Cohort Studies

Cohort Studies: Utility

-It is possible to study multiple exposures of interest to see which ones contribute to outcome of interest -for example what if the many exposures contribute to the incidence of heart disease of persons in the Framingham Study

Multiple Outcomes in a cohort study

-Study of multiple outcomes in a cohort study that starts w/exposed and unexposed groups -in a study that starts with a defined population, multiple exposures and multiple outcomes can be studied

EWhen to Perform A Cohort Study

-There is evidence of potential risk factors to be studied from other data (case-control, cross-sectional etc) -A new agent that may alter the risk of several disease is introduced [OC and ET (estrogen therapy)] -Where we can minimize the potential for loss to follow-up (short interval b/t exposure and disease)

Limitations: Prospective Cohort Studies

•Requires a large investment of time, human resources and is very expensive •Requires large sample sizes, unless the risk period is very long -Nurses Health Study 110,000 -Physicians 59,600 •Loss of subjects is a problem •Cannot use for rare diseases (use case-control) •Not easy to reproduce the findings, so it is difficult to look at consistency of effect --compare w/other papers, see if consistent results with your results

Analytic Bias

If the investigator and statisticians who are analyzing the data have any strong preconceptions of the findings it can introduce bias

Risk Ratio

RR= (a/a+b) / (c/c+d)

Temporality

Refers to the timing of information about cause and effect •did the info about cause and effect refer to the same point in time? •Or, was the info about the cause garnered before or after the info about the effect?

Retrospective Cohort Study

Subjects are selected on their exposure status and analyzed by exposure status. However, the time for the follow-up has already taken place and the subjects have either become diseased or not. The cohort is reconstructed from records -Look at those from cohort who got disease in the present, based on PAST exposure -Historical Prospective: Exposure group in past and present, look at those who got disease in future--ask more questions, then follow later and see if more diseased cases

Incidence Rate

• Calculation of the true rate of disease •Numerator: all incident cases of the disease over the follow-up period •Denominator: the sume of time contributed by the study subjects while they are still at risk for the disease -when a person gets the disease we stop counting their person-time • incidence rate in exposed: a/(a+b) •incidence rate in unexposed: c/(c+d)

What is the difference b/t absolute and relative measures?

•Absolute measures have greater public health importance b/c they tell us how many people are prevented from getting the disease if the exposure is removed •Relative measures are more useful in determining causality, how strong is the assoc seen?

Measures of Association

•Attributable risk=cumulative incidence in the exposed - cumulative incidence in the unexposed •the difference of risk in the exposed and unexposed groups •It provides the total increase in the incidence of disease attributable to the exposure •Conversely, if you could remove the exposure how many cases of disease you could avoid

Framingham Study

•Cardiovascular disease study •Residents were eligible if b/t 30 and 62 yeas of age--in the framingham area •Wanted a final sample size of 5000 •Exposures examined were smoking, obesity, hypertension, elevated cholesterol, and low levels of physical activity •Outcomes were new coronary events that were determined in patients examined very 2 years •Daily surveillance of the local hospital to assist in outcome determination (follow records, see if anyone died/dropped out)

Comparison (Non-Exposed Group) [control group]

•Cohort studies involve the comparison of disease rates b/t exposed and non-exposed groups •the comparison group is similar in demographics and geography to the exposed group, but lacks the exposure •in an occupational setting, several categories of exposure may exist

Cohort Study vs. Case-Control Study

•Cohort: -comparing: exposed people and un-exposed people -comparing the incidence rates of disease in: exposed people and non-exposed people •Case-Control: -comparing: people with the disease and people without the disease -comparing the proportions exposed in: people with the disease and people w/o the disease

Limitations of Other Study Designs

•Demonstrating temporality is a difficulty of most observational studies •Cross-Sectional and case-control study designs are based on exposure and disease info that is collected at the same time -advantage: efficient for generating and testing hypotheses -disadvantage: leads to challenges regarding interpretation of results •Cross-Sectional studies: -present difficulties in distinguishing the exposures from the outcomes of the disease, esp if the outcome marker is a biological or physiological parameter -ex, looking at Vit levels in a group of people, making sure they don't have high levels in blood before experiment •Case-Control studies: -Raise concerns that recall of past exposures differs b/t cases and controls

Exchangeability

•Fulfilling the criteria that the control group (the unexposed in a cohort) represents the disease experience of the exposed group had the exposed group not been exposed •We need an even distribution of potential confounders b/t the exposed and unexposed to achieve this •In other words--aside from the exposure itself, the exposed and unexposed groups are comparable in every relevant way (age, gender, smoking, obesity, etc)

Information Bias

•If the quality and the extent of information obtained is different for exposed versus the non-exposed group, it will lead to bias •This is particularly possible in the retrospective cohort design--relying on records

Assembling the Cohort

-be sure that all individuals in the cohort are at risk for disease; exclude others (e.g. women w/o ovaries from ovarian cancer study) -people w/a known or suspected history of the disease should be excluded -perhaps limit to a group at higher risk for disease (e.g. hip fracture study, limit to over age 65, don't want people to young, waiting for a very long time before develop disease) -perhaps collaborate w/other geographic areas to increase sample size and improve generalizability--allows to apply to other people not in geographic are studied in expt

Advantages: Retrospective Cohort Studies

-can study rare exposures -less expensive than a prospective cohort study -good for occupational studies -may be quick

Determining Exposure Status

-questionnaires -laboratory tests -physical measurements -special procedures -may need to repeat these at intervals throughout the study for changes in risk factors •Hawthorne effect--be careful asking about diet at every visit, they may begin to eat different foods--'the study itself influences bx'

Advantages of Prospective Cohort Studies

-the exposure to the potential cause or risk factor is assessed before the disease occurs -it is possible to estimate all measures of incidence and effect -it is possible to study several outcomes (diseases) to one cause •recall bias would not be a problem

Design of a Cohort Study

-to examine the possible relationship b/t an exposure and a disease -identify a group of individuals free of disease and identify them by their exposure -follow the cohort over time to determine the incidence of disease in the exposed versus not exposed •Find group of those exposed and those not exposed -follow over time, compare groups w/number of people who develop the disease and those that don't develop the disease

What is a cohort?

A population group, or subset thereof, that is followed over a period of time -cohort group members experience a common exposure assoc with a specific setting (e.g. an occupational cohort or school cohort) or they share a non-specific exposure assoc w/a general classification (e.g. a birth cohort--being born in the same year or era) -the term cohort is said to originate from the Latin cohors, which referred to one of ten divisions of an ancient Roman legion

Internal Validity

A study is internally valid if the assoc b/t exposure and disease seen in the study, is a good estimate of the "real life" exposure and disease assoc that exists in the study population •Internal validity is influenced or ruined by: -Confounding -Bias -Chance

Confounding

•In a cohort study the subjects have 'chosen' their exposure--so there is a greater chance for confounding (lack of exchangeability) •For example, if we were to study alcohol drinking and liver cancer, there is a RR=5 ( the risk of liver cancer is 5x greater among drinkers compared to non-drinkers) •However, more of the drinkers smoke than the non-drinkers so we don't know if the elevated risk is due to alcohol or confounded by smoking

Selection of a Comparison Group of Unexposed Individuals

•Internal comparison group: most likely when you start w/a population that you then evaluate (Framingham) •External comparison group: may be required for some studies where all are exposed (e.g. occupational exposure). Usually it consists of the general population which the exposed population comes from

Cohort Studies: Lost to Follow-Up

•Lost to follow-up if it occurs we hope that losses from each of the following groups is equal: a) exposed and gets disease b) not exposed and gets disease c) exposed and does not get disease d) not exposed and does not get disease •If losses from group (a) are greater, then the exposure and disease relationship will be underestimated •If losses from group (c) are greater we will overestimate the exposure and disease relationship •Unequal loss from these groups will result in bias of the RR and AR

Measurement of Disease

•Must be comparable for exposed and unexposed •Should be performed by persons unaware of exposure status (blinded) •Record ascertainment may be incomplete •Examination of cohort at predetermined intervals is ideal •Establish diagnostic criteria before the study begins

Limitations: Retrospective Cohort Studies

•Not good for rare diseases •Problems in ascertaining disease outcome •Problems in ascertaining the exposure -need to determine duration and intensity of exposure

Cohort Studies: Ascertainment Bias

•Outcome ascertainment bias •Since there is no blinding in the cohort study, there may be ascertainment bias in that the clinician determining disease status may be influenced by knowing exposure status of the study subject •Rationale for objective criteria for disease ascertainment -doctor looks harder for whatever the disease that they are looking for

Measures of Association

•Risk Ratio=cumulative incidence in exposed / cumulative incidence in unexposed •A risk ratio of 3 says that the risk of disease in the exposed is 3 times that of the unexposed •RR=1 exposure does not increase risk of disease •RR<1 exposure appears to be protective •RR>1 exposure increases the risk of disease •Risk ratio is a relative and not absolute measure

Cohort Studies

•Start with a group of subjects who lack a positive history of the outcome of interest and are at risk for the outcome •include at least two observation points: -one to determine exposure status and eligibility -two (or more) to determine the number of incident cases •permit the calculation of incidence rates •can be thought of as going from cause to effect •involve the collection of primary data--not using records, asking people in cohorts questions about themselves and collecting blood samples

Important Concepts

•Study population: the actual members of your study •Target population: the population to whom you would like to generalize your results •External validity or generalizability: a study has good external validity if you can generalize your results to other populations -who do you want to generalize to? -does the exposure disease relationship vary across diff populations

Power and Sample Size

•The larger the sample size of a study the more likely you are to see an exposure and disease assoc, given that the assoc really exists •The POWER of a study is the probability of finding a statistically significant assoc in the data, given that the assoc really exists in the study population -Increase chances of a positive finding if have large sample size

Cumulative Incidence

•The proportion of subjects who develop the disease over the follow-up period of the study •The average probability that an individual in the cohort will develop the disease over the follow-up period •Numerator: all incident cases of disease over the follow-up period •Denominator: all people at risk of the disease at baseline, all subjects at beginning since chosen as disease free •Should not calculate if persons are lost to follow-up

Prospective Cohort studies

•Prospective: -Subjects are identified by their exposure status at the beginning of the study and they are disease free and the disease status only becomes apparent at the end of the follow up period -exposure group in present, looking at those who got the disease in the future •Also called concurrent cohort •only individuals free of the disease of interest are included (subclinical disease) •Investigation goes through time along with the subjects in the trial •Disadvantage is the very long time to complete the study

Exposure-Based Cohort Studies

Definition: an exposure-based cohort is made up of subjects w/a common exposure, for example, workers exposed to lead during battery production •these studies overcome limitations of population-based cohort studies, which are not efficient for rare exposures •certain groups, such as occupational groups, may have higher exposures than the general population to specific hazards •Start with a defined population--ex, geographic -then not randomly assigned to 2 groups: exposed and not exposed -then see who gets the disease and who doesn't -Exposure-based cohort: used for rare diseases -population cohort: used with a large population

Cohort Start Point

Determine exposure/cause of disease •Subjects CHOSE their exposure -is there an assoc b/t a risk factor or characteristic and the development of a disease -is there a causal relationship b/t the risk factor and the disease?

Experimental v. Observational Studies

Experimental (Randomized Study): -Population--randomly allocated into exposed and not exposed •Observational (non-experimental, non-randomized) Study: -Population--not randomly allocated (e.g. self-selected) into exposed and not exposed -cohort studies

Exposure is not randomized

b/c the subjects were not randomized it is unclear whether the exposure caused disease -or whether some factor that leads to the exposure, for ex residing near the factory is the true exposure --hardest part of cohort study is teasing out what you want to look out from other potential exposures

Analysis

•When the length of follow-up varies •Survival analysis--product limit analysis =1 - probability of NOT developing the disease in any of the time intervals •Cox proportional Hazards model -paper examples -understand: taking into account the # of person years occur before developed disease •ex, product limit estimate: 0.395 -interpretation: in the absence of withdrawals due to death or loss to follow-up one would expect that 39.5% of the cohort would develop disease by time t -b/c of withdrawals only 25% of the cohort (5/20) has actually been observed to develop the disease

Combined Retrospective and Prospective Cohort

•You can combine the two designs •the major difference is the time frame •exposure is from past objective records, may gather more info over time •then cohort is followed up into the future

Active Follow-Up

•the investigator, through direct contact w/the cohort, must obtain data on subsequent incidence of the outcome (disease, change in risk factor, change in biological marker) •Accomplished through follow-up mailings, phone calls, or written invitations to return to study sites/centers -frequent contact is important in cohort study or will lose people