Computational recommendation system for parenteral nutrition prescription
Abstract:
Introduction: There is a drive in some countries to increase the use of multichamber bags as an alternative to compounded parenteral support for patients with chronic intestinal failure to preserve aseptic pharmacy capacity. If a required regimen is that of a multichamber bag plus one intravenous fluid product, then the potential combinations of available products can exceed 3000.
Method: A computer-based selection algorithm known as “PNMatch” was developed, which uses a “k-nearest neighbor” strategy for matching. After completion of the computational design for the algorithm, a staged testing and development process was undertaken, evaluating the algorithm against clinician prescribing. A matching assessment was then undertaken against existing prescriptions.
Results: Algorithm functionality was assessed through 20 parenteral nutrition formulation requests, all of which were successfully processed. Prescription selection demonstrated a 90% exact match rate, with one unmatched case and one nonimproved match. The mean time for prescription selection using PNMatch was 116 s. Multichamber bag requests were then evaluated. Most prescriptions matched closely with the existing formulations, achieving a Cohen kappa value of 0.8. Further analysis of 44 formulation requests showed successful product selection for 43 requests. The mean difference in volume was -75.3 ml, and for potassium, it was -4.9 mmol.
Conclusion: This is the first demonstration of the successful development and testing of a computer-based selection algorithm for the prescription of a multichamber bag-based regimen for patients with chronic intestinal failure. This has the potential to improve efficiency and reduce variability in parenteral support prescribing.
Reference:
Bond A, Harrison S, Lal S, Gerber L. Development of a computational recommendation system for parenteral nutrition prescription: An algorithm study. JPEN J Parenter Enteral Nutr. 2026 Jan 24. doi: 10.1002/jpen.70053. Epub ahead of print. PMID: 41578805.