Use of an artificial neural network (ANN) for classifying nursing care needed, using incomplete input data

Area of Science:

• Artificial Intelligence in Healthcare
• Machine Learning for Classification
• Health Informatics

Background:

• German nursing insurance uses a four-category disability classification system based on specific legal criteria.
• Obtaining complete information for classifying deceased individuals can be challenging.

Purpose of the Study:

• To evaluate an artificial neural network's (ANN) capability in determining disability categories.
• To assess ANN performance with varying input data characteristics (nature, quantity, quality).
• To determine the minimum required training data for effective ANN operation.

Main Methods:

• Retrospective observational study analyzing routine records of 14,000 adult nursing insurance clients.
• Training multiple ANNs with diverse input item sets (varying nature, number, quality) and training data sizes.
• Validation of ANN classification accuracy against expert assessments using kappa statistics.

Main Results:

• ANN achieved 80% correct classification (weighted kappa = 0.78) using all 30 input items.
• Performance remained strong with reduced input (weighted kappa = 0.63 with three items) and tolerated 20% missing values.
• A training set of 500 cases was sufficient for adequate performance.

Conclusions:

• The input item set exhibits redundancy, allowing ANNs to function effectively with subsets of data.
• ANNs serve as a valuable tool for quality control in disability classification.
• ANNs can provide satisfactory results for classifying deceased individuals with incomplete data.