Study design in etiologic research Flashcards
What are the three points in Design of data collection
- Experimental or observation
- Census or sampling
- Time
What are the four types of epidemiological research questions? Are they descriptive of causal?
- Diagnostic, predicts presence if disease (descriptive)
- Etiologic, explains occurance of disease (causal)
- Prognostic, predicts the course of a disease (descrptive)
- Therapeutic/intervention, defines medical action that should be taken (causal)
What is the design of data collection in Cohort studies?
- Non experimental
- Time
- Census (most of the time) or sampling
Definition of a cohort study
A group of subjects from whom data are collected and followed over time. A cohort can change in size, characeristics and membership over time. There is a difference in a fixed cohort and a dynamic population.
Describe a cohort with a fixed population.
Once a person experiences the defining event they remain part of the cohort as long as they are alive. A fixed cohort is changing over time in size (people die or are lost to follow up), age and gender distribution, but not in membership. E.g. birth cohort of babies born in 2017.
Describe a cohort with a dynamic population.
Persons can join and leave the cohort at every time. In a stable situation a dynamic cohort is not changing in determinants (age, sex) or size, but is changing in membership (new people replace old people). E.g. residents in a city, members of a health insurance plan
What is the meaning of a retrospective of prospective data collection?
In causal research, the study design is always prospective (determinant comes before outcome). In diagnostic research outcome and determinant occur at the same time. Data collection is prospective when data is collected after the research question and retrospective/historical if data is collected before the research question.
Strengths of a cohort study (5)
- able to calculate incidence in exposed/unexposed and true relative risk
- Best design for rare exposure
- Less subject to biases because outcome is not known. Since subjects are enroled before outcome, you can collect information on multiple determinants without recall bias
- multiple effects of exposure can be investigated
- temporal relationship exposure-outcome is clear
Limitations of a cohort study (7)
- Not suitable for rare outcomes
- Generally time-consuming (except retrospective cohort)
- Expensive (except retrospective cohort)
- Long latency period
- Change in exposure
- Lost to follow up (slective)
- Ethical considerations (because it takes a long time te perform)
Definition of domain?
The theoretical population to which the study findings apply (target population)
Definition of source population? And which two things are important for a source population in cohort studies?
The population from which the study population is sampled.
- The source population does NOT have to be representitive of any particular broader population
- Important is that the exposed and the unexposed groups come from the same source population in order to be comparable
Definition of study population?
Subjects that are selected from the source population and are observed by the investigator. Should be part of the domain and source population and is the result of selection criteria and logistic circumstances.
Characteristics exposure distribution (4) in cohort studies
- Exposure distribution in study population does not have to be representative for the exposure distribution in the source population or domain
- you can select al exposed and sample unexposed persons
- Power is highest when exposed and unexposed are of equal size (e.g. select)
- Because the exposure is measured before outcome, there is slection but not slection bias
What is the main difference between a follow-up (cohort) study and a case-control study?
In follow-up study the slection is based on exposure status, in case-control it is on outcome
Characteristics of follow-up in a cohort study (2)
- Presence or absence of risk factors are determined before outcome occurs (t>0)
- Follow-up is until the outcome, death, lost to follow up or end of the study
Reasons for follow-up in a cohort study (3)
- Determine if and when study endpoints happen
- Determine which member are currently under observation
- Obtain more information (change in determinant or confounders) –> completeness is essentioal because of loss of statistical pwer and could lead to bias when there is a relation between loss to follow-up and the outcome
Strengths of a cohort study (5)
- Able to calculate incidence in exposed/unexposed and true relative risk
- Best design for rare exposure
- Less subject to biases because outcome is not known. Since subjects are enroled before outcome, you can collect information on multiple determinants without recall bias
- multiple effects of exposure can be investigated
- temporal relationship exposure-outcome is clear
Prospective vs retrospective cohort
Retrospective is not neccessarily a negative qualification, but sometimes prospective data collection may provide data that are qualitativly or quantitavily better because it is collected after the research question. They are not synonymous to case-control and cohort!
What are the four typical summaries of exposure
- Current level (yes, no; ever, never; current dose)
- Average exposure (in a specific period)
- Duration of use (of a minimum dose)
- Cumulative exposure (sum of each exposure multiplied by the time; pack years)
What are the four typical summaries of exposure
- Current level (yes, no; ever, never; current dose)
- Average exposure (in a specific period)
- Duration of use (of a minimum dose)
- Cumulative exposure (sum of each exposure multiplied by the time; pack years)
Definition of induction time:
The time required for the effect of a specific exposure to become manifest, measured as the interval between first exposure - or achievement of cumulative exposure - and the occurence of overt disease.
- Study may be uninformative when follow-up is too short (or too long)
–> stratify for time since exposure to prevent dilution effect
Latency period
The time interval during which a disease is present but not detected.
- Important in early detection/screening –> lead-time
- If survival is compared between screen-detected cases and clinically detectes cases –> lead time bias
–> i.e. screen-detected cases seem to live longer just because they were detected earlier, even if earlier treatment does not improve survival.
What type of frequency and association measures are used in a fixed cohort and a dynamic population?
Fixed cohort
- Cumulative incidence (denominator is fixed number of people contribute to the cohort) (assumption = all subjects are followed for the same amount of time)
- risk ratio
- risk difference
- incidence density and rate ratio when you have to deal with competing risks, lost to follow-up (long follow-up time)
Dynamic population
- Incidence density (denominator is total time of follow-up)
- Rate ratio
- Rate difference
Which three assumptions does a rate have?
- incidence rate is assumed to be constant over time-window in which it is calculated
- If not: follow-up time should be defined in finer strata
- It is reasonable to be constant for a particular Age category
How do you calculate incidence density when follow-up is long and incidence vary with age
Split time period into categories for when the incidence rate can assumed to be constant.
- Assumption: Homogeneity of person time: experience of individuals is interchangeable. Ten months of 1 individuals is equal to 1 month of 10 individual’s
Thee aspects of vailidity
- Selection bias: in prognosis between exposed and unexposed
- Confounding bias: in determinants of the outcome
- Information bias: in complete, valid and precise collection of data
Thee aspects of vailidity
- Selection bias: in prognosis between exposed and unexposed
- Confounding bias: in determinants of the outcome
- Information bias: in complete, valid and precise collection of data
Selection bias in cohort studies (4)
External validity if
- the non-response is related to exposure status and to the disease status (will affect disease rates and rate ratio in an over- or underestimation), but unlikly in cohort studies because of unknown outcome at the start.
- if there is a relationship between lost to follow-up and determinant or outcome
- more a problem in case-control than cohort
- you can not adjust, you have to prevent. You can perform a sensitivity analysis - worst/best case scenario (all follow up do have the outcome or not)
Confounding bias in cohort studies
Effect estimate is distorted because mixed with effect of extraneous variable
- variable is associated with the determinant and the outcome, but not in a causal pathway
Dealing with confounding (3), (2)
Design: randomization, restriction/selection, matching
Data-analysis: stratification (rule of thumb 10%), multivariable analysis (many confounders)
- Confounders should be completely measured at baseline
- Keep in mind residual confounding (not measured confounders)
Effect modification deffinition (3)
= statistical interaction
- e.g. the effect of smoking depends on gender
- if present, results should be presented stratified
- Interaction should be considered before start of study
Information bias in cohort studies
Present when there are errors in obtaining information on exposure or outcome status, also called miclassification.
Differential misclassification
Misclassification of outcome differs between expsoure groups
- in a cohort study: differential misclasification of disease causes a distortion of the relation between exposure and outcome that results from exposure dependent misclassification
Non-differential misclassification
Under ascertaiment only (decreased sensitivity, but perfect specificity of disease mearurement method)
- cases are only percentage of actual cases and precentage is similar for expsoed and unexposed: underestimation of both CI’s but no difference in relative risk
–> relative risk unaffected
–> risk difference underestimated
Over ascertainment only (decreased specificity, but perfect sensitivity of disease mearurement method)
- dilution with subjects without disease
–> bias in relative risk towards null (underestimation
–> risk difference also underestimated
Dealing with information bias in cohort studies (2)
- use of standard techniques
- use blinding of researchers
Properties of causal diagrams (2)
- Directed: the arrwos can be interpreted as direction of the causal effects
- Acyclic: a variable cannot cause itself, either directly or through other variables (if you start in V1, you never come back to V1 only to V2, V3, V4…)
Definition cross-sectional study
Determinant and outcome are collected at the same point in time for each patient. Regarding to the theoretical design: a cross-sectional study van be etiologic or diagnostic. All individuals are observed only once.
Theoretical design of a cross-sectional study
- Observational
- t=0
- census