Causation and Validity Notes

NOTES NOTES CAUSATION & VALIDITY CAUSALITY osms.it/causality ▪ Consequential relationship between two events (e.g. A caused B) ▫ Contrast with correlation: association between two events ▪ Consequential relationship may be direct/ indirect ▫ Direct: event caused direct consequence which → effect (A → B) ▫ Indirect: initial event → another event → final effect (A → x → y → B) ▪ Correlation is not equal to causation ▫ Two correlated events may seem to have consequential relationship; sometimes due to random chance/ external factors/confounding (noncausal) variables ESTABLISHING CAUSALITY ▪ To establish causality between set of events, relationship must meet following criteria Temporality ▪ Cause happened before effect ▫ Event A followed by Event B ▫ Example: smoking → lung cancer 48 ▫ Example: the longer you smoke, the higher your risk of developing lung cancer Biologic coherence ▪ Causal mechanism for effect agrees with current knowledge ▫ Example: factually known that cigarettes contain carcinogenic agents Biologic plausibility ▪ Proposed mechanism of effect makes sense according to current knowledge ▫ Example: because we know cigarettes contain carcinogenic agents, it makes sense that cigarette-smoke exposure → higher probability of developing lung cancer Consistency with other knowledge ▪ Association has been shown repeatedly ▫ Example: it has been repeatedly proven that smoking confers higher risk of developing lung cancer Specificity ▪ Chances that effect is due to other causes ▫ Example: can there be another explanation for developing lung cancer besides exposure to cigarette smoke? Strength of association ▪ Relational closeness between two events ▫ Measured by relative risk, odds ratio, correlations, etc. ▫ Example: how closely is smoking related to developing lung cancer? Experimental evidence ▪ When you remove cause, effect disappears ▫ Example: if you stop smoking, your risk of developing lung cancer decreases Dose-response relationship ▪ More exposure to cause → greater effect ▫ Longer exposure to Event A → more risk of Event B Analogy ▪ Similar events have been proven to cause similar effects ▫ Example: smoking other substances has OSMOSIS.ORG

Chapter 6 Biostatistics & Epidemiology: Causation & Validity been known to cause lung pathology CAUSAL RELATIONSHIP TYPES Necessary and sufficient ▪ Presence of A required, present in adequate amounts to cause B ▫ Example: autosomal dominant mutation with complete penetrance both necessary, sufficient for disease to develop Necessary but not sufficient ▪ Presence of A required, not present in adequate amounts to cause B ▫ Example: heat required to cause burn, however, low heat will not cause burn; it is necessary but not sufficient Not necessary but sufficient ▪ Presence of A not required, but is enough to cause B ▫ Example: gunshot to head sufficient to cause death, however, not necessary, as there are many other causes of death Not necessary and not sufficient ▪ Presence of A not needed nor enough to cause B ▫ Example: urinary infection not necessary nor sufficient to cause pelvic inflammatory disease; urinary infection can be present without pelvic inflammatory disease, individual can have pelvic inflammatory disease without having urinary tract infection BIAS osms.it/bias ▪ Error in one step of study design/ conduction/analysis → results interpretation that is different from truth ▫ Many types of biases, no common classification SELECTION BIAS ▪ Errors made when choosing/following population to be studied ▫ Can occur at different stages of study ▫ Most commonly occurs when chosen sample is not representative of population MEASUREMENT BIAS ▪ AKA information bias ▪ Errors made when measuring data/results of interest ▫ Most commonly results in results misclassification which can be differential/non-differential Differential misclassification ▪ Error in measurement more likely to occur in one group than another ▫ Results of one group will be inherently different to other group’s results ▫ Example: blood glucose levels of groups measured by different machines; one gave accurate results, other reported inaccurate results Non-differential misclassification ▪ Measurement error likely to have occurred in both groups ▫ Results among two groups will not differ greatly ▫ Example: machine used to determine blood glucose levels for both groups was inaccurate OTHER BIAS TYPES ▪ Information gathering, management can → other bias types ▪ AKA information bias Procedure bias ▪ People allocated to different groups not treated identically ▫ Usually due to lack of blinding OSMOSIS.ORG 49

▫ Example: people in one group spend more time in hospital than other group Recall bias ▪ Awareness of event/effect influences individual’s recall of cause ▫ Most common in retrospective studies ▫ Example: after a person with cancer knows that radiation exposure is a cancer development risk factor, the person may place more emphasis on exposure to radiation than someone without cancer Lead-time bias ▪ Early diagnosis extends follow-up period, making it seem as if event being studied took longer to progress ▫ Example: early cervical cancer detection may make it seem as if cancer is less aggressive because of more time spent living with diagnosis Observer-expectancy bias ▪ When belief in intervention’s effectiveness interferes with reported treatment outcome ▫ Example: researcher’s belief in drug efficacy may interfere with reported results CONFOUNDING osms.it/confounding ▪ Occurs when external event is related to possible cause, outcome of interest but is not on causal pathway ▪ Example: study exploring relationship between exercising, overall health, we know that ▫ Exercising known to improve overall health ▫ Exercising associated with healthy lifestyle, but is not result of healthy lifestyle INTERACTION osms.it/interaction ▪ Combination of two/more factors changes disease incidence compared to influence they would have had individually ▫ Describes way multiple factors interact to produce event Synergism ▪ Refers to potentiation effect multiple factors may have on one another ▪ Example: 2 + 2 = 5 Antagonism ▪ Refers to inhibition effect multiple factors may have on one another ▪ Example: 2 + 2 = 3 50 OSMOSIS.ORG Figure 6.1 Biological interaction is when two exposures, like radon gas and cigarette toxins, work together to influence an outcome, like lung cancer.

Chapter 6 Biostatistics & Epidemiology: Causation & Validity Figure 6.2 A graph representing data collected from four groups with 100 people per group: those with no exposure to radon or cigarette toxins (A), those with exposure to only cigarette toxins (B), those with exposure to only radon (C), and those with exposure to both radon and cigarette toxins (D). The multiplicative scale was used to calculate the expected joint effect of radon and cigarette toxins based on their independent effects (columns B and C). These two exposures are said to have a synergistic interaction because observed relative risk > expected joint effect. If observed relative risk had been < expected joint effect, the interaction would have been antagonistic. OSMOSIS.ORG 51