Automated screening of potential organ donors using a temporal machine learning model

Area of Science:

• Medical Informatics
• Machine Learning in Healthcare
• Organ Transplantation

Background:

• Organ donation demand significantly exceeds supply.
• A substantial percentage of potential organ donors are not identified due to manual screening limitations.
• Current identification relies on manual processes by Organ Donation Organizations (ODOs).

Purpose of the Study:

• To develop and validate an automated machine learning system for identifying potential organ donors.
• To reduce the proportion of missed potentially eligible organ donors.
• To compare the performance of a neural network model against a logistic regression model.

Main Methods:

• Retrospective development and testing of a neural network model using routine clinical and laboratory time-series data.
• Training a convolutive autoencoder on longitudinal laboratory data.
• Integrating a deep neural network classifier.
• Comparison with a logistic regression model.

Main Results:

• The neural network model achieved an AUROC of 0.966, outperforming the logistic regression model (0.940).
• At a specific cutoff, both models showed similar sensitivity (84%) and specificity (93%).
• The neural network demonstrated robust accuracy across donor subgroups and in prospective simulations, unlike the logistic regression model.

Conclusions:

• Machine learning models, particularly neural networks, can effectively identify potential organ donors using routinely collected data.
• Automated screening systems have the potential to significantly improve donor identification rates.
• This approach offers a promising solution to enhance organ donation and transplantation efforts.