Becker's Clinical Quality & Infection Control

34 DATA INFORMATICS & ANALYTICS Penn Medicine researchers predict depression diagnoses from Facebook posts By Jessica Kim Cohen A Facebook user's language ticks may tip off a clinician to their likelihood of developing depression, accord- ing to a study published in the Proceedings of the National Academy of Sciences. Researchers from Philadelphia-based Penn Medicine and Stony Brook (N.Y.) University parsed through the Facebook statuses of more than 500,000 consenting patients, some of whom had been diagnosed with depression, to determine what they called "depression-associated language markers." Armed with these findings, the researchers created an algorithm they hoped would accurately predict a patient's likelihood of being diagnosed with depression, based only on language used in their Facebook statuses. The algorithm flagged Facebook statuses for language that referenced symptoms like sadness, loneliness and hostility through words like "tears" or "feelings." The algorithm also considered frequency of various words and phrases, such as how commonly a user posted a self-ref- erential status with first-person pronouns like "I" and "me" — these users were likely to be diagnosed with depression. "Social media data contain markers akin to the genome," Johannes Eichstaedt, PhD, founding research scientist of the World Well-Being Project at Penn and Stony Brook and a postdoctoral psychology fellow at Philadelphia-based University of Pennsylvania, said in a statement. "Depression appears to be something quite detectable in this way; it really changes people's use of social media." To evaluate the algorithm, the researchers applied it to the Facebook statuses of nearly 700 consenting patients who had visited an emergency department at a large urban academic medical center, 114 of whom had a diagnosis of depression listed in their medical records — and the algo- rithm proved accurate, even when only looking at a limited time frame. In fact, the algorithm could accurately predict a patient's likelihood of being diagnosed with depression when only analyzing Facebook statuses from three months before the first documentation of depression in their medical record, according to the researchers. Based on these findings, the researchers suggested social media networks, such as Facebook, could one day be used by clinicians to screen consenting patients for depression. "The hope is that one day, these screening systems can be integrated into systems of care," Dr. Eichstaedt said. "This tool raises yellow flags; eventually the hope is that you could directly funnel people it identifies into scalable treatment modalities." n Google creates AI to detect when breast cancer spreads By Jessica Kim Cohen G oogle's artificial intelligence algorithm LYNA accurately detects the spread of breast cancer — but the company is taking pains to highlight that the tool is meant to assist, not replace, human pathologists. LYNA, which stands for "LYmph Node Assistant," uses a type of AI modeled on how the human brain processes information, called "deep learning." A deep learning algorithm learns over time by extracting patterns from a data set. For Google's project, that means researchers trained LYNA on pathology slides from patients with metastatic breast cancer, or the stage of breast cancer where the dis- ease has spread from the primary tumor site to nearby lymph nodes. In an Oct. 12 post on the company's research blog, Google out- lined findings from two studies related to LYNA, which it calls a "proof-of-concept pathologist assistance tool." e first study, published in the Archives of Pathology and Labo- ratory Medicine, applied the AI algorithm to two new sets of pa- thology slides to assess LYNA's accuracy. Google reported LYNA correctly distinguished slides with metastatic cancer from slides without the disease 99 percent of the time in both data sets. e algorithm could also pinpoint the location of cancers and other "suspicious regions" within each slide, some of which were too small to be consistently detected by the human eye. Today, a pathologist's examination of a tumor using a micro- scope is considered the "gold standard" for cancer diagnosis, according to Google's blog post, despite the reality that small metastases, or micrometastases, are difficult to detect. Based on the findings from its first study, Google's research team sug- gested LYNA might prove a successful tool to highlight areas of concern for human pathologists to review before delivering their final diagnosis. e second study, published in e American Journal of Surgical Pathology, explored the potential clinical or workflow benefits the AI algorithm could provide. Six board-certified pathologists completed a simulated clinical assessment during the study, in which they reviewed images of lymph nodes for metastatic breast cancer both with and without assistance from LYNA. Using the AI algorithm improved diagnostic accuracy for the physicians, reducing the rate of missed micrometastases by a factor of two. e pathologists also reported that diagnosing micrometastases was "easier" with LYNA, according to the study results. With LYNA, the average time for a pathologist to review a slide was only one minute, compared to two minutes without the tool. "Encouragingly, pathologists with LYNA assistance were more accurate than either unassisted pathologists or the LYNA algo- rithm itself," according to Google's blog post. "is suggests the intriguing potential for assistive technologies such as LYNA to reduce the burden of repetitive identification tasks and to allow more time and energy for pathologists to focus on other, more challenging clinical and diagnostic tasks." n

• Cover