Learning Prognostic Models Using Disease Progression Patterns: Predicting the Need for Non-Invasive Ventilation in...

Area of Science:

• Neurology
• Data Science
• Medical Informatics

Background:

• Amyotrophic Lateral Sclerosis (ALS) is a progressive neurodegenerative disease leading to motor neuron loss and respiratory failure.
• Current treatments focus on symptom management and survival extension, with Non-invasive Ventilation (NIV) significantly improving quality of life and life expectancy.
• Predicting the need for NIV is essential for timely and preventive patient management.

Purpose of the Study:

• To develop prognostic models for predicting the need for NIV in ALS patients.
• To identify disease presentation and progression patterns using data mining techniques.
• To enhance model interpretability by incorporating identified patterns as predictive features.

Main Methods:

• Utilized itemset mining and sequential pattern mining on static and longitudinal patient data.
• Analyzed disease presentation patterns from diagnosis data and progression patterns from follow-up data.
• Developed prognostic models incorporating mined patterns to predict NIV need at 90, 180, and 365 days.

Main Results:

• The prognostic models demonstrated promising predictive capabilities for NIV need.
• Pattern evaluation identified bulbar function and phrenic nerve response amplitude as significant indicators of disease progression.
• Findings align with clinical knowledge regarding key biomarkers for respiratory insufficiency in ALS.

Conclusions:

• Data mining techniques can effectively uncover patterns in ALS progression relevant to NIV need prediction.
• Incorporating static and temporal disease patterns enhances prognostic model accuracy and interpretability.
• This approach supports proactive clinical decision-making for NIV management in ALS patients.