7 Internal Medicine and Hospital Medicine

Learning Objectives

This chapter examines AI applications for hospitalists, general internists, and subspecialists managing complex inpatients. You will learn to evaluate AI for hospital workflows, chronic disease management, clinical decision support, and discharge planning.

📋 Chapter Summary (TL;DR)

Key Applications: - Hospital early warning systems (deterioration prediction) - Readmission risk prediction - Length of stay forecasting - Chronic disease management (diabetes, heart failure, COPD) - Medication management and drug-drug interaction checking - Discharge planning optimization

Strong Evidence: Readmission prediction models, some early warning systems Weak Evidence: Most autonomous treatment recommendations Critical Need: Integration into hospital EHR workflows, avoiding alert fatigue

7.1 Major AI Applications in Hospital Medicine

7.1.1 1. Patient Deterioration and Early Warning

Epic Deterioration Index, WAVE, others - Predict cardiac arrest, ICU transfer Evidence: Variable - some benefit in RCTs, implementation-dependent Challenge: High false positive rates, alert fatigue

7.1.2 2. Readmission Risk Prediction

HOSPITAL score + AI enhancements Use case: Target high-risk patients for care transitions Evidence: Good prediction, but effective interventions remain elusive (Kansagara et al. 2011)

7.1.3 3. Chronic Disease Management

Diabetes: CGM data analysis, insulin dosing recommendations Heart failure: Remote monitoring + AI alerts for decompensation COPD: Exacerbation prediction from symptoms, spirometry

7.1.4 4. Medication Safety

Drug-drug interaction checking (traditional CDS + AI enhancements) Deprescribing recommendations for polypharmacy Personalized dosing (pharmacokinetics/pharmacodynamics)

7.1.5 5. Diagnostic Support

Differential diagnosis generation (Isabel, DXplain) Lab result interpretation (flag abnormals, suggest workup) Imaging interpretation (see Radiology chapter)

7.2 Implementation Challenges

EHR integration complexity
Workflow disruption in busy hospital environment
Handoffs between teams
Liability for AI-assisted decisions
Cost-benefit balance

7.3 The Clinical Bottom Line

Hospital AI is promising but immature compared to radiology AI. Most applications require physician oversight. Start with well-validated readmission prediction and deterioration models. Demand local validation. Monitor for alert fatigue.