Munro's STATISTICAL METHODS FOR HEALTH CARE

Using Researchand Statistics inHealth CareCHAPTER1OBJECTIVESAfter studying this chapter, you should be able to:1. Understand the role of research in developing knowledge for use in evidence-basedpractice guidelines.2. Discuss several ways that research can help policymakers.3. Describe the differences between descriptive and inferential statistics.4. Compose a study plan for the collection and analysis of data.HISTORICAL NOTENurses were at the forefront of the movement to use statistics to improve health care. Forexample, Florence Nightingale (1820–1910) used data from British army files to show howmost of the deaths in the British army during the Crimean War (1853–1856) were not causedby direct combat, but rather by illnesses contracted off the field of battle or as a result ofunattended wounds. Her statistical analyses convinced the British government to maintainfield hospitals and supply nursing care to soldiers. Nightingale passed her passion for statistics on to her namesake, Florence Nightingale David, the eldest daughter of her closestfriends. Florence Nightingale David (1909–1995) became a statistician and worked under KarlPearson. She produced the first edition of Tables of the Correlation Coefficient in 1938. DuringWorld War II, she used statistical models to help England prepare for German bombing raids.David later left England for the United States and founded the Statistics Department at theUniversity of California, Riverside in 1970 (Salsburg, 2001).3Munro Chap01.indd 311/7/2011 2:25:15 PM

4SECTION 1Obtaining and Understanding DataUNDERSTANDING THE ROLE OFEMPIRICAL RESEARCHNurses, allied health personnel, and publichealth professionals need a solid understanding of how information gained through observation, experience, or experimentation (e.g.,empirical knowledge), is generated becauseevidence-based practice has become the standard by which clinical and public health guidelines are produced (Andrews and Redmond,2004; McNaughton et al., 2004; Polit andBeck, 2008; Stevens, 2001). The widespreaduse of clinical guidelines since the 1980s has ledto significant improvements in the outcomes ofhealth care (Ahlqvist, Bogren, Hagman, et al.2006; Brooks, 2004; Penney and Foy, 2007).These guidelines depend on a systematic reviewof the research evidence (Stevens, 2001), which,in turn, requires a sound understanding of statistics and research methods (Klardie, Johnson,McNaughton, & Meyers, 2004; Meyers,Johnson, Klardie, & McNaughton, 2004). TheCochrane Collaboration produces the largest collection of clinical guidelines (CochraneCollaboration, 2010). This international nonprofit organization was founded in 1993 todevelop and disseminate systematic reviews ofhealth care interventions. In the United States,the U.S. Preventive Services Task Force (U.S.Preventive Services Task Force, 2007) hasassumed primary responsibility for developingevidence-based guidelines for health care.An example of an evidence-based clinicalguideline concerns the use of bed rest for backpain (Hagen, Jamtvedt, Hilde, Hinnem, 2005).Through a systematic review of the literature,Hagen and coworkers concluded that for peoplewith acute low back pain, bed rest is less effective than staying active. They also concludedthat for patients with sciatica, there is little orno difference in outcomes among those whostay active and those who rest in bed. Advancesin clinical and public health depend on practitioners such as Hagen and coworkers whoMunro Chap01.indd 4develop guidelines based on empirical research(McCormack, 2003). Taking a leading role inresearch, however, demands an understandingof how to conduct empirical research, includingcompetency in statistics.Research can also help policymakers to identify health care problems that may lend themselves to policy solutions. For example, theongoing nursing shortage is predicted to lastfor 10 years or more because of a number ofdemographic, environmental, and professionalfactors (Auerbach, Buerhaus, & Staiger, 2007).Health care organizations have responded tothis shortage by encouraging the immigrationof foreign nurses and calling for more graduates from nursing schools (Brush, Sochalski, &Berger, 2004). It appears, however, that somehealth care organizations are less concernedwith the educational level of nurses and moreconcerned with simply having more nursesof any educational background (e.g., ADN,diploma, BSN).The question suggested by this solution to thenursing shortage – whether the level of education of nurses in a given hospital affects patientoutcomes – was studied by a team of nurse, medical, and sociologic researchers, and the resultswere published in the Journal of the AmericanMedical Society (Aiken, Clarke, Cheung, et al.,2003). The findings of this study indicate thata 10% increase in the proportion of hospitalnurses with baccalaureate degrees is associated with a 5% decline in mortality after common surgical procedures. The researchers usedadvanced statistical models to account for manyfactors, other than the nurses’ education, thatmight explain the variation in hospital deathrates. In addition to the educational preparationof these nurses, the study also took into accounthow ill patients were on admission, the size ofthe hospital, the technological capacity of thehospital, whether or not it was a teaching facility, the board certification of the attending surgeons, and patient-to-nurse staffing ratios. Evenafter statistically controlling for all of these11/7/2011 2:25:16 PM

CHAPTER 1factors, there was a clear positive effect of nurseeducation level on quality of care. These findings determined that the level of nursing education is critical and that increasing the number ofnurses without concern for educational level hasserious implications for critically ill patients.TYPES OF RESEARCH STUDIESAND STATISTICSResearch studies serve many different purposes.Polit and Beck (2008) described the four mainpurposes of empirical research: description,exploration, explanation, and prediction andcontrol (considered one category). In general,research studies use two different categories ofstatistics to analyze the data collected: descriptive and inferential. Descriptive statistics aresimply numerical or graphical summaries ofdata, and may include charts, graphs, and simplesummary statistics such as means and standarddeviations to describe characteristics of a population sample. Inferential statistics are statisticaltechniques (e.g., chi-square test, the t-test, theone-way ANOVA) that allow conclusions to bedrawn about the relationships found among different variables in a population sample.Descriptive Studies andDescriptive StatisticsStudies whose primary purpose is descriptiveand explorative simply describe situations andevents. These studies use descriptive questionssuch as: “What is the marital status of peoplein the United States?” and “What is the averagelength of stay in the hospital after being admitted for an asthma attack?”. Descriptive statistics are typically used to analyze data in orderto answer these types of questions (see Chapter2 for more information about descriptive statistics). Table 1-1 illustrates the use of descriptive statistics to answer the question about themarital status of women in the United Statesby using data from the 2006-2008 AmericanMunro Chap01.indd 5Using Research and Statistics in Health CareTable 1-15MARITAL STATUS OF U.S.WOMEN AGE 15 YEARS ANDOLDERStatusPercentCurrently married50.2Never e: U.S. Census Bureau. (2010). 2006–2008 AmericanCommunity Survey 3-year estimates. S1201 Marital status.Retrieved from http://factfinder.census.gov/servlet/STTable?bm y&-geo-id 01000US&-qr name ACS 2008 3YR G00S1201&-ds name ACS 2008 3YR G00Community Survey (U.S. Census Bureau, 2000).As shown in Table 1-1, the survey found thatapproximately 50.2% of women were currentlymarried, 30.8% had never married, 10.6%were divorced, 2.2% were separated, and 6.3%were widowed (U.S. Census Bureau, 2010)Explanatory Studies andInferential StatisticsStudies that have the primary purpose of elucidating the relationships among variables are considered explanatory studies. Data for such studiesis often collected through observational studies,those in which the researcher just collects information about the study participants’ current,past or future status regarding the variables ofinterest without intervening in any way to changetheir status. The questions answered with thesetypes of statistics are usually more complex thanthose answered by descriptive statistics. Theirquestions and lines of inquiry are often based onestablished theories from the research literature.Explanatory studies depend on inferential questions [such as: “Are women who aresedentary during the third trimester of pregnancy more or less likely to have a cesareansection (C-section) than women who exercise regularly during the third trimester?” or11/7/2011 2:25:16 PM

6SECTION 1Obtaining and Understanding Data“Do people with health insurance have a longeror shorter hospital stay after being admitted foran asthma attack than people without healthinsurance?”]. Explanatory studies do not necessarily attempt to establish causality but oftenattempt to understand how variables are relatedto each other. For example, a question mightbe: “Does length of hospital stay for an asthmaattack differ depending on health insurance status?”. Inferential statistics are used to examinehow one variable is related to other variables; inother words, the relationship among variables(see Chapters 5 to 11 for more informationabout inferential statistics).An example of an explanatory study is oneconducted by Ludwig-Beymer and Gerc (2002),who examined the relationship between exercise behavior and receiving the flu vaccine in asample of 999 health care workers. Table 1-2shows the data from this study in a cross-tabulation table. A cross-tabulation table, sometimesreferred to as a cross-tab, is a way to display therelationship between two variables. Table 1-2shows that 48.1% of the health care workerswho exercised regularly received the influenzavaccine compared with 52.4% of those who didnot exercise regularly. Even though these numbers are not identical (48.1% versus 52.4%), astatistical test of probability (the chi-square test)Table 1-2indicates that they are not statistically different,meaning that the two groups did not differ intheir likelihood of obtaining the influenza vaccine more than one might expect just due torandom chance, and therefore the small difference we see can probably just be attributableto chance rather than to exercise habits (seeChapter 10 for more information about chisquare analysis).Prediction and Control Studies andInferential StatisticsPrediction and control studies seek to determinewhich variables are predictive of other variables and to determine causality (e.g., one eventcauses another to happen). Data for prediction and control studies are typically collectedusing quasi-experimental or experimental studydesigns in which researchers introduce an intervention (e.g., change one of the variables beingexamined) as these types of studies are thoughtto have better validity, making causal inferencemore solid than with purely observational studydesigns. True experimental designs include random selection and random assignment of studyparticipants to either the intervention groupor to one or more control groups that do notreceive the intervention. Quasi-experimentalRELATIONSHIP OF REGULAR EXERCISE TO OBTAINING A FLU VACCINATIONIN 999 HEALTH CARE WORKERSRECEIVED FLU VACCINATIONn (%)n (%)Row TotalsRegular ExerciseYesNoNumber in each gender groupYes235 (48.1)254 (51.9)489No267 (52.4)243 (47.6)510Column totals(number in each vaccination group)502497999Note: Chi-square p .18 (not statistically significant).Source: Data from Ludwig-Beymer P, & Gerc SC. (2002). An influenza prevention campaign: the employee perspective. Journal ofNursing Care Quality, 16(3), 1–12.Munro Chap01.indd 611/7/2011 2:25:16 PM

CHAPTER 1designs are similar to experimental designsexcept that they lack one or more of the following: random assignment to the intervention orcontrol group or, in some cases, a true controlgroup (Polit, 2008).Randomized control trials (RCTs) are considered experimental designs because studyparticipants are randomly assigned to an intervention group or a control group and followedforward in time to determine if the interventionimpacts on a specific health outcome. However,randomized control trials generally do notselect study participants randomly from thepopulation. Instead they have strict eligibilitycriteria that those interested in participating inthe study must meet before they are allowed toparticipate. Anyone not meeting these eligibilitycriteria is excluded from the study. This departure from the random selection of study participants from the general population may limit theexternal validity of the study; in other words,the study results may not be generalizable to thegeneral population.In health-related research, quasi-experimentaldesigns are often used, but their validity maynot be much better than that of observationalstudies; therefore, experimental studies areconsidered the gold standard for causal inference. As with explanatory studies, predictionUsing Research and Statistics in Health Care7and control studies use inferential statistics toanalyze the data and answer research questionsabout the relationship among variables.TEN STEPS TO BUILDINGA STUDY PLANAll studies, no matter what the purpose, needto be well-planned. For example, in studies inwhich many variables are measured, it is easy tolose track of the initial purpose of the study andto generate “results” that appear to be useful.These results, however, are meaningless unlessthey exist in the context of an organized lineof inquiry. Box 1-1 lists some of the commonmistakes researchers make when embarkingon research projects. These mistakes are oftenmade when there is no study plan or when theplan is insufficiently detailed. Papers resultingfrom studies that have inadequate study plansoften lack focus and clarity. Although severalwell-known methods for writing a study planare available, they all follow the same basicprinciples.A study plan is a written presentation of howthe researcher is going to obtain and analyze thenumerical data needed to answer the researchquestions. A good study plan keeps the analysis focused and relevant. It serves as the basisBOX 1-1 EIGHT COMMON MISTAKES IN RESEARCH1.2.3.4.5.6.7.8.Undertaking a research project without reviewing the existing literature on the subjectCollecting data without a well-defined plan, hoping to make sense of it afterwardsTrying to fit meaningful research questions to existing dataDefining terms in general or ambiguous languageFailing to base research on a sound theoretical foundationFailing to make explicit and clear the underlying assumptionsFailing to recognize the limitations of the approachFailing to anticipate rival hypotheses that would account for findings and that challengeinterpretations and conclusionsSource: Courtesy of Dr. Brenda Nichols, Dean of the College of Arts and Sciences, Lamar University, Beaumont,Texas.Munro Chap01.indd 711/7/2011 2:25:16 PM

8SECTION 1Obtaining and Understanding Datafor the introduction and methods section ofresearch papers after the data have been collected and analyzed. A study plan can also serveas the basis for the first sections of a dissertationor thesis. In addition, most grants require studyplans similar to the one presented here.The outline that a study plan should follow is summarized in Box 1-2 and describedon the following page. The method presentedhere is fairly standard, and similar ones can befound in guides to planning research (Ogdenand Goldberg, 2002; Wood, 2006). A studyplan begins with a statement of the researchquestion(s) that the study is trying to answer(i.e., the purpose of the study) and a shortdescription of the significance or importance ofthe question(s). The statement of purpose is theguiding force behind the entire research project, and the study should flow from it. A studyplan also needs a theoretical or conceptualframework on which research questions andhypotheses are based. This framework presentsa structured way of thinking about the interrelationships of the variables. Research questionsare either very specific or broadly conceptual.The hypotheses, however, must be very specificbecause they provide the guide for the analysisof the data. The study plan should define keyterms and variables, provide a description ofthe research design, and describe the sample andhow it was obtained. The plan should also statethe statistical techniques that will be used to testeach hypothesis. A good study plan lists anymajor assumptions and limitations of the studybeing described. And finally, a good study plancontains a brief description of how the findingsobtained from the study will be disseminated.Statement of the Purpose of the studyand Its SignificanceA study plan starts with a clear explanation ofthe purpose of the study and the significanceof the problem to be studied. This explanation should include the reasons why the studyis important and how the study fits into theexisting body of research. This section orientsresearchers and interested readers to the study.The statement of the problem should be nomore than two or three sentences, and shouldclearly articulate what the study is seeking toaccomplish. The rationale for the study shouldinclude a brief overview of the epidemiologyof the problem being addressed. It should alsoinclude a discussion of the monetary and nonmonetary costs of the problem to society, tothe health care system, and to people who havethe problem. It should also provide a review ofother studies in the literature that have examined similar issues. The rationale should thenBOX 1-2 TEN-STEP STUDY PLAN1.2.3.4.5.6.7.8.9.10.Munro Chap01.indd 8Statement of the problem and its significanceTheoretical or conceptual frameworkResearch questions to be answered by the studyList of hypotheses to be testedDefinitions of key terms and variablesDescription of the research designDescription of the sample and how it was obtainedDescription of the planned statistical analysisStatement of assumptions and limitationsDissemination plan11/7/2011 2:25:16 PM

CHAPTER 1explain the weaknesses

SIXTH EDITION Stacey B. Plichta, ScD, CPH Professor CUNY School of Public Health at Hunter College New York, New York Elizabeth A. Kelvin, PhD, MPH Assistant Professor CUNY School of Public Health at Hunter College New York, New York MMunro FM.indd i