Measures of Excess Risk PDF

Measures of excess risk Foundations of Epidemiology I  Goals To understand how to quantify measures of excess risk using incidence: – Risk difference – Risk ratio – Attributable risk – Attributable risk percent – Population attributable risk – Population attribut...

Measures of excess risk Foundations of Epidemiology I  Goals To understand how to quantify measures of excess risk using incidence: – Risk difference – Risk ratio – Attributable risk – Attributable risk percent – Population attributable risk – Population attributable risk percent To understand how to estimate the above without incidence measures  What is risk? An estimation of the likelihood of an outcome in a specified population, over a specified time period Risk = Cumulative incidence in the population (Ic) Ic = # of events in a specified time frame # in population at risk Assumes the population at risk remains constant  2008 Lung Cancer Risk in Females Population Cumulative Incidence (per 100,000) United States 36.2 Taiwan 28.7 For risk (cumulative incidence) calculation: Numerator = # of incident cases of lung cancer among females in population during observation period Denominator = total number of females without lung cancer in population at start of observation period -Can use total population when prevalent cases are negligible compared to whole population. Source: World Health Organization GLOBOCAN 2008  Most epidemiologists use “risk” for both cumulative incidence & incidence rate measures Both Cumulative Incidence and Incidence Rate can be used to calculate excess risk: – Risk Difference (Incidence Rate Difference) – Risk Ratio (Incidence Rate Ratio) – Attributable Risk (Attributable Rate) – Attributable Risk Percent (Attributable Rate %) – Population Attributable Risk (Pop Attr. Rate) – Population Attributable Risk Percent (Pop Attr. Rate %) 2008 Lung Cancer Incidence Rate in Females Population Incidence Rate (per 100,000 person-years) United States 36.2 Taiwan 28.7 For incidence rate calculation: Numerator = # of incident cases of lung cancer among females in the population Denominator = total observed person-time of females without lung cancer in that country -When prevalent cases negligible compared to whole population, can use total population * maximum observation time How do we quantify the association of an exposure with an outcome? Compare the incidence of outcome in exposed versus unexposed individuals Is there an excess risk of outcome among individuals exposed? Association does not always mean causation – We observe association, but infer causation – We’ll learn more about causation later in course  Calculating excess risk Exposed Unexposed = at risk at start of period = exposed = incident case during study period How does incidence compare in the populations? Common data table for dichotomous exposure & dichotomous outcome Outcome Exposure Present Absent Totals Yes a b a+b No c d c+d 𝑎 𝐼# = 𝐶𝑢𝑚𝑢𝑙𝑎𝑡𝑖𝑣𝑒 𝐼𝑛𝑐𝑖𝑑𝑒𝑛𝑐𝑒 𝑖𝑛 𝐸𝑥𝑝𝑜𝑠𝑒𝑑 = 𝑎+𝑏 𝑐 𝐼" = 𝐶𝑢𝑚𝑢𝑙𝑎𝑡𝑖𝑣𝑒 𝐼𝑛𝑐𝑖𝑑𝑒𝑛𝑐𝑒 𝑖𝑛 𝑈𝑛𝑒𝑥𝑝𝑜𝑠𝑒𝑑 = 𝑐+𝑑 𝑎+𝑐 𝐼! = 𝑇𝑜𝑡𝑎𝑙 𝐶𝑢𝑚𝑢𝑙𝑎𝑡𝑖𝑣𝑒 𝐼𝑛𝑐𝑖𝑑𝑒𝑛𝑐𝑒 = 𝑎+𝑏+𝑐+𝑑  Risk Difference RD = Incidence in Exposed - Incidence in Unexposed = Ie – Io Outcome Exposure Present Absent Totals Yes a b a+b No c d c+d 𝑎 𝑐 𝑅𝑖𝑠𝑘 𝐷𝑖𝑓𝑓𝑒𝑟𝑒𝑛𝑐𝑒 = − 𝑎+𝑏 𝑐+𝑑 RD = 0 no association between Exposure and Outcome RD > 0 Exposure positively associated with risk of Outcome RD < 0 Exposure negatively associated with risk of Outcome  In Class Exercise I: Risk Difference Exposed Unexposed = incident case Calculate & Interpret the Risk Difference  Risk Ratio (aka Relative Risk) 𝐼! 𝐼𝑛𝑐𝑖𝑑𝑒𝑛𝑐𝑒 𝑖𝑛 𝑒𝑥𝑝𝑜𝑠𝑒𝑑 𝑅𝑖𝑠𝑘 𝑅𝑎𝑡𝑖𝑜 = = 𝐼" 𝐼𝑛𝑐𝑖𝑑𝑒𝑛𝑐𝑒 𝑖𝑛 𝑢𝑛𝑒𝑥𝑝𝑜𝑠𝑒𝑑 Outcome Exposure Present Absent Totals Yes a b a+b No c d c+d 𝑎 𝑅𝑖𝑠𝑘 𝑅𝑎𝑡𝑖𝑜 = 𝑎 + 𝑐 𝑏 𝑐+𝑑 RR = 1 no association between Exposure and Outcome RR > 1 Exposure positively associated with risk of Outcome RR < 1 Exposure negatively associated with risk of Outcome  In Class Exercise II: Risk Ratio Exposed Unexposed = incident case Calculate & Interpret the Risk Ratio In Class Exercise III: Risk Difference vs. Risk Ratio Disease Incidence (per 100,000) Notation Exposure to E Population A Population B Ie Yes 40 90 Io No 10 60 Ie-Io Risk Difference Ie Risk Ratio Io (Relative Risk) Calculate, interpret, and compare the Risk Differences and Risk Ratios obtained for each population.  Choosing excess risk estimate Why might a researcher or public health practitioner choose to report Risk Difference vs. Relative Risk? Choosing excess risk estimate Relative Risk gives information about the strength of the association. Risk Difference gives information on potential public health impact.  Common data table for dichotomous exposure & incident cases over person- time of follow-up Exposure # Cases Person-time of follow-up Yes a PTe No c PTo 𝑎 𝐼# = 𝐼𝑛𝑐𝑖𝑑𝑒𝑛𝑐𝑒 𝑅𝑎𝑡𝑒 𝑖𝑛 𝐸𝑥𝑝𝑜𝑠𝑒𝑑 = 𝑃𝑇# 𝑐 𝐼" = 𝐼𝑛𝑐𝑖𝑑𝑒𝑛𝑐𝑒 𝑅𝑎𝑡𝑒 𝑖𝑛 𝑈𝑛𝑒𝑥𝑝𝑜𝑠𝑒𝑑 = 𝑃𝑇" 𝑎+𝑐 𝐼! = 𝑇𝑜𝑡𝑎𝑙 𝐼𝑛𝑐𝑖𝑑𝑒𝑛𝑐𝑒 𝑅𝑎𝑡𝑒 = 𝑃𝑇# + 𝑃𝑇" 𝑎 𝑎 𝑐 𝑃𝑇! 𝑅𝑎𝑡𝑒 𝐷𝑖𝑓𝑓𝑒𝑟𝑒𝑛𝑐𝑒 = − 𝑅𝑎𝑡𝑒 𝑅𝑎𝑡𝑖𝑜 = 𝑐 𝑃𝑇! 𝑃𝑇" 𝑃𝑇" Common data table for Case Control Study Outcome Exposure Present Absent Totals Yes a b a+b No c d c+d 𝑎 = 𝑂𝑑𝑑𝑠 𝑜𝑓 𝑒𝑥𝑝𝑜𝑠𝑢𝑟𝑒 𝑖𝑛 𝑡ℎ𝑜𝑠𝑒 𝑤𝑖𝑡ℎ 𝑜𝑢𝑡𝑐𝑜𝑚𝑒 𝑐 𝑏 = 𝑂𝑑𝑑𝑠 𝑜𝑓 𝑒𝑥𝑝𝑜𝑠𝑢𝑟𝑒 𝑖𝑛 𝑡ℎ𝑜𝑠𝑒 𝑤𝑖𝑡ℎ𝑜𝑢𝑡 𝑜𝑢𝑡𝑐𝑜𝑚𝑒 𝑑 #⁄ #⁄ #$ $ % 𝑂𝑑𝑑𝑠 𝑅𝑎𝑡𝑖𝑜 = %" = $⁄ = %& & &  In Class Exercise IV: Odds Ratio Case control Study: Cooking and lung cancer risk in Taiwanese women Lung cancer # meals cooked/day Present Absent 2+ 140 200 1 10 50 Calculate the odds ratio and interpret in words. Modified from Ko et al. Am J Epidemiol 2000;151:140-7  When is the Odds Ratio a Good Estimate of the Risk Ratio? When the cases and controls are drawn from the same source population When the exposure history of cases studied is representative of all people with the outcome in the source population When the exposure history of controls studied is representative of all people without the outcome in the source population When the disease is rare a will be small relative to b 𝑎 c will be small relative to d 𝑅𝑖𝑠𝑘 𝑅𝑎𝑡𝑖𝑜 = 𝑎 + 𝑏 𝑐 OR ≈ RR 𝑐+𝑑  Attributable risk for the exposed (AR) AR= Ie – Io Incidence (per 1,000) AR Unexposed Exposed Among exposed, what amount of the disease incidence is due to exposure? Should action be taken to eliminate the exposure?  Attributable risk for the exposed (AR) AR= ISame e – Io calculation as risk Incidence (per 1,000) difference. Why do we AR have 2 terms? Unexposed Exposed Among exposed, what amount of the disease incidence is due to exposure? Should action be taken to eliminate the exposure?  What is the difference between risk difference and attributable risk? Same value, but different interpretation! Risk difference = disease incidence associated with exposure among exposed individuals Attributable risk (AR) = disease incidence due to exposure among exposed individuals Attributable risk (AR) is only used for established, causal associations between exposure and disease  Attributable Risk (of exposed) Exposed Unexposed AR = 6/12 – 3/12 =3/12 = 0.25 Among exposed individuals, 25% develop the disease due to the exposure.  Attributable Risk % (of cases) What percent of D among the exposed is due to the exposure? Only used for established, causal associations between exposure and disease Also known as attributable portion AR%= Ie – Io x 100 Ie = RR-1 x 100 RR Attributable Risk % (of cases) Exposed Unexposed Of the exposed cases, 50% are caused by the exposure.  In Class Exercise V: AR and AR% What is the IT, IO, and IE? n=120 Calculate and interpret the Attributable Risk (AR) and Attributable Risk Percent (AR%) = incident case = exposed  In Class Exercise VI Cohort study of US female smokers and non-smokers followed for lung cancer Incidence per Lung Cancer Person-years 100,000 Cases Follow up person-years Ever smoker 2,000 800,000 250 Never smoker 150 600,000 25 What is the attributable risk for lung cancer among ever smokers compared to never smokers? Interpret in words. What is the attributable risk % for lung cancer among ever smokers compared to never smokers? Interpret in words. Modified from Freedman et al. Lancet Oncol 2008;9:649-656.  Vaccine Efficacy Special case of AR% Unvaccinated are exposed and vaccinated are unexposed Vaccine Efficacy = Iunvacc. – Ivacc. x 100 Iunvacc. Vaccine efficacy is the % reduction of disease in a vaccinated group of people compared to an unvaccinated group  In Class Exercise VII Vaccine Efficacy In a clinical trial of the varicella (chicken pox vaccine), the cumulative incidence of varicella in the unvaccinated children was 42.9% and the cumulative incidence of varicella in the vaccinated children was 11.8%. Calculate the vaccine’s efficacy and interpret in words.  Herd Immunity Oxford Vaccine Group: https://www.ovg.ox.ac.uk/news/herd-immunity-how-does-it-work Excess 1-year risk of lung cancer and CHD associated with smoking Is the association with smoking stronger for lung cancer or CHD? Is the health impact of smoking greater for lung cancer or CHD? Image: https://www.goredforwomen.org Excess 1-year risk of lung cancer and CHD associated with smoking Lung CHD Cancer Risk ratio 10 2.5  Excess 1-year risk of lung cancer and CHD associated with smoking Lung CHD Cancer Risk ratio 10 2.5 Interpretation: The 1-year risk of lung cancer is 10 times as high in smokers than it is in non-smokers, and the 1-year risk of CHD is 2.5 times as high in smokers than it is in non- smokers. Smoking is more strongly related to lung cancer incidence than CHD incidence.  Excess 1-year risk of lung cancer and CHD associated with smoking Lung CHD Cancer attributable 1/1,000 10/1,000 risk/yr Interpretation: The risk of lung cancer due to smoking (opposed to other causes) is 1/1,000 per year among smokers, and the risk of CHD due to smoking (opposed to other causes) is 10/1,000 per year among smokers. Smoking has a greater public health impact on CHD risk than lung cancer risk, among smokers.  Population Attributable Risk (PAR) Incidence (per 100) PAR Exposed In the total population, what amount of the disease incidence is due to exposure?  Compare to Attributable risk to the exposed (AR) 40 AR= Ie – Io Incidence (per 100) AR 10 Unexposed Unexposed Exposed Exposed Among the exposed, what amount of the disease incidence is due to exposure?  Population Attributable Risk The contribution of exposure to the overall incidence in the population Only used for established, causal associations between exposure and disease Addresses “Should we use resources to control a certain exposure or other exposures causing greater health problems in the population?” PAR= It – Io PAR = AR * pt = (Ie – Io) * pt Where pt= proportion of total population exposed  Population Attributable Risk IT = 0.2 24/120 IO = 0.1 8/80 IE =0.4 16/40 PAR = (IT – IO) = 0.2 – 0.1 = 0.1 Among those in the total population, 10% develop the outcome due to the exposure. PAR = AR * PT = 0.3 * 0.33 = 0.1 Population attributable risk % (of cases) What portion of cases in the population is caused by exposure? Only used for established, causal associations between exposure and disease PAR%= It – Io x 100 It = AR% x pc = RR-1 x pc x 100 RR Where pc= proportion of cases who are exposed  Population Attributable Risk % IT = 0.2 24/120 IO = 0.1 8/80 IE =0.4 16/40 What portion of cases in the population is caused by exposure?  Population Attributable Risk % IT = 0.2 24/120 IO = 0.1 8/100 IE =0.4 16/40 What portion of cases in the population is caused by exposure?  Cases pc= 16/24 AR% = 75% PAR% = AR% x pc = 75% * 2/3 = 50%  In Class Exercise VIII: PAR and PAR% Calculate and interpret the Population = incident case Attributable Risk (PAR) and Population = exposed Attributable Risk Percent (PAR%)  In Class Exercise IX If a given exposure has a PAR of 10 per 100 in the population, and an AR of 30 per 100 exposed in the population, what percentage of the population is exposed?  Population attributable risk Population attributable risk % Influenced by the prevalence of exposure in the population Can you apply a PAR or PAR% calculated in one population to another population? – Only if the prevalence of the exposure is the same in the two populations  In Class Exercise X Cohort study of US female smokers and non- smokers followed for lung cancer for 1 year Incidence per Lung Cancer Person-years 100,000 Cases Follow up person-years Ever smoker 2,000 800,000 250 Never smoker 150 600,000 25 What is the percent of lung cancer in US female smokers that is attributable to smoking? What is the percent of lung cancer in US females that is attributable to smoking? If 105,510 US women get lung cancer per year, how many cases could be avoided if smoking were eliminated? Modified from Freedman et al. Lancet Oncol 2008;9:649-656.  PAR% Example 60 Myocardial Infarction 50 Stroke PAR Percent 40 30 20 10 0 Hypertension Smoking Diabetes ApoB/ApoA1 ratio Modifiable Risk Factors 1. Interpret the population attributable risk percent for the various risk factors in words. 2. What is the sum of the PAR%s for MI due to all listed exposures? (Source: Endres, et al., Primary prevention of stroke: blood pressure, lipids, heart failure, Eur Hear J, 2011)  Estimating AR, AR%, and PAR% from Case Control Studies Case Control Study of Cooking and Lung Cancer Risk in non-smoking Taiwanese women (Ko, AJE, 2000) Photo: https://en.blog.kkday.com/20589/asia-taiwan-foodie-guide-tasty-eats Calculating AR from Case Control Studies You will not be tested on this. AR can only be calculated if: 1. Overall incidence of disease in the population It known or can be approximated; and 2. The frequency of exposure in controls pe reflects the frequency of exposure in the overall population that gave rise to the cases. 3. The OR is a good approximation of the RR Derivation on Text page 174 Calculating AR% from Case Control Studies The AR% can be estimated even if Ie and Io are unknown, but only if the OR is a good approximation of the RR. Calculating PAR% from Case Control Studies PAR% can be estimated only when OR is good approximation of RR, even if Ie and Io are not known. Only used for established, causal associations between exposure and disease 𝑃𝐴𝑅% = 𝐴𝑅% ∗ 𝑝! 𝑅𝑅 − 1 𝑃𝐴𝑅% = ∗ 𝑝! ∗ 100 𝑅𝑅 Where pc= proportion of cases who are exposed In Class Exercise XI: Estimating AR% from Case Control Study Lung cancer # meals Present Absent cooked/day 2+ 140 200 1 10 50 Calculate and interpret the AR% given that the OR is a good approximation of RR in this scenario. Modified from Ko et al. Am J Epidemiol 2000;151:140-7 In Class Exercise XII: Estimating PAR% from Case Control Study Lung cancer # meals Present Absent cooked/day 2+ 140 200 1 10 50 Calculate and interpret PAR% and compare to the AR%.  Measures of excess risk Measure Abb. Formula Question Risk ratio RR Ie / Io Is E related to D? Odds ratio OR ad/bc Is E related to D? Risk difference RD Ie - Io Is E related to D? Attributable risk to AR Ie - Io What % of the exposed the exposed develop D due to E? (pe=freq. of E in controls) Attributable risk % AR% AR/Ie*100 What % of cases among the exposed are RR-1 x 100 due to E? RR  Measures of excess risk Measure Abb. Formula Question Population PAR IT - Io What % of the Attributable Risk population develop D due to E? Population PAR% PAR/IT*100 What % of cases in Attributable Risk % the population are AR% x pc due to E? (pc=freq. of E in cases) Special excess risk measures  Mortality Rate Ratio # 𝑜𝑓 𝐹𝑎𝑡𝑎𝑙 𝐶𝑎𝑠𝑒𝑠 𝑀𝑜𝑟𝑡𝑎𝑙𝑖𝑡𝑦 𝑅𝑎𝑡𝑒 = 𝑇𝑜𝑡𝑎𝑙 𝑝𝑒𝑟𝑠𝑜𝑛 − 𝑡𝑖𝑚𝑒 𝑎𝑡 𝑟𝑖𝑠𝑘 𝑓𝑜𝑟 𝑑𝑒𝑎𝑡ℎ ∗ * Includes both prevalent cases and also healthy individuals at risk of developing and dying from the disease. 𝑀𝑜𝑟𝑡𝑎𝑙𝑖𝑡𝑦 𝑅𝑎𝑡𝑒 𝑓𝑜𝑟 𝐺𝑟𝑜𝑢𝑝 𝐴 𝑀𝑜𝑟𝑡𝑎𝑙𝑖𝑡𝑦 𝑅𝑎𝑡𝑒 𝑅𝑎𝑡𝑖𝑜 = 𝑀𝑜𝑟𝑡𝑎𝑙𝑖𝑡𝑦 𝑅𝑎𝑡𝑒 𝑓𝑜𝑟 𝐺𝑟𝑜𝑢𝑝 𝐵  Mortality Rate Ratio Age-Adjusted Heart Disease Mortality Rates in the U.S. for 2000-2014 (Data from CDC) Sex Mortality Rate (per 100,000) Female 131.8 Male 210.9 ABC.E Mortality Rate Ratio= = 1.60 BFB.G  Mortality Rate Ratio Age-Adjusted Heart Disease Mortality Rates in the U.S. for 2000-2014 (Data from CDC) Sex Mortality Rate (per 100,000 py) Female 131.8 Male 210.9 ABC.E Mortality Rate Ratio= = 1.60 BFB.G In the US between 2000-2014, the incidence rate of death from heart disease was 60% higher in males compared to females of a similar age. or In the US between 2000-2014, the incidence rate of death from heart disease in males was 1.6 times that of females of a comparable age.  Infant Mortality Rates Infant mortality rate is used to describe the frequency of deaths in children