Pharmaceutical Technology asked Jeff Elton, PhD, CEO of ConcertAI, about the role technology plays in performing CAPAs.
Performing corrective action and preventive action (CAPA) activities is often necessary to investigate a manufacturing deviation and prevent it from reoccurring. Pharmaceutical Technology asked Jeff Elton, PhD, CEO of ConcertAI, about the role technology plays in performing CAPAs.
PharmTech:How can technology be used to prepare and perform CAPAs? In investigating deficiencies, and/or developing action plans?
Elton (ConcertAI): Technology and AI [artificial intelligence] can, and should, be used to create and maintain effective CAPA plans, in clinical trials, real-world evidence (RWE), and medical devices ecosystem. In FDA’s three draft guidance documents published at the close of 2021 on the use of RWE, a number of key technical characteristics were noted as foundation. Upon integration of those across all technical environments interacting with study data, technologies can allow a study dataset to be considered a ‘package’ of data that includes details on its source, source relevance, reliability, etc. These same technologies also allow surveillance of data for anomalies and errors can assemble data to a specified cohort and create registries and comparative data sets assuring consistency with the study plan and any preliminary analyses.
PharmTech:How can technology be used to track the progress of CAPAs and the impact of changes on products and processes?
Elton (ConcertAI):GxP and other related processes can now all be electronically captured and maintained. This means that the technology development process for a dataset, technology enablement SaaS solution for clinical trials, or digital medical device can have the history and documentation available in a digital only version that further unpins critical certifications such as 21 Code of Federal Regulations Part11 that stands in assurance of reliability and reproducibility. Consequently, the prevention plan and any required corrective actions are fully available and integrated. We see these automation tools as improving accuracy, reducing manual effort, and fundamentally lowering risks.
PharmTech:How can technology be used to anticipate potential problem areas that might need a CAPA plan?
Elton (ConcertAI): A GxP plan is documented evidence that the system is fit for its intended use. This answers the question of ‘what will this system do?’ and ‘Is there evidence the system performed as expected?’. This underlies the CAPA, as we can automate the surveillance of performance relative to these expectations. The full lineage of data and metadata capture of performance would allow remediation of any failures to perform to these documented expectations. While some of this can be automated in design and capture, we see that these will advance towards intelligence systems where the deviations are identified by AI and ML tools that recognize departures from expectations, designs, and defined boundary conditions. While technology does not replace the need for human interventions whenever issues arise, it may certainly help prevent future problems by creating an ecosystem that is constantly evaluating itself against its desired inputs at various timepoints throughout the production and administration of life-saving care.
PharmTech:What kinds of data can be compiled during a CAPA, and how can these data improve product quality?
Elton (ConcertAI): Multiple types of data can be collected, including a validation plan, traceability matrix, conformance to design specifications, test case summary reports, GxP risk assessment outcomes, automated functional end-to-end test results, etc. In data-centric solutions, this can include confirmation of data ingestion, data mapping, data transformation—all aspects determining initial quality for data being used for any RWE analyses. One aspect that receives less attention is the interoperability and perpetuation of errors across multiple systems. Therefore, the more integrated the CAPA across systems or the more integrated the system across use-cases with a single CAPA the more robust the plan will be. The creation of the initial data also allows the automated assurance that the performance in-use is consistent with the data compiled during the CAPA development.
PharmTech:In what other ways can data found during a CAPA be used for pharmaceutical development and manufacturing?
Elton (ConcertAI): Anything that contributes to the reliability of the system or the data captured by a system contributes to clinical development. Often the data for a clinical trial or ongoing post-approval surveillance as part of a Post-approval Safety Study can include aspects of data generalizability, completeness, and reliability. The higher the confidence levels are on all of these parameters, the greater the ability to accept confirming data or the more quickly an action can be taken on less data.”
PharmTech:What kinds of data do regulators request from CAPA plans?
Elton (ConcertAI): This varies by the product, technology, or solution, but broadly:
- Product or QMS [quality management system] non-conformities—this is where automated GxP documentation, validation, and automated monitoring relative to the plan and conformance data can add enormous value.
- Internal bug reports (e.g., by developers)—again automated and exhaustive documentation that is maintained progressively for all aspects of the system and for documentation of conformance to expected performance can deliver a highly robust CAPA.
- Audit findings—some audits can now be run as a script. Others will be done on demand based on anomalies or where the criticality of the decision may be exceptionally high.
- Post-market surveillance findings, including trends—this is where the new RWE guidance for registries can change how this is done, increasing the scale of data used, allowing for rapid detection of anomalies, etc.
- Management review findings, including trends—This should be part of product development processes, acceptance process, and general releases. As we move towards digital and automated testing, it can become more ‘real-time’ versus period, discrete, or event-driven as it often is today.