Panacek Design And Performing Cohort Studies

Designing and Performing Cohort Studies Edward A. Panacek, MD, MPH Professor of Medicine University of California-Davis Medical Center (SAEM meeting sylabus: May, 2000)

Choosing the best design for each research question: It is time to stop squabbling over the “best” methods Sackett DL, Wennberg JE. BMJ. 1997;3315:1636. n Focusing on methods rather than questions has largely been arguing about the wrong things. n The question being asked (usually) determines the appropriate research strategy, not tradition. n Each method should flourish, because each has features that overcome the others limitations. n Which way of answering the question provides us with the most valid, useful answer?

Performing Clinical Research n There are many different “jobs” in clinical research – – – – – –

Prevalence proportion, incidence rates Measures of association (RR, OR) Effectiveness versus efficacy evaluation Benefits versus safety Endpoints versus outcomes Outcomes versus cost-effectiveness

n Should have many tools in your “research toolbox”

Points regarding study designs n Do not confuse scientific accuracy with clinical relevance n Well done cohort and case-control studies can be much more valuable than irrelevant clinical trials n There is much confusion regarding the definition of a “cohort study” – Has one core definition – Multiple actual study applications – Often used erroneously in presentations

Goals of this lecture n Be able to define the term “Cohort” n Describe different types of cohort studies n Contrast cohort to case-control studies and RCTs n List the advantages & disadvantages n List the main outcome measures used n Cite examples of classic cohort studies n Cite examples of cohort studies in EM literature

Dictionary definition of “cohort” n Latin: cohors – Enclosed yard or company of soldiers – All were the same type of soldier (e.g. calvary) – In the Roman armies, a band of 300-600 soldiers, constituting 1/10th of a Legion

Concept: A group of individuals that are all similar in some trait and move forward together as a unit

Epidemiology definition of “cohort” n Cohort: A group of individuals that share a common characteristic – Birth cohort : all individuals in a certain geographic area born in the same period (usually a year) – Inception cohort: all individuals assembled at a given point based on some factor, e.g. where they live or work – Exposure cohort: individuals assembled as a group based on some common exposure • e.g. radiation exposure during desert testing • e.g. asbestos exposure in the shipyards

Definitions of “cohort study” n The observation of a cohort (or cohorts), over time, to measure outcome(s) – AKA: Longitudinal, follow-up studies

They have 2 primary purposes: n Descriptive (measures of frequency) – To describe the incidence rates of an outcome over time, or simply describe the natural history of disease

n Analytic (measures of association) – To analyze associations between the rates of the outcomes and risk factors or predictive factors

Cohort studies versus Clinical trials (RCTs) n Randomization: – Cohort: no – RCT: yes

n Prospective: – Cohort: usually – RCT: yes

n Intervention: – Cohort: no, just the passage of time (observational) – RCT: yes

n Control of initial study conditions – Cohort: no – RCT: yes

Why use cohort studies instead of RCTs? n Unable to randomize – Impossible: genetic traits – Unethical: desperate disease (CA) – Illegal: effect of cocaine use during pregnancy

n Interested in incidence rates or predictors more than the effects of interventions – e.g. predictive role of initial BP in field in blunt trauma

n Field of investigation is immature n Limited research resources – time, money, subjects

Types of cohort studies n Single group (inception cohort) n Multiple groups (Double or Comparison cohort) – From the same inception cohort (internal controls) – Assembled separately (external controls)

n Prospective n Retrospective n Ambispective – Both prospective and retrospective components

Single group cohort study n AKA: Inception cohort n Structure: Assemble cohort based on some factor. Follow them over a set period of time. – Usually multiple observations for outcome(s) of interest

n Time frame: Usually prospective n Purpose: descriptive n Measures: Incidence rates, point prevalence

Example: Single group cohort Dawber TR, et al. An approach to longitudinal studies in a community: The Framingham study. Ann NY Acad Sci. 1963;107:539. n Began in 1948 with 5,209 participants n 5,123 spouses and children added in 1971 n Selected not based on exposures, but on stable pop.,

wide spectrum of occupations, single hospital, annual updated population lists n Allowed calculation of incidence rates and other

descriptive measures for many outcomes

Retrospective versus prospective cohort studies n Classification is based on the temporal relationship between the initiation of the study (sample defined) and occurrence of the outcome – i.e., outcome before initiation = retrospective

n However, both start by identifying and enrolling subjects based upon the presence or absence of the exposure (IV) of interest, without knowing the outcome at the time (even if retrospective) – i.e. subjects are free of the outcome (e.g. disease) at the time their exposure status is defined

Double group cohort study: Retrospective n If from within prior inception cohort; – AKA: “nested cohort study”

n Structure: Select exposed group and nonexposed group from pre-existing data base. – Obtain F/U information on numbers of outcomes

n Purpose: Compare the outcome rates in the 2 groups n Measures: Incidence ratios, Relative risks, Odds ratios ( can do single univariate comparison)

Example: Retrospective (nested) cohort study with internal controls Belanger CF, Hennekens CH. The nurses’ health study. Am J Nurs. 1978;78:1039

n 12,000 nurses surveyed at baseline and periodically thereafter n Collected information on many factors and outcomes over many years n Later, split group into those using oral contraceptives vs. not to compare outcomes – Compared rates of CA, AMI, etc – Addressed questions not formulated at study initiation

Example: Retrospective cohort study with external controls Enterline PE. Mortality among asbestos product workers in the US. Ann NY Acad Sci.1965;132:156

n Exposed : Asbestos workers identified from IRS tax returns (1948-51) n Unexposed:1. Cotton textile workers form IRS 2. General US matched population n Outcome: Death rates (from state health depts. ) n Measure: Death incidence rates in each group – overall and cancer specific → calculated rate ratios

Prospective cohort (double group) studies n The “classic” cohort study design n Sample defined prospectively during or before exposure and before outcome occurrence – “Exposure” can be many things (e.g. predictor variable) – Allows for more accurate measure of exposure/factor – Also allows for more accurate measure of potential confounding variables – Can have multiple measurements over time – Groups followed over time for development of the outcome

Example: Prospective (double) cohort study with internal controls Doll R, Hill AB. Mortality in relation to smoking: 10 years observation of British docs. Br Med J.1964;1:1399-1410. n Cohort: British doctors responding to a survey in 1950

– 65% response rate n Exposed: smokers ( and quantified amount) n Unexposed: non-smokers n Outcome:Lung Ca and death – Periodic F/U surveys and review of death records n Results: Increased risk with any smoking and a dose-

response relationship

Example: Prospective cohort study with “internal” comparison control group Paffenberger RS, et al. A natural history of athleticism and CV health. JAMA.1984;252:491-5. n Cohort: 16,936 Harvard alumni n Groups: high vs. low exercise groups n Measurements: college records and questionnaires at baseline and 10 years n Outcome: CHD data from questionnaires and death certificates n Results: RR for CHD = 1.5 if sedentary vs. if active

Example: Prospective cohort study with “external” comparison control group British Journal of Audiology. 1980s n Question: Is living under flight path hazardous? n Study group: Those living next to LAX airport n Controls: Other angelinos in different LA zip code n Outcome: Death rates in each zip code zone – From county health records

n Results: Higher per capita mortality rates by LAX – Problems???

Cohort studies: Importance of the comparison group n Unlike RCTs, cohort studies do not have randomization of study subjects n Therefore, they are more vulnerable to selection bias n This is usually not as issue with the exposure n Often a serious problem in terms of confounders n Increasing the size of the study can only partially help address the issue

Example: Prospective cohort study with “external” comparison control group Selikoff IJ, et al. Latency of asbestos dz among insulation workers in the US and Canada. CANCER. 1980;46:736+

n Exposed: 17,800 males in Asbestos Insulation Workers union in North America as of 1-1-67 n Unexposed: General population of males matched by age n Outcome: F/U for lung cancer rates through 1975 n Results: Positive assoc. between asbestos and lung CA

Ambispective cohort studies Two types: n Cohort created at time of study initiation but exposure in the past and outcomes both past and future – Ideal for evaluating exposures that may have both short-term and long-term effects

n Cohort assembled part retrospectively and part prospectively – Allows enrollment of more subjects/less time, using prospective evaluation to check for data completeness

Example: Ambispective cohort with both short-term and long-term outcomes Gunby P. Military looks toward 1985 in ongoing defoliant study. JAMA.1984;85:383. n Question: Are there deleterious effects of exposure to

agent orange in servicemen? n Exposed: 1264 exposed to defoliant spraying in Vietnam n Unexposed: 1264 who flew cargo missions at same time n Outcomes(retro): Medical problems during exposure time – e.g. dermatologic conditions, birth defects, liver problems n Outcomes (prospective): cancer rates up to 25 years later

Example: Ambispective cohort study with retro and prospective enrollment White RH, et al. Bleeding complications related to INR level in patients on warfarin. JAMA. n UC Davis anticoagulation clinic patients, since 1993, with continuing enrollment into the future n Exposed: those with elevated INRs n Unexposed: Pts with normal INRs n Outcomes: Bleeding complications

Cohort studies: Principal outcome measures n Crude: simple univariate comparison of rates or proportions between the 2 groups – Gives statistical but not clinical significance

n Descriptive: Incidence rates in the group(s) – Gives absolute measure of association but not comparisons

n Comparisons: Relative measures of association – Compares incidence rates between groups – Relative risk – Risk ratio (sometimes estimated by Odds Ratios)

Cohort versus case-control studies n Case-control studies start with the outcome and look back for exposures/factors – Outcome present = case – Outcome absent = control ( or referent subject) – Almost always are retrospective studies

n Cohort studies generally start with exposures and then follow the cases through time, for the outcomes – Exposure present = study subject – Exposure absent = control subject

Terminology confusion n Case-control studies often simply called “retrospective observational studies” – can occasionally be done prospectively – Best name = “Case-control (referent) study”

n Cohort studies often called “prospective observational studies” – Often performed retrospectively – Better name = “Exposed-unexposed study” • Unfortunately, this name has not caught on

Cohort studies: Matching n Pair matching – Each study subject is closely matched with a control subject on some specific factor – Requires special statistical tests in the analysis to adjust for the confounding effects of the matching

n Frequency matching – Each study subject or group of subjects are matched with controls on some category of a factor • e.g. by gender, or age within 5 years, smoker – Generally does not require special statistical tests in the analysis

Decisions about matching n Current statistical techniques allow adjustment for

confounders, so matching not as important as before n If have a known powerful confounder or one that is difficult

to measure precisely – Pair match on that confounder

n For most other possible confounders, better to just adjust

in the analysis n If match on a factor, less able to study its role in the dz. n Use frequency matching to prevent gross imbalances

between groups that would decrease the power of the study

Cohort studies in Emergency Medicine n Not as commonly used in EM as in primary care, occupational medicine, and cancer research n In EM, don’t usually perform long-term follow-up studies, unless doing epidemiologic research n However, very useful option for selected issues – – – –

Injury patterns and prevention research When unable to randomize When unable to get informed consent When the “F/U” period can be very short or can all be retrospective

The evolution of cohort studies n The classic cohort studies involved two components: – Exposed and unexposed groups – Longitudinal F/U over long time periods

n Neither of these elements seem well suited to EM research n However, cohort studies have evolved: – design components more flexibly applied – application of cohort studies expanded

Cohort studies: The element of “exposure” n The “classic” cohort studies compared an exposed group to an unexposed group n However, that is simply an extreme case of differences between two groups. n Other “differences” are also possible: – – – –

High exposure vs. low exposure Exposure 1 vs. exposure 2 Presence of factor 1 vs. factor 2 Intervention 1 vs. 2

Cohort studies: The element of “follow-up” n The original “classic” cohort studies involved long F/U periods n However, that is because the outcomes of interest were usually cancer and other conditions with long exposure-outcome timeframes n When the outcome follows closely after the exposure (or factor or intervention), the length of the “F/U” period is likewise short

EM example: Retrospective cohort with internal control group n Braun BL, et al. Marijuana use and medially attended

injury events. Ann Emerg Med.1998;32:353 n Cohort: Kaiser members undergoing multiphasic exams

1979-86 in SF or Oakland, aged 15-49 n Groups: Self-reported marijuana use (prior, current) as

exposed vs. the “never” category as the comparison group n Outcomes: Injury related clinic visits, hospitalizations and

fatalities n Results: Rate ratios not different between groups

EM example: Retrospective cohort study Tran P, Panacek EA. A comparison of norepinephrine and dopamine for treating TCA OD associated hypotension. Acad Emer Med. 1997;4:864-8.

n Cohort: All TCA OD pts requiring vasopressors Exposure 1: Dopamine as first vasopressor Exposure 2: Norepinephrine as first vasopressor

n Outcomes: BP response to normal range n Results: Norepi effective in all, dopamine in 60% – Relative risk for persistent hypotension with dopa= 4.8

EM example: Retrospective cohort study with internal control group Wintemute GJ. Criminal activity and assault-type handguns: A study of young adults. Ann Emerg Med.1998;32:44-50. n Cohort: 5,360 legal purchasers of handguns in Calif. in

1988, under age 25 n Group 1: purchased assault-type handguns n Group 2: purchased other types of handguns n Outcome: criminal activity during subsequent 3 years n Results: RR = 1.5-3.0 for criminal activity if purchased

assault-type handgun

EM example: Prospective cohort study (single group) Minogue MF, et al. Pts hospitalized after initial outpt treatment for CAP. Ann Emerg Med. 1998;31:376-80.

n Cohort: all patients with CAP initially treated as outpatients at 5 study centers n Outcomes: Hospitalization within 30 days n Results: Descriptive – % hospitalized – % CAP related – Identification of factors that may be predictive

EM example: Prospective cohort study Sakles JC... Panacek EA. Comparison of succinylcholine to rocuronium for RSI in ED. Acad Emer Med.1999;6:518. n Cohort: All ED pts undergoing RSI – Group 1: those receiving rocuronium as the NMB drug – Group 2: receiving succinylcholine as the NMB drug

n Outcomes: – Time to full relaxation & intubating conditions – Time to recovery and complications

n Results: Very similar in all parameters except recovery time. RR for complications = 1

Cohort studies: Strengths n The best way to study incidence of the outcome n Ideal for studying rare exposures (or initial conditions) n Unlike case-control studies: – The temporal sequence is clear – Can examine multiple effects from a single exposure n If prospective, minimizes bias in the measurement of

exposure n Much less expensive than RCTs n Sometimes the best or only ethical way to do the study – e.g. cannot or should not randomize

Cohort studies: Weaknesses n Inefficient for study of rare outcomes – Unless the attributable-risk is high for the exposure

n If prospective, can be nearly as resource expensive as RCTs n If retrospective, is dependent upon the adequacy of records n Because these are “follow-up” studies, validity of results is highly sensitive to losses to F/U

Cohort studies: Strategies to minimize “lost to F/U” n Exclude those likely to become “lost” – Planning to move – Unwilling to return

n Obtain complete tracking info – Address, phone #, SSN – Same for friend or close relative – Primary MD

n Maintain periodic contact – Reminders, updates

n Use secondary data sources for critical info – Death registries, Medicare records, voter/driver registration

Cohort studies: Follow-up issues n Is the duration of F/U appropriate for the outcome(s) of interest n How is the outcome of interest measured? – Validity and reliability of measure addressed?

n Is a high F/U rate (85%) been achieved? n Is there a comparison of the characteristics of the unavailable group to the followed group? – Not needed if very high F/U rates achieved

Cohort studies: Selecting the design n Retrospective cohort design – Can the question be answered with data that already exists? – If yes, this is by far the most economical approach n Prospective single cohort design – If goal is descriptive, measure incidence rates n Prospective double cohort design – When exposures need to be measured precisely – Outcomes are relatively common n Ambispective cohort study – Could study prospectively but would take too long to get enough data. Can use prospective data to QA the retro data

Cohort studies: final comments n Your research tool-box should have many tools n Cohort studies are one of the most important ones n Become a research conservationist – Don’t conspicuously consume research resources unless absolutely necessary – Save the RCTs for when the target is known precisely and the expense is warranted