'Evidence-Based' Studying

In ‘Clinical Epidemiology – A Basic Science for Clinical Medicine [1] the authors considered the process by which medical students learn diagnostic strategies for clinical examination and how this differs from the day-to-day strategies of the fully trained clinician [2]. They proposed that four strategies exist.

1. Exhaustion: Early in their training, students learn to seek all medical facts that could possibly contribute to an assessment of the patient are sought. This, the method of the novice, is prohibitively inefficient. This is because it views diagnosis as a two-stage process. First, collect all possible information. Then (and only then) begin to sift through it for relevant data.

2. Algorithms: The progression of the diagnostic process down but one of a large number of potential, pre-set paths by a method in which the response to each diagnostic enquiry automatically determines the next enquiry to be carried out and, ultimately, the correct diagnosis. It is supremely logical and is most often used when diagnosis of a common problem is delegated from senior to junior medical (or paramedical personnel) or when management strategies for uncommon problems are being considered.

3. Pattern Recognition (Gestalt): This occurs when a complex of symptoms and signs are pathognomonic and conform to a previously learned disease pattern.

4. Hypothetico-Deductive: This is the strategy used by most experienced clinicians. It comprises the formulation, from the earliest clues about the patient, of a "short-list" of potential diagnoses or actions, followed by the performance of evaluations that will best reduce the length of the list. With experience and training, many hypotheses spring forth by pattern recognition of a sort that generates multiple possibilities rather than a single very high probability. The key to efficient diagnosis is the absence of a finding that is virtually always present in the condition being considered. This enables the condition to be removed from the short list. Sub-routines from the strategy of exhaustion are used to search for relevant data. A thorough knowledge of the differential occurrence of key features of the conditions being considered characterises the superior diagnostician.

What makes a finding important and suitable for ‘hypothetico-deductive’ use [3]?

Ideally, findings should have

a) High inter-observer agreement (measured by the kappa statistic) and either

b) High sensitivity. When a finding of high sensitivity is negative, the diagnosis in question is ruled out (SnNOUT) [4] or

c) High specificity. When a finding of high specificity is positive, the diagnosis in question is ruled in (SpPIN) [4].

The threshold for SnNOUTs and SpPINs is not fixed, but probably lies in or around 95%. Any literature deemed to produce such figures should be critically appraised using EBM priniciples. Confidence intervals should be acceptable.

How might this be applied to the study of Radiology?

First, we should consider that our early resident training in film-viewing correlates with the strategy of ‘exhaustion’ and the goal of training is to reach a ‘hypothetico-deductive’ process of analysis. We learn many MCQ facts and long lists of differential diagnoses for our examinations. We learn pathognmonic findings (‘Aunt Minnie’ diagnoses) and use algorithms [5]. We try and learn many facts about many diseases. We do not yet, as EBM physicians do, consciously attempt to identify reliable findings, which if absent effectively exclude a diagnosis or if present effectively confirm it.

This is an area for further research in Radiologic-Pathologic correlation.

In the meantime, we suggest that if you are studying a differential diagnosis list, go to the literature / major textbooks and seek reliable findings that are almost invariably present in each condition. Link them to the condition in your mind. You will find this helpful during film-viewing, whether in examinations or in practice, as you will recall the findings with the diagnosis. Don’t worry about memorising all the findings in every condition – many will overlap between different conditions.

References

1. Sackett DL Haynes RB, Guyatt GH, Tugwell P, Clinical Epidemiology. A Basic Science for Clinical Medicine. 2nd Ed. ed. 1991, Boston / Toronto / London: Little, Brown and Company.

2. Sackett DL Haynes RB, Guyatt GH, Tugwell P, Diagnosis: Clinical Diagnostic Strategies, in Clinical Epidemiology. A Basic Science for Clinical Medicine. 1991, Little, Brown and Company: Boston / Toronto / London. p. 3-18.

3. Sackett DL Haynes RB, Guyatt GH, Tugwell P, The Clinical Examination, in Clinical Epidemiology. A Basic Science for Clinical Medicine. 1991, Little, Brown and Company: Boston / Toronto / London. p. 19-50.

4. Sackett DL, Strauss SE, Richardson WS, Rosenberg W ,Haynes RB, Diagnosis and Screening, in Evidence Based Medicine; How to Practice and Teach EBM. 2000, Churchill Livingstone: Edinburgh. p. 67-93. [ link ]

5. An imaging algorithm for the differential diagnosis of adrenal adenomas and metastases: MM McNicholas, MJ Lee, WW Mayo-Smith, PF Hahn, GW Boland and PR Mueller American Journal of Roentgenology 1995; 165:1453-1459. [ link ]