Bias Due to Unmeasured Covariates Alec Walker Confounding by Indication.

Slides:



Advertisements
Similar presentations
Helical CT Screening for Lung Cancer at Advanced Radiology Consultants
Advertisements

1 Case-Control Study Design Two groups are selected, one of people with the disease (cases), and the other of people with the same general characteristics.
Mouth Cancer Action Month
Presentations in this series 1.Introduction 2.Self-matching 3.Proxies 4.Intermediates 5.Instruments 6.Equipoise Avoiding Bias Due to Unmeasured Covariates.
Modern Mortality and Morbidity Differentials in the U.S. SOC 331, Population and Society,
Journal Club Alcohol and Health: Current Evidence July–August 2004.
Biology in Focus, HSC Course Glenda Childrawi, Margaret Robson and Stephanie Hollis A Search For Better Health Topic 11: Epidemiology.
Ethical issues and cancer screening. Efficacy The extent to which a specific intervention, procedure, regimen, or service produces a beneficial result.
Predictors of 5-year mortality of 1,323 patients newly diagnosed with clinical type 2 diabetes in general practice With special emphasis on self-rated.
HSTAT1101: 27. oktober 2004 Odd Aalen
The Nature of Disease.
Multiple Choice Questions for discussion
Utility of Post-Therapy Surveillance Scans in Diffuse Large B-Cell Lymphoma Thompson C et al. Proc ASCO 2013;Abstract 8504.
Female Cancer Fatima Reyadh Mrs. Timm 12A March 10, 2013.
Frank B. Hu et al N Engl J Med 2004; 351: Body-Mass Index and Relative Risk of Death from All Causes during 24 Years of Follow-up.
Thomas B. Newman, MD, MPH Andi Marmor, MD, MSEd October 21, 2010.

EPIB-591 Screening Jean-François Boivin 29 September
Presentations in this series 1.Overview and Randomization 2.Self-matching 3.Proxies 4.Intermediates 5.Instruments 6.Equipoise Avoiding Bias Due to Unmeasured.
Spurious Association Sometimes an observed association between a disease and suspected factor may not be real. e.g. A study was conducted between births.
Cognitive Impairment: An Independent Predictor of Excess Mortality SACHS, CARTER, HOLTZ, ET AL. ANN INTERN MED, SEP, 2011;155: ZACHARY LAPAQUETTE.
5 Most Common Cancers. 1 in 2 men and 1 in 3 women in Australia will be diagnosed with cancer before the age of 85.
Lecture 17 (Oct 28,2004)1 Lecture 17: Prevention of bias in RCTs Statistical/analytic issues in RCTs –Measures of effect –Precision/hypothesis testing.
Using HIV Surveillance to Achieve High Impact Prevention Irene Hall, PhD, FACE AIDS 2012 High-Impact Prevention: Reducing the HIV Epidemic in the United.
Prostate Cancer: A Case for Active Surveillance Philip Kantoff MD Dana-Farber Cancer Institute Professor of Medicine Harvard Medical School.
A role for lipids and statins in breast cancer risk and prevention? Dr. Mieke Van Hemelrijck Senior Lecturer in Cancer Epidemiology 3 August 2015.
 Small cell lung cancer is named based on the size of the cells that cause the cancer. It accounts for about 20% of lung cancer cases.  It also referred.
“The African American Prostate Cancer Crisis in Numbers”
Imran Ahmad, MD FACP. Consultant Medical Oncologist, KFSH &RC-Jeddah.
ELIGIBILITY: MRC/BHF Heart Protection Study Increased risk of CHD death due to prior disease: Myocardial infarction or other coronary heart disease; Occlusive.
A Case Study in the Scientific Method
Alec Walker September 2014 Core Characteristics of Randomized Clinical Trials.
Introduction to Survival Analysis Utah State University January 28, 2008 Bill Welbourn.
Presentations in this series 1.Introduction 2.Self-matching 3.Proxies 4.Intermediates 5.Instruments 6.Equipoise Avoiding Bias Due to Unmeasured Covariates.
CANCER CONTROL NHPA’s. What is it? Cancer is a term to describe a diverse group of diseases in which some of the cells in body become defective. The following.
Case-control study Chihaya Koriyama August 17 (Lecture 1)
The Good…. The Bad…. & The Ugly.. What is disease?  How many diseases are there?  Diseases vs Conditions…  Types of diseases.
The Good…. The Bad…. & The Ugly.. What is disease?  How many diseases are there?  Diseases vs Conditions…  Types of diseases.
Causal relationships, bias, and research designs Professor Anthony DiGirolamo.
 Volunteer bias  Lead time bias  Length bias  Stage migration bias  Pseudodisease.
Wayne Rosamond, et al. Circulation 2007;115; e69-e171.
Acknowledgements This report differs from the submitted abstract due to further subdivision of patients into analytic and non- analytic, and focus on the.
Evaluating Screening Programs Dr. Jørn Olsen Epi 200B January 19, 2010.
HPV and Cervical Cancer FAQ. What is cervical cancer? Cervical cancer is cancer of the cervix, the part of the uterus or womb that opens to the vagina.
5.2 Day 2: Designing Experiments
Presentations in this series 1.Overview and Randomization 2.Self-matching 3.Proxies 4.Intermediates 5.Instruments 6.Equipoise Avoiding Bias Due to Unmeasured.
The Good…. The Bad…. & The Ugly.. Disease concepts  How many diseases are there?  Health, Function, Disability, Impairment, oh my!  Is it a disease.
Basic Characteristics of the Study Population Identified in Pennsylvania Medicare and Pharmaceutical Assistance Contract for the Elderly, 1994–2000 Soko.
Stomach – UK July 2007.
BREAST CANCER Breast cancer  Breast cancer is one of the commonest causes of death in many developed countries in middle-aged women, and is becoming.
1 CONFIDENTIAL – DO NOT DISTRIBUTE ARIES mCRC: Effectiveness and Safety of 1st- and 2nd-line Bevacizumab Treatment in Elderly Patients Mark Kozloff, MD.
Reduced Lung Cancer Mortality Risk Among Breast Cancer Patients Treated With Anti- Estrogens Rapiti E et al. SABCS 2009;Abstract 35.
Stomach Cancer By: Justin, Ty, and Patrick. ·Stomach cancer should not be confused with cancers of the colon (large intestine), liver, pancreas, or small.
Transparency in the Use of Propensity Score Methods
Endoluminal Treatment of Barrett’s and Early Cancer Brant K. Oelschlager, MD University of Washington.
REDUCED LUNG-CANCER MORTALITY WITH LOW-DOSE COMPUTED TOMOGRAPHIC SCREENING The National Lung Screening Trial Research Team N Engl J Med 2011;365:
Carina Signori, DO Journal Club August 2010 Macdonald, M. et al. Diabetes Care; Jun 2010; 33,
Case control & cohort studies
Cell Biology & Cancer Unit Objective 1 Cancer types, incidence, pre-disposition, and risk factors Biomedical Technology.
Statin treatment and reduced risk of pneumonia in patients with diabetes EMW van de Garde, E Hak, P c Souverein, AW Hoes, JMM van den Bosch, HGM Leufkens.
Sex-specific trends in lung cancer incidence and survival : a population study of cases 호흡기 내과 R3 조주희 Thorax 2011;66: Camilla M T Sagerup,
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.
심 재 준심 재 준 Am J Gastroenterol 2007;102:
Screening Tests: A Review. Learning Objectives: 1.Understand the role of screening in the secondary prevention of disease. 2.Recognize the characteristics.
Liceo Scientifico F. Redi Stomach cancer Teacher: Student: Elisa Vannuccini Viola Faltoni.
Unmeasured Covariates and the Need for Randomization
Bleeding and cancer risk in patients with vascular disease COMPASS Steering Committee and Investigators.
Lung Cancer in Lewisham
Presenter: Wen-Ching Lan Date: 2018/03/28
Supporting Patients With Colorectal Cancer
Presentation transcript:

Bias Due to Unmeasured Covariates Alec Walker Confounding by Indication

TD U

TD

Confounders TD U

TD U Randomization

Confounders TD U Randomization Self-matching

7 A “Classic” Example: Cimetidine and Gastric Cancer

8 Does cimetidine cause stomach cancer?  Case reports of de novo appearance in 1982  Colin-Jones et al looked at data from ongoing work Persons treated with cimetidine in a 12-month window Matched to a comparison person  General practitioner  Age  Sex  Seen for another condition Examined the incidence of stomach cancer

9 Excess cases during follow-up Diagnosed before cimetidine treatment started Diagnosed within six months of starting cimetidine treatment Diagnosed more than six months after starting cimetidine treatment Controls Cases of “early” cancer Before study |  Study period  | After study Diagnosed before cimetidine treatment started Diagnosed within six months of starting cimetidine treatment Diagnosed more than six months after starting cimetidine treatment Controls Cases of “early” cancer Number of cases Colin-Jones et al. Cimetidine and gastric cancer: preliminary report from post-marketing surveillance study. Brit Med J 1982;285:

Hypotheses to account for excess cancers Colin-Jones and his coauthors suggested that  Stomach cancer incidence was only an artifact of treatment having come before diagnosis in disease was already present – The as-yet undetected disease caused the use of cimetidine and led to detected disease. They hypothesized that the effect would disappear with longer follow-up.

11 Excess persisted for years Deaths from Year1234 Ex- pected Malignant neoplasm of the stomach Malignant neoplasm of the trachea, bronchus and lung Diseases of the digestive system Colin-Jones DG. Postmarketing surveillance of the safety of cimetidine: mortality during second, third, and fourth years of follow up. Brit Med J 1985;291:1084-8

Hypotheses to account for excess deaths  Stomach cancer – The as-yet undetected disease caused the use of cimetidine and led to detected disease.

Hypotheses to account for excess deaths  Stomach cancer – The as-yet undetected disease caused the use of cimetidine and led to detected disease.  Lung cancer – Shared determinants. Cigarette smoking predisposes to persistence of stomach ulcer, which in turn leads to cimetidine use. The smoking also causes lung cancer.

Hypotheses to account for excess deaths  Stomach cancer – The as-yet undetected disease caused the use of cimetidine and led to detected disease.  Lung cancer – Shared determinants. Cigarette smoking predisposes to persistence of stomach ulcer, which in turn leads to cimetidine use. The smoking also causes lung cancer.  GI disease – Conditions that motivated cimetidine use led to death.

Hypotheses to account for excess deaths  Stomach cancer – The as-yet undetected disease caused the use of cimetidine and led to detected disease.  Lung cancer – Shared determinants. Cigarette smoking predisposes to persistence of stomach ulcer, which in turn leads to cimetidine use. The smoking also causes lung cancer.  GI disease – Conditions that motivated cimetidine use led to death. Under each of these hypotheses, cimetidine use was driven by unmeasured factors that also led to the outcomes. The argument was that cimetidine did not cause the deaths from GI disease, stomach cancer or lung cancer, but confounding created associations and the false appearance of causal relations.

But were any of these hypotheses correct?  Stomach cancer – The as-yet undetected disease caused the use of cimetidine and led to detected disease.  Lung cancer – Shared determinants. Cigarette smoking predisposes to persistence of stomach ulcer, which in turn leads to cimetidine use. The smoking also causes lung cancer.  GI disease – Conditions that motivated cimetidine use led to death. Colin-Jones and colleagues could not know for sure. Had they used a different study design, an answer might have been clear.

Mechanisms that underlie confounding by indication  Prognosis Doctors act in patients’ interest  Protopathic bias Treatment for symptoms of an undiagnosed disease  Shared risk factors The disease points back to its own risk-filled origins