Quiz · Hard
Applied questions: NNT, odds ratios, PPV, herd immunity, attack rates and recognising bias.
A trial finds a drug lowers 1-year stroke risk from 8% to 6%. What is the NNT?
50
20
25
2
ARR = 0.08 - 0.06 = 0.02, so NNT = 1/0.02 = 50. 25 would come from a 4% reduction; 2 is the percentage point change mistaken for NNT.
In a case-control study: a = 40 exposed cases, b = 60 exposed controls, c = 20 unexposed cases, d = 80 unexposed controls. What is the odds ratio?
2.00
1.33
2.67
4.00
OR = (40 x 80)/(60 x 20) = 3200/1200 = 2.67. 2.00 is a/c, which ignores the controls.
A screening test (sensitivity 95%, specificity 90%) is used in 1,000 people, 5% of whom have the disease. About how many positive results will be true positives?
About 95 of about 190 positives
About 48 of about 143 positives
About 50 of about 95 positives
About 48 of about 95 positives
Diseased = 50 -> TP = 47.5. Healthy = 950 -> FP = 95. Positives = 142.5, so PPV is about 33%. Many students forget to count false positives from the large healthy group.
Using the same test and population (sens 95%, spec 90%, prevalence 5%), what is the approximate PPV?
95%
90%
50%
33%
TP 47.5 / (47.5 + 95) = 0.33. PPV is far lower than sensitivity because of the low prevalence.
In a cohort, 30 of 200 exposed and 10 of 200 unexposed develop disease. What is the attributable risk percent among the exposed?
66.7%
33.3%
200%
10%
Ie = 0.15, Iu = 0.05; (0.15-0.05)/0.15 = 66.7%. 200% is the RR-1 mistaken for the percent among exposed (RR = 3).
The measles R0 is about 15. Which estimate of herd immunity threshold is closest?
15%
93%
67%
85%
1 - 1/15 = 0.933. Many choose 85% from lower R0 values, but R0 = 15 requires more than 90% immunity.
After a church supper, 40 of 100 people who ate the potato salad became ill and 5 of 50 who did not eat it did. What is the relative risk for potato salad?
8.0
0.25
4.0
1.25
Attack rate among eaters = 40/100 = 0.40; among non-eaters = 5/50 = 0.10; RR = 4.0. Using counts instead of rates (40/5 = 8) is the common error.
A study finds screening increased 5-year survival from 60% to 85%, yet death rates from the cancer did not change. The most likely explanation is:
Recall bias
Lead-time bias
Confounding by age
Berkson bias
Earlier detection lengthens survival from diagnosis without delaying death. Comparing disease-specific mortality is more reliable.
A researcher reports that countries with higher chocolate consumption have more Nobel laureates per capita and concludes that eating chocolate makes individuals smarter. What is the main error?
Ecological fallacy
Selection bias
Recall bias
Type II error
Group-level associations in ecological data cannot be assumed to hold for individuals; confounders such as national wealth are also unmeasured.
Which change would increase the statistical power of a trial?
Lowering the significance level from 0.05 to 0.001
Reducing the expected effect size
Increasing loss to follow-up
Increasing the sample size
More participants reduce random error. Lowering alpha, a smaller effect and more attrition all reduce power.
In a cohort study of asbestos workers, mortality is lower than in the general population. Which explanation best fits?
Length-time bias
Recall bias
Healthy worker effect
Misclassification of outcome
People able to work are healthier than the general population (which includes those too sick to work), biasing comparisons toward the null.
An investigator wants to know the true proportion of a town's adults with diabetes today. Which design fits best?
Case-control study
Cross-sectional survey
Randomized trial
Retrospective cohort
A cross-sectional survey measures prevalence at one point in time. The others do not estimate prevalence directly.
The 95% CI for an odds ratio is 0.8 to 2.4. Which interpretation is correct?
The association is not statistically significant at the 5% level
The exposure is protective
The exposure is harmful
The sample is too large
The interval includes 1 (no association). It's compatible with both slight protection and harm.
Non-differential misclassification of exposure in a case-control study will usually:
Bias the odds ratio away from 1
Have no effect on the odds ratio
Bias the odds ratio toward 1
Always produce a false positive finding
Errors that occur equally in cases and controls blur the difference between groups, moving the OR toward the null. Differential misclassification can bias in either direction.
Compare these two populations: Town A has crude mortality 12/1000 and Town B 8/1000, but Town A has many more elderly residents. After direct age standardization the rates are 7/1000 (A) and 8/1000 (B). What can you conclude?
Town A truly has higher mortality
Town B has the older population
Standardization introduced bias
The crude difference was due to age structure, not higher risk in Town A
After accounting for age, A is not worse than B. The crude gap was confounded by age.