Prospective Cohort Studies Frederick L. Brancati, MD, MHS Professor of Medicine & Epidemiology Director, Division of General Internal Medicine Osler Journal Club 2006 Visit Hopkins GIM at http://www.hopkinsmedicine.org/gim Background • • • • Physical activity lower CVD risk DHHS recommends life-long pursuits Sports differ in sustainability CVD benefits of individual sports uncertain The Johns Hopkins Precursors Study Over 1300 students (mainly white men) from the JHUSOM Classes of 1948-64. Baseline data collected in person in medical school. Followup data collected by yearly mailed questionnaires thereafter. Caroline Thomas, MD The Johns Hopkins Precursors Study Outline • Hypothesis: Tennis ability in youth predicts lower CVD risk in middle age • Design: Prospective cohort study • Setting: Johns Hopkins Precursors Study • Participants: 1019 male medical students • Data Collection: Extensive interview and physical assessment at baseline (early 20s); annual mailed follow-up questionnaires • Outcome: Incident CVD, including MI, CHD, CABG or PTCA, hypertensive heart disease, heart failure, & cerebrovascular disease • Analysis: Kaplan-Meier, Cox models Assessment of Sports Ability • How would you rate your overall ability in tennis (golf, football, baseball, basketball) during and before medical school? – No ability – Poor or fair ability – Good or excellent ability • No data on frequency, intensity, or subsequent participation Results Conclusions / Implications • Self-described tennis ability in young adulthood predicts lower CVD risk in middle age • Association of tennis to lower risk is – Graded (i.e. dose-response) – Independent of many possible confounders – Specific to tennis (as hypothesized) • Suggests promotion of tennis as a means to reduce CVD risk Strengths • • • • Prospective design Long-term follow-up Multiavariate analysis Blinded assessment of CVD Weaknesses • Observational studies can’t prove causality • Residual confounding is likely • Assessment of exposure was suboptimal – Ability, not activity – Single point, not repeated measures – Self-assessed, not objective • Sample limits generalizability Discussion Points • • • • What’s special about a cohort study? What are common obstacles? Can it be used for housestaff research? Can it ever be sufficient to change practice? • How do cohort studies relate to outcomes research? Taxonomy of Designs • • • • • Randomized Controlled Trial Prospective Cohort Study Case-Control Study Cross-Sectional Study Other Designs – Quasi-Experimental – Ecologic – Case Report The basic fighting unit was a cohort, composed of six centuries (480 men plus 6 centurions). The legion itself was composed of ten cohorts, and the first cohort had many extra men—the clerks, engineers, and other specialists who did not usually fight—and the senior centurion of the legion, the primipilus, or “number one javelin.” pro·spec·tive Pronunciation: pr&-'spek-tiv also 'prä-", prO-', prä-' Function: adjective Date: circa 1699 1 : relating to or effective in the future 2 a : likely to come about : EXPECTED <the prospective benefits of this law> b : likely to be or become <a prospective mother> “Prospective” in Epidemiology • Clearly defined cohort (group, sample) of persons at risk followed through time • Data regarding exposures (risk factors, predictors) collected prior to data on outcomes (endpoints) • Research-grade data collection methods used for purpose of testing hypothesis (?) Diagram of Hypothetical 6-Year Cohort Study to Identify Risk Factors for Facial Acne in Teenagers 1000 12-year-olds without acne 5 moved 50 with Acne 10 no answer 35 refused 900 15-year-olds without acne 10 moved 300 with Acne 40 no answer 48 refused 500 18-year-olds without acne 2 deaths 6-yr Follow-up Rate = 850/1000 = 85% Incidence Rate of Acne = 350/5475 PY = 63.9 per 1000 PY 350 incident cases of acne over 6 years Why Do A Cohort Study? • • • • • • Get incidence data Study a range of possible risk factors Establish temporal sequence Get representative data Prepare for randomized controlled trial Establish a research empire Types of Cohorts • • • • Occupational (e.g. Asbestos workers) Convenience (e.g. Precursors, Nurses) Geographic (e.g. Framingham, ARIC) Disease or Procedure – Natural History (e.g. Syncope, Lupus) – Outcomes Research (e.g. Dialysis, Cataracts) Sources of Cohort Data • Clinic Visits – – – – – Laboratory Assays Interview Physical Examination Imaging Physiologic tests • Home visits • Mailed materials • Telephone Interview • Medical Records • Administrative Data – – – – Medicare Medicaid Managed Care Veterans Admin • Birth Records • Death Certificates • Specimen Bank The Framingham Heart Study William Castelli, MD Recently Published Studies from the Johns Hopkins Precursors Study Outcome • Coronary Disease • • • • Type 2 Diabetes Hypertension Knee Osteoarthritis Depression Exposure -Anger, Depression, Gout, -Sports Ability -Blood pressure, Adiposity -Coffee -Knee injury -Insomnia What Might Explain Observed Relationship of Tennis Ability to Heart Disease Risk? • Tennis protects against heart disease • Men who like to play tennis are different – Thinner – Healthier Lifestyles – Higher Socioeconomic Status • Men who play tennis well are different – Taller, Thinner – Greater Cardiovascular Fitness • Chance (type I error) – Needs confirmation Hypothetical Causal Pathway Potential Confounders Healthier Men Choose Tennis Plays Tennis Healthier Men Play Tennis Well Plays Tennis Well Sustained Activity Thru Midlife Lower adiposity, Greater Fitness Lower BP, Lower LDL, Higher HDL Lower Risk of CHD Hypothetical Causal Pathway Grey Hair Higher Risk of CHD Potential Confounders Older Age Challenges in Cohort Studies • • • • • • Possibly long duration Possibly large sample size Need to recruit people “at risk” Drop outs, Deaths, Other losses Concern about residual confounding Multiple comparisons Type I error How to Exploit Cohort Design When Time is Short & Money is Scarce • • • • • • • Analyze existing data from another study Piggy-back onto on-going study Choose hospital-based cohort Choose short-term outcome Consider administrative data Consider public-use data Consider non-concurrent design