Mapping MOS Sleep Scale scores to SF‑6D utility index

Abstract

ABSTRACTObjective: Deriving preference scores for the Medical Outcomes Study (MOS) Sleep Scale would enable its use in cost–utility analyses. The objective of this study was to map scores of the MOS Sleep Scale to a preference-based health-state utility index (SF‑6D) scored from the SF‑36 Health Survey (SF‑36).Research design and methods: Three datasets were used: (1) the MOS study, a 4-year observational study of chronically ill patients, (2) a 7-week open-label, non-comparative clinical trial of an osmotic controlled-release oral delivery system (OROS) hydromorphone in the treatment of chronic low back pain (CLBP), and (3) a 6-week open-label randomized controlled trial of OROS hydromorphone in the treatment of pain associated with chronic osteoarthritis (OA). Various models were tested, where SF‑6D was regressed onto the Sleep Problem Index‑II (SLP9) in 1000 random half (developmental) samples of the MOS (n = 1413). The best fitting model was applied to the other 1000 random half (cross-validation) samples of the MOS (n = 1412), and to the two trial samples (n = 199 in the CLBP trial; n = 124 in the OA trial).Results: The best fitting model in the MOS samples included a quadratic term for the SLP9 which explained 34% of the variance in SF‑6D in the developmental samples. Errors in prediction were greatest at higher SLP9 scores. Addition of demographic and clinical variables to the model explained minimal incremental amounts of variance (< 5%) in SF‑6D scores. These results were replicated in the cross-validation MOS samples. In both developmental and cross-validation MOS samples, mean predicted and observed SF‑6D scores were nearly identical. When the mapping algorithm developed in the MOS was applied to the CLBP sample, mean predicted SF‑6D scores were 0.09 points higher than observed SF‑6D scores at both baseline and final visits, while changes in predicted and observed SF‑6D scores were identical.Conclusion: Results indicate that it is possible to map MOS SLP9 to SF‑6D yielding useable preference-based scores essential for cost–utility analyses. A limitation concerns the interpretation of SF‑6D scores estimated from SLP9 scores above 60, where the prediction errors increased considerably.