Digital Wisdom: AI as a Regulatory Imperative
From Data to Wisdom: The Ethics of Clinical Insight
The Data, Information, Knowledge, and Wisdom (DIKW) hierarchy models the transformation of raw facts into actionable insight. Data (raw figures) becomes Information (processed data with context), which is refined into Knowledge (applied information), finally maturing into Wisdom (applied, ethical judgment).
In the world of clinical research, we spend a disproportionate amount of time on the Data and Information parts, and we get to claim Knowledge through a successful Marketing Authorisation Application. This program touched on Wisdom, i.e., what do we do if the data tells us that an investigator could be cheating?
The program was part of a wider company agenda to build trust with society. Our client had two goals: to mitigate the risk of fraud in clinical research data for the organisation and to demonstrate leadership within the organisation for the application of advanced analytical techniques (in this case, predictive and analytical AI and ML) to the company’s operations.
Why Traditional Data Filters Fail to Catch Fraud
Fraud in clinical trials is something of an open secret: we know it goes on, but little is done to prevent it beyond removing affected data when it becomes obvious. The prevalence and distribution of fraud across the industry are not known, while the incentives are clear, and the regulators have started to realise this. Standard edit checks and manual reviews of clinical data look for errors or missing data. The fabrication and falsification of data, however, create data that looks deceptively normal, allowing it to pass through traditional filters. Our approach was to apply a combination of domain-derived logic and forensic accounting principles to the clinical dataset.
An “Always-On” Solution Sitting on Top of Existing Data Flows
At the scale of a multinational drug development portfolio, our approach enabled:
Site-Level Anomaly Detection: Instead of looking at individual data points, AI models compare an entire site's data against the broader study, therapeutic area, or regional datasets.
Continuous Portfolio Monitoring: Instead of "point-in-time" audits, the system runs continuously, flagging deviations in real-time.
The system identified site data entry errors not caught by standard checks at a number of sites, flagging identical records under different patient identifiers and other unusual data patterns (e.g., “too perfect” data). With the ability to identify studies and sites producing anomalous data, the client can decide if further investigation, e.g., a site audit, is warranted based on the signal. It is also possible to view anomalies by geography, therapeutic area, program, or even development stage to look for systemic issues using different views.
AI as a Regulatory Imperative
The system gave the client a ready-made, direct response to the US Department of Justice, which has recently communicated: a) that they expect Life Science companies to use data analytics to detect fraud; b) that if an AI capability for anomaly detection is not applied instead of random spot checks, this may be framed as “reckless disregard”; and c) that, with the establishment of the HSU in 2025, clinical data integrity is now a topic of focus and interest.
