Lecture 3: Biases in Pharmacoepidemiology Flashcards Preview

Pharmacoepidemiology > Lecture 3: Biases in Pharmacoepidemiology > Flashcards

Flashcards in Lecture 3: Biases in Pharmacoepidemiology Deck (13):
1

types of Non-comparability

Confounding in nature
Bias (created by the design/conduct of a study)

2

Types of Bias

Information (created by labeling of exposure and/or outcome)
Selection (created by how study population is created or followed)

3

Sources of Non-comparability ( Confounding in pharmacoepi)

Caused by risk factor for the disease that is also associated with use of the medication

4

Common confounders

Indication (or contraindication) for drug use
Severity of underlying disease
Physician preference
co-morbidities
co-medications

5

Sources of Non-comparability ( Labeling of exposure)

exposed labeled as unexposed or vice versa
This is due to complex exposure patterns (unclear about how to define exposed), medication compliance (affects reporting of medication use and thus EMR and Claims Databases--> does Rx = DRUG FILLED = DRUG TAKEN?)

6

Sources of Non-comparability ( Labeling of outcomes)

disease labeled as non-diseased or vice versa.
This is due to multiple outcomes in the same patient, reporting of outcomes, EMR and Claims databases (does Dx code = has outcome?)

7

Sources of Non-comparability ( sampling error and selection)

error in selecting unexposed study subjects (cohort) or controls (case-control)...in other words, an incorrect source population.
This is due to controls being selected DEPENDENT on exposure status [by 1) over-matching controls to cases, especially on unmeasured confounders 2) selecting healthy controls] or unexposed selected on factors related to baseline disease risk (so exposed and unexposed are from different sources)

8

Sources of Non-comparability ( attrition)

FU of study subjects is dependent on disease
This is due to long studies/long time between visits, also because study subjects may switch providers, move among health plans, geographically, die without knowledge of key providers, lose interest in participation or become hospitalized

9

Randomized Trial (Confounding and Bias)

Confounding: Randomization should evenly distribute known and unknown confounders (if randomization is broken, confounding occurs)
Bias: Susceptible to information and selection biases

10

Observational studies--cohort and case-control (Confounding and Bias)

Confounding: Occurs, so you must hypothesize, measure and control for confounders validly
Bias: susceptible to information and selection biases.

11

Randomized trial (Unique sources of bias)

non-compliance with randomized treatment and unblinding of randomized treatment.

12

Observational study (Unique sources of bias)

If not nested, case-control studies are susceptible (uniquely) to recall bias and bias from selection of controls

13

What is Random error?

Lack of power or chance findings. it includes Random confounding and random measurement error.