The VHA guideline group has done an excellent job of reviewing the available evidence and condensing it into a flexible practice guideline that is usable by clinicians and researchers. Adoption and dissemination of the guideline have been mandated by the VHA; plans are being made to monitor patients with diabetes and provide clinicians with feedback on their care.

Although the guideline states that patient preferences should be considered when goals are being set for glycemic control, it also provides numeric goals (for example, "patients who are likely to live a few years might have a target hemoglobin A_1c level of about 0.09" and "Many perhaps most, patients will have a target of about 0.08"). We would caution against letting these "treatment goals" metamorphose into "quality standards." Our experience with an ongoing study suggests that for many patients, treatment characteristics, rather than outcomes, may be the dominant factor in preference for therapy across the ranges of benefits predicted by our Markov model. To the extent that specific glycemic goals become quality standards, patient autonomy may be jeopardized by concerns over provider and institutional accountability.

Our primary purpose was to provide information for both patients and providers regarding outcomes and treatment characteristics before decisions about therapy are made. This is particularly important in diabetes, for which the treatments not only affect long-term outcomes but are also likely to affect patients' quality of life and can accrue substantial short-term costs. For example, in our preliminary work, patients with diabetes report that insulin therapy, which is often required to achieve glycemic goals set forth in guidelines, is perceived as extremely onerous compared with diet and oral hypoglycemic agents. Unfortunately, patients are not routinely presented with quantifiable information about treatments and outcomes, and the best means of providing such information to patients remains uncertain. We hope that ongoing research will contribute to our understanding of the effect of patient preferences on outcomes research and guideline development.