Fineout-Overholt, Ellen PhD, RN, FNAP, FAAN; Melnyk, Bernadette Mazurek PhD, RN, CPNP/PMHNP, FNAP, FAAN; Stillwell, Susan B. DNP, RN, CNE; Williamson, Kathleen M. PhD, RN
In July's evidence-based practice (EBP) article, Rebecca R., our hypothetical staff nurse, Carlos A., her hospital's expert EBP mentor, and Chen M., Rebecca's nurse colleague, collected the evidence to answer their clinical question: "In hospitalized adults (P), how does a rapid response team (I) compared with no rapid response team (C) affect the number of cardiac arrests (O) and unplanned admissions to the ICU (O) during a three-month period (T)?" As part of their rapid critical appraisal (RCA) of the 15 potential "keeper" studies, the EBP team found and placed the essential elements of each study (such as its population, study design, and setting) into an evaluation table. In so doing, they began to see similarities and differences between the studies, which Carlos told them is the beginning of synthesis. We now join the team as they continue with their RCA of these studies to determine their worth to practice.
RAPID CRITICAL APPRAISAL
Carlos explains that typically an RCA is conducted along with an RCA checklist that's specific to the research design of the study being evaluated—and before any data are entered into an evaluation table. However, since Rebecca and Chen are new to appraising studies, he felt it would be easier for them to first enter the essentials into the table and then evaluate each study. Carlos shows Rebecca several RCA checklists and explains that all checklists have three major questions in common, each of which contains other more specific subquestions about what constitutes a well-conducted study for the research design under review (see Example of a Rapid Critical Appraisal Checklist).
Although the EBP team will be looking at how well the researchers conducted their studies and discussing what makes a "good" research study, Carlos reminds them that the goal of critical appraisal is to determine the worth of a study to practice, not solely to find flaws. He also suggests that they consult their glossary when they see an unfamiliar word. For example, the term randomization, or random assignment, is a relevant feature of research methodology for intervention studies that may be unfamiliar. Using the glossary, he explains that random assignment and random sampling are often confused with one another, but that they're very different. When researchers select subjects from within a certain population to participate in a study by using a random strategy, such as tossing a coin, this is random sampling. It allows the entire population to be fairly represented. But because it requires access to a particular population, random sampling is not always feasible. Carlos adds that many health care studies are based on a convenience sample—participants recruited from a readily available population, such as a researcher's affiliated hospital, which may or may not represent the desired population. Random assignment, on the other hand, is the use of a random strategy to assign study participants to the intervention or control group. Random assignment is an important feature of higher-level studies in the hierarchy of evidence.
Carlos also reminds the team that it's important to begin the RCA with the studies at the highest level of evidence in order to see the most reliable evidence first. In their pile of studies, these are the three systematic reviews, including the meta-analysis and the Cochrane review, they retrieved from their database search (see "Searching for the Evidence," and "Critical Appraisal of the Evidence: Part I," Evidence-Based Practice, Step by Step, May and July). Among the RCA checklists Carlos has brought with him, Rebecca and Chen find the checklist for systematic reviews.
As they start to rapidly critically appraise the meta-analysis, they discuss that it seems to be biased since the authors included only studies with a control group. Carlos explains that while having a control group in a study is ideal, in the real world most studies are lower-level evidence and don't have control or comparison groups. He emphasizes that, in eliminating lower-level studies, the meta-analysis lacks evidence that may be informative to the question. Rebecca and Chen—who are clearly growing in their appraisal skills—also realize that three studies in the meta-analysis are the same as three of their potential "keeper" studies. They wonder whether they should keep those studies in the pile, or if, as duplicates, they're unnecessary. Carlos says that because the meta-analysis only included studies with control groups, it's important to keep these three studies so that they can be compared with other studies in the pile that don't have control groups. Rebecca notes that more than half of their 15 studies don't have control or comparison groups. They agree as a team to include all 15 studies at all levels of evidence and go on to appraise the two remaining systematic reviews.
The MERIT trial1 is next in the EBP team's stack of studies. As we noted in the last installment of this series, MERIT is a good study to use to illustrate the different steps of the critical appraisal process. (Readers may want to retrieve the article, if possible, and follow along with the RCA.) Set in Australia, the MERIT trial examined whether the introduction of a rapid response team (RRT; called a medical emergency team or MET in the study) would reduce the incidence of cardiac arrest, death, and unplanned admissions to the ICU in the hospitals studied. To follow along as the EBP team addresses each of the essential elements of a well-conducted randomized controlled trial (RCT) and how they apply to the MERIT study, see their notes in Rapid Critical Appraisal of the MERIT Study.
ARE THE RESULTS OF THE STUDY VALID?
Box. No caption avai...Image Tools
The first section of every RCA checklist addresses the validity of the study at hand—did the researchers use sound scientific methods to obtain their study results? Rebecca asks why validity is so important. Carlos replies that if the study's conclusion can be trusted—that is, relied upon to inform practice—the study must be conducted in a way that reduces bias or eliminates confounding variables (factors that influence how the intervention affects the outcome). Researchers typically use rigorous research methods to reduce the risk of bias. The purpose of the RCA checklist is to help the user determine whether or not rigorous methods have been used in the study under review, with most questions offering the option of a quick answer of "yes," "no," or "unknown."
Were the subjects randomly assigned to the intervention and control groups? Carlos explains that this is an important question when appraising RCTs. If a study calls itself an RCT but didn't randomly assign participants, then bias could be present. In appraising the MERIT study, the team discusses how the researchers randomly assigned entire hospitals, not individual patients, to the RRT intervention and control groups using a technique called cluster randomization. To better understand this method, the EBP team looks it up on the Internet and finds a PowerPoint presentation by a World Health Organization researcher that explains it in simplified terms: "Cluster randomized trials are experiments in which social units or clusters [in our case, hospitals] rather than individuals are randomly allocated to intervention groups."2
Was random assignment concealed from the individuals enrolling the subjects? Concealment helps researchers reduce potential bias, preventing the person(s) enrolling participants from recruiting them into a study with enthusiasm if they're destined for the intervention group or with obvious indifference if they're intended for the control or comparison group. The EBP team sees that the MERIT trial used an independent statistician to conduct the random assignment after participants had already been enrolled in the study, which Carlos says meets the criteria for concealment.
Were the subjects and providers blind to the study group? Carlos notes that it would be difficult to blind participants or researchers to the intervention group in the MERIT study because the hospitals that were to initiate an RRT had to know it was happening. Rebecca and Chen wonder whether their "no" answer to this question makes the study findings invalid. Carlos says that a single "no" may or may not mean that the study findings are invalid. It's their job as clinicians interpreting the data to weigh each aspect of the study design. Therefore, if the answer to any validity question isn't affirmative, they must each ask themselves: does this "no" make the study findings untrustworthy to the extent that I don't feel comfortable using them in my practice?
Were reasons given to explain why subjects didn't complete the study? Carlos explains that sometimes participants leave a study before the end (something about the study or the participants themselves may prompt them to leave). If all or many of the participants leave for the same reason, this may lead to biased findings. Therefore, it's important to look for an explanation for why any subjects didn't complete a study. Since no hospitals dropped out of the MERIT study, this question is determined to be not applicable.
Were the follow-up assessments long enough to fully study the effects of the intervention? Chen asks Carlos why a time frame would be important in studying validity. He explains that researchers must ensure that the outcome is evaluated for a long enough period of time to show that the intervention indeed caused it. The researchers in the MERIT study conducted the RRT intervention for six months before evaluating the outcomes. The team discusses how six months was likely adequate to determine how the RRT affected cardiopulmonary arrest rates (CR) but might have been too short to establish the relationship between the RRT and hospital-wide mortality rates (HMR).
Were the subjects analyzed in the group to which they were randomly assigned? Rebecca sees the term intention-to-treat analysis in the study and says that it sounds like statistical language. Carlos confirms that it is; it means that the researchers kept the hospitals in their assigned groups when they conducted the analysis, a technique intended to reduce possible bias. Even though the MERIT study used this technique, Carlos notes that in the discussion section the authors offer some important caveats about how the study was conducted, including poor intervention implementation, which may have contributed to MERIT's unexpected findings.1
Was the control group appropriate? Carlos explains that it's challenging to establish an appropriate comparison or control group without an understanding of how the intervention will be implemented. In this case, it may be problematic that the intervention group received education and training in implementing the RRT and the control group received no comparable placebo (meaning education and training about something else). But Carlos reminds the team that the researchers attempted to control for known confounding variables by stratifying the sample on characteristics such as academic versus nonacademic hospitals, bed size, and other important parameters. This method helps to ensure equal representation of these parameters in both the intervention and control groups. However, a major concern for clinicians considering whether to use the MERIT findings in their decision making involves the control hospitals' code teams and how they may have functioned as RRTs, which introduces a potential confounder into the study that could possibly invalidate the findings.
Were the instruments used to measure the outcomes valid and reliable? The overall measure in the MERIT study is the composite of the individual outcomes: CR, HMR, and unplanned admissions to the ICU (UICUA). These parameters were defined reasonably and didn't include do not resuscitate (DNR) cases. Carlos explains that since DNR cases are more likely to code or die, including them in the HMR and CR would artificially increase these outcomes and introduce bias into the findings.
As the team moves through the questions in the RCA checklist, Rebecca wonders how she and Chen would manage this kind of appraisal on their own. Carlos assures them that they'll get better at recognizing well-conducted research the more RCAs they do. Though Rebecca feels less than confident, she appreciates his encouragement nonetheless, and chooses to lead the team in discussion of the next question.
Were the demographics and baseline clinical variables of the subjects in each of the groups similar? Rebecca says that the intervention group and the control or comparison group need to be similar at the beginning of any intervention study because any differences in the groups could influence the outcome, potentially increasing the risk that the outcome might be unrelated to the intervention. She refers the team to their earlier discussion about confounding variables. Carlos tells Rebecca that her explanation was excellent. Chen remarks that Rebecca's focus on learning appears to be paying off.
WHAT ARE THE RESULTS?
Box. No caption avai...Image Tools
As the team moves on to the second major question, Carlos tells them that many clinicians are apprehensive about interpreting statistics. He says that he didn't take courses in graduate school on conducting statistical analysis; rather, he learned about different statistical tests in courses that required students to look up how to interpret a statistic whenever they encountered it in the articles they were reading. Thus he had a context for how the statistic was being used and interpreted, what question the statistical analysis was answering, and what kind of data were being analyzed. He also learned to use a search engine, such as Google.com, to find an explanation for any statistical tests with which he was unfamiliar. Because his goal was to understand what the statistic meant clinically, he looked for simple Web sites with that same focus and avoided those with Greek symbols or extensive formulas that were mostly concerned with conducting statistical analysis.
How large is the intervention or treatment effect? As the team goes through the studies in their RCA, they decide to construct a list of statistics terminology for quick reference (see A Sampling of Statistics). The major statistic used in the MERIT study is the odds ratio (OR). The OR is used to provide insight into the measure of association between an intervention and an outcome. In the MERIT study, the control group did better than the intervention group, which is contrary to what was expected. Rebecca notes that the researchers discussed the possible reasons for this finding in the final section of the study. Carlos says that the authors' discussion about why their findings occurred is as important as the findings themselves. In this study, the discussion communicates to any clinicians considering initiating an RRT in their hospital that they should assess whether the current code team is already functioning as an RRT prior to RRT implementation.
How precise is the intervention or treatment? Chen wants to tackle the precision of the findings and starts with the OR for HMR, CR, and UICUA, each of which has a confidence interval (CI) that includes the number 1.0. In an EBP workshop, she learned that a 1.0 in a CI for OR means that the results aren't statistically significant, but she isn't sure what statistically significant means. Carlos explains that since the CIs for the OR of each of the three outcomes contains the number 1.0, these results could have been obtained by chance and therefore aren't statistically significant. For clinicians, chance findings aren't reliable findings, so they can't confidently be put into practice. Study findings that aren't statistically significant have a probability value (P value) of greater than 0.5. Statistically significant findings are those that aren't likely to be obtained by chance and have a P value of less than 0.5.
WILL THE RESULTS HELP ME IN CARING FOR MY PATIENTS?
Box. No caption avai...Image Tools
The team is nearly finished with their checklist for RCTs. The third and last major question addresses the applicability of the study—how the findings can be used to help the patients the team cares for. Rebecca observes that it's easy to get caught up in the details of the research methods and findings and to forget about how they apply to real patients.
Were all clinically important outcomes measured? Chen says that she didn't see anything in the study about how much an RRT costs to initiate and how to compare that cost with the cost of one code or ICU admission. Carlos agrees that providing costs would have lent further insight into the results.
What are the risks and benefits of the treatment? Chen wonders how to answer this since the findings seem to be confounded by the fact that the control hospital had code teams that functioned as RRTs. She wonders if there was any consideration of the risks and benefits of initiating an RRT prior to beginning the study. Carlos says that the study doesn't directly mention it, but the consideration of the risks and benefits of an RRT is most likely what prompted the researchers to conduct the study. It's helpful to remember, he tells the team, that often the answer to these questions is more than just "yes" or "no."
Is the treatment feasible in my clinical setting? Carlos acknowledges that because the nursing administration is open to their project and supports it by providing time for the team to conduct its work, an RRT seems feasible in their clinical setting. The team discusses that nursing can't be the sole discipline involved in the project. They must consider how to include other disciplines as part of their next step (that is, the implementation plan). The team considers the feasibility of getting all disciplines on board and how to address several issues raised by the researchers in the discussion section (see Rapid Critical Appraisal of the MERIT Study), particularly if they find that the body of evidence indicates that an RRT does indeed reduce their chosen outcomes of CR, HMR, and UICUA.
What are my patients' and their families' values and expectations for the outcome and the treatment itself? Carlos asks Rebecca and Chen to discuss with their patients and their patients' families their opinion of an RRT and if they have any objections to the intervention. If there are objections, the patients or families will be asked to reveal them.
The EBP team finally completes the RCA checklists for the 15 studies and finds them all to be "keepers." There are some studies in which the findings are less than reliable; in the case of MERIT, the team decides to include it anyway because it's considered a landmark study. All the studies they've retained have something to add to their understanding of the impact of an RRT on CR, HMR, and UICUA. Carlos says that now that they've determined the 15 studies to be somewhat valid and reliable, they can add the rest of the data to the evaluation table.
Be sure to join the EBP team for "Critical Appraisal of the Evidence: Part III" in the next installment in the series, when Rebecca, Chen, and Carlos complete their synthesis of the 15 studies and determine what the body of evidence says about implementing an RRT in an acute care setting.
© 2010 Lippincott Williams & Wilkins, Inc.