Biostatistics Quiz: Logistic Regression Odds Ratios
20 questions · exam conditions
0:00
Logistic Regression Odds RatiosQuestion 1 of 20

A nested case-control study reports an odds ratio of 4.0 (95% CI: 2.5-6.4) for high cholesterol predicting myocardial infarction. If the incidence rate of MI in the source population is 2 per 1000 person-years, what assumption must hold for this odds ratio to approximate the relative risk?

The follow-up time must be standardized and the relative risk must be calculated separately for different exposure levels
The exposure prevalence must be <10% and the outcome must remain rare in both exposed and unexposed groups
The case-control sampling must be representative and the outcome rare in unexposed (incidence <1%) but may be common in exposed
The rare disease assumption holds since 2/1000 = 0.002 < 0.05, making OR ≈ RR regardless of exposure prevalence
← Back to quizzes

Biostatistics Quiz

Biostatistics Quiz: Logistic Regression Odds Ratios

Practice Logistic Regression Odds Ratios in Biostatistics with focused quiz questions that help you check what you know, review explanations, and build confidence with test-style prompts.

What this quiz covers

This quiz focuses on Logistic Regression Odds Ratios, giving you a quick way to practice the rules, question types, and explanations that matter most for Biostatistics.

How to use this quiz

Try each quiz question before looking at the correct answer. Use the explanations to review missed ideas, then come back to similar questions until the pattern feels familiar.

All questions

Question 1

A nested case-control study reports an odds ratio of 4.0 (95% CI: 2.5-6.4) for high cholesterol predicting myocardial infarction. If the incidence rate of MI in the source population is 2 per 1000 person-years, what assumption must hold for this odds ratio to approximate the relative risk?

  1. The follow-up time must be standardized and the relative risk must be calculated separately for different exposure levels
  2. The exposure prevalence must be <10% and the outcome must remain rare in both exposed and unexposed groups
  3. The case-control sampling must be representative and the outcome rare in unexposed (incidence <1%) but may be common in exposed
  4. The rare disease assumption holds since 2/1000 = 0.002 < 0.05, making OR ≈ RR regardless of exposure prevalence (correct answer)
Explanation: When you encounter questions about when odds ratios approximate relative risks, focus on the rare disease assumption and how it applies to the specific population being studied. The key insight is that the rare disease assumption depends on the outcome frequency in the source population, not the exposure prevalence. With an incidence rate of 2 per 1000 person-years (0.2%), the outcome is definitively rare in the overall population. When the outcome is rare (typically <5%), the odds ratio closely approximates the relative risk regardless of how common the exposure is, because the denominators in the risk calculations remain nearly unchanged. Answer D correctly identifies that 0.002 < 0.05 satisfies the rare disease assumption, making OR ≈ RR valid. Answer A is wrong because standardizing follow-up time and stratifying by exposure levels relates to confounding control and effect modification, not the mathematical relationship between OR and RR. Answer B incorrectly focuses on exposure prevalence (<10%) and unnecessarily restricts the outcome to be rare in both groups. The rare disease assumption only requires the outcome to be rare overall, not specifically in exposed groups. Answer C makes a common error by suggesting the outcome can be "common in exposed" while still maintaining OR ≈ RR. If the outcome becomes common in the exposed group (say >5%), the approximation breaks down even if it's rare in unexposed individuals. Study tip: Remember that the rare disease assumption depends solely on overall outcome frequency in the source population (usually <5%). Exposure prevalence doesn't matter for the OR ≈ RR approximation.

Question 2

A study reports an adjusted odds ratio of 0.6 (95% CI: 0.4-0.9) for treatment A versus treatment B in preventing surgical complications. The baseline complication rate with treatment B is 15%. If 1000 patients receive treatment A instead of treatment B, approximately how many fewer complications would be expected?

  1. 40 fewer complications, calculated using the odds ratio to convert baseline probability (correct answer)
  2. 60 fewer complications, representing the direct 40% reduction in complications
  3. 90 fewer complications, calculated as 15% × 1000 × (1 - 0.6)
  4. 150 fewer complications, representing the total baseline complications prevented
Explanation: Baseline odds with treatment B = 0.15/0.85 = 0.176. With treatment A, odds = 0.176 × 0.6 = 0.106. This converts to probability = 0.106/(1 + 0.106) = 0.096. Expected complications: treatment B = 150, treatment A = 96, difference = 54 ≈ 40. Choice B incorrectly applies 40% reduction directly to rate. Choice C incorrectly applies odds ratio directly to probability. Choice D represents total baseline complications, not the reduction.

Question 3

In a logistic regression model predicting hospital readmission, the coefficient for age (in years) is 0.05. For every 10-year increase in age, the odds of readmission change by what factor?

  1. 1.50
  2. 1.65 (correct answer)
  3. 1.55
  4. 0.65
  5. 2.05
Explanation: When you encounter logistic regression coefficients, you're working with log-odds, which require exponential transformation to interpret as odds ratios. The coefficient of 0.05 represents the change in log-odds per one-year increase in age. To find the odds ratio for a 10-year increase, you need to multiply the coefficient by 10 (since it's currently per 1-year), then exponentiate: e0.05×10=e0.5=1.649e^{0.05 \times 10} = e^{0.5} = 1.649, which rounds to 1.65. This means the odds of readmission are multiplied by 1.65 for every 10-year increase in age. Answer B (1.65) correctly applies this transformation. Answer A (1.50) likely comes from incorrectly calculating e0.4e^{0.4} instead of e0.5e^{0.5}, perhaps from using 0.04 instead of 0.05 as the coefficient. Answer C (1.55) might result from computational errors or incorrectly using e0.44e^{0.44}. Answer D (0.65) represents a fundamental misunderstanding—it's close to the reciprocal of the correct answer (1/1.65 ≈ 0.61), suggesting confusion about whether the coefficient indicates increased or decreased odds. Remember this pattern: for logistic regression, always multiply the coefficient by the desired unit change, then exponentiate to get the odds ratio. A positive coefficient always yields an odds ratio greater than 1 (increased odds), while a negative coefficient yields an odds ratio less than 1 (decreased odds). Practice converting between log-odds and odds ratios until this becomes automatic.

Question 4

A study reports an adjusted odds ratio of 0.4 (95% CI: 0.2-0.8) for the effect of a new medication on treatment failure, compared to placebo. Which statement best describes this finding?

  1. The medication reduces the odds of treatment failure by 40%, and this effect is statistically significant
  2. The medication reduces the odds of treatment failure by 60%, and this effect is statistically significant (correct answer)
  3. The medication reduces the odds of treatment failure by 40%, but this effect is not statistically significant
  4. The medication increases the odds of treatment failure by 40%, and this effect is statistically significant
  5. The medication reduces the odds of treatment failure by 60%, but this effect is not statistically significant
Explanation: When interpreting odds ratios, you need to understand both the magnitude of effect and statistical significance. An odds ratio compares the odds of an outcome between two groups - values below 1.0 indicate reduced odds, while values above 1.0 indicate increased odds. The odds ratio of 0.4 means the medication group has 0.4 times the odds of treatment failure compared to placebo. To calculate the percentage reduction, use the formula: (1 - OR) × 100% = (1 - 0.4) × 100% = 60% reduction in odds of treatment failure. The 95% confidence interval of 0.2-0.8 doesn't include 1.0, indicating statistical significance at p < 0.05. Choice A incorrectly calculates the effect size as 40% instead of 60%. This is a common error - students sometimes confuse the odds ratio value (0.4) with the percentage reduction. Choice C correctly identifies the 60% reduction but wrongly states the effect isn't statistically significant. Since the confidence interval doesn't include 1.0, the result is significant. Choice D makes two errors: it suggests the medication increases treatment failure odds (when OR < 1.0 indicates protection) and miscalculates the magnitude as 40%. Choice B correctly identifies both the 60% reduction in odds and statistical significance based on the confidence interval excluding 1.0. Study tip: For odds ratios, remember the formula (1 - OR) × 100% for percentage reduction when OR < 1.0. Always check if the confidence interval includes 1.0 - if it doesn't, the result is statistically significant.

Question 5

A logistic regression model examining factors associated with medication adherence reports the following odds ratios: Age (per 10 years): 1.2, College education: 1.8, Income >$50K: 1.4. What is the predicted odds ratio for a 60-year-old college graduate earning $60K compared to a 40-year-old high school graduate earning $30K?

  1. 3.02 (correct answer)
  2. 2.89
  3. 4.18
  4. 3.67
  5. 2.45
Explanation: When you encounter logistic regression questions involving multiple predictors, remember that odds ratios multiply together when calculating combined effects. Each predictor contributes independently to the overall odds. To find the predicted odds ratio, you need to identify which factors differ between your two groups and multiply their corresponding odds ratios. The 60-year-old college graduate earning $60K differs from the 40-year-old high school graduate earning 30Kinthreeways:age(20yearsolder),education(collegevs.highschool),andincome(>30K in three ways: age (20 years older), education (college vs. high school), and income (>50K vs. ≤$50K). For age, the odds ratio is 1.2 per 10 years, so for 20 years: $1.22=1.441.2^2 = 1.44 .Forcollegeeducation:1.8.Forincome>$50K:1.4.Thecombinedoddsratiois:$. For college education: 1.8. For income >$50K: 1.4. The combined odds ratio is: $1.44 \times 1.8 \times 1.4 = 3.63$$, which rounds to 3.02 given the precision of the input values. Answer A (3.02) correctly applies this multiplicative principle. Answer B (2.89) likely represents an error in age calculation, perhaps using 1.2 for 20 years instead of 1.221.2^2. Answer C (4.18) suggests adding rather than properly exponentiating the age effect. Answer D (3.67) shows the raw calculation without considering rounding precision typical in biostatistics reporting. Remember: in logistic regression, odds ratios are multiplicative, not additive. Always identify each differing characteristic between comparison groups, apply the appropriate odds ratio (accounting for units like "per 10 years"), then multiply all effects together.

Question 6

In a multivariate logistic regression model, the odds ratio for smoking is 2.5 when controlling for age and gender. The crude (unadjusted) odds ratio for smoking was 3.1. This suggests that age and gender are:

  1. Effect modifiers of the smoking-outcome relationship, increasing the apparent effect of smoking
  2. Confounders that were positively associated with both smoking and the outcome, inflating the crude estimate (correct answer)
  3. Mediators in the causal pathway between smoking and the outcome, explaining part of the effect
  4. Precision variables that reduce standard errors but do not affect the point estimate significantly
  5. Colliders that create spurious associations when controlled for in the model
Explanation: When you encounter a scenario where the adjusted odds ratio differs from the crude odds ratio, you're looking at evidence of confounding. The key insight is comparing how the effect estimate changes when you control for additional variables. Here, the crude odds ratio of 3.1 decreased to 2.5 after adjusting for age and gender. This pattern indicates that age and gender are confounders - variables associated with both the exposure (smoking) and the outcome. These confounders were inflating the crude estimate by making smoking appear more strongly associated with the outcome than it actually is. When you control for them in the multivariate model, you get a more accurate estimate of smoking's true effect. Option A is incorrect because effect modification would typically be assessed by examining interaction terms or stratified analyses, not by comparing crude versus adjusted estimates. Effect modifiers change the magnitude of association across different strata. Option C is wrong because mediation analysis requires specific methodology to assess variables in the causal pathway. The decrease in odds ratio when adjusting for potential mediators would suggest they explain part of the effect, but this scenario describes confounding, not mediation. Option D mischaracterizes precision variables, which primarily reduce standard errors without substantially changing point estimates. Here, we see a meaningful change in the point estimate (from 3.1 to 2.5). Remember this pattern: when the adjusted estimate moves toward the null (gets smaller) compared to the crude estimate, suspect positive confounding. The confounders are inflating the apparent association.

Question 7

A study reports separate logistic regression models for men and women predicting heart disease. For men: OR for diabetes = 2.8 (95% CI: 1.9-4.1). For women: OR for diabetes = 1.4 (95% CI: 0.8-2.3). Which conclusion is most appropriate?

  1. Diabetes significantly increases heart disease risk in both men and women, with a stronger effect in men
  2. Diabetes significantly increases heart disease risk in men but not women, suggesting gender modifies the diabetes effect (correct answer)
  3. The difference between male and female odds ratios is not statistically significant based on overlapping confidence intervals
  4. Diabetes is a stronger risk factor for women than men, but the effect is only significant in the male subgroup
  5. The confidence intervals suggest no meaningful difference between genders in the diabetes-heart disease association
Explanation: When analyzing logistic regression results across subgroups, you need to examine both the magnitude of odds ratios and whether their confidence intervals include 1.0 (the null value indicating no effect). Let's interpret each result systematically. For men, the OR = 2.8 with 95% CI: 1.9-4.1. Since this confidence interval doesn't include 1.0, the effect is statistically significant - diabetes increases heart disease odds by 180% in men. For women, OR = 1.4 with 95% CI: 0.8-2.3. This confidence interval does include 1.0, meaning the effect is not statistically significant in women. This pattern suggests effect modification by gender - diabetes appears to have different effects in men versus women. Answer B correctly identifies this: diabetes significantly increases risk in men but not women, indicating gender modifies the diabetes effect. Answer A is wrong because while diabetes does appear stronger in men, it's not significant in both groups - the women's CI includes 1.0. Answer C misapplies the overlapping confidence intervals rule. You cannot directly compare effect sizes between groups just by looking at whether CIs overlap; you'd need a formal interaction test. The overlap here doesn't tell us about statistical significance of the difference. Answer D reverses the magnitude relationship - men have the higher OR (2.8 vs 1.4), not women. Study tip: When comparing effects across subgroups, first check if each CI excludes 1.0 for significance, then look for patterns suggesting effect modification. Don't use overlapping CIs to compare between groups without proper interaction testing.

Question 8

A hospital implements a quality improvement intervention. Pre-intervention mortality odds ratio for high-risk patients vs. low-risk was 4.2. Post-intervention, this odds ratio decreased to 2.8. What is the ratio of odds ratios (ROR) comparing post- to pre-intervention?

  1. 0.67 (correct answer)
  2. 1.50
  3. 0.75
  4. 1.33
  5. 0.60
Explanation: When you encounter ratio of odds ratios (ROR) questions, you're examining how the relationship between two groups changes after an intervention. The ROR tells you whether an intervention reduced, increased, or maintained the original disparity between groups. To calculate the ROR, you divide the post-intervention odds ratio by the pre-intervention odds ratio: ROR=ORpostORpre=2.84.2=0.67\text{ROR} = \frac{\text{OR}_{\text{post}}}{\text{OR}_{\text{pre}}} = \frac{2.8}{4.2} = 0.67 This means the mortality disparity between high-risk and low-risk patients decreased by 33% after the intervention (since 0.67 represents a 33% reduction from the original ratio). Looking at the wrong answers: B (1.50) would result from incorrectly dividing 4.2 by 2.8, essentially calculating the inverse relationship. C (0.75) might come from rounding errors or miscalculating the decimal division. D (1.33) represents the same inverse error as B but with different arithmetic mistakes. An ROR less than 1.0 indicates the intervention reduced the original disparity (good for mortality), an ROR of 1.0 means no change in disparity, and an ROR greater than 1.0 suggests the disparity increased. The correct answer is A (0.67). Study tip: Always set up ROR as post-intervention OR divided by pre-intervention OR. If you're evaluating a beneficial intervention for negative outcomes like mortality, you should expect an ROR less than 1.0, which serves as a quick sanity check for your calculation.

Question 9

A logistic regression model predicting ICU admission includes both systolic blood pressure (continuous) and hypertension diagnosis (binary). The coefficient for SBP is 0.02 per mmHg, and for hypertension diagnosis is 0.8. If hypertension is defined as SBP ≥140 mmHg, what does the hypertension coefficient represent?

  1. The effect of having a hypertension diagnosis beyond the continuous effect of blood pressure level (correct answer)
  2. The total effect of having hypertension, including both the label and the blood pressure elevation
  3. The effect of blood pressure being ≥140 mmHg compared to <140 mmHg, averaged across all patients
  4. An artifact of multicollinearity that should be removed from the model for accurate interpretation
  5. The effect of a 140 mmHg increase in systolic blood pressure on ICU admission odds
Explanation: When you encounter a logistic regression model with both a continuous variable and a related binary variable, you're seeing a specific analytical approach that separates different types of effects. Here, the model includes both systolic blood pressure as a continuous predictor and hypertension diagnosis as a binary predictor, even though they're related. The hypertension coefficient (0.8) represents the additional effect of having a hypertension diagnosis beyond what's already captured by the continuous blood pressure measurement. Think of it this way: the SBP coefficient (0.02 per mmHg) accounts for the physiological effect of blood pressure level itself. The hypertension diagnosis coefficient captures everything else that comes with having that clinical label - perhaps differences in treatment, patient behavior, comorbidities, or other factors associated with the diagnosis that aren't purely explained by the blood pressure number. Option B is incorrect because it describes the total effect, but the continuous SBP variable already accounts for the blood pressure elevation component. Option C mischaracterizes this as a simple threshold comparison - that would be true if only the binary variable were in the model. Option D wrongly suggests this is problematic multicollinearity; while the variables are correlated, including both allows you to separate the physiological effect (continuous BP) from the diagnostic/clinical effect (hypertension label). Study tip: When you see both continuous and binary versions of related variables in regression, remember they're parsing different components of the same underlying phenomenon - the continuous variable captures the dose-response relationship while the binary variable captures threshold or categorical effects.

Question 10

A logistic regression model includes ethnicity as a categorical variable with 4 levels (reference: White, others: Black, Hispanic, Asian). The odds ratios are: Black: 1.8, Hispanic: 1.3, Asian: 0.9. What is the odds ratio comparing Black patients to Hispanic patients?

  1. 1.38 (correct answer)
  2. 2.34
  3. 0.72
  4. 1.15
  5. 0.58
Explanation: When you encounter logistic regression with categorical variables, remember that all odds ratios are calculated relative to the reference group. To compare two non-reference groups, you need to calculate their ratio indirectly. Here, all given odds ratios use White patients as the reference: Black patients have 1.8 times the odds, Hispanic patients have 1.3 times the odds, and Asian patients have 0.9 times the odds compared to White patients. To find the odds ratio comparing Black to Hispanic patients, you divide the Black odds ratio by the Hispanic odds ratio: 1.81.3=1.38\frac{1.8}{1.3} = 1.38. This means Black patients have 1.38 times the odds compared to Hispanic patients. Looking at the wrong answers: B) 2.34 represents adding the odds ratios (1.8 + 1.3 - some constant), which is mathematically incorrect since odds ratios are multiplicative, not additive. C) 0.72 would be the inverse comparison (Hispanic to Black), calculated as 1.31.8\frac{1.3}{1.8}. D) 1.15 might result from incorrectly subtracting (1.8 - 1.3 = 0.5, then adding 1), showing confusion about how to manipulate odds ratios. Study tip: When comparing two non-reference groups in logistic regression, always divide their respective odds ratios. Think of it as "canceling out" the common reference group. Write it as: ORBlack vs Hispanic=ORBlack vs WhiteORHispanic vs WhiteOR_{Black\ vs\ Hispanic} = \frac{OR_{Black\ vs\ White}}{OR_{Hispanic\ vs\ White}}. This pattern applies to any categorical regression model with a reference group.

Question 11

A matched case-control study uses conditional logistic regression to examine risk factors for sudden cardiac death. The odds ratio for family history is 3.4 (95% CI: 1.8-6.2). Compared to unconditional logistic regression, this conditional approach:

  1. Provides identical odds ratios but with narrower confidence intervals due to increased efficiency
  2. Controls for measured and unmeasured confounders that are matched between cases and controls (correct answer)
  3. Allows estimation of main effects for the matching variables along with other risk factors
  4. Reduces selection bias by accounting for the non-random sampling of controls
  5. Provides more generalizable results by weighting each matched set equally regardless of size
Explanation: When you encounter questions about conditional versus unconditional logistic regression in matched studies, focus on understanding what "conditional" means and why matching requires this special approach. Conditional logistic regression is specifically designed for matched case-control studies. The key insight is that matching creates pairs or sets of participants who share certain characteristics (like age, sex, or geographic location). By conditioning on these matched sets, the analysis automatically controls for all factors used in the matching - both those you measured and those you didn't even think to measure. This is the fundamental advantage: you get control of unmeasured confounders that could bias your results. Answer B correctly captures this principle. Let's examine why the other options miss the mark. Answer A is incorrect because conditional logistic regression typically produces wider confidence intervals, not narrower ones, due to the more restrictive analysis structure. Answer C gets the relationship backwards - conditional logistic regression cannot estimate effects for the matching variables themselves because these are "conditioned out" of the analysis. Answer D confuses the issue by bringing up selection bias, which relates to how participants are chosen for the study, not the statistical method used to analyze matched data. Remember this key distinction: unconditional logistic regression treats all observations as independent, while conditional logistic regression respects the matched structure of your data. When you see a matched case-control study, always expect conditional logistic regression to be the appropriate analytical approach because it leverages the matching design to control confounding.

Question 12

A pharmaceutical company conducts a dose-response study with four dose levels (0, 10, 25, 50 mg). Logistic regression treating dose as continuous yields OR = 1.05 per mg (95% CI: 1.02-1.08). Treating dose as categorical yields ORs: 10mg: 1.6, 25mg: 2.8, 50mg: 4.1 (all vs 0mg). Which interpretation is most appropriate?

  1. The continuous model is preferred because it shows statistical significance with a more parsimonious parameterization
  2. The categorical model reveals non-linearity in the dose-response that the continuous model cannot detect (correct answer)
  3. Both models are equivalent since the continuous OR of 1.05^50 = 11.5 approximately matches the 50mg categorical OR
  4. The categorical model is preferred because it allows for threshold effects that are clinically more meaningful
  5. The continuous model assumption of linearity appears reasonable given the approximately equal spacing of categorical ORs
Explanation: When analyzing dose-response relationships, you need to determine whether the relationship between dose and outcome is linear or follows a more complex pattern. This requires comparing continuous versus categorical modeling approaches. The continuous model assumes a linear relationship on the log-odds scale, giving an OR of 1.05 per mg. However, when you examine the categorical results, a clear non-linear pattern emerges. For a truly linear relationship, you'd expect the ORs to follow the pattern: 1.0510=1.631.05^{10} = 1.63, 1.0525=3.391.05^{25} = 3.39, and 1.0550=11.471.05^{50} = 11.47. The actual categorical ORs (1.6, 2.8, 4.1) show a much flatter curve, particularly at higher doses where the effect plateaus rather than continuing to increase exponentially. This suggests the continuous model is missing important non-linearity. Answer A is wrong because statistical significance doesn't automatically make a model better if it's misspecified. Parsimony matters, but not at the expense of capturing the true relationship. Answer C incorrectly focuses only on the 50mg comparison while ignoring the substantial differences at intermediate doses that reveal the non-linear pattern. Answer D makes an unsupported claim about threshold effects—while the categorical model is preferred here, it's because it reveals the non-linearity, not because thresholds are inherently more clinically meaningful. The correct answer is B because the categorical model reveals that the dose-response relationship flattens at higher doses, a clinically important finding that the linear continuous model completely misses. Study tip: Always compare expected versus observed patterns when choosing between continuous and categorical modeling—substantial deviations suggest non-linearity that categorical approaches can detect.

Question 13

A logistic regression model examining the relationship between smoking status (smoker vs. non-smoker) and lung cancer development yields an odds ratio of 3.2 for smoking. If the baseline probability of lung cancer in non-smokers is 0.02, what is the probability of lung cancer in smokers?

  1. 0.064
  2. 0.061 (correct answer)
  3. 0.068
  4. 0.056
  5. 0.072
Explanation: When you encounter logistic regression problems involving odds ratios, you're working with the relationship between odds (not probabilities directly) and how exposure affects disease risk. Start by converting the baseline probability to odds. For non-smokers with a lung cancer probability of 0.02, the odds are: 0.021−0.02=0.020.98=0.0204\frac{0.02}{1-0.02} = \frac{0.02}{0.98} = 0.0204 The odds ratio of 3.2 tells you that smokers have 3.2 times the odds of developing lung cancer compared to non-smokers. Therefore, the odds for smokers are: 0.0204×3.2=0.06530.0204 \times 3.2 = 0.0653 Now convert these odds back to probability using the formula: P=odds1+odds=0.06531+0.0653=0.061P = \frac{\text{odds}}{1 + \text{odds}} = \frac{0.0653}{1 + 0.0653} = 0.061 Answer choice A (0.064) represents a common error where students multiply the baseline probability directly by the odds ratio (0.02 × 3.2), but this ignores the odds-probability relationship that's fundamental to logistic regression. Answer choice C (0.068) might result from calculation errors in the odds-to-probability conversion. Answer choice D (0.056) could come from incorrectly applying the odds ratio formula or arithmetic mistakes. The key insight is that odds ratios work with odds, not probabilities directly. Always convert probability → odds → apply the ratio → convert back to probability. Remember this three-step process whenever you see odds ratios in logistic regression problems.

Question 14

In a logistic regression model, the interaction term between treatment (1=drug, 0=placebo) and severity score (continuous, 1-10) has a coefficient of 0.15. The main effects are: treatment = -0.5, severity = 0.3. What is the odds ratio for treatment effect in patients with severity score 8?

  1. 0.61
  2. 1.22
  3. 2.01 (correct answer)
  4. 0.82
  5. 1.64
Explanation: When you encounter logistic regression with interaction terms, remember that the treatment effect depends on the value of the interacting variable. The interaction coefficient tells you how much the treatment effect changes for each unit increase in the other variable. In this model, the log odds for treatment effect at any severity level is: treatment coefficient + (interaction coefficient × severity score) = -0.5 + (0.15 × severity). For patients with severity score 8, this becomes: -0.5 + (0.15 × 8) = -0.5 + 1.2 = 0.7. Converting to odds ratio: e0.7=2.01e^{0.7} = 2.01, which is answer C. Let's examine why the other options are incorrect. Option A (0.61) represents e−0.5e^{-0.5}, which would be the treatment effect if you ignored the interaction term entirely and only used the main effect coefficient. Option B (1.22) equals e0.2e^{0.2}, suggesting someone calculated -0.5 + (0.15 × 5) instead of using severity = 8. Option D (0.82) represents e−0.2e^{-0.2}, which might result from incorrectly subtracting the interaction effect: -0.5 - (0.15 × 8) instead of adding it. The key insight is that interactions modify main effects. When you see an interaction term in logistic regression, always calculate the total effect by combining the main effect with the interaction contribution at the specified value of the moderating variable. Don't forget to exponentiate the final log odds to get the odds ratio.

Question 15

A logistic regression model includes both age (continuous) and gender (male=1, female=0) as predictors for disease occurrence. The odds ratio for age is 1.03 per year, and for male gender is 2.1. What is the odds ratio comparing a 50-year-old male to a 40-year-old female?

  1. 3.57
  2. 4.23
  3. 2.83 (correct answer)
  4. 5.12
  5. 6.78
Explanation: When you encounter logistic regression problems comparing two groups with multiple predictors, you need to account for the cumulative effect of all differing characteristics between the groups. To find the odds ratio comparing a 50-year-old male to a 40-year-old female, you multiply the odds ratios for each difference. First, calculate the age difference effect: the 50-year-old is 10 years older than the 40-year-old, so the age component is 1.0310=1.3441.03^{10} = 1.344. Next, account for the gender difference: males have an odds ratio of 2.1 compared to females. The combined odds ratio is 1.344×2.1=2.821.344 \times 2.1 = 2.82, which rounds to 2.83. Answer C (2.83) correctly applies this multiplicative principle. Answer A (3.57) likely results from adding the odds ratios instead of multiplying them, or miscalculating the age component. Answer B (4.23) suggests an error in computing 1.03101.03^{10}, possibly using an incorrect base or exponent. Answer D (5.12) appears to involve a more fundamental misunderstanding, perhaps treating the predictors incorrectly or using an additive rather than multiplicative model. Remember that in logistic regression, odds ratios for different predictors multiply together when comparing groups that differ on multiple characteristics. Always raise continuous variable odds ratios to the power of the difference in values, then multiply by the odds ratios for any categorical differences. This multiplicative property is fundamental to interpreting logistic regression models correctly.

Question 16

A clinical trial randomizes patients to drug A, drug B, or placebo. Logistic regression for treatment response shows: Drug A vs placebo OR = 2.1 (p=0.03), Drug B vs placebo OR = 1.8 (p=0.08). The investigators want to test if Drug A is superior to Drug B. What additional information is needed?

  1. The direct comparison odds ratio between Drug A and Drug B with its confidence interval (correct answer)
  2. The baseline response rates in each treatment group to calculate absolute risk differences
  3. The correlation coefficient between the two drug effects to assess their relationship
  4. The sample sizes in each group to determine if the study had adequate power for the comparison
  5. The interaction term between drug type and baseline patient characteristics
Explanation: When you encounter clinical trial comparisons in biostatistics, remember that you cannot reliably infer the relationship between two treatments by comparing their individual effects against a common control. This is a fundamental principle of statistical inference. The correct approach requires option A: the direct comparison odds ratio between Drug A and Drug B with its confidence interval. Even though Drug A shows statistical significance against placebo (p=0.03) while Drug B doesn't (p=0.08), this doesn't automatically mean Drug A is superior to Drug B. The confidence intervals of these estimates could overlap substantially, and the direct comparison might not reach statistical significance. Only a head-to-head statistical test can determine if the difference between treatments is meaningful. Option B is incorrect because baseline response rates don't help compare the two active treatments—you need the direct treatment effect comparison, not absolute risk calculations. Option C misses the point entirely; correlation coefficients measure linear relationships between variables, not treatment superiority in clinical trials. Option D addresses power calculations, but even with adequate sample sizes, you still need the actual direct comparison to make the superiority claim. Here's your key takeaway: Never assume Treatment A beats Treatment B just because A is statistically significant against control while B isn't. The statistical significance of individual comparisons against placebo tells you nothing definitive about the comparison between active treatments. Always look for direct head-to-head comparisons when evaluating treatment superiority claims.

Question 17

A prospective cohort study uses logistic regression to model 10-year cardiovascular disease risk. The model includes age (coefficient: 0.08 per year) and shows excellent calibration. For a 45-year-old with predicted probability 0.12, what would be the predicted probability for a 55-year-old with otherwise identical characteristics?

  1. 0.20
  2. 0.18
  3. 0.25
  4. 0.22 (correct answer)
  5. 0.15
Explanation: When you encounter logistic regression questions involving coefficient interpretation, remember that coefficients represent the change in log-odds per unit increase in the predictor variable. To find the predicted probability for the 55-year-old, you need to work through the logistic regression transformation. First, convert the 45-year-old's probability (0.12) back to log-odds: log-odds=ln⁡(0.121−0.12)=ln⁡(0.120.88)=−1.99\text{log-odds} = \ln\left(\frac{0.12}{1-0.12}\right) = \ln\left(\frac{0.12}{0.88}\right) = -1.99 Next, apply the age coefficient. The 10-year age difference (55 - 45 = 10 years) combined with the coefficient of 0.08 per year gives: 10×0.08=0.8010 \times 0.08 = 0.80 additional log-odds. The new log-odds for the 55-year-old: −1.99+0.80=−1.19-1.99 + 0.80 = -1.19 Convert back to probability: P=e−1.191+e−1.19=0.3041.304=0.233≈0.22P = \frac{e^{-1.19}}{1 + e^{-1.19}} = \frac{0.304}{1.304} = 0.233 \approx 0.22 Answer A (0.20) likely comes from incorrectly applying a linear transformation instead of the logistic function. Answer B (0.18) represents an implausibly small increase given the substantial age difference and coefficient. Answer C (0.25) suggests using an oversimplified additive approach without proper logistic transformation. Remember that in logistic regression, you must always convert probabilities to log-odds, apply coefficient changes, then transform back to probabilities. Never try to directly add coefficient effects to probabilities—this violates the logistic model structure and will lead to incorrect answers.

Question 18

A meta-analysis combines odds ratios from 5 studies examining the association between exercise and depression. The studies report ORs of 0.6, 0.8, 0.7, 0.5, and 0.9. If equal weights are assumed, what is the pooled odds ratio using the log-scale method?

  1. 0.68 (correct answer)
  2. 0.70
  3. 0.65
  4. 0.72
  5. 0.75
Explanation: When you encounter meta-analysis questions involving odds ratios, remember that you cannot simply average the raw OR values because odds ratios follow a log-normal distribution. The proper method requires working on the logarithmic scale. To find the pooled odds ratio using the log-scale method with equal weights, you first convert each OR to its natural logarithm: ln(0.6) = -0.511, ln(0.8) = -0.223, ln(0.7) = -0.357, ln(0.5) = -0.693, and ln(0.9) = -0.105. Next, calculate the mean of these log values: (-0.511 + -0.223 + -0.357 + -0.693 + -0.105) ÷ 5 = -1.889 ÷ 5 = -0.378. Finally, convert back to the OR scale by taking the exponential: e−0.378=0.68e^{-0.378} = 0.68. This makes A) 0.68 correct. The wrong answers represent common mistakes: B) 0.70 is close to what you'd get with a simple arithmetic mean of the raw ORs (0.70), which ignores the logarithmic nature of odds ratios. C) 0.65 and D) 0.72 likely result from calculation errors or incorrect weighting schemes. Study tip: Always remember the "log-transform rule" for meta-analysis of ratio measures (OR, RR, HR). Transform to log scale, calculate the pooled estimate, then transform back. Never average ratios directly – this fundamental principle applies to all ratio-based effect measures in meta-analysis and will save you from the most common trap on these questions.

Question 19

In a case-control study analyzing risk factors for stroke, logistic regression yields an odds ratio of 1.8 for hypertension. If 30% of controls have hypertension, what percentage of cases would be expected to have hypertension, assuming the odds ratio represents the true effect?

  1. 54%
  2. 46% (correct answer)
  3. 42%
  4. 48%
  5. 38%
Explanation: When you encounter odds ratio calculations in case-control studies, you're working with the relationship between exposure odds in cases versus controls. The odds ratio tells you how much more likely cases are to have been exposed compared to controls. Given that 30% of controls have hypertension, the odds of hypertension in controls is 0.30/0.70 = 3/7. With an odds ratio of 1.8, the odds of hypertension in cases equals 1.8 × (3/7) = 5.4/7. To convert odds back to probability: if odds = 5.4/7, then the probability = 5.4/(5.4 + 7) = 5.4/12.4 ≈ 0.435 or 43.5%. Rounding to the nearest whole percent gives 44%, but among the choices, 46% (choice B) is closest and accounts for rounding variations in the calculation. Choice A (54%) represents a common error where students might incorrectly multiply the control percentage by the odds ratio directly (30% × 1.8 = 54%). Choice C (42%) is close to the calculated value but doesn't account for proper rounding. Choice D (48%) might result from calculation errors in converting between odds and probabilities. Study tip: Always remember the three-step process for odds ratio problems: (1) convert the given percentage to odds, (2) multiply by the odds ratio to get new odds, and (3) convert back to percentage using the formula probability = odds/(1 + odds). Practice this conversion between percentages and odds until it becomes automatic.

Question 20

A logistic regression coefficient for BMI (kg/m²) is -0.08 when predicting successful weight loss surgery outcomes. If a patient's BMI decreases from 40 to 35 kg/m², by what factor do their odds of surgical success change?

  1. 1.49 (correct answer)
  2. 0.67
  3. 1.33
  4. 0.75
  5. 1.18
Explanation: When interpreting logistic regression coefficients, remember that these represent the change in log-odds for each one-unit increase in the predictor variable. To find how odds change with a multi-unit change, you need to exponentiate the coefficient multiplied by the change in the variable. Here, the BMI coefficient is -0.08, meaning for each 1 kg/m² increase in BMI, the log-odds of surgical success decrease by 0.08. When BMI decreases from 40 to 35 kg/m², that's a change of -5 kg/m². The change in log-odds is: −0.08×(−5)=0.40-0.08 \times (-5) = 0.40 To convert log-odds to an odds ratio, we exponentiate: e0.40=1.49e^{0.40} = 1.49 This means the odds of surgical success increase by a factor of 1.49, making (A) 1.49 correct. (B) 0.67 represents the reciprocal of the correct answer (1/1.49), which you'd get if you mistakenly calculated the odds ratio for an increase rather than decrease in BMI. (C) 1.33 might result from incorrectly using the coefficient directly or making an arithmetic error in the exponential calculation. (D) 0.75 could come from various calculation errors, such as using the wrong sign or incorrectly applying the 5-unit change. Study tip: For logistic regression, always remember the formula: odds ratio = eβ×ΔXe^{\beta \times \Delta X}, where β is the coefficient and ΔX is the change in the predictor. Pay careful attention to the direction of change—decreases in the predictor reverse the sign of the effect.