I, too, support the underlying principles of EBM, recognize limitations, but wonder whether a way out of this impasse involves taking a Bayesean approach for incorporating EBM's various forms of evidence into practice decisions. Many years ago, authors like Pauker and Schwartz addressed the problem of balancing risk vs. benefit both quantitatively and qualitatively (see below), which brings to the fore issues of perception (individual perception of utility). Several of the papers described below provide examples. The last paper listed identifies a few more key issues (weak numeracy skills coupled with inadequate information presentation, making counterintuitive solutions all the more difficult to follow).
PARADIGM Database Search, 2006/12/01:
-----------------------------------------------
Pauker SG, Kassirer JP
The Threshold Approach to Clinical Decision Making
N ENGL J MED 1980;302(20):1109-1117
Bayesean approach weighted for information value & risk presented
to evaluate point at which expected value of witholding
treatment=expected value of treatment (test threshold) & where
expected value of testing=expected value of treating
(test-treatment threshold). If disease probability < test
threshold, risk of test exceeds benefit. If disease
probability>test-treatment threshold then therapy is indicated
but test to confirm dx isn't. Intermediate probability warrants
test before treatment. Examples point out that lack of
information value alone may negate value of test, as opposed to
risk itself.
Schwartz WB, Gorry GA, Kassirer JP, et al.
Decision Analysis and Clinical Judgement
AM J MED 1973;55:459-472
Describes application of Bayesean probabilities and utility
functions to clinical decisions, including decision trees for
typical clinical problems (whether or not to test, treat, use one
therapy or an alternative), sensitivity, specificity, and one
approach to generating a utility function. 14 references.
Kassirer JP, Moskowitz AJ, Lau J, et al.
Decision Analysis: A Progress Report
ANN INTERN MED 1987;106:275-91
Literature review (261 references) describing published models
for wide variety of clinical & public health problems. Problems
& proposed models are indexed in tables. General discussion of
advantages & disadvantages provided.
Goodman SN
Toward Evidence-Based Medical Statistics. 1: The P Value Fallacy,
2: The Bayes Factor
ANN INTERN MED 1999;130(12):995-1004, 1005-1013
Review explores application of statistical methods for deductive
and inductive reasoning, briefly tracing history of
Neyman-Pearson and Fisher frequentist approaches, confidence
intervals, and Bayesian analysis. Strengths and weakness of each
approach noted, recognizes confidence interval as a partial
solution, and advocates consideration of the Bayes factor
(likelihood ratio) as best strength of evidence component while
Bayes' equation provides a calculus for impact of evidence on
belief. Formula and table are provided to determine minimum
Bayes factor commensurate with Z-scores. SEE ALSO Richard
Royall's book (Statistical Evidence: A likelihood paradigm.
Monographs on Statistics and Applied Probability, #71, London:
Chapman & Hall; 1997) for a more mathematical statistics
treatment of the same topic (taking the position that frequentist
approaches are important in planning experiments but likelihood
ratios are more appropriate to interpret results, and not
equivalent to Bayesian methods).
Burkell J, Campbell DG
"What does this mean?" How Web-based consumer health information
fails to support information seeking in the pursuit of informed
consent for screening test decisions
J MED LIBR ASSOC 2005;93(3):363-73
Review (47 references) notes that almost half of North Americans
lack minimum quantitative literacy skills, and that problems in
interpreting predictive values have been recognized by cognitive
psychology & health sciences. Study of WWW resources for
screening test decisions about Down syndrome, prostate cancer &
breast cancer vs. criteria from General Medical Council standards
on elements of informed consent finds that resources located
through MedlinePlus "appeared to be substantially reliable,
correct, and created with the best of intentions" but incomplete
(less than half give sensitivity, 1 of 18 gives specificity, none
give prevalence to compute predictive value). Presents
recommendations for metadata systems.
----- Copyright Applied Epidemiology -----
--
David Birnbaum, PhD, MPH
Adjunct Professor
School of Nursing
University of British Columbia
Principal, Applied Epidemiology
British Columbia, Canada
|