Regression Analysis of Ordinal Panel Count Data in Recurrent Medication Non-adherence

TL;DR

This paper introduces a semiparametric proportional intensity model for analyzing ordinal panel count data, common in clinical trials, providing a robust and efficient method for inference and recurrence analysis.

Contribution

It develops a maximum sieve likelihood estimation approach using monotone splines for ordinal panel count data under a nonhomogeneous Poisson process model.

Findings

Method performs well with finite samples

Outperforms existing methods in simulations

Applied successfully to clinical trial data

Abstract

Panel count data arise in clinical trials when patients are asked to report their occurrences of events of interest periodically but the exact event times are unknown, only the count of events between two successive examinations are observed. Ordinal panel count data goes even further as the exact event counts are not observed, the only information available is rank of event counts, for example, 'never', 'sometimes' and 'always'. Currently, there is lacking of standard and efficient methods for analyzing this type of data. In this paper, we proposed a semiparametric proportional intensity model to analyze such data. We developed a maximum sieve likelihood estimation using monotone spline under the nonhomogeneous Poisson process model assumption for statistical inference. Simulation studies show that our method performs well with finite sample sizes and is relatively robust to model misspecification. In addition, we compared the proposed method with other competitors and the proposed method outperforms in various settings. Finally, we investigated the recurrence of medication non-adherence in a clinical trial on non-psychotic major depressive disorder using the proposed method.