#FutureReadyHealthcare
The safety of medicinal products is continuously monitored through various sources such as clinical data, electronic patient records, call centers, medical literature, and patient support programs for reporting Adverse Events (AEs). To handle the growing volume of AEs, a new and innovative solution is needed that prioritizes serious and clinically important cases. With industry already using technology to reduce manual labor, ground work is being laid to apply better and more efficient solutions. One approach is to implement a tiered case processing approach that groups and channels AEs based on complexity, relevance, and seriousness with the help of technology such as analytics and Artificial Intelligence. This approach when combined with an effective organizational design can improve efficiency, enhance risk management, and prioritize high-risk cases in pharmacovigilance systems.
Streamlining Case Management: The Role of Tiered Case Processing in Pharmacovigilance
The monitoring of the safety of medicinal products is an ongoing and important process throughout their lifecycle. The past decade witnessed a significant increase in the availability of sources for reporting Adverse Events (AEs), providing valuable insights into the safety profile of these products. Market Authorization Holders (MAHs) are now able to collect, evaluate, and report AEs from a variety of sources, including clinical data, electronic patient records, call centers, medical literature, and patient support programs, among others. This expansion of data sources has led to an increase in the number of Individual Case Safety Reports (ICSRs). The volume of Individual Case Safety Reports (ICSRs) is reaching an all-time high. This increase in data presents a challenge for safety operations but also an opportunity for the life sciences industry to optimize resource allocation and develop more efficient pharmacovigilance strategies to manage and utilize these insights.
In 2021, the number of ICSRs related to suspected adverse reactions collected and managed in EudraVigillance2 were 3.5million
Source: 2021annual report on EudraVigillance for the European parliament, the council and the commission.
From 2010 to 2022, adverse events reporting1 has increased at a CAGR of 9.9%
Source: FAERS public data base - AE reports from 2012 to 2022
The Potential of Tiered Case Processing in Pharmacovigilance
To effectively scale, a systematic approach is needed that efficiently allocates resources to serious and clinically important cases, while still allowing for a thorough assessment of all reported AEs. One solution is to group and channel AEs based on complexity, relevance, seriousness, and a variety of other factors. Taking this a step further, leveraging technology such as analytics and Artificial Intelligence based predictors can accelerate the process of systematically codifying cases right from the point of identification. This technology-enabled, tiered case processing approach is the key to the next wave of Pharmacovigilance ICSR case processing and automation, paving the way for the future of pharmacovigilance. This technology-enabled, tiered case processing approach is the key to the next wave of Pharmacovigilance ICSR case automation, paving the way for the future of pharmacovigilance.
This approach can potentially help to address four key challenges faced by pharmacovigilance systems, including:
■
Managing large volumes of data
■
Identifying and addressing potential safety concerns
■
Allocating resources effectively
■
Improving communication
Several potential benefits with tiered case processing in pharmacovigilance:
■
Improved efficiency: By case prioritization in pharmacovigilance, teams can allocate their resources more effectively and efficiently, potentially reducing the overall workload.
■
Enhanced risk management: Tiered case processing can help identify and address potential safety concerns more quickly, as higher-priority cases are reviewed and assessed more promptly.
■
Greater focus on high-risk cases: By prioritizing cases based on their potential risk, pharmacovigilance teams can focus more closely on those cases that pose the greatest potential risks to public health, allowing them to identify and address any safety concerns more quickly.
■
Enhanced communication: Tiered case processing can facilitate better communication between different levels of review and between different teams within the pharmacovigilance system, helping to ensure that all relevant information is considered when making risk management decisions.
How is the industry leveraging technology today to reduce the need for manual labor?
Currently, AE/ICSR case capture, processing, and medical review depend extensively on manual efforts. Technologies and corresponding examples of applied usage in PV case processing have started to be implemented, which has a lasting impact on the reduction of manual effort in the case capture and processing of the data.
These technologies and their application include:
Technology
■
Optical character recognition (OCR)
■
Robotic Process automation (RPA)
■
Natural language processing (NLP)
■
Natural language generation (NLG)
■
Analytics with artificial intelligence (AI)
Examples
■
Data extraction of AE information from flat files
■
Automation of data population into E2B R3 file data elements
■
Detection of relevant terms such as AEs or medications using medical ontologies
■
Creation of patient narratives from E2B data elements
■
Evaluation of common trends and recommendations applying confidence information
As a case in point, some Health Authorities have leveraged technology to enable AE reporting. For e.g., the MHRA modified its Yellow Card Scheme by significantly increasing the amount of AI tied to it in response to the volume of cases reported due to COVID-19. Similarly, the French Ministry of Health has implemented adverse drug monitoring technology on a national level to improve the safety of the country's vaccination program. The Medication Shield technology, rolled out by Synapse Medicine in collaboration with regional PV centers and the French National Agency for Medicines and Health Products Safety, was designed to increase the safety of COVID-19 vaccines in France by managing real-time adverse drug reaction signals on a pharmacovigilance reporting systems.3
Machine Learning (ML) has had limited use cases demonstrated so far in PV. However, the comfort generated through applying more fundamental automation hitherto has opened the industry to the opportunities that ML can provide. ML serves as the basis of tier-based case processing, which leverages analytics coupled with AI-based confidence associated with some or all patient data, signal data, external data, and even meta data to derive a recommendation. Case classification (profiling) based on historical evidence of the ICSR against a benchmark of similar known cases would expedite the clearance of cases to channel through the most well equipped resources, resulting in improved pharmacovigilance data management and quality while delivering time and cost reductions.
You can explore further in this Indegene report and know how industry is moving towards pharmacovigilance automation.
Areas of opportunity and how tiered case processing can be applied
Most recently (2021), the Uppsala Monitoring Center (UMC) 4 has developed an adjustment to analysis in signal detection in pharmacovigilance known as disproportionality which has been a common analytic approach toward post-marketing safety surveillance. However, this model works when looking at a single term and does not consider context provided by the reporter. A more recent analysis, called vigiGroup, is a novel clustering method to complement disproportionality analysis by grouping reports based on their co-reported terms, which increases insights. Cluster analysis such as this can enable data-driven discovery in PV and help identify adverse drug reactions despite differences in manifestation.
Case profiling/clustering applying analytics and recommendations from AI can further streamline the way cases are routed initially to dedicated resources best equipped to process them. This allows for a distribution of case types into a tiered case processing approach utilizing customized automation and human intervention playing a vital role in ensuring sufficient case evaluation and assessment of risk mitigation on pharmacovigilance. This approach can ensure a high touch effort where the machine’s confidence is low, and case complexity is high; therefore, having a more specialized team manage that case. This would be countered by cases where the machine-based confidence is high, and the cases are well profiled previously; therefore, minimal human touch or even no human touch of the case is required.
Data clusters in PV frequently occur in a conventional process, given that drugs are largely treated for specific indications creating common groups which can include information about the patient data such as seriousness, label listedness, concomitant diseases, medications, and causality to name a few. Case type classification typically also has merit given that trends on data completeness are difficult in literature or investigator-initiated research studies; therefore, these case types can be classified. Functional data from processing cases can help guide how to handle different case types based on the complexity of laboratory data, specific complex diseases, or manual intervention evidence.
Tiered Case Processing considers the following parameters:
■
Expectedness of the case
■
Seriousness of the case
■
Changes in severity
■
Abnormal secondary findings
■
Duration for which product is in the market
■
Overall risk profile of the product
Based on these parameters, subsets can be largely categorized as high-risk case reports and low-risk case reports; serious and non-serious cases; SUSARs (Suspected Unexpected Serious Adverse Reaction) and SAEs (Serious Adverse Event); AEs of special interests; pregnancy cases; etc. In specific scenarios, e.g., cases that are coming from commonly prescribed population groups and the AEs that are not serious and listed or already disqualified from prior signal analysis might be categorized as low-risk cases.
Once case profiling is complete, and by applying the AI-based recommendation engine, the cases are routed to a relevant qualified resource who performs further processing and medical review for the sufficiently supported risk of the cases. This approach follows the below considerations:
High-risk/Touch reports:
■
Adverse event(s) which are serious; suspected unexpected serious adverse reaction (SUSAR) events; cases with multiple events with a causal relationship/positive dechallenge or a recently approved drug with high instances of reported events; events affecting population who are not exposed to the product as per product label etc. will qualify for the high-risk category reports. For example, the use of Etanercept – a tumor necrosis factor receptor, may result in worsening of congestive heart failure or induce new-onset congestive heart failure, which is a serious or suspected unexpected serious adverse reaction (SUSAR) event. These kinds of reports will need an in-depth analysis and medical assessment.
■
Less structured cases like handwritten documents or reports with complex laboratory findings and varied diagnosis; data from non-interventional studies; clinical trial cases in pharmacovigilance, literature searches, etc will be handled by specialized teams of experts in conjunction with ML/AI technique usage.
Low-risk/Touch reports:
■
These include non-serious, lack-of-efficacy (LOE) reports, single related event(s), events that are already identified and listed in the product label, events affecting known age groups/populations, etc. For example, for Etanercept, the occurrence of Injection site reactions and Pyrexia are some of the very common adverse events which are listed and known to affect all age groups/populations. These kinds of reports require minimal human intervention and can be handled as one touch review.
■
Highly structured cases from electronic sources require minimal handling of data, such as XML-based files, license partner cases, and health authority (e.g., EVWEB) cases where the possibility of novel data is limited. A considerable amount of data can be mapped to fields in database and thus reducing human involvement for auto-populated fields.
■
Cases where historically common data points have all been received and handled with a high degree of quality.
Pharmacovigilance process flow chart: Depiction of a tiered case processing
Steps towards operational preparedness for successful adoption of new approach
Organizational design and resource qualifications are at the core of an effective and future ready Tiered Processing Operating Model. The qualifications of case processing and medical review resources in years of experience, quality outputs, and demonstrated medical aptitude should be much higher for high touch case types as compared to a resource qualified to process simple and low touch cases.
Life sciences organizations should consider the following levers to achieve the frictionless application of tier-based case processing:
■
Resource qualification
■
Roles and responsibilities
■
Quality
■
Training
■
Change management
■
Technical support
■
Documented processes/SOSs.
In a tiered approach operating structure, resources trained on simple case types will only receive those cases with highly predictable data. They can expedite cases more quickly than high-touch case types. Resources in this group should be well-trained in understanding the benefits of machine-based techniques like OCR, NLP, and NLG in conjunction with RPA/AI/Cognitive measures in highly structured cases. This will be highly effective and substantially reduce the turnaround time and cost of operations.
Training programs should be customized to consider the dependency on machine intervention and recommendations. Training curriculum for high-risk / touch reports should focus on medical judgement, complex medical situations, and unique scenarios, whereas low-risk cases focus on quality and consistency. A fully integrated training team well versed in change management and transformation initiatives ensures effective onboarding to technology, roles, and effective leverage of technology plus human roles are applied.
Assessment of productivity and quality is imperative to determine the success of the change management initiative by managerial staff. The change champions partnered with technical SMEs can ensure the effective adoption of this tiered processing model by supporting the training team, QA / compliance team, technical, and operational teams.
4 considerations to ensure proper implementation of technology and organizational readiness
■
Effectively documented methods to apply analytics and insights to drive case cluster analysis consistently.
■
Technology that can perform proper triaging of cases to various teams and a defined model to determine the appropriate classification of cases to route to the correct team. This requires a competency matrix to map skills to complexity in case types.
■
The ability of the organization and technology to apply AI and demonstrate the innate confidence the system has in presenting a recommendation for triaging as well as in the resulting data and the ability to present the data from clusters for proper data handling by applying the relevant insights.
■
Proper training, change and team management to prepare team members and ensure appropriate tier case handling and tiered teams are allocated effectively.
Make your move
Adopting a case profiling and tier-based case processing approach requires organizational readiness to support the adoption of this up-front, machine-enabled process. The initiation of a project at any organization will benefit from a pharmacovigilance roadmap approach that provides an opportunity to incubate this in a pilot model, allowing for demonstrated evidence of an effective classification model and the readiness of an organization to scale this model to a larger team and broader use cases.
Preparing the organization with a clearly defined use case, desired business impact, and expected outcomes and backing that plan with a change-ready organization is vital in the planning process. During the execution, a transformation team supporting change, training, quality, and pharmacovigilance compliance will ensure the effective adoption of this program.
Lastly, access to technology that can apply an analytics engine, leverage multiple automation and recommendation engine plus present data and ML/recommendation insights is imperative to the success of a program such as this. Leveraging a partner that understands the needs of scaling for transformation and implementing pharmacovigilance technology solutions through a roadmap approach will increase the probability of a successful tier-based processing initiative in PV operations.
References
1.
FAERS public data base- AE reports from 2012 to 2022
2.
2021 Annual Report on EudraVigilance for the European Parliament, the Council and the Commission
3.
COVID-19 vaccination campaign safety : France rolls out the “Medication Shield” of Synapse Medicine, 2021. Accessed on 16 August 2022.
4.
VigiGroup in Action – New Method Identifies Groups of Related Reports,2022. Uppsala Reports. https://uppsalareports.org/articles/vigigroup-in-action-new-method-identifies-groups-of-related-reports accessed on 16 August 2022.
5.
Gertrud Brunlöf, Carina Tukukino, and Susanna M Wallerstedt. Individual case safety reports in children in commonly used drug groups – signal detection. BMC Clin Pharmacol. 2008;8:1.
6.
E2B(R3) Electronic Transmission of Individual Case Safety Reports Implementation Guide — Data Elements and Message Specification(version3.01). https://www.fda.gov/media/81904/download accessed on 19 August 2022.
7.
Patrick Buckner, Deven Atnoor, Ph.D, and Mayank Thakkar. Conducting end-to-end pharmacovigilance workflows using AWS technologies. AWS for Industries.2020.
8.
Artificial Intelligence and Machine Learning in Pharmacovigilance – current use case and future opportunities. 2022 https://www.avenga.com/magazine/artificial-intelligence-machine-learning-pharma/ accessed on 19 August 2022.
9.
Peter Arlett, Sabine Straus, and Guido Rasi. Pharmacovigilance 2030; Clin Pharmacol Ther. 2020; 107(1): 89–91.
10.
How has pharmacovigilance changed since the COVID-19 pandemic? – posted by Kathryn Taylor, 2022. https://www.quanta-cs.com/blogs/2022-6/how-has-pharmacovigilance-changed-since-covid-19-pandemic accessed on 22 August 2022.
11.
Pharmacovigilance in Times of COVID-19, 2021.
https://arithmostech.com/pharmacovigilance-pandemic-covid-19/ accessed on 01 September 2022.
https://arithmostech.com/pharmacovigilance-pandemic-covid-19/ accessed on 01 September 2022.
12.
Pharmacovigilance Plan of the EU Regulatory Network for COVID-19 Vaccines, 2020. https://www.ema.europa.eu/en/documents/other/pharmacovigilance-plan-eu-regulatory-network-covid-19-vaccines_en.pdf accessed on 01 September 2022.
13.
Adverse event reporting: Power of AI enabled cognitive case processing and process automation by Prajeesh Pillai, 2021. https://tataelxsi.com/storage/insights/January2021/MjOH1I55BgBBhz2PpJmx.pdf. 2021 Nov 24
Authors
Vladmir Penkrat
Dr. Shubha Rao
Geethashree HS
