Abstract: Objective To explore the feasibility of autoregressive integrated moving average (ARIMA) model in predicting clinical platelet demand, and to provide basis for scientifically making the plan of blood collection and recruitment. Methods ARIMA model was established for the clinical usage of apheresis platelets monthly from January 2006 to June 2016 in Chongqing Central Blood Stations. The optimal model was used to predict the monthly clinical usage of apheresis platelets from July to December 2016. Results The optimal model for clinical usage of apheresis platelets was ARIMA(0,1,1)(1,0,1)₁₂. The model residual sequence autocorrelation function and partial autocorrelation function were basically within 95% confidence interval, and the Ljung-Box Q statistical results showed that there was no correlation between the residual error (P>0.05), indicating that the residual error sequence was white noise and the model passed the test. The actual values and the predicted values of the model were in the 95% confidence interval. Compared with the actual values of the clinical usage of apheresis platelets during the same period, the trend of curve change was basically the same, which indicates the prediction accuracy was high with a 7.5% mean relative error. Conclusion The optimal model ARIMA(0,1,1)(1,0,1)₁₂ can fit better the trend of clinical usage of apheresis platelets in time series.

Key words: autoregressive integrated moving average model, apheresis platelets, clinical demand prediction, Central Blood Stations