OBJECTIVES

A within-trial cost-effectiveness analysis (CEA) of the DPP was published in 2003. The authors appear to have ignored missing data and used ACA, reporting QALYs using the QWB to measure patient utilities for placebo (PBO), lifestyle (LS), and metformin (MET). The main motivation here is to estimate treatment utilities and QALYs using Manski bounds (MB) and compare these results with DPP findings.

METHODS

We had access to a 95% subset of the original data. We reproduced the original ACA results and then used MB to develop estimates for utilities and QALYs based on the QALY calculation method used in the original CEA. The minimum and maximum utility scores each year for each group were used to fill the missing value for participants in their respective groups for LMB and UMB. The percentage of participants with at least one missing utility data for QWB for years 1, 2 and 3 was 36%, 10 % and 48% respectively.

RESULTS

The yearly utility scores using the QWB originally reported in the DPP were [0.686, 0.675, 0.657], [0.687, 0.680, 0.673] and [0.703, 0.695, 0.692] for PBO, MET, and LS. The utility scores for LMB were [0.553, 0.631, 0.416], [0.560, 0.635, 0.476] and [0.560, 0.655, 0.496]. The analogous numbers for UMB were [0.797, 0.708, 0.826], [0.796, 0.715, 0.834] and [0.813, 0.725, 0.843].

QALYs using the QWB as originally calculated were 2.018, 2.040 and 2.090 for PBO, MET and LS. The LMB QALYs were 1.600, 1.671, 1.711; UMBs were 2.331, 2.345, 2.381. The minimum/maximum range versus PBO for LS was 0.050/0.781; for PBO versus MET the range was 0.014/0.745, indicating that even the relative rankings of QALYs cannot be known.

CONCLUSIONS

The missing data and MB estimates indicate considerable uncertainty even on QALY rankings of the treatments and certainly in any CEA based on them.