Artificial intelligence (AI) is often presented as the cure-all for modern healthcare: faster diagnoses, personalized treatments, reduced costs. In reality, AI in pharma is already powerful and capable of transforming care.
But as Amy M. FitzPatrick showed in her case study Beyond the Black Box: Practical AI for Rare Disease and Patient-Centric Innovation, the real challenge isn’t building models — it’s AI in Pharma implementation in real-world settings.
This raises a critical question: why AI fails in healthcare implementation, even when the technology works flawlessly.
This article explores what happens when AI works perfectly… and still fails patients.
One of the most promising use cases of AI in rare disease diagnosis is addressing the persistent issue of rare diseases diagnosis delay.
In inborn errors of immunity (IEIs), formerly known as primary immunodeficiencies, patients often face an average 8-year diagnostic delay. While the broader IEI category affects roughly 1 in 1,200 people, specific subtypes may impact only 1–2 per million.
The issue isn’t lack of data. It’s fragmentation.
Patients see multiple specialists and data sits across electronic health record (EHR) systems.
So the question becomes: Can AI identify these patients earlier?
The research team developed a Structured Query Language (SQL) script to scan EHRs for clinical features associated with a rare immunological condition. They mapped literature-validated features to ICD-10 codes and applied the query across 7 major U.S. hospitals, analyzing 17 million patient records.
The results were remarkable:
But the breakthrough for AI in Pharma was only on paper.
This case highlights the real challenges of AI in Pharma implementation in clinical settings. To understand why AI fails in healthcare implementation, we need to look beyond the model and into the system.
Not one institution was ready to act on the findings.
Barriers included:
Even more concerning, implementation costs disproportionately affect rural and under-resourced hospitals. Interoperability — the ability of systems to communicate — was one of the strongest predictors of AI adoption.
AI risked amplifying healthcare inequality.
Identifying patients is only step one.
If you call someone and tell them they may have a rare genetic disease:
Without these answers, even the most advanced AI in pharma implementation may increase anxiety rather than improve outcomes.
A strong Medical Affairs AI strategy is essential to bridge the gap between innovation and real-world impact.
Medical Affairs is uniquely positioned to connect data, systems, and patient care.
Support investigator-initiated studies (IISs) and real-world validation of AI tools. Collaborate with KOLs to test feasibility within existing data systems.
AI models using datasets like MIMIC-IV and UK Biobank demonstrate high predictive accuracy (~92%) in ICU settings. Predicting individual drug responses can reduce trial-and-error prescribing and prevent adverse events.
Reducing adverse events should be a strategic priority for pharma.
Patient-centric care requires understanding.
If patients cannot interpret product information, they cannot participate in decisions. Leveraging large language models (LLMs) to make SmPCs and PIs patient-friendly — and partnering with advocacy groups to combat misinformation — is a practical next step.
Machine learning can do more than find patients — it can reanalyze existing trial data.
Sex-disaggregated analyses may reveal differences in efficacy and adverse events that were previously overlooked. For example, certain therapies have shown dramatically different responses between men and women, raising questions about how we interpret pooled data.
Regulators may soon require more granular stratification.
Medical Affairs can lead this conversation before mandates arrive.
The biggest misconception about AI in healthcare is that technology is the bottleneck, but It isn’t.
The real barriers are:
In simple terms, the model identified patients that clinicians had missed — at scale.
AI has extraordinary potential in rare disease, oncology, ICU medicine, and beyond. Across discussions at events like Next Medical Festival in Brussels 2026, one theme is becoming clear: successful AI in Pharma implementation depends not on algorithms, but on execution.
Without workflows, funding, governance, and patient pathways, even the most accurate model cannot change outcomes.
Medical Affairs stands at the intersection of science, systems, and patient care — and a well-defined Medical Affairs AI strategy will be critical to translating innovation into impact.
The future of AI in healthcare will not be decided by algorithms, but by whether we can turn them into equitable, actionable, patient-centered practice.
AI in Pharma often fails not because of poor models, but due to implementation barriers like workflow gaps, limited staff, regulatory complexity, and lack of infrastructure.
Key challenges include data fragmentation, lack of interoperability, resource constraints, unclear governance, and limited readiness of healthcare systems to act on AI insights.
AI in rare disease diagnosis can analyze large-scale EHR data to identify patterns and flag high-risk patients earlier, helping reduce diagnostic delays.
Rare disease diagnosis delay is often caused by fragmented data, multiple specialist visits, and lack of awareness, making it difficult for clinicians to connect symptoms.
A strong medical affairs AI strategy helps bridge the gap between innovation and practice by supporting evidence generation, improving interoperability, and ensuring patient-centered adoption.
