Appendix A. Interpreting Evidence-Based Medicine

“Evidence-based medicine—is this something new?” asked my father, incredulously. “What were you practicing before?”

Like my father, our patients assume that we provide recommendations to them based on scientific evidence. The idea that there might not be relevant evidence or that we might not have access to that evidence has not even occurred to most of them. This is certainly not to imply that such evidence is the be-all and end-all of medical practice or that our patients would follow such recommendations blindly—rather, for me, it is a starting point from which to begin rational testing or outline a possible therapeutic plan.

The first time that I recall the term evidence-based medicine (EBM) being discussed was in the early 1990s.1,2 It seemed that we would need to develop skills in evaluating the published literature and determining its quality, validity, and relevance to the care of our patients. As a teacher and researcher, I was intrigued by the challenges of critically appraising articles and teaching this newfound skill to others. As a clinician, however, I was most interested in answering clinical questions and doing so in a compressed time frame. I need rapid access to tools or sources that provided summary answers to those questions tagged to information about the quantity and quality of the evidence and the consistency of information across studies.

There seemed to be many systems for rating literature but few that met the needs of the busy practitioner trying to make sense of individual clinical trials and the hundreds of both evidence-based and consensus-based guidelines that seemed to spring up overnight. In 2004, the editors of the U.S. family medicine and primary care journals and the Family Practice Inquiries Network published a paper on a unified taxonomy called Strength of Recommendation (SOR) Taxonomy that seemed to fit the bill (Figure A-1).3 This taxonomy made use of existing systems for judging study quality while incorporating the concept of patient-oriented (e.g., mortality, morbidity, symptom improvement) rather than disease-oriented (e.g., change in blood pressure, blood chemistry) outcomes as most relevant. A SOR A recommendation is one based on consistent, good-quality patient-oriented evidence; SOR B is a recommendation based on inconsistent or limited-quality patient-oriented evidence; and SOR C is a recommendation based on consensus, usual practice, opinion, disease- oriented evidence, or case series (Figure A-1 and A-2).

In this book, we made a commitment to search for patient-oriented evidence to support the information that we provide in each of the chapter sections (i.e., epidemiology, etiology and pathophysiology, risk factors, diagnosis, differential diagnosis, management, prevention, prognosis, and follow-up) and to provide a SOR rating for that evidence whenever possible. The bulleted format within these divisions would allow the practitioner to quickly find answers to their clinical questions while providing some direction about how confident we were that a recommendation had high-quality patient-oriented evidence to support it.

For example, a practitioner caring for a patient with severe chronic obstructive pulmonary disease (COPD) and frequent exacerbations who is already taking a combination long-acting β-agonist with an inhaled glucocorticoid asks, “What other options are available to reduce exacerbations?” This information can be found in the Management section of Chapter 56: Chronic Obstructive Pulmonary Disease, under Medications. Although the Global Initiative for Chronic Obstructive Lung Disease does not recommend mucolytics for routine use, authors of a Cochrane review, based on strong evidence SOR A, concluded that these medications produce a small decrease in the frequency of exacerbations (0.5 fewer exacerbations/year) and in disability days.4 Tiotropium (a long-acting anticholinergic) also reduces exacerbations and improves symptoms, but a recent metaanalysis concluded that mortality was increased with use of this medication SOR A.5 Theophylline also reduces exacerbations SOR A but is associated with nausea. The physician armed with this information can discuss the options with this patient and explore potential benefits and risks. Particularly in difficult cases where there are multiple options, the clinician's experience and the patient preferences are important aspects of shared decision making. One definition of EBM is “The integration of best research evidence with clinical expertise and patient values.”6

Several other concepts are used throughout the book that can assist practitioners in using evidence-based information and explaining that information to patients. Risk reductions from medical treatments are often presented in relative terms—the relative risk reduction (RRR) or the difference in the percentage of adverse outcomes between the intervention group and the control group divided by the percentage of adverse outcomes in the control group. These numbers are often large and use of them not only causes us to overestimate the importance of a treatment but misses its clinical relevance. A more meaningful term is the absolute risk reduction (ARR)—the risk difference between the two groups. This number can then be used to obtain a number needed to treat (NNT)—the number of patients that would need to be treated (over the same time as used in the treatment trial) to prevent 1 bad outcome or produce 1 good outcome. This is calculated as 100% divided by the ARR. NNT is more easily understood by us and our patients. See the NNT example in Box A-1.

In the above case, for example, the patient might ask how risky it could be to add tiotropium. As written in the chapter, the difference in death with use of this medication was 0.8% (2.4% vs. 1.6% on placebo) making the number needed to harm (NNH) over 1 year to be equal to 124 (for every 124 patients treated, 1 additional death might occur in 1 year).

Another term that is used in this book is the likelihood ratio (LR). This number, based on the sensitivity and specificity of a diagnostic test, is used to determine the probability of a patient with a positive test (LR+) having the disease or the probability of the patient with a negative test (LR−) not having the disease in question. The LR is defined as the likelihood that a given test result would be expected in a patient with the target disorder compared to the likelihood that that same result would be expected in a patient without the target disorder.6 The number obtained for the LR+ [Sensitivity/(100 − Specificity)] or the LR− [(100 − Sensitivity)/Specificity] can be multiplied by the pretest probability of disease to determine the posttest probability of disease. A nomogram (one can be found by visiting the website mentioned in reference 6) can be used to more easily work with these numbers to convert a pretest probability into a posttest probability. A LR+ over 10 is considered strong evidence to rule in disease while a LR− of less than 0.1 is strong evidence to rule out disease.

We both are privileged and cursed with practicing medicine in an information-rich environment. We have designed our Color Atlas to link evidence to clinical recommendations so that we can provide our patients the best science available. When the evidence is lacking, we make that clear and encourage you to engage in frank and honest discussions that lead to the shared responsibility for decisions. Our patients are justified in expecting science along with humanism—can we give them anything less?