By Editorial TeamA diagnostics company raised a fortune and then collapsed into a criminal conviction. That kind of spectacular failure leaves a clear map of what not to do, and for today’s healthcare AI vendors, the biggest lesson concerns the evidence gate.
The Theranos Record and Evidence Gate Bypass
The Theranos story is a cautionary tale about what happens when a diagnostics company bypasses standard clinical validation pathways. They made claims about revolutionary blood testing technology but skipped the part where independent review demands strong, peer-reviewed evidence, the kind of evidence that simply wasn’t there before their products were already touching patients. As Eric Topol pointed out, the company’s claims lacked scientific substantiation. They operated outside established medical device regulations, prioritizing market entry over clinical rigor. The lack of Peer-Reviewed Clinical Validation defined the entire case.
Expected Review Pathways for Health AI
For health AI, there are established review pathways you’re supposed to follow. The FDA 510(k) Pathway is the standard route for most medical device claims, where you have to prove your device is substantially equivalent to something already on the market (a “predicate device”). If your device is novel and low-to-moderate-risk with no predicate, the De Novo Classification path is your option. Either way, you’re submitting rigorous data to the FDA’s Center for Devices and Radiological Health (CDRH), whose records document a medical device claim’s expected journey from premarket submissions to postmarket surveillance. FDA guidance on medical device premarket submissions Inside this process, Peer-Reviewed Clinical Validation is a critical trust indicator, signifying independent scientific scrutiny of your product’s performance. On top of that, you’re expected to follow Good Machine Learning Practice (GMLP) principles for data management, model development, and performance monitoring, all while maintaining a Quality Management System (QMS) compliant with ISO 13485 that ensures consistent product quality and regulatory adherence.
Comparative Failures in Evidence Generation
Other companies faced similar evidence challenges. Just look at Olive AI and Pear Therapeutics, who sit in the same recorded failure set because each company treated evidence as optional or secondary. Olive AI, an AI healthcare automation company, faced scrutiny over claimed cost savings and efficiencies, but independent validation of these claims was limited. It’s no surprise they shut down in 2023, with their assets sold for parts, which showed the market demanded value. Pear Therapeutics developed prescription digital therapeutics (PDTs) that actually received FDA clearance. But getting payers to cover them and doctors to adopt them was another story entirely. The company’s collapse highlighted the difficulty in translating regulatory clearance into commercial success without strong, real-world evidence. Analysis of prescription digital therapeutic market challenges Regulatory compliance doesn’t guarantee market acceptance or clinical utility. It’s just the start.
Vendor Due Diligence: FDA Clearance/Approval History
If you’re a buyer, you have to do your own diligence by verifying a vendor’s FDA Clearance/Approval History. This isn’t optional. This history provides objective evidence of regulatory engagement, and it indicates if a product underwent a formal review that confirms specific safety and effectiveness claims. A vendor’s Data Room must contain complete documentation, including all FDA correspondence, clinical study reports, and QMS certifications. A Predetermined Change Control Plan (PCCP) is a positive signal for adaptive AI/ML devices because it allows predefined model modifications without new premarket submissions. What happens if they don’t have a PCCP? Every single model retrain could trigger a whole new 510(k), grinding everything to a halt. Real-World Evidence (RWE) also supplements key trial data. FDA framework for Real-World Evidence RWE strengthens both FDA submissions and payer narratives, and any vendor worth their salt should demonstrate clear strategies for its generation and integration.
Compliance as an Evidence Gate
Regulatory compliance is a critical gate, not just paperwork. This gate separates unproven claims from validated products. For health system safety officers, this means you must prioritize vendors who adhere rigorously to evidence-based pathways which includes peer-reviewed clinical validation and strong FDA clearance/approval history. A lack of adherence indicates significant risk. This lesson is critical for evaluating trustworthy AI healthcare platforms.
Frequently Asked Questions
What specific evidence pathways should we expect AI healthcare vendors to follow?
Vendors should follow established pathways like the FDA 510(k) pathway for substantial equivalence or the De Novo Classification pathway for novel devices. Both require rigorous data submission and demonstration of safety and effectiveness. Peer-Reviewed Clinical Validation is a critical trust indicator within these pathways.
What constitutes ‘Peer-Reviewed Clinical Validation’ and why is it important?
Peer-Reviewed Clinical Validation signifies independent scientific scrutiny of a product’s performance. It is a critical trust indicator within FDA pathways and helps ensure claims of revolutionary technology are scientifically substantiated. The absence of this validation defined the Theranos case, where claims lacked scientific rigor.
Beyond FDA clearance, what other documentation or principles should we look for from AI healthcare vendors?
Vendors should adhere to Good Machine Learning Practice (GMLP) principles, which guide data management and model development. A Quality Management System (QMS) compliant with ISO 13485 is also expected, ensuring consistent product quality. The presence of a Predetermined Change Control Plan (PCCP) for adaptive AI/ML devices is also a positive signal.
How can we verify a vendor’s claims regarding regulatory compliance and product performance?
Buyers can directly verify a vendor’s FDA Clearance/Approval History, which provides objective evidence of regulatory engagement. A vendor’s Data Room should contain comprehensive documentation, including FDA correspondence, clinical study reports, and QMS certifications. Vendors should also demonstrate clear strategies for Real-World Evidence (RWE) generation and integration.
