8 - Therapy Observational Designs Flashcards
What is the hierarchy of observational designs? (ie after RCT, within the observational category)
- prospective cohort
- prospective cohort w historical controls
- retrospective cohort
- case-control
- case series
what does a prospective cohort study look like? describe it.
- the only difference btw RCT is that there is no randomization now (people are either self-selected or have been selected into the the groups)
- now that there is no randomization, allocation concealment goes away (selection bias increases)
what do you look at in a prospective cohort study if you are worried about selection bias?
- look at the tables that describe the 2 groups
- if there is a big difference btw the 2 groups, could be due to a selection bias causing the outcome
- physicians should know what values prognostically should balance
- basically make sure groups are balanced
what does statistical analysis look like for a prospective cohort study? - what is the term called? what do we like to see?
- same as it would as for an RCT except:
- must adjust for confounders that might not be randomly allocated
- so if something is done to these random groups on purpose, dont control for them but if they might be random, we would want to adjust for them
- called adjusted or stratefied analysis
- we like to see consistency btw the 2 models (slide 8), if very inconsistent it would mean differences are potentially due to differences btw groups!
what does a prospective cohort with historical control look like? describe
- one treatment group receives the intervention prospectively and another treatment group had received it in the past (the control group)
what are we missing in a prospective cohort w historical control study?
- we are missing randomization as well as some control (bc for data in the historical group, we do not make the decisions about data collection etc)
- dealing with selection bias and also problems to do with this historical data (ie were all criteria, co-interventions etc the same?)
describe a retrospective cohort study
- again a sample is not randomly allocated to Tx or Ct group
- this time the question is coming after the data has already been collected
what problems do we face with a retrospective cohort study?
- same problems as w historical study (no randomization and relying on others’ data again)
- is there accurate standardized charting? similar protocol? similar outcomes?
- completeness status threatens validity of any conclusions you will try and make from this data (worry a lot about applicability due to lots of lost data)
what advantages do we face with a retrospective cohort study?
- advantage: usually gives a large database, ie from a network of clinicians (so large data gives us great precision, but worry about validity of participants included/ whether data was collected well enough to not introduce a bias)
describe a case control study
- selecting the cases/controls based on outcome rather than looking at people who received tx/ct first then looking at outcome (backwards)
- look backwards in time and try to compare the proportion of people who had the outcome and received the tx to the proportion who didn’t have the outcome and received the tx
who is the case and who is the control in a case control study
the person who has had the outcome is the case and the person who hasn’t is the control
what are the disadvantages to case control studies?
- previous concerns (no randomization and historical data collected)
- also we now have serious concern about causation (can’t show that one thing happened before the other)
- also only know the prevalence (not the incidence)
- no idea whether the sample is a good representation of the population or not
- no idea the size of the population we would have had to come across to get the n patients (can’t calculate anything that requires incidence!)
describe selection of incident cases (disease-positive) for case control studies
- incident cases are preferred but refers to “at time of diagnosis” not necessarily onset of disease
- so treatment could have occurred before onset of outcome or behaviour could have changed as a result of symptoms that occurred before diagnosis
- temporal relationships = difficult to have
for case control studies: the greater the time between the onset and diagnosis ___
the more difficult it is to differentiate btw factors that were present before outcome and those that appeared after outcome
describe selection of cases for non-incident (prevalent) cases in case control studies
- these are cases where people die or are cured quickly
- these may under represent those who die quickly or who are cured quickly
- potential for bias is too great (we don’t want to include this type of study!)
