Our results show that, as postulated, we were able to leverage digital tools, such as the EHR, to improve the timeliness, quality, and uniqueness of safety data collection leading to “high definition” actionable safety insights that are unlikely to have been attained otherwise.

There were several strengths to our methodology. First, we were able to show significant improvement in the timeliness of the exchange of AE information. Using traditional methods, the average time from AE occurrence to receipt of follow up was 106 days. In our PoC, we were able to demonstrate the 4 completed interactions (2 completed interactions including follow-up and 2 completed interactions that indicated the event was not an AE) took about 2 days on average, an overall time savings of 97%. Even when looking at just the 2 completed interactions the time savings was still 96% (10–12).

Second, we were able to demonstrate qualitatively that the resulting data, from the initial questions and bespoke follow-up, was significantly better than data obtained through the current process. In our study, both interactions (for which we received confirmation of the AE and answers to the initial and bespoke follow-up questions) were of sufficient quality for proper assessment of the event. For comparison, the study by Durrieu et al. found that only 12.7% of reports were well-documented (15). Additionally, our study demonstrated the value of PVEs having the ability to engage in near real-time, “active discussion” with HCPs. For example, the initial questions were focused, drug-event pair specific which allowed the HCP to provide the exact information the PVEs needed to perform an initial assessment as well as formulate highly focused follow-up questions to collect any missing information. With respect to the follow-up, in the two use cases both HCPs were asked about potential accidents or injuries that could have caused the bleeding, in one instance the HCP confirmed no accident whereas the other HCP confirmed the patient did have a recent injury, however the location was not in the area of the bleed. Additionally, the contextualized discussion led to unique insights that are contextualized to the specific drug event pair. For example, collecting history of renal disease to not only rule out an alternative etiology of the event but also assess proper drug elimination from the body. This highly focused exchange of information isn’t possible with the current process where PVEs are limited to whatever information the reporter wants to provide with the initial report, without influence from safety experts, and is further limited by the time it takes to collect follow-up, which can introduce recency bias into the additional data provided, if any.

Finally, our methodology addresses barriers to HCP participation. First, HCP burden was minimized by leveraging existing data from structured fields in the EHR and asking only the highly focused relevant questions that needed to be asked from a PVE perspective, specific to the exposure-outcome of interest. Secondly, our method raised awareness by providing summary data, such as FAERS, back to the HCP. Thirdly, it facilitated communication with the drug manufacturer, providing the HCP with the opportunity to engage with the PVEs.

There were limitations to our study. Unfortunately, with only 9 trigger events identified in 9 months and with only 4 completed interactions (2 completed interactions including follow-up and 2 completed interactions that indicated the event was not an AE), the number of completed interactions was lower than expected. It should be noted however that no trigger events were identified in the first 5 months, with all nine trigger events occurring in the last 4 months of the study with 6 in the last 6 weeks. This suggests a potential lead time may be required to recruit enough HCPs to reach a critical mass to enable this capability. If the study were expanded to include other EHR providers, we would need to enroll sufficient EHR providers to cover 30% US market share (5—fold increase) to achieve sufficient HCP participation for 10 completed interactions including follow-up, comparable to the magnitude of cases as detailed by Edwards et al. This level of participation, given the higher quality of data, would be sufficient to justify most regulatory actions. However, this extrapolation would be influenced by many factors e.g. frequency of prescribing of the drug and incidence rate of the AE of interest.

Second, our study was based on a US-based EHR, where the market is highly fragmented, and we only included 2 well known use cases. We are not sure if this capability would be generalizable outside of the US, although we don’t see any reason why not. Also, the EHR provider we used is focused on smaller independent practices many of which are primary care, so if this capability would work as well with larger practices and/or non-primary care providers is unknown, however there are no obvious reasons why this capability wouldn’t generalize to other EHR providers or similar digital communication tools based on the healthcare delivery setting. Finally, using such well known AEs as use cases may have prevented HCPs from consenting to participate. Given the current time pressures on healthcare it is not unreasonable to assume HCPs would allocate their time for value-add activities, such as less well known adverse events, that would most benefit the patient.

An area for potential improved participation might be to compensate HCPs for their participation. Although protecting patients is of paramount importance it is important to also realize the business side of medicine and the strained resourcing model that currently exists. If payments could be made in a transparent, ethical way then this may help facilitate participation, albeit in a potentially non-random way.

Although our study demonstrated value of the overall concept, there are a number of open questions that remain. The first question is around putting into place guardrails for a scaled implementation to ensure the information shared was only within the context of drug safety. These guardrails would define appropriate use to complement existing PV activities to support qualitative signal assessment for key safety concerns. These key concerns would likely represent specific drug-event pairs, such as those discussed in the products risk management plan and/or those that can significantly impact a drug’s benefit risk. Once this capability has yielded enough quality reports so a safety assessment can be reliably made, there maybe limited value in collecting any further data in this manner to inform benefit-risk. As it only takes a minimum number of high quality reports to support safety assessment we anticipate the volume of specific reports to be minimal, lower than 1.67 reports/HCP/month, based on the ASTER proof of concept study which enrolled 26 clinicians and reported 217 AEs over a 5 month period. Note that ASTER study was not restricted by Drug-Event pair, nor was it limited to key safety concerns therefore our estimate is likely to be conservative in comparison to restriction by drug-event pair. As an example of the rarity of a serious outcome, the incidence of myocarditis associated with covid-19 vaccines in the US population was estimated to be 1.27 per 100,000 patient days of exposure in the strata most impacted (men aged 18–25 years), and the longest risk period was 42 days, limiting the period of monitoring [25].

Although this capability offers significant potential to improve drug safety, it comes at the risk of it being leveraged inappropriately. Ensuring that there are no commercially related discussions as well as avoiding involvement in the actual delivery of healthcare is critical. We have shown how this can be done and that it was effective in this study, but moving forward there is a need to establish best practices within the healthcare, pharmaceutical, and regulatory communities to establish clear operational considerations to ensure this valuable communication channel stays open for the benefit of patients.

The next question is around the capability, in general. Although we have described our experiences using it for drug safety, this capability could generalize to any prospective data collection where timely interactive discussion is required or more likely to elicit key information while minimizing HCP/and patient burden. For example, could others leverage this capability to allow for increased clarity of benefits to ensure the most up to date benefit-risk profile is available. One risk of expanding beyond drug safety is that we can quickly overload already busy HCPs. Although our study demonstrated the minimal time required for HCP participation, an increase in the number of requests could quickly become burdensome to HCPs and discourage participation.