The Case Of The Half Pill: AI As Outsourcing Detective
By Louis Garguilo, Chief Editor, Outsourced Pharma
Damiano Dragone developed his perpetually positive perspective on artificial intelligence through decades of Pharmaceutical Quality work, and an early enthusiasm for the technology.
He’s always considered AI in our Quality systems and GMP environments, and how he might apply AI to his work, and working with CDMOs.
“We are inundated with more information than we can effectively use,” he says. Take, for example, the case of the half pill.
Chasing The Root Cause
The initial sponsor reaction is to place the problem’s origination at an external partner’s production or packaging line.
Hold on, says Dragone.
“The sponsor cannot treat this as someone else’s issue to investigate,” he says. “Operational activities are distributed across organizations, but accountability for product quality remains clearly defined at the sponsor, and that is not transferrable.”
Sponsor, CMO and any other involved parties must investigate both separately and together as a unit, come to an agreement on causes, and then corrective and preventive actions.
This “specific” half-pill complaint is not treated as an isolated deviation; tentacles to the cause can be widespread.
- Has a similar complaint occurred before, perhaps on another line, for another product or another customer? Or from product at a CDMO or location?
- Was there in fact a failure in tablet manufacturing? Did the packaging line’s vision system miss the defect?
- Was the system weakness missed during internal or client audits, or regulatory inspections?
“The objective is to solve today's deviation across the supply network as needed, and strengthen tomorrow's decision-making,” Damiano says.
Multiply the half pill scenario across thousands of our drug companies and partners, millions of product types, dosage forms, facilities, suppliers, audits, investigations, complaints … and the real scale of the challenges in our modern industry becomes obvious.
Considering all this, then, effective AI assistance has arrived just in time to help us raise the bar of quality assurance, safety, and to “add to our human creativity while we increase the complexity of our work.”
Dragone, who today heads his own consultancy, says AI’s practical value is in helping dispersed teams connect data from various areas; aid in the detection of patterns and anomalies; assess immediate and broader impacts; and allow for the response before a defect becomes a recurring failure.
Vigilant quality-trained human beings who know how to “talk to the machine,” as Dragone often refers to AI, is a drug-safety improvement at a welcome time.
Building The AI Model
AI is not magically connected to every machine or potentially relevant database to help solve the half-pill anomaly.
There’s hard work to accomplish first: individual professionals and organizations collectively must make available and define “the right inputs, then an AI model, and determine the information that model trains on,” while outside the AI model, establish rigorous human quality oversight.
In the half-pill case, this means drawing from all:
- known data on complaints and past remediations
- Quality systems and procedures
- manufacturing and packaging trains, and visual-inspection procedures and data
- relevant SOPs, validation protocols/records/ prior CAPAs
- historical market/consumer/patient data
For their part – and it remains a vital component – information technologists (IT) and AI specialists assist biopharma SMEs determine how to connect, structure, validate, secure, and maintain systems and models. Note: the biopharma SMEs are essential in AI training.
Again, for our half-pill case, for example an AI model might be trained on XX years of market complaints and deviations. The AI model is provided access to a CDMO’s production-line history, packaging data, vision-system records, manual inspection procedures, training records, and validation documentation.
This suggests today’s external partners must be open to this coordination and cooperation.
If blister inspection at a CDMO sometimes depends on trained operators rather than an automated vision system, that must be included in the data; otherwise, the AI model may miss a material part of the process and any problems.
Once your AI model is operating, it can near instantly answer pointed questions from an investigation team.
- Is the half pill linked to a specific product, line, shift, supplier, operator practice, inspection method, or validation weakness?
- Has this type of defect ever appeared before, if so when/where and what was the resolution to the problem?
The AI will make connections faster than any human team could by searching disconnected systems. The AI and trained quality professionals together push this to the optimal productivity.
AI Needs You
Dragone’s ideal for AI in the drug supply chain has been inspired by the potential for AI and Human interaction.
SMEs bring experience to this new intelligence, and again, we must learn how to talk to the machine.
Dragone says AI models suggest probable root causes, propose CAPA options, highlight missing evidence, and can draft language for investigation reports. But Human QA and SMEs critically review all output. He’s clear not to turn AI into “the decision-maker.”
This, he says, is “ethical behavior.”
Oversight never ends – on the ultimate behave of patients – no matter the plethora of data the AI derives its intuitions from.
“Human oversight requires understanding inputs and outputs, documenting and challenging the rationale behind AI-supported recommendations.”
If the AI proposes three root causes and the investigation team confers, the record should explain why. If the AI-Human team accepts, rejects, or modifies a proposed CAPA, the rationale should be traceable.
On the outsourcing front, sponsor and all partners must be able to demonstrate how an AI-supported recommendation, and any final decisions, were scientifically and procedurally justified – how they fit within our cGMP environments.
When we solve for the “half-pill event” as described above, our professionals become less pressured and better at their job.
AI isn’t the destination; better decisions are. AI and biopharma professionals throughout the supply chain enter together into a new “logic” and creativity ultimately on behalf of patients.
-----------
Also see: Don't Fear The Machine – A Biologist Explains AI In The BioPharma Industry