Devising a Missing Data Rule for a Quality of Life Questionnaire-A Simulation Study

Abstract

Objective: The aim of this study was to devise an evidence-based missing data rule for the Quality of Life Inventory-Disability (QI-Disability) questionnaire specifying how many missing items are permissible for domain and total scores to be calculated using simple imputation. We sought a straightforward rule that can be used in both research and clinical monitoring settings.

Method: A simulation study was conducted involving random selection of missing items from a complete data set of questionnaire responses. This comprised 520 children with intellectual disability from 5 diagnostic groups. We applied a simple imputation scheme, and the simulated distribution of errors induced by imputation was compared with the previously estimated standard error of measurement (SEM) for each domain.

Results: Using a stringent criterion, which requires that the 95th percentile of absolute error be less than the SEM, 1 missing item should be permitted for 2 of the 6 QI-Disability subdomain scores to be calculated and 1 missing item per domain for the total score to be calculated. Other, less stringent criteria would allow up to 2 missing items per domain.

Conclusion: Empirical evidence about the impact of imputing missing questionnaire responses can be gathered using simulation methods applied to a complete data set. We recommend that such evidence be used in devising a rule that specifies how many items can be imputed for a valid score to be calculated.