Duncan, who first went to Texas Health Presbyterian Sept. 25, was diagnosed with sinusitis, not Ebola on that visit. He was prescribed antibiotics, told to take Tylenol and discharged roughly 4½ hours after he arrived, the report states. Duncan returned Sept. 28, desperately ill, was admitted, subsequently infected two nurses who gave him care and died on Oct. 8.
In one of its earliest public reports of what went wrong, Dallas-based Texas Health Resources, Texas Health Presbyterian's parent organization, pointed the finger at its EHR and its configuration for not passing Duncan's travel history of having recently arrived from Ebola hotspot Liberia, from an intake nurse to a treating physician. In a statement only a few hours later, the hospital system walked back that assertion.
“Assigning blame to the EHR is not new and often reflects a reluctance to address the complex cognitive and/or performance issues involving front-line staff, especially those related to responsibility and accountability,” the report authors said in their account of the incident. “Acknowledging the reality that EHRs suffer from major usability and interoperability issues, they are only tools and not a replacement for basic history taking, examination skills and critical thinking.”
While the hospitals' EHR recorded and forwarded Duncan's travel history for example, it didn't do enough.
“The EHR didn't help,” said Dr. Hardeep Singh, the article's corresponding author and a patient-safety researcher at the Veterans Affairs Center for Innovations in Quality, Effectiveness and Safety. “It didn't aid in the process. We could have done better with the EHR. We could have had diagnosis support. We could have used (fewer) templates. It's not a single thing to point a finger at.”
Joining Singh in the research were Divvy Upadhyay, research associate at the Urban Institute's Health Policy Center in Washington, D.C.; and Dean Sittig, a professor in the School of Biomedical Informatics at the University of Texas Health Science Center at Houston and a member of the University of Texas at Houston-Memorial Hermann Center for Healthcare Quality and Safety.
The team of three medical informaticists pored over the publicly available record and, drawing on their own considerable experience and knowledge of how clinical IT systems work, wrote this report as a postmortem of those events.
“We've literally been reporters and the story has been changing every day,” said Singh, who is also a safety researcher at the Michael E. DeBakey VA Medical Center and Baylor College of Medicine, both in Houston.
According to an account by the Associated Press, which was provided by Duncan's family with copies of more than 1,000 documents of his patient records (which were unavailable to the three authors), the ED physician's note used a phrase saying Mr. Duncan was 'negative for fever and chills.'
This doesn't jibe with the subsequent release by the hospital of a chronology of Duncan's care that said he first presented at the ED with a fever of 100.1, which spiked while he was there to 103, then dropped to 101.2. The “negative” assessment, the authors suggested, “could have been selected erroneously from a series of predefined symptom options.”
Further on, the authors observed, the note read: “I have given patient instructions regarding their diagnosis, expectations for the next couple of days, and specific return precautions. The condition of the patient at this time is stable.”
“Because of the generic nature of this phrase and the use of a gender-neutral plural pronoun when referring to a patient, we believe this phrase was likely selected from a set of predefined patient instructions,” the authors wrote.
The researchers also noted that the nurse taking information about Duncan included his travel history, concluding “the biggest red flag in this case was the patient himself, an African with a foreign accent who reported he came from Liberia and presented with serious 'flu-like' symptoms.”
Typically, travel history is often a consideration in making an initial diagnosis, but it became critical information, in hindsight, in the Duncan case. The nurse recorded it using a template “designed to provide a high reliability nursing process to allow for the administration of influenza vaccine under a physician-delegated standing order,” the authors said.
“These highly constrained tools are optimized for data capture but at the expense of sacrificing their utility for appropriate triage and diagnosis, leading users to miss the forest for the trees,” the researchers said. “As this case illustrates, EHR-based clinical workflows often fail to optimize information sharing amongst various team members, leading to lapses in recognizing specific clinical findings that could aid in rapid and accurate diagnoses. “
Diagnostic errors are common in healthcare, the authors said, pointing to studies indicating diagnostic error “is thought to be in the range of 10% to 15% in the U.S., producing 12 million misdiagnosed adults each year.
An Institute of Medicine committee is working on a report about diagnostic error due to be released next year.
Stand-alone, computerized clinical-diagnosis-support systems have been available for decades, but work is just begun on maximizing their utility and ease of use by integrating these tools into regular clinical workflows supported by mainstream electronic health-record systems.
Computers and data analytics could be marshaled in other ways into the looming battle to address this complex problem of missed diagnoses, the authors said.
“Physicians need timely feedback on diagnostic accuracy if they are to learn from their actions,” they said. Today, however, “Such systems don't exist, nor are there any incentives to build them. Widespread use of EHRs and better data availability should facilitate development of these systems,” the authors said.
Follow Joseph Conn on Twitter: @MHJConn