Operational Pragmatism in ePRO-heavy trials

I was attending a site initiation visit when the clinical research associate (CRA) opened the meeting by saying something that I’ve heard versions of before, but this time it stuck with me differently. He said, essentially: we need educated patients for this study.

There’s a long series of electronic patient-reported outcomes (ePRO) built into the design, the study runs six years, and patient engagement is the backbone of the data. We need participants who are tech-savvy and ones who will stay on top of every requirement for the entire duration of the study.

Nothing mentioned by this CRA was objectionable or careless. He was doing his job and doing it well by protecting data integrity for a trial in which six years of missing or inconsistent ePRO entries would be a real and expensive problem.

But I kept circling back to one question: tech-savvy and educated relative to what? And more pointedly, are we describing the patients who can comply with the protocol, or are we describing the patients the study drug is actually meant to treat?

Those aren’t mutually exclusive and the gap between them is where I think our industry has a blind spot worth naming. The conflation hiding inside “engaged patient”

“We want engaged, educated, tech-savvy patients” sounds like a single, reasonable ask, but it isn’t. It’s actually several different attributes bundled together as if they were one thing:

• Digital literacy: Comfort with apps, portals, smartphones, navigating a login flow.
• Health literacy: Understanding what a symptom score means, why a question is being asked, how to self-report accurately.
• Language proficiency: Not just whether a translated form exists in patient’s native language, but whether the patient reads fluently enough to interpret nuance in a symptom descriptor.
• Disease burden and functional status: A patient whose disease itself impairs cognition, vision, fine motor control, or energy may struggle with ePRO completion regardless of education.
• Socioeconomic access: Device ownership, reliable connectivity, private space and time to complete forms, caregiver support.

When we shorthand all of this into “we need educated, tech-savvy patients,” the inclusion bar effectively shifts from meets the clinical inclusion criteria to meets the clinical inclusion criteria and also looks like someone who will be easy to retain. That second clause never appears in the protocol.

It shows up in screening conversations for whether a participant gets approached at all, in which the study coordinator decides whether a participant is “a good fit” before formal screening even starts.

The discomfort isn’t really about being unkind to less digitally fluent patients. It’s about what happens to the science when your enrolled population systematically diverges from your target population.

If the underlying condition disproportionately affects older patients, patients with lower formal education, rural patients, or patients with comorbidities that affect cognition or dexterity—which describes a meaningful share of chronic disease populations—then selecting for “tech-savvy and engaged” is not a neutral operational filter.

It’s a second, informal inclusion criterion stacked on top of the protocol’s actual one. You end up running an efficacy study on a population who doesn’t resemble the efficacy population you will face after approval.

That gap tends to surface later, and expensively, with real-world adherence rates that do not match trial adherence rates, post-marketing safety signals in subgroups barely represented in the pivotal data, and efficacy data that regulators or payers start to question once the drug is actually prescribed to the overall population it was meant for all along.

There’s also a quieter cost that rarely makes it into a CSR, as the patients excluded by this informal filter are often the same patients the disease burden literature tells us experience worse outcomes and less trial access in the first place. As such, we’re not just losing statistical representativeness, but we’re consistently steering trial opportunity away from the very groups most underserved by existing care pathways.

What never makes it into the data? This is the part I think gets missed most often, because it isn’t something a CRF captures. A few examples from how this plays out operationally:

• Pre-screening scenario: Long before a patient reaches formal eligibility screening, a coordinator has often already made an informal judgment about whether someone “seems like they’ll manage the app.” That judgment never appears in a screening log, it just shows up as an absence, the patient who was never approached.
• Caregiver-mediated entry as an invisible variable. A meaningful proportion of ePRO completion in real-world clinical care is caregiver-assisted, particularly for elderly patients or those with functional limitations. Most ePRO platforms don’t have a clean way to flag this, so trials often can’t tell after the fact how much of their “patient-reported” data was actually caregiver-reported, or how that may have shifted response patterns.
• Literacy versus translation. A translated form is not the same as an accessible form. Health literacy and reading level matter independently of which language the form is in, and very few protocols stratify or even track reading-level appropriateness of their ePRO instruments.
• Device and connectivity friction. In clinical research, this is usually treated as a logistics issue, not a data issue. When a site solves device or connectivity problems by simply not enrolling patients who lack reliable access, that’s a selection effect with downstream scientific consequences instead of an operational footnote.

Is the CRA’s instinct wrong? I don’t think so, and I think this is worth clarifying plainly. He is responding rationally to a genuine constraint: a six-year longitudinal design with heavy ePRO dependence will fail on data quality if retention collapses, and sponsors do penalize sites for exactly that because they lose crucial data points.

The instinct to protect participant engagement in clinical research is not the problem as there are various innovations supporting this aspect of studies and we all understand its importance. The engagement becomes a problem when it gets operationalized as a demographic proxy instead of being engineered into the protocol itself.

This reframe provides further insight regarding design upstream fix rather than downstream, toward blaming whoever is closest to the recruitment conversation.

These factors raise the question of where the fix actually belongs? A few structural changes would close a meaningful part of this gap without requiring anyone from study teams to compromise on data quality:

• Build low-tech parity into the protocol, not as an afterthought. Paper-ePRO hybrid options, interactive voice response as a fallback channel, or simplified large-text interfaces should be designed in from protocol development, not retrofitted after a site flags an access problem. If a fallback only exists because a coordinator improvised one, then it’s not a system, it’s a workaround and it will not be consistent across sites.
• Separate digital literacy support from eligibility. A patient who needs a one-time onboarding session to learn the ePRO app is not the same as a patient who is clinically unable to comply. Most protocols don’t distinguish between these, so study teams default to treating both as risk and self-select away from the first group along with the second.
• Make caregiver-assisted entry a tracked, legitimate data field. If a meaningful share of real-world ePRO completion is caregiver-mediated, the case report form should capture that explicitly rather than treating it as noise or hiding it as unreported missing data.
• Stratify health literacy and reading level alongside language, during instrument validation. Translation solves a language problem but it leaves comprehension unattended. Validated, literacy-graded versions of ePRO instruments would close a gap that translation alone leaves open.
• Track screen-fail and pre-screen-decline reasons with enough granularity to see this pattern. Most sites don’t currently have a field for “PI/coordinator judged patient unlikely to manage technology” because that judgment is rarely made explicit, even to the coordinator themselves. Without a way to see this pattern in aggregate, across sites, in feasibility data, it stays invisible by design rather than by intention.
• Include this as a feasibility assessment question. Feasibility assessments routinely ask whether a site has the patient population to enroll. They rarely ask whether the protocol’s engagement burden has been designed for the actual demographic profile of that population. That question belongs in feasibility, before the protocol is locked, not in a corrective amendment three years into enrollment.

This discussion is narrower and, I believe, more useful: when “engaged patient” quietly becomes shorthand for “educated, tech-savvy patient,” we have let an operational convenience become the unexamined eligibility criterion. The consequences of having such unaddressed inclusion criterion in the study is that the real-world population doesn’t match the trial population, and in outcomes data we then must explain rather than anticipate.

The fix isn’t asking PI and sites to compromise on data quality. It’s asking protocol designers to change their perspective regarding participant engagement in clinical research.

Currently, most teams would define this metric as something patients either do or do not have. In reality, participant engagement must be treated as something the clinical trial is responsible for building in.

About the Author

Rakshinda Mujeeb, PharmD, CRCP, ACRP-CP®, Research Projects Manager, Mediclinic Middle East.