Term
| Criteria for Causality (Surgeon's General) |
|
Definition
Temporality,consistency, strength, specificity, and plausibility, where
Specificity: factor only associated with the endpoint of interest and is only factor found to be associated with endpoint
Strength: how strong is measure of association
|
|
|
Term
|
Definition
| – group of individuals under observation in the study who are expected to have the same exposure-disease association as the source population |
|
|
Term
|
Definition
| same exposure-disease association as the target population and who can be enumerated |
|
|
Term
|
Definition
| Always based on person-time |
|
|
Term
|
Definition
| Nr. newly diagnosed cases that arose by the end of follow-up divided by the total number in the study population |
|
|
Term
|
Definition
1. Lack of changes in risk across calendar time
2. lost of follow-up have same conditional probability as those who remain |
|
|
Term
|
Definition
| proportion of population with the disease at a specified time |
|
|
Term
|
Definition
| All person time accumulates after beginning of study |
|
|
Term
|
Definition
| exactly same as study group except in exposure |
|
|
Term
| Can we study causality in real world? |
|
Definition
| NO, because comparison group is different from counterfactual group. |
|
|
Term
| When do odds approximate risk (R or p)? |
|
Definition
|
|
Term
|
Definition
• Observer (diagnostic). Typical sub-type is digit error in hypertension.
• Instrument (method)
• Reporting
• Recording and Entry
• Person (time varying) |
|
|
Term
| Measured (observed) value |
|
Definition
| True value+Error=True value+Bias+Random Error |
|
|
Term
|
Definition
| Systematic deviation from true value |
|
|
Term
|
Definition
--Rigorous and detailed protocol
--Reliable and accurate instruments
--Measure consistently and correctly
--Data checks
--QA and QC analyses
--Repeated measurements
--Reliability and validity studies |
|
|
Term
| How to assess extent of bias and random error |
|
Definition
Bias: comparison with gold standard
Random error: reliability studies |
|
|
Term
|
Definition
Binary: sensitivity/specificity
Continuous: correlation/regression |
|
|
Term
| Methods to assess reliability |
|
Definition
Discrete: correlation/regression, % agreement, Kappa
Continuous: correlation/regression,coefficient of variation, ICC |
|
|
Term
|
Definition
| VAR(between individuals)/(VarBetweenINd+VarWithin) |
|
|
Term
|
Definition
| =(Observed Agreement %) – Agreement Expected by Change Alone / 100 |
|
|
Term
| Positive Predictive Value |
|
Definition
|
|
Term
| Methods to reduce misclassification |
|
Definition
Paralel:positive if positive in at least one --> more sensitive
Sequential: positive if positive in both --> more sensitivity |
|
|
Term
| Threats to causal inference |
|
Definition
1. Lack of precision (n too small, association due to chance/bias)
2. Bias (confounding, information, selection)
3. Incorrect assessment of direction
4. Lack of external validity |
|
|
Term
|
Definition
| Sampling error, measurement error, changes over time |
|
|
Term
| Low sample size has as main consequences __ |
|
Definition
| High variance and distorted associations |
|
|
Term
|
Definition
| 1. (simple) random sampling 2. Assumptions of statistical model valid 3. No biases |
|
|
Term
|
Definition
|
|
Term
| Confounding (counterfactual definition) |
|
Definition
| Frequency of disease in counterfactual unexposed different from frequency of disease from observed unexposed. That is, different distribution of factors that cause the disease. |
|
|
Term
|
Definition
1. Cause disease. Caveat: may be associated (therefore not confounder)
2. Associated with exposure. Caveat: may happen by chance.
3. Cannot be in causal pathway between exposure and disease
Important baseline characteristics ARE confounders (as fulfill all 3 previous criteria) |
|
|
Term
|
Definition
Association in big group reversed when stratifying.
Happens --> OR and MH-OR different, but no association C-E
10% diff OR / MH-OR cutoff for confounding |
|
|
Term
| Do NOT choose confounders based on |
|
Definition
Collapsibility
P-values
Stepwise selection |
|
|
Term
| CHoose confounders based on |
|
Definition
Compare adjusted OR to crude OR (says whether exposure has effect or not) this gives direction of bias
Look at whether exposure levels vary by confounder (crude table)
Look at how disease varies by exposure, and assess how confounding is biasing such relationship |
|
|
Term
| Control for confounding (design) |
|
Definition
| Randomization, restriction, matching |
|
|
Term
| Control for confounding (analysis) |
|
Definition
Stratification
Standardization
MVR
Inverse probability weighting |
|
|
Term
| Non-differential confounding |
|
Definition
Does not vary by exposure, confounder and others
Biases towards null |
|
|
Term
|
Definition
Varies by expoisure, confounder and others
Over-or underestimates, can have high impact |
|
|
Term
| Regression to mean (controlling for) |
|
Definition
Design: repeat measurements, RCT
Analysis: compare observations at tail of distributions, ANCOVA |
|
|
Term
|
Definition
| Occurs when selection affected by exposure (or cause of exposure) AND outcome (case-control)/ cause of outcome |
|
|
Term
|
Definition
Source population does NOT have to be same as target population, BUT:
Exposure-outcome association should be same
Confounders/mediators shoudl be same
Same mechanisms (social) |
|
|
Term
|
Definition
| Population, variables measured, validated in language, feasible, consistent with measures of other studies |
|
|
Term
| Effect modification - Error |
|
Definition
Increases type I error (multiple comparisons). As interactions have low power, many consider p=0.10 as significant. However must combine higher p-value with Bonferroni.
Increases type II error (more difficult to identify interactions)
SE larger than for regular estimates |
|
|
