Undertriaging linked to worse outcomes in postop patients

NEW YORK (Reuters) – Among postoperative patients at increased risk for complications, those who were sent to low-acuity floors rather than to the intensive care unit (ICU) had higher morbidity and mortality in a cross-sectional study.

“This study was prompted by recognition that after major surgery, high-risk patients are occasionally admitted to general wards with infrequent vital sign and bedside assessments, which allow postoperative complications to progress to severe illness between assessments,” Dr. Tyler Loftus of the University of Florida Health in Gainesville told Reuters Health by email.

“Undertriage occurs when those providers underestimate a patient’s risk for adverse events, and occasionally when ICUs are full and cannot accommodate additional patients,” he said.

“Results from this study could be implemented in clinical settings by halting electronic order placement when those orders are triaging a high-risk patient to a general ward,” he said. “In these cases, the order could be escalated and reviewed by other providers, including those with critical care training.”

As reported in JAMA Network Open, Dr. Loftus and colleagues compared hospital morbidity and mortality among appropriately triaged ward admissions, undertriaged ward admissions and a risk-matched control group of ICU admissions.

Ward admissions were considered undertriaged if their estimated risk for hospital mortality or prolonged ICU stay (i.e.,48 hours or longer) was in the top quartile among all inpatient surgical procedures according to a validated machine-learning model available at surgical end time.

Among 12,348 postoperative ward admissions from University of Florida hospitals, 11,042 (89.4%; median age, 59; 54%, women) were appropriately triaged, whereas 1,306 (10.6%) were undertriaged and matched with a control group of 2,452 ICU admissions.

Compared with the control group, the undertriaged group was older (median age, 64 vs. 62), had increased proportions of women (49.7% vs. 44.0%), and had patients admitted with do-not-resuscitate orders before their first surgical procedure (4.1% vs. 1.1%). Ultimately, 207 undertriaged admissions (15.8%) had a subsequent ICU admission.

In the validation cohort of close to 21,000 admissions used to train the machine learning model, both hospital mortality and prolonged ICU stay estimations had areas under the receiver operating characteristic curve of 0.92.

The undertriaged group, compared with the control group, had a similar incidence of prolonged mechanical ventilation (2.5% vs. 2.2%); decreased median total costs for admission ($26,900 vs. $32,700); increased median hospital length of stay (8.1 days vs. 6.0 days); and increased incidence of hospital mortality (1.5% vs. 0.7%).

The undertriaged group also had increased discharges to hospice (1.8% vs. 0.6%); unplanned intubation (3.4% vs. 2.0%); and acute kidney injury (26.1% vs. 19.5%), as well as increased incidence of procedures such as red blood cell transfusion, arterial catheter placement, and central venous catheter placement.

Dr. Loftus said, “At surgery end time, machine learning algorithms can identify patients with high risk for adverse events who should be admitted to an ICU. Clinicians should also be aware of the opposite problem: overtriage of low-risk patients to ICUs. Low-risk patients occupying ICU beds can incur unnecessary expenses and create situations in which other patients with greater need of close surveillance are denied admission to a full ICU.”

Dr. Sherry Wren of Stanford University, coauthor of a related editorial, commented in an email to Reuters Health, “It is exciting to see the development of real-time clinical decision tools leveraging machine learning and large EHR datasets. These data help inform clinicians on new variables to consider when making a decision on admission location.”

“As a surgeon, I typically determine admission location primarily based on procedure type, comorbid medical conditions, urgency, intraoperative hemodynamic stability, and whether pressors or blood transfusions were necessary,” she said. “Therefore, it was a surprise to learn that anesthetic gas concentrations and duration of inhalation agents (key features considered in the model’s triage classifications) had such a strong effect on mortality and ICU stay. This opens up new discussions about this with my anesthesia colleagues.”