Comparison of complete-case and multiple imputation analysis in the investigation of the prognostic significance of parental reports of "asthma" and "wheeze" in kindergarten children

Abstract

Background: Missing data can be a source of selection bias in epidemiological studies if analysis is limited to cases with complete data and the data are not missing completely at random (MCAR). Multiple imputation is one method that has been recommended to overcome this bias. Aim: To compare the findings of complete-case and imputed-data analyses in examining whether parental-report of asthma and wheeze in the Longitudinal Study of Australian Children (LSAC) had any independent prognostic significance for future urgent health care utilisation in children. Methods:Children with parent-reported ever-doctor-diagnosed asthma were identified from Wave 1 of the LSAC kindergarten cohort (age 4-5 years). Children were classified as having “current wheeze” if their parent reported they had wheeze that lasted for a week or more in the preceding 12 months. Urgent health care utilisation at 2-year follow-up was defined as: any hospitalisations/ED attendances/>6 GP visits in the previous 12 months. We used Markov Chain Monte Carlo (MCMC) multiple imputation to create five imputed datasets. Odds ratios (ORs) were calculated for the complete-case and the five imputed datasets. A summary OR was derived from the five imputed-data estimates. Results:4,464 children were followed-up and 3,414 had complete data for all included variables. In the complete-case analysis of children with ever-diagnosed asthma at baseline, those who had current wheeze had a significantly higher risk of urgent health care utilisation in the 12 months before follow-up (OR=1.53 95% CI:1.12–2.10). Similar results were found for the imputed analysis (OR=1.49 95% CI:1.05–2.10). Conclusions Among children with diagnosed asthma, those with current wheeze were more likely to require future urgent health care than those without current wheeze. The finding that the OR based on the multiply-imputed data did not differ from the complete-case estimate indicates the data were MCAR and were not a source of selection bias in this case.

Acknowledgements: ACAM is an Australian Institute of Health and Welfare collaborating unit, funded by the Australian Government Department of Health and Ageing.