1. Statistics for clinicians Biostatistics course by Kevin E. Kip, Ph.D., FAHA Professor and Executive Director, Research Center University of South Florida, College of Nursing Professor, College of Public Health Department of Epidemiology and Biostatistics Associate Member, Byrd Alzheimer’s Institute Morsani College of Medicine Tampa, FL, USA 1

2. SECTION 7.1 Introduction to Meta-Analysis Meta analysis and measures of public health impact

3. SECTION 7.6 Introduction to Measures of Public Health Impact

4. Learning Outcome: Explain features of individual measures of public health impact.

5. Measures of Public Health Impact Attributable Risk (AR)Number Attributable Risk Percent (AR%)Percentage Population Attributable Risk (PAR)Number Population Attributable Risk Percent (PAR%)Percentage Number Needed to Treat (NNT)Number Number Needed to Harm (NNH)Number

6. Measures of Public Health Impact IMPORTANT! They all assume (require) that a cause-effect relationship exists between the exposure and the outcome.

7. Relative Risk vs. Attributable Risk Relative Risk:Measure of the strength of association, and indicator used to assess the possibility of a causal relationship. Attributable Risk:Measure of the potential for prevention of disease if the exposure could be eliminated (assuming a causal relationship).

8. Relative Risk vs. Attributable Risk Relative Risk: Etiology Attributable Risk: Policy decisions Funding decisions (e.g. prevention programs)

9. Attributable Risk: Refers to EXPOSED persons. Population Attributable Risk: Refers to both EXPOSED and NONEXPOSED persons. Measures of Public Health Impact

10. SECTION 7.7 Attributable Risk And Attributable Risk Percent

11. Learning Outcome: Calculate and interpret the measures attributable risk (AR) and attributable risk percent (AR%)

12. Attributable Risk (AR) Among the EXPOSED: How much of the disease that occurs can be attributed to a certain exposure? AR AR% This is of primary interest to the practicing clinician.

13. Attributable Risk (AR) AR = I exposed – I nonexposed = “Risk Difference” SmokeYesNo Yes8429163000 No8749135000 Develop CHD I SM = 84 / 3000 = 0.028 = 28.0 / 1000 I NS = 87 / 5000 = 0.0174 = 17.4 / 1000 (background risk) AR = (28.0 – 17.4) / 1000 = 10.6 / 1000

14. Attributable Risk (AR) AR = (28.0 – 17.4) / 1000 = 10.6 / 1000 Among SMOKERS, 10.6 of the 28/1000 incident cases of CHD are attributed to the fact that these people smoke … Among SMOKERS, 10.6 of the 28/1000 incident cases of CHD that occur could be prevented if smoking were eliminated.

15. Practice - Attributable Risk (AR) Hypertension Among smokers, calculate and interpret incident cases of hypertension that are attributed to their smoking. Among SMOKERS, SmokingDiseaseNo DiseaseTotalIncidence (I)(I) per 1000 Exposed52348400________ Not Exposed43406449________ AR________ AR = I exposed – I nonexposed

16. Practice - Attributable Risk (AR) Hypertension Among smokers, calculate and interpret incident cases of hypertension that are attributed to their smoking. Among SMOKERS, 34.2 of the 130/1000 incident cases of hypertension are attributed to the fact that these people smoke … Among SMOKERS, 34.2 of the 130/1000 incident cases of hypertension that occur could be prevented if smoking were eliminated. SmokingDiseaseNo DiseaseTotalIncidence (I)(I) per 1000 Exposed523484000.130130.00 Not Exposed434064490.09695.77 AR0.03434.23 AR = I exposed – I nonexposed AR = 0.130 – 0.096 = 0.034

17. Attributable Risk Percent (AR%) AR% = (I exposed – I nonexposed ) / I exposed = “Etiologic fraction” SmokeYesNo Yes8429163000 No8749135000 Develop CHD AR% = (28.0 – 17.4) / 28.0=37.9% I SM = 84 / 3000 = 0.028 = 28.0 / 1000 I NS = 87 / 5000 = 0.0174 = 17.4 / 1000 (background risk)

18. Attributable Risk Percent (AR%) AR% = (28.0 – 17.4) / 28.0= 37.9% Among SMOKERS, 38% of the morbidity from CHD may be attributed to smoking… Among SMOKERS, 38% of the morbidity from CHD could be prevented if smoking were eliminated.

19. Practice - Attributable Risk Percent (AR%) Metabolic Syndrome Among heavy drinkers, calculate and interpret the percentage of incident cases of the metabolic syndrome attributed to their drinking. Among HEAVY DRINKERS, DrinkingDiseaseNo DiseaseTotalIncidence (I) Exposed44326370_____ Not Exposed28392420_____ AR%__________% AR% = (I exposed – I nonexposed ) / I exposed AR% =

20. Practice - Attributable Risk Percent (AR%) Metabolic Syndrome Among heavy drinkers, calculate and interpret the percentage of incident cases of the metabolic syndrome attributed to their drinking. Among HEAVY DRINKERS, 43.9% of the morbidity from the metabolic syndrome may be attributed to heaving drinking … Among HEAVY DRINKERS, 43.9% of the morbidity from the metabolic syndrome could be prevented if heavy drinking were eliminated. DrinkingDiseaseNo DiseaseTotalIncidence (I) Exposed443263700.119 Not Exposed283924200.067 AR%0.43943.9% AR% = (I exposed – I nonexposed ) / I exposed AR% = (0.119 – 0.067) / 0.119 = 0.439 x 100 = 43.9%

21. SECTION 7.8 Population Attributable Risk and Population Attributable Risk Percent

22. Learning Outcome: Calculate and interpret the measures population attributable risk (PAR) and population attributable risk percent (PAR%)

23. Population Attributable Risk (PAR) Among the EXPOSED and NONEXPOSED (e.g. total population): How much of the disease that occurs can be attributed to a certain exposure? PAR PAR% This of interest to policy makers and those responsible for funding prevention programs.

24. PAR and PAR% Example: We want to estimate how much of the burden of diabetes among Tampanians is attributed to obesity.

25. PAR and PAR% CAUTION! In order to calculate PAR and PAR%, we have to be reasonably sure that the results of the study can be generalized to the population of Tampa. (e.g the incidence rates drawn from the sample approximate the incidence rates in the entire population).

26. Population Attributable Risk (PAR) PAR = I total – I nonexposed WeightYesNo Obese85036504500 Slim25052505500 Diabetes I T = 1100 / 10000 = 0.11 = 110 / 1000 I NE = 250 / 5500 = 0.0455 = 45.5 / 1000 (background risk) PAR = (110 – 45.5) / 1000 = 64.5 / 1000 1100890010000

27. Population Attributable Risk (PAR) PAR = (110 – 45.5) / 1000 = 64.5 / 1000 In Tampa, 64.5 of the 110/1000 incident cases of diabetes are attributed to obesity … In Tampa, 64.5 of the 110/1000 incident cases of diabetes that occur could be prevented with sufficient weight loss.

28. Practice – Population Attributable Risk (PAR) Hypertension Estimate how much of the burden of hypertension among Tampanians is attributed to high salt consumption. In TAMPA, Salt IntakeDiseaseNo DiseaseTotalIncidence (I) (I) per 1,000 _______ Exposed34380414_______ Not Exposed28524552_______ Total 62904 966 _______ PAR_______ PAR = (I total – I nonexposed ) PAR =

29. Practice – Population Attributable Risk (PAR) Hypertension Estimate how much of the burden of hypertension among Tampanians is attributed to high salt consumption. In TAMPA, 13.46 of the 64.18/1,000 incident cases of hypertension are attributed to having high salt consumption. In TAMPA, 13.46 of the 64.18/1,000 incident cases of hypertension could be prevented with elimination of high salt intake. Salt IntakeDiseaseNo DiseaseTotalIncidence (I) (I) per 1,000 82.31/1,000 50.72/1,000 64.18/1,000 Exposed343804140.08231 Not Exposed285245520.05072 Total 62904 966 0.06418 PAR0.0134613.46/1,000 PAR = (I total – I nonexposed ) PAR = (0.064 – 0.051) / 0.01346 = 13.46 per 1,000

30. Population Attributable Risk Percent PAR% = (I total – I nonexposed ) / I total WeightYesNo Obese85036504500 Slim25052505500 Diabetes PAR% = (110 – 45.5) / 110=58.6% 1100890010000 I T = 1100 / 10000 = 0.11 = 110 / 1000 I NE = 250 / 5500 = 0.0455 = 45.5 / 1000 (background risk)

31. Population Attributable Risk Percent PAR% = (110 – 45.5) / 110= 58.6% In Tampa, 59% of the cases of diabetes may be attributed to obesity in the population… In Tampa, 59% of the cases of diabetes could be prevented if Tampa residents lost sufficient weight.

32. Practice – Population Attributable Risk% (PAR%) Hypertension Estimate the percentage of the burden of hypertension among Tampanians that is attributed to high salt consumption. In TAMPA, 21% of the cases of hypertension may be attributed to having high salt consumption in the population. In TAMPA, 21% of the cases of hypertension could be prevented with elimination of high salt intake. Salt IntakeDiseaseNo DiseaseTotalIncidence (I) (I) per 1,000 _______ Exposed34380414_______ Not Exposed28524552_______ Total 62904 966_______ PAR%_______ PAR% = (I total – I nonexposed ) / I total PAR% =

33. Practice – Population Attributable Risk% (PAR%) Hypertension Estimate the percentage of the burden of hypertension among Tampanians that is attributed to high salt consumption. In TAMPA, 21% of the cases of hypertension may be attributed to having high salt consumption in the population. In TAMPA, 21% of the cases of hypertension could be prevented with elimination of high salt intake. Salt IntakeDiseaseNo DiseaseTotalIncidence (I) (I) per 1,000 82.31/1,000 50.72/1,000 64.18/1,000 Exposed343804140.08231 Not Exposed285245520.05072 Total 62904 966 0.06418 PAR%21.0% PAR% = (I total – I nonexposed ) / I total PAR% = (0.064 – 0.051) / 0.064 = 0.013 / 0.064 = 21.0%

34. Measures of Public Health Impact NOTE! Both attributable and population attributable risks should be cautiously interpreted. In reality, even if an exposure is causal, we do not know whether it truly contributed to disease occurrence in all exposed persons - - in some exposed persons, other causal factors may have been entirely responsible.

35. SECTION 7.9 Attributable Risk and Population Attributable Risk Percent for Case- Control Studies

36. Learning Outcome: Calculate and interpret the measures attributable risk (AR) and attributable risk percent (AR%) in Case Control Studies

37. l They are based on measures of incidence. l We can calculate incidence measures from case-control studies only under special circumstances. l Therefore, the AR and PAR cannot usually be calculated from case-control data. l However, for most case-control studies, we can calculate the AR% and PAR%. Measures of Public Health Impact

38. AR% (Case-Control Studies) (OR – 1) AR% =----------- x100 OR

39. Example: AR% (Case-Control Studies) SmokeYesNo Yes160120 No90200 Case-control study to evaluate the impact of smoking as related to bladder cancer. Bladder Cancer (160 / 90) OR = ------------ (120 / 200) = 2.96

40. Example: AR% (Case-Control Studies) Question: Among smokers, what proportion (percent) of bladder cancer cases can be attributed to their smoking habit? (OR – 1) AR% =-----------x100 OR AR% =((2.96 – 1) / 2.96) x 100 = 66.2%

41. Example: AR% (Case-Control Studies) l 66% of bladder cancer cases among smokers can be attributed to their smoking. l 66% of bladder cancer cases among smokers could be prevented if they had never taken up smoking. (Assuming there is a causal association between smoking and the development of bladder cancer).

42. Practice: AR% (Case-Control Studies) Esophageal Cancer Among smokers, estimate the percentage of esophageal cancer cases attributed to smoking. Among smokers, SmokingDiseaseNo DiseaseTotal Smoker6552116 Non-smoker288396684 OR_______ AR% _______ (OR – 1) AR% =-----------x 100AR = ----------- x 100= _______ OR OR = _____________

43. Esophageal Cancer Among smokers, 66% of esophageal cancer cases can be attributed to their smoking. Among smokers, 66% of esophageal cancer cases could be prevented if they had never taken up smoking. SmokingDiseaseNo DiseaseTotal Smoker6552116 Non-smoker288396684 OR1.692 AR% 40.9% (OR – 1)(1.692 – 1) AR% =-----------x 100AR = ----------- x 100=40.9% OR 1.692 OR = (65/288) / (52/396) = 1.692 Practice: AR% (Case-Control Studies) Among smokers, estimate the percentage of esophageal cancer cases attributed to smoking.

44. PAR% (Case-Control Studies) (P E ) (OR – 1) PAR% =-------------------- x 100 [(P E ) (OR-1)] + 1 where P E = proportion of exposed controls (assuming that the proportion of exposed controls is representative of the proportion exposed in the source population)

45. Example: PAR% (Case-Control Studies) SmokeYesNo Yes160120 No90200 Case-control study to evaluate the impact of smoking as related to bladder cancer. Bladder Cancer (160 / 90) OR = ------------ (120 / 200) = 2.96 P E = 120 / 320 = 0.375

46. Example: PAR% (Case-Control Studies) Question: In this study population, what proportion (percent) of bladder cancer cases can be attributed to smoking? (P E ) (OR – 1) PAR% =---------------------- x 100 [(P E ) (OR-1)] + 1 PAR% = (0.375) (2.96-1) [(0.375) (2.96-1)] + 1 x 100 = 42.4%

47. Example: PAR% (Case-Control Studies) l In this study population, 42% of bladder cancer cases can be attributed to smoking. l In this study population, 42% of bladder cancer cases could be prevented if people would not take up smoking. (Assuming there is a causal association between smoking and the development of bladder cancer).

48. Practice: PAR% (Case-Control Studies) Esophageal Cancer Among the study population, estimate the percentage of esophageal cancer cases attributed to smoking. In this study population, SmokingDiseaseNo DiseaseTotal Smoker6552116 Non-smoker288396684 OR_______ PE PE PAR% _______ PAR% =-----------x 100PAR%=______________ OR = _____________ P E = _____________ (P E ) (OR – 1) PAR% =-------------------- x 100 [(P E ) (OR-1)] + 1

49. Practice: PAR% (Case-Control Studies) Esophageal Cancer Among the study population, estimate the percentage of esophageal cancer cases attributed to smoking. SmokingDiseaseNo DiseaseTotal Smoker6552116 Non-smoker288396684 OR_______ PE PE PAR% _______ 0.116 x (1.692 – 1) PAR% =---------------------- x 100PAR% = 7.4% [(0.116) (1.692-1)] + 1 OR = (65 / 266) / (52 / 396) = 1.692 P E = 52 / (52 + 396) = 0.116 (P E ) (OR – 1) PAR% =-------------------- x 100 [(P E ) (OR-1)] + 1 In this study population, 7.4% of esophageal cancer cases can be attributed to smoking. In this study population, 7.4% of esophageal cancer cases could be prevented if people would not take up smoking.

50. SECTION 7.10 Attributable Risk versus Relative Risk

51. Learning Outcome: Differentiate between attributable risk and relative risk.

52. Relative Risk vs. Attributable Risk Smokers Non- smokersRRARAR% Lung cancer1401014.013092.9 CHD6694131.625638.3 Age-Adjusted Death Rates per 100,000

53. Relative Risk vs. Attributable Risk Smokers Non- smokers RRARAR% Lung cancer1401014.013092.9 CHD6694131.625638.3 Smoking has a much stronger association with lung cancer mortality than CHD mortality, however… death from CHD is much more common than lung cancer, hence higher attributable risk associated with smoking.

54. Issues in Prevention Policy An important question in prevention is whether the approach should target specific groups known to be at high risk, or extend to the general population as a whole. This depends largely on the nature of the exposure/disease relationship, and the distribution of the exposure in the population.

55. Percent of population Systolic Blood Pressure (mm Hg) The majority of the population has systolic blood pressure values in the normal range (< 140).

56. RR of CHD Death Systolic Blood Pressure (mm Hg) The risk of CHD increases steadily with higher systolic blood pressure levels.

57. % of excess CHD deaths Systolic Blood Pressure (mm Hg) The majority of excess CHD deaths occur largely in the high-normal range (130 to 159).

58. SECTION 7.11 Number Need to Treat

59. Learning Outcome: Calculate and interpret the measure Number Needed to Treat (NNT).

60. “Number Needed to Treat” Example: Assume: Treatment A = Drug Treatment B = Placebo Outcome = Colon Cancer within 5 Year Incidence A =0.15 Incidence B =0.20 NNT =1 / (I B – I A ) =1 / (0.20 – 0.15) =20 treated

61. Cannon (2006) No difference was observed for total or non-cardiovascular mortality………….. 32.3 – 28.8 = 3.5

62. Treating to New Targets (TNT) study Any Cardiovascular Event: Atorvastatin – 10 mg: 0.335 Atorvastatin – 80 mg: 0.281 Number Needed to Treat (NNT) NNT = 1 / (Incidence 10mg – Incidence 80mg ) = 18.5 Major Cardiovascular Event: Atorvastatin – 10 mg: 0.109 Atorvastatin – 80 mg: 0.087 NNT = 45.5 This is the best case scenario…….(e.g. initial run-in phase for trial eligibility).

63. Practice: Number Needed to Treat (NNT) Assume that a new type of immunotherapy, which is expensive, is more effective in treating ulcerative colitis compared to standard of care therapy. Calculate the number needed to treat to avoid a recurrence of disease with the more expensive new immunotherapy. Assume: Treatment A = New Immunotherapy Treatment B = Standard of Care Outcome = Recurrence of Ulcerative Colitis within 5 years Incidence A =0.20 Incidence B =0.35 NNT =1 / (I B – I A ) =_________ =______ treated patients

64. Practice: Number Needed to Treat (NNT) Assume that a new type of immunotherapy, which is expensive, is more effective in treating ulcerative colitis compared to standard of care therapy. Calculate the number needed to treat to avoid a recurrence of disease with the more expensive new immunotherapy. Assume: Treatment A = New Immunotherapy Treatment B = Standard of Care Outcome = Recurrence of Ulcerative Colitis within 5 years Incidence A =0.20 Incidence B =0.35 NNT =1 / (I B – I A ) =1 / (0.35 – 0.20) =6.7 (rounded up to 7) = treated patients