Background
Clinicians caring for infants with respiratory disease should make diagnoses, choose treatments, and counsel parents based on the best evidence available (preferably from high-quality research studies) while still using clinical judgment, patient values, and knowledge of the local circumstances. This approach, known as evidence-based medicine (EBM), discourages the sole use of less reliable “sources of truth” such as reasoning based on anatomy or pathophysiology, extrapolation from animal data, expert opinion that is not based on evidence, or adoption of a practice because peers and colleagues use it. Although the principles and steps of evidence-based practice are well described, it is common for diagnostic tests and treatments to be overused, underused, or misused. Therefore clinicians providing respiratory care should become proficient in using the tools and skills of evidence-based practice. The technical skills for EBM, listed in the order they are used, are framing a question; searching for evidence; assessing its quality; weighing risks, benefits, and costs; and implementation. In addition to these technical skills, several cognitive tools are also required to practice EBM, such as critical thinking, clinical reasoning, and decision making, particularly decision making in the face of uncertainty. EBM can be used to make decisions about the care of an individual patient, or for creating a guideline or protocol for the care of a defined category of patients (such as preterm infants with respiratory distress syndrome).
The Technical Steps of EBM
Formulating the Question
The first step in practicing EBM is to clearly delineate the clinical question. The components of a well-formulated question are patients, population, or problem (P); the intervention (I); the control or comparison (C); the outcomes of interest (O); and the type of study or Time frame (T). These can be remembered with the acronym PICOT. An example of a PICOT question is “In preterm infants of less than 28 weeks’ gestation with respiratory distress syndrome (P), does the prophylactic use of vitamin A (I) compared to placebo (C) reduce the risk of bronchopulmonary dysplasia (O) at 36 weeks post-menstrual age (T) ? The best type of study (T) to answer this question is a randomized controlled trial.”
Searching for the Evidence
Once a clear question is formulated, the clinician should then perform a search for all the relevant evidence. Collaborating with a medical librarian can ensure that the search is efficient and comprehensive, identifying all key published articles, abstracts, and reviews. The search can be performed using several electronic databases. The best known of these is MEDLINE (a bibliographic database of life sciences with a concentration on biomedicine), which can be accessed either using the free interface, PubMed ( www.pubmed.gov ), or through a proprietary interface such as Ovid. A particularly useful feature in PubMed is the “Clinical Queries” tab, which allows a user-friendly, quick, and focused search on a given topic that can help the clinician make informed decisions. There are three search filters available in Clinical Queries: clinical study categories (etiology, diagnosis, therapy, prognosis, and clinical prediction guides), systematic reviews, and medical genetics. Although MEDLINE contains millions of articles, it may still not contain all the relevant articles, and if an exhaustive search is essential, other databases such as CINAHL (an index of journal articles in nursing, allied health, biomedicine, and health care) and EMBASE (a biomedical and pharmacologic database of published literature) should also be searched. To identify unpublished abstracts (an important but often overlooked source of evidence), the proceedings and published abstracts (usually available online) of pediatric or neonatology conferences such as the Pediatric Academic Societies Meeting or the American Academy of Pediatrics National Conference and Exhibition should also be searched. Google Scholar is another Web-based search tool that searches the World Wide Web. In addition to published articles, it also identifies conference proceedings, books, and institutional repositories. It may serve as an adjunct but cannot replace a systematic search of a more comprehensive database such as MEDLINE. Other online sources of information include TRIP ( tripdatabase.com ), which is a clinical search engine designed to allow users to quickly find high-quality relevant evidence. Finally, the reference lists of full-text articles obtained from the electronic search should also be hand-searched to identify additional relevant articles.
Evidence identified through the search may fall into one of two categories: primary sources (i.e., original articles and abstracts) and reviews or summaries (“predigested sources”) of existing evidence on a given topic or question. Among such reviews, conventional narrative reviews (similar to textbook chapters) can provide useful background information about the topic but are subject to the biases and viewpoints of the authors. Therefore systematic reviews are preferred because their explicit methodology and transparency allow readers to replicate the methods and draw their own conclusions and inferences. The Cochrane Database of Systematic Reviews is a particularly useful source of high-quality systematic reviews in neonatology, and the full text of these reviews is available freely on the website of the National Institute of Child Health and Human Development ( www.nichd.nih.gov/Cochrane ). Only human studies/trials are included in Cochrane reviews—they do not cover studies in animal or mechanical models. For other primary articles and reviews, the full text of each article can be obtained from the journal’s website, without a fee in some cases, and in other cases by an individual subscription or an institutional library subscription.
The search is typically performed by entering keywords (e.g., using the Medical Subject Headings—MeSH—database in PubMed) and using Boolean operators (“OR,” “AND,” and “NOT”) to restrict the results to the most relevant articles. Custom search filters can be used to narrow the search by criteria such as study type, publication period, or type of journal.
In addition to performing searches for the evidence around specific clinical questions, clinicians should also develop good habits of keeping up-to-date with emerging evidence. With several thousand medical articles published each year, it is impossible for any clinician to read or even skim each published article or its abstract. Clinicians can keep up-to-date by subscribing to periodic updates from databases such as PubMed, RSS feeds, listservs managed by universities or scientific organizations, and table of content (TOC) alerts from journals. They can also regularly peruse websites such as www.neoknowledge.com and www.ebneo.org .
Once the search is completed and the relevant articles have been obtained, the next task is to evaluate the evidence to determine whether and how it can be used in decision making.
Evaluating Evidence about Therapy
Evaluating the Quality of Evidence
Determining the quality of evidence requires each article or abstract to be critically appraised, and to do this the clinician should be aware of the strengths and weaknesses of different study designs. Table 5-1 summarizes the most common types of study designs and the advantages and disadvantages of each. Observational studies are useful for hypothesis generation, and randomized trials are best for hypothesis testing. Most of our current knowledge about risk factors and exposures that result in disease or poor outcomes comes from observational studies. Most of our current knowledge about therapies comes from trials.
| Type of Study | Description | Advantages | Disadvantages |
|---|---|---|---|
| Intervention Studies (Trials) | |||
| Randomized controlled trial | Subjects are allocated to either the intervention (experimental) group or a comparison (control) group by a pure chance process. The two groups are followed prospectively for a specified period of time and then compared in terms of outcome measures specified at the outset. Can be a parallel group trial or crossover trial. Useful to study the efficacy of an intervention in preventing or altering the course of a disease and to identify causes or risk factors or subjects at high risk. | Controls for major biases. Likely to yield valid results. Useful to detect small differences between groups. | Results may sometimes not be generalizable. Complex and expensive to conduct. |
| Cluster randomized trial | Instead of individual subjects, an entire group or a neonatal unit or a community is randomly assigned to intervention and control groups. | Avoids inadvertent exposure of control subjects to intervention (“contamination”). | Usually unblinded. Potential for recruitment bias. |
| Can generate difficult ethical challenges. Requires statistical analysis that adjusts for nonindependence of observations within a cluster. | |||
| Non-randomized trial | Allocation of subjects to experimental intervention and control group occurs by nonrandom methods. Useful to study the efficacy of an intervention in preventing or altering the course of a disease and to identify causes or risk factors or subjects at high risk. However, can be subject to bias. | Can suffer from selection bias. | |
| Observational Studies | |||
| Cohort study | The course of a group of individuals is followed forward over time to monitor the natural history of a disease, to determine prognosis, or to identify causes of disease. Can be prospective or retrospective. The researcher does not influence the exposure of the subjects to treatment or other interventions; exposure is not intentional. Good design to determine the natural history of the disease and to identify causes or risk factors or subjects at high risk | Can establish causation, determines incidence. Can match for known confounders. | Needs considerable time and resources (although less expensive than randomized trials), controls may be difficult to find, difficult to study rare disorders, subject to bias. |
| Case–control study | Subjects with the disease or outcome of interest (cases) are compared to a group of subjects without the disease (controls). The frequency of causal or risk factors (exposures) in cases relative to controls is determined and expressed as the odds ratio. Always retrospective. Careful selection of controls is required to avoid bias. Useful to identify causes or risk factors or subjects at high risk. | Less expensive, can be performed quickly, requires fewer patients, useful for rare diseases and when interval between exposure and outcome is long, can study multiple exposures. | Improper selection of controls can introduce bias, inefficient for rare exposures, temporal relationship may be difficulty to establish, cannot derive incidence, subject to recall bias. |
| Cross-sectional study | Individuals with a defined disease, risk factor, or other condition of interest are identified at a point in time (a “snapshot”). Exposure and outcome are determined simultaneously. Good design to estimate prevalence of the condition, which is calculated as the number of individuals with the condition divided by the total number in the sample. Useful to identify causes or risk factors or subjects at high risk. | Inexpensive and easy to perform. | Cannot establish causation. Exposure and outcome may depend on recall. Sample sizes or groups could be unequal. |
| Ecologic study | An observational study is conducted at a population level rather than an individual level. Differences in outcome between populations or over time are related to population characteristics that could be risk (or preventive) factors. | Inexpensive. Can potentially use data from existing databases. | Unable to assess how many exposed subjects actually develop the outcome. Tends to overestimate the degree of correlation. |
| Case series | Description of features of a coherent/consecutive set of cases. | Provides useful description of rare conditions. | In the absence of controls, cannot assess risk factors and exposures or draw inferences about efficacy of treatments. |
| May be followed by clinical trials. | |||
| Case report | Reports a rare or interesting finding in a single patient. | Useful to raise awareness of a potential complication of treatment or an unusual presentation or course of a disease. | Does not provide generalizable knowledge. |
Earlier systems of grading the quality of evidence relied almost exclusively on overall study design and ranked the evidence in a pyramid based on study design, with systematic reviews occupying the apex of the pyramid (and comprising the best evidence) and reasoning from physiology or expert opinion occupying the base of the pyramid (the least reliable form of evidence).
Currently, the quality of evidence is best evaluated using the criteria of the GRADE (grading of recommendations, assessment, development, and evaluation) system ( Table 5-2 ). In this system, study design remains a critical, but not the sole, factor in judging the quality of evidence. Additional criteria are incorporated into a judgment of the quality of the evidence. Also, the quality of evidence is assigned for each outcome, not for each study.
| Quality of Evidence | Definition |
|---|---|
| High | Further research is very unlikely to change our confidence in the estimate of effect.
|
| Moderate | Further research is likely to have an important impact on our confidence in the estimate of effect and may change the estimate.
|
| Low | Further research is very likely to have an important impact on our confidence in the estimate of effect and is likely to change the estimate.
|
| Very low | Any estimate of effect is very uncertain.
|
| CER: This is the control event rate, the rate or incidence of event or disease in the control group | ||
| EER: This is the experimental event rate, the rate or incidence of event or disease in the experimental group | ||
| Relative risk (RR) = EER/CER. If RR < 1, then the therapy decreased the risk of outcome; if RR = 1, then the treatment had no effect; if RR > 1, then the treatment increased the risk of outcome. | ||
| Absolute risk reduction (ARR) or risk difference = CER − EER. An ARR of 0 signifies no treatment effect. | ||
| Relative risk reduction = 1 − RR. | ||
| Number needed to treat (NNT) = 1/ARR. | ||
| Odds ratio (OR): If a is the number of exposed cases, b is the number of exposed noncases, c is the number of unexposed cases, and d is the number of unexposed noncases, then: | ||
| OR = ( a / c )/( b / d ) = ad / bc . | ||
| DiseaseTest | Present | Absent |
| Positive | a | c |
| Negative | b | d |
| Sensitivity (true positives): a / a + b | ||
| Specificity (true negatives): d / c + d | ||
| Positive predictive value: a / a + c | ||
| Negative predictive value: d / b + d | ||
| Positive likelihood ratio (LR+) = sensitivity/(1 − specificity) | ||
| Negative likelihood ratio = (1 − sensitivity)/specificity | ||
| Posttest odds = pretest odds × LR | ||
Stay updated, free articles. Join our Telegram channel
Full access? Get Clinical Tree
