THE CLINICAL IMPACT AND COST-EFFECTIVENESS OF DIAGNOSTIC MODALITIES FOR INDETERMINATE BILIARY STRICTURES, INCLUDING NEXT-GENERATION SEQUENCING AND ARTIFICIAL INTELLIGENCE: A MICROSIMULATION ANALYSIS

Background:
Indeterminate biliary strictures persist as a diagnostic challenge, particularly due to the concern of delayed diagnosis of malignancy. Various strategies have been developed to improve the sensitivity of standard ERCP with brush cytology, including cholangioscopy with biopsy, next-generation sequencing (NGS), and cholangioscopy-based artificial intelligence (AI). However, the impact and cost-effectiveness of these strategies have not been compared.

Methods:
We developed a state-transition Markov microsimulation model with a base case of a patient with an indeterminate biliary stricture, comparing ERCP with brushing, brushing and FISH, cholangioscopy with brushing/biopsy, cholangioscopy with brushing/biopsy and AI (AI-BB) versus cholangioscopy with brushing/biopsy and NGS (NGS-BB). Prevalence of malignancy versus benign stricture, transition probabilities, cost, and performance characteristics of each modality were identified from published literature and public data sources. Patients diagnosed with malignancy by any means (AI, FISH, NGS, biopsy), were presumed localized and would undergo treatment for cholangiocarcinoma and surgery if resectable. To minimize false positive diagnoses, patients diagnosed by FISH/NGS/AI alone would undergo a repeat ERCP for confirmation. Outcomes included accurate diagnoses and missed diagnoses of malignancy after 3 ERCPs, number of ERCPs required for diagnosis of malignancy, and incremental cost-effectiveness ratio (ICER) with willingness-to-pay threshold $100,000/QALY.

Results:
Under all scenarios, brushing alone, cholangioscopy with biopsy/brushing, and FISH-based strategies were dominated (i.e., less effective and more costly) and removed from the analyses. In patients with an indeterminate biliary stricture, NGS-BB was the cost-effective strategy ($3,621/QALY) compared with AI-BB (when maximizing AI accuracy). NGS-BB identified 99.7% of patients with malignancy. The false positive diagnosis rate was 0.72% with AI-BB and 0.05% with NGS-BB. When maximizing the sensitivity of AI-BB, NGS-BB was the dominant strategy. The AI-BB strategy resulted in a higher false positive rate of 3.03% than the NGS strategy (0.05%). When maximizing the specificity of AI, AI-BB became the cost-effective strategy ($17,500/QALY). This strategy identified 99.5% of patients with malignancy. Under this scenario, adding AI could cost up to $131 and remain cost-effective at a WTP of $100,000.

Conclusions:
Among currently available diagnostic modalities for indeterminate biliary strictures, NGS-BB was cost-effective. AI-BB can be cost-effective, particularly if its diagnostic specificity is maximized. These results can inform model development and validation as well as implementation efforts in clinical practice. Future studies will be needed to evaluate the effect of combining AI and NGS for indeterminate biliary stricture management.