Article Text

PDF

Decision analysis
  1. Bob Phillips
  1. Evidence-based On Call, Centre for Evidence-based Medicine, University Dept of Psychiatry, Warneford Hospital, Headington OX3 7JX, UK; bob.phillipsdoctors.org.uk

    Statistics from Altmetric.com

    When we make a decision about a course of action—a diagnostic test, treatment, or other intervention—we weigh up more than just a single outcome. There could be beneficial outcomes, but the possibility of negative effects (adverse events, failures, repeat attendance, and so on) also needs to be considered. Diagnostic tests may give the wrong answer, and expose the patient to risks of non-treatment (or of inappropriate treatment). As clinicians, we instinctively assess the chances of the outcomes, weigh them, and conclude on a course of action.

    For example, in treating a sick child with pneumonia, one may use oral co-trimoxazole, an oral beta-lactam, intravenous penicillin, or intravenous cefuroxime. What is the best treatment to use? What does best mean: Most cures? Fewest side effects? Most cost effective? Most comfortable? There may be variations based on where you’re working—Australia or America, a rural clinic in the Kimberley or urban Adelaide hospital. Individual factors—allergy, HIV co-infection, likely support from parents—can all contribute to the decision.

    Decision analysis is a way of modelling all the factors and formally adding up the likely outcomes, and weighting these with values—be these costs, clinician, or patient centred measures of benefit (utilities). (See a previous Archimedes issue, “Economic analyses”1 for more on “utilities”.)

    For the clinician the full process can be difficult, time consuming, and monumentally boring. Where the practitioner can use such information is in taking analyses which have already been performed, appraising them, and using them in local practice.

    If no analysis exists, it may be worth doing a “back of the envelope” analysis. Knowing how good your current treatment is (taking costs and adverse effects into account) will let you know how effective a new treatment has to be to beat it. If you’re looking for a bedside diagnostic test, knowing in advance how many false negative and false positives you will accept informs you of the magnitude the likelihood ratios2 will need to reach. If the target you’ve set is unfeasible, then the five minutes spent thinking this through may have saved you hours of work.

    Now, it is stressed that decision analyses are models—not real. They provide approximations and guesses, and make transparent what occurs instinctively. In doing this, they make the decision process open to critical appraisal, rather than the decider open to criticism.

    References

    View Abstract

    Footnotes

    • Bob Phillips

    Request permissions

    If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.