Personalized AI assistant for mother and child based on intelligent dialogue systems with integration into the medical information system (review)

Abstract

This literature review presents international data and clinical experience using a personalized AI assistant, built using a retrieval-augmented generation (RAG) architecture and linked to a hospital’s medical information system (MIS), for continuous support of mothers and children during the perinatal and neonatal periods. The feasibility of this approach is undeniable given the need for information continuity between obstetricians/gynecologists, neonatologists, and patients in both planned and emergency settings. Currently, the perinatal and early pediatric care system lacks a sufficient number of objective, generally accepted, and standardized digital solutions that would ensure continuous information and analytical support for the patient and child at all stages of care. Meanwhile, the need for such tools may be due to the specifics of organizing medical care, which involves the interval nature of in-person visits and insufficient integration of data on the mother’s condition, the course of pregnancy, childbirth, and the subsequent development of the child. In the absence of systematic mechanisms for information interaction between the medical organization and the family between visits, patients are forced to turn to open Internet resources, the clinical reliability of which is often questionable. Based on a systematic analysis of the literature, it was shown that modern large language models (LLM) can ensure high accuracy in obstetrics (73—89%) and pediatrics (82—93%). The use of the RAG architecture allows us to reduce the frequency of hallucinations to 0—5.8% and increase accuracy to 96.4%. The article describes a search-augmented generation (RAG) architecture, including a virtual patient model, a multi-level urgency sorting system, and a mechanism for generating structured summaries for the physician. A study design is proposed to identify risk factors and develop management tactics that can be used to validate clinical efficacy and safety, continuity, monitoring, and management of patients during and after childbirth to reduce complications and side effects, as well as to improve the effectiveness of patient management, taking into account the requirements of Russian Federation regulatory documents. The results of this work can be used to develop recommendations and materials for optimizing and improving the quality of medical care, which can be an additional support for the use of AI, ensuring speed and timeliness regardless of the time of day.

Ähnliche Arbeiten