Reliability, Validity, and Bias in Epidemiological Studies

Question 1

compare precision versus accuracy

Answer 1

precision:
-measure of repeatability
-reliability

accuracy:
-degree to which a measurement reflect true status
-validity

Question 2

what types of errors can impact study results?

Answer 2

1. random errors (precision)
   -inaccurate transcription, typing errors, biological variation, etc.
   -CAN overcome with a large enough sample size
2. systematic errors (accuracy, bias)
   -programing, unclear definitions, not following protocols
   -canNOT overcome with increased sample size
   -categories: selection bias, information bias, confounding
   -all observational studies have built in bias that will impact the study results to some degree

Question 3

describe types of validity

Answer 3

1. internal validity: ability of a study to measure what it intends to measure
   -MUST at least have internal validity to even hope extrapolate to another population, if not, your study is just trash
2. external validity: ability of a study to be extrapolated to a target population

Question 4

describe selection bias

Answer 4

1. absence of comparability between groups being studied
2. can prevent via:
   -using a probability sampling strategy
   -this is challenging to do in real life though

Question 5

describe information bias

Answer 5

1. incorrect determination of exposure, outcome, or both
2. differential misclassification:
   -errors in data collection varies by study group resulting in unpredictable bias (can either be away or toward the null)
   -if get insignificant results, you have no idea what it means
3. non-differential misclassification:
   -errors in data collection are the same for each group, resulting in “noise” and a bias towards the null
   -prefer this because if you get insignificant results, you have probably just underestimated
4. how to prevent:
   -standardize protocol
   -pre-test equipment (validated)
   -blind researchers/observers to avoid bias

Question 6

describe confounding

Answer 6

1. mixing or blurring of effects
   -association between exposure and outcome deviates from true value due to the influence of other variables
2. features of a confounding variable:
   -variable associated with exposure of interest
   -variable associated with outcome of interest
   -not in causal pathway (not an intervening variable)
3. how to prevent:
   -randomizing
   -matching
   -restriction

Question 7

describe non-probability versus probability sampling

Answer 7

non-probability: consecutive, convenience, or judgement samples
-key features: unknown probability of inclusion, prone to bias, not necessarily representative

probability sampling: simple random, stratified random, cluster, or multi-stage sampling
-key features: known probability of inclusion, not necessarily an equal probability, representative of a larger population