Lecture 2: Study Designs in Pharmacoepidemiology Flashcards
Explain the Paradigm shift that Nicole speaks of in this lecture
The shift was from timely submission of documents to quality pharmacovigilance processes. The approach to postmarking drug safety went from reactive to proactive.
What are some epidemiology data sources?
Published literature (scientific literature, publicly available reports), Existing data sources (EMRs, Patient/drug/population registries, RCTs), and primary data collection (for De Novo studies).
Describe the spectrum of research approaches
Goes from controlled to real-world and from experimental research to observational research. The study types ranging from least real-world but most valid to most real-world, but difficult validity are: RCTs, Pragmatic designs (Large simple trials), Non-randomized primary data epi studies (cohort based, case control), cross sectional, secondary data sources
What are three things that cause invalid results in pharmacoepidemiology?
Random error, confounding (related to why ppl take the drug and the outcome), and Bias (information–how it is measured or labeled and Selection–how people are recruited or retained).
Sources of non-comparability
Nature, Sampling error & selection, labeling exposure, attrition and labeling disease
What are 5 requirements for a valid result in Epidemiology?
1) Exposed and unexposed must be comparable with respect to underlying disease risk
2) the exposure must be measured correctly
3) disease/outcome measured correct;y among exposed and unexposed
4) factors that contribute to confounding must be measured/controlled for correctly
5) exposed and unexposed should not be subject to Loss to followup.
What is invalidity?
Non-comparability–the probability of disease in the unexposed is not a good substitute for the counterfactual risk. So we won’t be able to identify the causal effect of the exposure from the association between exposure and disease.
There are alternative explanations for the E/D relationship aside from E causing D.
Non-comparability occurs when the counterfactual (disease in unexposed does not equal the proxy of the counterfactual). The more far off the proxy for the counterfactual is from the counterfactual, the further the non-comparability will be.
What is the counterfactual?
The disease experience of the exposed had they not been exposed.
Structure of Randomized control trial
Population N is randomized to exposure and then the exposure groups are are followed for disease. Investigator controls exposure, so he/she splits confounding variables equally. More on RCT: Randomization: Yes, Sample size: small, Inclusion criteria: Narrow Questionnaires: Complex/lengthy Endpoints: All Required patient visits: Yes Physicians/Investigators: Clinical research/academic centers Site monitoring: Frequent Continue FU if Tx discontinued: No Naturalistic: No
What are some study designs with observational FU?
Large simple trials, cohort studies (disease and product registries), case-control studies, cross-sectional studies.
Structure of Cohort study
Population N is self selected into exposure and then the exposure groups are followed for disease.
Investigator does not control exposure, validity must be avoided through design/control, patients should not be lost to FU, measure exposure and outcome accurately.
Types of cohort studies in pharmacoepi
Pre-approval (studies of people with the indications of interest which look at natural progression of the disease, epidemiology and risk factors for the indications) and Post approval (studies of people taking the drug of interest [drug utilization/adherence, comparative safety/effectiveness, and active surveillance
What are similarities & differences between RCT and cohort study?
1) when the study period begins
2) whether exposure is randomly assigned or self-selected
3) RCT avoids confounding, cohort study is subject to it
4) Both are subject to bias
Structure of Case-Control study
1) Case-cohort study (controls are sampled from the overall N population)
2) Incidence density study (controls are sampled/selected after exposure groups have been determined)
3) Cumulative incidence study (controls are selected/sampled from after disease occurs)
Exposure is not under control of investigator, confounding must be avoided through design/control, no loss to FU, exposure and outcome measured accurately, controls must represent the same population that gave rise to cases.
Hierarchy of study designs: a few notes
Cohort study (and thus, RCT) underlies all case-control designs, the only differences: CC includes only sample of non-diseased, when and how controls are sampled, cohort study and case-control subjecting to confounding and all including RCT are subject to bias.
What measure of association does a case-cohort provide?
Pseudo RR
What measure of association does a incidence density study provide?
Pseudo IR
What measure of association does a cumulative incidence density study provide?
OR
More on primary data Observational Studies
Randomization: No, Sample size: Larger, Inclusion criteria: Broad Questionnaires: Brief Endpoints: Hard Required patient visits: No Physicians/Investigators: community based Site monitoring: minimal Continue FU if Tx discontinued: NA Naturalistic: Yes
More on secondary data observational studies
Randomization: No, Sample size: Larger, Inclusion criteria: Broad Questionnaires: NA Endpoints: Hard Required patient visits: NA Physicians/Investigators: community based Site monitoring: NA Continue FU if Tx discontinued: NA Naturalistic: Yes
Large Simple trials are:
They were conceived to study effectiveness of interventions, but more recently used for comparative safety. They are a hybrid between randomized clinical trial and observational cohort study. Participants are randomized to treatment groups, with follow-up as per routine practice.
Its different from RCT in the sense that it has observational follow-up like an observational study (cohort), while RCT has tightly controlled FU. They are similar to RCT in the sense that it is randomized.
More on LST
Randomization: Yes, Sample size: Larger, Inclusion criteria: Broad Questionnaires: Brief Endpoints: Hard Required patient visits: No Physicians/Investigators: community based Site monitoring: minimal Continue FU if Tx discontinued: Yes Naturalistic: Yes
RCT, Observational study and LST (generalizability and validity)
RCT maximizes validity, but limited generalizability
Observational study maximizes generalizability and validity depends on choice of design and control for bias
LST maximizes validity and generalizability.
RCT and LST
RCT underlies the LST. Follow-up is more observational in LST and more reflective of real world 9both studies avoid confounding, but LST more subject to non comparability after randomization because of simplicity.
Case-Crossover
Similar to case-control, includes only cases (avoids selection bias), each case serves as his or her own control
eliminates confounding by factors that do not change over time.
Case window (in case crossover)
The hypothesized at-risk time period preceding the event
Control window (in case crossover)
another time period of the same length, but not overlapping with the case window
What do you need to know about validity with Case crossover?
Requires precise knowledge about: time at risk (case window), the effect period of the exposure
Case selection must be independent of exposure
Exposure data quality must be equivalent for case and control windows
Exposure time trend must be absent
New User Design
Patients are new-users of the study medication–follow up begins at initiation of therapy.
Data for confounders collected prior to initiation
Power is limited in this study and retrospectively obtaining long-term drug use information may be difficult.