How AI‑Powered Smart Triage Is Revolutionizing Patient Routing in JD Health’s Internet Hospital

Background and Policy Context

Patients often mis‑select medical departments due to limited health knowledge, leading to delayed diagnosis, higher misdiagnosis rates, and increased psychological and financial burdens. JD.com’s Internet Hospital introduced an AI‑based smart triage service that matches simple patient descriptions to the appropriate department, improving efficiency and care quality.

Medical AI, built on internet infrastructure, leverages NLP, deep learning, computer vision, and big‑data services to lower costs and enhance outcomes. National policies such as the 2017 "New Generation AI Development Plan" and the 2018 "Internet+ Healthcare" directive mandate widespread online triage and guidance services by 2020.

Industry Pain Points

China’s healthcare system faces severe supply‑demand imbalance: an aging population of 249 million, over 300 million chronic‑disease patients, and an 8.7% annual incidence rise, contrasted with only 3.6 million practicing physicians for a daily outpatient volume of 23 million. Consequently, appointment difficulty and department mismatches are common.

Online registration has surged, yet patients still frequently select wrong departments due to limited medical knowledge. Manual triage is slow, costly, and error‑prone, wasting resources. The COVID‑19 pandemic amplified these issues, with peak daily online users exceeding 8 million.

Solution Overview

JD Health’s research team combined AI and big‑data techniques with a professional triage knowledge base and disease knowledge graph to develop an intelligent triage service. Users simply describe their symptoms in natural language, and the system recommends the most suitable department, seamlessly integrating into pre‑consultation workflows.

The service continuously learns from scenarios, diseases, and user feedback to expand coverage across all departments, delivering precise matches while dramatically reducing operational costs compared to manual triage.

Core Technologies

Triaging Knowledge Base & Rule Engine: Built on national standards, the knowledge base extracts rules from authoritative clinical guidelines, literature, and JD’s own user scenarios, with validation from full‑time physicians to ensure professional accuracy.

Medical Information Recognition Algorithm: A suite of NLP models processes de‑identified data to simulate patient intent, iteratively training to improve recognition capabilities.

Multimodal Fusion Triage Algorithm: Combines knowledge matching, deep‑learning neural networks, domain‑specific adjustments, and hierarchical voting. Models cross‑validate each other to produce final triage results that balance accuracy, coverage, and explainability.

Application Example

During the first half of 2020, JD’s smart triage operated 24/7 with zero downtime, zero manual triage, and no long waiting times. It matched tens of thousands of daily users with thousands of doctors across 20 primary and 80 secondary specialties, delivering immediate, professional consultations during the pandemic.

Future Vision

Platformization: Extend the smart triage service to hospital websites, apps, WeChat public accounts, mini‑programs, and third‑party registration platforms, reducing mis‑appointments and enhancing overall patient experience.

Scenario Expansion: Evolve the triage knowledge base into a comprehensive medical‑health knowledge graph covering disease management, risk identification, dialogue scripts, follow‑up pathways, and health education. This will enable intelligent assistance throughout the entire care cycle—pre‑consultation, in‑consultation, post‑consultation, and long‑term health management.