Q&A Set Based on the MELD-Plus Study
Based on: Kartoun U, Corey KE, Simon TG, Zheng H, Aggarwal R, Ng K, Shaw SY. The MELD-
Plus: A generalizable prediction risk score in cirrhosis. PLoS ONE. 2017;12(10):e0186301.
https://doi.org/10.1371/journal.pone.0186301
Graduate-Level Exam: MELD-Plus Study
1. Explain the rationale behind using adaptive LASSO instead of traditional LASSO or
Ridge regression in the context of MELD-Plus.
Answer: Adaptive LASSO assigns different penalty weights to different coefficients,
enabling consistent variable selection while reducing bias. It is particularly useful when
predictor importance varies, as in EMR data.
2. Discuss the trade-offs between AUROC and clinical interpretability in developing
predictive models like MELD-Plus.
Answer: Higher AUROC indicates better model performance, but complex models may
reduce interpretability. MELD-Plus balances this by using clinically familiar variables,
ensuring both high accuracy and usability.
3. The authors used a 70/30 train-test split. Discuss alternative validation strategies and
when they might be preferred.
Answer: Alternatives include k-fold cross-validation or bootstrapping. These are
preferred when data are limited or when estimating model stability across different
subsets.
4. Evaluate the generalizability of MELD-Plus based on the use of the IBM Explorys
dataset. What limitations remain despite this external validation?
Answer: While external validation on Explorys shows generalizability, limitations
include demographic, coding, and care variability. Differences in data capture and
healthcare practices can affect performance.
5. Propose a modification to MELD-Plus that could make it suitable for outpatient
cirrhosis risk monitoring.
Answer: Incorporating time-series data from outpatient visits, such as trends in labs and
vitals, and including additional outpatient-specific variables (e.g., medication adherence)
could improve relevance.
6. Critically assess the implications of using mean imputation for missing data in this
study. What alternative would you recommend?
Answer: Mean imputation is simple but can bias estimates and underestimate variability.
Alternatives like multiple imputation or predictive mean matching would better preserve
relationships among variables.
Based on: Kartoun U, Corey KE, Simon TG, Zheng H, Aggarwal R, Ng K, Shaw SY. The MELD-
Plus: A generalizable prediction risk score in cirrhosis. PLoS ONE. 2017;12(10):e0186301.
https://doi.org/10.1371/journal.pone.0186301
Graduate-Level Exam: MELD-Plus Study
1. Explain the rationale behind using adaptive LASSO instead of traditional LASSO or
Ridge regression in the context of MELD-Plus.
Answer: Adaptive LASSO assigns different penalty weights to different coefficients,
enabling consistent variable selection while reducing bias. It is particularly useful when
predictor importance varies, as in EMR data.
2. Discuss the trade-offs between AUROC and clinical interpretability in developing
predictive models like MELD-Plus.
Answer: Higher AUROC indicates better model performance, but complex models may
reduce interpretability. MELD-Plus balances this by using clinically familiar variables,
ensuring both high accuracy and usability.
3. The authors used a 70/30 train-test split. Discuss alternative validation strategies and
when they might be preferred.
Answer: Alternatives include k-fold cross-validation or bootstrapping. These are
preferred when data are limited or when estimating model stability across different
subsets.
4. Evaluate the generalizability of MELD-Plus based on the use of the IBM Explorys
dataset. What limitations remain despite this external validation?
Answer: While external validation on Explorys shows generalizability, limitations
include demographic, coding, and care variability. Differences in data capture and
healthcare practices can affect performance.
5. Propose a modification to MELD-Plus that could make it suitable for outpatient
cirrhosis risk monitoring.
Answer: Incorporating time-series data from outpatient visits, such as trends in labs and
vitals, and including additional outpatient-specific variables (e.g., medication adherence)
could improve relevance.
6. Critically assess the implications of using mean imputation for missing data in this
study. What alternative would you recommend?
Answer: Mean imputation is simple but can bias estimates and underestimate variability.
Alternatives like multiple imputation or predictive mean matching would better preserve
relationships among variables.
