Study Guide PDF

Studies: Observational ○ Case study - looking at one person ○ Descriptive - describing a pop. ○ Ecological - observing entire pop. (ex - smoking rate in boston) ○ Cross-sectional - snapshot/single point in time (ex - survey) ○ Case-control - subjects...

Studies: Observational ○ Case study - looking at one person ○ Descriptive - describing a pop. ○ Ecological - observing entire pop. (ex - smoking rate in boston) ○ Cross-sectional - snapshot/single point in time (ex - survey) ○ Case-control - subjects selected based on disease outcome/status Ex: Ppl w/ lung cancer and without, ask ab smoking habits ○ Cohort studies - subjects selected based on exposure status Prospective: follow over time to see if exposure → disease outcome (longitudinal) Experimental ○ Randomized control trial (RCT) - subjects randomly assigned to exposures/treatments Calculations: RR OR ○ Risk Ratio - Incidence of exposed vs incidence of unexposed RR: (A/A+B) / (C/C+D) Significance: If p-value X 1: Positive association (increased risk/incidence), RR < 1: Negative association (decreaased risk/incidence) ○ Odds Ratio - (A/C) / (B/D), case-control Interpretation - compared to unexposed subjects, exposed subjects have 1.33 times the odds of getting the disease or a 33% increased odds of getting the disease Interpretation: Compared to non-smokers, smokers have 0.44 times the odds of developing lung cancer or a 66(?)% decreased odds of getting the disease. SS SP ○ Sensitivity - a / (a + c) ○ Specificity - d / (b + d) PPV NPV ○ PPV - a / (a + b) ○ NPV - d / (d + c) Biases: Study biases ○ Selection bias - study participants different from source population Differential participation (when willingness to participate related to both exposure/disease status) Differential loss to follow up Bias if study participants exit a study for reasons related to both exposure/disease ○ Information bias - Bias if wrong info collected from/about study participants(ex - wrong classification) Recall bias - if people with disease remember/report their exposure differently (ex - more often/less often) than ppl w/o disease Interview bias - if there's systematic difference in soliciting, recording, or interpreting information ○ Healthy worker effect - better health in occupational cohort studies than average (bc workers more healthy) Screening ○ Lead-time bias: earlier detection appears to increase survival time, but death occurs at same time (**screening >effective than is) ○ Length bias: Screening detects slower-growing less aggressive disease more, (+overdiagnosis/unncessary treatment) Controls: 1. Nested controls from a cohort population/study (sub-set of full source pop from existing cohort pop.) 2. Population-based controls (controls from pop.) 3. Hospital or clinic-based controls (controls from hospital/clinic) Confounding v. effect modifier: Bias direction ○ Underestimation (towards null) Ex: Null (RR 1.0), Biased (RR 1.4), Truth (RR 1.9) ○ Overestimation (away from null) Ex: Null (RR 1.0), Truth (RR 2.0), Biased (RR 2.6) Stratified analysis ○ OR of 1: No effect, not confounder (?) ○ OR the same - confounder ○ OR different - effect modifier, not confounder Hypothesis testing/statistics 3 main steps ○ Specify null and alternative hyoptheses ○ Determine compatibility of study results with null hypothesis (statistical tests) ○ Decide whether to reject or not reject the null hypothesis Type I and II Errors ○ Type I error (alpha error) Incorrect rejection of a null hypothesis ○ Type II error (beta error) Fail to reject null hypotheses when it's false ○ Can use the 2x2 table above when unsure Confidence interval ○ Interpretation 1. Range of possible values within which the true magnitude of effect lies with a stated level of certainty Ex: "Assuming no bias/confounding, we have 95% confidence the true measure of association lies somewhere inside the interval from 1.2 to 1.7" 2. Range of hypotheses that are compatible with the observed data "Assuming no bias/confounding, the results from study A are consistent with the hypothesis that the strength of the association lies between 1.2 to 1.7" ○ Statistical significance If the 95% confidence interval does not include the null value (relative risk of 1.0): statistically significant results Interpreting statistical significance this way may result in misleading conclusions If a confidence interval includes 0 or 1 in the range, it's not significant Because at one point the numerator and denominator were equal (if RR is 1) Whether something is significant can be told by the p-value or the confidence interval. Causation: Characteristics ○ Positive (presence of causative exposure) or negative (lack of preventive exposure) Causative exposures - smoking, illicit drugs, air pollution Preventive exposure - taking vitamins, exercising, eating low fat diets Attributes ○ Association: causal factor must occur together with its effect ○ Time order: cause must precede effect (either proximate or distant in time) ○ Direction: asymmetrical relat. bw cause and effect Ex: Prenatal smoking causes low birth weight, but low birth weight Xcauses prenatal smoking Hill's guidelines ○ 1: Strength of association Stronger associations more indicative of exposure being causal Ex: Lung cancer and smoking ○ 2: Consistency If assoc. repeatedly observed in diff circumstances, then assoc. more likely to be causal ○ 3: Specificity Cause leads to single disease, not multiple diseases (also given effect has a single cause, not multiple causes) ○ 4: Temporality Exposure must precede occurrence of the disease ○ 5: Biological gradient If different doses of exposure are associated w/ diff associations with the outcome, then exposure is likely a cause If assoc. strength increases as exposure level increases, association more likely to be causal ○ 6/7: Plausability/coherence Biological or social models should exist to explain an association Association should not conflict with current knowledge of natural history and biology of disease Screening: Suitable test ○ Ideally, inexpensive, easy to administer, minimal discomfort, high level of validity/reliability ○ Valid test: Does what's intended to do (correctly classify ppl), Reliable test: Gives you same results on repetition