A tutorial on dynamic risk prediction of a binary outcome based on a longitudinal biomarker

Bayley trajectories predict school readiness better than single assessments in formerly very preterm preschoolers

BACKGROUND

Development of children born very preterm (VPT) is evaluated using the Bayley Scales of Infant Development. Early Bayley scores may not predict later outcomes. We studied whether VPT Bayley trajectories in the early years predicted school readiness better than single assessments.

METHODS

We prospectively evaluated 53 VPT at 4-5 years using standardized measures of school readiness, including the domains of cognition, early mathematical and literacy abilities, and motor skills. Predictors were Bayley-III scores obtained 1-5 times/child between 6 and 35 months. Linear mixed models (LMM) with random effects extracted estimated random effect for slope (change in Bayley score/1 year) and fixed+random effect sum for the intercept (initial Bayley score) for each participant, to then evaluate 4-5-year outcomes prediction.

RESULTS

Variability of individual trajectories prevailed across developmental domains. For the initial LMM, adding Bayley change to models with only initial score improved model fits for several Bayley-III domains. Models containing estimates for initial Bayley scores and Bayley change explained significantly more variability in school readiness scores (21-63%) than either variable alone.

CONCLUSION

Neurodevelopmental follow-up of VPT is more relevant to school readiness when children are assessed multiple times in the first 3 years. Neonatal intervention research could use early trajectories rather than single timepoints as outcomes.

IMPACT

This study is the first to examine individual Bayley scores and trajectories to predict school readiness of formerly preterm children at 4-5 years. Modeling demonstrated extreme variability of individual trajectories compared to the group's average trajectories. Models containing initial Bayley scores and Bayley change over time explained more variability in preschool readiness than either variable alone. Using the Bayley to predict future school readiness is enhanced by administration across multiple follow-up visits and inclusion of change across the first 3 years. Follow-up care models and clinical trial design for neonatal interventions may benefit from a trajectory-based approach to outcomes evaluation.

A Bayesian learning model to predict the risk for cannabis use disorder

BACKGROUND

The prevalence of cannabis use disorder (CUD) has been increasing recently and is expected to increase further due to the rising trend of cannabis legalization. To help stem this public health concern, a model is needed that predicts for an adolescent or young adult cannabis user their personalized risk of developing CUD in adulthood. However, there exists no such model that is built using nationally representative longitudinal data.

METHODS

We use a novel Bayesian learning approach and data from Add Health (n = 8712), a nationally representative longitudinal study, to build logistic regression models using four different regularization priors: lasso, ridge, horseshoe, and t. The models are compared by their prediction performance on unseen data via 5-fold-cross-validation (CV). We assess model discrimination using the area under the curve (AUC) and calibration by comparing the expected (E) and observed (O) number of CUD cases. We also externally validate the final model on independent test data from Add Health (n = 570).

RESULTS

Our final model is based on lasso prior and has seven predictors: biological sex; scores on personality traits of neuroticism, openness, and conscientiousness; and measures of adverse childhood experiences, delinquency, and peer cannabis use. It has good discrimination and calibration performance as reflected by its respective AUC and E/O of 0.69 and 0.95 based on 5-fold CV and 0.71 and 1.10 on validation data.

CONCLUSION

This externally validated model may help in identifying adolescent or young adult cannabis users at high risk of developing CUD in adulthood.

Early prediction of post-molar gestational trophoblastic neoplasia and resistance to methotrexate, based on a single serum human chorionic gonadotropin measurement

BACKGROUND

Clinicians are unable to provide individualized counseling regarding risk of progression for patients with a complete hydatidiform mole (CHM). We developed nomograms enabling early prediction of post-molar gestational trophoblastic neoplasia (GTN) and resistance to methotrexate (MTX) based on a single serum human chorion gonadotropin (hCG) measurement.

METHODS

We generated two nomograms with logistic regression: to predict post-molar GTN, and MTX resistance. For patients with high probability to progress to post-molar GTN or MTX resistance, we determined hCG cut-offs at 97.5% specificity to select patients for additional- or adjustments in current treatment.

RESULTS

The nomograms had a good to excellent ability to distinguish either between patients with uneventful hCG regression versus progression to post molar GTN, or between patients cured by MTX versus patients in whom resistance would occur. At 97.5% specificity, we identified 66% (95%CI 56-75) of the 149 patients who would progress to post-molar GTN, four weeks after initial curettage. For patients treated with MTX, we identified 55% (95%CI 23-83) of the 43 patients who would become resistant, preceding their third course at 97.5% specificity.

CONCLUSION

The nomograms and cut-off levels can be used to assist in counseling for patients diagnosed with CHM.

A tutorial on dynamic risk prediction of a binary outcome based on a longitudinal biomarker

Dynamic risk predictions based on all available information are useful in timely identification of high‐risk patients. However, in contrast with time to event outcomes, there is still a lack of studies that clearly demonstrate how to obtain and update predictions for a future binary outcome using a repeatedly measured biomarker. The aim of this study is to give an illustrative overview of four approaches to obtain such predictions: likelihood based two‐stage method (2SMLE), likelihood based joint model (JMMLE), Bayesian two‐stage method (2SB), and Bayesian joint model (JMB). We applied the approaches to provide weekly updated predictions of post–molar gestational trophoblastic neoplasia (GTN) based on age and repeated measurements of human chorionic gonadotropin (hCG). Discrimination and calibration measures were used to compare the accuracy of the weekly predictions. Internal validation of the models was conducted using bootstrapping. The four approaches resulted in the same predictive and discriminative performance in predicting GTN. A simulation study showed that the joint models outperform the two‐stage methods when we increase the within‐ and the between‐patients variability of the biomarker. The applicability of these models to produce dynamic predictions has been illustrated through a comprehensive explanation and accompanying syntax (R and SAS^®).