Learning from tragedy – improving diagnosis through case reviews

In this issue of Diagnosis, we are initiating a new feature: “Case Reports of Diagnostic Error”. Many journals feature case reports, with the most well-known being the “Case Records of the Massachusetts General Hospital” that have appeared in the New England Journal of Medicine for most of the past century. These are addressed to health care professionals and meant to be instructional, focusing on the disease and its differential diagnosis, while also exploring the pathophysiology and clinical features. The purpose of the case reports published in Diagnosis is novel and completely different: the focus of these reports is not the disease, but the diagnostic process. The goal is learning how to improve that process, thereby avoiding harm related to diagnostic error for future patients. There is another unique feature of these reports that has been lacking in traditional reports: they will include the perspectives (as much as possible) of the patient or family, the providers involved in the patient’s care and other major stakeholders. This is fundamental, as the hope of every family of every patient harmed by diagnostic error is that a similar error never happens to another person. Thus, sharing these stories is an important step in fulfilling these families’ aching desires.

The first case in this series reports the tragic death of Julia Berg, a 15-year-old girl, who died after an unnecessary operation for presumed cholecystitis. The actual diagnosis, discovered post-mortem, was Epstein-Barr virus (EBV) infection with hepatitis, that is, mononucleosis. What can be learned from this case about improving diagnosis?

The first lesson is the value of learning from an in-depth analysis of a single case. The diagnostic process is uniformly idiosyncratic, and a nuanced approach must be taken when analyzing the diagnostic process that unfolded in a given case. This case highlights the importance of skillfully navigating all the steps of the diagnostic process; it is conceivable, if not likely, that Julia may have survived her EBV infection if greater attention had been paid to her presenting complaints and laboratory findings. All the necessary facts were available to make the correct diagnosis, but because her history of fatigue was dismissed, and her striking lymphocytosis was not appreciated as being salient, the correct diagnosis was not considered. The team locked in on an ‘obstructive’ pattern of liver function tests, and interpreted this as representing a diseased gall bladder. Lymphocytosis would be uncommon in cholecystitis while it is a classic finding in EBV infection – in this case, the finding that did not fit really mattered. In case series of diagnostic error, these breakdowns in appreciating the salience of findings, and synthesizing the available information, are the most common elements contributing to error [1], [2]. There is great educational value in highlighting when findings don’t fit – and encouraging health care professionals to ask “What if the finding that doesn’t fit really matters?” with every patient.

The next lesson from this case is the value of the autopsy in revealing the correct diagnosis. Although autopsies have virtually disappeared in many centers, there are a few remaining centers interested in performing these procedures. In 2014, a specialty center in Europe reported a major diagnostic discrepancy rate of 23% [3], and a series of 334 cases from the United States found a 10% discrepancy rate [4]. Thus even in the ‘high tech’ era, the autopsy has value in identifying diagnostic errors, and remains the gold standard as a research tool. Given these striking discrepancy rates, we believe it is a moral obligation to ensure that providers and families know the actual diagnosis affecting patients. Reviving the autopsy as a tool for learning was a major recommendation of the National Academy of Medicine Report on Improving Diagnosis in Health Care [5] that has yet to be acted upon. There are multiple reasons for autopsies not being performed, but these factors are completely surmountable and restoring an appreciable rate of autopsies is central to improving our understanding of our diagnostic outcomes. Autopsies are likely one of the most effective means of learning from harmful diagnostic (and other) errors and it is unacceptable to forego them for reasons of expediency.

A third lesson relates to the what can be learned (and what cannot be learned) about diagnosis from administrative data. There is a growing trend to study diagnostic outcomes from ‘big data’, that is, using administrative data derived from the electronic record and associated coding linked to each encounter. This approach has been used successfully, for example, to study diagnostic error associated with stroke [6]. Julia’s case is a powerful reminder, however, of the limitations imposed by looking just at administrative data. Her death certificate lists the cause of death as fulminant EBV infection and recent cholecystectomy. Would a computer algorithm have interpreted these discordant two facts as representing a diagnostic error? In short, our current coding practices and the many inaccuracies captured on death certificates create serious limitations on the value of trying to learn from aggregated analyses that rely on this information [7]. It is likely that complex algorithms will need to be developed to search for “signals” that signify diagnostic error in these large datasets until a code that designates a case as a diagnostic error (or missed opportunity) is developed. This does not yet exist but would have substantial value for research and learning, not to mention transparency and culture change.