Measures
Primary outcome measures were utilization: ED visits, all-cause unplanned admissions, and 30-day readmissions. Secondary outcome measures included 2 types of quality indicators(QIs, see Appendix for scoring): (a) gaps in care measures that captured whether providers were following standards of care for diabetes, ischemic vascular disease, and prevention; and (b) patient composites which reflected patient illness severity for diabetes and cardiac disease. Higher scores indicated more gaps in care or greater disease severity. Both types of QI measures required at least a 12-month window and thus could not be computed for patients engaged with the CCT who had only a 6-month follow-up period (as their scores could reflect their pre-CCT status). In addition, comprehensive care was denoted by the provision of depression screening on the PHQ-2 [14], and whether HgA1c was greater or equal to 9.0 in diabetics served as an additional patient outcome measure. Other secondary outcome measures included practice joy via the Well-Being in the Workplace Questionnaire (WWQ) [15] and patient satisfaction from CAHPS-CG 12-month survey with PCMH items [16].
Data Collection
Utilization and quality data were extracted from the network’s hospital and outpatient electronic medical records (EMR). Practice staff were emailed every 6 months and asked to anonymously complete the WWQ via Survey Monkey [17]. At baseline, patient satisfaction surveys were distributed in the practice, and patients had the option of anonymously completing them during their visit or returning them in a prepaid envelop. While not recommended for CAHPS, this procedure had been internally used with success previously. At 12 months, the same survey was mailed to a random sample of patients with a prepaid return envelope.
At the practice level, utilization and QI data were only available for patients from 4 of the 6 practices: data were not available for 1 non-EMR practice and there was negligible variation in utilization from the pediatric practice. For the patient level analyses, utilization was available for patients in all 6 practices.
Statistical Analyses
To test whether outcomes were improved relative to a comparison group following the introduction of CCTs into the practices (practice level analyses) or CCT engagement (patient level analyses), mixed models analyses of variance with repeated measures on Time (pre- vs. post-CCT) were conducted with SAS [18] PROC MIXED and PROC GENMOD for continuous and dichotomous outcomes, respectively. To determine whether there was greater improvement in the CCT groups, all models included the interaction between Time and Group (CCT versus no CCT) in addition to their main effects. At the practice level, high-risk and non-high-risk patients were analyzed separately. And, at the patient level, CCT-MNGT and CCT-DCREC patients were analyzed separately with results adjusted for patient’s age. Some variables were not normally distributed. The quality variables were able to be normalized with natural log transformations, but utilization variables had to be dichotomized into “any” versus “none” due to severe skewness, inflated 0s and larger-valued counts. Practice joy and patient satisfaction can only be reported at the practice level (responses were aggregated within each practice because anonymous responses do not permit linking specific respondents over time and different patients were sampled across measurement occasions) and non-parametric tests (Wilcoxon signed ranked tests and tests of dependent proportions) were used to test for change over time given the small sample size ( n = 6).