Errata and comments on
Ann Aschengrau and George Seage III
Essentials of Epidemiology in Public Health
Sudbury MA: Jones and Bartlett
3r edition, 2013 (http://www.jblearning.com/catalog/9781284028911/)


The following errata have been noted:

  • Practice problem 6:
    The answers to A and D are given as "true". In question 6A, that statement can be read (1) as referring to typical risks of CHD and lung cancer and typical risk ratios for smoking or (2) as a general statement that a disease with a much higher incidence will necessarily lead to a greater risk difference. The latter proposition is not true in general, although it would often be the case. Since the text lists the answer as true, I assume the authors mean for us to use typical incidences (e.g., about 0.01 per year for CHD in male nonsmokers aged 60+ years (my guess) and about 0.0001 for lung cancer in male nonsmokers aged 60+ years - also a guess). Typical RR’s for heavy male smokers would be about 1.5 for CHD and about 10.0 for lung cancer. With these numbers, the risk difference for CHD (1.5 x 0.01 - 0.01 = 0.005) is greater than that for lung cancer (10 x 0.0001 – 0.0001 = 0.001 – 0.0001 = 0.0009). To have it come out the other way with these incidences, you’d need risk ratios like 1.1 for CHD (risk difference 0.011-0.01 = 0.001) and 15 for lung cancer (risk difference 0.0015 – 0.0001 = 0.0014).

    In quesiton 6D, “excess risk” is defined at the top of page 68 (“excess relative risk”) as RR-1. So if excess risk = RR-1 = 15%, then RR=1.15. The percentage of cases among persons exposed to unchlorinated water that are attributable to the lack of chlorination (or that would be prevented by avoiding unchlorinated water) would be (RR-1)/RR = 0.15/1.15 = 0.13, or 13%.
  • Chapter 6
    In the middle or page 144, the text reads "... it is perfectly ethical to conduct an observational study by commparing women who choose to drink during pregnancy with those who decide not to do so." Though that may be the case for some circumstances, many observers would argue that, as a minimum, if investigators have direct interaction with the women they should advise the women about the dangers to the baby from alcohol consumption during pregnancy. The scenario is not unlike a cohort study of HIV transmission in discordant couples, where it would certainly be required to counsel study participants about the importance of using condoms to prevent infection.

    At the top of page 145, "minimal resources" should probably be "minimum resources". I would also question the characterization of bias as "an error committed by the investigator in the design or conduct of a study that leads to a false association between the exposure and disease." Bias can lead to under- or over-estimation of an association, and bias can result even in the absence of investigator error. For example, most studies involve bias toward the null due the limitations of measurement instruments and limits on accuracy of recall.

    The last sentence of the first paragraph on page 145 is confusing to me. Random error can affect observed results in various ways, and is not particularly likely to "lead to a false association".

Chapter 8, Cohort Studies

The following are comments, rather than errata.

  • Chapter 8, Page 205, "natural experiments":
    A&S appear to apply the term "natural experiments" to most any cohort study "because the investigator acts as a disinterested observer merely letting nature take its course." In my experience (e.g., John Last's Dictionary of Epidemiology, 3rd ed), "natural experiment" refers to a situation where the assignment of exposure appears to be effectively random. The classic example is John Snow's study in which he compared death rates from cholera between households whose water was supplied by Southwark and Vauxhall Company or from the Lambeth Company (see page 18-21 in A&S).

  • Chapter 8, Page 206, first paragraph:
    Many epidemiologists would regard the statement "it is perfectly ethical to conduct an observational study by comparing women who choose to smoke during pregnancy with those who choose not to do so" as correct only if the investigators recommended that the study participants quit smoking and offered resources and/or referrals (e.g., see Norma Kanarek and Marty S. Kanarek, Smoking cessation in clinical trials and public health studies: a research ethical imperative. Ann Epidemiol Dec 2007;17(12):983-987). Similarly, both observational and experimental studies of HIV risk have traditionally counseled participants to use condoms even though fewer transmissions means less study power or having to recruit more participants.

  • Chapter 8, Page 206, 2nd paragraph:
    "A classic cohort study examines ... a single exposure." I've seen similar statements in other sources, but I'm not sure that the concept is a useful one. Some cohort studies are defined on the basis of exposure status, but other cohort studies, including some "classics" like the Framingham Study and other cardiovascular disease cohort studies conducted inthe mid-20th century (e.g., Evans County, Tecumseh, Chicago Western Electric, ...) were created from communities or occupational settings rather than defined by "an exposure". Baseline examinations in these studies were extensive, providing data on a large number of "exposures". In my experience, what defines a cohort study is the identification of a collection of people who are at risk for the occurrence of one or more outcomes of interest and the follow-up of these people to detect such occurrences. A key characteristic is that exposures are ascertained at or before the beginning of the follow-up period, so that there is some confidence that the measured exposure(s) reflect the person's status before the disease has occurred. Often, physiologic specimens are stored, permitting later identification of exposures that can nevertheless be linked to the pre-disease condition. There is a class of cohort studies (A&S appear to call these special cohorts - page 214) that are designed around a group of people with an exposure. An example is studies of asbestos workers. The exposures studied in these instances are often rare in the general population, so that the identification of exposed persons is a fundamental step in designing the study.

  • Chapter 12 (Random Error), Pages 331:
    Reads: "SD is the standard deviation of the sample mean" (p331, end of next-to-last paragraph)
    Actually, SD represents the standard deviation of the population. A standard error is the standard deviation of the distribution of a statistical estimator, such as a mean, proportion, rate, risk difference, or risk ratio. If such an estimator is calculated repeatedly for a large number of samples from the same population, then the estimator will have a probability distribution whose mean provides the best estimate of the population value. The standard deviation of that distribution is the standard error of the estimator. When the estimator is the mean, then that standard deviation is called the standard error of the mean, SEM. The formula for SEM is the population standard deviation divided by the square root of the sample size. When the population standard deviation is unknown, it is estimated from the sample. Similarly, the standard error of a proportion (e.g., a prevalence) estimated from a sample of size n is the square root of p(1-p)/n. The standard error describes how far away the proportion calculated from the sample is likely to be from the population proportion it estimates.
  • Chapter 12 (Random Error), Confidence interval example calculation on page 331:
    The 95% confidence limits for the cumulative incidence of mortality among subjects from Steubenville are given as 20.4% to 22.6% in the last paragraph of page 331. The correct values are 19.3% to 23.7%. If one calculates the confidence interval but forgets to multiply the standard error by 1.96, one obtains results very close to the values in the text. Thanks to Vanessa Miller and Christine Owre for spotting the error.
  • Chapter 12 (Random Error), Confidence interval example calculation on page 331:
    The 95% confidence limits for the cumulative incidence of mortality among subjects from Steubenville are given as 20.4% to 22.6% in the last paragraph of page 331. The correct values are 19.3% to 23.7%. If one calculates the confidence interval but forgets to multiply the standard error by 1.96, one obtains results very close to the values in the text. Thanks to Vanessa Miller and Christine Owre for spotting the error.
  • Chapter 12 (Random Error), Pages 334-335:
    In the first formula on page 334 (tdf = ...) the first n2 should be n1.
    In the second formula on page 334 (sp2 = ...), the denominator "n1 + n2 - 2" should be enclosed in parentheses: "(n1 + n2 - 2)" to properly indicate the order of operations.
    In the calculation at the top of page 335, the numerator should be enclosed in parentheses: "(69.1 - 63.7)" to properly indicate the order of operations, and the denominator within the radical should be "(1/20 + 1/20)".

Return to EPID600 home page


Vic Schoenbach, 5/26/2009, ..., 9/15/2013, 11/20/2014, 10/15/2017