Epidemiology Review Questions:

1. A team of researchers hypothesize that watching violent, noneducational cartoons might lead to epilepsy in childhood. Children with and without epilepsy are compared on the basis of hours spent watching violent, noneducational cartoons. Which of the following statements best characterizes the assessment of data in such a study? A. Absolute and relative measures of risk can be derived. B. The difference in level of exposure to a putative risk factor is the basis for comparison. C. The risk ratio can be calculated directly. D. The temporal association between exposure and outcome can be established with certainty. E. The use of healthy controls ensures external validity. 2. The researchers in question 1 do not find statistically significant evidence that cartoons produce epilepsy in childhood. However, determined to show that violent, noneducational cartoons cause harm, they hypothesize that perhaps viewing this type of programming leads to attention deficits. They assemble two groups of children: those who watch violent, noneducational cartoons frequently and those who watch only other types of television programming. Which of the following is a characteristic of such a study? A. Additional risk factors cannot be assessed as the study progresses. B. Internal validity is independent of confounders. C. The two study groups are assembled on the basis of their outcome status. D. The most appropriate measure for comparing outcomes in this study would be an odds ratio. E. The temporal association between exposure and outcome is uncertain.

1. B. This is a case control study. The cases are children who have epilepsy, and the controls are (or should be) children who do not have epilepsy but whose characteristics are otherwise similar to those of the cases. The difference in the level of exposure to a putative risk factor (violent, noneducational cartoons) is the basis for comparison. The use of healthy controls does not ensure that the results will pertain to a general population, so external validity is uncertain (E). In a case control study, the risk ratio cannot be calculated directly, but must be estimated from the odds ratio (C). The temporal relationship between exposure and outcome usually is not known with certainty in this type of study (D); exposure and disease already have occurred at the time of enrollment. Only relative measures of risk can be calculated based on a case control study (A). A cohort study is required to obtain absolute risk. 2. A. This is a cohort study. The two groups are assembled on the basis of exposure status (i.e., risk factor status), not outcome status (C), and are then followed and compared on the basis of disease status. Only the risk factors, or exposures, identified at study entry can be assessed as the study progresses. The development of attention deficits in this case are the "disease" or outcome of interest. The temporal association between exposure and outcome should be established to prevent bias (E); in a cohort study, subjects are generally free of the outcome (disease) at the time of enrollment. The inclusion of subjects who developed attention deficits before the defined exposure would bias the study, so these subjects should be excluded. In all epidemiologic studies, a thorough attempt must be made to control confounders because confounding compromises internal validity (B). The most appropriate means of comparing the two groups in this study is the risk ratio, or relative risk of attention deficits (D). The control group should have been otherwise similar to those who watch violent, noneducational cartoons except that they did not watch this programming.

1. This is calculated as c/(a + c). A. Accuracy B. Alpha error C. Beta error D. Bias E. Cutoff point F. False-negative error rate G. False-positive error rate H. Positive predictive value I. Precision J. Random error K. Sensitivity L. Specificity

1. F. The false-negative error rate is equal to c/(a + c). In a 2 × 2 table, cell c represents the number of participants who have a false- negative test result (participants in whom the disease of interest is present, but the test result is negative). Cell a represents the number of participants who have a true-positive test result. The sum of cells a and c represents all participants who have the disease; this is the denominator and population from which false-negative results are derived. The false-negative error rate is the ratio of participants who have false-negative results to all participants who have the disease, including both those detected and those missed by the diagnostic test.

11. Assume that the risk of death in patients with untreated pneumonia is 15%, whereas the risk of death in patients with antibiotic-treated pneumonia is 2%. Assume also that the risk of anaphylaxis with antibiotic treatment is 1%, whereas the risk without treatment is essentially 0%. What is the number needed to treat (NNT) in this scenario? A. 1.0 B. 7.7 C. 13 D. 39.4 E. 100 12. Concerning the information given in question 11, what is the number needed to harm (NNH) in this scenario? A. 1.0 B. 7.7 C. 13 D. 39.4 E. 100 13. Concerning the information given in question 11, what would be the net result of intervention in this scenario? A. 1.0 patient saved for each patient harmed B. 7.7 patients harmed for each patient saved C. 13 patients saved for each patient harmed D. 39.4 patients harmed for each patient saved E. 100 patients saved for each patient harmed

11. B. The NNT is calculated as 1 divided by the absolute risk reduction (ARR) associated with an intervention. In this scenario the intervention (antibiotic treatment) reduces the risk of death from 15% to 2%, so ARR is 13%, or 0.13. Dividing 0.13 into 1 yields 7.7. This implies that 1 life is saved, on average, for every 7.7 patients treated with antibiotics. The 1.0 (A) represents the percent absolute risk increase of anaphylaxis with antibiotics. The 100 (E) is the number needed to harm (see explanation to answer 12). The 13 (C) is the percent ARR of death with antibiotics. The 39.4 (D) is a nonsense distracter. 12. E. When an intervention increases (rather than decreases) the risk of a particular adverse outcome, the number of patients treated, on average, before one patient has the adverse outcome is the NNH. To calculate the NNH, the absolute risk increase (ARI) associated with the intervention is divided into 1. In this scenario there is a 1% risk of anaphylaxis (the adverse outcome) when antibiotic treatment (the intervention) is used, and a 0% risk of anaphylaxis when antibiotic treatment is not used. Antibiotic treatment increases the risk of anaphylaxis by an absolute 1% (0.01). The NNH is 1/0.01, or 100. On average, 100 patients would need to be treated before 1 patient, on average, is harmed by the intervention. Incorrect answer choices (A-D) are explained under answer 11. 13. C. If the NNT were greater than the NNH, the intervention would harm more patients than help. Ignoring the issue that the type and degree of harm and help may differ, if the NNH were greater than the NNT, the intervention in question would help more patients than it harmed. In this scenario the NNH (100) is greater than the NNT (7.7); the procedure provides net benefit because fewer patients need to be treated before one is helped than need to be treated before one is harmed. In this case, dividing the ARR (13%) by the ARI (1%) yields 13 as the number helped for each one harmed. When the ARR is greater than the ARI, there is net benefit, with more than one patient helped for each one harmed. By the same reasoning, when the ARI exceeds the ARR, there is net harm, with more than one patient harmed for each one helped. (The actual numbers in this case are fictitious and used for illustration purposes only.) Answers A, B, D, and E are nonsense distracters having numerical values corresponding to the metrics explained in answer 11.

14. You randomly assign clinical practices to charting with a new electronic medical record (intervention) or charting using paper records (control), and you test whether physicians are satisfied or not (outcome) after 3 months. Of 100 subjects assigned to each condition, you find evidence of satisfaction in 12 in the EMR group and 92 in the paper group. An appropriate measure of association in this study is the: A. Incidence density B. Power C. Likelihood ratio D. Odds ratio E. Risk ratio 15. Concerning the information given in question 14, the value of the measure identified is: A. 0.08 B. 0.13 C. 1.3 D. 8.1 E. 13.1

14. E. The study described is a cohort study, in which subjects are assembled on the basis of exposure status (charting method) and followed for outcome (physician satisfaction). The risk ratio is the outcome measure of a cohort study with a dichotomous outcome. The odds ratio (D) is used in case control studies rather than cohort studies and is in essence the "risk of having the risk factor." The likelihood ratio (C) is used to evaluate the utility of a diagnostic test. Incidence density (A) is an outcome measure of events per subject per unit of time (e.g., per 100 person-years) and is particularly useful when events may recur in individual participants. Power (B) is used to assess the capacity of a study to detect an outcome difference when there truly is one, given a particular sample size; power is used to avoid beta error (type II or false-negative error). 15. B. To calculate the risk ratio (relative risk), the risk of the outcome in the exposed is divided by the risk of the outcome in the unexposed. In this study the outcome is physician satisfaction; the exposure is electronic medical record (EMR) charting; the risk of the outcome in the exposed is 12/100, or 0.12; and the risk of the outcome in the unexposed is 92/100, or 0.92. Dividing 0.12 by 0.92 yields 0.13. This implies that the "risk" of physician satisfaction in those exposed to the EMR was 0.13 times the risk of that outcome in the control group. Note that an "exposure" can increase or decrease the "risk." In this case the "risk" of satisfaction decreases with EMR implementation, and this may be easier to interpret by considering the inverse ratio: risk of unexposed to risk of exposed, or 0.92/0.12 = 7.7; in other words, the risk of satisfaction is 7.7 times greater among physicians using paper charts than among physicians using an EMR, 3 months after EMR implementation. Note also that the outcome in question need not be an adverse outcome, despite the connotation of the term "risk." Answers A, C, D, and E are nonsense distracters.

2. During a given year, 12 cases of disease X are detected in a population of 70,000 college students when those 12 students present for medical attention. Many more students have mild symptoms of the disease and do not seek care. Of the 12 detected cases, 7 result in death. The ratio of 7/12 represents: A. The case fatality ratio B. The crude death rate C. The pathogenicity D. The standardized mortality ratio E. 1-prevalence 3. In regard to question above, to report the incidence rate of disease X, it would be necessary to know: A. Nothing more than the data provided B. The pathogenicity C. The infectiousness of the disease D. The duration of the clinical illness E. The midyear population at risk 4. In regard to the original question, to report the prevalence of disease X, it would be necessary to know: A. The cure rate B. The duration of illness C. The number of cases at a given time D. The number of losses to follow-up E. The rate at which new cases developed

2. A The case fatality ratio for a particular condition is the number of deaths caused by the condition, divided by the total number of identified cases of the condition in a specified population. In this example, the case fatality ratio is 7/12, which, expressed as a percentage, is 58.3%. The crude death rate (B) is the number of deaths caused by the condition, divided by the midperiod population. Pathogenicity (C) is indicated by the proportion of infected persons with clinical illness. The standardized mortality ratio (D) is the number of observed deaths in a population subgroup, divided by the expected deaths based on a reference population. The term (1-prevalence) is not meaningful (D). 3. E The incidence rate is the number of new cases in a specified population, during a specified period, divided by the midperiod population at risk. In the information provided in question 2, it does not say whether 70,000 represents the population at the midpoint or at the beginning of the observation period, and it does not say whether the entire population is at risk for disease X. Sometimes, incidence rates are based on the total population, although not everyone is at risk, because there is no convenient way to distinguish the susceptible from the immune population. An example would be incidence rates of hepatitis B in the United States; only the unimmunized population would truly be at risk, but the rate might be reported with the total population in the denominator. 4. C By definition, the prevalence of a condition is the number of cases in a specified population at a particular time. If this information is known, nothing else is required to report the prevalence. The prevalence is influenced by the duration of illness (B), the cure or recovery rate (A), the cause-specific death rate, and by immigration and emigration. These are factors that influence the number of cases in the study population at any particular time. The number of losses to follow-up (D) and the rate at which new cases develop (E) are not needed to report a prevalence; these have more to do with incidence metrics.

3. An example of this type of systematic distortion of study data is weighing subjects while they are fully dressed. A. Biologic plausibility B. Confounder C. Effect modifier D. External validity E. Internal validity F. Intervening variable G. Measurement bias H. Necessary cause I. Recall bias J. Sufficient cause K. Synergism

3. G. In contrast to random error, measurement bias is a systematic distortion of study data. Random error produces some measurements that are too large, some that are too small, and perhaps some that are correct. Such error would contribute to variability within groups, limiting an investigator's ability to detect a significant difference between groups. Random error reduces the power of a study to show a true difference in outcome. When error is systematic, rather than random, statistical power may be preserved, but the study's validity (i.e., its most critical attribute) is threatened. Consider a study of weight loss in which the control participants and the intervention participants were weighed fully clothed and with shoes on at enrollment. After a weight loss intervention, the control participants were again weighed fully dressed, but the intervention participants were weighed after disrobing. This would be a biased and invalid measure of the intervention and resultant weight loss. Bias may threaten internal validity, external validity, or both.

4. A study is conducted to determine the effects of prescription stimulant use on an individual's willingness to bungee jump. A total of 500 individuals are assembled on the basis of bungee-jumping status: 250 are jumpers and 250 are not jumpers. Of the 250 jumpers, 150 report prescription stimulant use. Of the 250 nonjumpers, 50 report prescription stimulant use. Most of the nonjumpers take anxiolytics. Which of the following statements is true? A. Jumpers and nonjumpers should be matched for prescription stimulant use. B. The absolute and relative risks of bungee jumping with the use of prescription stimulants can be determined from this study. C. This is a cohort study. D. This study can be used to calculate an odds ratio. E. Unanticipated outcomes can be assessed in this study. 5. Considering the information given in question 4, what is the absolute difference in the risk of jumping between those using prescription stimulants and those not using these drugs? A. 0.4 B. 0.67 C. 67 D. 100 E. It cannot be calculated because this is a case control study 6. Considering the information given in question 4, the odds ratio calculated from this study would give the odds of: A. Jumping among stimulant users to jumping among nonstimulant users B. Jumping among stimulant users to nonjumping among nonstimulant users C. Nonjumping among stimulant users to nonstimulant use among jumpers D. Stimulant use among jumpers to stimulant use among nonjumpers E. Stimulant use among nonjumpers to nonstimulant use among jumpers 7. Concerning the information given in question 4, the odds ratio in this study is: A. 0.2 B. 0.6 C. 2 D. 5 E. 6 8. Concerning the information given in question 4, the results of this study indicate that: A. Bungee jumping and stimulant use are associated. B. Bungee jumping and stimulant use are causally related. C. Bungee jumping influences one's need for prescription stimulants. D. The use of prescription stimulants influences a person's tendency to bungee jump. E. There is no association between anxiolytic use and bungee jumping.

4. D. The study described is a case control study (not cohort, C). The cases are individuals who have a history of bungee jumping, and the controls are (or should be) individuals who have no history of bungee jumping but whose characteristics are otherwise similar to those of the cases. The cases and controls should not be matched for prescription stimulant use (A) because this is the exposure of interest. Overmatching is the result when cases and controls are matched on the basis of some characteristic or behavior that is highly correlated with the exposure of interest. Overmatching precludes the detection of a difference in exposure, even if one truly exists. The odds ratio is calculated from a case control study and is used to estimate relative risk. Absolute risk cannot be determined from a case control study (B). In this type of study, the outcome defines the group at study entry; although unanticipated risk factors can be assessed, only the outcome variables chosen as the basis for subject selection can be evaluated (E). 5. E. Absolute risk cannot be determined from a case control study because the groups do not represent the population from which they were drawn. All numerical choices (A-D) are incorrect for this reason. Only the odds ratio can be calculated from a case control study. 6. D. The odds ratio in a case control study is the odds of the exposure in cases relative to controls. The exposure in this study is prescription stimulant use. Cases are those who bungee jump, whereas controls are those who do not. The odds of using prescription stimulants among cases relative to controls is the outcome of interest. The odds of jumping among stimulant users to jumping among nonstimulant users (A) could also be an odds ratio of interest, but it would come from a different case control study, with a different design, to answer a different question. Specifically, such a study would assemble cases (stimulant users) and controls (nonstimulant users) and look for the exposure (whether they bungee jump or not). The other answers (B, C, and E) are nonsense distracters. 7. E. The formula for the odds ratio (OR) is (a/c)/(b/d). This is algebraically equivalent to ad/bc, where a represents cases with the exposure, b represents controls with the exposure, c represents cases without the exposure, and d represents controls without the exposure. Displaying the data from the study in a 2 × 2 table is helpful. 8. A. The odds of using prescription stimulants are six times as great in cases as they are in controls in this study. There is an apparent association between stimulant use and tendency to jump. There may in fact be no true causal relationship (B) if the findings are (1) caused by chance, (2) biased (e.g., if cases are interviewed more thoroughly or differently than controls with regard to psychotropic use), or (3) confounded by an unmeasured variable (e.g., alcohol consumption). Moreover, even if the relationship were causal, the found association would not tell us anything about the direction; in other words, we cannot determine whether bungee jumping predisposes to stimulant use (C), or vice versa (D), because exposure and outcome are measured simultaneously in a case control study, and the temporal sequence is uncertain. As for anxiolytics, these drugs may or may not be associated with bungee jumping (E). This study did not assess anxiolytic exposure, and the question gives no information about anxiolytic use among jumpers.

6. The US President invites a group of legislators to a formal luncheon at the White House. Within 24 hours, 11 of the 17 diners experience abdominal pain, vomiting, and diarrhea. The President does not eat the salmon and feels fine. Of the 11 symptomatic guests, 4 have fever and 7 do not; 5 have an elevated white blood cell count and 6 do not; 6 ate shrimp bisque and 5 did not; 9 ate salmon mousse and 2 did not; and 1 goes on to have surgery for acute cholecystitis resulting from an impacted calculus (stone) in the common bile duct. Of the 11 symptomatic guests, 10 recover within 3 days; the exception is the senator who underwent surgery and recovered over a longer period. The guests at this luncheon had shared no other meals at any time recently. The fact that 11 of 17 diners become sick: A. Is a coincidence until proven otherwise B. Represents a disease outbreak C. Is attributable to bacterial infection D. Is not an outbreak because the usual pattern of disease is unknown E. Should be investigated by the CDC 7. Considering all the details from question 6, the attack rate is: A. 4/11 B. 5/11 C. 9/11 D. 1/17 E. 11/17 8. Considering all the details from question 6, the earliest priority in investigating the phenomenon would be to: A. Close the kitchen temporarily B. Define a case C. Perform a case control study D. Perform stool tests E. Submit food samples to the laboratory 9. Considering all the details from question 6, the best case definition for the guests' disease would be: A. Abdominal pain, vomiting, and diarrhea within 24 hours of the luncheon B. Acute viral gastroenteritis C. Staphylococcal food poisoning D. The onset of abdominal pain and fever after the luncheon E. An elevated white blood cell count 10. Considering all the details from question 6, and suspecting that the disease may be the result of a common- source exposure involving contaminated food, the investigators attempt to determine which food is responsible. Their initial task is to: A. Analyze food specimens from the luncheon in the laboratory B. Close the kitchen C. Examine the kitchen and interview the food preparers about their techniques D. Interview luncheon attendees to find out what they ate E. Perform a case control study

6. B Although there is no formal surveillance of these luncheons, clearly the situation described represents the unexpected occurrence of disease; almost two-thirds of diners becoming sick after a meal does not represent a usual pattern (D). The usual pattern of disease in this case does not derive from surveillance but from common experience. Nobody expects so many people to sicken simultaneously after lunch; this "attack rate" clearly represents more than coincidence (A). The CDC would not investigate so small and localized an outbreak unless requested to do so by local public health officials (E). The illness presented here could have been caused by bacterial infection (C), but other microorganisms, toxins, and even deliberate poisoning are also on the list of possibilities. 7. E. The attack rate is the proportion (yes, it is called a "rate" but is actually a ratio or proportion) of exposed persons who become ill. Given the information in question 6, the exposure at this stage (before the investigation) is most inclusively and best defined as "participation in the luncheon." As the outbreak investigation proceeds, the definition of exposure could change, for instance, if a particular food, such as the salmon mousse, were implicated. The denominator used to calculate the attack rate would then change to represent only the persons exposed to the contaminated food. There were 17 diners, so 17 is the attack rate denominator. The numerator is the number of people who became ill. The question suggests that the illness is abdominal pain, vomiting, and diarrhea, and that 11 people had these symptoms. Thus the attack rate is 11/17. However, if the case definition is further restricted to include only those affected individuals who additionally had fever, or only those who had an elevated white blood cell count, the numerator would change, and the attack rate would be 4/17 or 5/17, respectively. The proportion of ill individuals having fever was 4/11 (A). The proportion of ill individuals having an elevated white blood cell count was 5/11 (B). The proportion of ill individuals eating the salmon mousse was 9/11 (C). The proportion of all diners (ill and not ill) requiring surgery was 1/17 (D). 8. B. Establishing a diagnosis or a case definition is the earliest priority in an outbreak investigation. Investigators cannot investigate the origin of a problem until they precisely define the problem. Laboratory tests (D and E), control measures (A), and case control studies (C) are usually performed later in an investigation, after hypotheses have been generated regarding transmission and cause. 9. A. The case definition is required to distinguish between cases and noncases of the disease under investigation (i.e., to distinguish who is "ill" and who is "not ill" by strict criteria). The ideal case definition would permit the inclusion of all cases and the exclusion of all noncases, providing perfect sensitivity and specificity (see Chapter 7). The case definition should include salient features about time, place, and person. The relevant time is "within 24 hours of the luncheon" (not before the luncheon, not days later). The relevant place is "the luncheon" (i.e., all cases must have been in attendance at the event). The relevant person is an attendee at the lunch event (implied), and someone having symptoms—in this case "abdominal pain, vomiting, and diarrhea." The onset of abdominal pain and fever after the luncheon (D) could also be a case definition, but this definition is not as specific as to symptoms or time and therefore is not as good; it will necessarily be less specific and tend to produce more false-positives (i.e., it will tend to label more "noncases" as "cases"; see specificity and positive predictive value in Chapter 7). An elevated white blood cell count (E) could be part of a case definition, but without additional criteria for place and time and other associated symptoms, this single criterion would be quite poor as a case definition. Acute viral gastroenteritis (B) and staphylococcal food poisoning (C) are on the differential (list of possible diagnoses) for the affected guests' illness, but these are clinical syndromes with highly variable presentations and symptoms. A good case definition needs to specify precise symptoms explicitly; neither acute viral gastroenteritis nor staphylococcal food poisoning does this. 10. D. Even before developing hypotheses about the source of infection, investigators would have to characterize this outbreak by time, place, and person. In this fairly simple example, virtually all this information is in the case definition. The time and place of exposure were the lunch hour at the White House, and the cases were the symptomatic guests. With sufficient information to generate the hypothesis that this outbreak is caused by contaminated food, the investigators can begin to determine what food to implicate. To do so, they must find out what foods were eaten preferentially by the symptomatic guests. A simple interview of all attendees would be an initial step. It is also possible, however, that the outbreak of "abdominal pain, vomiting, and diarrhea" is unrelated to the food served at the luncheon. For example, the symptoms might relate to a problem with a lavatory, with only guests who used the restroom during the luncheon affected. Until a specific food cause is firmly suspected, analyzing food specimens from the luncheon in the laboratory (A), closing the kitchen (B), and examining the kitchen and interviewing the food preparers about their techniques (C) would be premature. A case control study (E) would be useful to help establish an association between the symptoms and the presumed exposure, but before such a study can be performed, the illness (i.e., specific symptoms) and the exposure (i.e., specific foods or other sources) must be established. Finding out what everyone ate is the logical starting point to see if any food cause can be implicated.

9. You decide to investigate the stimulant use-bungee jumping association further. Your new study again involves a total of 500 individuals, with 250 in each group. This time, however, you assemble the groups on the basis of their past history of stimulant use, and you prospectively determine the incidence rate of bungee jumping. You exclude subjects with a prior history of jumping. Over a 5-year period, 135 of the exposed group and 38 of the unexposed group engage in jumping. The relative risk of bungee jumping among the group exposed to stimulant use is: A. 2.1 B. 3.6 C. 4.8 D. 6 E. Impossible to determine based on the study design 10. Now considering the information provided in question 9, among bungee jumpers, what percentage of the total risk for jumping is caused by stimulant use? A. 0% B. 3.6% C. 10% D. 36% E. 72%

9. B. This is a cohort study, and the relative risk (or risk ratio) can absolutely be determined (not E). 10. E. The percentage of total risk caused by an exposure among those with the exposure is the attributable risk percent in the exposed, or AR%(exposed) . This can be calculated using the risk ratio (RR), The AR% in this case indicates that among those who use stimulants, almost three-fourths of the total risk for bungee jumping is attributable to the medication, assuming the found association is real and not an artifact of chance, bias, or confounding. Note that these data are fictitious.

10. While the pork industry lobbied aggressively against dubbing the novel H1N1 influenza virus "swine flu," substantial evidence supported that this wholly new genetic variant of influenza developed from confined animal feed operations associated with commercial pig farming. The novel H1N1 virus resulted from: A. Antigenic shift B. Antigenic drift C. Antibody shift D. Antibody drift E. Antisocial rift

A The H1N1 virus represents a major genetic change from the predecessor influenza variants from which it evolved. Reassortment of surface antigens from several strains of influenza across several host species (including birds and pigs) likely contributed to the virus's development. Such novel admixture of surface antigens resulting from the combination of existing viral strains is known as antigenic shift. Antigenic drift (B) is a more minor alteration in surface antigens resulting from mutations within a single virus strain. Antibody shift (C) and antibody drift (D) are not defined phenomenon. Antisocial rift (E) might occur between competing researchers vying for the title of world's leading influenza epidemiologist.

5. This is the tendency of a measure to be correct on average. A. Accuracy B. Alpha error C. Beta error D. Bias E. Cutoff point F. False-negative error rate G. False-positive error rate H. Positive predictive value I. Precision J. Random error K. Sensitivity L. Specificity

A. Accuracy is the ability to obtain a test result or study result that is close to the true value. Accuracy is to be distinguished from precision, which is the ability to obtain consistent or reproducible results. An accurate result may not be precise. Repeated blood pressure measurements in a group of patients might lead to a correct value of the mean for the group, even if poor technique caused wide variation in individual measurements. A precise result may be inaccurate. The same mean weight might be obtained for a group of participants on consecutive days, but if each measure were obtained with the participants clothed, all the results would be erroneously high. Accuracy and precision are desirable traits in research and diagnostic studies.

6. This is present if it is possible to conceive of an underlying mechanism by which an apparent cause could induce an apparent effect. A. Biologic plausibility B. Confounder C. Effect modifier D. External validity E. Internal validity F. Intervening variable G. Measurement bias H. Necessary cause I. Recall bias J. Sufficient cause K. Synergism

A. Before one variable (a "cause" or "exposure") can be thought to induce another variable (the "effect" or "outcome"), the relationship must pass the test of biologic plausibility (i.e., the relationship must make sense biologically and mechanistically). In determining biologic plausibility, an open mind is essential because what is implausible now may become plausible as the science of medicine advances.

1. The basic goal of epidemiologic research, especially with observational data, is to: A. Describe associations between exposures and outcomes B. Identify sources of measurement error and bias C. Establish direct causality D. Maximize external validity E. Reject the alternative hypothesis

A. Virtually all epidemiologic research involves the study of two or more groups of participants who differ in terms of their exposure to risk factors or some defined outcome. The basic goal of the research is to compare the frequency of different exposures and outcomes between groups to look for associations. Most epidemiologic studies are not sufficient to determine causality (C). Identifying sources of measurement error and bias (B) and having external validity (D) are desirable but are not basic goals of doing the research. All research sets out to reject the null hypothesis, looking for evidence in support of the alternative hypothesis (E).

4. An official from the state department of public health visits outpatient clinics and emergency departments to determine the number of cases of postexposure prophylaxis for rabies. The official's action is an example of: A. Active surveillance B. Case finding C. Outbreak investigation D. Screening E. Secondary prevention

A. Whenever a public health official visits health care delivery sites to assess the pattern of a disease or condition, the assessment constitutes active surveillance. The question does not provide enough information to determine whether an "outbreak" (C) of rabies exposure is under consideration. Case finding (B) refers to testing a group of patients for a disease that is not yet clinically apparent; an example is taking chest x-ray films or electrocardiograms in a group of patients being admitted to the hospital without cough or chest pain, perhaps before surgery. Screening (C) is an effort to identify occult (asymptomatic) disease or risk factors in a population. Secondary prevention (E) is an effort to prevent clinical consequences of a disease after the disease is identified. The administration of postexposure rabies vaccines might qualify as secondary prevention, but a public health official's review of cases in which the vaccines were given would not qualify as secondary prevention.

1. Epidemiology is broadly defined as the study of factors that influence the health of populations. The application of epidemiologic findings to decisions in the care of individual patients is: A. Generally inappropriate B. Known as clinical epidemiology C. Limited to chronic disease epidemiology D. Limited to infectious disease epidemiology E. Limited to biologic mechanisms rather than social and environmental considerations

B Clinical practice is devoted to the care of individual patients. Outcomes in individual patients in response to clinical interventions cannot be known, however, until after the interventions have been tried. The basis for choosing therapy (or for choosing diagnostic tests) is prior experience in similar patients, rather than knowledge of what would work best for the individual. The use of clinically applied statistics, probability, and population-based data to inform medical decisions is known as clinical epidemiology. Clinical epidemiology pertains to all clinical care. Its findings and applications are not limited to infectious diseases (D) or chronic diseases (C), or to biologic mechanisms (E). Social and environmental considerations are highly relevant to clinical decision making (E). Far from being inappropriate (A), the application of epidemiologic principles to patient care is fundamental to evidence-based practice and is supportive of robust clinical decisions.

9. Evaluation of which of the following potentially preventable causes of disease is most likely to raise ethical concerns? A. Dietary intake B. Genetic susceptibility C. Immunization status D. Smoking status E. Social support networks

B Currently available technology permits the identification of genetic susceptibility to some diseases (e.g., breast cancer). Because the ability to modify genetic risk is not nearly as great as the ability to recognize it, ethical concerns have been raised. Little good may come from informing people about a risk that they cannot readily modify, whereas harm, such as anxiety or increased difficulty and expense involved in obtaining medical insurance, might result. In contrast, social support networks (E), immunization status (C), dietary intake (A), and smoking status (D) all are factors that can be modified to prevent disease and are therefore less ethically problematic.

8. Attempts to eradicate a disease through widespread immunization programs may be associated with potential adverse effects. Which of the following adverse effects is correlated with the effectiveness of a vaccine? A. The emergence of resistant strains of infectious agents to which the vaccine is targeted B. The loss of the natural booster effect C. The occurrence of infection in younger age groups D. The occurrence of allergic reactions E. The risk of disorders of the autism spectrum

B When an infectious disease is common, numerous individuals become infected and develop immunity to the causative agent. When their immunity begins to wane and they come into contact with infected individuals, they are reexposed to the causative agent. This environmental reexposure boosts their immune system so that their protective immunity is maintained. The phenomenon is called the natural booster effect. One of the potential adverse effects associated with a widespread immunization program is the loss of the natural booster effect. If a vaccine is highly efficacious, it markedly reduces the chances of exposure to the causative agent in the environment. If exposure does occur after the vaccine's effect in an individual has declined with time, the vaccinated individual is at risk for infection. This is why booster vaccines are given at specified intervals. Other potential adverse effects associated with widespread immunization, such as allergic reactions (D), are rare and idiosyncratic and are unlikely to be correlated with the effectiveness of the vaccine. Widespread vaccination tends to delay exposure to the wild-type pathogen and tends to shift the disease to older age groups, rather than younger ones (C). Similar to antibiotics, vaccines are ineffective in some individuals. In contrast to antibiotics, which act by killing or arresting the growth of pathogens, vaccines act by sensitizing the immune system to pathogens and do not promote the emergence of resistant strains (A). An enormous body of high- quality evidence now refutes that there is any link between vaccines and autism spectrum disorders (E), and previous studies and authors suggesting this association have been discredited.

1. A study involves tracking a condition that can recur in individuals over time (e.g., "heartburn" or dyspepsia). Which of the following measures would allow the authors of the study to make full use of their collected data? A. Attributable risk B. Incidence density C. Period prevalence D. Point prevalence E. Proportional hazards

B Incidence density is a measure reported in terms of the frequency (density) of a condition per person-time (e.g., person-days, person-months, or person-years). It is a composite measure of the number of individuals observed and the period of observation contributed by each. For example, 10 individuals observed for 1 year each would represent 10 person-years of observation. One individual observed for 10 years would also represent 10 person- years of observation. Incidence density allows all data to be captured and reported, even when some individuals are lost to follow-up before the end of the observation period. Thus the measure allows investigators to make full use of their data without losses of information inevitable with less-inclusive metrics. Incident density is particularly useful when an event can recur in an individual during the observation period. In contrast, period prevalence (C) and point prevalence (D) fail to capture recurrent events over time because they describe events only during a given period (period prevalence) or at a given point in time (point prevalence). An individual with multiple recurrences of a condition would be indistinguishable from an individual having only a single occurrence with these metrics. Proportional hazards (E) is a statistical method used to characterize the effects of multiple variables on the risk of a binary outcome, such as survival. Attributable risk (A) expresses the extent to which a single factor is responsible for a particular outcome in a population.

1. This must be associated with the exposure and the outcome. A. Biologic plausibility B. Confounder C. Effect modifier D. External validity E. Internal validity F. Intervening variable G. Measurement bias H. Necessary cause I. Recall bias J. Sufficient cause K. Synergism

B. A confounder is a third variable that is associated with the exposure variable and the outcome variable in question but is not in the causal pathway between the two. Obscures relationship. Cigarette smoking is a confounder of the relationship between alcohol consumption and lung cancer. Cigarette smoking is associated with the outcome, lung cancer, and with the exposure, alcohol consumption, but is not thought to be involved in the pathway through which alcohol could lead to lung cancer. If an investigator were to assess the relationship between alcohol consumption and lung cancer and not take smoking into account, alcohol would be found to be associated with an increased risk of lung cancer. When the study has controls in place to account for smoking (i.e., when varying degrees of alcohol consumption are compared in participants with comparable cigarette consumption), the association between alcohol and lung cancer essentially disappears. Almost all the increased risk that seemed to be attributable to alcohol is actually caused by cigarettes. In other words, heavy drinkers might develop lung cancer, but not because they consume a lot of alcohol. They have lung cancer because they tend to smoke. The alcohol is relatively innocent when it comes to lung cancer, and the apparent association is caused by confounding. Note that if cigarette consumption did not vary with alcohol exposure, there would be no apparent relationship and no confounding.

3. This is a type I error. A. Accuracy B. Alpha error C. Beta error D. Bias E. Cutoff point F. False-negative error rate G. False-positive error rate H. Positive predictive value I. Precision J. Random error K. Sensitivity L. Specificity

B. A false-positive error is also known as a type I error or alpha error. These interchangeable designations refer to situations in which the data indicate that a particular finding (test result or study hypothesis) is true when actually it is not true. Typically, results attributable to chance or random error account for alpha error, although alpha error may result from bias as well. The value of alpha used in hypothesis testing specifies the level at which statistical significance is defined and specifies the cutoff for p values. Under most circumstances, the conventional level of alpha employed is 0.05 (see Chapter 9). The more stringent the value assigned to alpha (i.e., the smaller alpha is), the less likely that a false-positive result would occur. The clinical situation often dictates basic criteria for setting the value of alpha. A false-positive error may be more tolerable in a study involving new therapies that are desperately sought for severely ill patients than in a study involving a preventive intervention to be applied to a healthy population.

3. The members of a public health team have a continuing interest in controlling measles infection through vaccination. To estimate the level of immunity in a particular population in a quick and efficient manner, what type of study should they conduct? A. Case control study of measles infection B. Cross-sectional survey of vaccination status C. Randomized trial of measles vaccination D. Retrospective cohort study of measles vaccination E. Ecologic study of measles in the population

B. Clarifying the research question lays the foundation for selecting a specific and useful study design. In this example, the study should answer the following questions: Is there adequate immunity to measles in the population? If immunity is inadequate, how should vaccination policy be directed to optimize protection of the community? A cross-sectional survey of vaccination status (or antibody titers, or both) of community members would be the most expedient, cost-effective means to obtain answers to questions concerning the allocation of public health resources to prevent the spread of measles through vaccination. Cross-sectional surveys are often useful in setting disease control priorities. Answer choices A, C, and E are about studies investigating measles infection, not immunity. A retrospective cohort study of measles vaccination (D) might be appropriate for looking at the effectiveness of vaccination in the community (i.e., comparing measles incidence in those vaccinated and those not) but would not be appropriate to estimate the level of immunity in a particular population.

7. Which of the following measures the chance of having a risk factor? A. Kappa B. Odds ratio C. p value D. Relative risk E. Risk ratio

B. The odds ratio, usually derived from a case control study, indicates the relative frequency of a particular risk factor in the cases (i.e., participants with the outcome in question) and in the controls (i.e., participants without the outcome in question). The outcome already has occurred; the risk for developing the outcome cannot be measured directly. What is measured is exposure, which is presumably the exposure that preceded the outcome and represented a risk factor for it. The odds ratio may be considered the risk of having been exposed in the past, given the presence or absence of the outcome now. The odds ratio may be considered the risk of having the risk factor. The odds ratio approximates the risk ratio (E) when the disease in question is rare. The relative risk (D) and the risk ratio are the same measurement (the terms are interchangeable). Kappa (A) is a measure of interrater agreement. The p value (C) is a measure of statistical significance, the probability of finding an effect as great or greater by chance alone (the smaller the p value, the greater the statistical significance).

3. The risk of acquiring an infection is 300 per 1000 among the unvaccinated and 5 per 1000 among the vaccinated population. Approximately 80% of the population is exposed to the pathogen every year. Which of the following would be true? A. The absolute risk reduction is 295. B. The relative risk reduction is 5900%. C. The population attributable risk percent is 27. D. The number needed to treat is 6.2. E. The number needed to harm is 38.

B. To determine relative risk reduction, we can first calculate an absolute risk reduction (ARR). ARR = Risk(exposed) - Risk(unexposed) . Those exposed to the vaccine have a risk of 5/1000, or 0.005. Those unexposed to the vaccine have a risk of 300/1000, or 0.3. The ARR is thus 0.005 - 0.3 = -0.295 (not A). Dividing ARR by Risk(exposed) yields the relative risk reduction (RRR), or -0.295/0.005 = 59 = 5900%. In other words, vaccinated people are 59 times less likely to acquire the infection than unvaccinated people. The number needed to treat (NNT) is simply 1/ARR, or 1/-0.295, or -3.4 (not D). In other words, only four people would have to be vaccinated for one to benefit. The number needed to harm (E) cannot be calculated from the information given, since no adverse events of the vaccine are reported. Likewise, the population attributable risk percent (PAR%) cannot be calculated (C). One of two formulas can be used to determine PAR%. In the case described in the question, it is impossible to calculate the PAR% because we have not been told the proportion of the population exposed (i.e., vaccinated) for the first formula, and we do not know the Pe for the second formula. To determine the Risk(total) for the first formula, we need to know the proportion of the population vaccinated (in whom the risk of infection is 5 per 1000) and the proportion unvaccinated (in whom the risk is 300 per 1000). With exposure to a preventive measure, such as a vaccine in this case, risk is reduced among the exposed. The risk of the outcome is less among the exposed than among the unexposed, so the PAR would be negative.

2. Arizona, Colorado, and New Mexico report cases of an unexplained respiratory tract illness with a high case fatality ratio. Which of the following is most reliably true regarding this event? A. The cases represent an epidemic. B. The identification of the cases is an example of active surveillance. C. It is appropriate for the CDC to investigate the cases. D. The seemingly new cases may be an artifact of improved health department surveillance. E. If the illnesses represent an endemic disease, the cases do not constitute an outbreak.

C 2. C. The respiratory tract illness described in the question was the hantavirus pulmonary syndrome. Because fatalities were reported from more than one state, the investigation of the illness would fall within the jurisdiction of the Centers for Disease Control and Prevention. The question implies that the case reports were unsolicited (this was an "unexplained" illness that people happened to notice, not a well-established illness that people were looking for). Thus the cases would be an example of passive surveillance rather than active surveillance (B), as through improved monitoring by a health department (D). The question does not provide enough information to determine whether the reported cases constitute an epidemic (A) or an outbreak. To make this determination, the illness would need to be characterized sufficiently to ascertain whether a sudden change in the pattern of disease had occurred. Even if the condition were endemic (E), this would not exclude the possibility of an outbreak (i.e., unexpected increase in disease activity).

4. For an infectious disease to occur, there must be interaction between: A. Behavioral factors and genetic factors B. The agent and the vector C. The host and the agent D. The vector and the environment E. The vector and the host

C C. The minimum requirement for disease transmission to occur is the interaction of the agent and the host in an environment that enables them to come together. A vector of transmission (B, D, and E) may or may not be involved. Likewise, behavioral factors and genetic factors (A) may or may not be involved.

5. An example of the iceberg phenomenon would be: A. The primary prevention of colon cancer B. Giving a medicine that only partially treats an illness C. Widely publicized fatalities caused by emerging swine flu D. Conducting field trials in northern latitudes E. When cold temperatures favor disease outbreaks

C When a new disease (e.g., H1N1 swine flu) emerges, the first cases that are reported tend to be the most severe cases. Continued investigation reveals that these cases are merely the "tip of the iceberg." Many more cases that are generally less severe are hidden from view initially, just as the bulk of an iceberg lies below the water and is not seen initially. The primary prevention of colon cancer (A) may address only the "tip of the iceberg" for that disease; a medicine that treats only some of the symptoms of an illness (B) may address only the "tip of the iceberg" for that illness. However, such prevention and treatment describe different "iceberg" situations than that of the named epidemiologic phenomenon. Although cold temperatures (E) could favor certain disease outbreaks (e.g., by causing more people to spend more time indoors close to each other), such situations have little to do with the iceberg phenomenon. Conducting field trials in northern latitudes (D) could result in the sighting of icebergs, but such trials also have little to do with the iceberg phenomenon.

9. This is useful for studying trends in the causes of death over time. A. Age-specific death rate B. Case fatality ratio C. Cause-specific death rate D. Crude birth rate E. Direct standardization of death rate F. Incidence rate G. Indirect standardization of death rate H. Infant mortality rate I. Prevalence rate J. Standardized mortality ratio K. Standardized rate

C. For rates to be compared, the denominators (or populations) must be comparable. To compare events that are similar, the numerators must be defined in the same way. Cause-specific rates provide numerator data based on comparable diagnosis. Death rates may be similar in two populations, but the deaths may be caused by different conditions, with differing implications for public health management. Cause-specific rates would be useful in attempting to set priorities for funding of public health resources.

2. This alters the nature of a true relationship between an exposure and an outcome. A. Biologic plausibility B. Confounder C. Effect modifier D. External validity E. Internal validity F. Intervening variable G. Measurement bias H. Necessary cause I. Recall bias J. Sufficient cause K. Synergism

C. In contrast to a confounder, an effect modifier does not obscure the nature of a relationship between two other variables; rather, it changes the relationship. Consider the effect of age on the pharmacologic action of the drug methylphenidate. In children the drug is used to treat hyperactivity and attention-deficit disorder, both of which are conditions in which there is too much activity. In adults the same drug is used to treat narcolepsy, a condition characterized by extreme daytime somnolence and, in essence, a paucity of energy and activity. There are true, measurable, and different effects of methylphenidate on energy and activity levels in children and in adults. The effects are not confounded by age but altered by it. This is effect modification. In contrast to confounders, which must be controlled, effect modifiers should not be controlled. Instead, effect modifiers should be analyzed to enhance understanding of the causal relationship in question.

6. A potential bias in screening programs that is analogous to the late-look bias is: A. Spectrum bias, because the cases are clustered at one end of the disease spectrum B. Retrospective bias, because the severity of illness can only be appreciated in retrospect C. Length bias, because cases lasting longer are more apt to be detected D. Selection bias, because the program selects out severe cases E. Selection bias, because the program selects out asymptomatic illness

C. Length bias, the tendency to detect preferentially long-lasting cases of subclinical illness whenever screening is conducted, is analogous to late-look bias. Length bias occurs for the same reasons as late-look bias (see the answer to question 5). A screening program usually is prospective and would not be subject to any type of "retrospective bias" (B). The intent of any disease screening program is to detect asymptomatic and unrecognized cases. Selection bias (D and E) refers to the selection of individuals to participate, and screening programs may be subject to selection bias. This would limit external validity but is not a comparable effect to the late-look bias. Spectrum bias (A) is said to occur if the clinical spectrum (variety) of the patients used to measure the sensitivity and specificity of a diagnostic test differs from the clinical spectrum of the patients to whom the test would be applied, so that the values obtained are inaccurate in the test's actual use.

11. As the sensitivity increases, which of the following generally occurs? A. The cutoff point decreases B. The false-negative error rate increases C. The false-positive error rate increases D. The specificity increases E. The statistical power decreases

C. Sensitivity is the capacity of a test to identify the presence of disease in diseased individuals. The more sensitive a test is, the more likely it is to yield positive results (whether disease is actually present) and the less likely it is to yield negative results. This explains why an increase in sensitivity typically is accompanied by an increase in the false-positive error rate. An increase in sensitivity is also typically accompanied by a decrease in the false-negative error rate (B), which is why a negative result in a highly sensitive test is more likely to be a true negative. As a test's sensitivity increases, its specificity often decreases (D). Specificity is the capacity of a test to identify the absence of disease in nondiseased individuals. The more specific a test, the more readily it yields negative results (whether the disease is actually absent). The cutoff point (A) influences sensitivity and specificity and is generally chosen to strike the "optimal" balance between the two, with optimal dependent on the context and following no general trend. Mathematically, statistical power = 1 - beta. Beta is false-negative error, and (as previously noted) an increase in sensitively tends to lower false- negative error. With a smaller beta, statistical power (E) would increase.

13. The clinicians in a primary care clinic do not know the prevalence of Chlamydia trachomatis infection in their community. In screening patients for this infection, they plan to use a test that performed with a sensitivity of 75% and a specificity of 75% in clinical trials. When the clinicians use the test to screen patients for C. trachomatis in their own community, which of the following could they use to help them interpret a positive test result? A. Kappa B. Phi C. LR+ D. Odds ratio E. RR 14. What is the value for the measure specified in question 13? A. 2.6 B. 3 C. 5 D. 8.4 E. 161.

C. The likelihood ratio for a positive test result (LR+) is calculated as the test's sensitivity divided by its false-positive error rate (1 - specificity). The LR+ can be used to estimate the reliability of a positive test result when the prevalence of disease is unknown. It is useful in this context because the components of the LR+ are contained within the columns of a 2 × 2 table and are independent of disease prevalence. The odds ratio (D) is used to evaluate the outcome of a case control study. The risk ratio, or RR (E), is used to evaluate the outcome of a cohort study. Kappa measures agreement between observers (A), and phi measures the strength of association of a chi-square value (B). B. The likelihood ratio for a positive test result (LR+) is a ratio of the sensitivity of the test to the false-positive error rate. The sensitivity is 0.75. The false-positive error rate = (1 - specificity) = (1 - 0.75) = 0.25. Thus the LR+ is 0.75/0.25, or 3.

9. In a case control study that is being planned to study possible causes of myocardial infarction (MI), patients with MI serve as the cases. Which of the following would be a good choice to serve as the controls? A. Patients whose cardiac risk factors are similar to those of the cases and who have never had an MI in the past B. Patients whose cardiac risk factors are similar to those of the cases and who have had an MI in the past C. Patients who have never had an MI in the past D. Patients whose cardiac risk factors are dissimilar to those of the cases and who have had an MI in the past E. Patients whose cardiac risk factors are unknown and who have had an MI in the past

C. The usual goal of a case control study is to determine differences in the risk factors seen in the participants with a particular outcome (which in this example is myocardial infarction) and the participants without the outcome. If the two groups of participants were matched on the basis of risk factors for the outcome (i.e., had similar cardiac risk factors, as in answer choices A and B), the potentially different influences of these factors would be eliminated by design. Matching on the basis of known (established) risk factors to isolate differences in unknown (as yet unrecognized) risk factors is often appropriate. However, if the cases and controls resemble one another too closely, there is a risk of overmatching, with the result that no differences are detectable, and the study becomes useless. If the case control study is used to measure the effectiveness of a treatment, it would be appropriate for the cases and controls to be similar in risk factors for the disease. In trying to elucidate possible causes (i.e., possible risk factors) for MI, however, participants should be dissimilar in risk factors so that differences can be explored in association with the outcome. Recruiting patients with unknown (and thus potentially similar) risk factors (E) would be undesirable for this reason. Likewise, it would be undesirable to include people who have ever had an MI in the past (B, D, and E), because they have the outcome of interest; it is impossible to see what factors are associated with having an MI versus not having an MI when everyone has had an MI.

3. Before quitting smoking, Tim, his cigarettes, and his tobacco smoke represent: A. Agent, host, and environment, respectively B. Agent, environment, and vector, respectively C. Vector, agent, and vehicle, respectively D. Host, vehicle, and agent, respectively E. Vehicle, vector, and agent respectively

D D. A vehicle is an inanimate carrier of an agent of harm (e.g., the gun for a bullet, the syringe for heroin, the cigarette for tobacco smoke). Host, vector, and agent are defined in the explanation to question 2. Environment is the context and conditions in which the host, agent, and vehicle or vector interact. Given these explanations, answer choices A, B, C, and E miss the mark.

7. Herd immunity refers to: A. Immunity acquired from vaccines developed in herd animals B. Immunity naturally acquired within confined herds of animals or within overcrowded human populations C. The high levels of antibody present in a population after an epidemic D. The prevention of disease transmission to susceptible individuals through acquired immunity in others E. The vaccination of domestic animals to prevent disease transmission to humans

D Herd immunity not only prevents the immunized individuals in a population from contracting a disease, but it also prevents them from spreading the disease to nonimmunized individuals in the population. Herd immunity prevents disease transmission to susceptible individuals as a result of acquired immunity in others. The characteristics of a particular infection and a particular population determine the level of prevailing immunity required to limit the spread of a disease. The presence of a highly infectious illness (e.g., measles) in a population with multiple exposures (e.g., college students) requires that nearly everyone be immunized to prevent transmission. Herd immunity does not refer to herd animals (A), herds of animals (B), or vaccinating animals to prevent disease transmission to humans (E). It likewise does not refer to overcrowded populations of humans (B) or the high levels of antibodies present in a population after an epidemic (C).

5. An article highlighting the long-term consequences of inadequately treated Lyme disease is published in a medical journal. After a summary of the article appears in popular newspapers and magazines, patients with vague joint pains begin insisting that their physicians test them for Lyme disease. Cases in which the test results are positive are reported as cases of Lyme borreliosis. This represents: A. Outbreak investigation B. An epidemic of Lyme borreliosis C. A change in reporting that would underestimate incidence D. A change in surveillance that would overestimate the prevalence E. A change in screening that would underestimate the likelihood of an outbreak

D Whenever public attention is focused on a particular health problem, an increase in the number of reported cases of the problem is likely. However, even when available diagnostic tests are almost perfect (with sensitivities and specificities approaching 100%; not the case for Lyme disease tests), if the prevalence of the actual disease in the tested population is low, the majority of positive test results will be false positive (see Chapter 13, Bayes theorem). Thus the number of false-positive cases of Lyme disease will necessarily increase as more people with vaguely characterized clinical syndromes are tested. This increase in "Lyme disease" reporting represents an epidemic of Lyme disease testing, not an epidemic of the disease itself (B). The increased testing does not represent an outbreak investigation (A), because the question does not specify what the expected patterns of the disease are or suggest that the increased testing is being performed to investigate new and unusual patterns in disease. Increased testing would tend to overestimate the likelihood of an outbreak (E) and overestimate prevalence, not underestimate incidence (C).

10. This is differential error. A. Accuracy B. Alpha error C. Beta error D. Bias E. Cutoff point F. False-negative error rate G. False-positive error rate H. Positive predictive value I. Precision J. Random error K. Sensitivity L. Specificity

D. Bias is differential error because it distorts data in a particular direction (e.g., weighing participants after a meal, taking blood pressure readings after caffeine ingestion or exercise, performing psychometric testing after alcohol ingestion). In contrast to bias, random error is nondifferential error because random error is equally likely to produce spuriously high or spuriously low results.

2. Studies may be conducted to generate or test hypotheses. The best design for testing a hypothesis is a: A. Case control study B. Cross-sectional survey C. Longitudinal ecologic study D. Randomized controlled trial E. Retrospective cohort study

D. Randomized controlled trials (including RCFTs and RCCTs) are experimental studies and represent the gold standard for hypothesis testing. These studies are costly in time and money, however. They are best reserved for testing hypotheses that already are supported by the results of prior studies of less rigorous design (i.e., observational studies). Often the RCT is the final hurdle before a hypothesis is sufficiently supported to become incorporated into clinical practice. (Unfortunately though, RCTs sometimes produce results opposite what would be expected given observational evidence.) Observational studies such as case control studies (A), cross-sectional surveys (B), longitudinal ecologic studies (C), and retrospective cohort studies (E) are appropriate for hypothesis generation and, sometimes (depending on their design and purpose) for testing certain hypotheses. For none of these designs, however, is it ever appropriate both to generate and to test the same hypotheses in any single study.

5. When the study sample adequately resembles the larger population from which it was drawn, the study is said to have this. A. Biologic plausibility B. Confounder C. Effect modifier D. External validity E. Internal validity F. Intervening variable G. Measurement bias H. Necessary cause I. Recall bias J. Sufficient cause K. Synergism

D. The external validity of a study is determined by the resemblance of the study population to the larger population from which it was drawn. In a well-designed study of antihypertensive therapy in the prevention of stroke in middle- class white men, the validity of the findings for middle-class white women and for men and women of other socioeconomic and ethnic groups is uncertain. Although internal validity defines whether a study's results may be trusted, external validity defines the degree to which the results may be considered relevant to individuals other than the study participants themselves. Another term used to discuss external validity is generalizability. The greater a study's external validity, the more applicable or generalizable it is to populations outside of the study population.

12. Two radiologists interpret 100 mammograms. They agree that the results are normal in 70 mammograms and abnormal in 12 mammograms. In the remaining 18 cases, the first radiologist thinks that results are normal in 6 mammograms and abnormal in 12 mammograms, whereas the second radiologist thinks just the opposite. The value of an appropriate measurement of their agreement that takes into consideration chance agreement is: A. 6% B. 16% C. 26% D. 46% E. 86%

D. To assess the measure of agreement, kappa, the investigator should begin by setting up a 2 × 2 table and placing the number of "abnormal" (positive) results and "normal" (negative) results in cells a, b, c, and d: N is the sample size; Ao is the observed agreement; and Ac is the total agreement attributable to chance, which is equal to the sum of cell a agreement expected by chance plus cell d agreement expected by chance. N is given as 100. Ao is calculated as a + d, so here it is 12 + 70 = 82.

6. Which of the following is beyond the scope of activities undertaken by epidemiologists? A. Analyzing cost effectiveness B. Establishing modes of disease transmission C. Studying how to prevent disease D. Providing data for genetic counseling E. Rationing health care resources

E Although the responsibilities of epidemiologists may include analyzing cost effectiveness (A) and recommending allocation of health care resources, the rationing of these resources is a political task and falls outside the purview of epidemiologists. Establishing modes of disease transmission (B) and identifying means of preventing disease spread (C) are integral aspects of epidemiology. Genetic epidemiologists participate in genetic counseling (D).

1. An outbreak of disease should be reported to the local or state health department: A. Only if the diagnosis is certain B. Only if the disease is infectious C. Only if the disease is serious D. Only if the outbreak involves at least 10 people E. Always

E E. By definition, an outbreak represents a deviation from the expected pattern of health and disease in a population. Any outbreak may reveal a new health threat or point to the breakdown of some aspect of the system designed to protect health in the population. Any seeming deviation from the expected pattern of disease (i.e., any apparent outbreak) should always be reported, regardless of whether the diagnosis is certain (A). Waiting for definitive diagnoses to confirm suspicion may cause dangerous delay; initial suspicions will suffice to initiate the appropriate investigations toward characterization and control of the threat. Outbreaks of infectious (B) and noninfectious disease may be equally dangerous and important, and a seemingly mild disease may represent a serious health threat to certain vulnerable populations. Thus all unusual disease patterns should be reported, even if the disease does not seem serious (C) and even if only a few people are affected. Indeed, for rare diseases, even a single case might be cause enough for alarm (e.g., smallpox). Having at least 10 affected individuals (D) is not necessary.

3. Cases of "flesh-eating" group A streptococcal disease are reported in a defined population. Which of the following types of information would be most helpful for determining whether these cases represent a disease outbreak? A. The clinical features and methods of diagnosing the disease B. The disease vector and reservoir C. The exact location and timing of disease onset D. The incubation period and pattern of disease transmission E. The usual disease patterns and reporting practices

E E. Steps in the evaluation of an outbreak include establishing a diagnosis; developing a case definition; determining whether an outbreak exists; characterizing the outbreak by time, place, and person; developing and testing hypotheses; initiating control measures; and initiating follow-up surveillance. To complete all these steps, considerable information about the disease in question is required. An early determination about the probability of an outbreak depends most, however, on knowing the usual disease patterns and knowing the reporting practices that bring the disease to attention. In this case, if group A streptococcal fasciitis generally does not occur at all in the population, the occurrence of even one correctly reported case might represent an outbreak. Because the disease is severe, the regular occurrence of unreported cases would be unlikely, so even without active surveillance, reporting probably would be fairly reliable and complete. Although the information provided is extremely limited, this scenario would strongly suggest the occurrence of an outbreak. The clinical features and methods of diagnosing the disease (A) and exact location and timing of disease onset (C) are important for establishing the case definition and may contribute to the degree of confidence with which one determines "cases." The disease vector and reservoir (B) and the incubation period and pattern of disease transmission (D) are important in understanding a disease process and may contribute to the case definition and hypotheses about causes.

2. Tim has a severe heart attack at age 58. The near-death experience so scares Tim that he quits smoking. Tim's wife is also scared into quitting smoking even though she feels fine. Tim's son resolves never to start smoking, seeing what cigarettes have done to his dad. The act of not smoking for Tim, Tim's wife, and Tim's son represents: A. Host, vector, and agent effects, respectively B. Herd immunity C. Tertiary prevention for Tim's son D. Tertiary prevention, primary prevention, and secondary prevention, respectively E. Tertiary prevention, secondary prevention, and primary prevention, respectively

E Prevention means intervening to interrupt the natural history of disease. Primary prevention represents the earliest possible interventions to foil disease before it even begins (e.g., Tim's son never starting to smoke). Secondary prevention is thwarting the progression of established disease that has not yet declared itself with symptoms or outward signs (e.g., Tim's asymptomatic wife quitting smoking). Tertiary prevention is slowing, arresting, or even reversing obvious or symptomatic disease to prevent worsening symptoms and further deterioration (e.g., Tim quitting smoking after his heart attack). Answer choices C and D get distinctions between these stages of disease prevention wrong. Herd immunity (B) is the prevention of disease transmission to susceptible individuals through acquired immunity in others. Herd immunity does not apply to the scenario involving Tim and his family. Likewise, definitions for host (i.e., susceptible individual), vector (i.e., unaffected carrier), and agent (i.e., medium of harm) (A) do not apply to the characterizations of Tim and his family.

5. This is calculated after the two populations to be compared are "given" the same age distribution, which is applied to the observed age-specific death rates of each population. A. Age-specific death rate B. Case fatality ratio C. Cause-specific death rate D. Crude birth rate E. Direct standardization of death rate F. Incidence rate G. Indirect standardization of death rate H. Infant mortality rate I. Prevalence rate J. Standardized mortality ratio K. Standardized rate

E In direct standardization the age-specific death rates (ASDRs) are available for the populations to be compared. The age distribution of a hypothetical "standard" population (often consisting of the sum of the populations under comparison) is derived. The ASDRs from each of the populations are applied to the hypothetical age distribution, and these summary rates may be compared because they are free of age bias. Crude death rates may be low in developing countries because the population is relatively young; standardization helps correct age bias to better represent the data.

9. The closer this is to the upper left corner of an ROC curve, the better it is. A. Accuracy B. Alpha error C. Beta error D. Bias E. Cutoff point F. False-negative error rate G. False-positive error rate H. Positive predictive value I. Precision J. Random error K. Sensitivity L. Specificity

E. A cutoff point is used to distinguish normal from abnormal test results. If an upper-limit cutoff point is set too high, the normal range includes many participants with disease. If the upper- limit cutoff point is set too low, many normal participants are said to have "abnormal" test results. The optimal cutoff point, where all cases of disease are detected with no false-positive results, is represented by the upper left corner of the ROC curve. Such a point virtually never exists and is at best approximated. The closer the cutoff point is to the upper left corner of an ROC curve, the greater the sensitivity, and the lower the false-positive error rate, for a given test.

5. Cross-sectional surveys are subject to the Neyman bias, or late-look bias. This may be explained as the tendency to: A. Detect only the late stages of a disease, when manifestations are more severe B. Detect only the cases of a disease that are asymptomatic C. Find more disease in older cohorts D. Detect fatalities preferentially E. Detect the more indolent cases of a disease preferentially

E. At any given moment, the prevalence of disease in a population is influenced by the incidence of disease (the frequency with which new cases arise) and the duration. Diseases of long duration are more apt to accumulate in a population than are diseases that run a short course (resulting in either full recovery or death). Even within the categories of a particular illness, such as prostate cancer, the prevalence of the more indolent cases (i.e., the slowly progressive cases) is likely to be higher than the prevalence of the aggressive cases. This is because the indolent cases accumulate, whereas the aggressive cases result in a rapid demise (and death removes people who would otherwise be considered from investigator consideration). A cross-sectional survey preferentially detects the indolent cases that tend to accumulate and misses the cases that have recently occurred but already resulted in death. This is the late-look bias: One is looking too late to find aggressive cases that already have led to death. Thus because of late-look or Neyman bias, investigators will preferentially miss (not detect) fatalities (D) and severe, late-stage disease (A). The age of the cohorts (C) or symptoms of disease (B) are not relevant concerns.

4. This defines normal and abnormal test results. A. Accuracy B. Alpha error C. Beta error D. Bias E. Cutoff point F. False-negative error rate G. False-positive error rate H. Positive predictive value I. Precision J. Random error K. Sensitivity L. Specificity

E. In clinical medicine, there are relatively few pathognomonic findings (findings considered definitive of a particular disease and expressed in dichotomous terms as the presence or absence of the disease). Most tests produce results that do not indicate the presence or absence of disease with absolute certainty. The cutoff point indicates the value beyond which a test result is considered abnormal, leading to a diagnosis or suggesting the need for further testing. The cutoff point chosen is influenced by the situational priorities. An initial screening test should be highly sensitive, so the cutoff point for the upper limit of normal values should be set low. A follow-up test should be highly specific, so the cutoff point for the upper limit of normal values should be set high.

4. A published study purported to show that a variety of symptoms were more common among participants with a history of suboptimally treated Lyme disease than among controls who had no history of Lyme disease. The data were obtained largely by a survey of the participants. The study is most likely subject to which of the following distortions? A. Ecologic fallacy, length bias, and lead-time bias B. Intervention bias, random error, and length bias C. Late-look bias, measurement bias, and length bias D. Lead-time bias, late-look bias, and selection bias E. Selection bias, recall bias, and random error

E. In testing study hypotheses, bias must be diligently avoided. Although participants often can be randomized in a prospective study, they cannot usually be randomized in a retrospective study, such as the study on Lyme disease described in the question. In almost any retrospective study, there is a risk of selection bias (i.e., if participants are allowed to choose, those who most believe their symptoms are related to the exposure in question may want to participate). In the quantification of symptoms, many of which are subjective, measurement bias is possible (i.e., the firm believers in the cause of their symptoms may report their symptoms as being more severe). Additionally, in any retrospective study in which participants are surveyed, the category that defines their role in the study (as a member of the "exposed" group or member of the "nonexposed" group) may influence their recall of relevant events. In the study described, the participants with a history of Lyme disease might be more likely to recall symptoms suggestive of the disease or its late complications than would the controls, who have no history of disease and might dismiss or fail to remember episodes of minor joint pain. Recall bias is one of the most important limitations of case control studies. Random error is possible in any study and is never completely avoided. Lead- time bias is germane only to screening programs and the study of the time course of a disease, such as the time between the diagnosis and the outcome (e.g., death). Lead-time bias is the tendency for early detection to increase the interval between detection and the outcome without altering the natural history of the disease. Length bias refers to the tendency to detect preferentially more indolent cases of disease in a population screening program; it is not germane to the study in question. Similarly, late-look bias, the tendency to detect preferentially less severe cases of disease in a population after patients with severe disease have been removed from consideration (e.g., through death), is also not germane to the study question. The ecologic fallacy refers to the tendency to presume that an association found in a population (e.g., frequency of an exposure and frequency of a disease) is operative in individuals. Because the study in question is a study of individuals, the ecologic fallacy is irrelevant. The study is not an intervention study, however, and "intervention bias" is not an established form of bias. Any form of bias can invalidate the findings of a study. Thus investigators must be vigilant in preventing bias (which can generally only be addressed prospectively—i.e., before a study is done). Given the explanations here, answers A, B, C, and D are incorrect.

8. A case control study may have a particular advantage over a cohort study when the disease in question is: A. Fatal B. Indolent C. Infectious D. Virulent E. Rare

E. The groups in a case control study are assembled on the basis of the outcome. If the outcome is rare, this design is particularly advantageous. Risk factors in a defined group with the rare outcome can be assessed and compared with risk factors in a group without the outcome. Two potential problems of conducting a cohort study for a rare disease are that too few cases would arise to permit meaningful interpretation of the data and that a very large sample size would be required, resulting in great (often prohibitive) expense. Considerations of fatality (A), indolence (B), infectiousness (C), and virulence (D) are not relevant.

11. This is present when study results are obtained in an unbiased manner. A. Biologic plausibility B. Confounder C. Effect modifier D. External validity E. Internal validity F. Intervening variable G. Measurement bias H. Necessary cause I. Recall bias J. Sufficient cause K. Synergism

E. The most important criterion on which a study is judged is its internal validity. If study results are obtained in an unbiased manner, the study is said to have internal validity. If the study is biased, it lacks internal validity, and its results are unreliable and meaningless. Consider, hypothetically, that in a study of prostate cancer, the outcome in men treated with orchiectomy (surgical removal of testes) is found to be worse than the outcome in men treated with orange juice. If men debilitated by illness had been assigned to orchiectomy, whereas men with no overt signs of illness had been assigned to orange juice therapy, the better outcome seen with orange juice therapy would be invalid because of the biased design of the study. Internal validity can be present only if bias is eliminated. Even if results are internally valid, however, it may not be possible to generalize results. To generalize, the investigator must have externally valid results, as discussed in answer 5.

6. This is the number of new cases over a defined study period, divided by the midperiod population at risk. A. Age-specific death rate B. Case fatality ratio C. Cause-specific death rate D. Crude birth rate E. Direct standardization of death rate F. Incidence rate G. Indirect standardization of death rate H. Infant mortality rate I. Prevalence rate J. Standardized mortality ratio K. Standardized rate

F Incidence, or incident cases, is merely the number of new cases. To generate a rate, the number of new cases over a specified period is divided by the population at risk at the midpoint of the study period (recall the differences between risks and rates) and multiplied by a constant, such as 1,000 or 100,000, to facilitate expression.

7. This is a means or the means by which the causal factor leads to the outcome. A. Biologic plausibility B. Confounder C. Effect modifier D. External validity E. Internal validity F. Intervening variable G. Measurement bias H. Necessary cause I. Recall bias J. Sufficient cause K. Synergism

F. An intervening variable, or mediator or mediating variable, often represents an important mechanism by which an initial causal variable leads ultimately to a particular outcome. High-fructose corn syrup is related to obesity, but the connection between the two is largely indirect. For example, studies find fructose blunts levels of leptin, a hormone that regulates appetite, which can lead to overeating and weight gain. In this case, leptin is an intervening or mediating variable in one causal pathway between fructose and weight gain. Fructose also is shuttled directly to the liver, where it is converted to triglycerides (fat), ultimately driving insulin resistance, hyperinsulinemia (high insulin in the blood), and weight gain. Triglycerides and insulin are intervening or mediating variables in another causal pathway between fructose and weight gain.

7. This is used if age-specific death rates are unavailable in the population whose crude death rate is to be adjusted. A. Age-specific death rate B. Case fatality ratio C. Cause-specific death rate D. Crude birth rate E. Direct standardization of death rate F. Incidence rate G. Indirect standardization of death rate H. Infant mortality rate I. Prevalence rate J. Standardized mortality ratio K. Standardized rate

G When the age-specific death rates are unknown for the populations to be compared, direct standardization of rates is not feasible. Indirect standardization applies the death rates from a reference (e.g., US) population to the study populations. The reference rates are applied to the age distribution of the study populations, and the number of deaths expected in each group is calculated by assigning the reference population mortality rates to the study population. The actual number of deaths in the study groups is compared with the expected number of deaths if they had the same age-specific death rates as the reference population.

8. This is absolutely required for a disease to occur, but it will not necessarily produce the disease. A. Biologic plausibility B. Confounder C. Effect modifier D. External validity E. Internal validity F. Intervening variable G. Measurement bias H. Necessary cause I. Recall bias J. Sufficient cause K. Synergism

H. A necessary cause is a factor that is required for a disease to occur. A necessary cause does not invariably lead to the disease, but the disease will certainly not occur unless the necessary cause is present. In the case of an infectious disease, exposure to a pathogen is always a necessary cause. Some people exposed to the pathogen may fail to acquire the disease, however, because of robust immunity, limited exposure, or other factors. Thus exposure to a pathogen is necessary but not sufficient. Necessary cause can be applied more broadly than disease scenarios. For example, impact between two vehicles is necessary to have a fatal head-on collision but is not necessarily sufficient to produce the fatal result. The speed of the crash, the angle of the collision, whether the drivers were restrained, and whether air bags deployed, all influence the ultimate result.

10. This is often used as an overall index of the health status of a country. A. Age-specific death rate B. Case fatality ratio C. Cause-specific death rate D. Crude birth rate E. Direct standardization of death rate F. Incidence rate G. Indirect standardization of death rate H. Infant mortality rate I. Prevalence rate J. Standardized mortality ratio K. Standardized rate

H. The infant mortality rate (IMR) is influenced by various aspects of maternal and fetal care, including nutrition, access to prenatal medical care, maternal substance abuse, the home environment, and social support networks; it is often used as an overall index of a country's health status. If infants are able to thrive, the overall health of a nation is thought to be adequate, whereas if infants are failing, and the resulting IMR is high, and the overall health of a nation is thought to be poor.

8. This is a nondifferential error. A. Accuracy B. Alpha error C. Beta error D. Bias E. Cutoff point F. False-negative error rate G. False-positive error rate H. Positive predictive value I. Precision J. Random error K. Sensitivity L. Specificity

J. Random error is nondifferential error because it does not distort data consistently in any one direction. Random error may produce some measurements that are too high and others that are too low. The mean may or may not be distorted by random error. Random error affects precision. Bias is differential error because it produces measurements that are consistently too high or too low. Bias affects accuracy.

9. This is a systematic distortion in outcome assessments in retrospective studies; it is eliminated by a prospective design. A. Biologic plausibility B. Confounder C. Effect modifier D. External validity E. Internal validity F. Intervening variable G. Measurement bias H. Necessary cause I. Recall bias J. Sufficient cause K. Synergism

I. Recall bias is a systematic distortion found especially in retrospective studies. For example, in a case control study of congenital anomalies, the mothers of children born with any congenital anomaly might be more likely to recall being exposed during pregnancy to toxic substances (e.g., medications) than would the mothers of children born without congenital anomalies. This differential recall produces systematic error. In a prospective study, exposure is established at enrollment, before the subjects are distinguished on the basis of outcome (indeed, before the outcomes have occurred). Any differences in recall between groups would therefore likely be random, not systematic. Recall bias is minimized in a prospective study.

10. This is sufficient to produce disease. A. Biologic plausibility B. Confounder C. Effect modifier D. External validity E. Internal validity F. Intervening variable G. Measurement bias H. Necessary cause I. Recall bias J. Sufficient cause K. Synergism

J. A sufficient cause is one that, if present, is all that is necessary to cause a particular disease or condition. However, a sufficient cause need not be necessary. For example, menopause is sufficient to cause infertility, but is not necessary to cause infertility. Hysterectomies, genetic disorders, and anatomic abnormalities are also sufficient to cause infertility but likewise not necessary.

8. This is the observed total deaths in a population, divided by the expected deaths in that population, multiplied by 100. A. Age-specific death rate B. Case fatality ratio C. Cause-specific death rate D. Crude birth rate E. Direct standardization of death rate F. Incidence rate G. Indirect standardization of death rate H. Infant mortality rate I. Prevalence rate J. Standardized mortality ratio K. Standardized rate

J. The standardized mortality ratio is often derived from indirect standardization methods. The observed number of deaths in a population is divided by the expected number of deaths in that population, based on the reference population. The numerator and denominator usually are multiplied by 100 to avoid describing the ratio as a fraction.

2. This is the ability of a test to detect a disease when it is present. A. Accuracy B. Alpha error C. Beta error D. Bias E. Cutoff point F. False-negative error rate G. False-positive error rate H. Positive predictive value I. Precision J. Random error K. Sensitivity L. Specificity

K. Sensitivity is calculated as the number of true cases of the disease detected by the test (cell a in a 2 × 2 table) divided by all true cases (cell a + cell c).

6. This is calculated as a/(a + c). A. Accuracy B. Alpha error C. Beta error D. Bias E. Cutoff point F. False-negative error rate G. False-positive error rate H. Positive predictive value I. Precision J. Random error K. Sensitivity L. Specificity

K. Sensitivity is defined as the ability of a test to detect true cases of a disease when disease is present. Sensitivity is calculated as the number of true cases of the disease detected by the test (cell a in a 2 × 2 table) divided by all true cases (cell a + cell c).

4. This is a multiplicative effect between exposure variables. A. Biologic plausibility B. Confounder C. Effect modifier D. External validity E. Internal validity F. Intervening variable G. Measurement bias H. Necessary cause I. Recall bias J. Sufficient cause K. Synergism

K. When the combined effect of two or more variables on an outcome is greater than the sum of the separate effects of the variables, their interaction is called synergy or synergism. Cigarette smoking and asbestos exposure have a synergistic effect on the risk of lung cancer. If the relative risk or risk ratio for lung cancer in smokers is X and if the relative risk for lung cancer in asbestos workers is Y, the relative risk in those with both exposures is closer to X × Y than to X + Y.

7. This is the ability of a test to exclude a disease when it is absent. A. Accuracy B. Alpha error C. Beta error D. Bias E. Cutoff point F. False-negative error rate G. False-positive error rate H. Positive predictive value I. Precision J. Random error K. Sensitivity L. Specificity

L. Specificity is the capacity of a test to show positive results only in participants who have the disease in question. Alternatively stated, specificity is the ability of a test to exclude a disease when it is absent. A test that is highly specific is a good "rule-in" test, because it reliably indicates the presence of the disease when the result is positive. Specificity is calculated as the number of participants with true-negative test results (cell d in a 2 × 2 table) divided by the total number of participants who do not have the disease (cell b + cell d).