From May 2004 to May 2006, we conducted face-to-face interviews with patients without dementia who were aged ≥18 years, living with diabetes, and attending clinics affiliated with an academic medical center (University of Chicago, Chicago, IL) and physician offices affiliated with a suburban hospital (MacNeal Hospital, Berwyn, IL). Prospective subjects were initially identified through clinic scheduling software based on ICD-9 codes for diabetes (i.e., 250.xx). Randomly identified patients were sent study recruitment letters. Letters were followed by a telephone call. We performed a screening telephone minimental status examination and excluded patients with scores ≤17 (14
). We successfully contacted 2,990 patients, and 2,398 of these patients were eligible for the study. A total of 910 patients (38% of eligible subjects) scheduled interviews, and 701 patients (29% of eligible subjects) completed interviews. The average of age of subjects who completed interviews did not differ from that of other eligible patients.
Interviews took ~1 h and were conducted by trained interviewers in English or Spanish. All Spanish interview materials were professionally translated and back translated. We elicited utilities using the time-tradeoff method (15
). For each time-tradeoff elicitation, patients were given a description of a hypothetical health state and asked to consider life in that state. The text of all health state descriptions is included in an online appendix (available at http://dx.doi.org/10.2337/dc07-0499
). The health state descriptions were based upon our prior study of diabetes-related health state utilities (13
) and existing descriptions in the literature. Health state descriptions were reviewed with the clinical faculty at the University of Chicago and pilot tested with patients. During the time-tradeoff elicitation, patients were asked to give their preference for 10 years in the health state of interest and a shorter period of time in perfect health. Using the ping-pong method, patients were asked a series of iterative questions where the time in perfect health was systematically altered by yearly increments and questioning was stopped, when the patient was indifferent between a given time choice. The point at which the patient was indifferent between the time choices was used to calculate the utility score (e.g., if 6 years of life in perfect health = 10 years with an amputation, the utility = 0.60). To minimize the effects of order response bias, the order of utility assessments was randomly allocated along two dimensions of the health states: 1
) complication states versus treatment states and 2
) severe/intensive states versus intermediate/conventional states.
The descriptions of several complication health states were based on previous descriptions of life with complications found in the utility literature (blindness [16
], diabetic retinopathy [symptomatic] [16
], end-stage renal disease on hemodialysis [17
], amputation [18
], and major and minor stroke [19
]). When such descriptions were not available we developed health state descriptions based on clinical experience and from published descriptions of life with such complications (angina-stage II Canadian Heart Association [20
], diabetic neuropathy [symptomatic] [18
], and diabetic nephropathy [21
For each treatment state, we described the daily experience of treatments, the laboratory testing associated with treatments, and the likelihood of side effects. Patients were asked not to consider long-term effects of treatments on complications but to focus on the daily quality-of-life effects of treatments. We based our description of intensive and conventional glucose control on the treatment protocols and patient experiences of the U.K. Prospective Diabetes Study (UKPDS) (3
). With intensive glucose control, patients were told that they would be more likely to be given multiple oral agents and insulin, that the frequency of major hypoglycemic episodes would be higher, and that the need for self-glucose of monitoring would be greater to achieve A1C <7% in comparison with conventional glucose control (A1C = 7.9%). Similarly, we used the UKPDS blood pressure trial protocols as the basis for descriptions of intensive and conventional blood pressure control (2
). Patients were told that with intensive blood pressure control they would be more likely to be given three to four blood pressure agents compared with conventional blood pressure control. Descriptions for the remaining treatment states were based on data from the medical literature (e.g., aspirin [22
] and cholesterol-lowering medication [23
We also queried patients about their perceptions of quality of life with comprehensive diabetes care, which we described as the combination of cholesterol-lowering medication, aspirin, intensive blood pressure control, intensive glucose control, diet, and exercise. This combination represented care that was both comprehensive in breadth but also intensive in terms of risk factor goals. We also asked patients about a state we called the comprehensive care with polypill state. This state was identical to the comprehensive diabetes care state except that the number of pills taken per day was reduced by the use of the polypill.
After utility elicitation, patients were asked about their overall health status, current medications, relationship with their physician, beliefs regarding medications, and willingness to take more medications. Medical records were abstracted for data on current medications, comorbidities (Charlson comorbidity index [24
]), and risk factor levels. We performed a 10% rereview and found moderate to excellent agreement among abstractors. The intraclass correlation coefficient for A1C was 0.92. κstatistics for the presence of complications ranged from 0.59 to 0.79.
All analyses were performed using SAS statistical software (release 8.1; SAS Institute, Cary, NC). We describe the distribution of utilities using the mean, median, mode, SD, skewness, and kurtosis provide graphical illustration of the distributions of utility scores. Paired t tests were used to compare multiple health state utilities ascertained from the same individuals. Wilcoxon’s rank-sum tests were used for comparisons of utilities across subgroups.