Skip to content

Minimizing the Threats of Antimicrobial Resistance and Infections Associated with Antibiotic Use

The introduction of effective antibiotics in the 1940s ushered in an era of optimism with rapid declines in deaths due to infections. Since that time, however, antibiotic resistance has emerged rapidly, and too few antibiotics are making it through the development pipeline. That means once curable infections could soon become more virulent and even untreatable.

Click here to watch Disruptive Dozen: 12 AI Technologies That Will Reinvent Care

Indeed, the decline in the usefulness of antibiotics is troubling on multiple fronts, but perhaps one of the most serious is the accompanying rise in infections with multidrug resistant organisms, particularly in hospitals and other health care settings. The danger lies not just in the “super bugs” themselves, but also in another infection linked to antibiotic use, Clostridium difficile colitis.

C. difficile causes life-threatening diarrhea, primarily in those recently treated with antibiotics. While there are good treatments for C. difficile, as many as 25% of patients will relapse. The bacteria can survive in the environment for months, increasing the risk of transmission. Strikingly, in the period from 2000 to 2007, deaths from C. difficile rose 400%. In 2011, nearly half a million Americans became infected with C. difficile and roughly 30,000 patients died within a month of diagnosis. Moreover, C. difficile represents the leading cause of healthcare-associated nosocomial infection in the U.S., costing acute care facilities approximately $5 billion each year.

Researchers are now working on multiple fronts to combat this formidable health threat. That includes efforts to leverage machine learning and other artificial intelligence (AI) approaches. For example, researchers in Massachusetts and Michigan are developing a tool that uses clinical data from patients’ electronic health records (EHRs) to create an AI-based risk score that reflects a patient’s likelihood of developing a C. difficile infection. The score incorporates diverse types of data, including medications, procedures, health care settings, health care staff, lab results, vital signs, patient demographics, patient history, and admission details. Not only does this approach integrate thousands of variables, it also models changes in risk during the course of an inpatient hospital stay, making it possible to create daily estimates of C. difficile infection risk for each patient. These risk predictions can then be used to develop targeted interventions, such as infection control strategies and antimicrobial stewardship, to help minimize the dangers posed by this formidable bacterium.

Methicillin-resistant Staphylococcus aureus (MRSA) is another pathogen that poses significant health care challenges. In hospitals, MRSA is associated with serious and invasive conditions, including pneumonia, surgical site infections, and sepsis. More than 80,000 invasive MRSA infections occur annually in the U.S. These infections are more lethal, costly, and difficult to treat than those involving non-resistant forms of S. aureus: it is estimated that MRSA costs the U.S. health care system approximately $10 billion each year.

Investigators are turning to AI to help tackle these challenges, too. For example, a research team in Massachusetts is using machine learning to identify the earliest signs of MRSA infection by mining data in patients’ EHRs, including both clinical and non-clinical information. Their method seeks to identify MRSA cases before they are diagnosed, giving physicians advance warning and helping them more effectively target preventive and therapeutic measures. On a similar front, researchers in the U.K. recently developed a machine-learning method to predict MRSA virulence based solely on the microbe’s genome sequence. By analyzing different genomic features, including single nucleotide polymorphisms and small structural changes, the team’s method could pinpoint clinical isolates that are likely to produce severe complications in patients. Such an automated tool, when combined with EHR-based approaches like those described above, could provide an additional layer of information to help clinicians better tailor treatment according to the needs of individual patients. That kind of precision is essential to preserving the long-term success of our current — and future — antibiotic arsenal.

For more information about Dr. Shenoy’s research, please contact Partners HealthCare Innovation by clicking here.