Designing Remote Dose-Escalation Check-Ins

A titration check-in is not a symptom diary; it's a decision input. Every field on the form should map to one of three outputs: hold the dose, escalate as planned, or de-escalate and flag for review. I've sat in enough protocol design meetings to know the instinct that fights this. Someone always wants to add a field because it would be "good to know." If a data point doesn't change one of those three outcomes, it doesn't belong on the form, no matter how clinically interesting it looks in isolation. The core of the check-in has to capture timing first: date and time of the last dose, whether it was taken on schedule or delayed. Titration protocols for drugs like semaglutide or tirzepatide run on fixed intervals, typically four-week steps, and a patient who's three days late to their injection is functionally on a different protocol than one who took it on time. I've reviewed forms that bury this question below a wall of symptom checkboxes. It's backwards. Timing should be question one, not question seven. Severity comes next, and this is where a lot of forms fail by asking the wrong kind of question. "Are you experiencing nausea?" is a yes-or-no that tells a clinician almost nothing. A functional check-in asks for frequency, duration, and interference with daily activity, ideally on a scale that's been validated somewhere, even a blunt one. The Common Terminology Criteria for Adverse Events, CTCAE, grades adverse events 1 through 5, mild through death, and it was built for oncology trials decades ago. Its triage logic transfers almost perfectly to titration, because it forces every reported symptom into a bucket that maps directly to an action. A handful of digital health platforms have adapted CTCAE-style grading specifically for GLP-1 and biologic step-ups, and the reason is simple: it gives a nurse reviewing forty overnight submissions a common language. Nobody has to guess what "pretty bad" means at 6 a.m. Then there's trajectory. Is the symptom new, worsening, stable, or improving compared to the last check-in? This single field does more triage work than almost anything else on the form. A grade 2 nausea that's held steady across three check-ins is a different clinical picture entirely from a grade 2 that was a grade 1 last week, even though the severity number looks identical on paper. Static severity readings without trajectory are close to useless for a titration decision. You're comparing snapshots without knowing which direction the film is running. Last, the form needs an open text field, small and unstructured, sitting underneath the structured questions. Patients report things nobody thought to ask about, a strange metallic taste, a sleep pattern that's gone sideways, a mood shift they can't quite name. The structured fields catch the expected adverse event profile. The free text catches the tail, and the tail is where the surprises live. ## Why asynchronous triage is a different discipline than a clinic visit In a clinic, a nurse or physician sees the patient, hears tone of voice, watches how they move, asks a follow-up question in real time. None of that exists in an asynchronous check-in. The entire clinical judgment has to happen off a form filled out by someone with zero medical training, submitted whenever they happened to remember, and reviewed by a clinician who's never met them and never will. Two mechanisms close that gap: structured escalation rules and time-boxed review windows. The escalation rules are a decision tree the clinical team builds ahead of time, so that certain combinations, say grade 3 GI symptoms paired with a heart rate above threshold, trigger an automatic flag that pulls a human in immediately rather than waiting for the batch review cycle. Lower-grade, stable, or improving symptoms route into a standard queue, reviewed within a set window, often 24 hours for anything non-urgent. Building the tree in advance removes the ambiguity about what counts as urgent. Nobody wants a reviewer making a judgment call at 11 p.m. about whether "moderate" chest tightness needs a callback now or in the morning; that decision should already be made by the protocol, not by whoever's on shift. Time-boxed windows matter because asynchronous doesn't mean unbounded. A check-in that sits unreviewed for 48 hours defeats the entire model, since remote monitoring only works if someone's watching closely enough to catch a problem before the next dose comes due. Programs that publish their review SLAs, even just internally, tend to hold themselves to a tighter standard than the ones that leave the timeline loose and hope for the best. Then there's the question of who actually does the reviewing. Most remote titration programs I've seen use tiered review: a nurse or PA handles the first pass on standard submissions, physician review gets reserved for flagged or escalated cases. This isn't a new idea. Telephone triage lines have run this exact model for decades, for the same reason it works here: most check-ins are unremarkable, and routing every single one to a physician wastes a resource that's already scarce. That efficiency only holds if the tiering rules stay conservative about what qualifies as "standard," though. Erring toward escalation costs a few extra minutes of physician time. Erring toward under-escalation costs a missed adverse event, and those two costs are not remotely comparable. ## The failure modes worth naming Two mistakes show up again and again in programs that get this wrong. The first is over-collection. Teams new to remote titration tend to ask for everything: blood pressure, weight, mood, sleep, appetite, injection site photos. More data feels safer. It isn't. Every additional field is friction, and friction drives drop-off in check-in completion, which is a worse outcome than a slightly thinner dataset. A patient who abandons the form gives the clinical team zero information. A patient who answers four sharp questions gives them exactly what they need. Fewer, better questions beat more, weaker ones, every time I've seen this tested. The second mistake is treating every check-in identically regardless of where the patient sits in the schedule. The first check-in after an initial dose is hunting for a different signal than the check-in right before a scheduled escalation. Early check-ins should weight tolerability heavily, since the job is catching an adverse reaction before the next step up. Pre-escalation check-ins should weight readiness, confirming that side effects from the current dose have settled enough to justify moving forward. Run the identical form at both points and you're asking questions that don't matter at that stage, which dilutes attention away from the ones that do. ## Where this is heading None of the underlying logic here is new. Structured severity grading, trajectory tracking, tiered asynchronous review; this is borrowed almost wholesale from oncology symptom management and post-surgical remote monitoring, fields that spent years working out how to keep a patient safe between visits without a clinician in the room. What's changed is scale. GLP-1 titration programs are running this model across patient populations that dwarf anything a single oncology practice ever managed remotely. That scale is forcing the protocols to move faster and automate more than their predecessors ever had to. A single oncology nurse triaging a few dozen post-chemo check-ins a week is a fundamentally different operation than a platform processing thousands of GLP-1 check-ins a day, and the second one doesn't survive on good intentions alone. It survives on a form that asks the right four questions, an escalation tree that removes guesswork, and a review window somebody actually enforces.