Lecture Learning Objectives

1. Lecture Learning Objectives • Review epidemiology research methods • Learn how risk factors for disease are identified • Understand the difference in study designs • Relate study design to strength of the scientific evidence

2. Lecture Tools • Hyperlinks in blue take you to • Web sites with additional information on a topic • References for additional review • Often no more information will be offered in this lecture on that topic so use of those links is highly recommended

3. Descriptive Epidemiology • Case Studies or Case Series • Published reports - single or multiple cases of disease • Weak data because only cases discussed • No evidence to compare cases with controls • Surveillance • Monitoring of disease or behaviors • Different methods allow for more analysis -- national surveys

4. Analytic Epidemiology Studies • Observational studies • Cross-sectional surveys • Case-control • Cohort • Experimental or Intervention studies • All of these study types collect information about outcomes and exposures • Outcome -- a disease or possibly behavior of interest • Exposure -- a characteristic being examined as a risk factor or cause of disease, or a preventive factor for disease • Exposures must precede outcomes to be a risk or preventive factor

5. Study Variables - Categorical • Nominal • Groups of persons based on a characteristic • Gender, marital status, occupational status, disease status • Ordinal • Also categorical but there is an order to the groupings • Disease severity -- mild, moderate, severe • Age groupings

6. Study Variables - Categorical • Binary of dichotomous • yes or no, • white or non-white, • disease or no disease, • exposed or not exposed • Multichotomous • White, black, Asian, other • 18-34, 35-54, 55+

7. Study Variables - Continuous (numeric) • Number of cigarettes per day or per week • Hair nicotine level • Cotinine level • Number of times use ecstasy in a week • Likert scale -- agree totally (1) to disagree totally (5) • Scale calculation for rebelliousness

8. Very Brief Statistics • Statistics differ for different types of variables -- continuous versus categorical • Continuous variables • Means - T-test or ANOVA compares means in different groups • Correlation looks at relationship between 2 continuous variables • Linear regression based on correlation - control for confounding with continuous outcomes • Categorical variables • Frequencies • Chi Square compares frequency distributions in two or more groups • Logistic regression controls for confounding with categorical outcome

9. Analytic Epidemiology Studies • All of these study types end up with four basic groups of participants based on exposure and disease • 2 X 2 table

10. Analytic Epidemiology Studies • Goal is to determine ‘strength of association’ between exposures and outcomes • Does the presence of the exposure increase the level of disease or behavior? • Does poor parenting increase drug use in teens? • Does the presence of the exposure decrease the level of disease or behavior? • Does a strong parent-child relationship decrease drug use in teens? • Does the use of methadone decrease heroin use? • Association -- when the presence of factor 1 statistically increases or decreases the level of factor 2 we say the two factors are associated

11. Very Brief Statistics • Association -- measured in different ways • Correlation • Measures association between two continuous variables • Does not give a good indication of the ‘strength’ of an association • Relative Risks or Odds Ratios • Measures association by looking at rates of disease or frequency of exposures among groups • Gives a measure of the ‘strength’ of the association • Allows you to prioritize risks

12. More Very Brief Statistics • Most studies are done on samples, so all results are only estimates • We can never prove an association -- we can just provide evidence to support or refute the hypothesis of an association • We use statistics to give us an idea of how TRUE a finding is • P-value or probability value -- the probability that the finding is due to chance alone

13. Sample vs Population

14. Research Errors

15. Research Errors

16. Type I or Alpha Errors • To report that an association between an exposure and outcome exists when in truth it does not • Worst type of error to make • So -- we use an alpha level of 0.05 • If a statistical procedure reports a p-value of <0.05, then we say there is “less than a 5% probability the results are due to chance alone” • Confidence level = 1-alpha or 95% • There is a 95% probability that the finding is true • the level of confidence in the results • confidence we are not making an alpha error

17. Type II or Beta Errors • To report that an association between an exposure and outcome does not exist when in truth it does • The least serious error to make • In order to minimize this type of error, we increase our sample size -- the larger our sample, the less likely we are to miss something • Statistical calculation is done to tell you a minimum sample size needed depending on the study question • We use a Beta error of 20% • Study power = 1-Beta error • So -- a study with 80% power will make a Beta error only 20% of the time

18. Observational Studies • Individuals are observed -- no manipulation of exposures by researchers • Data gathered from participants about exposures and disease • Retrospective -- ask questions about past history of exposure and disease onset • Prospective -- ask questions at baseline and follow persons over time • Used primarily to identify risk factors but are important in situations where randomized, controlled trials (RCTs) are impossible or unethical

19. Observational Studies - Basic Types • Cross-sectional • Case-Control • Cohort • All gather data on exposures and disease and look for associations • Some variations / mixed methods also seen

20. Cross-Sectional Studies • Usually surveys • Gather information about both exposures and diseases • Cannot confirm which came first -- exposure or disease • Cannot confirm causation but can indicate strength of associations between exposures and diseases

21. Cross-Sectional Studies Mike Edmond, Population Medicine Lecture, VCU

22. Cross-Sectional Survey Data • Prevalence -- can only look at the presence or absence of disease or behavior at the time the question is asked • If survey sample is representative of a population -- prevalence of disease or behavior among participants can be applied to that larger population • No information on incidence of disease • No temporal information to determine causation -- can only determine association

23. Case-Control Studies • Exposure data is retrospective • Compare: • Frequency of exposure among persons with disease -- “cases” • Frequency of exposure among persons without disease -- “controls” • http://www.pitt.edu/~super1/lecture/lec8591/index.htm

24. Case-Control Studies • **Develop hypothesis of relationship between exposure and outcome -- can be causative or preventive • Alcohol is a risk factor for lymphoma • Identify ‘cases’ based on outcome • Persons with a disease of interest • Persons with a behavior of interest • Identify ‘controls’ • Persons similar to cases except no disease of interest • Might match on already known risk factors -- often age, sex

25. Case-Control Studies Mike Edmond, Population Medicine Lecture, VCU

26. Case-Control Studies • Ask same questions of both participant groups • Retrospective data gathering regarding the presence of potential exposures • Can look at many exposures with one outcome • Look for higher frequency of exposure among cases compared to controls if exposure a risk for disease • Look for lower frequency of exposure among cases compared to controls if exposure is protective

27. Case-Control Studies-Data • Frequency of exposure by different characteristics • Compare sub-groups of your sample • Identify the odds of exposure among cases and compare to odds of exposure among controls to get a measure of association • Odds Ratio -- measure of association

28. Cohort or Longitudinal Studies • Best of the observational study types • Identify cohorts with and without exposure of interest AND disease free • Follow over time to identify disease incidence in both cohorts • Is incidence of disease higher in the exposed cohort? • http://www.pitt.edu/~super1/lecture/lec8581/index.htm

29. Cohort Studies • Develop hypothesis of relationship between exposure and outcome • Identify a cohort of persons who are exposed but do NOT have the outcome of interest • Identify a comparison cohort who do NOT have exposure and do NOT have outcome of interest • Another option is to identify a large cohort of persons and ask about exposure -- then divide the large group into exposed and not-exposed

30. Cohort Studies Large cohort of persons Mike Edmond, Population Medicine Lecture, VCU

31. Cohort Studies • Start with known exposure status so can confirm exposure came first and look forward for disease onset • Because persons are all disease freeat start -- can calculate incidence of disease in both groups -- IR • If exposure is a risk factor / cause of disease, incidence will be higher in exposed group than the non-exposed • If exposure is protective, incidence will be lower in exposed group • Comparison of IR gives you a measure of association between exposure and disease -- Relative Risk

32. Measures of Association and Burden • Goal of most epidemiology studies • Relative Risk (RR), Odds Ratio (OR) • Measures to determine if an “exposure” is statistically associated with increased (or decreased) disease • Gives an idea of the strength of the association-- stronger association more likely to be causative • Attributable Risk (AR), Population Attributable Risk (PAR) • Give an indication of risks that cause greatest burden in populations • http://www.bmj.com/epidem/epid.3.dtl#pgfId=1001728

33. Relative Risk (RR) • Is the best way to determine if something is a risk factor • Also called Risk Ratio or Rate Ratio • Ratio of the incidence rate (IR) of disease in exposed to the IR of disease in unexposed • Used with cohort studies • RR = IR of lung cancer in smokers IR of lung cancer in non-smokers

34. Cohort Study - RR • Entire population divided into 4 groups - a, b, c, d • IR of exposed = a / a + b • IR of unexposed = c / c + d

35. Cohort Study - RR • Entire population divided into 4 groups - a, b, c, d • IR of exposed = a / a + b • IR of unexposed = c / c + d

36. Relative Risk (RR) • RR = IR in exposed IR in unexposed • RR = 1 means no difference in rate of disease -- so exposure is not a risk • RR > 1 means more disease in exposed -- indicates exposure is a risk factor for disease • RR < 1 means less disease in exposed -- indicates exposure not a risk factor and might protect against disease

37. Odds Ratios (OR) • Estimate of a Relative Risk • Used in case-control studies where incidence and rates of disease cannot be calculated • Same interpretation as RR • OR = 1 means no difference in odds of exposure -- exposure is not a risk • OR > 1 means more exposure in diseased -- indicates exposure is a risk factor for disease • OR < 1 means less exposure in diseased -- indicates exposure not a risk factor and might protect against disease

38. Case-Control OR • Ratio of odds (or probability) of exposure in cases to odds of exposure in controls

39. Case-Control OR • Ratio of odds (or probability) of exposure in cases to odds of exposure in controls a/c = a X d b/d b X c

40. Measures of Association -- Confidence • Remember - calculated ORs or RRs is only an estimate • So we calculate 95% confidence intervals • RR +/- CI (or OR +/- CI) -- from minimum to maximum value --gives an estimate of what the TRUE RR is 95% of the time • Be sure a researcher reports an RR or OR WITH CIs. • If the CIs include ‘1’ than it is possible the RR includes ‘1’ and NO increased risk can be confirmed

41. Measures of Association -- Confidence • If our RR is 2.0 • Suggests that exposure doubles the risk for disease • In a large study, might get a CI = 0.5 • RR=2.0 with a true range of 1.5 - 2.5 • Include ‘1’? • In a small study, might get a CI = 1.0 • RR=2.0 with a true range of 1.0-3.0 • Include ‘1’? • Conclusion? • Some researchers use p-values - if the p-value is <0.05 -- you assume the CIs do NOT include ‘1’

43. Levels of Evidence for Interventions • Anecdote • Personal experience • Respected authority • Published case report • Published descriptive studies • Observational • Cross-sectional • Case-control • Cohort

44. Cigarette smoking, alcohol drinking, and risk of lymphoid neoplasms: results of a French case control study. Monnereau et al. Cancer Causes Control (2008) 19:1147-1160 • Abstract • Objective: To study potential role of smoking and alcohol in lymphoid neoplasms (LN). • Methods: A case–control study that included 824 cases and 752 hospital controls aged 18–75 years was conducted. Cases were newly diagnosed with non-Hodgkin’s or Hodgkin’s lymphoma, multiple myeloma, or lymphoproliferative syndrome (LPS). Controls were matched with the cases by gender, age, and center. • Results: Overall, smoking was not related to LN. However, average tobacco consumption tended to be inversely related to non-Hodgkin’s lymphoma (NHL), LPS, and the hairy cell leukemia (HCL) subtype, with a significant negative trend for the latter (OR of 0.4, 0.2, 0.1 for consumptions of B10, 11–20, [20 cig/day). An inverse association between ‘ever drinking’ and Hodgkin’s lymphoma (HL: OR = 0.5 [0.3–0.8]) and NHL (OR = 0.7

46. Hair et al. . Risky Behaviors in Late Adolescence: Co-occurrence, Predictors, and Consequences Journal of Adolescent Health 45 (2009) 253-261 • Purpose:Advances in research have broadened our understanding of the risky behaviors that significantly threaten adolescent health and well-being. Advances include: using person-centered, rather than behavior-centered approaches to examine how behaviors co-occur; greater focus on how environmental factors, such as family, or peer-level characteristics, influence behavior; and examination of how behaviors affect well-being in young adulthood. Use of nationally representative, longitudinal data would expand research on these critical relationships. • Methods:Using data from the National Longitudinal Survey of Youth, 1997 cohort, a nationally representative sample of adolescents who are being followed over time, the present study: (1) identifies profiles of risky behaviors, (2) investigates how environmental characteristics predict these profiles of risky behaviors (e.g., delinquency, smoking, drug use, drinking, sexual behavior, and exercise), and (3) examines how these profiles of risky behaviors relate to positive and negative youth outcomes. • Results:Four risk profiles were identified: a high-risk group (those who report high levels of participation in numerous behaviors), a low-risk group (those who engage in very few risky behaviors), and two moderate risk-taking groups. We found that profiles with any negative behaviors were predictive of negative outcomes. • Conclusions:It is important for practitioners to examine health behaviors in multiple domains concurrently rather than individually in isolation. Interventions and research should not simply target adolescents engaging in high levels of risky behavior but also adolescents who are engaging in lower levels of risky behavior. http://www.bls.gov/nls/nlsy79.htm