# Integrated Implementation Strategies to Promote the Use of AI-Assisted Diagnostic Software for Lung Nodule Screening in China: Process Evaluation Based on the RE-AIM Framework

## TL;DR

This study evaluates how AI-assisted diagnostic software can be effectively used in a Chinese secondary hospital for lung nodule screening, using a referral system and data exchange with a tertiary hospital.

## Contribution

The study introduces an integrated, context-specific implementation strategy for AI diagnostic software in secondary hospitals, supported by a referral system and cloud-based data transfer.

## Key findings

- AI software detected 65.2% of lung nodules compared to 32.4% in the historical control group.
- 88% of eligible patients were successfully referred to a tertiary hospital using a cloud-based data transfer tool.
- Referral adherence and data transmission success were high, but documentation and feedback use were limited due to workflow disruptions.

## Abstract

While artificial intelligence (AI)–assisted diagnostic software holds promise for improving diagnostic efficiency and reducing disparities in health care delivery, its effective implementation in lower-tier health care settings remains limited in China. Most existing studies have focused on algorithm performance, while real-world implementation strategies remain underexplored, particularly in resource-constrained clinical environments.

This study aimed to design, implement, and evaluate an integrated, context-specific strategy to facilitate the effective implementation of AI-assisted diagnostic software for pulmonary nodule screening in a secondary hospital within China’s hierarchical health care system.

A prospective process evaluation was conducted in a secondary hospital in Beijing, supported by a collaborating tertiary referral center. The implementation strategy integrated AI software for computed tomography–based pulmonary nodule analysis into the diagnostic workflow of the secondary hospital, enabling initial screening and identification of suspected cases. Patients meeting referral criteria were referred to the tertiary hospital through a structured mechanism facilitated by a cloud-based data transfer tool, which enabled the return of diagnostic feedback and ensured continuity through a bidirectional referral and feedback system. Short-term implementation outcomes were evaluated using the RE-AIM framework, focusing on feasibility, adoption, and areas for improvement.

During the study period, 85.6% (1105/1291) of chest computed tomography scans were analyzed using the AI software, with a significant increase in the pulmonary nodule detection rate compared to the historical control group (65.2% vs 32.4%, P<.001). Among eligible patients, 88% (22/25) completed referral to the tertiary hospital, indicating a high level of adherence to the referral protocol. Moreover, 90.9% (20/22) of imaging data were transmitted successfully via the data transfer tool, facilitating timely diagnosis. However, several challenges remained, including the low rate of fully documented referral records (28%) and minimal use of diagnostic feedback by referring physicians. These limitations were largely attributed to disruptions in routine clinical workflows due to inadequate integration of the data transfer tool with existing hospital systems and continued reliance on manual data entry.

This study demonstrated the feasibility and potential value of deploying AI-assisted diagnostic software in a secondary hospital when supported by a tailored referral mechanism and interhospital data exchange systems. The findings highlighted the critical role of referral adherence, information infrastructure, and feedback mechanisms in optimizing the clinical utility of AI technologies. Further multicenter research is warranted to assess the generalizability, cost-effectiveness, long-term sustainability, and scalability of the implementation strategies across diverse health care settings.