2009 Joint Statistical Meetings Washington, DC August 1-6, 2009 Interpreting Differential Effects in Light of Fundamental Statistical Tendencies James.

Slides:



Advertisements
Similar presentations
How would you explain the smoking paradox. Smokers fair better after an infarction in hospital than non-smokers. This apparently disagrees with the view.
Advertisements

2008 Joint Statistical Meetings Denver, Colorado, August 2-7, 2008 Evaluating the Size of Differences Between Group Rates in Settings of Different Overall.
Comparing Two Proportions (p1 vs. p2)
American Public Health Association 138 th Annual Meeting Denver, Colorado Nov. 6-10, 2010 The Emerging European Acceptance of Scanlan’s Rule in Health.
Copyright ©2005 Brooks/Cole, a division of Thomson Learning, Inc. Relationships Between Categorical Variables Chapter 12.
Confidence Intervals, Effect Size and Power
Race and Socioeconomic Differences in Health Behavior Trajectories Across the Adult Life Course ACKNOWLEDGEMENTS This research was supported by the grant.
The Misunderstood Relationship Between Declining Mortality and Increasing Racial and Social Disparities in Mortality Rates James P. Scanlan (Presented.
Family Size and Family Structure Lecture 12 Subtitle: Trends in Births and Births Rates.
Incorporating considerations about equity in policy briefs What factors are likely to be associated with disadvantage? Are there plausible reasons for.
BS704 Class 7 Hypothesis Testing Procedures
Compare Outcomes Using all the above specific categories, we could compare 0-4 year-old male Asian mortality rates for asthma with 0-4 Asian female rates.
Hypothesis Testing. Outline The Null Hypothesis The Null Hypothesis Type I and Type II Error Type I and Type II Error Using Statistics to test the Null.
Measuring population development from social cohesion perspective by women and men according to the Census data Urve Kask Statistics Estonia.
Fundamentals of Hypothesis Testing: One-Sample Tests
Multiple Choice Questions for discussion
Measurement Problems in the National Healthcare Disparities Report American Public Health Association 135 th Annual Meeting & Exposition, Nov. 3-7,2007,
Evidence-Based Medicine 3 More Knowledge and Skills for Critical Reading Karen E. Schetzina, MD, MPH.
Comparing Two Population Means
Evidence Based Medicine
Lecture 7 Introduction to Hypothesis Testing. Lecture Goals After completing this lecture, you should be able to: Formulate null and alternative hypotheses.
The Impact of Inequality on Personal Life Chances Roderick Graham Fordham University.
Kansas Department of Health and Environment Center for Health Disparities 2008 Health Disparities Conference Topeka, Kansas, Apr. 1, 2008 Measuring Health.
Measures of Association
January 31 and February 3,  Some formulae are presented in this lecture to provide the general mathematical background to the topic or to demonstrate.
© 2008 McGraw-Hill Higher Education The Statistical Imagination Chapter 10. Hypothesis Testing II: Single-Sample Hypothesis Tests: Establishing the Representativeness.
Copyright ©2006 Brooks/Cole, a division of Thomson Learning, Inc. Relationships Between Categorical Variables Chapter 6.
Understanding Variations in Group Differences that are the Results of Variations in the Prevalence of an Outcome American Public Health Association 134.
16 th Nordic Demographic Symposium Helsinki, Finland, 5-7 June 2008 Measures of Health Inequalities that are Unaffected by the Prevalence of an Outcome.
Royal Statistical Society 2009 Conference Edinburgh, Scotland 7-11 September 2009 Measuring Health Inequalities by an Approach Unaffected by the Overall.
7 th International Conference on Health Policy Statistics, Philadelphia, PA, January 17-18, 2008 Can We Actually Measure Health Disparities? James P. Scanlan.
Educational Research Chapter 13 Inferential Statistics Gay, Mills, and Airasian 10 th Edition.
Measuring Healthcare Disparities Third North American Congress of Epidemiology Montreal, Quebec, June 21-24, 2011 James P. Scanlan Attorney at Law Washington,
British Society for Population Studies 2007 Annual Conference St. Andrews, Scotland, Sep 2007 Methodological Issue in Comparing the Size of Differences.
1 Multivariable Modeling. 2 nAdjustment by statistical model for the relationships of predictors to the outcome. nRepresents the frequency or magnitude.
Overview of Regression Analysis. Conditional Mean We all know what a mean or average is. E.g. The mean annual earnings for year old working males.
Organization of statistical research. The role of Biostatisticians Biostatisticians play essential roles in designing studies, analyzing data and.
The Inflammatory Breast Cancer Cancer Registry Paul H. Levine, M.D. Paul H. Levine, M.D. The George Washington University School of Public Health and Health.
CROSS SECTIONAL STUDIES
BIOSTATISTICS Lecture 2. The role of Biostatisticians Biostatisticians play essential roles in designing studies, analyzing data and creating methods.
The Nutrition Transition Program The University of North Carolina at Chapel Hill Ethnic Differences in the Association Between Body Mass Index and Hypertension.
Sources of Increasing Differential Mortality among the Aged by Socioeconomic Status Barry Bosworth, Gary Burtless and Kan Zhang T HE B ROOKINGS I NSTITUTION.
Nursing 4604L Kimberly A. Rogers, RN Healthcare for an Aggregate at Risk Males in Pasco County, Florida Coronary Heart Disease Among Males In Pasco County,
Objectives (Chapter 20) Comparing two proportions  Comparing 2 independent samples  Confidence interval for 2 proportion  Large sample method  Plus.
Reducing Tobacco Intake Lowers Risk of Lung Cancer in Heavy Smokers Slideset on: Godtfredsen NS, Prescott E, Osler M. Effect of smoking reduction on lung.
Summary of “A randomized trial of standard versus intensive blood-pressure control” The SPRINT Research Group, NEJM, DOI: /NEJMoa Downloaded.
20. Comparing two proportions
for Overall Prognosis Workshop Cochrane Colloquium, Seoul
Cigarette Smoking in the United States
Hypothesis Testing.
Long-Term Follow-Up of Patients Receiving Lung-Volume-Reduction Surgery Versus Medical Therapy for Severe Emphysema by the National Emphysema Treatment.
Health and Human Services National Heart, Lung, and Blood Institute
Does Public Health Insurance Affect How Much People Work?
The Misinterpretation of Health Inequalities in the United Kingdom
The Misinterpretation of Health Inequalities in Nordic Countries
Hypothesis Testing Review
Basic Statistics Overview
James P. Scanlan Attorney at Law Washington, DC, USA
Long-Term Follow-Up of Patients Receiving Lung-Volume-Reduction Surgery Versus Medical Therapy for Severe Emphysema by the National Emphysema Treatment.
Sampling Distributions
Kanguk Samsung Hospital, Sungkyunkwan University
James P. Scanlan Attorney at Law Washington, DC, USA
Part 3: Weighting Estimation Samples Frank Porell
Increasing Disparity: The Scanlan Effect
Contributions to Cardiovascular Disease Prediction in the Women’s Health Study* Nancy R. Cook, et al. Circulation 2007;115:
9th International Conference on Health Policy Statistics
Measuring Health Disparities in Healthy People 2010
Relationships among Inflation, Interest Rates and Exchange Rates
Peterborough 1 PCN Data pack August 2019.
Presentation transcript:

2009 Joint Statistical Meetings Washington, DC August 1-6, 2009 Interpreting Differential Effects in Light of Fundamental Statistical Tendencies James P. Scanlan Attorney at Law Washington, DC, USA Oral at

Summary 1. Factors that similarly affect two groups with different baseline rates of an outcome will tend to show a larger proportionate effect on the outcome for the group with the lower base rate but a larger proportionate effect on the opposite outcome for the other group 2. True subgroup effects can only be identified by determining the degree to which a factor shifts each group’s risk distribution

References Measuring Health Disparities page on jpscanlan.com – especially the Solutions tab Scanlan’s Rule page on jpscanlan.com – especially the Subgroup Effects tab Can We Actually Measure Health Disparities? (Chance 2006) Race and Mortality (Society 2000) Divining Difference (Chance 1994)

An illogical expectation There exists a tendency to regard it as somehow normal that a factor that similarly affects two groups’ susceptibilities to an outcome will cause the same proportionate change in the outcome rates for each group and to regard anything else as a differential effect (subgroup effect, interaction)

One reason the expectation is illogical Where two groups have different base rates of experiencing an outcome, a factor that has the same proportionate effect on each group’s rate of experiencing the outcome necessarily has a different proportionate effect on each group’s rate of avoiding the outcome E.g., Group A (10/90); Group B (20/80)

A logical expectation A factor that similarly affects two groups with different baseline rates of an outcome will tend to show a larger proportionate effect on the outcome for the group with the lower base rate but a larger proportionate effect on the opposite outcome for the other group Why is that logical?

Interpretive Rule 1/Scanlan’s Rule 1 When two groups differ in their susceptibility to an outcome, the rarer the outcome, the greater tends to be the relative difference in experiencing it and the smaller tends to be the relative difference in avoiding it.

Fig 1. Ratio of (1) Disadvantaged Group (DG) Fail Rate to Advantaged Group (AG) Fail Rate at Various Cutoff Points Defined by AG Fail Rate

Fig. 2. Ratios of (1) DG Fail Rate to AG Fail Rate and (2) AG Pass Rate to DG Pass Rate at Various Cutoff Points Defined by AG Fail Rate

Fig 3. Ratios of (1) Bl Rate of Falling below Various Income Levels to Wh Rate of Falling below Level and (2) Wh Rate of Falling above Level to Bl Rate of Falling above Level

Fig. 4. Ratio of (1) Bl to Wh Rate of Falling above Various SBP Levels and (2) Wh to Bl Rate of Falling below the Level (NHANES , , Men 45-64)

Implications of SR1 As an adverse becomes less prevalent relative difference in experiencing it tend to increase while relative differences in avoiding it tend to decline (e.g., test failure, poverty, hypertension, mortality). Relative differences in adverse outcomes tend to be larger among advantaged subpopulations while relative differences in favorable outcomes tend to be smaller among such subpopulations (e.g. race differences in health outcomes among the college-educated, race and SES differences in health outcomes among the young, SES differences in health outcomes among British Civil servants) Factors that decrease/increase the prevalence of an outcome will tend to have greater proportionate effect on group with lower base rate but greater proportionate effect on the opposite outcome for the other group

Problems Arising from SR1 When we observe standard patterns of changes in differences between group rates as prevalence of an outcome changes, how do we determine whether the difference between the groups’ situations actually changed in a meaningful way and how do we quantify the difference at each point in time (in a way that is unaffected by overall prevalence)? How do we identify genuine differential effects?

Estimated Effect Size Estimated effect size (EES) = estimated difference, in terms of percentage of a standard deviation, between means of hypothesized underlying, continuously- scaled normal distributions of factors associated with experiencing an outcome, derived from outcome rates of each group (see JSM 2008 and Solutions and Solutions Database tabs on jpscanlan.com)

Table 2 Illustration of the Solution (in terms of an adverse outcome decreasing in overall prevalence) Period Yr 1 Yr 2 AG Rate60% 42% DG Rate77% 60% Measures of Difference Change Direction AdvRatio Increase FavRatio Decrease EES (z) Decrease

Illustrations from Two Perspectives Perspective 1: Identifying/evaluating differential effects of factor on two groups Perspective 2: Comparing the size of the difference between the rates of two groups according to the presence or absence of a factor

Table 3. Comparison of Effects of Hypertension Control on Heart Attack Risk of Women and Men with Similar Risk Factor Profiles (A65,TC300,HDL50,NS, NM) (Perspective 1) Gender SBP 120 Risk SPB150 Risk Adverse Reduction Favorable IncreaseEES F4.0%7.0%42.9%3.2%0.28 M14.0%19.0%26.3%6.2%0.21

Table 4. Comparison of Gender Differences in Heart Attack Risk for Women and Men with Similar Risk Factor Profiles, by with and without Hypertension Control (A65,TC300,HDL50,NS, NM) (Perspective 2) SBP Fem Rate Male Rate Adverse Ratio Favorable RatioEES %14.0% %19.0%

Table 5: Comparison of Black-White Difference in Lung Cancer among 29 Cig Per Day (Haimon NEJM 2006) (Perspective 2) Cig/DayWhite RateBlack Rate Adverse Ratio Favorable RatioEES <110.7%2.2% >293.1%5.4%

Table 6. Comparison of Effects of Decreasing Smoking from >29 to <11 Cig Per Day on Whites and Blacks (Haimon NEJM 2006) (Perspective 1) (cor. 9/20/14) Race<11 CPD>29 CPD Adverse Reduction Favorable IncreaseEES W0.7%3.1%77.2%2.5%0.60 B2.2%5.9%63.4%4.0%0.42

Table 7. Comparison of Effects of Beta Blockers on Mortality Among Heart Patients at Different Ages (Gottlieb NEJM 1998) (Perspective 1) AgeBeta RateNo Beta Rate Adverse Reduction Favorable IncreaseEES <7011.3%18.7%39.6%9.1%0.34 >8022.6%33.1%31.7%15.7%0.32

Table 8. Comparison of Age Differences in Mortality Among Patients Treated and Not Treated with Beta Blockers (Gottlieb NEJM 1998) (Perspective 2) Treatment Status<70 Rate>80 Rate Adverse Ratio Favorable RatioEES Beta11.3%22.6% NoBeta18.7%33.1%

Issues Contrary literature regarding the patterns Clinical significance issue Statistical significance issues Re spurious patterns Re solutions approach Absolute minimum issue