Relative vs Absolute Risk How to Interpret Epidemiological

Relative vs. Absolute Risk: How to Interpret Epidemiological Data Sam Zakhari, Ph. D. Senior Vice President, Science Distilled Spirits Council, USA June 3, 2019

What is Risk? RISK is the possibility of losing something of value: • physical health • social status • financial wealth • data (cyber risk) • emotional well-being ……………. . Hazard: Potential Source of Harm 2

How to Measure Risk? Focus on Health Risk …. . 3

Epidemiological (Observational) Studies Exposure Association Disease Not exposed Exposed Patterns of disease in Specified populations (time, place, personal characteristics) 4

Absolute Risk • Risk of disease occurring over a period of time • It is a probability • Absolute risk = incidence rate • e. g. , 1 in 20 develop colorectal cancer over lifetime 5

Absolute Risk…Cancer 66% of Cancer Mutations is Due to Error in Copying DNA April 4, 2017 by Dr. Francis Collins NIH Director https: //directorsblog. nih. gov/ 6

Relative Risk…Lung Cancer Compares risk of disease between exposed and non-exposed groups 1 25 75 Smokers 25% 99 1% RR = 25% ____ = 25 1% Non-smokers 7

Odds Ratio(OR) • Determines how strongly exposure to risk factor is associated with outcome Smoking A/B OR = C/D Lung Cancer Yes No Yes A C No B D OR = 1: Exposure to RF does not affect odds of outcome OR > 1: Exposure to RF associated with higher odds of outcome OR < 1: Exposure to RF associated with lower odds of outcome 8

RR and OR in Context of AR • RR and OR convey useful information, however, they MUST be interpreted in the context of absolute risk • If the AR is low (e. g. , 0. 02%, or 1 in 5, 000), a 100% increase in RR would result in 2 in 5, 000 incidence; high statistical, but low clinical significance • On the other hand, if AR is high (e. g. , 30%, 1, 500 in 5, 000), then a small increase in RR (e. g. , 1. 2%) will result in extra 300 cases, a high clinical significance. A statistically significant result may not necessarily be clinically important 9

Classical Epidemiological Studies. . . • Epidemiologic studies run the risk of appearing to give results more precise and conclusive than are warranted e. g. , 10% increase in BC risk with each drink • Increased relative risk does not necessarily mean that an individual will definitely get the disease • Relative risk sounds more alarming; authors may manipulate the reader by using alarming phraseology to upturn emotional impact 10

Epidemiologists Express Risk as… Relative Risk 11

Breast Cancer: Absolute vs. Relative https: //www. breastcancer. org/symptoms/understand_bc/risk/understanding Risk 1667 40 -yr old women Absolute risk = 1. 45% Light drinking increase in absolute risk = 0. 06% 24 BC Cases 1 BC Case Light drinking increases Relative Risk by 4%; (We don’t know if this is causal) i. e. Absolute risk = 1. 45 X 1. 04 = 1. 51%, an increase in absolute risk 0 f 0. 06% 12

Classical Epidemiological Studies Show Association but Not Causation. . Except when the relative risk is very high, e. g. , smoking, infectious diseases 13

Why Classical Epidemiology Alone Is Not Sufficient to Establish Causation? 14

General Population 1. NO ONE SIZE FITS ALL The premise of classical epidemiology: patients with the same diagnosis have similar causes and disease progression despite variations in molecular pathology (Not True) 15

The Real World Different genetic background, life-style, diet, culture, environmental conditions (water and air pollution, endemic diseases, viruses, parasites, etc), comorbid conditions, education, access to health care 16

Every Individual is Unique … 17

Personalized Medicine … 18

Why Epidemiology is Not Sufficient to Establish Causation? 2. Methodological Problems • Self Report Under-reporting (Klatsky et al (2014) Moderate alcohol intake and cancer: the role of underreporting. Cancer Causes Control. doi: 10. 1007/s 10552 -014 -0372 -8) • Average Consumption (7 X 1 =/ 1 X 7) • Do not Consider Molecular Mechanisms, and Other Risk Factors for Cancer Formation 19

Under. Reporting “What I drink and what I tell the pollsters I drink are two different things” 20

Epidemiology vs. Molecular Biology A B Exposure Disease Epidemiology A C B Molecular Biology 21

Cancer(s) are Complex, Multifactorial Diseases Age, Family History, Inherited genetic mutations • Lifestyle • Immunosuppression • Environment • Infections • Obesity • Radiation • Sunlight • Tanning beds • DNA damage • Chronic inflammation • Diet • Viruses • Carcinogens • Hormones • Circadian rhythm Cancer patient These are the forces that precipitate or prevent cancer 22

The Lancet Study …No Safe Level https: //doi. org/10. 1016/S 0140 -6736(18)31310 -2 Study • Based on “estimates” of current drinking, alcohol attributable death, and adjusted alcohol sales. • It uses 23 health outcomes associated with alcohol use. • The level of alcohol consumption that minimized harm was zero. Ripple • • • No alcohol safe to drink, global study confirms ’No amount of alcohol is good for your overall health, global study says No safe level of alcohol, new study concludes No Amount Of Alcohol Is Good For Your Health, Global Study Says No amount of alcohol is safe, health experts warn No healthy level of alcohol consumption, says major study There's "no safe level of alcohol, " major new study concludes No amount of alcohol — not even 1 glass of wine — is safe, global study says 'No Safe Limit': Even One Drink a Day Increases Risks Safest level of alcohol consumption is none, worldwide study shows There’s no ‘safe’ level of alcohol consumption, study finds 23

Press Releases and Media BMJ 2014; 349: g 7015 doi: 10. 1136/bmj. g 7015 Of the press releases: Exaggerated News Stories • 40% contained exaggerated advice • 33% contained exaggerated causal claims • 36% contained exaggerated inference from animal research to humans 50% 81% 86% 24

Take Home Messages 1 2 3 • RR convey useful information, however, it MUST be interpreted in the context of absolute risk • A statistically significant result may not necessarily be clinically important • Association does not mean causation • To posit causation, epidemiological study need to dovetail with molecular biology Sam 25

szakhari@discus. org 202 -682 -8837 26