[Effect of combination model on fitting cancer mortality and prediction]

Objective: To reduce the cancer burden in the Jinchang cohort and provide evidence for developing cancer prevention strategies and performing effectiveness evaluation in the Jinchang cohort. We are fitting thirteen years of cancer mortality data from the Jinchang cohort by using six kinds of predicting methods to compare relative fitness and to select good predicting methods for the prediction of cancer mortality trends. Methods: The mortality data of cancer in Jinchnag cohort from 2001-2013 were fitted using six kinds of predicting methods: dynamic series, linear regression, exponential smoothing, autoregressive integrated moving average (ARIMA) model, grey model (GM), and Joinpoint regression. Weight coefficients of combination models were calculated by four methods: the arithmetic average method, the variance inverse method, the mean square error inverse method, and the simple weighted average method. Results: The cancer mortality was fitted and compared by using six kinds of forecasting methods; the fitting precision of the Joinpoint linear regression had the highest accuracy (87.64%), followed by linear regression (87.32%), the dynamic series (86.99%), GM (1, 1) (86.25%), exponential smoothing (85.72%) and ARIMA (1, 0, 0) (81.98%), respectively. Prediction accuracy of the combination model derived from GM (1, 1) and linear regression (>99%) was higher than that of the combination model derived from ARIMA (1, 0, 0) and GM (1, 1). The combination model derived from the GM (1, 1) and linear regression, with weight coefficients based on the arithmetic average method and the mean square error inverse method, had the best prediction effect of the four weight calculation methods. Conclusion: Prediction accuracy of the combination model, with accuracy >95%, was higher than that of the single prediction methods.