L4: Causality and statistical interpretation Flashcards
Criteria for causation?
Criteria for causation: Building a circumstantial case for likely causes of disease, injury, or early death
When case is strong enough 🡪 RCT for more confirming evidence
However, not all questions can be answered with an RCT because impractical, costly, or unethical
How to build your case: Find statistical association
Results of confounding and effect modification account for
Judge strength of scientific evidence
🡪 Austin Bradford Hill criteria
bradford hill’s criteria for assessing causality?
Strength of association
Consistency
Specificity
Temporality
Biological gradient
Plausibility
Coherence
Experiment
Analogy
explaining bradford hills criteria?
strength of association: The strength of association refers to how strong that relationship is between the exposure and the disease.
Coherence: a casual interpretation should not conflict with knowledge about the natural history of the disease and its known pathophysiology
Experiment: manipulation of the presumed cause should result in a lower rate of disease
Plausability: findings are explainable by existing biological knowledge and can be altered
Biological gradient: a dose-response gradient. More exposure should be associated with a higher rate of disease. A linear gradient is the best evidence, but other patterns of association may be causal.
Temporality: the effect must occur after the exposure, especially for disease that develops slowly.
Specificty: association is limited to one disease and a single group of locationw ith no other likely explanation. However, disease can have more than one case.
Consistency: association is repeatedly observed by different persons and in different places, circumstances and times.
analogy: Analogy refers to comparing the exposure-disease relationship you’re studying with other similar relationships that are already established. If a similar exposure has been shown to cause the same disease, it may strengthen the case for causality in the new exposure.
For example, direct smoking is strongly associated with lung cancer. If there is weaker evidence that second-hand smoke also causes lung cancer, the analogy to direct smoking might be used to support the idea that second-hand smoke could also cause the disease.
interpreting conclusion paragraphs of studies?
If interpreting conclusion paragraph of a study: give statement on study design, causality. Perhaps if observational study mentions other studies then you can say causality. Bring in redford hill critera.
causality and causation?
Causality (in the statistical sense) means that when we intervene, the chances of different outcomes are systematically changed.
Causation is difficult to establish statistically, but well-designed randomised trials are the best available framework.
Principles of blinding, intention to treat and so on have enabled large-scale clinical trials to identify moderate but important effects.
Observational data may have background factors influencing the apparent observed relationship between an exposure and an outcome, which may be either observed confounders or lurking factors.
Statistical methods exist for adjusting for other factors, but judgement is always required as to the confidence with which causation can be claimed.
Randomized Controlled Trial (RCT) and Causality:
An RCT is considered the gold standard for claiming causality because:
The randomization process helps ensure that groups are comparable at the start of the study.
The only difference between the groups is the exposure (treatment or intervention), making it easier to attribute differences in outcomes to the exposure itself.
reducing information bias?
In a double-blind study, both the participants and the researchers (or those administering the treatment) do not know which treatment the participants are receiving.
This helps prevent information bias, which could occur if knowing the exposure (treatment or intervention) affects the way researchers or participants interpret outcomes.
For example, if a doctor knows a patient is receiving a specific treatment, they might unconsciously (or consciously) interpret or record the outcome differently, leading to biased results.
open label bias?
Open-Label Bias:
In an open-label trial, both the doctor and the patient know which treatment is being given.
This can lead to bias because the doctor’s knowledge of the treatment may influence their behavior, their interpretation of the outcomes, or the way they interact with the patient.
In a surgical trial comparing two procedures, for example, the doctor knows which procedure the patient is receiving, and this knowledge might affect how the doctor measures the patient’s recovery, or how they treat the patient post-surgery. Legally, the doctor might have to inform the patient about what surgery they’re going to undergo, but ideally, this shouldn’t affect the study outcome.
Ppdac?
PPDAC:
This stands for Problem, Plan, Data, Analysis, Conclusion, which is a systematic approach to designing and conducting research.
The PPDAC framework is a method to plan and carry out research while ensuring internal validity. The focus during the “Plan” phase is on identifying potential confounders (factors that could distort the true relationship between exposure and outcome) and effect modifiers (factors that may change the strength or direction of the relationship between exposure and outcome).
The “Define study cohort” and “Avoid selection bias” steps emphasize creating a representative study population and minimizing biases that could harm the internal validity of your study.
types of variables?
Variables: Variable: any measurement that can take on a different value in different circumstances
Continuous variables: Measurements that can be made, at least in principle, to arbitrary precision (e.g., height and weight)
Binary variables: Yes/No questions, Male/Female, Alive/Dead, etc
Categorical variables: measures that can take on two or more categories which may be:
Unordered categories: a person’s country of origin, the colour of a car
Ordered categories: such as the rank of military personnel
Numbers that have been grouped: such as levels of obesity
Height- continuous but to make categorical: short, medium, tall however this is also ordinal, to make binary- 2 categories e.g: everyone blow 1.40m everyone above
BMI- continuous variable
descriptive statistics?
Descriptive statistics: summarising continuous variables
There are three basic interpretation of the term ‘average’:
Mean: sum of the numbers divided by the number of cases
Median: the middle value when the numbers are put in order
Mode: the most common value
Slide 34: prospective cohort study- following people overtime
Descriptive statistics
Need to understand pop and for analysis. Gives you idea for the distribution of the dataset. A watyof picking up any potential errors in the data set.
