Right-Sizing QI Collaboratives for Real-World, Resource-Lean Residency Settings

Dr Petrilli and colleagues offer a clear, practice-grounded framework for launching multi-institutional quality improvement collaboratives.¹ Our Family Medicine Midwest Scholarly Activity Collaborative, a multistate effort supported by an American Board of Family Medicine Foundation Residency Learning Network Planning Grant, found much to agree with in their six steps, and a few areas where our experience suggests alternative approaches for resource-lean settings. Our comments are intended as complements to the authors’ steps and examples.

Aligning with organizational priorities can secure leadership support, but Healthcare Effectiveness Data and Information Set (HEDIS)-anchored topics are often already “owned” by health systems, creating redundancy and uneven starting points across sites. In our 2024-2025 hypertension project, some systems had mature initiatives while others had none, complicating shared aims and comparisons. We encourage quality improvement (QI) collaboratives to pick a high-yield topic with low institutional ownership, which might mean avoiding HEDIS and existing pay-for-performance measures. Examples include goals to reduce unnecessary referrals, because grooved systems can achieve excellent or even superior outcomes in primary care, such as diagnosis of adult ADHD in primary care. 

Another goal selection strategy for resource-lean QI collaboratives is to adopt a “replicate-a-proven-single-site project” in which a QI project completed at one site can be implemented at other sites. This capitalizes on the collaborative structure by allowing the inception site to serve as a resource for the subsequent sites. These parameters for goal selection encourage each residency to implement an innovative change package locally, then the collaborative works to synthesize the lessons learned across sites.²^,³

A five-role, multidisciplinary team is aspirational. Smaller community residencies struggle to secure adequate QI time for even one person. We agree with Dr Petrilli et al that several tasks will need to be accomplished, and identifying those tasks is essential. Our experience was that only one person from each residency site was able to participate in the collaborative meetings, and that person needs to ensure that they can accomplish the tasks within their site.

We concur that IRB and data-use agreements become rate-limiting once cross-site identifiable data are involved. We recommend designing the collaborative so that analysis occurs locally using common operational definitions, with aggregate, deidentified summaries and qualitative change packages exchanged centrally. Each site can conduct its QI project prior to collaborative IRB approval because local QI projects can proceed without IRB approval. This two-stage process preserves learning while minimizing IRB complexity and institutional reluctance to share granular data.⁴

The figure presented by Dr Petrilli et al is a helpful scaffold, but our multisite project required far more time to converge on a topic, harmonize measures, and clear approvals, which dilutes the tempo of classic rapid-cycle plan-do-study-act (PDSA) processes. A two-stage plan (as suggested here) may fit real-world constraints: Stage 1 focuses on within-site adaptation of a change package; Stage 2 synthesizes cross-site results and dissemination. This structure maintains momentum, reduces coordination drag, and still yields generalizable insights.²^,³

In short, Dr Petrilli et al advance the conversation; we suggest an additional modified pathway for programs with leaner teams and heterogeneous systems. We would welcome further detail on the authors’ total project duration, data-sharing strategy, and measurable outcomes as such information would help programs calibrate effort against expected benefit.