Diagnosis by Search Engine in a CPC

Case 22-2015 [1] marks a sad nadir for the Journal’s clinicopathological conferences. The main problem is that the discussant, after a two-paragraph differential diagnosis, used an Internet search engine to extend the differential, justifying this maneuver because it was “a challenging case.”

While physicians often do need help, and should not be afraid to seek it, it’s easy to see why this case’s approach ill suits a teaching exercise…. Every large hospital has at least one gray-haired master diagnostician on its staff who serves as a walking, talking “search engine” for tough cases. Asking that diagnostician to help with a challenging case is no different than asking a computer search engine, but this is not the strategy on which CPCs are based. No CPC in the past has ever had the discussant give up after just two paragraphs and ask a master diagnostician to bail him out. Yet that is what happened here.

Perhaps the editors believe that comments from the patient’s attending physician served as a replacement for the traditional diagnostic discussion. Of course this cannot be true. In a diagnostic-type of CPC (there are now many types of CPC), the whole point of the discussion is to have a clinician deconstruct the case in a prospective vein. Otherwise, the exercise becomes merely a memoir for the attending physician, and that is what we have here: entire categories of disease eliminated by weak heuristics (e.g. no infection because no response to antibiotics – systemic mycosis, anyone?) with little of the intellectual rigor that makes the diagnostic CPCs so great, and none of the nuance associated with disease entities that makes the CPCs so interesting.

There are multiple other unfortunate aspects to this CPC.

First, by employing a deus ex machina (literally) to reach the right diagnosis, the NEJM editors and the discussant forgot the main purpose of the CPCs, as stated more than 50 years ago by Castleman and Dudley: “The clinicopathological conferences remain an exercise in deductive reasoning and clinicopathological correlation. It is less important to pinpoint the correct diagnosis than to present a logical and instructive analysis of the pertinent conditions involved” [2]. The present CPC not only has comparatively miniscule teaching value, but it actually negates the leadership role that CPCs have in the past served. It serves only as an exemplar of intellectual laziness.

Second, and worse, the discussant’s diagnostic approach was fundamentally flawed. When the search engine’s first 10 results all pointed to the same disorder, the discussant abandoned the search for other explanations, claiming “there are no other obvious diagnostic considerations” and committing the single most common cognitive error in diagnosis: premature closure [3]. Medicine has great mimics that one ignores at one’s peril [4]. There is also the ever-present possibility that Occam’s razor fails, and that two disease entities are at work. Moreover, if 10 of the first 10 search engine results point to the same diagnosis, then the skill in using a search engine in this case would have been using it to identify a second and third item for the differential. The discussant was too lazy to overcome any of these pitfalls.

Third, the discussant’s lazy approach meant that relevant information related to the case was omitted from the CPC discussion. For example:

• He picked hyperferritinemia as one of the key terms presented to the search engine. Although the literature on adult onset Still disease does indeed frequently mention high ferritin levels, the discussant should have clarified that ferritin is simply acting as an acute phase reactant, meaning that several other lab tests could yield the same diagnostic tip-off [5]. It would also have been useful to note the sensitivity and specificity of hyperferritinemia.

• How did the discussant pick the terms to present to the search engine? Supplying a different set of terms, picked from the patient’s many clinical abnormalities, leads to much different search engine results. Such a “sensitivity analysis” was an obvious topic for discussion.

• The discussant did not address the fact that some search engines (e.g. Google) adjust the search results presented to a user based on the user’s past search history [6]. So, a rheumatologist is going to bias Google to return rheumatology results. It may not have mattered much in this case, but any person in medicine who uses a search engine for clinical purposes should know of this phenomenon.

Fourth, the diagnostic approach cannot be replicated in non-English languages because on-line resources in other languages will not only be different, but less comprehensive. Medicine is a universal profession and until now the CPCs have been universal teaching tools when translated into other languages. Marginalizing a large fraction of the world’s physicians is good for neither patients nor the profession.

Fifth, and most disheartening, the discussant described clinical reasoning as a “diagnostic test,” as if it were as far outside the reach of bedside medicine as an MRI scan or a laparotomy.

I can’t decide whether the NEJM published this CPC as a gimmick, or to serve notice that we are in a brave new world where careful thought is unnecessary in clinical medicine, or simply because they have never had a case of adult onset Still disease in a CPC and needed a novel mechanism to make it happen (because no pathologist involvement is present).

Although it was a game effort, the central mistake was forgetting the heritage of CPCs. As Castleman and Dudley also stated in 1960: “The clinicopathological conference has been adversely termed a `guessing game,’ but this is not true if the case has been selected wisely. The properly chosen case lends itself to a discussion of the differential diagnosis.” In this instance, there was effectively no discussion of the differential.