An Unconditional Test for Change Point Detection in Binary Sequences with Applications to Clinical Registries

Area of Science:

• Statistical methodology
• Biostatistics
• Change point analysis

Background:

• Change point detection in binary sequences has a long history, with prior research focusing on asymptotic and exact conditional methods.
• Existing exact conditional tests can be overly conservative, particularly with small sample sizes.

Purpose of the Study:

• To develop a novel exact unconditional test for change point detection in binary sequences.
• To demonstrate the superior power and efficiency of the new test compared to existing methods.

Main Methods:

• An exact unconditional test was developed by treating the total number of events as random.
• Efficient numerical calculation methods were devised.
• Adaptations of Berger and Boos methods were employed to address nuisance parameters.
• The new test was compared against Worsley's exact conditional test via Monte Carlo simulations.
• The method was applied to investigate changes in pin site infections in patients with external fixators.

Main Results:

• The unconditional test effectively controls the type I error rate at the nominal level.
• The new test demonstrates uniform superiority in power compared to Worsley's exact conditional test, especially in small sample scenarios.
• A significant beneficial effect of a new treatment procedure for pin site care was identified in the patient cohort.

Conclusions:

• The developed exact unconditional test is an effective and user-friendly tool for change point problems in binary sequences.
• The test is particularly recommended for small sample size applications.
• The methodology provides a valuable alternative to existing, more conservative, conditional tests.