Behavioral Science Flashcards
Retrospective case-control vs retrospective cohort
Case control compares a group with a disease and without a disease to look at exposures. End up with odds ratio (odds of getting disease with exposure vs without). Cohort study compares a group with a given exposure vs without, and sees how many developed disease. Looks at relative risk (risk of developing disease associated with exposure)
Relative risk vs odds ratio
Relative risk is incidence exposed/incidence unexposed. Used is cohort study because have incidence data (new people who get disease in time period). Odds ratio, don’t have incidence because not looking over period of time, just whether they are a case or not. So have to do odds that group with disease was exposed to a risk factor/ odds that group without disease was exposed to risk factor. (Cases with exposure/ cases without exposure)/(noncases with exposure/ noncases without exposure). If prevalence is low, the two are about equal.
Odds ratio for drug study
(Cases with treatment/ cases without treatment)/ (non-cases cases with treatment / noncases without treatment). Think of treatment as exposure.
Phase I study: sample and purpose
Sample: small number of HEALTHY volunteer. Purpose: safety, toxicity, pharmacokinetics. Seeing how drug works on healthy people
Phase II study: sample and purpose
Sample: Small number of PATIENTS with disease of interest. Assess: treatment efficacy, optimal dosing, and AE’s. Does the drug do what we want?
Phase III study: sample and purpose
Sample: Large number of patients randomly assigned to treatment or placebo (standard of care). Compares new treatment to current standard of care.
Phase IV: study sample and purpose
Postmarketing surveillance trial of patients after approval to detect rare or long-term AE’s.
Crossover study
Have placebo group and treatment group and then a washout period, then they switch. Allows people to serve as own controls.
Sensitivity
TP/(TP+FN). 1- FN rate. Out of the people who have the disease, how many tested +? Can you pick it up? How good at you at finding positives? If you are bad, then a positive is more likely to be a true positive, and you’ll probs get a lot of false negatives.
Specificity
TN/(TN + FP). 1 - FP rate. Out of the people who don’t have the disease, how many tested -? If you pick it up, is it specific, or could it be something else? How good are you at finding negatives? If you are bad, then a negative is more likely to be a true negative, and you’ll probs get a lot of false positives.
PPV
TP/(TP + FP). Given that the test is positive, how likely is it that the person has the disease? Varies with prevalence.
NPV
TN/(TN + FN). Given that the test is negative, how likely is it that the person doesn’t have the disease? Varies with prevalence.
Relative risk
Risk of developing disease in exposed group/ risk of in unexposed group. (Exposed with disease/ total exposed)/(unexposed with disease/total unexposed). Compare with odds ratio: (exposed with/exposed w/o)/unexposed with/unexposed w/o). For RR we have prevalence, and thus can divide by the total. Use for cohort studies.
Attributable risk
Difference in risk between exposed and unexposed. (Exposed with disease/total exposed) - (Unexposed with disease)/ total unexposed. Like RR except minussing instead of dividing.
You are given the RR and need to calculate the AR…
RR-1/RR. The math works out, trust me. You can do it out too if you want.
Absolute risk reduction
Now we are comparing treatment with control. Risk with treatment - risk without treatment. Absolute because not worried about what is attributable to the exposure, just looking at the change in risk overall.
Number needed to treat vs number needed to harm
Number needed to treat is number of patients who need to be treated for 1 to benefit. Number needed to harm is number of patients needed to be exposed for 1 to be harmed. 1/attributable risk.
Effect modification vs confounding
Effect modification is external variable impacts effect of risk factor on disease status. Ex: risk of smoking increases the risk of disease, but only for women. Vs. confounding, where being a woman increases both the risk for smoking AND risk of disease, so can’t tell if it is the smoking or being a woman causing the effect.
Precision vs accuracy
Precision = reliability, all trials have nearly same outcome. Accuracy = validity, how true the test is (bias?)
Random error vs systematic error in changing precision and accuracy
Random error will change precision (will be different every time). Systematic error will change accuracy (won’t be as true)
Berkson’s bias
A selection bias from selecting hospitalized patients as a control. Think of Berks –> hospital
Late-look bias
Using a survey to study a fatal disease (only patients still alive will be able to answer). Gathering of info at inappropriate time.
Procedure bias
Subjects in different groups not treated the same. For example, more attention is paid to treatment group, stimulating greater adherence.
