Reducing referrals in primary care through specialist-supported telehealth: quasi-experimental evidence from Northeast Brazil.

Background: Synchronous tele-interconsultations - real-time interactions linking primary care physicians and hospital-based specialists - have been proposed as a way to support clinical decision-making and, potentially, to build clinical competencies over time through repeated exposure to specialist reasoning. However, causal evidence on whether such programs reduce reliance on specialist referrals remains limited, particularly in low- and middle-income country settings characterized by large territories and uneven specialist distribution. Methods: We conducted a quasi-experimental evaluation of the TeleNordeste project, a large-scale telehealth initiative implemented in Northeast Brazilian states under a public-private partnership. Using staggered difference-in-differences methods, our primary aim was to estimate the treatment effect on primary care referral rates, using a panel of primary care units observed from 2018 through 2025. After applying a municipality-level restriction to reduce spillover contamination, the analytical sample comprised 6,825 primary care units: 3,918 never-treated comparison units, 1,002 low-dose units (1-4 cumulative telehealth interactions), and 1,905 high-dose units (≥5 interactions). Secondary analyses explored whether estimated effects were more consistent with immediate care-flow reorganization or workplace learning. Results: Any exposure to the program was associated with a statistically significant reduction in referral rates (-0.356 percentage points, no-covariate specification; Conclusions: Participation in the TeleNordeste program was associated with meaningful reductions in referrals to specialized care. Although immediate case resolution during tele-interconsultations provides a plausible explanation for this association, workplace learning cannot be confirmed or excluded with the available data. In the absence of a clear dose-response gradient, these mechanisms cannot be disentangled, and the results indicate that program participation itself can strengthen primary care case resolution and reduce reliance on specialist referrals. Interpretation of mechanisms is limited by potential under-registration of referrals and by the use of primary care units, rather than individual physicians, as the unit of analysis. Future evaluations would benefit from collecting physician-level identifiers and strengthening audits of referral-registration completeness, thereby enabling a more detailed assessment of how tele-interconsultations affect clinical practice and referral decisions over time.