Comparison of Two Methods for Estimation of Work Limitation Scores from Health Status Measures

Research output: Contribution to journalConference abstract in journalResearchpeer-review

To compare two methods for estimation of Work Limitations Questionnaire scores (WLQ, 8 items) from the Role Physical (RP, 4 items) and Role Emotional scales (RE, 3 items) of the SF-36 Health survey. These measures assess limitations in role performance attributed to health (emotional, physical, or both) and different breadth of impact (work vs. work and other activities). We compared WLQ estimates based on an item response theory crosswalk (Method1) and a regression imputation (Method 2). Such estimates can expand the information from studies using only the SF-36 measure, and can inform future data collection strategies.

We used data from two independent cross-sectional panel samples (Sample1, n=1382, 51% female, 72% Caucasian, 49% with preselected chronic conditions, 15% with fair/poor health; Sample2, n=301, 45% female, 90% Caucasian, 47% with preselected chronic conditions, 21% with fair/poor health). Method 1 used previously developed and validated IRT based calibration tables. Method 2 used regression models to develop aggregate imputation weights as described in the literature. We evaluated the agreement of observed and estimated WLQ scale scores from the two methods and their ability to discriminate among known groups of patients.

Estimated scores from the two methods were strongly correlated (r=.99). Estimated and observed scale scores had strong correlations (r= .68 for RE and r=.76 for RP). Observed and estimated WLQ from both methods successfully differentiated between levels of self reported general health and between patients with and without chronic conditions. For both methods the estimated WLQ means from SF36RP score were closer (and not statistically different for Method1) to the observed WLQ means than estimated WLQ scores from the SF36 RE scale.

Our results suggest that both methods provide useful WLQ estimates for group level analysis. Method 1 appears slightly more accurate than Method 2, but is computationally more complex.
Original languageEnglish
JournalValue in Health
Issue number7
Pages (from-to)A709
Publication statusPublished - Nov 2015
EventISPOR 18th Annual European Congress - MiCo - Milano Congressi, Milano, Italy
Duration: 7 Nov 201511 Nov 2015
Conference number: 18


ConferenceISPOR 18th Annual European Congress
LocationMiCo - Milano Congressi

ID: 172891177