Estimation and prediction system for multi-state disease process: Application to analysis of organized screening regime

Rationale, aims and objectives: The disease progression of cancer and non-malignant chronic disease often involve a multi-state transition. However, estimation of parameters and prediction regarding the multi-state disease process are complex. This study aimed to develop an estimation and prediction system with a computer-assisted software using SAS/SCL as a platform to predict the risk of any outcome arising from the underlying multi-state process with or without the incorporation of individual characteristics. Method: The computer-aided system is first constructed following the theoretical framework of stochastic process. The functions provided in this software include model specification, formulation of likelihood function, parameter estimation, model validation and model prediction. An example of breast cancer screening for a high-risk group in Taiwan was used to demonstrate the usefulness of this software. Results: The natural history of breast cancer of a three-state disease process has been demonstrated. Two suspected risk factors, late age at first full-term pregnancy and obesity, were considered by the form of the proportional hazard model. Formulation of intensity matrix, likelihood function, assignment of initial values, and parameter constraint and estimation were successfully demonstrated in model specification. Model validation suggested a good fit of the constructed model. The application of model prediction enables one to project the effectiveness of organized screening by different inter-screening intervals from a policy level or from an individual basis. Conclusions: A computer-aided estimation and prediction system for multi-state disease process was developed and demonstrated. This system can be applied to data with the property of multi-state transitions in association with events or disease.