Practicing Evidence-Based Neonatal-Perinatal Medicine

Suzanne M. Lopez, Kathleen A. Kennedy and Jon E. Tyson

This chapter focuses on five strategic processes in practicing evidence-based neonatal-perinatal medicine: (1) asking a focused clinical question; (2) searching MEDLINE, the Cochrane Library, and other sources for high-quality evidence (primary reports and systematic reviews); (3) critically appraising the retrieved evidence for its validity; (4) extracting the data; and (5) applying the results to patient care. The role of the Cochrane Collaboration in the preparation, dissemination, and timely updating of systematic reviews of evidence from randomized clinical trials is highlighted. Strategies for promoting evidence-based clinical practice are presented.

Evidence-based medicine (EBM) has been described as “the conscientious, explicit, and judicious use of current best evidence in making decisions about the care of individual patients.”³¹ The practice of EBM requires efficient access to the best available evidence that is applicable to the clinical problem.

It is essential, however, to make two disclaimers. First, not every clinical decision can be based on strong evidence because such evidence might not exist. In a study at McMaster University Medical Centre, the primary intervention for the primary diagnosis of admissions to the neonatal intensive care unit (NICU) was evaluated for category of supporting evidence as a treatment. Only 34% of cases were prescribed treatment based upon good randomized controlled trial (RCT) evidence. The majority of patients’ primary (96%) and secondary (86%) diagnoses were managed with interventions based upon sound evidence.⁵ Similar findings were found by Ellis and co-workers¹¹ in the care provided on a general medicine inpatient service in England. The estimated principal treatments prescribed for patients’ primary diagnoses were based on strong evidence from RCTs in about 50% of cases, convincing non-RCT evidence in about 30% of cases, and no substantial evidence in about 20% of cases. These findings were based on evaluations of institutions emphasizing the practice of EBM. The proportion would undoubtedly be substantially lower than 50% in other institutions. Many widely used therapies have not been well evaluated with respect to either effectiveness or safety.¹

Second, evidence provides a necessary but insufficient ground for clinical decisions. Clinical expertise is no less important under the evidence-based approach; an accurate history, physical examination, and clinical diagnosis are crucial to a properly directed search for evidence that is directly applicable to the patient’s problem. In addition, for some treatment decisions, it is essential to consider the values and preferences of parents with respect to the probable clinical outcomes of the treatments being considered for their infant.

Asking a Focused Clinical Question

A focused clinical question should contain the following elements:

• Patients of interest

• Treatment or exposure of interest

• Nature of any comparisons to be made

• Primary outcome of interest and other important outcomes

The exact form of a focused clinical question depends on whether the question concerns treatment or prevention, etiology, diagnosis, or prognosis.15,41 For questions concerning treatment or prevention, a focused question has the following form: In (patient, problem, or risk factor) does (treatment of interest) compared with (control or alternative treatment) reduce (adverse outcome[s])?

Two examples follow: (1) In women carrying fetuses of 24 to 34 weeks’ gestation who are at risk of delivering, does corticosteroid (dexamethasone or betamethasone) compared with no treatment reduce the incidence of respiratory distress syndrome (RDS) in their infants? (2) In infants ≥36 weeks with hypoxic ischemic encephalopathy, does hypothermia treatment using either whole body hypothermia or selective head cooling, compared with no hypothermia treatment, reduce the frequency of moderate to severe neurodevelopmental impairment or death at 18 to 24 months of age?

Armed with a focused clinical question based on an accurate delineation of the clinical problem, the treatment alternatives being considered, and the important clinical outcomes, a targeted search can be conducted for valid evidence that is applicable to the problem.

Finding Evidence

Sources of Evidence

Clinical evidence that is relevant to problems in neonatal-perinatal medicine is appearing at an accelerating rate and can be found in journals, conference proceedings, online databases, and other sources. Many published reports provide only weak evidence because strong research designs were not used. Evidence-based recommendations are constantly changing as new evidence becomes available. The challenge for a busy clinician is to be able to identify evidence that is valid, up-to-date, and applicable to the clinical problem using strategies that are comprehensive and yet efficient. These strategies are usually directed at retrieving primary reports and systematic reviews.

Recent review articles might seem like an efficient source of best available evidence. Because most review articles do not use explicit review methods, however, systematic reviews (discussed later) are a better source of summarized evidence. Although textbooks can provide valid evidence that is based on systematic methods of review, very few textbooks (except books that focus on evidence-based practice 7,12,27,34) require contributors to use explicit and systematic methods when reviewing evidence and making treatment recommendations. There tends to be a long time gap between the appearance of new evidence and its impact on therapeutic recommendations found in textbooks.² In neonatal-perinatal medicine and other fields in which new evidence is rapidly accumulating, it is especially important to be able to access systematic reviews that are frequently updated.

Efficient Strategies for Searching for Evidence

Primary Reports

Primary reports that are relevant to neonatal-perinatal medicine are published in numerous journals. Most of these journals are indexed in MEDLINE, but additional reports may appear in journals indexed in other computerized databases, including CINAHL and EMBASE. With access to the Internet, one can now search MEDLINE for clinical evidence using PubMed; other databases maintained by the National Library of Medicine also can be accessed. PubMed can be accessed at www.ncbi.nlm.nih.gov/pubmed. An increasing number of full-text articles are available through PubMed Central (accessible through PubMed).

To define the topic of a search, one uses Medical Subject Headings (MeSH terms), text words, or a combination, combining them appropriately in a Boolean search with AND or OR (a medical librarian can quickly teach the logic of this). Help is also available online in the PubMed tutorial. Search terms for the patient population, the intervention, the comparison, the outcome of interest, or all of these may be included.

Often the clinician finds that a MEDLINE search based only on topic descriptors yields a long list of reports that he or she does not have time to scan or read. Busy clinicians need to prune potentially cumbersome lists by incorporating into the search a strategy for limiting the retrieval to reports that are likely to be of high methodologic quality and more likely to provide valid evidence. This strategy includes using methodologic filters that have been validated against hand-searching 18,44 to identify articles that, depending on the type of focused question posed, have the methodologic quality attributes shown in Table 9-1. These methodologic filters are used together with topic descriptors (with the use of AND) so that only articles that are clinically relevant and satisfy the methodologic criteria are retrieved.

TABLE 9-1

Searching MEDLINE for Sound Clinical Studies Using Methodologic Filters

Type of Question	Criterion Standard for Methodologic Quality
Treatment	Random or quasi-random allocation of participants to treatment and control groups
Etiology	Formal control group using random or quasi-random allocation; nonrandomized concurrent controls; cohort analytic study with matching or statistical adjustment; or case-control study
Diagnosis	Provision of sufficient data to calculate sensitivity and specificity of the test, or likelihood ratios
Prognosis	Cohort of subjects who, at baseline, have the disease of interest, but not the outcome of interest

Modified from Haynes RB, et al. Developing optimal search strategies for detecting clinically sound studies in MEDLINE. J Am Med Inform Assoc. 1994;1(6):447-458.

By choosing different methodologic filters, the clinician can maximize either the sensitivity (for comprehensiveness) or the specificity (for fewest methodologic false-positive results) of his or her search. To do this, one uses PubMed’s Clinical Queries page (click on Clinical Queries on the PubMed page or access directly at http://www.ncbi.nlm.nih.gov/pubmed/clinical. After entering the clinical search terms, one is asked to click on the category of the question one is asking (therapy, diagnosis, etiology, prognosis, or clinical prediction guide) and on whether the scope of the search should be broad or narrow. If a clinician is reviewing a topic and wants to be comprehensive in retrieval of sound clinical studies, he or she would select a broad filter. If the clinician has limited time and wants urgent access to perhaps only one or two reports that are likely to be methodologically sound, he or she would select a narrow filter.

Systematic reviews 6,36 are distinguished from other types of reviews by the rigor of the review methods. The objectives and methods are explicitly planned a priori, and they are documented in the review. A review without a methods section is unlikely to be a systematic review.

Systematic reviews of RCTs attempt to identify all trials that have compared a defined therapy against an alternative in a defined population. Trials are included or excluded from the review on the basis of methodologic rigor (without consideration of the trial results). If the populations and the contrasting interventions are similar, the results may be summarized quantitatively by calculating a typical effect based on the results of all eligible trials. This latter step, called a meta-analysis, increases the precision of the estimates of treatment effect. A meta-analysis is not a necessary part of a systematic review, however; if there is clinical or statistical heterogeneity across trials, it may be inappropriate to calculate a typical effect. Systematic reviews can be found in MEDLINE by limiting the publication type to “Meta-Analysis” or by using PubMed’s Clinical Queries (Systematic Reviews column).

An example follows: The clinician wishes to find a systematic review, with meta-analysis, of studies of women at risk for preterm delivery that assesses the effect of antenatal corticosteroids on the incidence of RDS in their infants. Using PubMed, the search terms are entered: corticosteroid AND respiratory distress syndrome. The search is limited by publication type to meta-analysis. Alternatively, topic descriptors could be entered into the Clinical Queries page and the output would be viewed under “Systematic Reviews.”