Between August 2002 and May 2004, we enrolled participants in a 12-month open, randomized, single-center, controlled trial with a parallel group design. Patients with type 2 diabetes were randomly assigned to a group receiving usual care or a group receiving usual care plus Web-based care management. Participants, physicians, and care managers could not be feasibly blinded to group assignment after randomization. The study was approved by the University of Washington (UW) Institutional Review Board.
We conducted the trial at the UW General Internal Medicine Clinic (GIMC), a teaching clinic that provides care to 7,707 patients. The clinic is staffed by 25 faculty and 48 resident providers and employs a nurse practitioner to provide case-management services to chronic-disease patients.
Electronic medical record data were used to identify potential participants 18 to 75 years old, whose most recent GHb in the prior 12 months was ≥7%, and who had made at least two visits to GIMC during the prior year. We excluded patients who had participated in the pilot study of the intervention, had major psychological illness, were non–English speaking, had a resident as a primary physician, or were followed primarily in a specialty clinic.
Following an invitation letter, the study coordinator contacted potential participants by phone to assess study eligibility. Exclusion criteria assessed during the phone interview included lack of Internet access and cognitive, language, or hearing impairment severe enough to preclude participation. At the end of the recruitment phone call, the study coordinator invited eligible participants to participate. Participants initially provided oral consent over the phone.
Allocation to the study group was concealed from the study coordinator and the participant until after the recruitment phone call. Following initial oral consent, the study coordinator consulted the allocation assignment table. Participants were randomly assigned in equal numbers to the two groups. The study's statistician used a computer random number generator to create a random number table in a nonblocked sequence. For participants in the intervention group, the study coordinator arranged for in-person, follow-up written consent and an intake visit. Participants in the intervention and usual-care groups did not receive additional incentive for participation in the study. Baseline data for all participants were from automated data in the electronic medical record.
The study aimed to improve glycemic control, used a care manager (3
), and targeted four key domains in Wagner's Chronic Care Model (4
): self-management support for patients, delivery system design, clinical information systems, and clinical decision support (). A complete description of the module is elsewhere (1
Design of Web-based collaborative care intervention
Participants in the care management group met with the care manager for a 1-h visit. The care manager used a collaborative-care approach consisting of four components: 1) defining problems, 2) setting goals, 3) providing access to services that teach skills needed to carry out medical regimens, guide health behavior change, and provide emotional support, and 4) following up actively. The care manager and patient reviewed the patient's electronic medical record and collaboratively created an action plan.
The care manager introduced participants to the Web-based program and encouraged them to review online medical records, send blood glucose readings weekly, and send secure e-mail as needed. She responded to patients’ messages Monday through Friday, reviewed blood glucose levels at least once per week, adjusted hypoglycemic medications, and conferred with the primary care provider as needed.
All participants received primary care from a physician who was board certified in internal medicine at the UW GIMC. All providers used the same electronic medical record, which included patient-specific reminders for measurement of GHb to <7.0%.
The primary outcome was the absolute change in GHb between baseline and end of the 12-month study period. Secondary outcomes included total plasma cholesterol and systolic and diastolic blood pressure. The most recent GHb in the 12 months before randomization was used for baseline measurements. Participants were called 12 months after randomization for a GHb test if one had not been obtained between 9 and 12 months postrandomization. We used the GHb measure closest to 12 months after randomization and no earlier than 9 months or later than 15 months after randomization. GHb values were rapid immunoassay tests from a Bayer Laboratories DCA-2000+ analyzer (Siemens Medical Solutions Diagnostics, Tarrytown, NY). Secondary analysis examined the percent of individuals with GHb <7%.
Post hoc analyses evaluated secondary outcomes related to overall care of diabetes but not targeted by the intervention, including blood pressure and total cholesterol, which were collected as part of usual care. Total cholesterol was used instead of LDL cholesterol measurements because of its more common availability; values were measured using enzymatic analytic chemistry on a Beckman Synchron. Blood pressure was measured using an aneroid sphygmomanometer at GIMC. To maximize the power, absolute change in systolic and diastolic blood pressure and total cholesterol was evaluated between the groups. Exploratory analysis investigated the relationship of GHb with number of participants’ page views of the electronic medical record and number of blood glucose level uploads.
Health care utilization was measured by total number of outpatient encounters with health care providers and inpatient days at the UW Medical Center and affiliated hospitals and clinics during a 2-year period that included the 12 months before study enrollment and the 12-month intervention period. Outpatient encounters were further divided into specialty and primary care encounters.
Electronic medical record access was measured by the number of page views by section of the Web-based medical record. E-mail use by a participant was defined as one or more messages initiated to or in reply to the case manager (6
). Individual counts of e-mails were not available for analysis. Study staff collecting outcomes were not blinded to group assignment.
The trial was designed to have 80% power to detect a difference of 0.5% in GHb concentration (two-sided significance level of P
<0.05; SD of mean GHb 1.26; mean change in Z
score SD in GHb levels 0.87) (7
). Intention-to-treat analysis of the main trial outcome included all randomly allocated participants with available outcome data. Three participants in the intervention group and six in the usual-care group did not have a follow-up GHb. Primary analysis used linear regression with change in GHb as the dependent variable, adjusted for age and sex and for baseline GHb (8
). Sensitivity analyses included a single imputation method for missing GHb follow-up measures using the baseline GHb observation carried forward, an average of baseline GHb among all participants carried forward, and an average of all available post GHb by study group.
Secondary outcomes used the same analysis procedure as in the primary analysis. Utilization analyses used Student's t tests to compare differences in utilization between intervention and usual-care groups, including differences at baseline and follow-up, and in the changes from baseline to follow-up.