Using prior information to individualize start item selection when assessing physical functioning with the EORTC CAT...

Computerized adaptive test (CAT) provides individualized measurement, using the patient's previous responses to select the next most informative item. However, the first item, the start item, is usually not individualized as no score estimate is available a priori. The European Organisation for Research and Treatment of Cancer (EORTC) CAT Core covers 15 health-related quality of life domains. Scores for one domain may be used to obtain initial score estimates for another domain. We assessed the potential for using such cross-domain information to individualize start item selection for the EORTC CAT Core physical functioning.

The potential for predicting physical functioning (PF) scores from each of the 14 other domains using linear regression was assessed in an international, mixed sample comprising 10,084 cancer patient assessments. Using Monte Carlo CAT simulations, the impact of individually selected PF start items vs. fixed start item for CAT measurement precision was assessed.

Depending on the domain predicting PF, the correlation of predicted and observed PF scores ranged 0.25-0.71 and the predicted PF scores were within 1SD of the observed PF scores for 57-85% of the patients. The CAT simulations showed that individually selected start items improved measurement precision for the initial steps of CATs. The application of individual start items had trivial or no impact on measurement precision when the CAT asked three or more items.

Simple linear regression may provide useful cross-domain predictions. Using individualized start items may increase measurement precision of the EORTC CAT Core for the initial steps of CAT which may be of relevance for short CATs.