CONCERN

COmmunicating Narrative Concerns Entered by RNs (CONCERN) early warning system: a pragmatic cluster-randomized controlled trial

Paciente / Población Intervención / Exposición Comparación Desenlace

In 60,893 adult hospital encounters on 74 cluster-randomised clinical units across two health systems, the CONCERN machine-learning early warning system, which mines real-time nursing surveillance documentation reduced in-hospital mortality and length of stay compared with usual care.

N
60.893
Diseño
Pragmatic cluster RCT (74 units, two health systems)
Desenlace
In-hospital mortality (instantaneous risk) and length of stay (co-primary)
Relevancia
1
ResultadoIn-hospital mortality aHR 0.64 (95% CI 0.53–0.78; P<0.0001), a 35.6% lower instantaneous risk of death; length of stay aIRR 0.91 (95% CI 0.90–0.93). Sepsis aHR 0.93 (95% CI 0.86–0.99); unanticipated ICU transfer aHR 1.25 (95% CI 1.09–1.43).
Rossetti SC, et al. Nat Med. 2025;31(6):1895-1902. 10.1038/s41591-025-03609-7
Discusión y crítica

The first cluster-randomised trial to show a machine-learning early warning system reducing in-hospital mortality, and notable for its signal: patterns in how nurses document, rather than vital signs or laboratory values alone. The rise in unanticipated ICU transfers is consistent with earlier escalation but is also a cost. Unblinded, pragmatic and limited to two health systems over one year; mortality was analysed as instantaneous risk, and generalisability to settings with different nursing documentation cultures is untested.