Week 1: What is your research question? Flashcards
How is a prediction model achieved?
Building a multivariable regression model (“prediction model”) that describes relationship between multiple predictors and outcome.
In causal (etiologic) models, you need to control for what and why?
Need to control for confounding variables distorting the effect of the exposure on the outcome.
What is a key difference when it comes to including variables in prediction vs. causal models?
Anything that can improve prediction of the outcome can be included in the model, regardless of causal relationship.
An odds ratio alone doesn’t provide adequate information to evaluate predictive ability
because of what two reasons?
Predictive value is influenced by how frequently a risk factor occurs in the population.
Many risk factors associated with increased risk of disease occur very commonly in the population. A risk factor may have an extremely high odds ratio, but be so rare in the population that it is not a useful tool to predict adverse outcomes at the population level.
We can evaluate a predictive model’s performance through which two concepts?
Risk stratification capacity: Ideal clinical prediction model would divide the population into ‘minimal risk’ or ‘high risk’ groups, allow surveillance and interventions to be appropriately focused.
Discrimination: how well can the model’s predictions separate those who have an outcome from those who don’t.
How do predictive and causal models differ with regards to developing statistical models?
Predictive = data-driven
Causal = expert causal knowledge
How do measures differ between predictive and causal models?
Predictive = measures of performance (sensitivity, specificity, area under the curve)
Causal = measures of association (e.g., OR, RR, RD, HR…)
How do you calculate sensitivity and specificity?
Sensitivity = true positives / total with disease x 100
Specificity: total negatives / total with disease x 100
How do you calculate positive predictive value and negative predictive value?
Positive predictive value = true positives / total testing positive x 100
Negative predictive value = true negatives / total testing negative x 100
Descriptive epidemiology seeks to characterize what?
Distributions of health, disease, and harmful or beneficial exposures in a well-defined population, including any meaningful differences in distribution, and whether that distribution is changing over time.
Descriptive/predictive/causal question: What proportion of people who attend the emergency department with a whiplash injury completely recover within 3 months?
Descriptive.
Descriptive/predictive/causal question: How well does a set of simple clinical measures predict the likelihood of recovery within 3 months?
Predictive.
Descriptive/predictive/causal question: Are people who receive education and reassurance more likely to recover in 3 months than people who receive a neck brace and advice to rest?
Causal.
