1 EPI235: Epi Methods in HSR April 12, 2007 L4 Evaluating Health Services using administrative data 3: Advanced Topics on Risk Adjustment and Sensitivity.

Slides:



Advertisements
Similar presentations
Agency for Healthcare Research and Quality (AHRQ)
Advertisements

Comparator Selection in Observational Comparative Effectiveness Research Prepared for: Agency for Healthcare Research and Quality (AHRQ)
Controlling for Time Dependent Confounding Using Marginal Structural Models in the Case of a Continuous Treatment O Wang 1, T McMullan 2 1 Amgen, Thousand.
Department of Health and Human Services Center for Drug Evaluation and Research Review of Epidemiologic Studies on Cardiovascular Risk with Selected NSAIDs.
Improving health worldwide George B. Ploubidis The role of sensitivity analysis in the estimation of causal pathways from observational.
Bias in Clinical Research: Measurement Bias
Sensitivity Analysis for Observational Comparative Effectiveness Research Prepared for: Agency for Healthcare Research and Quality (AHRQ)
“Personality, Socioeconomic Status, and All-Cause Mortality in the United States” - Chapman BP et al. Journal Club 02/24/11.
PHSSR IG CyberSeminar Introductory Remarks Bryan Dowd Division of Health Policy and Management School of Public Health University of Minnesota.
ODAC May 3, Subgroup Analyses in Clinical Trials Stephen L George, PhD Department of Biostatistics and Bioinformatics Duke University Medical Center.
Estimation and Reporting of Heterogeneity of Treatment Effects in Observational Comparative Effectiveness Research Prepared for: Agency for Healthcare.
From last time….. Basic Biostats Topics Summary Statistics –mean, median, mode –standard deviation, standard error Confidence Intervals Hypothesis Tests.
Chapter 10 Simple Regression.
Measures of association
1 EPI235: Epi Methods in HSR March 31, 2005 L2 Evaluating Health Services using administrative data 1: Introduction to Risk Adjustment (Dr. Schneeweiss)
Introduction to Biostatistics
Cohort Studies Hanna E. Bloomfield, MD, MPH Professor of Medicine Associate Chief of Staff, Research Minneapolis VA Medical Center.
Validation of predictive regression models Ewout W. Steyerberg, PhD Clinical epidemiologist Frank E. Harrell, PhD Biostatistician.
Chapter 14 Inferential Data Analysis
THREE CONCEPTS ABOUT THE RELATIONSHIPS OF VARIABLES IN RESEARCH
Sample Size Determination Ziad Taib March 7, 2014.
As noted by Gary H. Lyman (JCO, 2012) “CER is an important framework for systematically identifying and summarizing the totality of evidence on the effectiveness,
Covariate Selection for Observational Comparative Effectiveness Research Prepared for: Agency for Healthcare Research and Quality (AHRQ)
STrengthening the Reporting of OBservational Studies in Epidemiology
Unit 6: Standardization and Methods to Control Confounding.
Comparing high-dimensional propensity score versus lasso variable selection for confounding adjustment in a novel simulation framework Jessica Franklin.
BASIC STATISTICS: AN OXYMORON? (With a little EPI thrown in…) URVASHI VAID MD, MS AUG 2012.
Advanced Statistics for Interventional Cardiologists.
Intervention Studies Principles of Epidemiology Lecture 10 Dona Schneider, PhD, MPH, FACE.
Improved Treatment of Ischemic Heart Disease and Disability and Death in the Elderly Kate Stewart Mary Beth Landrum David Cutler Academy Health June 27,
Reduced Risks of Neural Tube Defects and Orofacial Clefts With Higher Diet Quality Carmichael SL, Yang W, Feldkamp ML, et al; National Birth Defects Prevention.
FDA Approach to Review of Outcome Measures for Drug Approval and Labeling: Content Validity Initiative on Methods, Measurement, and Pain Assessment in.
CHP400: Community Health Program- lI Research Methodology STUDY DESIGNS Observational / Analytical Studies Case Control Studies Present: Disease Past:
 Is there a comparison? ◦ Are the groups really comparable?  Are the differences being reported real? ◦ Are they worth reporting? ◦ How much confidence.
Quality Through the Eyes of the Patient: State-of-the-Art Concepts Paul D. Cleary, Ph.D. April 10, 2001 Quality Through the Eyes of the Patient: State-of-the-Art.
Sensitivity Analysis for Residual Confounding
Article Review Cara Carty 09-Mar-06. “Confounding by indication in non-experimental evaluation of vaccine effectiveness: the example of prevention of.
Slides to accompany Weathington, Cunningham & Pittenger (2010), Chapter 3: The Foundations of Research 1.
Plymouth Health Community NICE Guidance Implementation Group Workshop Two: Debriding agents and specialist wound care clinics. Pressure ulcer risk assessment.
Approaches to the measurement of excess risk 1. Ratio of RISKS 2. Difference in RISKS: –(risk in Exposed)-(risk in Non-Exposed) Risk in Exposed Risk in.
1 G Lect 14M Review of topics covered in course Mediation/Moderation Statistical power for interactions What topics were not covered? G Multiple.
LEADING RESEARCH… MEASURES THAT COUNT Challenges of Studying Cardiovascular Outcomes in ADHD Elizabeth B. Andrews, MPH, PhD, VP, Pharmacoepidemiology and.
Copyright © 2010, 2007, 2004 Pearson Education, Inc. All Rights Reserved. Section 7-1 Review and Preview.
1 Risk Assessment Tests Marina Kondratovich, Ph.D. OIVD/CDRH/FDA March 9, 2011 Molecular and Clinical Genetics Panel for Direct-to-Consumer (DTC) Genetic.
Interpreting observational studies of cardiovascular risk of NSAIDs. Richard Platt, MD, MS Harvard Medical School and Harvard Pilgrim Health Care HMO Research.
How confident are we in the estimation of mean/proportion we have calculated?
Case-Crossover Studies.
Measuring covariate data_Presentation (November 14, 2007) 1 Measuring covariate data in subsets of study populations: Design options Jean-François Boivin,
Chapter 8 Parameter Estimates and Hypothesis Testing.
Case-Control Studies Abdualziz BinSaeed. Case-Control Studies Type of analytic study Unit of observation and analysis: Individual (not group)
© Copyright McGraw-Hill 2004
2007Mar201 Journal Club for Analysis of Complex Datasets Cui Y, Shu X-O, Gao Y-T, Cai H, Tao M-H, Zheng W. Assocation of ginseng use with survival and.
1 EPI235: Epi Methods in HSR April 5, 2005 L3 Evaluating Health Services using administrative data 2: Advanced Topics in Risk Adjustment (Dr. Schneeweiss)
Instructor Resource Chapter 15 Copyright © Scott B. Patten, Permission granted for classroom use with Epidemiology for Canadian Students: Principles,
Matching. Objectives Discuss methods of matching Discuss advantages and disadvantages of matching Discuss applications of matching Confounding residual.
CASE CONTROL STUDY. Learning Objectives Identify the principles of case control design State the advantages and limitations of case control study Calculate.
Bayesian methods in epidemiological research JONAS BJÖRK, LUND UNIVERSITY. 5 FEBRUARY 2016.
Transparency in the Use of Propensity Score Methods
How to Conduct a Meta-Analysis Arindam Basu MD MPH About the Author Required Browsing.
Analysis of Mismeasured Data David Yanez Department of Biostatistics University of Washington July 5, 2005 Biost/Stat 579.
Challenges to the Epidemiology of Aging: The REasons for Geographic And Racial Differences in Stroke Study George Howard, DrPH UAB School of Public Health.
Methodological quality assessment of observational studies Nicole Vogelzangs Department of Psychiatry & EMGO + institute.
Systematic review of the potential adverse effects of caffeine consumption in healthy adults, pregnant women, adolescents, and children: Cardiovascular.
Inference: Conclusion with Confidence
OHDSI Method Evaluation
Date:2017/10/03 Presenter: Wen-Ching Lan
Effect of Obesity on In-Hospital Mortality in Patients with Cardiogenic Shock Complicating AMI Obesity is paradoxically associated with favorable mortality.
Interpreting Basic Statistics
Chapter 3 Hernán & Robins Observational Studies
Presentation transcript:

1 EPI235: Epi Methods in HSR April 12, 2007 L4 Evaluating Health Services using administrative data 3: Advanced Topics on Risk Adjustment and Sensitivity Analysis (Dr. Schneeweiss) Risk adjustment in studies using administrative databases is limited to observed confounders. Dr. Schneeweiss will illustrate theory and practice of assessing the sensitivity of epidemiologic risk estimates towards unobserved confounding. An interactive Excel program will be used for illustration. Background reading: Walker AM: Observation and inference, Chapter 9. Epidemiology Resources, Newton Lower Falls, Schneeweiss S, Glynn RJ, Tsai EH, Avorn J, Solomon DH. Adjusting for unmeasured confounders in pharmacoepidemiologic claims data using external information: The example of COX2 inhibitiors and myocardial infarction. Epidemiology 2005;16:17-24.

2 Unmeasured (residual) Confounding  Confounding factors that are not measured are hard to adjust for in observational analyses  If unadjusted they lead to residual confounding

3 Unmeasured (residual) Confounding: [smoking, healthy lifestyle, etc.] Drug exposure Outcome RR EO OR EC RR CO CUCU CMCM

4 Unmeasured Confounding in Claims Data  Database studies are criticized for their inability to measure clinical and life-style parameters that are potential confounders in many pharmacoepi studies  OTC drug use  BMI  Clinical parameters: Lab values, blood pressure, X-ray  Physical functioning, ADL (activities of daily living)  Cognitive status

5 Strategies to Discuss Residual Confounding  Qualitative discussions of potential biases  Sensitivity analysis  SA is often seen as the ‘last line of defense’  A) SA to explore the strength of an association as a function of the strength of the unmeasured confounder  B) Answers the question “How strong must a confounder be to fully explain the observed association”  Several examples in Occupational Epi but also for claims data Greenland S et al: Int Arch Occup Env Health 1994 Wang PS et al: J Am Geriatr Soc 2001

6 Foot-in-Mouth Award (Economist ‘04): “… there are known knowns; there are things we know we know. We also know that there are known unknowns; that is to say we know that there are some things we do not know. But there are also unknown unknowns – the ones we don’t know we don’t know. …, it is the latter category that tend to be the difficult ones.” (Wisely unknowing) Donald Rumsfeld

7 Notation

8 A simple sensitivity analysis  The apparent RR is a function of the adjusted RR times ‘the imbalance of the unobserved confounder’  After solving for RR we can plug in values ofr the prevalence and strength of the confounder:

9 A made-up example  Association between TNF-a blocking agents and NH lymphoma in RA patients  Let’s assume and observed RR of 2.0  Let’s assume 50% of RA patients have a more progressive immunologic disease  … and that more progressive disease is more likely to lead to NH lymphoma  Let’s now vary the imbalance of the hypothetical unobserved confounder

10 Bias by residual confounding

11

12 Pros and cons of “Array approach”  Very easy to perform using Excel  Very informative to explore confounding with little prior knowledge Problems:  It usually does not really provide an answer to a specific research question  4 parameters can vary -> in a 3-D plot 2 parameter have to be kept constant  The optical impression can be manipulated by choosing different ranges for the axes

13 Same example, different parameter ranges

14 Conclusion of “Array Approach”  Great tool but you need to be honest to yourself  For all but one tool that I present today:  Assuming conditional independence of C U and C M given the exposure status  If violated than residual bias may be overestimated Drug exposure Outcome RR EO OR EC RR CO CUCU CMCM Hernan, Robins: Biometrics 1999 ?

15 More advanced techniques  Wouldn’t it be more interesting to know  How strong and imbalanced does a confounder have to be in order to fully explain the observed findings? RR CO OR EC

16 Example: Wang et al: JAGS 2001;49:1685 Zolpidem use and hip fractures in older people. The issue: Are there any unmeasured factors that may lead to a preferred prescribing of zolpidem to people at higher risk for falling and fracturing? > Frailty is a hard to measure concept in claims data RR CO OR EC ARR PCPC

17 How do we do that?  We want to express as a function of, ARR, P C, P E OR EC RR CO Walker AM: Observation and Inference. Epidemiology Resources Inc., Newton, 1991

18

19

20

21 Example: Psaty et al: JAGS 1999;47:749 CCB use and acute MI. The issue: Are there any unmeasured factors that may lead to a preferred prescribing of CCB to people at higher risk for AMI? OR EC RR CO ARR = 1.57 ARR = 1.30

22

23 Caution!  Psaty et al. concluded that it is unlikely that an unmeasured confounder of that magnitude exists  However, the randomized trial ALLHAT showed no association between CCB use and AMI  Alternative explanations:  Joint residual confounding may be larger than anticipated from individual unmeasured confounders  Not an issue of residual confounding but other biases, e.g. control selection?

24 Pros and cons of “Rule Out Approach”  Very easy to perform using Excel  Meaningful and easy to communicate interpretation  Study-specific interpretation Problems:  Still assuming conditional independence of C U and C M  “Rule Out” lacks any quantitative assessment of potential confounders that are unmeasured

25 External Adjustment  One step beyond sensitivity analyses  Using additional information not available in the main study  Often survey information

26 Strategies to Adjust residual con- founding using external information  Survey information in a representative sample can be used to quantify the imbalance of risk factors that are not measured in claims among exposure groups  The association of such risk factors with the outcome can be assess from the medical literature (RCTs, observational studies) Velentgas et al: PDS, under review Schneeweiss et al: Epidemiology, in press 2004

27 How do we do that?  We want to express ARR as a function of,, ARR, P C, P E OR EC RR CO Walker AM: Observation an Inference. Epidemiology Resources Inc., Newton, 1991

28

29 Example: COX-2 inhibitors use and MI  Ray et al., Lancet 2002:  >25mg roficoxib vs. non NSAID users, RR=1.9 ( )  Medicaid patients, new users  Solomon et al., Circulation in press:  >25mg roficoxib vs. non NSAID users, RR=1.6 ( )  Medicare patients with drug coverage through PACE  Can these associations be due to confounding by factors not measured in claims data?  e.g. BMI, OTC aspirin use, smoking, education etc.

30 In our example: RofecoxibAcute MI RR EO From Survey data in a subsample From medical literature OR EC RR CO [smoking, aspirin, BMI, etc.] CUCU CMCM

31 Where can we get detailed information on unmeasured confounders?  MCBS: Medicare Current Beneficiary Survey  Representative Sample  12,000 Medicare beneficiaries each year (majority > 65y)  Face-to-face interview in beneficiary’s home  ‘Cost and Use’ file include drug utilization  98% response rate  >95% data completeness  Low cost ($900 / year)  Readily available, but 2-year lag time)

32 Unobserved confounders in our example  Independent predictors of MI:  Aspirin use  Smoking  BMI  Educational attainment  Income status  Expl. 2: Independent predictors of hip fracturs:  Cognitive impairment  Physical impairment  Restrictions in ADL (Rubinstein L)

33 Our survey population  1999 MCBS  Restricted to >64 years  Restricted to community sample (no proxi interviews)  N = 8,785

34 Distribution of unmeasured confounders among drug users

35 Celecoxib vs. Rofecoxib users …

36 Literature estimates of RR CO

37 Calculating bias

38 More contrasts

39 What does it mean?  Ray et al.: RR of 1.9 is an underestimation of the unconfounded RR by 5% (max)  So the effect estimate corrected for 5 unobserved confounders would be about 2.0  Solomon et al.: RR of 1.6 would move to 1.7

40 Sensitivity of Bias as a Function of a Misspecified RR CD : Obesity (BMI >=30 vs. BMI<30)

41 Sensitivity towards a misspecified RR CO from the literature: OTC aspirin use (y/n)

42

43 Variance estimate of externally adjusted parameters

44 Summary External Adjustment  This method provides a quantitative assessment of the effect of selected unobserved confounders  Easy to use (Excel program available from author)  MCBS is available from CMS for $900 per annual survey  Should be more frequently used in Pharmacoepi studies using claims data

45 Limitations (1)  Example is limited to 5 potential confounders  No lab values, physical activity, blood pressure etc.  What about the ‘unknow unknowns’?  We currently explore NHANES ’99/’00  Lab values, dietary suppl. (Ca 2+ ),  Drug data quality?  To assess the bias we assume an exposure–disease association of 1 (null hypothesis)  The more the truth is away from the null the more bias in our bias estimate  However the less relevant unmeasured confounders become

46 Limitations (2)  Validity depends on representativenes of sampling with regard to the unmeasured confounders  We could not consider the joint distribution of confounders  Limited to a binary world

47 Solving the Main Limitations  Need a method  That addresses the joint distribution of several unmeasured confounders  That can handle binary, ordinal or normally distributed unmeasured confounders  Lin et al. (Biometrics 1998) :  Can handle a single unmeasured covariate of any distribution  But can handle only 1 covariate  Sturmer, Schneeweiss et al. (AJE 2005 in press) :  Propensity Score Calibration can handle multiple unmeasured covariates of any distribution