- Comparison is essential component of epidemiology
Design of Case Control Study
- group of individuals with disease à cases
- group of people without disease à control
- if exposure is related to disease à prevalence of history of exposure will be greater among cases than among controls
- first select then measure past exposure
- case control study cannot estimate prevalence of disease
- difference between cohort / case control
o case control à begins with people with disease and compares them to controls
o cohort à begins with group of exposed people and compares them to group of non exposed people
o difference is not prospective / retrospective
§ can have retrospective cohort study
- example
o artificial sweeteners and bladder cancer
o cigarette smoking and lung cancer
§ not just exposed à exposure data further stratified in terms of dose
Selection of Cases and Controls
- Selection of cases
o if cases selected from specific hospital à cases may be biased due to referral pattern
- incident or prevalent cases
o newly diagnosed or not
o best to use incident cases
§ any risk factors we may identify in a study using prevalent cases may be more related to survival with the disease than to development of disease
· eg, if people die soon after diagnosis, they will be under represented in study that uses prevalent cases – such a study is more likely to include longer term survivors
· even if include only incident cases à excludes patients who die before diagnosis
o Selection of Controls
§ if we conduct a case control study and find more exposure in the case than in the controls, we would like to be able to conclude there is an association
· way controls are selected is major determinant of whether conclusion is valid
§ should they be similar in all ways to cases other than having disease
§ or
§ representative of all persons in population from which cases are selected (reference population)
o Source of controls
§ non hospitalized persons as controls
· ideally, probability sample
· neighborhood control à resident in neighborhood where case lives
· best friend control
§ hospitalized patient as control
· captive population
· but they represent sample of ill defined reference population
· attractive to compare hospitalized cases to hospitalized controls from same institution – presumably come from same reference population
· all other patients vs other patients of specific diagnosis
· will not be clear if it is the cases or controls who differ from general population
· option – not exclude any diagnosis groups from control, but to analyse study data separately for different sub groups
o Problems in Control Selection
§ Example
· case control study of cancer of pancreas
o apparent dose response relationship between coffee consumption and cancer of pancreas, particularly women
o difficult to know whether disease is caused by coffee consumption or by some factor closely related to coffee consumption
o coffee consumption closely related to smoking
o problem – controls were other patients of attending doctor of cases à doctors specialize à therefore their other patients were exposed to preventive talks etc
§ cannot assume controls intake of coffee was representative of general population.
§ this observed difference in coffee consumption between pancreatic cancer cases and controls may be result not of cases drinking more coffee than expected, but rather of controls drinking less coffee than expected.
§ when a difference in exposure is observed between cases and controls
· was level of exposure observed in controls really level expected in the population in which study was carried out
· or
· controls may have particularly high or low level of exposure that might not be representative of level in pop
o Matching
§ selecting controls so they are similar to cases in certain characteristics
· group matching
o frequency matching
§ eg, if 25% of cases are married, select 25% of controls who are married.
· individual matching
o each case individually matched to control
§ 25 yo white female
§ problems
· practical
o difficult to match on too many characteristics
· conceptual
o once we have matched on a characteristic à we cannot study that characteristic
§ by matching, we ensure the same prevalence of that factor in cases and controls
o unplanned matching
§ ie neighbourhood matching à also implicit matching on other factors
· eg, social / economic
Problems of Recall
- limitations in recall
- recall bias
- Limitations in Recall
o subject does not always have correct information
§ circumcision example
- Recall Bias
o eg, mother of abnormal child will try harder to identify reason for abnormality
Use of Multiple Controls
- controls of same type
o why not just increase number of cases
§ relatively infrequent disease
· increase power of study by increasing number of controls.
- multiple controls of different types
o eg, add neighborhood control to hospital control
o first decide which comparison will be bible
When is a case control study warranted?
- when association shown, next step is to do cohort study
- rare disease
- length of time between exposure and disease
Case Control Studies Based in a Defined Cohort
- case control study embedded in cohort study
o pop identified and followed over time
o baseline data obtained
o small percentage of subjects develop disease / most do not
§ subjects who develop disease à cases
§ select controls from those who did not get disease
- cohort based case control studies
o nested case control studies
§ controls are a sample of individuals who are at risk of the disease at the time each case of the disease develops.
§ as each case/s develops, same number of controls are selected.
§ cases and controls matched on calendar time and length of followup
o case – cohort studies
§ controls are randomly chosen from defined chort with which the study began
§ controls not individually matched
- advantages of embedding a case control study in a defined cohort
o all data obtained before disease has developed
o cost saving
Other Study Designs
- Case Crossover Design
o diseases with acute outcomes
o each person serves as own control
o case identified
§ exposure determined for short period preceding (at risk period)
§ exposure also determined over control time which is more remote from event
- Cross Sectional Design
o both exposure and disease outcome are determined simultaneously
o prevalence study – cases that existed at time of study
o association
§ calculate prevalence
· of disease
· of exposure
§ association may be with survival after disease rather than with risk of developing disease
§ not possible to establish temporal relationship
No comments:
Post a Comment