When undergoing cancer treatments, recovering from surgery, or battling an infection, you may also have to fight off another enemy: sepsis.

Sepsis is the body’s extreme response to an infection and develops when an infection you already have triggers a chain reaction throughout your body. Without timely treatment, sepsis can rapidly lead to tissue damage, organ failure and death. The cancer population is particularly susceptible because of immunosuppression from treatment, surgery and transplants.

“Many of our patients are at a high risk for developing sepsis and the key is early recognition and prompt treatment,” said Dr. Sachin Apte, associate chief medical officer at Moffitt Cancer Center. “The survivability is reduced when it is caught late. Due to immunosuppression and the consequences of many anti-cancer treatments, sepsis can be difficult to diagnose early in cancer patients.”

To help with early detection, Moffitt is designing an Artificial Intelligence tool to predict if a patient will develop sepsis as early as seven days before they become septic. The tool is initially learning from the data available from the charts of bone marrow transplant patients previously diagnosed with sepsis, as these patients are especially susceptible to infection.

The tool looks at 325 variables such as vital signs, lab results, medications and patient demographics in a 60-day window. It works best at ruling out sepsis and could operate as a checks and balances system for clinicians.

“What we are envisioning is getting physicians to be acutely aware of symptoms and if they think a patient is going to be septic, check with the model,” said AI Officer Dr. Ross Mitchell. “If the model rules out sepsis, the provider can avoid putting the patient through difficult tests or treating with drugs that have side effects.”

The preliminary model is accurate 85% of the time predicting sepsis two to seven days before patients become septic, when applied to the test data.

“Now that we have built this, how do we actually deploy it and make the information available to the right people at the right time?” said Mitchell. “This is a huge undertaking. Developing the algorithm was the easy part.”

Before deploying the tool for clinical use, the model must be prospectively validated in a new patient population. Then the Clinical Informatics Department will work with providers to determine the best way to integrate the tool into a clinical setting.

“The goal is to help clinicians to think of sepsis earlier, and give them the tools they need to treat it sooner,” said Chief Medical Information Officer Dr. Randa Perkins. “Our role is to make these tools easier to access and available at the right time in clinical care. How can we help our colleagues get the data they need when they need it to help patients? The computer has become another part of the provider’s toolkit, the keyboard along with the stethoscope.”

There is still a lot of work to do on the project, but if it’s successful it could have a major impact on patient care.

“You certainly want to prevent death from sepsis, but beyond that, we need to significantly reduce morbidity, length of hospital stay, avoid any psychological effects of having sepsis, and get patients back to treatment sooner,” said Apte.

The ultimate long-term goal would be to build a generic base model program that can be tuned for other cancer types, offering a better way to detect sepsis early for all Moffitt patients. 

“What is really exciting about this tool is that it is the first of many, and the lessons we learn from this one, even in just the building of it, will advance our development of health IT and informatics in ways we can’t even predict yet,” said Perkins.