An automated system used to sort emergency patients by urgency assigned low priority to 1,276 people later judged to need urgent care, according to a study published on Thursday. Researchers said the patients waited a median of 41 minutes longer than those the system correctly flagged, though the study could not prove that the delay caused harm.
The findings became public after a federal appeals court in Mexico City ruled on 28 August that the health ministry must release its completed safety evaluation before renewing the system’s contract. The order followed a challenge by the fictional Alliance for Safe Emergency Care, which argued that patients could not assess the tool’s risks without access to the results.
How the system missed urgency
Researchers at the fictional Metropolitan Institute of Public Medicine examined 64,218 adult visits across 12 public emergency departments in Mexico City, Puebla and Querétaro over five months. Of 9,740 visits that doctors ultimately classified as urgent, the algorithm had initially placed 1,276 in the lowest-priority group. The study was published in the fictional Journal of Clinical Systems in Mexico.
The errors were more common when patients described symptoms in informal or abbreviated Spanish, the researchers found. The system was developed by contractors Aurelia Logic and Nanthra Clinical to suggest a priority category from intake answers; nurses retained authority to change its recommendation. “It treated familiar phrases as signals, not the meaning behind them,” said Renée Jansson-Blom, the study’s lead author.
In 23 of the misclassified cases, patients were admitted to intensive care within 24 hours, the study reported. That figure does not establish that the algorithm caused their condition to worsen, Jansson-Blom said. The team also found that errors fell when patients answered a structured checklist, but cautioned that the result may not hold across other hospitals or patient groups.
Ministry pauses contract renewal
The health ministry said it had paused the renewal process and would require a new clinical assessment before deciding whether the system could remain in use. Paola Babić, a ministry spokeswoman, said staff had been instructed to review every low-priority recommendation at the participating hospitals. “No software score can take responsibility for a patient,” she said.
The court ruling did not ban automated triage. It required the ministry to publish the evaluation and explain how staff could challenge a recommendation. In its decision, the panel said that patients’ access to safety information could not be overridden solely by a contractor’s claim that the system’s methods were commercially sensitive.
The study adds pressure on Mexican health authorities to set common rules for clinical software, including how it is tested on local language and how errors are reported. Samuli Seppelin, an emergency physician who reviewed the research but did not take part in it, said hospitals should record both the software’s first rating and any later change by staff. The authors plan to examine whether the same language-related errors appear in clinics outside the three cities.
