Comparative Study on Estimation Methods of Proportional Hazard Models for Interval-Censored Data

Abstract

Purpose: In this study, we compare the estimation methods of interval-censored data using both simulated and real data. Many past studies have used fixed sample sizes in their simulation studies. We performed the best possible simulation study. Method: The methods include Finkelstein’s method with Piecewise and Spline and imputation methods (i.e., Efron’s method in the Cox model). Results: If the interval-censored data do not overlap, the same estimation results are obtained regardless of the assignment point for the estimation of the Cox model. The overlapping data also did not significantly affect the accuracy of the estimation. On the other hand, Finkelstein’s method showed differences in estimation depending on the two estimation methods of the baseline survival function. Although it was not possible to determine which method had better power, the Spline method had a smaller absolute error than the Finkelstein method. A comparison of Cox’s and Finkelstein’s methods showed that Finkelstein’s method was superior in terms of power. Conclusion: Interval-censored data is a form of data that can be found in a variety of fields. In this study, we compared estimation methods for interval-censored data, and the usefulness of Finkelstein’s method can be seen from simulation studies.