Model building techniques and assumption testing MCQs With Answer

Model building techniques and assumption testing are central to clinical pharmacokinetics and therapeutic drug monitoring. This short quiz series is designed for M.Pharm students to reinforce core concepts of population PK model development, parameter estimation methods, residual error structures, covariate selection strategies, and diagnostic tests used to verify model assumptions. Each question focuses on practical decision points — choice of compartmental vs noncompartmental approaches, error model selection, handling below‑quantification limit data, goodness‑of‑fit and predictive checks, and statistical criteria for model comparison. Use these MCQs to test conceptual understanding and prepare for applied modelling tasks in therapeutic drug monitoring and PK/PD research.

Q1. Which residual error model is most appropriate when measurement error increases proportionally with concentration?

• Additive error model
• Proportional error model
• Mixed additive and proportional error model
• Log‑normal (multiplicative) error model

Correct Answer: Proportional error model

Q2. In NONMEM or population PK, which estimation method handles nonlinear mixed effects and is robust for sparse data?

• First‑order (FO)
• First‑order conditional estimation with interaction (FOCE‑I)
• Ordinary least squares (OLS)
• Nonlinear least squares (NLS)

Correct Answer: First‑order conditional estimation with interaction (FOCE‑I)

Q3. Which diagnostic plot best identifies model misspecification related to structural bias over the concentration range?

• Observed vs. predicted concentrations (DV vs PRED)
• Histogram of individual covariates
• Kaplan‑Meier survival plot
• Covariate distribution boxplot

Correct Answer: Observed vs. predicted concentrations (DV vs PRED)

Q4. What does a high eta shrinkage indicate about individual parameter estimates?

• High precision and strong individual information
• Low shrinkage and well‑estimated individual effects
• Estimates are pulled toward the population mean due to limited information
• No effect on model reliability

Correct Answer: Estimates are pulled toward the population mean due to limited information

Q5. Which statistical criterion is appropriate for nested model comparison in population modelling?

• Akaike Information Criterion (AIC) only
• Likelihood ratio test based on objective function value (OFV)
• Bayesian Information Criterion (BIC) only
• Mean squared error (MSE)

Correct Answer: Likelihood ratio test based on objective function value (OFV)

Q6. During covariate model building, forward inclusion followed by backward elimination primarily helps to:

• Select the single best covariate and ignore others
• Control type I and II error to refine a final covariate set
• Automatically transform covariates to normality
• Estimate residual variability more accurately

Correct Answer: Control type I and II error to refine a final covariate set

Q7. Which criterion indicates potential multicollinearity between covariates in a regression or PK model?

• High condition number or variance inflation factor (VIF)
• Low residual standard error
• High OFV drop with covariate inclusion
• Large interindividual variability estimate

Correct Answer: High condition number or variance inflation factor (VIF)

Q8. What is the primary purpose of a visual predictive check (VPC) in model evaluation?

• To compute bootstrap confidence intervals for parameters
• To compare observed data distribution to simulated predictive intervals
• To test normality of residuals using Shapiro‑Wilk
• To perform likelihood ratio tests for nested models

Correct Answer: To compare observed data distribution to simulated predictive intervals

Q9. Which normalized residual is recommended to assess independence and homoscedasticity in nonlinear mixed effects models?

• WRES (weighted residuals)
• NPDE (normalized prediction distribution errors)
• Raw residuals (DV‑PRED)
• Log residuals

Correct Answer: NPDE (normalized prediction distribution errors)

Q10. When handling concentrations below the limit of quantification (BLQ), which approach uses likelihood contribution of censored data rather than simple omission?

• Replace BLQ with zero
• Omit BLQ observations from the dataset
• M3 method (censored data likelihood approach)
• Replace BLQ with half the LOQ

Correct Answer: M3 method (censored data likelihood approach)

Q11. A combined (additive + proportional) residual error model is beneficial when:

• The assay error is purely constant across concentrations
• Error is negligible and can be ignored
• Low‑concentration measurements show constant error while high concentrations show proportional error
• Only log‑transformed data are analyzed

Correct Answer: Low‑concentration measurements show constant error while high concentrations show proportional error

Q12. Which method provides a nonparametric estimate of parameter uncertainty by resampling the dataset?

• Likelihood profiling
• Bootstrap
• FOCE estimation
• Kolmogorov‑Smirnov test

Correct Answer: Bootstrap

Q13. In model identification, structural unidentifiability often arises from:

• Insufficient sampling times only
• Parameter correlations or redundant parameterization in the model
• High interindividual variability alone
• Poor assay sensitivity only

Correct Answer: Parameter correlations or redundant parameterization in the model

Q14. What does a nonrandom pattern in individual weighted residuals vs time suggest?

• Model assumptions are fully satisfied
• Possible model misspecification such as wrong structural model or time‑varying bias
• Perfect fit to individual data
• Only a problem with data entry

Correct Answer: Possible model misspecification such as wrong structural model or time‑varying bias

Q15. Which strategy is preferred to evaluate predictive performance for external datasets?

• Internal goodness‑of‑fit plots only
• Cross‑validation or external validation using an independent dataset
• Only reporting OFV from the original dataset
• Using the same data for model building and testing without partitioning

Correct Answer: Cross‑validation or external validation using an independent dataset

Q16. The covariance step in NONMEM that fails to estimate standard errors usually indicates:

• Excellent parameter identifiability
• Problems such as parameter non‑identifiability, high correlations, or numerical instability
• The model has the best possible fit
• Insufficient residual variability

Correct Answer: Problems such as parameter non‑identifiability, high correlations, or numerical instability

Q17. When is log‑transformation of concentrations commonly applied in PK modelling?

• To linearize exponential error structures and stabilize variance when multiplicative errors dominate
• To make distributions more skewed
• To remove the need for any residual error model
• Only when data are categorical

Correct Answer: To linearize exponential error structures and stabilize variance when multiplicative errors dominate

Q18. Which of the following is NOT an assumption typically tested in residual diagnostics for mixed effects models?

• Normality of random effects
• Independence of residuals
• Homoscedasticity (constant variance) of residuals
• Instrument calibration procedure validity

Correct Answer: Instrument calibration procedure validity

Q19. Which modelling change is most appropriate if visual predictive checks show underprediction at peak concentrations?

• Remove all covariates from the model
• Modify the structural model (e.g., add a second compartment or transit absorption) or adjust absorption rate parameters
• Reduce the number of observations used for fitting
• Switch to noncompartmental analysis only

Correct Answer: Modify the structural model (e.g., add a second compartment or transit absorption) or adjust absorption rate parameters

Q20. Which metric summarizes both goodness‑of‑fit and model complexity and can be used for non‑nested model comparison?

• Objective Function Value (OFV) only
• Akaike Information Criterion (AIC)
• Eta shrinkage alone
• Raw residual sum of squares

Correct Answer: Akaike Information Criterion (AIC)

Download