A Statistical Framework for Single Subject Design with an Application in Post-stroke Rehabilitation

TL;DR

This paper introduces a Bayesian statistical framework for single subject design in post-stroke rehabilitation, enabling clinicians to determine if a patient’s behavior matches that of a healthy population through a novel null distribution approach.

Contribution

It presents a new Bayesian method using posterior predictive draws to assess single subjects against healthy population data, addressing a key clinical testing challenge.

Findings

Method performs satisfactorily in simulation studies

Provides error rate estimates and confidence intervals

Enables evidence-based clinical decision making

Abstract

This paper proposes a practical yet novel solution to a longstanding statistical testing problem regarding single subject design. In particular, we aim to resolve an important clinical question: does a new patient behave the same as one from a healthy population? This question cannot be answered using the traditional single subject design when only test subject information is used, nor can it be satisfactorily resolved by comparing a single-subject's data with the mean value of a healthy population without proper assessment of the impact of between and within subject variability. Here, we use Bayesian posterior predictive draws based on a training set of healthy subjects to generate a template null distribution of the statistic of interest to test whether the test subject belongs to the healthy population. This method also provides an estimate of the error rate associated with the decision and provides a confidence interval for the point estimate of interest. Taken together, this information will enable clinicians to conduct evidence-based clinical decision making by directly comparing the observed measures with a precalculated null distribution for such measures. Simulation studies show that the proposed test performs satisfactorily under controlled conditions.