The number of adults with identified diabetes in 1995 and eligible for inclusion in the analysis each subsequent year was 5432 in 1995, 4835 in 1996, and 4451 in 1997. Of these, 4339 (79.8%) in 1995, 3941 (81.5%) in 1996, and 3767 (84.6%) in 1997 had at least one HbA1c test. provides additional information about these patients and compares patients with no HbA1c tests during the study period with patients with one or more HbA1c tests during the study period. Notably, 67% of all patients with diagnosed diabetes mellitus had two or more tests during the study period, and 44% had three or more tests.
Characteristics of Study-Eligible Patients Who Did or Did Not Have at Least One HbA1c Test During the 36-Month Study Period, Multispecialty Health Care Practice, Minnesota, 1995–1997
Patients linked to a PCP were compared with those not linked to a PCP. These two groups of patients had no differences in sex (47% in both groups were female, P = .60), but those linked to a PCP were older (60.4 years vs 55.3 years, P < .01), had higher Charlson comorbidity scores (1.75 vs 1.02, P < .01), and had higher HbA1c values in the first year (8.3% vs 8.1%, P < .01). In addition, those linked to a PCP were more likely to have had an HbA1c test. For 1573 person-years (about 13% of all person-years), no link to a PCP could be made. These person-years were excluded from final analysis after preliminary models indicated that their exclusion had only a minimal effect on the analyses.
shows the intercept-only model for the dependent variable of year 1 HbA1c value. Most of the variance in HbA1c value is clearly at the patient level. In the full model with covariates entered at the patient and physician levels, most variance is again at the patient level, with insignificant variance at the physician level, and a small but significant amount of variance at the clinic level. Similar models were constructed with the HbA1c value in year 2 and then in year 3 as the dependent variables, and the same patterns of effect on variance were observed (data not shown). The amount of variance related to the clinic level remained small throughout the study period.
Variance and Change in HbA1c Values at the Levels of Clinic, Physician, and Patient, Multispecialty Health Care Practice, Minnesota, 1995–1997
also shows the multivariate models comparing change in HbA1c value in intercept-only models and in models that include patient and physician covariates. In these models, only small and insignificant variance components were associated with the physician and clinic levels of the model.
presents more detailed data from the full model predicting change in HbA1c. This model explained 11.8% of the variance in change in HbA1c value. Significantly more improvement in HbA1c was noted among the oldest patients and in those who had classes of glucose-lowering medications added during the study period.
Multivariate Modela Evaluating Relationship of Patient and Physician Covariates to Change in HbA1c Value From 1995 Through 1997, Multispecialty Health Care Practice, Minnesota
Additional models were run using HbA1c test rates, low-density lipoprotein cholesterol test rates, and diabetes eye examination rates as the dependent variables. In each of these analyses, the pattern of results was the same as the results for HbA1c values. In each model, more than 80% of total variance was at the patient level, with much smaller and usually insignificant amounts of variance at the physician and the clinic levels.