Why AI Cannot Currently Find a Cure for Primary Sclerosing Cholangitis
AI cannot find a cure for PSC because the fundamental disease pathophysiology remains unknown, there are no validated therapeutic targets to model, and AI requires high-quality data with clear endpoints—none of which exist for PSC. 1
The Core Problem: Unknown Disease Biology
The 2022 EASL guidelines explicitly state that "the aetiology of PSC is unknown" and emphasize that the search for causal factors must continue 1. AI, regardless of its sophistication, cannot discover cures when the basic disease mechanisms are unclear. The guidelines identify critical knowledge gaps:
- Limited understanding of cholangiocyte biology in the context of PSC's characteristic fibrotic lesions
- Insufficient granularity on immune function in vivo in livers and bile ducts of PSC patients
- Unclear interactions between gut microbiome and biliary physiology, despite this being a promising research area
AI is fundamentally a pattern-recognition and prediction tool—it cannot generate novel biological insights from a vacuum. Without understanding what drives PSC progression, AI has no therapeutic targets to optimize or predict.
The Data Quality Problem
The 2025 position paper on AI barriers in hepatology identifies why AI struggles even when applied to liver diseases 2:
- AI adoption in hepatology remains limited despite numerous publications and regulatory approvals
- Integration into clinical workflows faces substantial barriers, including lack of validated endpoints
- AI literacy among clinicians is insufficient for proper implementation
For PSC specifically, the data challenges are severe:
Missing Clinical Endpoints
The EASL guidelines emphasize that drug development pipelines are "significantly hampered by the shortage of consensus endpoints with proven relevance to risk of liver transplantation and death" 1. AI cannot optimize treatment strategies when there's no agreement on what constitutes treatment success.
Inadequate Disease Markers
Among PSC's greatest unmet needs are:
- Lack of accurate markers of PSC severity
- Lack of sensitive tools for early detection of cholangiocarcinoma
- Inability to detect early PSC manifestations with minimal fibrosis 1
AI requires robust input data and clear outcome measures—PSC provides neither.
What AI Could Theoretically Help With (But Doesn't Yet)
The guidelines identify five research priority areas where AI might eventually contribute 1:
- Early PSC diagnosis tools - detecting minimal fibrosis manifestations
- CCA detection - identifying neoplasia at curative stages
- Endpoint clarification - measuring PSC severity for clinical trials
- Pathophysiology research - searching for causal factors
- Clinical trial design - implementing patient-reported outcomes
However, the 2025 AI barriers paper notes that "clinical AI models must prove efficacy and trustworthiness before adoption" 2. For PSC, we haven't even reached the stage of having models to test.
The Pharmaceutical Industry Problem
The lack of validated endpoints leads to "a low level of interest from the pharmaceutical industry" in PSC drug development 1. This creates a vicious cycle:
- No clear therapeutic targets → No drug development
- No drug trials → No data for AI to analyze
- No AI insights → No new therapeutic targets
Current Treatment Reality
The evidence confirms that no medical therapy has been proven to improve transplant-free survival or overall survival in PSC patients 3, 4. Ursodeoxycholic acid is widely used but only improves biochemical parameters without proven survival benefit 3. Liver transplantation remains the only curative option 1, 3.
What Needs to Happen First
Before AI can meaningfully contribute to finding a PSC cure, the field needs:
- Basic science breakthroughs understanding PSC pathophysiology through novel technologies like single-cell resolution assessments and organoid-based models 1
- Validated biomarkers for disease severity and progression
- Consensus clinical endpoints for measuring treatment efficacy
- Large-scale data repositories with standardized collection methods
- Integration of AI into clinical trials as recommended by the EASL AI Taskforce 2
AI is a powerful tool for pattern recognition in complex datasets, but it cannot substitute for fundamental biological discovery. The question isn't why AI can't find a cure—it's that medicine hasn't yet provided AI with the biological understanding, data quality, or validated targets necessary to even attempt the task for PSC.