Pharmacometrics

Partners in Publishing

Most medical writing focusses on the traditional trial narrative: phase II happens, p-values follow, adverse events are tabulated. That template doesn’t hold for model-based work, and the gap it leaves is rarely about vocabulary. It shows up as drafts that misstate what a simulation actually supports, review cycles spent correcting basic compartmental reasoning, and eventually the point where a modeller decides it’s faster to write the manuscript themselves.

I exist to close that gap.

Because when you hand over a modelling report under non-disclosure, a conference poster, a slide deck, or raw output, what you want to come back is something you can actually submit.

I don't need a maths class! I just need the maths!

To me, maths is a language for describing how quantities move, nothing more mystical than that. In pharmacometrics, I understand it to describe what happens once a drug enters a body: how it distributes, how the body clears it, how it acts once it gets there, and what a differential equation predicts will happen next. It’s not just about defining ADME in a sentence.

Take a one-compartment model, where the body is treated as a single well-stirred tank. Or take a two- or three-compartment model, splitting that into central and peripheral spaces to capture distribution lag; that’s a choice that shows up in the concentration-time curve as a second or third exponential term, not a stylistic preference. Absorption, similarly, can be first-order, zero-order, or delayed by a lag time or a chain of transit compartments. Elimination can be linear until clearance saturates, and the kinetics turn Michaelis-Menten. This is not generally taught at BSc level.

Nor is NONMEM. Fixed effects, between-subject variability, residual error, θ, Ω, Σ — estimated by FOCE-I, SAEM, or a Bayesian engine like Stan or Monolix depending on the group’s preference. I don’t need the estimation method explained to understand why a switch from FOCE to SAEM changed the shrinkage on a random effect, or why a covariate got dropped for correlating with clearance rather than driving it; I just need you to be clear that that is what happened.

The plots carry the same weight. A visual predictive check isn’t an illustration, it’s the model’s prediction interval tested against what was actually observed, and a systematic miss at the tails is a structural problem, not a formatting one. Goodness-of-fit plots, residuals against time or population prediction, are read for bias and misspecification, not admired for their axes.

Now I want to be clear that none of this makes me a modeller, and I’m not trying to be one. It simply means that when you’re in the room explaining why a covariate stayed or a compartment got added, that actually helps me to help you, because now I can actually place that within the critical context of the narrative.

Non-Disclosure to Publication

It all starts with paperwork. I sign the non-disclosure agreement, get access to whatever secure CMS you’re working from, and you show me what you have so far: modelling reports, clinical trial reports, slide decks, conference posters, raw data where that’s appropriate. Some projects start further along: a first draft already exists, and the job is to take it from there rather than build it from source material.

Either way, the research becomes something to make submittable, and for a modelling paper specifically, that means ironing out the places where a simulation gets misread before it ever reaches a journal. For instance, a reviewer will sometimes challenge a prediction as if it claimed to be an observation, “you can’t possibly know that happens”, when the model was never asserting it as fact.

This is a communication problem. It was stating a conditional: IF a covariate behaves this way, IF exposure crosses this threshold, THEN the model predicts this outcome. That distinction, between a claim and a scenario, is exactly the kind of thing a traditional writer won’t catch before submission, and exactly the kind of thing that turns into a major revision if it isn’t caught. I’ll catch it in internal review, where a sentence can still be reworded, rather than in reviewer comments, where it has to be defended.

The same goes for the parameters themselves. A dropped covariate, a fixed rather than estimated θ, a between-subject variability term that didn’t shrink the way a reviewer expected, each of those needs a reason in the text, not just a number in a table, or a reviewer will ask the question the manuscript should have already answered.

Once it’s ready, I become the point of contact with the journal, submission, correspondence, the months of waiting that follow. When comments come back, the split is straightforward: you handle whatever requires new modelling, I handle the narrative. The idea is that what comes back stays minor, when the internal review already caught what a reviewer was going to ask.

Prior projects include areas like...

One project inferred liver-stage parasitemia in malaria chemoprophylaxis directly from blood-stage infection data, modelling both single-bite and multiple-bite mosquito exposure scenarios rather than assuming a single infection event.

Another translated a gene therapy’s efficacy data for wet macular degeneration from monkey data into a human dosing prediction, where the translation itself, not just the therapy, was part of the modelling problem we solved.

A third built a single predictive model unifying wild-type, LS, and YTE Fc variants of an FcRn-binding monoclonal antibody, capturing how each variant’s altered receptor binding extends the antibody’s half-life by reducing clearance.

Research doesn't need bottlenecks

You’ve built the model. What happens next, the drafting, the internal review, the journal, the months of correspondence, doesn’t need to be the part that slows everything down, or the part you end up doing yourself. I’ll take it from where it currently stands, and get it published.