicated time series data than linear regression models; moreover, it has not beencompared with segmented regression analysis methods. We applied the INGARCH model and segmentedregression analysis methods (Poisson regression analysis (PREG) and generalized least squares (GLS)) to realpsychotropic drug prescription data in Japan and then discussed the statistical behavior of these methods. We usedpsychotropic drug prescription data from a hospital in Japan for the ITSA. Several administrative policies for theprevention of multidrug use of psychotropic drugs have been enforced in recent years, and we evaluated theeffects of the policies on the four types of psychotropic medicines (antidepressants, antipsychotic drugs, anxiolytics,and sleeping drugs). The test results differed according to the methods of analysis used. Segmented regressionanalysis methods by GLS and PREG fit linear regression models to the data but did not necessarily model the realtime series data well. Conversely, INGARCH could model the more complicated time series behavior; thus, theresults suggested that INGARCH can model more various types of count time series than segmented regressionanalysis.
