Linking Evidence-Based Medicine Therapeutic Summary Measures to Clinical Decision Analysis

Iztok Hozo, † Gary H. Lyman ‡

In This Article

Abstract & Introduction

Objective: Evidence-based medicine (EBM) seeks to improve clinical practice by evaluating the quality of clinical evidence and ensuring that only the "best" evidence from clinical research is used in the management of individual patients. EBM has contributed to our understanding of the meaning of the benefit and harm of treatment as reported in the literature, and it is often promoted as an aid to clinical decision making. However, EBM therapeutic summary measures reflect only a single dimension of clinical decision making. The purpose of this work is to show how EBM therapeutic summary measures can be effectively incorporated into medical decision making.
Design: The effective application of the therapeutic summary measures advocated by EBM requires their integration into the framework of clinical decision analysis. Clinical decision analysis involves not only the identification and specification of the probabilities of clinical events but also the assessment of their relative values or utilities. We present here several analytic models for the integration of EBM therapeutic summary measures within the framework of clinical decision analysis.
Main results: As expected, our analysis demonstrated that treatment should never be administered if its harm is greater than its efficacy, which is generally expressed as relative risk reduction. Likewise, a diagnostic test should never be ordered if the therapeutic harm is greater than the therapeutic efficacy. Intervention is always favored if the number needed to treat to avoid one adverse outcome (NNT) is smaller than the number needed to treat to harm one individual (NNH). When faced with a choice between two therapeutic options, the action threshold above which an intervention is favored can be expressed in terms of the harm inflicted (H) as H x NNT or NNT/NNH. If a patient's preferences are taken into account as relative value judgments (RV) of adverse events relative to that of therapeutic events, the action threshold is defined as NNT x (RV/NNH).
Conclusions: In the setting of clinical decision making, EBM summary measures derived from population studies can be effectively used to define diagnostic and therapeutic action thresholds that may help in the management of individual patients.

Key Words: evidence-based medicine, decision analysis

Evidence-based medicine (EBM) has emerged as a powerful problem-oriented approach to the practice of medicine that seeks to improve patient care by considering the quality of clinical evidence. [1] In the category of therapeutics, the main focus of EBM is to evaluate treatment effect usually expressed as one of several therapeutic summary measures. [2] EBM has been advanced as an important tool in clinical decision making that may aid physicians in selecting one treatment alternative over another. [3,4] Recommendations are often made concerning the preferred management strategy on the basis of a comparison of the benefits and harms associated with competing treatment alternatives. [5,6] However, it is unclear how this understanding of treatment benefit and harm should actually relate to a specific clinical decision. For example, should we choose treatment that is more efficacious or one that is less harmful? What is the minimal therapeutic benefit at which a treatment is still worth administering? What is the maximal acceptable harm at which a treatment is still considered worthwhile? Meaningful answers to these questions, as well as the application of EBM to everyday clinical practice, can be achieved by linking therapeutic summary measures to the methods of formal decision analysis. [7,8]


Comments on Medscape are moderated and should be professional in tone and on topic. You must declare any conflicts of interest related to your comments and responses. Please see our Commenting Guide for further information. We reserve the right to remove posts at our sole discretion.
Post as: