Healthcare Analytics Predictive Analytics & Trends Flashcards

A predictive model assigns a 73% probability of 30-day readmission to a patient. This output is best described as a:

Answer: Calibrated probability score

A calibrated probability score outputs the likelihood of an event, allowing risk stratification rather than just a yes/no prediction.

Which approach is used in healthcare predictive analytics to address severe class imbalance, such as when only 2% of patients experience the outcome of interest?

Answer: SMOTE (Synthetic Minority Over-sampling Technique)

SMOTE generates synthetic examples of the minority class to balance the training dataset, improving model sensitivity to rare but important outcomes.

A health analytics team wants to segment patients into distinct risk groups without a predefined outcome label. Which technique is most appropriate?

Answer: K-means clustering

K-means clustering is an unsupervised technique that groups patients based on feature similarity without requiring predefined outcome labels.

The Optum, IBM MarketScan, and CMS claims databases are commonly used in healthcare predictive analytics primarily because they:

Answer: Offer large-scale longitudinal patient data across diverse populations

These claims databases contain longitudinal records for millions of patients across diverse settings, making them valuable for training and validating population-level predictive models.

Which regulatory consideration is most relevant when deploying an AI-based predictive model as a clinical decision support tool in a US hospital?

Answer: FDA Software as a Medical Device (SaMD) oversight

The FDA's SaMD framework regulates AI/ML-based clinical decision support tools that meet the definition of a medical device, requiring appropriate clearance or approval.

Predictive models that learn patient-specific treatment effect heterogeneity to recommend individualized interventions are known as:

Answer: Heterogeneous treatment effect models or uplift models

Uplift or heterogeneous treatment effect models estimate how much a specific treatment benefits each individual patient, enabling personalized intervention decisions.

Which trend in healthcare predictive analytics involves using large pre-trained language models fine-tuned on clinical text to improve tasks such as diagnosis prediction and risk stratification?

Answer: Clinical foundation models (e.g., GatorTron, BioGPT)

Clinical foundation models are large language models pre-trained on vast clinical corpora and fine-tuned for downstream tasks, achieving strong performance on prediction and NLP tasks.