Bloodstream infection subtypes and characteristics comparing solid organ transplant and nontransplant populations

Bloodstream infections (BSIs) are common and are associated with high mortality rates. Few studies have examined the heterogeneity of clinical characteristics in solid organ transplant (SOT) recipients with BSIs. We used machine learning (ML) to identify clinically distinct subtypes in SOT and non-SOT BSI patients. We applied unsupervised ML to clinical variables collected within 48 hours of index BSI diagnosis, clustering 15 550 patients from a major medical center in Houston, Texas. Using k-means++, we identified 3 major subtypes (α, β, and γ). Patients in cluster α were older, predominantly male, and required more vasopressor and ventilator support compared with other clusters. SOT patients in cluster α included more liver and lung transplant recipients and developed BSI closer to the transplant date (165 days, P < .01). Propensity score matching was applied to compare the mortality in SOT and non-SOT groups. Although SOT patients exhibited higher rates of drug-resistant pathogens, SOT patients had a lower 30-day mortality compared with non-SOT patients overall (9.0% vs 12.8%, P = .01), driven by cluster α specifically (18.3% vs 33.1%, P < .01). ML approach demonstrated the potential to identify distinct phenotypes of BSIs in both SOT and non-SOT patients, which may support precision medicine and contribute to future clinical research.