Examining productivity losses associated with health related quality of life using patient data

This report presents the findings of a study that estimates productivity loss due to poor health, using days off paid employment and normal activities, EQ-5D and International Classification of Disease (ICD) chapter, as well as socio-demographic data. The results can be used to predict the level of productivity loss associated with EQ-5D values and specific disease diagnosis in economic evaluation.

Methods

Data was used from a prospective survey of inpatients discharged from a hospital in Wales, United Kingdom from April 2002 to January 2009. The number of days off paid employment (N=51,326) and the number of days off normal activities (N=55,726) in the last six weeks were estimated separately using the health of the patient measured by EQ-5D, ICD chapter and other health and socio-demographic data. A variety of different regression models were estimated that are appropriate for the distribution of the number of days off paid employment/normal activities dependent variable, which has large spikes at 0 (zero days off paid employment/normal activities) and 42 days (maximum number of days off paid employment/normal activities in the recall period of 6 weeks). Models were compared based on their predictive ability.

Results

The preferred model that most accurately reflected the distribution of the data was the zero-inflated negative binomial with variable inflation which performed well at approximating the spikes at 0 and 42 days. The results indicated that better health in terms of EQ-5D scores was associated with lower productivity loss in both paid employment and normal activities that did not include employment. Comorbidities were associated with greater productivity loss. ICD chapter codes had varying associations with days off paid employment and normal activities, depending on the model and dependent variable. Increasing age was associated with greater productivity loss but at a decreasing rate, while being female was associated with lower productivity loss.

Discussion

The preferred model can be used to predict changes in productivity associated with changes in the health of the patient measured using EQ-5D, and these results can be applied to existing datasets to inform economic evaluation. Limitations include recall bias and response bias of the productivity data, and restrictions of the dataset to exclude some patient groups.