Primary Care Diabetes Bundle Management: 3-Year Outcomes for Microvascular and Macrovascular Events | Page 2

Published Online: June 26, 2014
Frederick J. Bloom Jr, MD; Xiaowei Yan, PhD; Walter F. Stewart, PhD; Thomas R. Graf, MD; Tammy Anderer, PhD; Duane E. Davis, MD; Steven B. Pierdon, MD; James Pitcavage, MS; and Glenn D. Steele Jr, MD
During the study period, individuals were included as diabetes cases if they met Healthcare Effectiveness Data and Information Set (HEDIS) criteria19 and had 4 or more encounters with a diabetes International Classification of Diseases, Ninth Revision (ICD-9) code on different dates during the study period. The earliest encounter with a diabetes ICD-9 code was taken as the diabetes baseline date. Patients had to have at least 180 days of follow-up after their diabetes baseline date to contribute to the analysis.


We identified all inpatient and outpatient claims during the 5-year study period related to the following ICD-9 codes: retinopathy (ICD- 9 codes 250.5x and 362.xx); amputation (ICD-9 codes 89.5x, 89.6x, 89.7x, and 99.7x); MI (ICD-9 code 410.xx); and stroke (ICD-9 codes 430, 431.xx-438.xx). The following numbers of cases were identified: 2518 retinopathy; 138 amputation; 3824 MI; and 2112 stroke. Incident microvascular or macrovascular events were identified as the earliest documentation of 1 of the above codes if there was also at least 1 year of follow-up prior to the first documentation of the claims event. This “clean-out period” restriction was imposed to eliminate prevalent events among those who became new members of GHP. In sensitivity analysis, we also used a 2-year clean-out period. The findings were essentially the same. Using the 1-year criteria, we identified 519 retinopathy, 17 amputation, 714 MI, and 445 stroke incident events.

Chronic kidney disease, 1 of 3 serious microvascular diseases associated with diabetes, was not evaluated as an outcome because it is not reliably reported in insurance claims data unless a patient has end-stage disease.

Statistical Analysis

Because patients were not randomized, we used a 2-step method to protect against inferential threats attributable to baseline differences in risk factors. First, logistic regression was used to calculate a propensity score20 for each eligible diabetes patient to balance disease burden for members of the DS and NDS groups at the beginning of intervention (January 1, 2006), which was denoted as the index date. If patients had pure postintervention exposure, then the index date was set to be the same as the baseline date. Claims data from the date of enrollment to the index date were used to derive the propensity score based on patient gender and age at index date, as well as claims documentation of comorbidities (mild liver disease, peripheral vascular disease, heart failure, cerebrovascular accident, pulmonary disease, severe liver disease, renal disease, malignant cancer, cardiovascular disease, and dialysis) up to the index date. The propensity score was used to match DS and to NDS diabetes patients (± 0.2 of SD of the propensity score, an optimal caliper usually used in propensity score–matching studies21) and to balance both groups on future risk of events of interest. Second, to account for differences in person-time of observation in the postsystem-of-care periods, we calculated the probability of being “exposed” to the post-system period. To this end, each patient’s age at diabetes baseline date, duration of person-time in the preintervention period, and comorbidities documented before the index dates were included in the conditional logistic regression model. Conditional logistic regression was used to control for matched pair variation. A score obtained for each subject represented the likelihood of postsystem-of-careexposure.

Person-time was excluded until a patient reached 40 years of age. Cumulative risk was estimated using the weighted Cox Proportional Hazards regression model, which combined a traditional Cox Proportional Hazard regression model with inverse probability weighting (IPW). The inverse of the previously described postsystem exposure score was used as a weight, along with age at the index date and gender. A patient was censored from observation if they discontinued as a GHP member, died as a GHP member, or incurred their first unique major microvascular or macrovascular event as previously described. The C statistic was used to measure discriminatory capability of the model and calculated as described by Pencina and colleague’s method.22 The cumulative hazard of each outcome was predicted by the model. Statistical analyses were performed using SAS version 9.2 (SAS Institute Inc, Cary, North Carolina) for the time-to-event analysis and relative risk comparison.

The number of patients needed to treat (NNT) was calculated based on risk reduction, which was estimated from the above Cox proportional hazard model; that is, the 3-year survival rate was estimated from the model for DS and NDS group, respectively, and the difference represented the 3-year risk reduction. The NNT is the inverse of the risk reduction.


The initial study cohort of 16,086 GHP members met the criteria for diabetes diagnosis. (Figure 1). After applying exclusion criteria (<180-day follow-up, age <40, only preintervention exposure or prevalent outcome), the sample was reduced by 3421 of the 8355 patients in the DS group clinics and by 3634 of the 7734 in the NDS group clinics. After applying propensity score–matching criteria, 4095 DS subjects were matched to 4095 NDS subjects. Before matching (Table 2), DS group members were significantly older, had proportionately more of their person- time in the postintervention period, and had a longer follow-up period than NDS members. After matching, DS and NDS members still differed on duration of follow-up time in the postintervention period. The 2 groups were similar on the risk score derived at baseline and up to the time of the index date. The adjusted hazard ratio (Table 3) for MI was 0.77 (95% CI, 0.65-0.90); for stroke, 0.79 (95% CI, 0.65-0.97); and for retinopathy, 0.81 (95% CI, 0.68- 0.97). The risk of the macrovascular and microvascular events were significantly lower among DS patients compared with NDS patients (Figure 2 includes cumulative hazard plots). These hazard plots show that the reduction in macrovascular outcomes (MI and stroke) is most pronounced after year 1 of the DS, and that the microvascular outcome retinopathy is evident after approximately 18 months. The risk of developing retinopathy, MI, and stroke among DS patients was significantly less than that of NDS patients in the 3-year follow-up period. The NNT to prevent 1 event over 3 years was 82 (95% CI, 37-133) for MI; 178 (95% CI, 57-681) for stroke; and 151 (95% CI, 47- 512) for retinopathy.

PDF is available on the last page.

Issue: June 2014
More on AJMC.COM