The possibility of obtaining a large, nationally representative sample of primary care clinicians (physicians, physician's assistants, and nurse practitioners) makes the VHA health system an enticing setting to conduct implementation and outcomes research. With careful planning, a systematic yet flexible approach, and a multidisciplinary staff, it is possible to recruit a nationwide sample of primary care clinicians employed in the VHA's community-based outpatient clinics. Over approximately two years, we were able to recruit 401 clinicians representing 168 clinics and 48 facilities in 26 states and Puerto Rico and the Virgin Islands. These groups accounted for 73% of all eligible facilities, over 75% of their associated clinics, and 42% of their clinicians.
Most GRTs do not report a response rate as they have a target number of "groups" or practices to recruit for the purposes of statistical power  and do not identify, or at least report, a sampling denominator. Our facility and clinic response rates were much higher than the 27% of nursing homes in a GRT study of osteoporosis fracture prevention  or the 33% of practices in a managed-care organization's study to increase chlamydia screening . Our response rate is similar to non-GRT studies where the purpose was to obtain a population-based nationwide sample. For example, the National Institutes of Health-funded Coronary Artery Risk Development in Young Adults (CARDIA) study has been following an initial cohort of 5,115 community-dwelling healthy young adults first recruited in 1985 for nearly 25 years. The initial 1985 recruitment for CARDIA resulted in a 55% response rate . CARDIA has significantly contributed to our scientific knoweldge, having resulted in over 400 peer-reviewed publications. More recently, the National Institues of Health-funded Cancer Outcomes Research Consortium (CanCORS) was established in 2001 to obtain a representative, population-based sample to study the processes and outcomes of patients with newly diagnosed lung or colorectal cancer . The approximately 10,000 cancer patients recruited with a population-based approach represent about 50% of the underlying target population. As the recruitment methodologies of GRTs become more refined, their findings will be highly generalizable.
Recruiting for GRTs and for RCTs can be viewed under similar theoretical perspectives, including Choo's general model of information use identifying major elements that influence information-seeking behavior  and the work of Christensen and Armstrong involving diffusion of innovation , which includes "disruptive" effects. In the VA MI-Plus study, recruitment involved two groups of clinicians: (1) physicians to identify a local PI and (2) clinicians to log on and participate in the intervention. These clinician groups may have different elements that influence their participation. Local PIs had to complete necessary IRB training and submit applications through R&D and IRB committees for study approval at their facility. Even with the parent site (Birmingham) preparing necessary packages in the second phase, obtaining these approvals could be quite time consuming. There was no direct compensation to these individuals. Reasons to participate, as cited by another GRT , may include the desire to improve their clinical practice or an interest in contributing to medical knowledge in general, but these benefits must exceed any perceived disruptive effects. In comparison, at the clinic level, a clinician simply had to log on to enroll and thus be classified as participating.
In comparing phase 1 and phase 2 recruiting, we found, as have others [21, 22], that physician-to-physician recruiting gave a much greater yield and that prior personal contacts did not have a substantial effect. We also learned that recruitment strategies needed to change over time in order to achieve recruitment targets. Similarly, Ellis et al.  used 10 different nonrandomized strategies over 11 months to recruit sufficient practices in the GLAD HEART study, a total of 61 practices, all within one US state. In a review of recruitment rates and strategies across studies conducted in one medical center, Johnston et al.  found considerable variation in recruitment rates despite similar strategies and staffing. Number of recruited practices ranged from 30 to 137; most required over nine months to recruit and most had not planned for the time needed. They found personal connections helpful and have suggested that these personal connections can be developed during the recruiting process. We also found that buy-in from participants (the use of local PIs to champion the study) and a flexible recruitment strategy enhanced recruitment, findings consistent with those of Johnston et al. 
Minimization of possible disruptive effects for the clinician may have facilitated recruitment in our study. First, VHA's use of EHRs made it possible to extract patient records without interfering with office flow. Also, randomization and analysis was at the clinic level, thus low-performing individual clinicians were not at risk of being identified. Similarly, the use of EHRs and clinics as the randomization unit enabled the recruitment of 20 practices in 14 states for a multimethod GRT . The Ornstein study relied on academic detailing and site visits, components that may be disruptive from the theoretical perspective and expensive or impractical for a nationwide study. Interestingly, the parallel MI-Plus study involving primary care clinicians in Alabama and Mississippi [25, 10, 8] had a much lower participation rate (13%) for clinicians , perhaps because these clinicians lacked EHRs and viewed manual chart abstraction as disruptive to their practices.
Between the first and second phases of our recruitment, the amount of time required to obtain facility approval of the study protocol decreased from a median of 255 days to 94 days. This 63% reduction was primarily attributed to the addition of an experienced IRB staff member at the Study Coordinating Center that allowed for the implementation of a more systematic and structured approach to IRB management. The complexity and sheer volume of work needed to coordinate IRB approval for 48 participating facilities cannot be overstated. The majority of facilities required R&D approval prior to IRB submission, and obtaining R&D approval constituted the bulk of the facility approval time, with IRB approval requiring only an additional two to four weeks. This may be misleading in that many R&D committees wanted "the essence of the IRB packet" to review, thus, an IRB specialist is invaluable in facilitating R&D approval as well.
Establishment of the recently implemented central IRB in the VHA (an IRB approved by a central office to cover all participating facilities in a multisite study) should enhance the efficiency, cost, and attractiveness of conducting nationwide GRTs within the VHA. Use of single-study IRB cooperative agreements in the (beta)-Carotene and Retinol Efficacy Trial (CARET) in a university setting reduced the average time to complete IRB approval from over six months to one month for each of many substudies . Even with a central IRB, we anticipate, as have others [27–29], that a dedicated research assistant or IRB specialist is advised in the planning of any large GRT within or external to the VHA. In 2005, with an established protocol and experienced staff, it took approximately six months from initial contact at a facility to enroll an associated clinic; half of this time (three months) was for facility approval, which perhaps can be reduced to one month with the central IRB recently implemented by the VHA. One challenge that will remain, even with a central IRB, is getting PIs to do requisite training in research practices (e.g., good clinical practices, privacy, and security training) needed for IRB approval. This required substantial effort from our study staff, primarily that of the IRB specialist. In an era of ever-increasing regulatory oversight, we believe that this will persist as a substantial task that should be planned for when designing studies and budgeting personnel. A database of and for VHA researchers to register and complete the approval and training necessary to do VHA research should facilitate the recruitment process.
Our conclusions regarding the importance of a functional, truly interdependent relationship between the study PI and the clinical research coordinator echo those of other teams . The success of our study would not have been possible without a close collaboration between these two members of the research team. Evaluating the value-added contribution of such a position should be an important future consideration.
Our experience suggests that using a recruitment approach that seems counterintuitive might be warranted. Our initial efforts to recruit local PIs focused on high-yield targets (i.e., personal contacts), largely due to initial anxiety on the part of the recruitment team of cold calling. While recruiting based on familiarity might have made us feel better, the cold peer-to-peer calling successfully recruited many local PIs and proved less difficult and more efficient than anticipated. We might have saved time and improved study efficiency by expending more energy on cold calling local PIs early and getting the recruitment process started and saving the "easy" recruits for later. Anecdotally, cold calling individual clinicians to log on was not nearly as successful a recruitment tool as cold calling for local PIs. This observation may be a result of being able to offer the facility of local PIs a site distribution of funds ($2,500) to cover costs of participating, while we could not offer clinicians any similar distribution of funds for participation in the study owing to VHA policy.