Types of study designs: from descriptive studies to randomized controlled trials Kirsten Bibbins-Domingo, PhD, MD Assistant Professor of Medicine and of Epidemiology and Biostatistics University of California, San Francisco

Objectives To understand the difference between descriptive and analytic studies To identify the hierarchy of study designs, and the strengths and weakness of each design To be able to apply different study designs to the same research question

Types of Studies Descriptive Studies Observational Analytic Studies Cross Sectional studies Case Control studies Cohort studies Experimental Studies Randomized controlled trials

Hierarchy of Study Types Analytic Descriptive Case report Case series Survey Observational Cross sectional Case-control Cohort studies Experimental Randomized controlled trials Strength of evidence for causality between a risk factor and outcome

Descriptive studies Getting a “lay of the land” Surveys (NHIS, MCBS) “How many men in the U.S. filled Viagra prescriptions in 2004?” Describing a novel phenomena Case reports or case series Viagra-associated serous macular detachment. Sildenafil-associated nonarteritic anterior ischemic optic neuropathy.

Descriptive studies Cannot establish causal relationships Still play an important role in describing trends and generating hypotheses about novel associations The start of HIV/AIDS research Squamous cell carcinoma in sexual partner of Kaposi sarcoma patient. Lancet. 1982 Jan 30;1(8266):286. New outbreak of oral tumors, malignancies and infectious diseases strikes young male homosexuals. CDA J. 1982 Mar;10(3):39-42. AIDS in the "gay" areas of San Francisco. Lancet. 1983 Apr 23;1(8330):923-4.

Analytic Studies Attempt to establish a causal link between a predictor/risk factor and an outcome. You are doing an analytic study if you have any of the following words in your research question: greater than, less than, causes, leads to, compared with, more likely than, associated with, related to, similar to, correlated with

Hierarchy of Study Types Analytic Descriptive Case report Case series Survey Observational Cross sectional Case-control Cohort studies Experimental Randomized controlled trials Strength of evidence for causality between a risk factor and outcome

Research Question Is the regular consumption of Red Bull associated with improved academic performance among U.S. medical students?

Rationale “functional drink” designed for periods of mental and physical exertion. performance, concentration, memory, reaction time, vigilance, and emotional balance Taurine + glucuronolactone + caffeine

Background Alford C, Cox H, Wescott R. The effects of red bull energy drink on human performance and mood. Amino Acids. 2001;21(2):139-50. Warburton DM, Bersellini E, Sweeney E. An evaluation of a caffeinated taurine drink on mood, memory and information processing in healthy volunteers without caffeine abstinence. Psychopharmacology (Berl). 2001 Nov;158(3):322-8. Seidl R, Peyrl A, Nicham R, Hauser E. A taurine and caffeine-containing drink stimulates cognitive performance and well-being. Amino Acids. 2000;19(3-4):635-42. Horne JA, Reyner LA. Beneficial effects of an "energy drink" given to sleepy drivers. Amino Acids. 2001;20(1):83-9. Kennedy DO, Scholey AB. A glucose-caffeine 'energy drink' ameliorates subjective and performancedeficits during prolonged cognitive demand. Appetite. 2004 Jun;42(3):331-3.

Great idea, but how do you get started…. Interesting, novel, and relevant, but… You only have 25,000 dollars to start investigating this question. What is feasible?

Study Design #1 Cross-sectional study of UCSF medical students taking USMLE Step 2 Questionnaire administered when registering for USMLE 2 Primary predictor: self-report of >3 cans Red Bull per week for the previous year Covariates: Age, sex, undergraduate university, place of birth Outcome: Score on USMLE Step 2

Cross-sectional study: structure Red Bull consumption USMLE Score time

Cross-sectional Study: Descriptive value: How many UCSF medical students drink Red Bull? What is the age and sex distribution of UCSF medical students who drink Red Bull? Analytic value: Is there an association between regular Red Bull consumption and test scores among UCSF med students? Univariate Multivariate (controlling for “confounders”) Other cross-sectional surveys: AAMC California Health Interview Survey (NHIS, CHIS) National Health and Nutrition Exam Survey (NHANES)

Cross-sectional Study: Pluses + Prevalence (not incidence) + Fast/Inexpensive - no waiting! + No loss to follow up + Associations can be studied

Measures of association Risk ratio (relative risk) A A + B C C + D Disease Yes No Risk Factor A B C D

Cross-sectional study: minuses - Cannot determine causality Red Bull consumption USMLE Score time

Cross-sectional study: minuses - Cannot determine causality ACE inhibitor use and hospitalization rates among those with heart failure Heart failure patients with a documented DNR status and mortality time

Cross-sectional study: minuses - Cannot determine causality - Cannot study rare outcomes

What if you are interested in the rare outcome? The association between regular Red Bull consumption and… A perfect score on the USMLE – Step 2 Graduating top 1% of the medical school class Acceptance into a highly selective residency ANSWER: A Case-Control study

Study Design #2 A case-control study Cases: 4th year med students accepted to residency in “highly selective specialty X”. Controls: 4th year med students who applied but were not accepted. Predictor: self-reported regular Red Bull consumption Additional covariates (age, sex, medical school, undergraduate institution)

Case control studies Investigator works “backward” (from outcome to predictor) Sample chosen on the basis of outcome (cases), plus comparison group (controls)

Case-control study structure present TARGET CASES Medical students accepted to highly selective residencies ACTUAL CASES 4th year UCSF students who matched in “highly selective specialty X” Red Bull consumption YES Red Bull consumption NO TARGET CONTROLS All unsuccessful applicants to highly selective residency programs ACTUAL CONTROLS 4th year students who failed to match in “highly selective specialty X” time

Case control studies Determines the strength of the association between each predictor variable and the presence or absence of disease Cannot yield estimates of incidence or prevalence of disease in the population (why?) Odds Ratio is statistics

Case-control Study: pluses + Rare outcome/Long latent period + Inexpensive and efficient: may be only feasible option + Establishes association (Odds ratio) + Useful for generating hypotheses (multiple risk factors can be explored)

Case-control study-minuses Causality still difficult to establish Selection bias (appropriate controls) Caffeine and Pancreatic cancer in the GI clinic Recall bias: sampling (retrospective) Abortion and risk of breast cancer in Sweden Cannot tell about incidence or prevalence Studies of diagnostic tests: Sensitivity, specificity Positive predictive value, negative predictive value

Measures of association Disease Yes No Test A B C D Sensitivity = A/A+C Specificity = D/B+D PPV = A/A+B NPV = D/C+D

Case-control - “the house red” Rely tampons and toxic shock syndrome: High rates of toxic shock syndrome in menstruating women Suspected OCPs or meds for PMS Cases: 180 women with TSS in 6 geographic areas Controls: 180 female friends of these patients and 180 females in the same telephone code Tampon associated with TSS (OR = 29!) Super absorbency associated with TSS (OR 1.34 per gm increase in absorbency) Led to “RELY” brand tampons being taken off the market.

Where are we? Preliminary results from our cross-sectional and case-control study suggest an association between Red Bull consumption and improved academic performance among medical students What’s missing? - strengthening evidence for a causal link between Red Bull consumption and academic performance Use results from our previous studies to apply for funding for a prospective cohort study!

Study design #3 Prospective cohort study of UCSF medical students Class of 2009 All entering medical students surveyed regarding beverage consumption and variety of other potential covariates Survey updated annually to record changes in Red Bull consumption Outcomes: USMLE Step 1 score, USMLE Step 2 score, match in first choice residency

Cohort studies A cohort (follow-up, longitudinal) study is a comparative, observational study in which subjects are grouped by their exposure status, i.e., whether or not the subject was exposed to a suspected risk factor The subjects, exposed and unexposed to the risk factor, are followed forward in time to determine if one or more new outcomes (diseases) occur Subjects should not have outcome variable on entry No new subjects allowed in after initial recruitment The rates of disease incidence among the exposed and unexposed groups are determined and compared.

Elements of a cohort study Selection of sample from population Measures predictor variables in sample Follow population for period of time Measure outcome variable Famous cohort studies Framingham Nurses’ Health Study Physicians’ Health Study Olmsted County, Minnesota

Prospective cohort study structure The present The future Top USMLE scorers Everyone else time

Strengths of cohort studies Know that predictor variable was present before outcome variable occurred (some evidence of causality) Directly measure incidence of a disease outcome Can study multiple outcomes of a single exposure (RR is measure of association)

Weaknesses of cohort studies Expensive and inefficient for studying rare outcomes HERS vs. WHI Often need long follow-up period or a very large population CARDIA Loss to follow-up can affect validity of findings Framingham

Other types of cohort studies Retrospective cohort Identification of cohort, measurement of predictor variables, follow-up and measurement of outcomes have all occurred in the past Much less costly than prospective cohorts Investigator has minimal control over study design

Other types of cohort studies Nested case-control study Case-control study embedded in a cohort study Controls are drawn randomly from study sample Double cohort Used to compare two separate cohorts with different levels of exposure to predictor variable (e.g., occupational groups)

What type of study is this? Among individuals with coronary disease, what is the association between baseline levels of B-type natriuretic peptide and subsequent risk of heart failure? Among individuals presenting to heart failure clinic, what is the association between self-reported symptoms and risk of hospitalization for heart failure? Using data from HERS (RCT of HRT in women with coronary disease): Determine the risk factors for developing incident heart failure among women without heart failure at baseline. Determine whether HRT is associated with mortality among women with heart failure. Determine genetic markers for development of heart failure among black women in HERS.

Hierarchy of Study Types Analytic Descriptive Case report Case series Survey Observational Cross sectional Case-control Cohort studies Experimental Randomized controlled trials Strength of evidence for causality between a risk factor and outcome

What distinguishes observational studies from experiments? Ability to control for confounding Confounder Predictor Outcome Examples: sex (men are more likely to drink red bull and men are more likely to match in neurosurgery) Undergraduate institution (students from northwest school are more likely to drink red bull and also more likely to score higher on USMLE)

But we measured all of the potential confounders……. In a prospective cohort study you can (maybe) measure all potential known confounders, but… You can’t control for unanticipated or unmeasured confounders

Study design # 4 Randomized controlled trial of daily Red Bull consumption among entering UCSF medical students Class 2009 Randomized to daily consumption of Red Bull vs. daily consumption of placebo Outcomes: USMLE Step 1 score, USMLE Step 2 score, match in first choice residency

Randomized controlled trials Investigator controls the predictor variable (intervention or treatment) Major advantage over observational studies is ability to demonstrate causality Randomization controls unmeasured confounding Only for mature research questions

Basic Trial Design Treatment Dx No Dx Control Dx No Dx Placebo Population Treatment Dx No Dx Randomization Sample Control Dx No Dx Placebo For example, select persons with high LDL-cholesterol from among patients presenting to the metabolic clinic at the SFVA and treat with lovastatin to see if this results in lower LDL cholesterol. Note can have an uncontrolled trial or time series design. These have no concurrent control group and can produce the wrong answer due to learning, regression to the mean or secular trends. Note, for example, that it is common to have a 30-50% improvement in depression scores, pain scores, hot flash scores. Can have a controlled trial without randomization. This addresses learning, regression to the mean and secular trends, but introduces the issue of baseline differences in the groups. This design is really no different from a double cohort study. Can have a randomized controlled trial without placebo (blinding). This allows (as will be discussed later) for co-intervention and biases outcome ascertainment. Also note that these three elements - treatment, randomization and blinding result in the major ethical concerns regarding randomized trials.

Steps in a randomized controlled trial Select participants high-risk for outcome (high incidence) Likely to benefit and not be harmed Likely to adhere Measure baseline variables Randomize Eliminates baseline confounding Types (simple, stratified, block)

Steps in a randomized controlled trial Blinding the intervention As important as randomization Eliminates co intervention biased outcome ascertainment biased measurement of outcome Follow subjects Adherence to protocol Lost to follow up Measure outcome Clinically important measures Adverse events

What is Blinding? Single blind - participants are not aware of treatment group Double blind - both participants and investigators unaware Triple blind - various meanings persons who perform tests outcome adjudicators safety monitoring group

Why blind?: Co interventions Unintended effective interventions participants use other therapy or change behavior study staff, medical providers, family or friends treat participants differently Nondifferential - decreases power Differential - causes bias

Why blind?: Biased Outcome Ascertainment or adjudication If group assignment is known participants may report symptoms or outcomes differently physicians or investigators may elicit symptoms or outcomes differently Study staff or adjudicators may classify similar events differently in treatment groups Problematic with “soft” outcomes investigator judgement participant reported symptoms, scales

Analysis of randomized controlled trial Analyzed like cohort study with RR Intention to treat analysis Most conservative interpretation Include all persons assigned to intervention group (including those who did not get treatment or dropped out) Subgroup analysis Groups identified pre-randomization

High Quality Randomized Trials Tamper-proof randomization Blinding of participants, study staff, lab staff, outcome ascertainment and adjudication Adherence to study intervention and protocol Complete follow-up

Hierarchy of Study Types A study type of every budget, purpose and research question Analytic Descriptive Case report Case series Survey Observational Cross sectional Case-control Cohort studies Experimental Randomized controlled trials Strength of evidence for causality between a risk factor and outcome