Knowledge of statistical principles prepares advanced practice registered nurses (APRNs) to independently evaluate the credibility and meaning of research so they can provide high-quality, evidence-based care. One type of statistical analysis not often used in nursing literature is the measure of association and risk.1 However, these measures are common in medical and epidemiology studies, and therefore, heavily influence evidence-based practice at the APRN level.
What do APRNs need to know about measures of association and risk? How can measures of association and risk inform APRN clinical practice? This article presents basic concepts and definitions of common risk indexes and provides case study illustrations of how these measures can be used in clinical practice.
Case study 1
Mr. T is a 62-year-old male with osteoarthritis in both knees and shoulders. He presents with a complaint of joint pain that is no longer controlled with over-the-counter (OTC) acetaminophen. He states that he has taken ibuprofen 800 mg three times a day in the past with good results and that he would like a prescription for that medication. In addition to the OTC acetaminophen, he has tried nonpharmacologic measures, such as maintaining a healthy weight, daily bicycle riding, and occasional use of a transcutaneous electrical nerve stimulation unit.
His past medical history is significant for hypertension and coronary artery disease. He manages both of these conditions with lifestyle measures and is not on any other medications. His primary care NP is concerned about the risk associated with high-dose, long-term, nonsteroidal anti-inflammatory drugs (NSAIDs) considering his history of hypertension and coronary artery disease.2
Long-term use of NSAIDs is common in primary care patients, and because of this, the APRN evaluated current research to determine the safest and most effective NSAIDs for the patient (see Relative risk evidence for case study 1). This meta-analysis presents the evidence using relative risk (RR).
Measures of risk
There are underlying concepts and definitions that must be understood to correctly interpret any statistical test. RR and odds ratio (OR) are calculated from categorical data, which is simply data that consists of categories or groups and not numerical measurements (for example, the presence or absence of a disease). Categorical data are ordinal if there is an order to the categories and nominal if not. Note that common mathematical procedures such as averaging do not make sense for categorical data (see Levels of measurement: Definition and examples).1,3
Epidemiology research frequently investigates the relationship between two categorical variables: the exposure variable (which measures whether the patient was exposed to a certain factor) and the outcome variable (which measures whether or not the patient experienced some outcome of interest). Although various study designs exist, two commonly used research methods in epidemiology involve the use of OR and RR: the prospective cohort design and the retrospective case-control study.
In prospective cohort studies, individuals are studied over time, and the exposure variable is measured before the outcome is known.3 In contrast, retrospective case-control studies start with cases that have the outcome of interest (for example, some disease) and compare their exposure to a specific factor to that of a control group of subjects who do not have the disease. RR is the index used in prospective cohort studies because the risk of experiencing the outcome can be calculated for both exposed and unexposed individuals.
The researcher starts with the exposure variable and determines how many of the exposed and how many of the unexposed individuals developed the outcome of interest. ORs are used in retrospective case-control studies because having started with the outcome variable, the researchers cannot calculate the risk of developing the outcome (often some disease) for exposed and unexposed individuals; however, they can compare the prevalence of the exposure variable for diseased individuals and controls.3 Thus, ORs are indirect measures of risk. Only in situations where the outcome of interest is rare (less than 10%) for both the exposed and unexposed groups can the RR and OR be interpreted in the same light.3
Relative risk
RR compares the likelihood of the outcome of interest in the exposed population with the likelihood of the outcome in the general or control population. More specifically:
To illustrate this concept, consider the following 2 x 2 table giving the results of a long-term, randomized clinical trial of the effects of rofecoxib on cardiovascular health (see Relative risk).4 In this study, patients were randomly assigned to either take 25 mg of rofecoxib daily or a placebo.
The risk of a cardiac event in the control group is 0.924%, whereas the risk of a cardiac event in the rofecoxib group is 2.41%. The RR of a cardiac event is computed by simply dividing these two probabilities: RR = 0.0241/0.00924 = 2.61. The probability of the outcome in the exposed population is divided by the probability of the outcome in the control population. Therefore, if the RR is less than 1, exposure decreases the risk of the outcome; if the RR is greater than 1, exposure increases the risk of the outcome. Thus, individuals who took rofecoxib were 2.61 times more likely to experience a cardiac event than individuals who were given a placebo. Rofecoxib has since been voluntarily removed from the market by the manufacturer (http://www.fda.gov/NewsEvents/Newsroom/PressAnnouncements/2004/ucm108361.htm).
RR is a quantity that is easy to understand and interpret; however, this quantity is only meaningful when the researcher has a random sample of the exposed and control populations and then observes whether the outcome of interest occurs (for example, in a cohort study). In many studies, this is logistically difficult (for example, if the outcome of interest is a rare event or takes many years to observe), so the researcher compares a sample of individuals who have experienced the outcome of interest (the cases) with a sample of individuals who have not experienced the outcome of interest (the controls) and computes an OR for various risk factors.
Case study 1 resolution
Long-term NSAID use as presented in case 1 is a common clinical concern encountered by primary care NPs. The NP in this case located a current meta-analysis that evaluated studies on NSAIDs and adverse outcomes. This meta-analysis found that ibuprofen and diclofenac demonstrated an increased risk for both major coronary events and gastrointestinal (GI) complications.2 However, naproxen did not demonstrate an increased risk for major coronary events, but it was associated with an increased risk for adverse GI events.2
The patient presented in case 1 has a low risk for GI complications but a higher risk for coronary events. Therefore, naproxen was selected for this patient instead of ibuprofen. The patient will be tested for Helicobacter pylori, and a proton pump inhibitor will be added to reduce the risk of GI bleeding if the test is positive, indicating an increased risk for gastric ulcer formation.5
The patient will receive education regarding the signs and symptoms of GI complications, will be instructed to stop taking naproxen, and advised to immediately notify the NP if any of these occur. This NP possessed the knowledge and skills necessary to correctly interpret measures of association and risk to make informed, evidence-based clinical decisions.
Case study 2
Ms. K worked in an internal medicine office as an NP. She noticed a significant increase in the number of patients in the practice diagnosed with sleep apnea. She began a review of the literature to identify recommendations for screening for this disorder and found a review of the literature based on her exact clinical question: "Which clinical signs and symptoms are predictive of obstructive sleep apnea (OSA)?"6
The authors reviewed nine cross-sectional studies of individuals diagnosed with sleep apnea. Cross-sectional studies look at groups of individuals at one point in time to evaluate the association between health-related states. They are useful for identifying those at greatest risk for a disease.3 Although both RR and OR can be computed in a cross-sectional study, using the OR allows the researcher to take into account confounding variables (and is therefore used more frequently than RR in cross-sectional studies in epidemiology).7,8
Odds ratio
The odds of an event occurring is the probability that the event occurs divided by the probability that the event does not occur. Stated more simply, it is the number of occurrences of the event divided by the number of nonoccurrences. For example, when rolling a fair, six-sided die, the odds of rolling a 6 are 1:5, whereas the probability of rolling a 6 is 1/6. An OR compares the odds of exposure in the diseased/case group with the odds of exposure in the nondiseased/control group.
For example, consider the following 2 x 2 table, which provides the results of a case-control study for a population of snorers; the cases group consisted of 126 snorers with OSA and the controls consisted of 65 snorers without OSA.9 The odds of awakening with choking are 1.52 for snorers with OSA and 0.912 for snorers without OSA. Therefore, the OR is 1.52/0.912 = 1.67 (see Odds ratio).
This means that the odds of awakening with choking are 1.67 times higher for snorers with OSA, indicating that choking may be a predictor of OSA. An OR greater than 1 indicates a positive association between exposure and disease, whereas an OR less than 1 indicates a protective relationship between exposure and disease. If the outcome of interest is very rare (which is common for case-control studies), the OR can be thought of as an approximation of the RR.7,10
Case study 2 resolution
Although the studies reviewed by Ghuman, Ludwig, and St. Anna had differences in methodology and some conflicting findings, Ms. K was able to find useful information from the review.6 She determined that she would conduct further screening among her patients with larger body mass indexes (BMIs), waist-to-hip ratios, and neck circumferences.
She would also ask her patients if anyone ever told them they had pauses in their breathing at night or if they had ever awakened feeling like they had been holding their breath.6 Ms. K also made the decision to continue to search the literature for additional studies and reviews on early recognition and diagnosis of OSA (see OR evidence for case study 2).
Case study 3
Mr. J is a master's prepared APRN currently practicing in an internal medicine office. This year, he decided to return to school to obtain his Doctor of Nursing Practice (DNP) degree. His epidemiology class coincided with the first documented cases of Ebola transmission in the United States. The students in the course were very interested in Ebola, so this topic was selected for the upcoming weekly discussion in the course.
Mr. J must develop a question for the class to discuss based on this epidemic. In order to prepare for this discussion and obtain a baseline understanding of this epidemic, Mr. J selected a current article published by the World Health Organization (WHO) Ebola Response Team (see Evidence for case study 3).
Confidence intervals
The values reported for RR or OR are single-point estimates of a population parameter; interval estimates (confidence intervals [CI]) provide a bigger picture of the range of values for that population parameter.11 CIs are commonly calculated at 90%, 95%, or 99% confidence levels. One way to interpret the CI is to imagine that you drew 100 different samples from the population. Using a 95% CI, the true RR or OR in the population could be expected to be within the CI range 95 out of 100 times; using a 99% CI, one would expect the population parameter to be within the CI range 99 out of 100 times. As the CI range becomes larger, the risk of error becomes smaller.
CIs can be calculated around any statistic. An RR or OR of 1 indicates that exposure does not increase or decrease the risk of the outcome. Because of this, a CI for RR and OR that contains 1 within the range of values is considered not significant or more likely to be due to chance.
Case study 3 resolution
The 2013 to 2014 Ebola epidemic in West Africa is the largest to date with over 4,507 probable cases as of September 2014 and a mortality of 70.8% (95% CI, 69 to 73 [95% CI = the sample mean +/- 1.96 X standard error of the mean of the distribution sample]).12 This public health emergency fueled efforts to develop a vaccine against the virus. Once a vaccine is approved, initial supplies will most likely be limited. Although the age-group most affected by Ebola is 15 to 44, (60.8%), the group with the highest mortality is age 45 and older (OR 2.47; 95% CI, 1.79 to 3.46).12
Individuals between ages 15 and 44 have an OR for mortality less than 1 (OR 0.48; 95% CI, 0.36-0.62) and are therefore less likely to die from the disease.12 Based on this information from the WHO surveillance study, the question Mr. J developed for the class discussion is as follows: If an Ebola vaccine becomes available in limited supply, outside of healthcare workers, which group should be a priority for vaccination? The group with the highest percentage of cases (ages 15 to 44) or the age-group with the highest mortality (age 45 and older)?
Moving forward
With the plethora of evidence being produced in healthcare-related disciplines, it is imperative to identify resources for point-of-care reference of evidence-based publications. APRNs should consider access to such resources an essential condition to employment negotiations.
The fact that RR, OR, and CIs are common in medical and epidemiology research make it essential for APRNs to understand the correct use and interpretation of studies that report these statistics. NPs must develop skills to evaluate the strength of evidence and to determine-with confidence-which findings should be incorporated into strong, evidence-based clinical practice and which ones require further evaluation.
REFERENCES