Imputation of missing values by integrating neural networks and case-based reasoning

Area of Science:

• Medical Informatics
• Artificial Intelligence in Healthcare
• Neonatal Research

Background:

• Missing data in medical databases, especially when not missing at random, poses significant challenges for developing robust prediction models.
• Accurate physiologic parameters are crucial for effective patient management and outcome prediction in neonatal intensive care units (NICUs).

Purpose of the Study:

• To develop and present an advanced data imputation approach for physiologic parameters in a neonatal intensive care unit (NICU) database.
• To address the issue of non-random missing data by incorporating individualized patient information into imputed values.

Main Methods:

• Integration of artificial neural networks (ANNs) and case-based reasoning (CBR) to impute missing physiologic data.
• Utilizing relevant patient data within the NICU database to inform the imputation process.

Main Results:

• Successfully replaced missing values in the NICU database using the integrated ANNs and CBR approach.
• The imputation method incorporates individualized case information, enhancing the relevance of imputed data.

Conclusions:

• The proposed data imputation method effectively handles non-random missing data in NICU databases.
• Combining ANNs and CBR offers a promising strategy for improving the quality of medical data for prediction modeling.