Sepsis Early Warning System SepsisCatch: Alerting Clinicians 6 Hours Before Deterioration

SepsisCatch, jointly developed by DeepMind Health in the UK and Imperial College London, is an AI early warning system for sepsis. The system analyzes subtle patterns in electronic medical records and can issue an alert six hours before sepsis deterioration, with 89 percent accuracy.

Sepsis is one of the leading causes of death among ICU patients worldwide. The clinical challenge is early recognition: a patient can deteriorate rapidly from a mild infection into life-threatening shock. Traditional inflammatory markers such as procalcitonin and C-reactive protein lag, often rising significantly only after the patient's condition has already deteriorated.

SepsisCatch's training data comes from eight years of electronic medical records at 15 NHS hospitals in the UK, covering about 4.2 million inpatient stays. The model learned to identify subtle features in the pre-sepsis phase, including reduced heart rate variability, slight rises in respiratory rate, minor blood pressure fluctuations, and small trends in lab results.

The system is deployed by integrating with hospital HIS systems to analyze the records of every inpatient in real time. When the model's assessment of a patient's sepsis risk crosses a threshold, the system immediately sends an alert to the assigned nurse's mobile device, along with the specific adverse indicators and recommended next steps.

In the pilot phase, SepsisCatch was deployed for 18 months at 12 NHS hospitals in the UK. Results show sepsis-related in-hospital mortality fell 23 percent and ICU transfer rates fell 18 percent. That kind of improvement is unprecedented in traditional sepsis management.

The medical community's reaction to SepsisCatch is mixed. Supporters see it as best practice for AI-assisted medicine, with high alert accuracy and no significant additional burden on clinicians. Anthony Gordon, professor of critical care medicine at Imperial College London, noted that sepsis care is all about time, and AI alerts can buy a valuable early intervention window.

Critics worry about the algorithm's interpretability. When SepsisCatch raises an alert, clinicians sometimes have trouble understanding its reasoning. The DeepMind Health team responded that the company is developing visualization tools to help doctors understand the specific abnormal indicators behind each alert.

On the business model, SepsisCatch is currently offered to NHS hospitals as a subscription at 12 pounds per bed per month. DeepMind Health has signed a framework agreement with the NHS and expects to extend the service to all NHS hospitals by the end of 2026.

On international rollout, DeepMind Health has signed agreements with Kaiser Permanente in the US, Singapore's National Healthcare Group, and the Canadian Health Information Research Institute. These organizations will localize and deploy SepsisCatch in their countries, with expected coverage of more than 100,000 hospital beds.

SepsisCatch's next development direction is to support early warning for more critical conditions. Domhnall Carlin, head of research at DeepMind Health, said the team is exploring similar technology for acute kidney injury, cardiac arrest, and acute respiratory distress syndrome.