© 2026. Houston Methodist, Houston, TX. All rights reserved.
back2.svgforward2.svg
fb_share.svgtw_share.svgin_share.svg
NURSING SCIENCE
Article Review:
A Randomized Controlled Trial of Artificial Intelligence-Based Analytics for Clinical Deterioration

By Astrid Wu, BSN, RN, PCCN – Nursing Science Fellow – Houston Methodist Hospital

Science-1.jpg
Title:
A Randomized Controlled Trial of Artificial Intelligence-Based Analytics for Clinical Deterioration
Published: Feb. 5, 2026, in Scientific Reports
Level of Evidence: AACN Level of Evidence C
What was the purpose?
Clinical deterioration is often preceded by changes in vital signs, but delayed recognition can increase morbidity and mortality. There is limited evidence from randomized controlled trials of predictive analytics. This study evaluated whether CoMET artificial intelligence (AI), which analyzes cardiorespiratory data, charted vital signs and laboratory results, then passively displays patients’ risk score and increased event-free hours among patients in acute care cardiology and medical-surgical units. The authors hypothesized that CoMET could facilitate early identification of patients at risk of adverse outcomes and increase the number of hours free from clinical deterioration.
What was the population studied?
The population studied was patients admitted to an 85-bed acute-care cardiology and medical-surgical unit from Jan. 4, 2021, to Oct. 4, 2022. The units included medical-cardiology, cardiothoracic surgery and medical-surgical patients, including some solid-organ transplant patients. Because the authors wanted to test CoMET in an authentic clinical setting, all patients admitted to these units were enrolled in the study.
Was the setting comparable to Houston Methodist? Were the nurses like our nurses?
The study setting was reasonably comparable to that of Houston Methodist because both are academic hospital environments that care for complex cardiac and medical-surgical patients and use continuous monitoring and interdisciplinary care. However, the authors did not describe nurses’ experience, education or staffing ratios, so it is unclear whether the participating nurses were comparable to nurses at Houston Methodist.
Did the authors use the appropriate methods?
The authors used a cluster-randomized controlled trial in which 11 room-based clusters were assigned to either the intervention or control group. Clusters were re-randomized every two months using a Latin square design to maintain balanced assignments. This design was appropriate because it allowed clinicians to identify intervention patients without adding complexity or creating false reassurance when a rising risk score was not visible. Enrolling all eligible patients also reflected a real-world acute care environment.
The primary outcome was hours free from clinical deterioration within 21 days, including emergency ICU transfer, emergent intubation, cardiac arrest, emergent surgery or death. Patients remained eligible for event-free hours while clinically stable. Those with non-emergent ICU transfers, planned surgeries or bed changes were censored because these events did not represent unexpected deterioration. This approach preserves construct validity and reduces data contamination from confounding events.
What were their findings?
The results showed no statistically significant difference in the primary outcome between the full cohort and those at risk for clinical deterioration. The median number of event-free hours was 504 in both groups, indicating that most patients completed the 21-day study period without a qualifying deterioration event. Only 5.3% of patients experienced a clinical deterioration event.
Among patients who experienced a deterioration event, those in the display group experienced more events than those in the control group. There were also no significant differences in death, emergency ICU admission, emergency intubation, cardiac arrest or emergency surgery. Patients whose CoMET scores increased substantially had an average hospital stay of approximately 6.8 days, compared with 3.4 days among patients without a substantial increase. This finding suggests that the AI system may have been effective in identifying patients with greater illness severity, even though displaying the scores did not improve the primary clinical outcome.
Did the findings make sense?
The findings were reasonable. A risk score alone may not improve outcomes unless clinicians receive standardized education and follow a clear response protocol. For the system to be effective, clinicians must notice the score, understand its meaning, communicate the information and take timely action. Because CoMET was a passive display without required responses, clinicians may not have used the information consistently.
Other factors also affected the findings. The low deterioration rate made it difficult to detect statistically significant differences, and the COVID-19 pandemic disrupted staffing and hospital resources. Approximately 11% of patients moved between display-on and display-off beds and were censored. Clinicians appeared to move sicker patients into CoMET-equipped beds, suggesting they perceived value in the display. However, these transfers disrupted randomization, made the groups less comparable and may have biased the results toward finding no difference.
How did things change?
The AI display did not produce a statistically significant change in event-free hours or other major patient outcomes compared with usual care. However, clinician behavior may have changed because some clinicians moved sicker patients into CoMET display beds. The display may therefore have influenced clinical decision-making even though the study did not demonstrate better patient outcomes.
The study also showed that a higher AI risk score was associated with longer hospitalizations and greater patient risk. Future implementation may be more effective if the system includes actionable alerts, standardized escalation procedures and clearer guidance about how nurses and other clinicians should respond to changes in risk.
How is this important for nursing?
This study is important for nursing because nurses continuously assess patients and monitor vital signs, trends and lab results. With increasingly complex workloads, AI-based predictive systems may assist nurses in the early detection of subtle deterioration. However, the study demonstrates that technology alone cannot replace nursing judgment, nor is it most effective when integrated into established clinical workflows.
For AI tools to be useful in nursing practice, nurses must receive appropriate, standardized education on interpreting risk scores and have a clear procedure for responding to an increase. For example, a significant increase could prompt the nurse to reassess the patient, verify vital signs, review recent laboratory results, notify the provider or activate the rapid response team when appropriate.
Overall, this study provides valuable evidence that a passive AI display alone is insufficient to improve patient outcomes. Future research should examine how nurses interpret AI-generated risk scores, how the information affects clinical decision-making and whether active alerts combined with standardized nursing interventions can lead to earlier treatment and reduced clinical deterioration.
Reference:
Keim-Malpass, J., et al. (2026). A randomized controlled trial of artificial intelligence-based analytics for clinical deterioration. Scientific Reports, 16, 7345. https://doi.org/10.1038/s41598-026-39051-z