Notes and References to Second Opinions on Medicine in the Media

1. Clinical epidemiology is the discipline that applies information on groups of patients to the everyday practice of medicine. It applies and extends techniques of epidemiology to medical practice. Epidemiology studies the presence of disease in populations and the factors that influence and cause disease; evaluates prevention measures and therapies; and informs public policy decisions.

2. Chen WY, Rosner B, Hankinson SE, et al. Moderate Alcohol Consumption During Adult Life , Drinking Patterns, and Breast Cancer Risk. JAMA, November 2, 2011-Vol 306, No. 17 Pages 1884-1890.

3. A textbook based report is a report written in a style used to write a term paper in college or graduate school. The report is thorough, different viewpoints are considered. References and footnotes are provided. The footnotes and references provide transparency of the author’s thinking and an audit trail for the author’s conclusions. Readers or listeners are thus provided the opportunity to evaluate the validity of the report.

4. The Relative Risk based statistic used in the newscast is calculated by using the following formula: 1 – { [ (number of cases of breast cancer in moderate drinkers) / (number of moderate drinkers)] / [ (number of breast cancer cases in non-drinkers) / (number of non-drinkers)]}, while the Risk Difference statistic is calculated by this formula: [(number breast cancer cases in moderate drinkers) / (number of moderate drinkers)] – [( breast cancer cases in non drinkers) / (number of non-drinkers)]. Note the technical name for the Relative Risk based statistic is Relative Risk Reduction. I avoided mentioning its technical name in the text to avoid confusion since the text is discussing an example of increased risk. Unfortunately the terminology in epidemiology can be inconsistent and confusing. 

The above can be expressed in algebraic symbols. Let: a = number of breast cancer cases in moderate drinkers; c = number of breast cancer cases in non-drinkers; tmd = total moderate drinkers; and tnd = total non-drinkers. Thus we have 

Relative Risk Reduction = 1– [(a/tmd)/(c/tnd)]

Risk Difference = (a/tmd) – (c/tnd)

See footnote 9. below for comments pertaining to the calculation of the Risk Difference for this study.

5. Epidemiology 4th ed. by Leon Gordis 2009 published by SaundersElsevier, Philadelphia, See chapters 11 and 12 regarding the use and calculation of the Relative Risk and the Risk Difference. The synonymous term Attributable Risk is used in this text instead of the term Risk Difference.

6. Epidemiology in Medicine by Charles H. Hennekens and Julie E. Buring 1987 published by Lippincott Williams & Wilkins, Philadelphia, . See chapter 4 especially page 87 and pages 93 through 95. The synonyms Risk Difference and Attributable Risk are mentioned on page 87. The last paragraph of page 93 discusses the use of the Relative Risk and the Risk Difference ( the text uses the synonym Attributable Risk).

7. Activ Epi Companion Textbook by David G. Kleinbaum, Kevin M, Sullivan, and Nancy D. Barker 2003 Third Corrected Printing 2005 published by Springer Science + Business Media LLC, New York, . See lessons 5 and 6, especially pages 103 and 133. The authors use the term Risk Difference and do not make use of the synonymous term Attributable Risk.

8. Users’ Guide to the Medical Literature: A Manual for Evidenced Based Medicine 2nd ed. Edited by Gordon Guyatt, DrummendRennie, Maureen O. Meade, and Deborah J. Cook 2008, published by the McGraw-Hill Companies, Inc., New York, This book is the bible of evidence based medicine. See chapter 7, especially page 90. The authors also use the synonym Absolute Risk Reduction for Risk Difference. The term Relative Risk Reduction is explained on page 90 along with an example of its calculation. Also, on page 90 is the beginning of a discussion of Relative Risk vs Risk Difference.

9. The calculation of the Risk Difference in this study is complicated by the data analysis of the study. The data was analyzed by the statistical method called survival analysis. The authors of the study2 did not report the Risk Difference nor Kaplan-Meier survival curves that would allow for the calculation adjustments recommended by Altman and Anderson.10 Sufficient information was given to perform a less exact calculation described by Barratt et al in reply to de Lemos11 for large randomized controlled trials with good follow-up (the study discussed here is a large prospective cohort study with good follow-up). The calculation method of Barratt et al11 gives a Risk Difference of 0.4% which is less than 1 in 100. The authors did report a statistic called the Percent Attributable Risk. They reported a Percent Attributable Risk of 3%. The calculation of the Percent Attributable Risk gives a larger number than the Risk Difference. The Risk Difference (synonym Attributable Risk) is the proportion in the numerator in the Percent Attributable Risk calculation while the denominator is the proportion of cancer cases in moderate drinkers (page 88 Hennekens6). Thus the Risk Difference cannot be greater than 3%. Textbooks usually refer to the Percent Attributable Risk as attributable risk percent, attributable proportion, or etiologic fraction (page 88 Hennekens6). Unfortunately the terminology in epidemiology can be inconsistent and confusing. Hopefully in the future, journals will require authors to report Risk Diferences.

10. Altman DG, Anderson PK. Calculating the number needed to treat for trials where the outcome is time to an event. BMJ DEC.4 1999, 319:pages 1492-1495.

11. de Lemos ML, Barratt AL, Wyer PC, Guyatt G, Simpson JM. NNT for studies with long-term follow-up. Letters CMAJ MAR. 1, 2005, 172(5): pages 613-615.

12. This study is a prospective cohort study because the moderate drinking group members and non-drinking group members were not randomly assigned to their groups. The members of each group self-selected to their group. It is prospective in that the outcomes had not occurred when the study was initiated. A cohort study can determine the presence of an association however further evidence is required to determine if that association is a cause-effect association. Randomized controlled trials can determine cause-effect associations.

13. Survey methods were utilized. Drinking behavior was assessed by written responses to questions on a questionnaire which was sent out periodically. Health status was assessed primarily the same way.

14. Dabigatran (Pradaxa) for stroke prevention in patients with non-valvular atrial fibrillation  NPS RADAR/AUGUST 2011

15. Connolly SJ et al. Dabigatran versus warfarin in patients with atrial fibrillation. New Engl J Med 2009;361;1139-51.

16. Users’ Guide to the Medical Literature: A Manual for Evidenced Based Medicine 2nd ed. Edited by Gordon Guyatt, Drummend Rennie, Maureen O. Meade, and Deborah J. Cook 2008, published by the McGraw-Hill Companies, Inc., New York, See pages 91-93.

17. Center for Evidence Based Medicine The  term Absolute Risk Reduction (ARR) is a synonymous with  Risk Difference (RD).

For more information about Dr. LaRiccia visit:
Email address for Dr. LaRiccia: