Discussion about this post

User's avatar
Huw Llewelyn MD FRCP's avatar

Thank you for summarising so clearly the basics of the current state of play on appraising diagnostic test studies. However, there have been concerns about these methods for some time (e.g. Limitations of Sensitivity, Specificity, Likelihood Ratio, and Bayes' Theorem in Assessing Diagnostic Probabilities: A Clinical Example, Moons et al (Epidemiology 1997; 8, 12-17). At best the above indices allow provisional comparison between different diagnostic tests (especially for epidemiologists e.g. PCR and LFD tests for Covid-19). However, they leave open some important questions for clinicians, e.g.:

1. How do you decide when a disease should be ASSUMED to be present (in the form of a diagnosis based on criteria) especially as the diagnostic (not disease) criteria will be on a spectrum of severity? Should we be basing this on probabilities of outcomes in RCTs conditional on estimated disease severity on recruitment?

2. How do you assess the ability of test results of varying degrees of severity to create helpful lists of differential diagnoses?

3. How do you interpret the non-dichotomised numerical values of test results required to differentiate between differential diagnoses? Ratios of pairs of sensitivities based on dichotomised results created by thresholds (as in Pauker & Kassirer, 1980) are used sometimes.

I explore some of these issues in the Oxford Handbook of Clinical Diagnosis (http://oxfordmedicine.com/view/10.1093/med/9780199679867.001.0001/med-9780199679867-chapter-13 ). I am currently completing the 4th edition.

Expand full comment
Michael Sikorav MD's avatar

There is an error in the NPV as its d/c+d isn't it ?

Expand full comment
1 more comment...

No posts