Using a Sample Size Calculation Framework for Clinical Prediction Models When Developing and Selecting Mapping Algorithms Based on Linear Regression.

Abstract

To propose using a framework for calculating the sample size for clinical prediction models when developing and selecting mapping algorithms from a health-related quality-of-life (HRQOL) measure onto the score of a preference-based measure (PBM) using linear regression. The framework was summarized for health economics researchers. Mapping studies that mapped the European Organization for Research and Treatment of Cancer Quality of Life Questionnaire Core 30 onto the EQ-5D-3L index using linear regression were evaluated in terms of sample size. The required sample size for each study was calculated using 4 criteria: global shrinkage factor ≥ 0.9, difference between the apparent and adjusted R2 ≤ 0.05, multiplicative margin of error in the estimated residual standard deviation ≤ 1.1, and absolute margin of error in the estimated model intercept ≤ 0.025. Ten mapping studies were identified. The information required to calculate the sample size was successfully extracted from previous mapping studies. Four of 10 mapping studies did not have sufficient sample sizes. Further extension of this framework to other regression approaches used in mapping studies is necessary. The sample size should be considered when developing and selecting a mapping algorithm based on linear regression. No recommendation or guidance is available for the sample size to develop and select a mapping algorithm from a health-related quality-of-life measure onto the score of a preference-based measure.This research proposes using a framework for calculating the sample size for clinical prediction models in sample size consideration for mapping algorithms using linear regression.A survey showed that the information required to calculate the sample size could be successfully extracted from previous mapping studies and that 4 of 10 mapping studies did not have sufficient sample sizes.