OPOs are using AI to improve both operations and clinical decision-making in organ donation. Through real-world examples, Southwest Transplant Alliance (STA) and New England Donor Services (NEDS) demonstrated how AI tools, such as automated referrals and predictive models, can help identify more donor opportunities, reduce manual workload, and support staff in making faster and more accurate decisions. Innovations are being put into practice to meet the growing challenges in the field.
Challenges
- Growing demands on clinical staff
- Increasing volumes of unstructured data
- The need for faster, more accurate decision-making in donor identification and Donation After Circulatory Death (DCD) assessment
- Limitations of manual workflows that hindered consistency and response time across cases
Objectives of the Clinical Intervention
- STA
- To automate the referral process to identify potential donors earlier in the clinical workflow and reduce delays in evaluation
- To streamline the intake and analysis of clinical data to ease the burden on staff and improve operational efficiency
- NEDS
- To equip staff with predictive tools that support more informed and accurate decisions regarding DCD viability
- To enhance the consistency and timeliness of decision-making across cases through standardized, data-driven processes
Approaches and Actions Taken
- STA
- Implemented real-time referral triggers in hospital EMRs using FHIR APIs and CDS Hooks, enabling referrals at the time of admission
- Introduced natural language processing (NLP) to find key insights from unstructured clinical data, reducing manual review and improving coordination
- NEDS
- Built a DCD viability prediction model using TensorFlow and multiple linear regression
- Analyzed over 4,000 records to identify 20 key clinical predictors from an initial set of 54
- Integrated the model into a real-time dashboard to generate probability scores that support case-level decision-making
Findings (Data and Results)
- STA reported a 45% increase in referrals during initial implementation: Staff and model predictions aligned 95% of the time.
- NEDS’s model achieved 73% overall accuracy, occasionally outperforming staff: The model was most effective in cases of very early or late expiration.
Future Directions
STA is scaling its approach thoughtfully, aligning staffing models with hospital expansion. NEDS is incorporating direct EMR feeds for richer data and exploring new predictive variables such as injury location and eye movement to improve model performance.
