“A Community-Engaged Approach to Address Collateral Findings in Embedded Research”

In In their article Morain and Largent suggest looking “beyond the investigator-participant dyad” to understand the ethical obligations in embedded research using Electronic Health Record (EHR) data (Morain & Largent, 2022). We agree with this general claim but believe their analysis of the cases misses important known problems with EHRs that shift what solutions are required. In every case the interpretation of results emerging from data analysis assumes the cause is likely the result of individual patient neglect or substandard care. We irritate this assumption by providing alternative system and patient/community-based reasons for deviation from care norms in the record.

EHR data can be vague, incomplete, or mistaken (Bell et al, 2022) for many reasons. Gaps in EHR data can be based on idiosyncrasies of local EHR data entry practices. New employees and trainees learn from their peers about the language norms and shortcuts they can take in the EHR platform, whether this relates to quick-text functionality or copy-and-paste procedures (Brown, 2014). Sometimes there are user-errors or software errors where information is mixed up (van der Bij et al, 2017; Bell et al, 2020). Sometimes the EHR makes documenting nuance or differences from a process or discussion difficult (form structure within the record, for example).

Furthermore, various system-based contextual features could justify deviation from standards of care. Healthcare systems create norms of practice relevant to their resources, the communities they serve, their employees, etc. Standards of care are not universally agreed (Moffett & Moore, 2011) and systems and their providers may not be up to date with the latest guidance (which may not be settled or comprehensive), have sufficient resources available, or have other practical issues that require different approaches.

Variations from the standard of care may also be based on community- and patient-specific acceptance or access to certain kinds of interventions or non-medical support. The patient-centered care model (Hansen, Walters and Howes, 2016) means that strict guidance should not always be followed but rather adapted to the individual needs of patients . Culturally responsive hospitals in settings with populations with specific needs may justifiably implement practices that deviate from the norm. Although these deviations might be justifiable by community standards, they may not be clearly documented as such because of limitations in the EHR systems.

Using EHR data requires recognition of both these issues. First, recognizing (and accounting) for the quality of the data found in EHR records in these kind of embedded studies. Second, recognizing practices in the embedded context can vary in relation to patient, community, and systems-based values and constraints from what might be considered the ideal and that such variation may be justified.

This does not mean that community-based collateral findings in large EHR data are always insignificant. The cases that Morain and Largent (2022) raise could plausibly signify collateral findings, but, may given the discussion above also signal a localized EHR entry idiosyncrasy or error and/or patient- or community-related issues of care. For example, in the third case study titled “poor emergency medical care”, Morain and Largent suggest that suturing an animal wound on a child is against standard of care. However, opinion on suturing or non-suturing of animal wounds has mixed evidence (Cade, Low and Head, 2018) meaning physicians’ practices probably vary across the country (and within health systems potentially) and parents may make arguments for why sutures in a particular instance may be preferable. For the first case study titled “untreated atrial fibrillation”, it’s worth noting that the recommendation in the guidelines (based on varying degrees of evidence) regarding initiation of medication includes many caveats and references to shared decision-making around benefit and risks of various medical interventions (January et al, 2019). It is possible that various untreated patients elected not to have medical intervention, or they had contraindications to some interventions, or other reasons may exist that justify variation.

One way to resolve these issues while being responsive to the differences that Morain and Largent (2022) raise related to this kind of embedded EHR data research (obligations, communication, various types of costs) is to recognize that not only is a relationship with the health system needed, but also one with the population the health system serves in the form of a community advisory board (CAB). All three parties should be connected and this should begin prior to the project start right through to completion, where completion means all parties have agreed to a plan of action (where necessary) and dissemination in relation to the findings. This recognizes that the relationship here is between the researcher, the health system, and the population it serves. This triadic relationship is one that can be both mutually supporting in ensuring appropriate information sharing for contextualization of EHR data (and other factors), but also reinforces and enables responsibility and accountability between parties when dealing with collateral findings.

At the onset of an embedded research project, both the health system and CAB can advise researchers on possible local norms that affect the interpretation of their work so that variation in data is not taken at face value. Researchers can communicate to the health system and CAB what the purpose and process of the research is and the possibility of collateral findings. This allows a conversation to happen at the beginning of the project between the community and health system mediated by the researchers about what is an acceptable response to different kinds of findings (individual negligence, large scale deviation from norm, etc.) that may be refined when actual findings arise. These discussions build transparency and accountability into the foundations of the project between the health system, researchers, and the community. EHR data used in the context of embedded research allows a clinical audit to occur and both the health system and CAB should know that and plan for audit related outcomes from the beginning.

Morain and Largent (2022) state a key difference is the lack of consent that these kinds of projects engender through their waivers. Working with a CAB provides some sense of community endorsement of the work and causes a direct sense of obligation to those served by the health system. It also gives the CAB the power to determine what is the appropriate response to findings. This relates to the point that Morain and Largent (2022) raise about communicating collateral findings. CABs may suggest outreach or individual contact depending on the nature of the finding(s) that may arise. Responses will be negotiated with health systems, but through this kind of process are more likely to be in line with what the health systems’ populations find acceptable.

Lastly, Morain and Largent (2022) raise the issue of costs. They raise issues of not just financial costs, but also costs to potential partnerships between researchers and health systems by disclosing collateral findings. We acknowledge that health systems who experience the outcomes of uncovering problems may not engage in research again but that is not a reason to avoid this work or communicate that health systems have responsibilities when issues are discovered. We are of the view that good relationships and communication between the triadic parties in the set-up of projects connects people to a shared vision of what the research is supposed to achieve and informs all parties of the constraints each is working under, and their goals, etc. This communication has the potential to alleviate distrust by communities, or bad feelings by health systems but depends on good practices in collaborative partnerships and community engagement.

EHR embedded research offers opportunities to certain research questions that otherwise might not be possible. Explicitly acknowledging the limitations of EHR data and the variation in care norms is required to do this work practically. The waivers of consent and the nature of questions that are generally used in these setting may remove obligations to individual “research subjects”, but do not remove obligations – they transfer to the health system and its population. The best way to manage such obligations in our view is through the creation of a CAB that is in conversation with the health system and researchers from the very beginning. This triadic relationship can support transparent communication, contextualization of data, and accountable and agreed approaches to collateral findings.