Centre for Health Evidence: Home » Users' Guides to EBP
|
Centre for Health Evidence: Home » Users' Guides to EBP |
Antonio L. Dans, Leonila F. Dans, GH Guyatt, S. Richardson, for the Evidence-Based Medicine Working Group
Based on the Users Guides to Evidence-based Medicine and reproduced with permission from JAMA. (1998;279(7):545-9). Copyright 1998, American Medical Association.
You are the attending physician on duty when a poor 45 year old man presents to the emergency room of a general hospital in the Philippines. He has experienced severe chest pain for two hours, associated with clammy perspiration. Physical examination reveals a blood pressure of 110/70 mmHg, a heart rate of 92, a normal 1st heart sound and clear lungs. An electrocardiogram discloses 3mm ST segment elevation in the inferior leads. As intravenous lines are placed, and the patient is prepared for admission to the coronary care unit, you consider whether you should offer this patient a thrombolytic agent. Though your response is that the impecunious patient cannot afford the treatment, you ponder the right course of action in a richer patient. As your duty ends that night, you resolve to prepare for the next patient admitted for an acute myocardial infarction (MI), by retrieving the best evidence on the use of thrombolytics.
Streptokinase is the only thrombolytic agent that your patients might afford. You therefore confine your search to this drug, trying to locate the best trial or, if possible, a meta-analysis. Using Grateful Med software, you select Myocardial Infarction (MI) from the list of Medical Subject Headings (MESH) used to index articles. On the second subject line, you use the term streptokinase (MESH). You limit your search to English-language articles, and to find quantitative reviews or original studies, you use the terms "meta-analysis" or "randomized controlled trial" as the publication type.
You retrieve a systematic meta-analysis of randomized trials that you find deals only with effectiveness [1] and not toxicity. You therefore also review a single trial, ISIS-2 [2], that you choose on the basis of its size (17,000 patients), strong design (including double-blinding), and the wide variety of settings in which the study was undertaken. You refer to earlier User's Guides to evaluate the validity of the studies [3] [4], as well as the magnitude and precision of the treatment effects and toxicity [5]. The articles pass the validity criteria and the magnitude of the treatment reduced the even rate from 17.4% to 12.8% [1]. This outweighs the potential harm of "bleeds requiring transfusion", which occurred in 0.5% of streptokinase-treated patients, compared to 0.2% in the placebo group in ISIS-2.
An answer to your last question, how you can apply the results to your patients, does not come easily. Asians comprised a small minority of the patients in the trials, and you are uncertain about your hospital staff's ability to cope with technical requirements for administering the drug or dealing with complications.
As clinicians look more often to randomized controlled trials (RCTs) to guide their clinical care, they must decide how to apply RCT results to individual patients in their practice setting. This Users Guide addresses this issue of applicability, which involves the implications of the trial results for patient care. Applicability is closely related to concepts of generalizability and external validity, but is broader in its scope, including issues related to the overall impact of treatment in individual patients. In considering applicability, clinicians must decide whether the biology of the treatment effect will be similar in patients they are facing, their patients' risk of the target event that treatment is designed to prevent and of the side effects that may accompany treatment, and their own ability to deliver the intervention in a safe and effective manner [6].
Clinicians managing patients who differ economically, racially, culturally from those recruited in typical clinical trials face particular challenges in addressing applicability. Such patients include those from the inner cities of North America, the Native American reservations, or from less industrialized countries. Clinicians seeing these patients cannot afford to repeat every trial simply because of doubts regarding applicability. The end result is that applicability becomes a "fait accompli" -- an issue that may often be ignored rather than confronted.
Earlier in this series, we addressed the applicability problem in the User's Guide for articles about therapy or prevention [3]. "A better approach than rigidly applying the study's inclusion and exclusion criteria is to ask whether there are compelling reasons why the results should not be applied to the patient. A compelling reason usually won't be found, and most often you can generalize the results to your patient with confidence".
Physicians may encounter problems in following this advice. We didn't give a good definition of a "compelling reason", or provide guidelines on how to systematically address the question. In this paper, we correct these deficiencies by presenting a set of guidelines for evaluating the applicability of the results of RCTs to populations other than the participants. We present the guides as questions that probe for situations when clinicians may be forced to reject applicability. We phrase the questions so that a "Yes" answer will lead clinicians to suspect a problem of applicability. Table 1 summarizes the guides, categorizing them into biologic issues (that help us decide if the treatment can work), socio-economic issues (that help us decide if the treatment will work), and epidemiologic issues (that help us decide how efficient the treatment will be). As we discuss each issue, we will offer sources of information that will help physicians answer their questions.
Diseases with a single name may represent conditions with important pathophysiologic differences. These differences can sometimes lead to diminished treatment responses due to 1) divergence in pathogenetic mechanisms or 2) biologic differences in the causative agent. Hypertension in blacks, which has been observed to be relatively responsive to diuretics, and unresponsive to beta-blockers [7], provides an example of the former. This selective response reflects a state of relative volume excess that investigators now theorize may have served protective functions in their hot and arid ancestral environments [8].
Malaria provides an example of a condition that may vary because of biologic differences in the causative agent. Malaria treatment protocols vary depending on drug resistance patterns [9]. In these examples, clinicians should anticipate variation in response to treatment, and should temper hasty conclusions regarding the applicability of trial results.
To address our scenario of applicability of streptokinase to the treatment of MI in the Philippines, we reviewed case series of autopsies done on Filipino MI patients [11]. Pathologic changes in the coronary arteries and myocardium were similar to those noted among North Americans [12], while non-atherosclerotic causes of coronary disease were very rare. Clinical surveys have demonstrated that Filipinos share the same risk factors for coronary disease [13] as North Americans [14]. Thus, we can be confident that disease pathogenesis is similar.
Between-population differences in response to treatment may arise from a) differences in drug metabolism, b) differences in immune response, or c) differences in environmental factors that affect drug toxicity. Differences in drug metabolism may directly influence the efficacy of a treatment regimen. If they are not identified, slow metabolizers of a drug could face the risk of greater toxicity, while a significant decrease in efficacy might occur in rapid metabolizers. Such differences are usually based on genetic polymorphism in the activity of metabolizing enzymes. A well-known example is hepatic N-acetyl transferase, an enzyme with increased activity among Asians [15]. For this reason, clinicians offer higher drug dosages for agents such as isoniazid, hydralazine, and procainamide. Other examples of genetic polymorphism include pseudocholinesterase activity in the metabolism of suxamethonium, and glucose-6-phosphate dehydrogenase activity in the metabolism of sulfonamides and other drugs [16].
Differences in patients' immune response may also modulate treatment effect. Hemophilus influenza vaccine, for example, has a lower efficacy in Alaskan natives than in non-native populations [17]. Finally, environmental factors may affect response to therapy. For instance, the incidence of thyroid dysfunction from amiodarone differs in low versus high iodine environments [18].
The patient's perspective may also merit specific consideration, if costs (e.g. in travel) reduce access to care. Also, some patients may not be able to participate in community care programs if these impose major costs in terms of informal nursing support in the home. In some countries, most notably the USA, patients may also be responsible for a sizeable proportion of their health care bills. Many economic analysts do not track all of these costs, owing to the time and effort required. However, the patient's perspective is partially integrated into the analysis by measuring the outcomes of therapy, such as impact on quality of life.
Although we found no studies evaluating the pharmacokinetic profile of streptokinase when given to Filipinos, post-marketing studies show that Filipinos experience the same reperfusion arrhythmias and the bleeding complications when given streptokinase at the same dose as North Americans [19]. These studies provide some assurance of similarities in the response to adverse effects of treatment.
To the extent that groups of people exhibit different compliance with treatment, clinicians may expect variation in treatment effectiveness. Variability in compliance between populations may stem from a) resource limitations in a particular setting, or b) less obvious attitudinal or behavioral idiosyncrasies. Both types of problems may, for example, affect the safety of outpatient administration of anticoagulants. Neither indigent patients nor their society may be able to afford repeated clinic visits and tests for treatment monitoring. Alcoholics, whatever their financial situation, may be less likely to comply with monitoring. Inadequate monitoring, whatever the reason, increases bleeding risk from over-anticoagulation, shifting the balance between benefit and harm (even to the point where harm outweighs benefit).
In this guide, provider compliance refers to a host of diagnostic tests, monitoring equipment, interventional requirements and other technical specifications that clinicians must satisfy to safely and effectively administer a treatment. Financial conditions in a health care center, access to equipment, technologic expertise, and availability and skill of health personnel may influence treatment effectiveness. For instance, while carotid endarterectomy may benefit low-risk patients when surgery-associated stroke is low, the net effect for such patients in centres with higher surgery-associated stroke rates may be an increase in adverse outcomes [22].
In less-industrialized countries many hospitals and clinics do not have easy access to sophisticated equipment and problems of provider compliance are therefore common. For example, while rheumatic atrial fibrillation remains a common problem in Asian countries, very few laboratories in rural areas perform the tests necessary for titration of warfarin dose. This limitation is likely to reverse the critical balance between effectiveness and safety of treatment.
Because of experience regarding availability of equipment, laboratory tests and health personnel resources, practitioners themselves are a good source of information regarding feasibility of interventions. Clinicians' assessments can be supplemented by formal quality of care assessments, and post-marketing surveillance of adverse effects. Whatever the source of information, a thorough understanding of the technical requirements for safe and effective administration should guide decisions regarding the ability to comply.
Administration of streptokinase carries potential hazards, foremost of which is catastrophic bleeding. Facilities for emergency administration of cryoprecipitate, fresh frozen plasma or whole blood must be available [23]. In hospitals without efficient blood banking systems, it may be difficult to cope with bleeding emergencies. This increases the potential hazards of treatment and may tip the balance between benefit and harm.
When satisfied that biologic, social or economic differences do not compromise applicability, the clinician must examine patient characteristics that can influence either the magnitude of the benefit or the risks of treatment (and thus, the trade-off between the two) [24]. The last two guides address these issues.
The presence of other conditions in a particular locality may affect treatment efficiency in two possible ways - as a) competing diagnostic possibilities, or b) competing etiologies of outcome. The management of pneumonia in developing countries provides an example of a competing diagnostic possibility.
The acute respiratory infection (ARI) management protocol includes a symptom-driven algorithm for differentiating pneumonia from non-pneumonia. This protocol identifies children who need antibiotics and has proven effective in reducing mortality from pneumonia among children under five years of age [25]. Unfortunately, similarities exist in the clinical presentation of pneumonia and malaria. In malaria endemic areas, clinicians may expect an increase in false positive "pneumonias". These false positive patients will not respond to antibiotics for pneumonia, and a delay in instituting antimalarial treatment may result. If the drop in accuracy is large enough, the balance between harm and benefit will change. To resolve this issue, investigators have initiated a study to determine if the ARI protocol can maintain its effectiveness in malaria endemic areas [26].
Competing causes of target events may also affect the magnitude of benefit. An example comes from the management of acute myocardial infarction in some Filipino hospitals. A recent study disclosed 30 in-hospital deaths in a cohort of 149 patients admitted to a charity hospital [27]. On the basis of results from the meta-analysis, clinicians might expect streptokinase to reduce this 20% death rate by 25% [1]. However, a closer look at the local data shows a contrast with the original studies in which virtually all deaths were a direct result of cardiac ischemia. In the Philippine study, non-cardiac causes (mostly pneumonia with sepsis) were responsible for 11 of the 30 deaths. Streptokinase will not reduce mortality in such patients. Adequate antibiotic coverage may result in a greater (and more economical) reduction in mortality for patients who develop pneumonia.
In addition to reducing benefit, other morbidity may affect the magnitude of risk. Surgical mortality may increase in malnourished patients, shifting the balance between benefit and risk. On occasion, other morbidity can also work in the opposite direction -- increasing efficiency. For example, a patient with a large infarct in whom the clinician is considering warfarin may also have atrial fibrillation. Since anticoagulation reduces stroke risk in such patients, the presence of atrial fibrillation strengthens the indication for treatment.
Cohort studies provide the most reliable information on co-morbid conditions. In the MI scenario, we used data from the local study of 149 charity patients to evaluate the impact of other morbid conditions [24]. As we noted, we can expect streptokinase to prevent around 5 of the 19 cardiac deaths (but none of those from other causes), and the absolute reduction in all-cause mortality is a decline from 30/149 (20.1%) to 25/149 (16.8%).
In our User's Guide on therapy, we addressed the relation between a patient's risk of an adverse event and the magnitude of the treatment impact. Because the issue is so important in assessing applicability of trial results, we will review it in detail.
In the therapy User's Guides, we introduced the notion of number needed to treat (NNT). Thinking about NNT requires an understanding of the concepts of relative risk (RR), relative risk reduction (RRR) and absolute risk reduction (ARR). Readers desiring a full discussion of these concepts can refer back to the earlier paper [3]. Because it estimates the number of patients who need to receive treatment (with implications about the associated toxicity and cost) in order to prevent one adverse event, clinicians can use the NNT to consider a treatment's efficiency.
The NNT is the inverse of the ARR resulting from a particular treatment in a particular group of patients. If a patient's risk without treatment is 20%, then we expect 20 in a hundred untreated patients will suffer an adverse event. When we administer a treatment with a RRR of 10%, only 18 treated patients will experience adverse events. Thus, for every 100 patients treated, we prevent 2 events, and the NNT is 100/2 or 50. If the expected event rate in untreated patients is cut by half to 10%, and the RRR remains the same, in treating 100 patients we will prevent only 1 adverse event, and the NNT will double to 100/1 or 100.
This reasoning, and much of what follows, assumes that RRR remains constant across subgroups. While testing this assumption can be difficult [28], there are situations in which the assumption will fail, and clinicians should be alert to this possibility [29] [30]. Fortunately, however, in most instances the assumption will not introduce important inaccuracies in the NNT [31] [32].
One source of differences in expected event rates is country of origin and residence. Keys and colleagues compared the 20-year incidence of coronary deaths in the United States, five European countries and Japan [33]. They found an extremely low incidence of coronary death in the Japanese cohort, despite correction for baseline differences in recognized risk factors. Similar results have been observed in preliminary reports of the ongoing Multinational Monitoring of Cardiovascular Disease and their Determinants project [34]. In this study, involving 39 centres from 26 countries, east Asians showed a much lower incidence of coronary death than their western counterparts. Age standardized mortality rates for coronary heart disease were lowest in Japanese (40/100,000), and highest in North Ireland 414/100,000).
Thinking of the NNT, this ten-fold difference in incidence among the Japanese would translate to a ten-fold increase in the NNT for a drug preventing coronary deaths. This decrease in efficiency may warrant a reconsideration of applying the results of a trial to low risk patients. We consider the issue of balancing costs and effects in our user's guides for determining a level of recommendation [35].
Cohort studies on the course of disease in untreated patients can provide excellent risk data, and such studies are even more useful when they define subsets of patients at varying risk. Of 424 Filipinos with MI eligible for streptokinase (but in whom the drug was not administered) who participated in a cohort study conducted in 9 centers in Metro Manila, 37 (11.1%) suffered cardiac deaths [24]. This provides a good estimate of the expected event rate. If streptokinase had been given, it would have prevented 25% of the deaths, reducing the absolute mortality rate to 8.3%. Thus, 2.8% of the those otherwise destined to die would have been spared (the ARR), and the NNT is 100/2.8 or approximately 36 patients.
The expected event rates varied in patient subpopulations [24]. Young patients with small infarcts had a much lower expected mortality (and thus much larger NNTs) than old patients with large infarcts. Using prognostic information from these various subgroups, we constructed Table 2 which shows the expected mortality according to age and left ventricular wall involvement, and the corresponding NNT to save one life in each group. As the table shows, NNT can range from 16 (when treatment is applied to patients with a poor prognosis) to as much as 179 (when treatment is applied to patients with a good prognosis).
Varying patient risk will impact on benefit no matter what the environment in which you practice. Even if you work in the Western tertiary care environment in which investigators conducted their original studies, you will still face high and low risk patients. The critical trade-off between risk and benefit may well vary in these patients groups, mandating different treatment decisions [32].
These guides address the task of applying the results of clinical trials done on restricted, specially selected populations to other groups. Although inspired by the predicament in less-industrialized countries, the guides are relevant to all situations where clinicians must make decisions regarding applicability. By breaking down the problem into specific questions, we have provided guides for busy clinicians who make daily attempts to strike a balance between making "unjustifiably broad generalizations and being too conservative in one's conclusions" [4].
When clinicians suspect limited applicability (ie, when a response of "yes" is encountered for any of the questions), what can they do? This will depend on whether or not the anticipated differences are important, and if important, whether they are remediable. For example, differences in disease pathophysiology (guide number 1) do not always mean that applicability is limited. Management of a cataract, for instance, will probably be the same regardless of the cause. Differences in treatment response (guide number 2) can sometimes be accommodated by altering administration of a treatment (such as adjusting the dose of a drug). Education, training, provision of necessary equipment, and other attempts at optimizing compliance may address problems in patient and provider compliance (guide numbers 3 and 4).
For differences in co-morbid conditions or expected target event rates (guide numbers 5 and 6) the clinician's response will depend on the difference observed. If an increase in efficiency is anticipated (as when disease prognosis is worse or the incidence of an adverse outcome is greater), a recommendation to treat can be more easily accepted. A decrease in efficiency, on the other hand, should lead clinicians to be more cautious in accepting a treatment recommendation.
When the answer to one or more of the Guide questions is Yes, and the differences noted are important and not easily remediable, clinicians should not assume that the trial results can be readily applied. In these instances, an additional RCT may be warranted.
What should we recommend regarding thrombolytic use for the Filipino patient admitted for acute MI? There is no reason to believe that Filipinos have a different disease pathogenesis or a different response to treatment with thrombolytics (Guides 1 and 2). Patient compliance will not be an important issue since the drug is given intravenously as a single dose (Guide 3). The technical requirements for administration are often, but not always, available, and when they are not the risks may outweigh the benefits of thrombolytic administration (Guide 4).
Two issues remain to be resolved, both dealing with the magnitude of treatment impact. Pneumonia is an important co-morbid condition, accounting for one third of deaths, at least in some charity hospitals (Guide 5). However, rates of cardiac death are still sufficiently high (11.1%) that the relative risk reduction we can achieve with streptokinase (28%) will result in an NNT of 32 for the overall population (Guide 6). For sub-groups of patients, however, the NNT will range from 16 to 179, depending on the age and the size of the infarct (Table 2). Should we recommend the routine use of streptokinase among Filipinos presenting with acute MI? The guides have brought us closer to an answer. We have confirmed applicability of the thrombolytic data on the effectiveness of streptokinase, but only in centers with adequate blood-banking facilities. We have also refined estimates of treatment impact, based on knowledge of the course of disease among Filipinos. However, the cost of the drug is approximately US $250 per treatment, in a country in which the average ANNUAL per capita income is $720 [36].
The judgement about whether to give streptokinase will depend on who pays for the treatment (in the Philippines, usually the patients themselves), patient and family values, what resources are available (usually very limited in our charity hospital setting), and the competing needs (for example, the need for antibiotics because of a high incidence of pneumonia, in turn a result of overcrowding in the wards). For equally applicable treatments, our final decision may differ for a much less costly, but equally effective and applicable treatment, such as aspirin for our MI patient.
Table 1: Biologic Issues
|
Table 2: Baseline mortality rate (MR) without treatment and estimated number needed to treat (NNT) to save one life using streptokinase in Filipinos with acute MI, tabulated according to age and wall involvement.
|
||||||||||||||||||||||||||||||
4. Guyatt GH, Sackett DL, Cook DJ, for the Evidence-Based Medicine Working Group. Users' guides to the medical literature, II. how to use an article about therapy or prevention, A. are the results of the study valid? JAMA 1993; 270: 2598-2601.
5. Guyatt GH, Sackett DL, Cook DJ, for the Evidence-Based Medicine Working Group. Users' guides to the medical literature, II. how to use an article about therapy or prevention, B. what were the results and will they help me in caring for my patients? JAMA 1994; 271: 59-63.
6. Friedman LM, Furberg CD, DeMets DL. Fundamentals of Clinical Trials, 2nd ed. Little, Mass: PSG Publishing Co Inc. 1985.
7. Falkner B, Kushner H. Effect of chronic sodium loading on cardiovascular response in young blacks and whites. Hypertension 1990; 15: 36.
8. Wilson TW. History of salt supplies in West Africa and blood pressure today. Lancet 1986; 1: 784.
9. World Health Organization. World malaria situation in 1992, part 1. Wkly Epidemiol Rec 1994; 69: 309-314.
10. Davis CE. Generalizing from clinical trials. Control Clin Trials 1994; 15: 11-14.
11. Canlas MM, Dominguez AE, Abarquez RF. Ten year review of the clinicopathologic findings of coronary artery disease at the University of the Philipines, Pilipine General Hospital (1969-1978). Phil J Int Med 1980; 18: 65-74.
12. Roberts WC, Potkin BN, Solus DE et al. Mode of death, frequency of healed and acute myocardial infarction, number of major epicardial coronary arteries severely narrowed by atherosclerotic plague, and heart weight in fatal atherosclerotic coronary artery disease: analysis of 889 patients studied at necropsy. J Am Coll Cardiol 1990; 15: 196-202.
13. Balgos AA, Lopez MB, delos Santos E, et al. The significance of risk factors in myocardial infarction - a 2 year retrospective study at the University of the Philipines, Philipines General Hospital. Phil J Cardiol 1984; 12: 104-108.
14. Farmer JA, Gotto AM. Dyslipidemia and other risk factors for coronary artery disease. In: Braunwald E, ed. Heart Disease - a Textbook of Cardiovascular Medicine. 5th ed. Philadelphia, Pa: WB Saunders Co; 1997: 1126-1160.
15. Horai Y, Ishizaki t. Pharmacogenetics and its clinical implication: N-acetylation polymorphism. Ration Drug Ther 1987; 21: 1-7.
16. Goodman GA, Rall TW, Nies AS, Taylor P. Principles of therapeutics. 8th ed. New York, NY: Pergamon Press Inc; 1991; 71-73.
17. Ward J, Brenneman G, Letson GW, Heyward WL. Limited efficacy of a Haemophilus type b conjugate vaccine in Alaska Native infants: the Alaska H influenzae Vaccine Study Group. N Engl J Med 1990; 323: 1415-1416.
18. Martino E, Safran M, Aghini-Lombardi F, et al. Environmental iodine intake and thyroid dysfunction during chronic amiodarone therapy. Ann Intern Med 1984; 101: 28-34.
19. Dela Paz AG, Pineda NE, Justiniani RP et al. Thrombolysis in acute myocardial infarction. Philipine J Cardiol 1988; 17: 185-188.
20. Bulatao J. split-Level Christianity. Manila, Philippines: University of Sto Tomas Press, 1966.
21. Raja SN, Williams S, McGee R. Multideminsional health locus of control beliefs and psychological health for a sample of mothers. Soc Sci Med 1994; 39: 213-220.
22. Barnett HJM, Eliasziw M, Meldrum HE, Taylor DW. Do the facts and figures warrant a 10-fold increase in the performance of carotid endarterectomy on asymptomatic patients? Neurology 1996; 46: 603-608.
23. Gersh BJ, Opie LH. Antithrombotic agents: platelet inhibitors, anticoagulants and fibrinolytics. In: Opie LH, ed. Drugs for the Heart. 3rd ed. Philadelphia, Pa: WB Saunders Co, 1991.
24. Glasziou PP, Irwig LM. An evidence based approach to individualizing treatment. BMJ 1995; 311: 1356-1359.
25. Sazawal S, Black RE. Meta-analysis of intervention trials on case management of pneumonia in community settings. Lancet 1992; 340: 528-533.
26. Sharp SJ, Thompson SG, Altman DG. The relation between treatment benefit and underlying risk in meta-analysis. BMJ 1996; 313: 735-738.
27. Rothwell PM. Can overall results of clinical trials be applied to all patients? Lancet 1995; 345: 1616-1619.
28. Bailey KR. Generalizing the results of randomized clinical trials. Control Clin Trials 1994; 15: 15-23.
29. Oxman AD, Guyatt GH. A consumer's guide to subgroup analysis. Ann Intern Med 1992; 116: 78-84.
30. Yusuf S, Wittes J, Probstfield J, Tyroler HA. Analysis and interpretation of treatment effects in subgroups of patients in randomized clinical trials. JAMA 1991; 266: 93-98.
31. Keys A. Seven Countries: a Multivariate Analysis of Death and Coronary Heart Disease. Cambridge, Mass: Harvard University Press, 1980.
32. World Health Organization Monitoring of Cardiovascular Disease and Their Determinants (MONICA). WHO MONICA Project: assessing CHD mortality and morbidity. Int J Epidemiol 1989; 18(suppl 3, pt 1): S38-S45.
33. Guyatt GH, Sackett DL, Sinclair JC, Hayward R et al. Users' guides to the medical literature, IX. a method for grading health care recommendations. JAMA 1995; 274: 1800-1804.
© 2001 Evidence-Based Medicine Informatics Project
© 2001 Centre for Health Evidence.
Home.
Users' Guides to EBP.
Webmaster.
Disclaimer.