https://www.jameslindlibrary.org/articles/paradigm-lost-carl-liebermeister-and-the-development-of-modern-medical-statistics/

-------------------

Feinstein AR. The inadequacy of binary models for the clinical reality 
of three-zone diagnostic decisions. J Clin Epidemiol 1990;43:109–13. 
Coste J, Pouchot J. A grey zone for quantitative diagnostic and 
screening tests. Int J Epidemiol 2003;32:304–13. 
  
The first paper discusses the inappropriate use of a binary framework for 
clinical diagnostic decisions, and the second paper outlines an approach to 
defining the limits of a 'grey zone'. Not sure if it is what you are looking 
for, but an interesting read nonetheless.   -Bethany Shinkins


FH 22Aug13:

I wanted to stir the pot a bit to bring up something that I feel strongly about after working in medical diagnostic research since 1980.  The traditional approach to teaching MDs about probabilistic diagnosis, incorporating sensitivity and specificity, has created significant problems clinically, including:

- the tendency to dichotomize the disease instead of considering severity and likely progression, and labeling patients as "diseased" when it is harmful to do so (see the attached editorial)

- the tendency to dichotomize the test output when the vast majority of medical tests yield a continuous spectrum of results (even the pregnancy test, which could have been developed to yield an "uncertain - retest" category).

- a failure to recognize that sensitivity and specificity almost always vary with patient characteristics [they might have been unifying attributes of tests had they been constant]

- the need to adjust sens. and spec. for workup bias whereas "forwards" disease probabilities require no such adjustment.


Mammography screening for breast cancer is a good case in point.  90% of positive mammograms are false positive, not just because mammography isn't perfect but because the disease and test results are dichotomized.  "False positives" could be avoided if we used probability to the fullest and didn't simply label the result as "positive".  The relevant Bi-rads category used in mammography reading is defined as the radiographer's impression is the risk of cancer is between 3% and 96%.  How strange!  Now wonder that assessment of the utility of mammography is so confusing.

We have made a lot of progress in thinking of prognosis in terms of multivariable risk models, probability of surviving xx years, and estimated life expectancy.  In my estimation it is time to put diagnosis in the context of forward (prospective) probabilities of disease taking into account pre-test variables and test outcomes, simultaneously.  On that note, I've attached example nomograms for differential diagnosis of bacterial vs. viral meningitis, and for diagnosis of coronary artery disease.