Abstract PS03-06: Clinical implementation of artificial-intelligence-assisted detection of breast cancer metastases in sentinel lymph nodes: saving costs and time (the CONFIDENT-B trial)

Abstract

Abstract Background Metastases in sentinel lymph nodes (SN) of breast cancer (BC) patients are strongly associated with a worse survival and consequently guide treatment. If metastases are absent upon pathologist’ assessment of the regular hematoxylin and eosin (HE)-slide, additional immunohistochemistry (IHC)-stains are performed to ensure that no metastases are missed. However, these stains come at high additional costs, which may exceed specimen reimbursement. Fortunately, digital pathology is becoming more common, thereby creating an avenue of opportunities for artificial intelligence (AI) assistance. Although the number of studies on (promising) AI-algorithms increases exponentially, studies on actual clinical implementation are lacking. In this single-center prospective trial, we investigated to which extent an artificial intelligence (AI)-assisted clinical workflow for the detection of SN-metastases reduces IHC-use, while maintaining current diagnostic safety standards. Methods We enrolled 190 SN-specimens of 182 patients with invasive or in situ BC from September 2022 to May 2023. SN-specimens were allocated bi-weekly to either the control-arm (n=90) or the intervention-arm (n=100). In the control-arm, SN-specimens were digitally assessed according to the current clinical workflow, while pathologists in the intervention-arm assessed the SN-specimen with the output the ‘Metastasis-Detection-App’ (Visiopharm©) available. In both groups, IHC was performed in all morphologically negative cases. Main outcome was the relative risk (RR) of IHC-use per detected case of SN-metastases. Case-mix adjustment was performed by log-binomial regression. Results Overall, 59 SN-specimens contained metastases (31.1%). AI-assistance resulted in a significantly lower risk of IHC-use per detected case of SN-metastases (adjusted RR: 0.680, 95% CI: 0.347-0.878). Besides preventing IHC-use, thereby reducing costs, AI-assisted pathologists also spent significantly less time on their assessment of the SN-specimen (3m:41s vs. 6m:04s, p = 0.028). Furthermore, the sensitivity of AI-assisted pathologists was up to 30% higher. The AI-assisted pathologists missed two cases of micro-metastases in the intervention arm, one of which was in retrospect highlighted by the algorithm, while in the other case the tumor cells were located in a heavily cauterized area of the HE-slide and therefore only visible on the (serial) IHC-slides. In the control arm, the algorithm in retrospect picked up all micro- and macro-metastases and nearly half of the isolated tumor cells (ITC). In addition, all participating pathologists stated that AI was easy to use, that they felt confident using AI, and that besides saving them time, AI made their work more enjoyable. Cost reductions on IHC by AI-assistance depend on laboratory policy (i.e. when and on how many levels IHC is performed), but, at a cost of €25,- per IHC-stain, range from €1.500 to €3.500 per 100 SN’s in a scenario where IHC is performed in all morphologically negative cases. In a scenario where IHC is only performed in patients in whom finding ITC has clinical consequences (i.e. patients who received neoadjuvant treatment), cost savings on IHC range from €7.500-€12.500 per 100 SNs, depending on laboratory policy. Conclusion AI-implementation for the detection of SN-metastases in BC-patients leads to a significant reduction of IHC-use and subsequent costs, while saving pathologists time and making their work more enjoyable. Importantly, AI-implementation during this trial was safe and patients were not at risk of an inferior diagnosis. By doing this trial alone, an estimated €3,000 on IHC-use was saved. Such tangible cost savings are crucial to build a viable business case for AI implementation in diagnostic pathology. Citation Format: Carmen van Dooijeweert, Rachel Flach, Natalie ter Hoeve, Celien Vreuls, Roel Goldschmeding, Jan-Erik Freund, Paul Pham, Tri Nguyen, Elsken Van der Wall, Geert Frederix, Nikolas Stathonikos, Paul Van Diest. Clinical implementation of artificial-intelligence-assisted detection of breast cancer metastases in sentinel lymph nodes: saving costs and time (the CONFIDENT-B trial) [abstract]. In: Proceedings of the 2023 San Antonio Breast Cancer Symposium; 2023 Dec 5-9; San Antonio, TX. Philadelphia (PA): AACR; Cancer Res 2024;84(9 Suppl):Abstract nr PS03-06.