TUTORING CHEST PAIN DIAGNOSIS: RELATIVE CONTRIBUTION OF CASES, CONTRASTS, AND PROTOTYPES

   Objective. To discover the most effective way to teach diagnosis, we compare the effects of tutorial materials using textbook descriptions of diseases, versus feature lists relevant to a single disease, or to pairs of diseases, and we characterize the impact of the number of case examples.   

 

   Method. In group sessions, students in graduate physician assistant or physical therapy programs or in undergraduate nursing programs studied a chest pain diagnosis tutorial with information and case examples about six diseases. For each participant, each disease was taught with a different format, crossed with either 0, 1, or 4 case examples, using 18 randomly assigned booklets. The formats were: Textbook (T) description including clinical appearance, physiological basis, and epidemiology; Prototype (P), a list of clinical features ranked by sensitivity; Contrast (C), features of a pair of easily confused diseases, grouped into sets of shared versus discriminating features; the T&P and C&P combinations; and a control. 205 students completed pre- and post-tests (to diagnose 2 or 3 cases per disease, multiple choice), and 159 completed a followup test after at least 2 weeks.

 

   Results.  For the two diseases that each participant studied with no case examples, the mean proportion of test items correctly diagnosed improved from .553 to .597, or 4.4%, averaging over all 6 tutorial formats. With one example case, accuracy improved 17.7% (F(1,199)=16.99, p<.0005); with four examples, it improved 19.1% (NS better than 1 example). Follow up improvements were 5.6%, 13.5% (F(1,153)=9.09, p=.003), and 13.8% (NS), respectively.

   Three formats (C, C&P, and T&P) produced the highest gains in proportion of correct diagnoses, significantly better than the control (all p<.01). Two other formats (P and T) were only marginally significantly (p<.20, two-tailed) better than control, in the expected direction. The first set was marginally more effective than the second set (all p<.10 or p<.20). At followup, there were no significant differences.

 

   Conclusion.  Studying example cases promoted learning and retention of diagnostic knowledge. Four examples were no better than one. Formats that contrast features between diseases produced better immediate learning than control, as did the text book extract with the symptom list. The text book extract alone, and the feature list for just one disease, were less effective. Format differences did not persist after 2 weeks.