Evidence-based medicine

Evidence based medicine is the practice of solving the clinical problems in one's practice by judicious and systematic use of the medical literature. This includes framing questions rightly and searching the right kind of literature. Thereafter, the available evidence needs to be evaluated for the validity, strength and effect size. Finally, the results are examined for applicability to the current problem which requires a detailed knowledge of the clinical setting, patient profile and the issues related to cost and harm. The present communication deals with these issues in a step-wise manner in order to stimulate readers to practise this important art.

vidence-based medicine (EBM) means using the are defined and then the literature is consulted to resolve medical literature as a tool to solve the individual these questions. Dissecting the question into its component patient's or society's diagnostic, prognostic or parts to facilitate finding the best evidence is a fundamental therapeutic problems. The fact that practice should be EBM skill. Most questions can be divided into three parts: based on evidence presented in the medical journals is 1. The population -the relevant patients/subjects. widely recognized but less commonly followed. Important 2. Intervention or exposure -the management strategies reasons for this include inaccessibility of good quality we are interested in comparing or the potentially medical literature, paucity of time to evaluate the sea of harmful exposure about which we are concerned. literature and lack of determination and competence in 3. Comparison group -the alternative option which could assessing its relevance or validity to a specific patient.
be no intervention or some other conventional management strategy. One goal of EBM is to be aware of the evidence on which 4. Outcome -the patient-relevant consequences or the one's practice is based, the soundness of the evidence and exposure in which we are interested. the strength of the inference the evidence permits. This strategy requires a clear delineation of the relevant Constructing a searchable question that allows you to use question(s); a thorough search of the literature relating to the medical literature to generate an answer requires an the questions; a critical appraisal of the evidence and its in-depth understanding of the clinical issues involved in applicability to the clinical situation; and a balanced patient management. It requires problem-solving and application of the conclusions to the clinical problem. The analytical skills besides basic clinical knowledge about the objective is to make efficient use of the published literature condition. The awareness of cost and psychosocial issues to help with patient care. The present communication is also important. attempts to present the basics of EBM in a simple, practical and step-wise manner and is not intended to serve as a A-45 year-old multiparous women who is nondiabetic and nonhypertensive presents with low backache for duration of three months. The backache intermittently becomes severe requiring frequent analgesics. The physical and neurological examination is entirely normal. X-ray of the spine is normal.

Initial question
Should we undertake magnetic resonance imaging (MRI) of the spine before deciding management for this woman?

Digging deeper
The key features of this patient are her middle age, moderately severe backache, normal physical and IJO -January -March 2007 / Volume 41 / Issue 1 neurological examination and a normal initial radiograph. Alternative investigational strategies address issues of the patient having any specific vertebral/spinal pathology and proceeding according to the result of this investigation.
What outcomes are we trying to influence in our choice of investigation? We would like her to be relieved of pain and disability. The reason we wish to detect vertebral/spinal lesion if they are present is that if a resectional surgery or any other specific management may benefit the patient. Thus, the primary outcome of interest is alteration of routine management (e.g., physiotherapy, analgesics etc.) Shah D, et al.: Evidence based medicine modify the prognosis (such as obesity or comorbidity) and follow them for the target outcome such as postsurgical complications or survival.
The information could be in the form of textbooks, individual study/studies, the systematic review of all the available studies, a synopsis of individual studies and systems of information. Most of these resources are now electronically available for quickly finding out the answers to the clinical questions. To find answers to general background medical questions (such as physiology, mechanism of action of an intervention or diagnostic approach to a clinical condition), referring to a textbook

Improved (searchable question)
that is well referenced and updated frequently is likely to A searchable question would specify the relevant patient be faster and more rewarding. For questions dealing with population, the management strategy or exposure and the specific patient's problem such as risk and benefits of a patient-relevant consequences of that exposure as follows: particular treatment (foreground questions), the most Patients: Middle-aged women with chronic low backache efficient approach is to begin with a prefiltered evidence-Intervention: MRI for management based resource [ Table 1] in which someone has already Outcome: Alteration of management or surgical need.
done the exercise of reviewing the literature and choosing Bearing the structure of the question in mind-patient, only the methodologically strongest studies. Looking for a intervention or exposure and outcome-is extremely helpful systematic review article on your topic addresses a targeted in arriving at an answerable question. Once the question clinical question using strategies that decrease the likelihood is posed, the next step in the process is translating the of bias. Framing the searchable question provides you with question into an effective search strategy.
the best choice of keywords for search. Carefully modifying your search strategy by limiting the search to desired STEP 2: SEARCHING THE LITERATURE sections such as diagnostics or therapeutics would help to deal with the problem of 'too much material'. If a search After the question dealing with a patient's management is of these resources does not provide a satisfactory answer clearly defined, you need to search the literature. There to a focused clinical question, it is the time to turn to are four fundamental types of clinical questions: those MEDLINE. It is an attractive database for finding involving therapy (effect of different treatments on patient information because of its comprehensiveness and free beneficial outcome), harm (potentially harmful effects on accessibility. However, a thorough knowledge of the patient function), diagnosis (strength of a test to differentiate structure of database indexing, use of medical subject between those with or without a target condition) and headings (MeSH) and combining various search results is prognosis. To answer questions about a therapeutic issue, essential for performing an effective search. Detailed we identify randomized controlled trials (RCTs) in which a information on searching MEDLINE and other reference process analogous to flipping a coin determines databases is available elsewhere. [1][2][3][4] participants' receipt of an experimental treatment or a control or standard treatment and subjects are followed

APPLICABILITY
should also be consulted to address the issues of harm. However, for many potentially harmful exposures, randomly allocating patients is neither practical nor ethical. For these exposures, the best way is to identify observational studies in which personal choice determines whether people are exposed or not exposed. For diagnostic purposes, studies evaluate the test by comparing its performance with a gold standard or reference standard in diagnosing or excluding a particular condition. For prognosis, investigators identify patients who belong to a particular group (such as patients undergoing surgery or patients with cancer) with or without factors that may  might be disastrous. As discussed earlier, RCTs are the most unbalanced between two groups, the outcome will be commonly used and important tools to address therapeutic biased, either overestimating or underestimating the issues. When evaluating medical literature for clinical treatment effect. Typically, observational studies are known application, the study should be assessed using three to overestimate (and occasionally underestimate) treatment discrete steps. 5 effect in comparison to randomized controlled trials. 9,10 Though attempts are made to match patients in (i) Are the results valid?
observational studies, the power of randomization in This issue deals with the credibility of the study and whether balancing the two groups is far greater with respect to both the results represent an unbiased estimate of the treatment the known and the unknown determinants of outcome. effect. This depends on whether the study was designed and conducted in a way that justifies claims about the (b) Was randomization concealed? If those determining therapeutic benefits or harms of a treatment regimen. The the patient allocation to a treatment or control group are validity of the study can be evaluated by finding out aware of the arm to which a patient will be allocated, they whether the persons exposed to a particular intervention may knowingly or unknowingly enroll sicker or less sick or control had similar prognosis at the beginning of the patients to either the treatment or control group. 11,12 This study and whether the groups were still similar with respect behavior defeats the purpose of randomization and is likely to prognostic factors throughout the study.
to yield a biased result. The effective methods of (a) Were subjects randomized? Comparison of outcomes randomization concealment are pharmacy coding of drugs among nonrandomized cohorts of patients who, for various and central concealment in which the recruiter makes a reasons, did or didn't undergo an intervention frequently call to a center to find out the arm of allocation before leads to false conclusions and randomized trials in the past assigning each subject to a particular intervention group. have generated surprises by contradicting the results of these less rigorous trials. [6][7][8] For example, randomized trials demonstrated that steroid injections do not ameliorate facet-joint back pain 7 and plasmapheresis does not benefit patients with polymyositis. 8 The studies in which patient or physician preference determines whether a patient receives treatment or control (or another treatment) often yield biased outcomes. This is because the outcome of interest (like morbidity or mortality) is influenced by many causes of which treatment is only one. The patient's age, severity of the disease condition, dietary factors, lifestyle, presence of co-morbid conditions and a number of other known and unknown factors influence the frequency with which a trial's target outcome occurs. If these factors prove (c) Were subjects analyzed in the groups to which they were randomized? Omitting results of subjects who did not take the assigned treatment or who took alternative treatment will bias the results and ruin the process of randomization. The reason for this is that such behaviors are often shown by patients who are different from the rest of the group in respect to the prognostic factors such as disease severity. 13,14 For example, in comparison of medical or surgical therapy for a particular condition, some patients randomized to surgery never undergo the operation because they are too sick or because they suffer the outcome of interest (e.g., stroke or myocardial infarction) before they are operated. If such patients who otherwise had poor prognosis (as they were sicker) are included in the control arm but excluded from the surgical arm, even a useless surgical therapy will appear to be effective! In reality this apparent benefit has not come from surgery but from the systematic exclusion from the surgical group of those with the poorest prognosis. This process of analyzing all patients in the same group to which they were originally randomized (and not based on the treatment they actually received) is called an intention-to-treat analysis. This process ensures that the known and unknown prognostic factors are equally distributed in two groups and the effect seen is actually the result of the treatment assigned.
Shah D, et al.: Evidence based medicine study results. Strict cutoffs like 10% or 20% are often misleading. In assessing whether the study results would have been the same if there was no loss to follow-up, we first assume worst case scenario. This means assuming that all patients allocated to the treatment arm and lost to followup suffered from undesired outcome (like death or continuing pain) and all patients allocated to the control group and lost to follow-up had the desired outcome (like survival or pain-free). If assuming this worst case scenario does not alter the magnitude of the treatment effect, the results are valid even with loss to follow-up. If however, this substantially alters the results you have to judge whether the patients lost to follow-up were similar with respect to important prognostic factors such as disease (d) Were patients in two intervention groups similar with severity. This lessens but does not eliminate the possibility respect to known prognostic factors? Although the process of different rate of target outcome in those who are lost. of randomization should take care of this factor, sometimes by chance more patients with a particular risk factor might (ii) What are the results? get allocated to one group. The smaller the trial, more is This involves the interpretation of the size and precision of the chance that prognostic imbalance might be created.
the study results; and whether and how the results could Therefore, we should check whether the study subjects in be applied to the desired clinical setting. 17 the two groups were similar in respect to at least the known prognostic factors before the commencement of (a) How large is the treatment effect? Authors often report intervention. If the differences are large, the validity of the the magnitude of benefit in several ways. If in a controlled results is compromised unless statistical measures have trial, 5% of those in the treatment group and 10% in control been employed to permit adjustment for the baseline group suffer from an adverse outcome (say death), the differences.
absolute risk reduction or risk difference between the proportion who died in the control group and the (e) Was blinding done? Patients who know that they are proportion who died in the treatment group is 0.10-0.05 taking a treatment they believe to be effective may feel, = 0.05 or 5%. Another commonly reported measure of perform or report better even if there is no biological action. treatment effect is relative risk reduction which measures The impact of this placebo-effect or reporting bias on study the percentage by which the treatment reduces the risk of results is best minimized by ensuring that patients are event relative to that occurring among control patients. In unaware of the nature of the treatment received and the this example, the treatment reduced relative risk of death control group patients benefit from these effects to the same by (0.1-0.05)/0.10, i.e., 0.5 or 50%. Whenever the authors extent as actively treated patients.
report that a certain treatment is x% more effective, we need to examine what the authors mean. Whether they Similarly, if the clinicians or/and outcome assessors are mean absolute risk difference or relative risk reduction has aware of the nature of the treatment received, they might a great bearing on interpretation as often relative risk prognostically imbalance the groups after randomization reduction results in subjective impression of a larger by conscious or unconscious differential administration of treatment effect. co-interventions and by differentially interpreting the outcomes, respectively. 15 Effective blinding eliminates these possibilities and improves the validity of the study results.
(f) Was follow-up complete? The subjects which are lost to follow-up often have a different prognosis than those who continue as the reason they are lost may be that they are suffering from adverse effects or on the other hand they might be doing well and so did not return to get assessed. The greater the number of patients getting lost to followup, the more is a study's validity potentially compromised. 16 How much follow-up loss is acceptable depends on the (b) How precise is the estimate of treatment effect? In the above example we have assumed an absolute value of the treatment effect calculated from the observations which is called a point estimate. However, this estimate is unlikely to be precisely correct as the true value lies somewhere in its neighborhood. This is calculated by confidence interval (CI) which is a range of values within which one can be confident that a parameter value lies. 18 A figure of 95% CI is arbitrarily used which means the range in which the true relative risk reduction value will lie 95% of the times. If the lower value of the 95% CI of the RRR does not have clinical IJO -January -March 2007 / Volume 41 / Issue 1 significance, then the trial has not really helped us to decide whether to offer the new treatment or not. In fact, the results could be consistent with an adverse effect of the treatment if the lower value of the 95% CI is in negative despite the point estimate being very high (50% in the above example).
In general, smaller the number of patients enrolled in a trial, the wider is the 95% CI and less precise are the results.
(iii) How can results be applied to patient-care?
The final task in EBM is the process of applying the results of the literature to the clinical practice. 19,20 This involves a Shah D, et al.: Evidence based medicine prevent one adverse outcome or produce one positive outcome. 21 To calculate this, one needs to know the risk of the adverse event if left untreated. Suppose the risk of death due to a disease condition is 1 in 500 (0.2% or 0.002). If the intervention reduces the risk of death by 50% (RRR of 0.5), the risk of death after treatment becomes 0.1% or 1 in 1000. The absolute risk reduction (ARR) due to this treatment strategy becomes 0.2-0.1% = 0.1% or 0.001. The inverse of this ARR is equal to the number of such patients we would have to treat to prevent one event-the NNT. In this example, we need to treat 1000 patients to detailed knowledge of the clinical setting, patient profile save a single life. If the drug otherwise is toxic, the number and the issues related to cost and harm.
harmed might become more than the lives saved. On the other hand, if the risk of death due to a particular condition (a) Are patients in my practice similar to those in the study?
is high, the ARR would become larger with same RRR and The profile of patients in our clinical settings might be NNT would be smaller making the drug more useful. different from those included in studies. These patients Knowing the NNT helps clinicians in the process of trading might be more or less sick or suffering from a co-morbid off the benefits and risks associated with the management condition like malnutrition or obesity. If our patients would options. have met the inclusion criteria of the study in question, the results of the study can be safely applied to them. If

BEYOND THERAPEUTICS
however, the patients differ in some aspect, we should examine carefully if there is a compelling reason why the In the foregoing, we have dealt mainly with therapeutic results should not be applied to them. If a compelling reason issues particularly in Step 3 and Step 4. Application of is not found, it is usually safe to generalize the results to EBM is equally important to other issues like those of harm your patients.
(adverse events of therapy), diagnosis 22 The process of evaluating evidence and clinical application (b) Were all clinically important outcomes considered? The for these is almost similar, though we need to examine reason we offer treatment to the patients is that it will different kind of studies, say observational studies for issues improve outcomes important for them like reducing related to harm and sensitivity/specificity studies for symptoms like pain or disability, reducing long-term performance of a diagnostic test. Similarly, systematic complications, avoiding hospitalization or preventing death. reviews need to be examined for their validity of results Many a times, the studies report benefit of treatment in and clinical applicability. 24 Thorough description of these terms of nonfunctional outcomes, for example, improved is beyond the scope of this article and readers are results of bone densitometry as a result of hormone encouraged to consult detailed texts and reviews on these replacement therapy in osteoporesis or improvement in topics. 22 (c) Are the likely benefits worth the potential harm and cost? It needs to be seen whether the potential treatment benefits are worth the efforts put by the health system and the patient. A 50% reduction in the relative risk of death may sound very impressive but may be of very minimal impact on the patient or your practice, particularly if the risk of death otherwise is very low. This is illustrated by a term 'Numbers needed to treat' (NNT) which denotes the number of patients who must receive an intervention to clinicians and academicians has led to negative as well as positive reactions. The need to develop new skills in searching and critical appraisal of the literature can be daunting. Second, busy clinicians have limited time to master and apply these new skills and the resources required for instant access to evidence are often woefully inadequate in clinical settings. Providing evidence-based care directed toward maximizing patients' quality of life often increases the costs of their care. Also, evidence that EBM "works" has been late and slow to come.
However, the earlier sections have clarified some "pseudolimitations" that arise from misunderstandings of the definition of EBM. An examination of the definition and steps of EBM quickly dismisses the criticisms that it denigrates clinical expertise, is limited to clinical research, ignores patients' values and preferences or promotes a cookbook approach to medicine. In fact, it involves all these issues in a more systematic and extensive manner.

Obstacles to EBM
It is feared that nonimplementation or partial implementation is likely to be the fate of many interventions based on EBM. 31 There are several behavioral obstacles to implementation of evidence in practice. 32 Firstly, much of the science is seen in practice as inconclusive or as