A carregar...
Publicação
Dissecting medical referrals in health services
| Nome: | Descrição: | Tamanho: | Formato: | |
|---|---|---|---|---|
| 1.88 MB | Adobe PDF |
Orientador(es)
Resumo(s)
Background: Medical referrals between primary care (PC) and specialist care (SC) physicians are essential to integrated healthcare delivery but often encounter inefficiencies arising from communication gaps, coordination issues, and organi zational structure. While prior studies have explored patient-sharing networks, limited quantitative evidence exists on how physicians’ professional relationships, such as co-training, co-affiliation, or network embeddedness, relate to referral behavior. This study develops a data-driven framework combining network anal ysis and graph neural networks (GNNs) to examine how professional-network structure characterizes referral formation. Methods: The study analyzed five years of consultation data from a large Portuguese private healthcare system. Two complementary networks were con structed: a directed referral network linking PC to SC physicians based on patient transitions within a 30-day window, and an undirected professional network representing shared institutional and educational ties. Graph-based embedding models (GraphSAGE, Attri2Vec, Node2Vec) were trained to predict referral links, using physician demographic and network-based inputs. Feature importance was evaluated using Shapley Additive Explanations (SHAP), and model embeddings were visualized using UMAP to assess structural coherence. Results: Incorporating professional-network information improved referral pre diction accuracy across all models. SHAP analyses identified degree and eigen vector centrality as dominant predictors of referral probability, indicating that physicians with broader professional connectivity and greater embeddedness were more likely to receive referrals. Age and gender effects were minor and secondary to structural network factors. Visualization of learned embeddings revealed that professional-network information produced coherent physician clus ters aligned with hospital co-affiliations and cross-role interactions, demonstrating that professional networks capture referral proximity and organizational cohesion. Conclusions: Physicians’ professional-network positions are strongly associated with referral behavior and provide incremental predictive value beyond demographics alone. Embeddedness and centrality reflect relational mechanisms of trust, reputation, and coordination that relate to access to specialist care. Incorporating network analytics into referral management can inform policy interventions aimed at improving transparency, balancing referral loads, and promoting equitable access to specialist services.
Descrição
Publisher Copyright: © The Author(s) 2026
Palavras-chave
Medical referrals Professional networks Social network analysis Graph Neural Networks Link prediction Health services research Healthcare coordination