Singapore General Hospital developing AI to prevent antibiotic resistance
Singapore General Hospital is developing an AI solution to determine the necessity of prescribing antibiotics, reduce their use, and identify the most appropriate ones for each patient.
Developed with DXC Technology, the Augmented Intelligence in Infectious Diseases (AI2D) is an AI model currently covering pneumonia cases. It was built using deidentified clinical data, including X-rays, clinical symptoms, vital signs, and infection response trends, of around 8,000 SGH patients between 2019 and 2020. It covered seven broad-spectrum intravenous antibiotics commonly prescribed against pneumonia.
FINDINGS
Last year, the research team led by the SGH Pharmacy division, conducted a pilot validation study of the AI model, comparing it against 2,000 pneumonia cases in 2023.
In a paper, SGH and DXC noted that the AI2D helped cut down cases to review by three times (624 out of 2,012). It also raised the possibility of identifying cases requiring intervention among that heap of review cases by almost 12%, compared to 4% in a full manual review process. Moreover, analysing data in one case, which took 20 minutes in a manual review, was done in “less than a second.”
In the study, the AI model demonstrated 90% accuracy in determining whether administering antibiotics is necessary in a given pneumonia case; the study also revealed that almost 40% of antibiotics prescribed for those cases might have been unnecessary.
WHY IT MATTERS
SGH said that pneumonia accounts for 20% of all infections treated at their hospital, with antibiotic prescriptions written the most out of all infections. Patients tend to be hospitalised between two to nine days on average, costing the government up to SG$5,000 (over $3,500) for each government-subsidised patient admitted to a public hospital.
Half of acute care hospitals worldwide are potentially prescribing the wrong antibiotics to patients, resulting in antimicrobial resistance. At SGH, for example, the hospital unnecessarily prescribed 20%-30% of broad-spectrum intravenous antibiotics, based on a 2018 antibiotic use udit. In Singapore, up to 30% of hospital-acquired infections are believed to have become resistant to broad-spectrum antibiotics.
Hospitals are establishing antimicrobial stewardship programs to tackle this growing global concern by preventing the overuse of antibiotics and identifying where these – especially narrow-spectrum antibiotics – are best recommended. Such programs can potentially help reduce hospital stays, minimise deaths and readmissions, and save patient and hospital costs.
In carrying out antimicrobial stewardship programs, using automation and AI could be the best approach, as teams would require insights in real-time at the point of prescription. AI would help identify review cases and prioritise ones requiring intervention. Automation would allow hospitals to expand the coverage of their antibiotic use audit, providing thorough insights into their usage.
The research team is now preparing a comparative study involving 200 SGH inpatients to test the effectiveness of their AI model for reducing antibiotic use. Later, they will work on determining the most effective antibiotic against pneumonia, and then build the same model covering urinary tract infection, another common hospital-acquired infection.
THE LARGER TREND
Outside Singapore, the Taiwanese hospital, China Medical University Hospital, has been recognised for its effort to address antimicrobial resistance by utilising AI. First implemented in 2021, its Intelligent Anti-Microbial System is a four-in-one AI-powered platform that identifies drug-resistant strains, predicts and monitors sepsis and death, recommends drug doses, and compares drug-drug interactions. This specific application of AI in the EMR system was highlighted in the hospital’s revalidation for Stage 7 of the HIMSS Electronic Medical Record Adoption Model in November last year.
ON THE RECORD
“Well-intended doctors are constantly balancing the risks and benefits of antibiotic use. It is often hard to tell definitively that patients will benefit from it based on clinical assessment, patient-specific factors, or condition severity alone. If not prescribed promptly for those who genuinely need them, it may lead to dire complications. Yet, misusing antibiotics contributes to antibiotic resistance, posing challenges for future infection treatment,” said Dr Piotr Chlebicki, AI2D project member and a senior consultant at SGH Department of Infectious Diseases.