Sparse item testing of clinical scales in neurology trials to alleviate burden to patients.

Abstract

Neurology trials typically rely on composite scales for measuring symptom severity. Completing all items in a long scale can be burdensome for patients, caregivers, and trial personnel. To test the hypothesis that sparse item testing, aided by item-response modelling, can preserve the power for detecting treatment effect in a controlled trial. UPDRS (Unified Parkinson's Disease Rating Scale) Part III (motor examinations) data from a placebo-controlled trial (N = 391) of ropinirole were analysed with a longitudinal item-response model. Symptom severity was estimated directly from item scores as a latent variable, without needing the total score. This enabled sparse item testing. With the symptom severity as a clinical endpoint, the potential power loss for detecting treatment effect due to the sparse testing was assessed by simulation. When each patient took 18 of all 27 tests in UPDRS Part III at each study visit, there was no appreciable power loss. Reducing four visits to three also had negligible effects on power. A threefold reduction of the total tests that each patient needed to do throughout the trial, from 108 to 27, only compromised power slightly, e.g., from 92 to 87% at N = 160. These findings show that using the symptom severity derived from item scores as the endpoint allows sparse testing to drastically reduce trial burden without incurring major power loss. This benefit would multiply for indications like Alzheimer's disease where modern trials often require patients to be tested on multiple scales at several times.