Researchers are finding new ways to make sure physicians make the right call Geisinger Medical Center doctors including Dennis Torretti, middle left, and Benjamin Hohmuth, middle right, are developing a system to improve diagnostic accuracy and identify patients who may be experiencing errors. Doctors are developing novel solutions to make sure they come up with the right diagnoses. A flood of new initiatives by researchers, physicians, health-care systems, nonprofits and malpractice insurers is yielding new insights and approaches. These include sophisticated computer programs, some that use artificial intelligence to help analyze and diagnose tough cases, and others that scan records for errors such as missed test results and appointments. Advanced technologies aren’t just bringing the processing power of big data and machine learning to bear. They are also allowing more doctors to share their knowledge—including lessons they’ve learned from their own diagnostic mistakes. Misdiagnoses are a leading cause of medical malpractice suits in the U.S., and they don’t have simple causes or solutions. Most people will experience at least one diagnostic error in their lifetime, sometimes with “devastating consequences,” according to a landmark 2015 report by the National Academies of Sciences, Engineering and Medicine. For decades, inaccurate or delayed diagnosis has been “a blind spot in the delivery of quality care,” said the report, which is broadly credited in scientific circles with inspiring the recent multipronged push for improvements. “There is a huge opportunity to put knowledge and tools into the hands of clinicians to help them make a better or more timely and accurate diagnosis,” says Janet Corrigan, head of the patient-care program at the nonprofit Gordon and Betty Moore Foundation, which is funding several projects to improve the accuracy of medical diagnoses. Here are some of the most promising efforts: Spotting Mistakes A Pennsylvania hospital is experimenting with the first formal computerized program to track, measure and analyze doctors’ diagnostic mistakes—sometimes as they are happening. The Safer Dx Learning Lab is being implemented at Geisinger Medical Center, the flagship of Geisinger Health System, based in Danville, Pa., which has 13 hospitals, a health plan and a medical school. The project is funded by a Moore Foundation grant and is led by Hardeep Singh, chief of health policy, quality and informatics at the Michael E. DeBakey VA Medical Center and an associate professor at Baylor College of Medicine in Houston. Once fully installed, the system will scan electronic medical records to identify patients who are experiencing potential or actual diagnostic mistakes or delays, based on certain clinical criteria, such as abnormal chest X-rays or CT scans for which there is no documented follow-up. Flagged records will then be reviewed for confirmation, analysis and further action. For example, Geisinger will look at patients who return to the ER unexpectedly within 48 hours after discharge, or a hospital patient who was unexpectedly transferred to an intensive-care unit to see what happened and why. “Knowledge from systematically analyzing missed opportunities in correct or timely diagnosis will inform improvements and create a learning health system for diagnosis,” Dr. Singh says. Dr. Singh’s research suggests doctors often miss common conditions with significant potential for harm. A 2013 study led by Dr. Singh reviewed 190 diagnostic errors detected by unexpected returns of patients after an initial primary-care visit. A large variety of common diseases such as pneumonia, cancer and urinary-tract infections were missed on the initial visit. Many errors were related to problems with medical history-taking, physical exams, ordering of diagnostic tests for further work-up and failure to review previous records. Other issues included clinicians failing to initiate referrals to specialists and failures in follow-up and tracking of diagnostic tests. “We could ask the question in every patient diagnosed with cancer, was the appropriate diagnosis made in a timely fashion, or were there opportunities, on review, to have made this correct diagnosis earlier?” says Dennis Torretti, associate chief medical officer at Geisinger Medical Center. Geisinger also is adding a confidential hotline reporting system where clinicians can call and leave a recorded message. By reducing the reporting burden, the system aims to gather information on several types of opportunities that are currently being missed. That information will be analyzed by the learning lab to develop improvements, Dr. Singh says. Similarly, Geisinger will ask patients how it can do a better job learning about their experiences related to diagnosis. Current patient-satisfaction surveys, he says, don’t ask questions that could identify issues related to missed diagnoses. “We all see patients who had to see doctors several times before a diagnosis was achieved and had a feeling all along that something was being missed, or were discharged from the ER and felt the diagnosis didn’t explain their symptoms,” says Benjamin Hohmuth, principal investigator on the Safer Dx project and associate chief medical informatics officer at Geisinger. Harnessing patient feedback from such experiences could help identify system failings and make sure patients get follow-up faster when a breakdown occurs, Dr. Hohmuth says. Help from AI Many online tools have been developed to analyze symptoms and help doctors arrive at a diagnosis—with mixed results. Now a system that uses AI can synthesize opinions from many doctors into a single cohesive perspective. The Human Diagnosis Project has developed an electronic consulting system that lets participating doctors enter a patient’s background, symptoms, test results and other findings, then invites expert doctors to review the case, suggest a diagnosis and recommend next steps. Users can include specific questions, such as, “Does this patient need an MRI?” As doctors post their responses, the AI-enhanced software combines and analyzes all the input, weighted by each doctor’s relative expertise. More than 6,000 doctors from 500 institutions and 40 medical specialties have been building the system since 2014, backed by an alliance that includes medical schools, medical specialty boards and research experts. The aim is “not only to deliver the best possible diagnosis for the patient at hand, but to form a record that doctors can draw on for similar cases in the future,” says Shantanu Nundy, a primary-care doctor at Mary’s Center, a community-health center in Washington, D.C., and director of the nonprofit arm of Human Dx, which makes the system available for social, academic and noncommercial uses. “You often hear that if you talk to five doctors you get six opinions, which leaves the patient at a loss for what to do next,” says Dr. Nundy. “Human Dx takes multiple doctors’ opinions and synthesizes them into one.” Researchers at Johns Hopkins, University of California, San Francisco and Harvard are testing the performance, cost and training value of the system. “Generally, doctors can diagnose better than computers, but doctors supported by technology like this system could help improve the accuracy of clinical decisions across the board,” says David Westfall Bates, professor of health policy and management at Harvard T.H. Chan School of Public Health and chief innovation officer at Brigham and Women’s Hospital in Boston. Sanjay Desai, director of the Osler Medical Training Program at Johns Hopkins, is experimenting with Human Dx as a tool to help evaluate residents’ skills in making a diagnosis. “One of the biggest challenges is trying to measure a physician’s ability to clinically reason,” says Dr. Desai. If successful, such a system could make medical expertise more widely available to patients who lack the means or access to affordable specialist care. Sharing Lessons Another effort to improve practices and training, though not necessarily through a single tool, is a network in Massachusetts including Brigham and Women’s Hospital; Crico, a Harvard-owned malpractice insurer; and the Boston-based second-opinion service Best Doctors. The network, known as Pride, short for Primary Care Research in Diagnostic Errors, plans to identify, analyze and classify diagnostic errors and delays with the help of electronic medical records, to develop and share interventions that can overcome diagnostic errors and delays, especially in primary care. The network will disseminate insights and lessons to doctors through forums, newsletters, journals and webinars. Among the advances it hopes to deliver are screening tools to uncover issues related to drugs or exposures that can lead to missed or delayed diagnosis, such as patients who are on multiple drugs, or who may have been exposed to toxins on the job or at home. It also plans to help doctors avoid ordering unnecessary and wasteful tests by developing “principles of conservative diagnosis,” says Gordon Schiff, associate director of Brigham and Women’s division of general internal medicine and quality and safety director at Harvard Medical School’s Center for Primary Care. Another key goal is to identify “diagnostic pitfalls” such as when a doctor mistakes one disease for another, or presumes an existing disease is responsible for new symptoms, or fails to monitor evolving symptoms, Dr. Schiff says. By reviewing malpractice claims and safety reports and conducting focus groups with specialists, Brigham and Women’s has compiled a list of more than 400 disease-specific pitfalls, which will be used to identify specific problems that are ripe for improvement. Crico, meanwhile, is also funding a project overseen by Dr. Schiff to develop five tools to help avoid pitfalls that leave doctors open to malpractice claims. For example, studies have shown that as few as 40% to 50% of orders for colonoscopies result in patients’ getting the test. In response, the project plans to develop and test “loop-closing” tools for electronically tracking doctors’ recommendations of tests and procedures that aren’t carried out. The project will also develop educational tools to help patients understand the need to follow up after doctors’ appointments and tests. “We don’t want to abdicate our responsibility to follow up on test results,” says Dr. Schiff, “but we have to tell patients it is their responsibility to play a role. If you haven’t heard anything after a test in two weeks, get back to us.” Better thinking Some medical schools are creating programs to overcome pitfalls in thinking that can affect a doctor’s diagnostic skills and become ingrained early in medical training. The University of Pittsburgh Medical Center has developed videos of cases with actors posing as patients. The videos feature expert commentary, animation and Q&As. The aim isn’t only to improve the clinical reasoning skills of new doctors, but to make them aware of the prevalence, causes and consequences of diagnostic errors, according to William Follansbee, director of the UPMC Center for Medical Decision Making and a professor of cardiovascular medicine. In making diagnoses, doctors may rely on shortcuts or simple rules of thumb based on what they’ve seen in past cases, Dr. Follansbee says. They may also anchor their thinking in an initial impression of a case or prematurely make a diagnosis before pursuing alternative possibilities. The most important first step in making a diagnosis, Dr. Follansbee says, is to ask good questions in a careful medical history. He developed the “two-steps-back checklist for diagnostic decision making,” which calls on doctors to take two steps back before they make a diagnosis—unless it is an obvious one—and ask themselves a series of questions. These include, what else it could be, whether patients are taking, or not taking, medications, and whether a test will provide answers they need. “These might seem like common, simple questions we all know,” he says, “but they are frequently overlooked.” Source
