Predicting future changes in the work ability of individuals receiving a work disability benefit: weighted analysis of longitudinal data.

Objectives Weighted regression procedures can be an efficient solution for cohort studies that involve rare events or diseases, which can be difficult to predict, allowing for more accurate prediction of cases of interest. The aims of this study were to (i) predict changes in work ability at one year after approval of the work disability benefit and (ii) explore whether weighted regression procedures could improve the accuracy of predicting claimants with the highest probability of experiencing a relevant change in work ability. Methods The study population consisted of 944 individuals who were granted a work disability benefit. Self-reported questionnaire data measured at baseline were linked with administrative data from Dutch Social Security Institute databases. Standard and weighted multinomial logit models were fitted to predict changes in the work ability score (WAS) at one-year follow-up. McNemar's test was used to assess the difference between these models. Results A total of 208 (22%) claimants experienced an improvement in WAS. The standard multinomial logit model predicted a relevant improvement in WAS for only 9% of the claimants [positive predictive value (PPV) 62%]. The weighted model predicted significantly more cases, 14% (PPV 63%). Predictive variables were several physical and mental functioning factors, work status, wage loss, and WAS at baseline. Conclusion This study showed that there are indications that weighted regression procedures can correctly identify more individuals who experience a relevant change in WAS compared to standard multinomial logit models. Our findings suggest that weighted analysis could be an effective method in epidemiology when predicting rare events or diseases.