Unformatted text preview: CONTRIBUTIONS OF DIFFERENT TYPES OF EVIDENCE TO THE CONCLUSION OF A VALID STATISTICAL ASSOCIATION AND JUDGMENT OF CAUSALITY Charles H. Hennekens, MD,DrPH
Sir Richard Doll Research Professor Charles E. Schmidt College of Biomedical Science & Center of Excellence Florida Atlantic University (FAU) Disclosure
I am funded by the Charles E. Schmidt College of Medicine at Florida Atlantic University (FAU). I have served as Principal Investigator on two investigator initiated research grants funded to FAU by Bayer testing the effects of aspirin dose on platelet and inflammatory biomarkers as well as nitric oxide formation. I serve as an independent scientist in an advisory role to investigators and sponsors as Chair of Data and Safety Monitoring Boards for Actelion, Amgen, Anthera, Bristol-Myers Squibb, and Sunovion and as a Member of Data and Safety Monitoring Boards for AstraZeneca, Bayer , British Heart Foundation, Canadian Institutes of Health Research and Lilly. I serve as an independent scientist in an advisory role to the U.S. Food and Drug Administration, U.S. National Institutes of Health, Children's Services Council of Palm Beach County and UpToDate. I serve as an independent scientist in an advisory role to legal counsel for GlaxoSmithKline and Stryker. I serve as speaker for the Association for Research in Vision and Ophthalmology, Baptist Health South Florida, National Association for Continuing Education, PriMed, and the International Atherosclerosis Society. I receive royalties for authorship or editorship of three textbooks. I receive royalties as co-inventor on patents concerning inflammatory markers and cardiovascular disease which are held by Brigham and Women's Hospital. I have an investment management relationship with The West-Bacon Group within SunTrust Investment Services who has discretionary investment authority. I do not own any common or preferred stock in any pharmaceutical or medical device company. Death is inevitable but premature death is not Sir Richard Doll Totality of Evidence Basic research (why) Epidemiology (whether) Descriptive studies case reports case series ecological studies Analytic studies observational case-control cohort randomized trials Hennekens CH. Epidemiology in Medicine. Boston, Mass: Little, Brown & Co.;1987. ADVANTAGES AND DISADVANTAGES
Advantage: Precision Disadvantage: Questionable relevance to free living humans Epidemiology
Advantage: Relevance to free living humans Disadvantage: Imprecision QUESTIONABLE RELEVANCE OF BASIC RESEARCH TO FREE LIVING HUMANS
Who would have guessed that Homo sapiens would share with the humble guinea pig the unenviable distinction of being unable to synthesize ascorbic acid or with armadillos a susceptibility to the bacterium that causes leprosy or that intestinal cancer usually occurs in the large intestine of humans and the small intestine of sheep? Professor John Cairns QUESTIONABLE RELEVANCE OF BASIC RESEARCH TO FREE LIVING HUMANS
In basic research over 750,000 chemicals have the potential to cause cancer but less than 7500 have any direct relevance to humans Professor Sir Richard Peto QUESTIONABLE RELEVANCE OF BASIC RESEARCH TO FREE LIVING HUMANS
When the Harvard researcher was asked how to explain the discrepancies between the excess bladder cancer among Canadian rats fed 15 gallons of artificial sweeteners daily and no association between artificial sweeteners and bladder cancer in Massachusetts humans who drank about 1-2 cans of soda containing artificial sweeteners daily he replied that there must be systematic differences between the Canadian rats and the Massachusetts humans. Columbus Georgia Ledger QUESTIONABLE RELEVANCE OF BASIC RESEARCH TO FREE LIVING HUMANS I guess it takes a researcher from Harvard to put 2 and 2 together. The New Yorker Magazine STATISTICAL ASSOCIATION AND CAUSE AND EFFECT RELATIONSHIPS Statistical Association: A matter of fact like death and taxes Cause and Effect Relationships: A matter of opinion like truth and beauty STATISTICAL ASSOCIATION
A valid statistical association can be inferred from an analytic study after exclusion of CHANCE and BIAS and CONFOUNDING as plausible alternative explanations for the observed findings CHANCE Chance refers to the probability of findings at least as extreme as in your data when the null hypothesis is true Any P value depends on both the magnitude of association as well as the size of the sample. P values and confidence limits evaluate the role of chance but not bias or confounding Don't let the glitter of the t-table detract from the quality of the fare Professor Sir Austin Bradford Hill BIAS Bias may be defined as any systematic error in an analytic study that results in an incorrect estimate of the magnitude of association Selection bias refers to any error that arises in the process of identifying the study population Observation bias includes any systematic error in the measurement of information on exposure or outcome. CONFOUNDING Confounding may be viewed as a mixing of the effect of the exposure under study on the disease with that of a third factor. This third factor must be associated with the exposure and, independent of the exposure, be a risk factor for the disease. THE NEED FOR LARGE SCALE RANDOMIZED EVIDENCE
For hypotheses testing of large effects (i.e. smoking and lung cancer where RR=20, or even smoking and CHD where RR=2.0) randomized evidence is neither necessary nor desirable For small to moderate effects (i.e.10-50%) the amount of uncontrolled
and uncontrollable confounding inherent in all case control and cohort studies is as big as the effect size so large scale randomized evidence is crucial.
Hennekens CH, DeMets D: The need for large scale randomized evidence without undue emphasis on small trials, their metaanalyses or subgroup analyses JAMA 2009;302:23612362. SUBGROUP ANALYSES Subgroup analyses are no longer randomized and have lower sample sizes and should be viewed, at best, as hypothesis formulating and, at worst, as rubbish. The biggest danger in interpretation of subgroups is acting as if they provide serious evidence. If you torture the data enough they will confess. Professor Sir Richard Peto Intention to Treat (ITT) Analyses of Randomized Trials A randomized trial tests the offering of the treatment not the treatment ITT analyses preserve the randomization and avoids the introduction of confounders, only some of which are known and knowable Intention to Treat (ITT) and Analyses of Compliers: Coronary Drug Project Trial of Clofibrate and Mortality
MORTALITY ITT Compliers with Clofibrate Compliers with Placebo Multivariate analyses
(controlling for 40 confounders) 18.0% 15.0% 15.1% 16.4% 19.5% 24.6% 28.2% 25.8% META-ANALYSIS Should be considered more as hypothesis formulating than as hypothesis testing May provide the best estimate of effect that should be tested a priori in a large scale randomized trial designed to test the question Utility is dependent on the quality and comparability of the data from its component trials. Will reduce the role of chance but may introduce bias and confounding
Hennekens CH, DeMets D: The need for large scale randomized evidence without undue emphasis on small trials, their meta-analyses or subgroup analyses JAMA 2009;302:2361-2362. Meta-analysis is to analysis as meta-physics is to physics Professor William P. Castelli CAUSE AND EFFECT RELATIONSHIPS
A judgment based on the totality of evidence which includes Strength of association Consistency of findings Biological credibility Temporal sequence Dose-response THE NEED FOR LARGE SCALE RANDOMIZED TRIALS TO PROVIDE A SUFFICIENT TOTALITY OF EVIDENCE Calcium channel blockers and risk of myocardial infarction Antioxidant vitamins and risks of cardiovascular disease and cancer Cyclooxygenase-2 inhibitors and risks of cardiovascular disease Rosiglitazone and risk of myocardial infarction Ezetimibe and risk of cardiovascular disease Numbers of Randomized Patients in Completed Trials of Lipid Modifying Drugs on Clinical Cardiovascular Disease Outcomes Statins Nicotinic Acid Omega-3-FA Fibrates Gemfibrozil Fenofibrate Ezetimibe 90,056 2,835 11,324 2,531 9,795 0 French Fries
20 years ago Today 210 calories 2.4 ounces 610 calories How manycalories are in6.9 ounces these fries? Calorie difference: 400 Calories
How to burn* 400 calories: Walk 2 hour 20 minutes
*Based on 130-pound person. Darwinism and Risk of Cardiovascular Disease Walking the Dog The United States is the fattest society in the world and likely to be the fattest in the history of the world. Professor Charles H. Hennekens New York Times A bit of cultural news ..... After a 2 year loan to the United States, David returns to Italy Proud Sponsors Established Risk Factors for CHD
Blood cholesterol 10% = 20%-30% in CHD High blood pressure 5-6 mm Hg = 42% in Stroke = 16% in CHD Cigarette smoking Cessation = 50%-70% in CHD Body weight BMI<25 vs BMI>27 = 35%-55% in CHD Physical activity 20-minute brisk walk daily = 35%-55% in CHD "We must all hang together, or assuredly we shall all hang separately." Benjamin Franklin July 4, 1776 GOALS OF HEALTH CARE PROVIDERS AND ACADEMIC RESEARCHERS
Maximize benefit and minimize risk which is not to be confused with avoidance of risk. Make clinical decisions based on the totality of evidence not dependence on particular subgroups of particular studies. Avoid misstatements of benefit to risk ratios which may increase publicity, academic promotions and grant support in the short run but confuse colleagues and frighten patients and make it more difficult to conduct high quality research ( COX-2 inhibitors and glitazones) When the totality of evidence is incomplete it is appropriate to remain uncertain. Hennekens CH, DeMets D: The need for large scale randomized evidence without undue emphasis on small trials, their meta-analyses or subgroup analyses JAMA 2009; 302:23612362. ...
View Full Document
This note was uploaded on 02/23/2012 for the course STAT 312 taught by Professor Staff during the Fall '11 term at Rutgers.
- Fall '11