Evaluating the use of prior information to individualise start item selection for the EORTC CAT core

Computerized adaptive tests (CATs) provide individualised measurement, using score estimates based on the patient’s prior responses to select the next most informative item. However, as no score estimate is available at the outset, the start item is typically not individualised. The European Organisation for Research and Treatment of Cancer (EORTC) CAT Core covers 15 health-related quality of life (HRQoL) domains. We explored whether scores from one domain could be used to obtain initial score estimates and hence, individualised start items for another domain, thereby improving measurement. For each HRQoL domain, we evaluated the ability to predict scores using each of the 14 other domains in a large international sample of cancer patients (N = 10,084). Using simulations, we compared the impact of individualised versus standard fixed start item on CAT measurement precision. Across domains, predicted scores were within one standard deviation of the observed score in 72–89% of the assessments (mean = 83%), with predicted-observed correlations ranging 0.31–0.72 (mean = 0.55). The impact of individualised start items varied by domain and score level but typically improved reliability in the initial steps of CATs (first 3 items), particularly for patients with extreme scores. Cross-domain predictions can be used to generate initial score estimates for individualised start item selection. Simulations suggested that individualised start items lead to improvements in measurement precision, particularly for short CATs (up to 3 items) and patients with extreme scores. Individualised start items based on cross-domain predicted scores is planned to be incorporated into the EORTC CAT Core toolbox. The study investigated how to make dynamic health questionnaires smarter and more personalised for people with cancer. Dynamic questionnaires (called CATs) adapt to the individual based on the person’s answers to questions. However, the first question of such questionnaires is the same for everyone because there is no information yet about the individual. The aim was to explore whether the score for one health aspect (like fatigue or pain) could provide a ‘good guess’ about the score for another aspect (like physical functioning). If so, such estimates could be used to tailor the first question to the individual thereby hopefully make assessments more relevant and accurate. This was explored for a dynamic health questionnaire, the EORTC CAT Core. Using data from over 10,000 cancer patients our hypothesis was confirmed: the new approach resulted in start questions, that were more relevant for the individual (compared to starting with the same question for all) and are therefore likely to improve assessment relevance, accuracy, and efficiency.