What are the best calculations to predict hospital readmission?

From the Guidelines

The best calculations to predict hospital readmission include the LACE index, HOSPITAL score, and PREADM score, with the LACE index being particularly effective in calculating risk based on Length of stay, Acuity of admission, Comorbidities, and Emergency department visits in the past six months 1. The LACE index is a useful tool in identifying patients at high risk of readmission, with a score above 10 indicating high readmission risk.

• The HOSPITAL score evaluates Hemoglobin levels, discharge from an Oncology service, Sodium levels, Procedure during hospitalization, Index admission Type, number of Admissions in the last year, and Length of stay.
• The PREADM score incorporates Previous hospitalizations, Residence in a nursing facility, ED visits in the past six months, Age, specific Diagnoses, and number of Medications. These tools should be integrated into discharge planning, with high-risk patients receiving interventions like medication reconciliation, follow-up appointments within 7 days, clear discharge instructions, and post-discharge phone calls, as supported by recent studies 1. The use of these calculations is effective because they identify modifiable risk factors and target resources to patients most likely to benefit from enhanced transitional care, ultimately reducing readmission rates and healthcare costs. Key factors contributing to readmission include male sex, longer duration of prior hospitalization, number of previous hospitalizations, number and severity of comorbidities, and lower socioeconomic and/or educational status, with scheduled home health visits and timely ambulatory follow-up care reducing readmission rates 1.

From the Research

Predictive Models for Hospital Readmission

• Various models have been proposed to predict hospital readmission risk, including logistic regression, survival analysis, and machine learning techniques 2
• A systematic review of 77 studies found that logistic regression and survival analysis were the most widely used techniques, but machine learning techniques are becoming increasingly popular 2
• Machine learning techniques, such as decision tree-based methods and support vector machines, have been shown to improve prediction ability over traditional statistical approaches 2, 3

Performance of Predictive Models

• The performance of predictive models for hospital readmission varies significantly, with Area Under the ROC Curve (AUC) values ranging from 0.54 to 0.92 2
• A study using machine-learned features found that the LACE readmission prediction model had an AUC of 0.66, while a machine learning model had an AUC of 0.83 3
• Another study found that the use of machine learning algorithms, such as tree-based methods and neural networks, can achieve an AUC above 0.70 4

Challenges and Solutions in Predictive Modeling

• Hospital readmission prediction is facing many challenges, including data variety and complexity, data imbalance, locality and privacy, model interpretability, and model implementation 5
• Technical solutions have been proposed to address these challenges, including the use of machine learning techniques and the development of new datasets and resources 5
• A critical appraisal of published readmission models found that many models had weaknesses in their development, including failure to internally validate and failure to account for readmission at other institutions 6

Machine Learning in Predicting Hospital Readmissions

• Machine learning provides great opportunities in the prediction of hospital readmission, with many studies using electronic health records, population-based data sources, and administrative claims data 4
• The most common machine learning algorithms used for hospital readmission prediction are tree-based methods, neural networks, regularized logistic regression, and support vector machines 4
• Further research is needed to compare the performance of machine learning algorithms for hospital readmission prediction 4