AI to Combat Hospital-Acquired Infections – A Revolution for Patient Safety

By Claire Paris, MD MBA FHM, VP of Medical Affairs and Chief Medical Officer, UNC Lenoir Healthcare

If we could save upwards of $30 billion a year on avoidable healthcare costs, why wouldn’t we? This is what the CDC estimates that hospital-acquired infections cost annually. 1 in 25 patients will suffer a hospital-acquired infection—many of these result in actual harm to the patient. For example, a central line or urinary catheter left in place too long causes an infection. Frustratingly, hospitals can also have these infections identified that may not be true infections, but fall into the NHSN criteria. These are costly in terms of unnecessary testing, financial penalties for hospitals, and lower publicly reported scores.

AI would alter the approach with an infection prediction with increased diagnostic accuracy. It could help discern acute inflammation from infection.

AI is poised perfectly to help us predict the patients that will get these infections through the use of predictive analytics scanning vast amounts of data combined with real time monitoring of the patient’s vital signs and other data points. This would allow us to mitigate those risks by removing or replacing the problematic lines. It can also predict multidrug-resistant organisms that could put a patient at risk. Multidisciplinary teams all have their roles in preventing these infections and AI suggestions and recommendations could be targeted towards the members of these teams.

AI would alter the approach with an infection prediction with increased diagnostic accuracy. It could help discern acute inflammation from infection.

While the risk of hospital-acquired infections depends on the hospital’s infection control practices, and those steps taken to reduce the risk, patient factors opposing these efforts are also at play which include immune status, recent antibiotic use, frequent visits to healthcare facilities, length of stay (LOS), major procedures, age, ventilatory support and intensive care stays. It seems quite plausible that artificial intelligence (AI) could identify risk factors and generate a score. Steps could be suggested and taken to mitigate infections by keeping devices out as much as possible and guiding clinician care decisions.

Cleveland Clinic investigators recently presented that AI could very accurately predict multi drug-resistant organisms days prior than a culture is available. It is exciting to think that we can use AI to predict and tailor antibiotics and isolation precautions towards these days ahead of a final culture. Taking antimicrobial stewardship to the next level to get patients appropriately treated earlier will save lives, time and money.

AI has recently been used to model new designs of urinary catheters to block the migration of bacteria towards the bladder. Catheters were made consistent with these designs, creating an obstacle course of geometric designs inside the catheter that blocked the migration of bacteria upstream. The design was optimized for E. coli, and testing showed that after 24 hours the bacterial burden was 1/100 of that of traditional Foley catheter design. This is exciting that we can use AI technology to predict the behavior of microbes and design ways to inhibit their growth and migration.

Machine learning (ML) algorithms have demonstrated value in predicting clostridium difficile infection with just 6 hours of data. With almost 30,000 deaths per year related to c. difficile infections, early diagnostics to treat, identify those at risk and isolate to prevent spread would be an incredible advance to saving lives.

CLABSI (central line-associated bloodstream infection) could be predicted allowing physicians to remove the lines prone to infection and avoid those consequences. Suggestions of treatment based on probability and risk would help discern true CLABSI from blood culture contamination.

The support for antimicrobial stewardship that AI could provide would adjust the approach toward treating infections. Currently, the physician evaluates data for the likelihood of an infection. Cultures are taken, the results of which will not be available for several days, and empiric antibiotics are started. When cultures and sensitivities are available, antibiotics are sometimes changed based on those results, or de-escalated. Have we then given a patient a toxic or broad-spectrum antibiotic for a few days that was unnecessary? Have we bred more resistance? AI would alter the approach with an infection prediction with increased diagnostic accuracy. It could help discern acute inflammation from infection. (Is this sepsis or something else?) The correct antibiotic could be chosen which would eliminate the need to de-escalate or change, and the best duration of therapy would be suggested.

Enhanced cleaning and sterilization practices could be suggested by algorithms to identify and mitigate risks with equipment and high touch areas in the healthcare setting.

Patient and caregiver education and engagement could be enhanced through AI based on their medical conditions or procedures to provide targeted and relevant material about their care and infection prevention practices. This would certainly foster a collaborative culture of safety, and mitigate the spread of infections.

I dream of a day when a patient is admitted to the hospital, using AI tools, we are able to get predictive scores on the likelihood of hospital-acquired infections or other complications. More informed decision-making can be made based on the probability instead of blindly pan-culturing when not needed or leaving devices in patients at high risk. Imagine that when a patient comes to the hospital, your risk of CAUTI, CLABSI or hospital-acquired pressure ulcer is present and available to the admitting team so that decision-making to reduce or eliminate this outcome altogether can be made. A most likely diagnosis and risks are presented along with the likely infection cause/ Is it MDRO or not? The right antibiotics are given and a short LOS gets the patient home safely and efficiently. Lives will be spared, and billions of dollars potentially saved.