Becker's Clinical Quality & Infection Control

35 DATA INFORMATICS & ANALYTICS Seattle database tracks superbugs, helps hospitals find best drugs to fight them By Megan Knowles T o better equip hospitals in the fight against antibi- otic resistance, a group of physicians and scientists in Seattle launched a nonprofit organization in September to develop a data- base of antibiotic-resistant bacteria, the Seattle Times reported. The group, Antibiotic Resistance Monitoring, Analysis and Diagnostics Alliance, or ARMADA, said it wants to "get a grasp on the situation" by building a database that tracks and "fingerprints" bacterial strains. "Right now a patient comes to the doctor with an infection, and the doctor knows a lot about the patient, but doesn't know anything about the bug," said ARMADA founder Evgeni Sokurenko, MD, PhD. "How a patient will react to an infection will be tracked in their medical records, but nothing is being tracked about the bug, even though the bug will repeat itself, and no one will know what it has done in other patients." ARMADA will gather information on how superbugs react to a variety of antibiotics. By giving this data to hospitals and microbiology labs, providers can immediately find the most effective drug for patients. The group has collected data for 30,000 strains so far and aims to increase this number tenfold in the next few years. The project receives funding from the National Institutes of Health and partners with Seattle Children's Hospital, Seattle-based Harborview Medical Center and Oakland, Ca- lif.-based Kaiser Permanente. n Stanford scientists use bioinformatics to trace hospital-acquired bloodstream infections to patients' digestive tracts By Harrison Cook R esearchers at Stanford (Calif.) University found hospital-ac- quired bloodstream infections often begin in a patient's digestive tract, according to a small study pub- lished by Nature Medicine. For the study, researchers used bioinformatics software to analyze blood and stool samples from 30 pa- tients who developed a bloodstream infection after receiving a bone-mar- row transplant between October 2015 and June 2017. Researchers also analyzed each patient's entire gene sequence. Many of the bloodstream infections started in a patient's body, often in the large intestine. Researchers found 15 patients' stool samples re- vealed detectable levels of the same bacterial strain responsible for their bloodstream infections. "Until now, we couldn't pinpoint those sources with high confidence," Ami Bhatt, MD, assistant professor of hematology and genetics at Stanford University, said in a news release. "That's a problem because when a patient has a bloodstream infection, it's not enough simply to admin- ister broad-spectrum antibiotics. You need to treat the source, or the infection will come back." Researchers found little evidence of a pathogen from one patient's blood- stream infection matching bacterial strains in other patients' blood or stool. "I don't think we're passing around active infections among one another as often as has been assumed," Dr. Bhatt said. "Our results suggest that people are the most likely source of their own infections. Maybe we need to get rid of this idea of 'catching' others' infections, and give more thought to the health of our own resident microbial ecosystems." n NYU uses Google's AI to identify lung cancer By Jessica Kim Cohen R esearchers at New York University in New York City trained an artificial intelligence algorithm developed by Google to distinguish between two common types of lung cancer, Wired reported. e researchers used Google's Inception v3 — an open-source deep learning algorithm — and trained it to differentiate between cancerous and healthy tissue images using thousands of images collected by e Cancer Genome Atlas, a public database of patient tissue samples. ey also taught it to distinguish between two types of lung cancer, which are similar in appearance: adenocarcinoma and squamous cell carcinoma. e researchers tested the retrained deep learning algorithm on image samples from cancer patients at NYU, and found it accurately diagnosed patients 83 to 97 percent of the time — only using images of the tumor, and without including information from sequencing or other tests. "I thought the real novelty would be not just to show the AI is as good as humans, but to have it provide insights a human expert would not be able to," Aristotelis Tsirigos, PhD, a pa- thologist at NYU School of Medicine and a lead author on the new study, told Wired. "ese cancer-driving mutations appear to have microscopic effects that the algorithm can detect." e researchers plan to continue to train the algorithm with data from additional sources and are considering seeking FDA approval. n

• Cover