The Role of AI in Transforming Pharmacovigilance in Clinical Trials
Emerging technology and patient centricity are currently some of the biggest topics surrounding clinical research. Here we discuss some of the ways in which artificial intelligence (AI) can be utilised to transform key steps in clinical trials to increase clinical trial efficiency, improve patient safety and the patient experience.
Enrolment:
Among the many reasons why clinical trials are terminated, the single biggest contributor is insufficient patient intake, accounting for 55% of trial terminations [1]. This is followed by lack of efficacy which only accounts for 15% of terminated trials [1]. Doctors can recommend clinical trials they are aware of to their patients, but otherwise the onus is often on the patients to find clinical trials themselves. This could be by looking through government websites, medical research charities and seeking out patient recruiters for clinical studies. However, this can be overwhelming and time-consuming, especially for those who are unfamiliar with these resources and the complicated medical jargon.
Finding suitable candidates with comparable confounders and underlying conditions from the very start of a trial is also important to identify safety issues as soon as possible. There is also a need to increase focus on recruiting more diverse patient populations in clinical studies in order to reduce the disparity in health outcomes between different patient ethnicities. Several barriers that prevent ethnic minorities from participating in clinical trials have been identified – one of them being a lack of clinical trial awareness [2]. AI tools can improve the enrolment process for both the patient and clinical researchers, and to help mitigate these current issues in healthcare.
For example, natural language processing (NLP) is being used to extract and analyse patient health records in the US to find suitable study candidates for clinical trials [3]. The software can analyse structured and unstructured data, including doctor’s notes, pathology reports and other medical data in free-text form. All extracted data is used to form a unified patient graph which can then be matched against the eligibility criteria of ongoing trials. This level of precise automation allows eligible candidates to be identified within minutes rather than months and relies less on the patient to find suitable trials for themselves. With such high processing rates, more potential candidates can be identified and contacted, increasing clinical trial awareness and recruitment outreach to more diverse populations.
Monitoring:
A requirement for patients participating in clinical trials is having to regularly visit the research site for check-ins. These visits have traditionally been considered essential to enable investigators to assess the effects from the investigational medicinal product (IMP) and to monitor the patient’s health. However, the frequent travelling can put a strain on the patient’s time and potentially finances, depending on the methods of travel reimbursement. This has also been highlighted as a barrier for ethnic minorities participating in clinical trials [2]. The disparity in socioeconomic factors between different ethnicities means that the burdens associated with clinical trial participation are more significant for minority groups. These burdens not only deter people from entering clinical trials, but also increases the risk of patient drop-out during the study.
To lessen the burden on the patient, patient monitoring may be done at regular, but less frequent, check-in appointments. This then poses a different issue of patient safety. Any asymptomatic changes in a patient’s health will likely be undetected until the patient attends their next check-in appointment and the clinician performs their routine tests, thus delaying the detection of potentially harmful developments and any necessary medical intervention.
With AI, remote patient monitoring (RPM) is now becoming more common in clinical trials. This means that physiological measurements can be taken continuously in real time from patients through the use of wearables and biosensors. The information can be collected without any patient input and then analysed by AI-enhanced software. Implementing this technology helps in a number of ways: patient participation in a clinical trial becomes more convenient, more data can be collected which allows more conclusions to be drawn from a study, time and resources are saved for the investigators from analysing such large volumes of data, and patient safety is improved through continuous tracking.
Wearable devices may be used to track physiological changes in the user [4]. The technology can detect and flag key physiological changes which may be indicative of disease progression and predictive insights on the user’s health can be made so that pre-symptomatic medical attention can be given.
As an example of industry and service provider collaboration, Chugai have made use of the Biovitals platform to develop a digital solution for objectively assessing the pain associated with endometriosis [5]. By using biosensors and AI-based algorithms, the severity of pain can be quantified, thus reducing the variability that is commonly seen with traditional methods of pain assessments. Although the validity of this solution is still being investigated in a study, this development demonstrates the novel ways in which AI can be utilised to improve the way clinicians can remotely conduct difficult and subjective assessments.
The use of technology is not just limited to the monitoring of routine physiological changes. Speech signals can be processed using an app to detect changes in neurological health using auditory phenotyping [6]. Samples can be analysed at any time, providing real-time updates of neurological disease progression.
These RPM methods, aided by the use of AI, can also be applied to clinical trials to not only improve the patient experience, but to also vastly improve patient safety by minimising the risk of adverse events and reactions, and investigators would be able to collect more valuable patient data.
Adherence:
Good adherence rates in clinical trials are essential in obtaining reliable data which valid conclusions can be drawn from. For many studies, once patients are enrolled, they will receive the study drug or placebo to take home. They will often be asked to maintain a patient diary to document when the study drug has been taken. This process is heavily dependent on the patients and is susceptible to human error.
A reason for non-adherence in a clinical trial may be as simple as the patient forgetting to take the medication. Even in instances where the patient has remembered to take the medication, and has done so correctly, they may forget to document this in their diary. When the patient re-visits the research site for check-ins, this diary will be checked and if there are blanks, the investigator will have to rely on the patient’s memory of events. AI can be used in different ways to maintain patient adherence remotely in clinical trials.
One way in which this has been done is by using visual phenotyping to assess adherence, acting as an interactive medical assistant which patients can use through their smartphones [7]. Patients will be able to take a video of themselves taking the medication and from the video, the software will be able to confirm that the correct person has taken the correct medication.
Non-adherence driven by fear and misunderstandings about the drug is also an issue in clinical trials. AI-based conversational chatbots can be used to address this. They can be used to perform daily check-ins to take the medication and can also use NLP to respond to any queries and concerns a patient might have about the drug. The software could then pull up the relevant information for the patient. By making the educational information so easily accessible and personalised, patient non-adherence driven by fear of the treatment can be tracked and resolved.
AI and machine learning can also be used to predict which patients are at risk of non-adherence, thus allowing the trial coordinators to proactively intervene in such cases to maintain high adherence rates.
Pros of AI in clinical trials:
· Improves enrolment: automated processes can increase clinical trial enrolment rates, find optimal candidates and increase outreach to minority groups.
· Efficient patient safety monitoring: any issues a patient may have during a trial can be predicted or immediately detected through real-time RPM and, therefore, can be resolved sooner.
· Improved compliance: automated reminders can be given to patients, adherence can be tracked and those at risk of drop-out can be detected, prompting intervention.
· Patient-centric solutions: making the clinical trial processes more convenient, personalised and informative for the patient.
· Improved efficiency overall: With more processes becoming automated, this will give employees more time to focus on other more urgent tasks.
Cons of AI in clinical trials:
· AI technology can be expensive to begin implementing.
· AI technology could have a steep learning curve for employees.
· Patient-centric processes involving AI require patients to have access to smart technology which they can competently use. This may be an issue for older populations and those from lower socioeconomic backgrounds.
· Data protection and security concerns surrounding third party organisations which have access to patient information.
· AI is not perfect, so processes should not be built on the assumption that the technology will be error free.
· AI interventions in clinical trials may inflate efficacy through optimal study populations and dosing, thus widening the gap between drug efficacy and real-world effectiveness.
Although there are many positives to come from implementing AI technology in clinical trials, there are important issues that need to be overcome as outlined above. AI interventions would undoubtedly increase the speed and efficiency of clinical research which could help bring new life-changing treatments to the market sooner, but the biggest problem would be whether the observed outcomes from such clinical trials will be reflective of real-world situations.
Points to consider:
· The observed efficacy in clinical trials will likely be inflated with such strong technology interference. This may require increased numbers of post-authorisation studies to ensure safety and efficacy remain as expected.
· Observed clinical outcomes could be reflective of drug effectiveness if the same AI interventions are also used by real-world patients.
o However, accessibility to such technology in the real-world will need to be considered, especially for those from lower socioeconomic backgrounds.
· If pre-symptomatic changes in patient health in a clinical trial are always detected early enough for medical intervention, the number of adverse events may be reduced and gaps in the drug safety profile may be formed. This may need to be reflected in post-market risk management plans.
· If non-adherence always prompts AI interference and reminders, genuine reasons for non-adherence may go undetected (e.g. IMP formulation is unpleasant or difficult to swallow).
· Additional regulation and understanding around the technology used in clinical trials will be essential.
In this rapidly evolving digital era, the pharmaceutical industry is constantly looking for new and innovative ways to make clinical trials more efficient, accurate and streamlined for patients. It seems that AI technology will be fundamental in transforming clinical trial processes however, from the points raised above, it is clear that there are complexities in the discussions surrounding science and ethics that arises from using AI in clinical trials. For it to be used successfully to improve the clinical trial processes, patient safety and healthcare for the general population, it is paramount that careful considerations are made on the way AI technology is regulated, there is a clear understanding of what contexts it can be used in and when it is best to use human judgement instead.
References
1. Enrollment issues are the top factor in clinical trial terminations
https://www.pharmaceutical-technology.com/comment/reasons-for-clinical-trial-termination/
2. Increasing Diversity in Clinical Trials: Overcoming Critical Barriers
https://www.sciencedirect.com/science/article/pii/S0146280618301889
3. Deep 6 AI
4. Biovitals by Biofourmis
https://www.biofourmis.com/biovitals/
5. Chugai and Biofourmis Partner to Develop an Objective Assessment of Pain Using Digital Technology
https://www.chugai-pharm.co.jp/english/news/detail/20200722113000_744.html
6. Aural Analytics
7. AiCure
Drive Phase PV is a pharmacovigilance consultancy and provider of pharmacovigilance services, supporting patient safety throughout the life-cycle of our clients' products. Contact us to discuss developing your PV processes and technology requirements. Whether via SOP development, database implementation or migration we deliver a tailored packaged to seamlessly fit to your company structure and strategy.