AI predicts presence of Pseudomonas, MRSA in sputum cultures
Click Here to Manage Email Alerts
Artificial intelligence can help predict the chance of resistant organisms appearing in sputum cultures in the ER, helping to guide therapy, according to findings presented during the ASM Microbe virtual meeting.
“AI can be used to predict certain tests which require significant time until they are available so that doctors can commence more targeted therapy earlier,” Joongheum Park, MD, an internal medicine physician and hospitalist at Beth Israel Deaconess Medical Center in Boston and cofounder of AvoMD Inc., told Healio.
“We believe that the same methodology can be applied to multiple clinical targets. Examples include predicting co-bacterial infection in those with COVID-19 — you can start antibiotics to cover bacterial infection for those patients.”
Park and colleagues developed a machine learning model to estimate sputum culture results for pneumonia. They analyzed 1,677 admissions from a single tertiary care center in Boston. The researchers included pneumonia patients who were admitted from the ER to the ICU and had positive sputum culture within 72 hours of their admission. They used input features including demographics, medical history, basic laboratory test results and initial vital signs.
Of the admissions analyzed, 26% of sputum cultures grew MRSA and 15% grew Pseudomonas. The researchers’ prediction model achieved an area under the receiver operating characteristic curve — a measurement that predicts the probability that a patient who experiences an event will show a higher predicted risk than a patient who does not — of 0.64 for Pseudomonas and 0.67 for MRSA.
“Certain traditional risk factors were not included in our model. For example, patients who received recent antibiotics were known to be at increased risk of having drug-resistant bacteria,” Joowhan Sung, MD, a hospitalist at MedStar Southern Maryland Hospital, told Healio. “If the information on recent antibiotic use can be added to the model, prediction performance can likely further improve.”
Sung said there are two major drawbacks to starting patients on empiric broad-spectrum antibiotics.
“One drawback is the drug’s side effect. For example, Vancomycin is one of the most commonly prescribed antibiotics to treat MRSA infection, but it has kidney toxicity,” Sung said. “Kidneys of patients with active infection are already vulnerable to the damage, so we do not want to use it unless we have some degree of clinical suspicion of MRSA infection. The second drawback is the potential development of drug-resistant bacteria. The more patients that receive broad-spectrum antibiotics, the more likely they will develop a drug-resistant infection in the future.”
Park noted that AI can be utilized to expedite the use of certain therapies.
“Doctors should be able to make more efficient clinical decisions with better outcomes if we can predict the results in advance,” Park said. “For example, we chose pneumonia, where sputum culture can help identify causal bacteria, but growing the bacteria takes several days to finish — you might be too late when they are available. If we can predict the results earlier with AI, then we should be able to implement targeted antibiotic therapy earlier.”
Collapse
Sung J, et al. Virtual sputum culture: Artificial intelligence (AI)-assisted prediction of sputum culture result in patients with pneumonia in the emergency room. Presented at: ASM Microbe; June 18-20, 2020 (virtual meeting).
Click Here to Manage Email Alerts