Objectives To assess healthcare costs of patients screened for cervical spine (C-spine) fractures using CT, and estimate the change in in-hospital costs if an artificial intelligence (AI) algorithm for C-spine fracture detection would assist the radiologist as concurrent reader. Materials and methods This retrospective, early health technology assessment included 2321 consecutive patients (2007–2014; median age 49 years; 61% male) screened for C-spine fractures using CT, of whom 219 patients with fractures. Healthcare costs were calculated per diagnostic category (true positive, true negative, false positive and false negative) based on the diagnosis made by the attending radiologists, and by AI analysis, compared to the reference standard. The potential diagnostic accuracy measures with 95% confidence intervals (CI) for diagnoses made by radiologists assisted by AI and the potential average cost per diagnostic category in this scenario were estimated. Results Radiologists identified 193/219 fractures and 2085/2102 scans without fractures. AI identified 23 fractures and 16 scans without fractures that had been misclassified by the radiologists. This resulted in a potential sensitivity of 216/219 (98.6%, 95% CI: 95.7–99.6, 10.5% increase compared to radiologists) and specificity of 2101/2102 (100.0%, 95% CI: 99.7–100, 0.8% increase compared to radiologists) for the AI-assisted scenario. The total cost for the AI-assisted scenario was €60,862 (0.3%) higher than for radiologists alone. Conclusion In this scenario analysis, the addition of AI as concurrent reader to radiologists was estimated to potentially increase sensitivity by 10.5% and specificity by 0.8% together with a 0.3% increase in in-hospital costs due to more detected fractures. Key Points <Emphasis Type="BoldItalic">Question</Emphasis> The impact of artificial intelligence to detect cervical spine fractures on CT on in-hospital healthcare resource use and costs is unknown. <Emphasis Type="BoldItalic">Findings</Emphasis> Assistance by AI could potentially improve radiologists’ sensitivity and specificity by 10.5% and 0.8%, respectively, for an additional healthcare cost increase of just 0.3%. <Emphasis Type="BoldItalic">Clinical relevance</Emphasis> Using artificial intelligence as concurrent reader to radiologists to detect cervical spine fractures on CT could potentially increase diagnostic accuracy to nearly 100% with a minimal increase in in-hospital healthcare resource costs. Graphical Abstract

Potential change in healthcare costs of implementing artificial intelligence for detecting cervical spine fractures on CT: an early health technology assessment

期刊论文
发表日期：

2026年1月5日

查看全文

Gaby J. van den Wittenboer,
Ingrid M. Nijholt,
Ingelin Kvamme,
Chris van Lieshout,
Mario Maas,
Martijn F. Boomsma

摘要

Objectives

To assess healthcare costs of patients screened for cervical spine (C-spine) fractures using CT, and estimate the change in in-hospital costs if an artificial intelligence (AI) algorithm for C-spine fracture detection would assist the radiologist as concurrent reader.

Materials and methods

This retrospective, early health technology assessment included 2321 consecutive patients (2007–2014; median age 49 years; 61% male) screened for C-spine fractures using CT, of whom 219 patients with fractures. Healthcare costs were calculated per diagnostic category (true positive, true negative, false positive and false negative) based on the diagnosis made by the attending radiologists, and by AI analysis, compared to the reference standard. The potential diagnostic accuracy measures with 95% confidence intervals (CI) for diagnoses made by radiologists assisted by AI and the potential average cost per diagnostic category in this scenario were estimated.

Results

Radiologists identified 193/219 fractures and 2085/2102 scans without fractures. AI identified 23 fractures and 16 scans without fractures that had been misclassified by the radiologists. This resulted in a potential sensitivity of 216/219 (98.6%, 95% CI: 95.7–99.6, 10.5% increase compared to radiologists) and specificity of 2101/2102 (100.0%, 95% CI: 99.7–100, 0.8% increase compared to radiologists) for the AI-assisted scenario. The total cost for the AI-assisted scenario was €60,862 (0.3%) higher than for radiologists alone.

Conclusion

In this scenario analysis, the addition of AI as concurrent reader to radiologists was estimated to potentially increase sensitivity by 10.5% and specificity by 0.8% together with a 0.3% increase in in-hospital costs due to more detected fractures.

Key Points

Question The impact of artificial intelligence to detect cervical spine fractures on CT on in-hospital healthcare resource use and costs is unknown.

Findings Assistance by AI could potentially improve radiologists’ sensitivity and specificity by 10.5% and 0.8%, respectively, for an additional healthcare cost increase of just 0.3%.

Clinical relevance Using artificial intelligence as concurrent reader to radiologists to detect cervical spine fractures on CT could potentially increase diagnostic accuracy to nearly 100% with a minimal increase in in-hospital healthcare resource costs.

Graphical Abstract