Improving evidence-based primary care for chronic kidney disease: study protocol for a cluster randomized control trial for translating evidence into practice (TRANSLATE CKD)

Conduct a cluster randomized controlled trial of point-of-care CDS plus the full TRANSLATE model of practice change, versus CDS alone in promoting evidence-based care in primary care practices for all patients with an eGFR <60 and >15 ml/min/1.73 m² confirmed with repeat testing over three or more months. (CKD stages three and four).

CDS practices using the TRANSLATE model will provide a greater degree of evidence-based guideline-concordant care for CKD than CDS-only practices.

Patients with stage three and four CKD in facilitated practices will have slower CKD progression than patients in CDS-only practices.

Patients with stage three and four CKD in facilitated practices will have significantly lower all-cause mortality than stage three and four patients in CDS-only practices.

The process evaluation will determine through qualitative methods the fidelity of the facilitated TRANSLATE program; find the challenges and enablers of the implementation process, and the contextual factors that contribute to TRANSLATE decisions and strategies; and translate lessons learned into pragmatic ‘best practices’ for future facilitation and dissemination.

The intervention of CDS plus TRANSLATE is more cost-effective than the intervention of CDS alone.

Only practices providing ambulatory primary care as their principal function, located in non-hospital settings, employing at least one primary care physician, with a minimum of 2,000 patients seen in the prior year, are eligible. ‘Practices’ are defined as distinct office locations that belong to organizations with one or more practice sites. Candidate practices are drawn from members of the Distributed Area Research and Therapeutics (DARTNet) members. These members have point of care decision support for preventive care and multiple chronic diseases. CKD will be added to the chronic disease management reminder system.

Patient-level data will be drawn from the population of all active patients at each practice and will include all patients diagnosed with stage three CKD. Patients are considered to meet stage three CKD criteria if at least two consecutive eGFR measurements at least three months apart fall below 60 ml/min. At the inception of the study, we used data from the preceding 12 months to identify eligible patients (eGFR <60 and >15 ml/min) with additional patients added to the analytical dataset as they meet study criteria. These criteria were also used to initiate the clinical decision support algorithms.

Recruitment for the study began in April 2012. The recruitment goal is to enroll 36 practices total, randomized into the two study arms. To date, 29 practices have been enrolled in the study. 25 of these practices have been randomized in two waves. The intervention period for the first wave of 18 practices began in January of 2013. The intervention for the second wave of seven practices will begin in June of 2013. We plan to recruit and randomize all additional practices by the end of 2013. Consent was obtained at multiple levels; a practice administrator signed a practice agreement and data use agreement for each individual practice organization. From each individual practice site (cluster), informed consent was obtained from a site coordinator and a lead physician at each site, as representatives of the cluster. Consent was obtained prior to randomization.

In cases where two or more selected practices were drawn from the same multi-practice organization, contamination was limited by checking for clinician overlap. If overlap was minimal, organization was used as a stratification variable; if overlap was potentially problematic in terms of contamination, practices were constrained to be in the same group. From the pool of available practices, pre-study information from electronic health record (EHR) data on baseline performance characteristics related to CKD or in general and variables which may influence outcome measures, including practice and patient panel characteristics as well as pre-study clinical measures were collected (full-time equivalent physicians, mean HbA1c, % diabetic, % stage four CKD, % of diabetics with HbA1c >9, mean GFR, mean SBP, % with SBP >130, % with SBP >140, % African American, % Hispanic, % uninsured) [34]. Stratification variables included geographic region and organization. All possible combinations of the 18 eligible practices into two groups of nine were generated [35] using the IML procedure in SAS [36], retaining order so that the first designated group would be assigned to the intervention, the remaining practices to controls (n = 48,620). Only randomizations that were balanced on stratification variables were retained (n = 1,728). For each randomization, a balance criterion (defined as the sum of squared differences on standardized variables between control and intervention groups) was computed. After examining the distribution of the balance criterion, a maximum allowable difference between the groups was established and an optimal set of randomizations identified (n = 136). From this set, one was chosen using a random number generator and practices were assigned to treatment or control arms. This process was repeated for the next group of seven practices, including a dummy practice in the randomization procedure to accommodate the uneven number of practices. The remaining practices will be randomized using the same protocol. As with all CRCTs, patients are not randomized per se; they are assigned to treatment conditions along with the medical practices in which they receive care.

Given the CKD national prevalence rate of 13% and an expected average of 5,500 active patients per practice, we expect roughly 715 CKD patients per practice (this estimate is supported by a recent analysis of >110,000 individuals with stage three or four CKD drawn from over 100 DARTNet practices.) With 18 practices per arm and a minimum of 200 patients per practice (smallest practice estimated at 2,000 patients with a conservative 10% prevalence of CKD over life of study) there will be a minimum of 3,600 patients per arm (actual expected sample size is over 14,000 per arm.) A sample size of 3,600 per arm will provide >80% power to detect a 0.18 effect size difference between two arms at a single time point if the intra-class correlation (ICC) is 3%. This effect size was assumed based on previous results from the diabetes TRANSLATE study [32]. In terms of change over time, a sample size of 3,600 will provide >80% power to detect a small linear trend effect, increasing from 0 at baseline to 0.2 standard deviation (SD) at final follow up, with four observations per person and an ICC of 3%, with a random effects structure with random intercept and random slope [37].

Four elements of the TRANSLATE method will be implemented in both groups, while the remaining ones will apply to facilitated practices only. All interventions will occur at the cluster level, and there will be no randomization or specific intervention undertaken for individual patients. The CDS-only practices will have CKD decision support algorithms based on the Kidney Disease Outcomes Quality Initiative (KDOQI guidelines added to their existing CDS. There will be three separate vendors that will be providing point of care support for the study. Computer Integrated Networks of America (CINA), Health Metrics Systems (HMS), and an EMR with integrated CDS Medent. All practices will receive one session of introductory academic detailing [38] concerning the rationale for the algorithms. They will also be provided related technical support on request. Facilitated practices will receive the CKD-specific CDS, as well as a practice facilitator to assist them in the implementation of the full TRANSLATE model (Table 1).

The facilitators' overall role is to help the practices implement the TRANSLATE model to improve guideline-concordant care for CDK. The facilitator will assist with the site coordination, physician champion's needs, audit and feedback, team approach, and education. Two practice facilitators are each assigned to half of the CDS+ facilitation practices. Throughout the course of the intervention period, the facilitators will engage in virtual interactions with the practices, in which they will work with practice quality improvement (QI) teams to identify and solve issues specifically related to their care of patients with CKD and form and maintain relationships with each practice. Facilitation objectives include: assisting practices in setting goals to implement CKD guidelines; helping practice teams, strategize, test, and implement change; facilitating meetings and fostering a continuous QI culture; serving as a liaison for data and performance feedback; and sharing best practices and linking intervention practices.

Due to the national scope of this project, facilitation contacts will occur ‘virtually’ using videoconferencing (GoToMeeting®), phone, and e-mail. General communication will entail a minimum of once monthly teleconferences with each practice’s site coordinator to review the status of the project, monthly virtual QI team meetings, monthly calls between the academic mentors and the lead physician, periodic learning collaboratives, and a benchmarking webinar every six months with the academic detailing team.

From practice EHRs, we will collect the measurements outlined in Table 2. Medication fulfillment data will be collected from Surescripts RxHub through CINA or the practice EHR. Death will be determined from information in practice EHRs or from linkages to the National Death Registry. We will check for deaths among patients who have not made any visits in the prior 12 months in the first two years of the intervention and in the prior six months in the final year of the intervention. Medicare claims data will be obtained before and after the intervention to allow for a proper cost analysis.

To determine the effectiveness of the TRANSLATE elements, it is crucial to first determine the degree of adoption by the practice, and then systematically document the challenges and enablers of the implementation process, including the role of facilitation and the contextual factors that contribute to success. Qualitative measures will be collected in tandem with the implementation of facilitated TRANSLATE elements to support real-time learning and a reiterative process of evaluation (Table 3).

Semi-structured interviews will be conducted with the physician champions at baseline and endpoint, asking about their care of patients with CKD and workflow processes in their office. Members of the evaluation team will also conduct site visits at endpoint with both intervention and comparator practices.

E-mail and other communication between practice facilitators and the sites will be analyzed by the evaluation team. Facilitators will also collect notes of meetings, and practice activities; and submit this information to the evaluation team on a regular basis.

The hypothesis is that the intervention of CDS plus the facilitated-TRANSLATE is more cost-effective than the intervention of CDS alone, so the economic analysis will focus on the TRANSLATE elements in delaying CKD progression and reducing mortality due to CKD in the practice settings of primary care physicians. Because there is not a true control group (no clinic practice has no intervention), the proposed project will examine the relative costs of these interventions.

Three types of costs will be included into the total costs of each intervention: the cost of providing services in the clinic practices; the cost of treating CKD; and the other healthcare expenses for patients. We plan to survey a sample of randomly selected practice clinics to measure the cost of providing CKD services. The investigators have created and used a spreadsheet and guideline for collecting cost data from practices for previous studies [39, 40]. The practice facilitation activities of TRANSLATE are expected to lead to more costs in the intervention practices. The cost of training clinic practices to use the CDS system and functions of TRANSLATE will also be included into this type of cost, and the cost of training (time) will be measured based on the hourly wages of physician or medical staff. The cost of treating CKD and other healthcare costs for patients will be based on the claims data that we acquire from DARTNet.

The effectiveness measures will include the degree of evidence-based guideline-concordant care for CKD patients (a patient-level score based on the percentage of goals achieved); CKD progression for patients with stage three and four CKD; and all-cause mortality rate.

The outcome for this analysis will be eGFR measurements over time. There will be multiple eGFR measures per patient over the duration of the study. We will use general linear mixed-effects models to estimate the rate of decline in eGFR (random intercept, random slope) and the degree to which the baseline covariates predict eGFR. Time for each observation will be coded as days since baseline, converted to months to aid interpretability. The primary hypothesis of difference in slope between treatment groups, adjusting for socio-demographic and clinical covariates, will be tested.

All-cause mortality will be confirmed using the National Death Index to determine the exact date of death. The outcome for the analysis will be time from baseline to death. Patients who are alive at the end of the study period will be censored at the end of the follow-up time. Assumptions of the proportional hazards model will be checked for each variable. Covariates will include baseline eGFR, defined as the mean of the last two eGFRs prior to study entry, as well as socio-demographic and clinical characteristics.

Qualitative data for the process evaluation will be analyzed using an immersion-crystallization approach [41, 42]. This content-driven approach allows the data to speak for itself, as researchers immerse themselves in the data repeatedly to identify themes that emerge [41]. Themes are identified around concepts that are expressed repeatedly in the text and constantly compared back to the data to ensure that they represent the data accurately. Once themes are identified using this method, an additional step in the analysis will be to compare the themes with the elements of the TRANSLATE model to determine how well the data complement the interventional model and how the data can illuminate the process of practice adoption of each element.

Analysis will compare the effectiveness-cost ratios between the two arms to examine which intervention is more cost-effective. Because of the available claims data, electronic medical records data, and mortality data from the National Death Index, the analysis will calculate the quality-adjusted life years (QALYs) lost due to CKD for both interventions. The QALYs will be the most important effectiveness measure in the economic analysis, because QALYs can be valued as a dollar amount. In the United States, $50,000 per QALY is a decision rule that is often used to guide interpretation of cost-effectiveness analyses. We will also do sensitivity analysis by replacing this decision rule with $25,000 or $100,000. We will compare the effectiveness (dollar values of QALYs) with the cost, and determine whether the effectiveness is larger than the cost for each intervention. If the effectiveness is larger than the cost, these interventions will have cost savings. If both interventions will produce cost savings, the proposed project will compare the cost savings for these two interventions and determine which intervention has a high cost saving amount.