Re: Overcoming overuse: the way forward is not standing still—an essay by Steven Woloshin and Lisa M Schwartz
The views expressed here in the BMJ [1] and in the NEJM [2] do not offer a clear way forward. Experienced doctors have always been aware of the trade-off between over-diagnosis and under-diagnosis, over-treatment and under-treatment. The eternal problem is ‘too much uncertainty’ when trying to predict outcomes with and without interventions. It is clearly important to share decisions with patients, but that will not reduce the degree of uncertainty for predictions either. We need a more sophisticated understanding of the diagnostic and prognostic processes and better directed research.
EBM currently assesses tests and their results by their ability to predict ‘gold standard’ diagnostic test results. The latter are also used to recruit patients for RCTs. The problem is that these ‘gold standards’ are adopted without any real supporting evidence for their predictive value, so it is not surprising that there is over-diagnosis and related problems. Assuming constant relative risk reduction and extrapolating of risk over time additively also undermine the interpretation of RCTs as they only provide approximations for low probabilities found in epidemiology. Bayes simple rule with specificity and likelihood ratios is also an inadequate model of diagnostic reasoning [3,4,5].
There will have to be a more sophisticated approach to the problem and better directed research if there is to be progress.
Rapid Response:
Re: Overcoming overuse: the way forward is not standing still—an essay by Steven Woloshin and Lisa M Schwartz
The views expressed here in the BMJ [1] and in the NEJM [2] do not offer a clear way forward. Experienced doctors have always been aware of the trade-off between over-diagnosis and under-diagnosis, over-treatment and under-treatment. The eternal problem is ‘too much uncertainty’ when trying to predict outcomes with and without interventions. It is clearly important to share decisions with patients, but that will not reduce the degree of uncertainty for predictions either. We need a more sophisticated understanding of the diagnostic and prognostic processes and better directed research.
EBM currently assesses tests and their results by their ability to predict ‘gold standard’ diagnostic test results. The latter are also used to recruit patients for RCTs. The problem is that these ‘gold standards’ are adopted without any real supporting evidence for their predictive value, so it is not surprising that there is over-diagnosis and related problems. Assuming constant relative risk reduction and extrapolating of risk over time additively also undermine the interpretation of RCTs as they only provide approximations for low probabilities found in epidemiology. Bayes simple rule with specificity and likelihood ratios is also an inadequate model of diagnostic reasoning [3,4,5].
There will have to be a more sophisticated approach to the problem and better directed research if there is to be progress.
References
1. Woloshin S, Schwartz LM. Overcoming overuse: the way forward is not standing still. BMJ 2018;361:k2035
2. NEJM Rosenbaum L. The less-is-more crusade—are we overmedicalizing or oversimplifying? N Engl J Med2017;377:2392-7. . doi:10.1056/NEJMms1713248 pmid:29236644
3. Llewelyn H, Ang AH, Lewis K, Abdullah A. The Oxford handbook of clinical diagnosis;.3rd ed. Oxford University Press, 2014: 615-64. http://oxfordmedicine.com/view/10.1093/med/9780199679867.001.0001/med-97...
4. Llewelyn H. The way forward from “rubbish” to “real” EBM in the wake of Evidence Live 2015. http://blogs.bmj.com/bmj/2015/04/28/huw-llewelyn-the-way-forward-from-ru...
5. Llewelyn, H. Sensitivity and specificity are not appropriate for diagnostic reasoning. BMJ 2017; 358 doi: https://doi.org/10.1136/bmj.j4071. https://www.bmj.com/content/358/bmj.j4071
Competing interests: No competing interests