Ways of Using the Medical Literature

This book is about using the medical literature. But not, as we describe in the following section, in the ways medical students most typically use it.

Background and Foreground Questions

There are several reasons that medical students, early in their training, seldom consult the original medical literature. First, they are not usually responsible for managing patients and solving specific patient problems. Even if they attend a school that uses problem-oriented learning as an educational strategy, their interest is primarily in understanding normal human physiology and the pathophysiology associated with a patient's condition or problem. Once they have grasped these basic concepts, they will turn to the prognosis, available diagnostic tests, and possible management options. Finally, when students are presented with a patient-related problem, their questions are likely to include, for example, what is diabetes, why did this patient present with polyuria, and how might we manage the problem.

By contrast, experienced clinicians responsible for managing a patient's problem ask very different sorts of questions. They are interested less in the diagnostic approach to a presenting problem and more interested in how to interpret a specific diagnostic test; less in the general prognosis of a chronic disease and more in a particular patient's prognosis; less in the management strategies that might be applied to a patient's problems and more in the risks and benefits of a particular treatment in relation to an alternative management strategy.

Think of the first set of questions, those of the medical student, as background questions; think of the second set as foreground questions. In most situations, you need to understand the background thoroughly before it makes sense to address issues in the foreground.

On her first day on the ward, a medical student will still have a great deal of background knowledge to acquire. However, in deciding how to manage the first patient she sees, she may well need to address a foreground issue. A senior clinician, while well versed in all issues that represent the background of her clinical practice, may nevertheless also occasionally require background information. This is most likely when a new condition or medical syndrome appears (consider the fact that as recently as 20 years ago, experienced clinicians were asking, "What is the acquired immunodeficiency syndrome?") or when a new diagnostic test ("How does PCR work?") or treatment modality ("What are COX-2 inhibitors?") is introduced into the clinical arena. At every stage of training and experience, clinicians' grasp of the relevant background issues of disease inform their ability to identify and formulate the most pertinent foreground questions for an individual patient.

Figure 1A-1 represents the evolution of the questions we ask as we progress from being novices (who pose almost exclusively background questions) to being experts (who pose almost exclusively foreground questions). This book is devoted to how clinicians can use the medical literature to solve their foreground questions.

Browsing and Problem Solving

Traditionally, clinicians subscribed to a number- - sometimes a large number- - of target medical journals in which articles relevant to their practice were likely to appear. They would keep up to date by skimming the table of contents and reading articles relevant to their practice. One might label this the browsing mode of using the medical literature.

Traditional approaches to browsing have major limitations of inefficiency and resulting frustration. Picture a clinician with a number of subscriptions placing journals in a pile on her desk awaiting browsing review. She may even be aware that less than 10% of articles that are published in the core medical journals are both high quality and clinically useful. Unable to spend sufficient time to browse, she finds the pile growing until it becomes intimidating. At this point, she tosses the whole pile and starts the process again.

Although it is somewhat of a parody, most experienced clinicians can relate easily to this scenario. Physicians at every stage of training often feel overwhelmed by the magnitude of the medical literature. Evidence-based medicine offers some solutions to this problem.

Browse Secondary Journals. Perhaps the most efficient strategy is to restrict your browsing to secondary journals. For internal and general medicine, ACP Journal Club (http://www.acponline.org/journals/acpjc/jcmenu.htm) publishes synopses of articles that meet criteria of both clinical relevance and methodologic quality. We describe such secondary journals in more detail later in this section.

Many specialties and subspecialties do not yet have devoted secondary journals. This is likely to be a temporary phenomenon, at least for the major specialties. In the meantime, you can apply your own relevance and methodologic screen to articles in your target journals. Most clinical publications serve a dual purpose: as a forum for both investigator-to-investigator communication and investigator-to-clinician communication. However, only the latter articles will be directly relevant to your practice. Part 1 of this book is devoted to providing the tools that will allow you to screen journals for high-quality, relevant evidence. When you have learned the skills, you will be surprised both at the small proportion of studies to which you need to attend-- and at the efficiency with which you can identify them.

Operate in a Problem-solving Mode. Another part of the solution to the overwhelming-amount-of-literature problem is for clinicians to spend more of the time they have available for consulting the literature in what we call a problem-solving mode. Here, questions raised in caring for patients are defined and then the literature is consulted to resolve these questions. Whether you are operating in the browsing mode or problem-solving mode, this book can help you to judge the validity of the information in the articles you are examining, gain a clear understanding of their results, and apply them to patients.

The remainder of this section focuses on skills you will need to use the literature effectively when you are in the problem-solving mode.

Framing the Question

Clinical questions often spring to practitioners' minds in a form that makes finding answers in the medical literature a challenge. Dissecting the question into its component parts to facilitate finding the best evidence is a fundamental EBM skill. Most questions can be divided into three parts.

1. The population. Who are the relevant patients?

2. The interventions or exposures (diagnostic tests, foods, drugs, surgical procedures, etc...). What are the management strategies we are interested in comparing, or the potentially harmful exposure about which we are concerned? For issues of therapy or harm, there will always be two or more parts to this: the intervention or exposure and a control or alternative intervention(s) or exposure(s).

3. The outcome. What are the patient-relevant consequences of the exposure in which we are interested?

We will now provide examples of the transformation of unstructured clinical questions into the structured questions that facilitate use of the medical literature.

Example 1: Diabetes and Target Blood Pressure

A 55-year-old white woman presents with type 2 diabetes mellitus and hypertension. Her glycemic control is excellent on metformin and she has no history of complications. To manage her hypertension, she takes a small daily dose of a thiazide diuretic. Over a 6-month period, her blood pressure hovers around a value of 155/88 mm Hg.

Initial Question: When treating hypertension, at what target blood pressure should we aim?

Digging Deeper: One limitation of this formulation of the question is that it fails to specify the population in adequate detail. The benefits of tight control of blood pressure may differ in diabetic patients vs nondiabetic patients, in type 1 vs type 2 diabetes mellitus, as well as in those with and without diabetic complications. We may wish to specify that we are interested in the addition of a specific antihypertensive agent. Alternatively, the intervention of interest may be any antihypertensive treatment. Furthermore, a key part of the intervention will be the target for blood pressure control. For instance, we might be interested in knowing whether it makes any difference if our target diastolic blood pressure is less than 80 mm Hg vs less than 90 mm Hg. The major limitation of the initial question formulation is that it fails to specify the criteria by which we will judge the appropriate target for our hypertensive treatment. The target outcomes of interest would include stroke, myocardial infarction, cardiovascular death, and total mortality.

Improved (Searchable) Question: A searchable question would specify the relevant patient population, the management strategy and exposure, and the patient-relevant consequences of that exposure as follows:

Patients: Hypertensive type 2 diabetic patients without diabetic complications

Intervention: Any antihypertensive agent aiming at a target diastolic blood pressure of 90 mm Hg vs a target of 80 mm Hg

Outcomes: Stroke, myocardial infarction, cardiovascular death, total mortality

Example 2: Suspected Unstable Angina

A 39-year-old man without previous chest discomfort presented to the emergency department at the end of his working day. Early that day he had felt unwell and nauseated; he had had a vague sensation of chest discomfort and had begun to sweat profusely. The unpleasant experience lasted for about 2 hours, after which the patient felt tired but otherwise normal. At the end of his work day, feeling rather nervous about the episode, he came to the emergency department. The patient has no family history of coronary artery disease. He has had hypertension for 5 years that is controlled with a thiazide, has a 15-pack-year smoking history, and has a normal lipid profile. His physical examination, electrocardiogram (ECG), creatine kinase level, and troponin I level are all normal.

Initial Question: Can I send this man home or should I admit him to a monitored hospital bed?

Digging Deeper: The initial question gives us little idea of where to look in the literature for an answer. We can break down the issue by noting that the patient has suspected unstable angina. However, a number of distinguishing features differentiate him from other patients with possible unstable angina. He is relatively young, he has some risk factors for coronary artery disease, his presentation is atypical, he is now pain free, there is no sign of heart failure, and his ECG and cardiac enzymes are unremarkable.

The management strategies we are considering include admitting him to a hospital for overnight monitoring or sending him home with the appropriate follow up, including an exercise test. Another way of thinking about the issue, however, is that we need to know the consequences of sending him home. Would discharge be a safe course of action, with an acceptably low likelihood of adverse events? Thinking of our question that way, the exposure of interest is time. Time is usually the exposure of interest in studies about patients' prognosis.

What would be our objective in admitting the patient to a coronary care unit? By doing this, we will not be able to prevent more distant events (such as a myocardial infarction a month later). We are interested primarily in events that might occur during the next 72 hours, the maximum time the patient is likely to be monitored in the absence of complications. What adverse events might we prevent if the patient is in a hospital bed with cardiac monitoring? Should he develop severe chest pain, cardiac failure, or myocardial infarction, we would be able to treat him immediately. Most important, should he develop ventricular fibrillation or another life-threatening arrhythmia we would be able to administer cardioversion and save his life.

Improved (Searchable) Question: A searchable question would specify the relevant patient population, the management strategy and exposure, and the patient-relevant consequences of that exposure as follows:

Patients: Young men with atypical symptoms and normal ECG and cardiac enzymes presenting with possible unstable angina

Intervention/Exposure: Either admission to a monitored bed vs discharge home, or time

Outcomes: Severe angina, myocardial infarction, heart failure, or arrhythmia, all within the next 72 hours

Example 3: Squamous Cell Carcinoma

A 60-year-old, 40-pack-year smoker presents with hemoptysis. A chest radiograph shows a parenchymal mass with a normal mediastinum, and a fine needle aspiration of the mass shows squamous cell carcinoma. Aside from the hemoptysis, the patient is asymptomatic and physical examination is entirely normal.

Initial Question: What investigations should we undertake before deciding whether to offer this patient surgery?

Digging Deeper: The key defining features of this patient are his non-small-cell carcinoma and the fact that his history, physical examination, and chest radiograph show no evidence of intrathoracic or extrathoracic metastatic disease. Alternative investigational strategies address two separate issues: Does the patient have occult mediastinal disease, and does he have occult extrathoracic metastatic disease? For this discussion, we will focus on the former issue. Investigational strategies for addressing the possibility of occult mediastinal disease include undertaking a mediastinoscopy or performing a computed tomographic (CT) scan of the chest and proceeding according to the results of this investigation.

What outcomes are we trying to influence in our choice of investigational approach? We would like to prolong the patient's life, but the extent of his underlying tumor is likely to be the major determinant of survival and our investigations cannot change that. The reason we wish to detect occult mediastinal metastases if they are present is that if the cancer has spread to the mediastinum, resectional surgery is very unlikely to benefit the patient. Thus, in the presence of mediastinal disease, patients will usually receive palliative approaches and avoid an unnecessary thoracotomy. Thus, the primary outcome of interest is an unnecessary thoracotomy.

Improved (Searchable) Question: A searchable question would specify the relevant patient population, the management strategy and exposure, and the patient-relevant consequences of that exposure as follows:

Patients: Newly diagnosed non-small-cell lung cancer with no evidence of extrapulmonary metastases

Intervention: Mediastinoscopy for all or chest CT-directed management

Outcome: Unnecessary thoracotomy

Another way of structuring this question is as an examination of the test properties of the chest CT scan. Looking at the problem this way, the patient population is the same, but the exposure is the CT scan and the outcome is the presence or absence of the target condition, mediastinal metastatic disease. As we will subsequently discuss (see Part 1C2, "Diagnostic Tests"), this latter way of structuring the question is less likely to provide strong guidance about optimal management.

These examples illustrate that constructing a searchable question that allows you to use the medical literature to generate an answer is often no simple matter. It requires an in-depth understanding of the clinical issues involved in patient management. The three examples above illustrate that each patient may trigger a large number of clinical questions, and that clinicians must give careful thought to what they really want to know. Bearing the structure of the question in mind – patient, intervention or exposure, and outcome – is extremely helpful in arriving at an answerable question.

Once the question is posed, the next step in the process is translating the question into an effective search strategy. By first looking at the components of the question, putting the search strategy together is easier.

Searching for the Answer

In this section, we will introduce you to the electronic resources available for quickly finding the answers to your clinical questions. We will demonstrate how the careful definition of the question, including specification of the population, the intervention, and the outcome, can help you develop a workable search strategy. However, you must also consider a fourth component. What sort of study do you hope to find? By sort of study, we mean the way the study is organized or constructed--the study design.

Determining Question Type

To fully understand issues of study design, we suggest that you read the entire Part 1 of this book. Following is a brief introduction.

There are four fundamental types of clinical questions. They involve

Therapy: determining the effect of different treatments on improving patient function or avoiding adverse events

Harm: ascertaining the effects of potentially harmful agents (including the very therapies we would be interested in examining in the first type of question) on patient function, morbidity, and mortality

Diagnosis: establishing the power of an intervention to differentiate between those with and without a target condition or disease

Prognosis: estimating the future course of a patient's disease

To answer questions about a therapeutic issue, we identify studies in which a process analogous to flipping a coin determines participants' receipt of an experimental treatment or a control or standard treatment, the so-called randomized controlled trial or RCT (see Part 1B1, "Therapy"). Once the investigator allocates participants to treatment or control groups, he or she follows them forward in time looking for whether they have, for instance, a stroke or heart attack – what we call the outcome of interest (Figure 1A-2).

Ideally, we would also look to randomized trials to address issues of harm. However, for many potentially harmful exposures, randomly allocating patients is neither practical nor ethical. For instance, one could not suggest to potential study participants that an investigator will decide by the flip of a coin whether or not they smoke during the next 20 years or whether they will be exposed to potentially harmful ionizing radiation. For exposures like smoking and radiation, the best one can do is identify studies in which personal choice, or happenstance, determines whether people are exposed or not exposed. These observational studies provide weaker evidence than randomized trials.

Figure 1A-3 depicts a common observational study design in which patients with and without the exposure of interest are followed forward in time to determine whether they experience the outcome of interest. For smoking or radiation exposure, one important outcome would likely be the development of cancer.

For establishing how well a diagnostic test works (what we call its properties or operating characteristics) we need yet another study design. In diagnostic test studies, investigators identify a group of patients who may or may not have the disease or condition of interest (such as tuberculosis, lung cancer, or iron-deficiency anemia which we will call the target condition). Investigators begin by collecting a group of patients whom they suspect may have the target condition. These patients undergo both the new diagnostic test and a gold standard (that is, the test considered to be the diagnostic standard for a particular disease of condition; synonyms include criterion standard, diagnostic standard, or reference standard). Investigators evaluate the diagnostic test by comparing its classification of patients with that of the gold standard (Figure 1A-4).

A final type of study examines patients' prognosis and may identify factors that modify that prognosis. Here, investigators identify patients who belong to a particular group (such as pregnant women, patients undergoing surgery, or patients with cancer) with or without factors that may modify their prognosis (such as age or comorbidity). The exposure here is time, and investigators follow patients to determine if they experience the target outcome, such as a problem birth at the end of a pregnancy, a myocardial infarction after surgery, or survival in cancer (Figure 1A-5).

One of the clinician's tasks in searching the medical literature is to correctly identify the category of study that will address her question. For example, if you look for a randomized trial to inform you of the properties of a diagnostic test (as opposed to whether patients benefit from its application), you are unlikely to find the answer you seek.

Think back to the questions we identified in the previous section. Determining the best strategy for managing hypertension is clearly a treatment issue. However, we may also be interested in rare and delayed adverse effects of the medications we use to lower blood pressure, which is an issue of harm.

Considering the second scenario we presented, we can formulate the question in two ways. If we ask, How likely is myocardial infarction or death among young men with symptoms suggestive but atypical of unstable angina? the issue is one of prognosis. If we ask, What is the impact of alternative management strategies, such as admission to a coronary care unit or discharge? we are interested in treatment and would look for a randomized trial that allocated patients to the alternative approaches.

We can also formulate the question from the third scenario in two ways. If we ask, How well does CT scanning of the chest distinguish between non-small-cell lung cancer patients with and without mediastinal metastases? we would look for a study design that can gauge the power of a diagnostic test (see Figure 1A-4). We might also ask, "What is the rate of unnecessary thoracotomy in non-small-cell lung cancer patients who go straight to mediastinoscopy vs those who have CT scan-directed management?" For this treatment issue, we will seek a randomized trial (see Figure 1A-2).

Is Searching the Medical Literature Worthwhile?

Because our time for searching is limited, we would like to ensure that there is a good chance that our search will be productive. Consider the following clinical questions:

Example: In patients with pulmonary embolism, to what extent do those with pulmonary infarction have a poorer outcome than those without pulmonary infarction?

Before formulating our search strategy and beginning our literature search to answer this question, we should think about how investigators would differentiate between those with and without infarction. Since there is no reliable way, short of autopsy, of making this differentiation, our literature search is doomed before we even begin.

Example: Consider also a 50-year-old woman who has suffered an uncomplicated myocardial infarction 4 days previously and who asks, before discharge home, when she can resume sexual intercourse.

Were we to formulate a question that would allow us to address her inquiry, its components would look something like this.

Patients: Women after uncomplicated myocardial infarction

Intervention: Advice to resume intercourse as soon as so inclined vs waiting, say, 8 weeks

Outcomes: Recurrent infarction, unstable angina, cardiovascular and total mortality, health-related quality of life

Type of question: Therapy, therefore we would look for a randomized trial.

How likely is it that investigators have conducted a randomized trial of this question? Highly improbable. It is slightly less implausible that investigators have conducted an observational study of timing of return to sexual intercourse (here, patients would report when they had returned to sexual intercourse and investigators would compare outcomes in those who had started early vs those who had waited until later).

These two examples illustrate situations in which you will not want to use the medical literature to solve your patient management problems. The medical literature will not help you when there is no feasible study design that investigators could use to resolve the issue. Your search will also be fruitless if there is a feasible design, but it is very unlikely that anyone has taken the time and effort to carry out the necessary study. Before embarking on a search, carefully consider whether the yield is likely to be worth the time expenditure.

Sources of Evidence

You can look to local specialists, subspecialists, and more experienced clinical colleagues not only for opinion, but also for evidence to address your clinical problem (see Part 1A, "Introduction: The Philosophy of Evidence-Based Medicine"). Their experience and advice are particularly crucial when the medical literature is unlikely to be helpful. Furthermore, experts who stay current on the latest evidence in their field may be able to quickly provide you with the most relevant citations.

Clinicians will not need this book to advise them to consult respected colleagues – they do not neglect this source of data. Where clinicians might need help is in the use of online resources. We focus on online rather than print products because they are generally easier to search and more current than print products (Table 1A-3). With the relatively recent appearance of many of the resources we recommend, however, little research specifically addresses their relative merits. The approaches we describe reflect our own experiences and those of our colleagues working individually or with medical trainees.

Selecting the Best Medical Information Resource

What is the optimal medical information resource? To a large extent, it depends on the type of question that you have and the time you have available. During the late 1980s, observational studies suggested that clinicians could identify one to two unanswered questions per patient in an outpatient setting and up to five per patient in a hospital setting. More recent studies in family practice in the United Kingdom and the United States have found the rate of questions arising in patient care to be 0.32 question per patient.

Be sure to match your question to the source of information that could likely provide the most appropriate answer. To take extreme examples, MEDLINE is not the best source of information on gross anatomy, and the hospital information system is the best place to provide laboratory data for a specific patient. Table 1A-4 summarizes the types of questions that clinicians ask, along with the optimal study designs, online sources of data, and MEDLINE searching terms to match the methodologic type.




To answer focused foreground clinical questions, the most efficient approach is to begin with a prefiltered evidence-based medicine resource such as Best Evidence, the Cochrane Library, or Clinical Evidence (see Table 1A-3). By prefiltered, we mean that someone has reviewed the literature and chosen only the methodologically strongest studies. The authors of these products have designed them in such a way as to make searching easy. The sources are updated regularly -- from months to a couple of years -- with methodologically sound and clinically important studies.

Textbooks. To find answers to general background medical questions, prefiltered evidence-based medicine resources are unlikely to be helpful. Referring to a textbook that is well referenced and updated frequently is likely to be faster and more rewarding. UpToDate and Scientific American Medicine are updated regularly – from months to years, depending on the rapidity with which important new evidence is accumulating; they are heavily referenced so that you can assess how current the material is and you can even read the original articles. Other textbooks available in electronic formats, such as Harrison's Principles of Internal Medicine, can also provide valuable general background information. Additionally, new textbooks that are entirely Internet based, such as emedicine, are now available. As texts become more evidence based and routinely are updated as new evidence is published, they will provide an increasingly important source of answers to foreground as well as background questions. Our own experience suggests that UpToDate and Clinical Evidence are already well along the path to becoming evidence-based sources to answer foreground questions.

MEDLINE. MEDLINE, the bibliographic database maintained by the US National Library of Medicine, is useful primarily to answer focused foreground questions. The size and complexity of this database, however, make searching somewhat more difficult and time consuming. As a result, we recommend using MEDLINE only when searching prefiltered sources has proved fruitless (or when prior knowledge suggests, before beginning the search, that prefiltered sources will prove barren).

We will now review the databases suitable for answering a specific clinical question, illustrating their use with the example of the optimal blood pressure target level in patients with diabetes.

Using Prefiltered Medical Information Resources

A good starting point in the evidence-seeking process is to look for a systematic review article on your topic. A systematic review addresses a targeted clinical question using strategies that decrease the likelihood of bias. The authors of a rigorous systematic review will have already done the work of accumulating and summarizing the best of the published (and ideally unpublished) evidence. You will find both Best Evidence and the Cochrane Library useful for finding high-quality systematic reviews quickly and effectively. Both are also good sources to consult for original studies.

Best Evidence

Cochrane Library


Clinical Evidence

Using Unfiltered Medical Information Sources


The World Wide Web

Clinical Application

The health sciences literature is enormous and continues to expand rapidly. To the extent that this reflects ongoing research and the identification of potential improvements for patient care, this is very promising. At the same time, however, it makes the task of locating the best and most current therapy or diagnostic test more challenging. The emergence of new information products specifically designed to provide ready access to high-quality, clinically relevant, and current information is timely and encouraging.

Finding the articles that address your clinical question requires 5 to 30 minutes, depending on the resource you use or your experience with systematic searching. A full assessment of the validity and applicability requires an additional half-hour. The UKPDS study and the HOT study are the closest matches to your patient and the clinical situation. The studies show a clear reduction of diabetes-related mortality with tight blood pressure control in persons with type 2 diabetes mellitus and hypertension. You decide to set target systolic blood pressure at less than 150 mm Hg and target diastolic blood pressure at 80 mm Hg.


1. Haynes RB. Loose connections between peer reviewed clinical journals and clinical practice. Ann Intern Med. 1990;113:724-728.

2. Oxman AD, Sackett DL, Guyatt GH, for the Evidence-based Medicine Working Group. Users' guides to the medical literature, I: How to get started. JAMA. 1993;270:2093-2095.

3. Richardson WS, Wilson MC, Nishikawa J, Hayward RSA. The well-built clinical question: a key to evidence-based decisions. ACP Journal Club. 1995;123:A-12.

4. McKibbon KA, Richardson WS, Walker Dilks C. Finding answers to well built clinical questions. Evidence-Based Medicine. 1999;6:164-167.

5. Sackett DL, Straus SE. Finding and applying evidence during clinical rounds: the "evidence cart." JAMA. 1998;280:1336-1338.

6. Covell DG, Uman GC, Manning PR. Information needs in office practice: are they being met? Ann Intern Med. 1985;103:596-599.

7. Osheroff JA, Forsythe DE, Buchanan BG, Bankowitz RA, Blumenfeld BH, Miller RA. Physicians' information needs: analysis of questions posed during clinical teaching. Ann Intern Med. 1991;114:576-581.

8. Barrie AR, Ward AM. Questioning behaviour in general practice: a pragmatic study. BMJ. 1997;315:1515.

9. Ely JW, Osheroff JA, Ebell MH, et al. Analysis of questions asked by family doctors regarding patient care. BMJ. 1999;319:358-361.

10. Tight blood pressure control and risk of macrovascular and microvascular complications in Type 2 diabetes: UKPDS 38. BMJ. 1998;317:703-713.

11. Fuller J, Stevens LK, Chaturvedi N, Holloway JF. Antihypertensive therapy in diabetes mellitus (Cochrane Review). In: The Cochrane Library, Issue 4, 1999. Oxford: Update Software.

12. Hansson L, Zanchetti A, Carruthers SG, et al, for the HOT Study Group. Effects of intensive blood-pressure lowering and low-dose aspirin in patients with hypertension: principal results of the Hypertension Optimal Treatment (HOT) randomized trial. Lancet. 1998; 351:1755-1762.

13. The Canadian Lung Oncology Group. Investigation for mediastinal disease in patients with apparently operable lung cancer. Ann Thorac Surg. 1995;60:1382-1389.

14. Greenhalgh T. How to read a paper: the Medline database. BMJ. 1997;315:180-183.

15. McKibbon A. PDQ. Evidence-based Principles and Practice. Hamilton: Decker; 1999.

16. Haynes RB, Wilczynski N, McKibbon KA, et al. Developing optimal search strategies for detecting clinically sound studies in MEDLINE. J Am Med Informatics Assoc. 1994;1:447-458.

17. Wilczynski NL, Walker CJ, McKibbon KA, Haynes RB. Assessment of methodologic search filters in MEDLINE. Proc Annu Symp Comp Appl Med Care. 1994;17:601-605.

18. Devillé WL, Bezemer PD, Bouter LM. Publications on diagnostic test evaluation in family medicine journals: an optimal search strategy. J Clin Epidemiol. 2000;53:65-69.

19. Bankwala Z, Swenson LJ. Unstable angina pectoris: what is the likelihood of further cardiac events? Postgrad Med. 1995;98:155-158,161-162,164-165.