References for the PK-platelet ITP working document

What the literature has, what it does not, and what to read in what order

A review of published dose and PK-platelet modelling in immune thrombocytopenia, the reading queue for the PK-platelet dose selection specification, and a status marker on every source recording whether the claim drawn from it has been checked.

Six sources have been read; the rest were assembled on 2026-09-09 from web search results and abstracts. Entry 1, the IWG 2009 consensus paper, and entries 22 to 25, the FDA, EMA and PMDA reviews of rilzabrutinib, fostamatinib and efgartigimod in ITP, were read in full and are ✅, as is the PAGE 2009 poster behind entry 26. Every other marker is ❌, including the numbers Section 7 and Section 16 of the specification quote: the ianalumab 3 against 9 mg/kg comparison, the rituximab low-dose response rates, the efgartigimod immunoglobulin G reduction, and the four-compartment platelet model structure all come from search summaries rather than from the papers. A search summary is weaker evidence than an abstract, which is itself weaker than the paper. Treat each entry as a prediction about what its source says, and correct the entry when the source is read.

The working document is the specification, and the folder’s other documents are on the project index.

Status markers

• ❌ Not checked. No claim in the working document has been verified against this source. Where the working document characterizes it, the characterization comes from an abstract or a search result.
• ⚠️ Transcribed, unverified. A number, table or model structure in the working document came from this source but has not been checked against it.
• ✅ Checked. Verified against the source.

On a reading-queue entry, ✅ means the question posed against that entry has been answered, by reading the paper or by deciding it is out of scope, and the note on the entry says which. On a source entry below the queue, ✅ keeps the stricter meaning: the claim the specification draws from it has been checked against the source. None is ✅ yet.

What the search found

No published model links drug concentration to platelet count for any drug in immune thrombocytopenia (ITP) that works by stopping antibody-mediated platelet destruction. The search covered three mechanisms that do that: B-cell depletion (ianalumab, rituximab, obinutuzumab, obexelimab, telitacicept, povetacicept), plasma-cell depletion (mezagitamab), and blocking the destruction step itself (fostamatinib on spleen tyrosine kinase, rilzabrutinib and orelabrutinib on Bruton tyrosine kinase, efgartigimod and rozanolixizumab on the neonatal Fc receptor). Each was paired with population pharmacokinetic and pharmacodynamic (PK/PD) terms. What came back was clinical trial reports, models whose pharmacodynamic endpoint stops at a mechanistic biomarker, and one exposure-response analysis against platelet count that was reported as inconclusive.

The regulatory reviews, read in full, change the answer from “not found” to “attempted three times, and here is what happened.” Entries 22 to 25.

• Rilzabrutinib, FDA 2025. An indirect exposure-response model on 10,652 weekly platelet counts from 305 patients “adequately characterized platelet counts.” It exists, it fit, and it was built after the dose was chosen, on data almost entirely at that one dose, to confirm it. Its continuous endpoint rose with exposure; its dichotomous primary endpoint did not.
• Fostamatinib, FDA 2018 and EMA 2019. “None of the applicant’s attempted models including semi-mechanistic, direct response, and indirect response model were successful for the platelet count analysis as a continuous variable.” Logistic regression at one visit on 79 patients found a slope; the same slope was absent at the primary endpoint window. The FDA called the early slope “notable” and the CHMP said a relationship “has not been demonstrated.”
• Efgartigimod, PMDA 2024. argenx built an IgG-to-platelet model on the Phase 2 data and could not explain the between-patient variation with it, so the Phase 3 dose was chosen on IgG reduction alone, and PMDA accepted that.

So the gap is real and it has a shape. At Phase 2 sample size the model did not identify (efgartigimod); at Phase 3 sample size with two doses it did not fit as a continuous model (fostamatinib); at pooled Phase 2 and 3 sample size with one dose it fit and showed a flat dose-response on the endpoint that counts (rilzabrutinib). None was used to pick a dose. That is the specification’s question, asked three times by sponsors with more data than a Phase 1 study has, and the answer was no each time.

Why the gap exists is argued in Section 19 of the specification, in seven ranked reasons, each drawn from an entry below. The short form: the drugs with published PK-platelet models in ITP all raise platelet production, which gives a graded dose-response with a lag of days, and the drugs in this project’s class lower platelet destruction through an unmeasured autoantibody with a peak-response lag of 14 to 180 days.

What does exist splits into five groups, and the specification borrows from four of them.

What the literature supplies and what it does not

Group	State of the literature	What the specification takes
PK-platelet models in ITP, for thrombopoietin receptor agonists	Mature. Lifespan and transit-compartment models fit to serial platelet counts, several drugs, regulatory use	The platelet-side structure and its fixed parameters (Section 8.3)
Model-based dose selection in ITP	Exists, for a neonatal Fc receptor blocker, anchored on immunoglobulin G rather than on platelets	The precedent that this is done and accepted (Section 13)
PK-B-cell models for B-cell depleting agents	Mature, in other diseases. Turnover models on circulating B cells, plus tissue receptor occupancy where circulating counts saturate	The depletion structure and the saturation argument (Sections 7, 8.2)
PK-IgG models for FcRn blockers	Mature. Receptor occupancy drives immunoglobulin G suppression, fit across dose groups, used to select a Phase 3 ITP dose	The precedent, and the shape of a model that stops one link short (entries 3, 4)
The join: drug → B cells → autoantibody → platelets, in ITP	Not found	Nothing. This is the gap the project sits in

Three findings from the search bear directly on whether the project’s answer will be yes, each unverified and each with an entry below.

Dose-response looks flat over the ranges that have been studied. Ianalumab at 3 and 9 mg/kg in ITP gave 12-month freedom-from-treatment-failure probabilities of 51% and 54% against 30% for placebo, a three-fold dose range separated by three percentage points (entry 14). Rituximab at 100 mg weekly and at 375 mg/m² weekly give pooled overall response rates around 60% and 69%, a four-fold dose range (entry 18). If both hold up on reading, the plateau in ITP starts low and the model’s job is to find its bottom edge rather than to climb a slope.

The one worked example of model-based dose selection for this exact molecule class used a second biomarker precisely because circulating B cells saturate. The ianalumab Phase 2b dose selection in primary Sjögren’s syndrome (entry 2) paired a circulating B-cell turnover model with a hypothesis-driven tissue receptor occupancy model, and reportedly used the pair to space three doses. That is the specification’s Section 7 argument, already made by someone else, for a different indication, and accepted by a regulator.

The one exposure-response analysis found against platelet count in ITP for a drug that blocks destruction came back inconclusive. Fostamatinib inhibits spleen tyrosine kinase and therefore blocks the phagocytosis step directly. Responders had higher active-moiety concentrations than non-responders, but the review reports the correlation between exposure and achievement of the platelet endpoint as uncertain, while exposure against hypertension was significant (entry 10). Read that entry before deciding how much weight the result carries: an inconclusive analysis of a drug titrated to effect says less than an inconclusive analysis at fixed dose.

Exposure-response confounding is a documented failure mode for exactly this study shape. The review literature on time-dependent clearance in therapeutic antibodies reports that single-dose-level designs carry a high probability of a false-positive exposure-response finding (entry 19). An escalation with a single-dose expansion cohort is that design. This is why Section 11 makes dose rather than exposure the independent variable in the selection rule.

Where to start

Read in this order. Entries 1 and 22 to 26 are already read, and the review sections above carry what they said.

• Entries 22 to 25 first, and they are already read. The regulatory reviews answer the project’s premise directly. Re-read only for a quotation.
• Entry 1, already read. Section 3 of the specification is what it produced. Re-read only to check a definition.
• Entry 26, the eltrombopag ITP model, before building anything. It has the mixture, the fixed maturation chain, and the response-guided titration simulation that Sections 8.3, 8.4 and 15 of the specification propose. The poster has been read; the paper has not, and the paper is where the degradation rate in ITP is.
• Entry 2 next, and it may settle the project’s shape. It is the closest published analogue: model-based Phase 2b dose selection for the same molecule, in a different autoimmune disease. Budget one hour. If it turns out that tissue receptor occupancy did the work and circulating B cells did not, Section 17’s first failure mode is confirmed before any code is written.
• Entries 3 and 4 together, for the precedent that a PK/PD model has selected a dose in ITP and what it was anchored on. Entry 4 is the published model and entry 3 is its application, so read 4 for the equations and 3 for the decision. An hour.
• Entry 10 before deciding to build the project. It is the one published attempt at the thing this project proposes, and it did not work. Half an hour on why.
• Entries 5 and 6 together, as Milestone 1. They are the platelet-side structure. The deliverable is a parameter table, so read for the equations and the estimates, not for the conclusions.
• Entry 7 for the ITP-specific platelet control problem. It applies a romiplostim PK/PD model to choosing a dose profile in an ITP patient, which is the closest thing found to the specification’s question with the drug class swapped.
• Entry 8 for the B-cell side, as Milestone 2, and entry 9 for the transduction delay prior in Milestone 3.
• Entries 16 and 17 for what ITP dose-ranging designs actually look like. Entry 17 is the only randomized two-dose comparison found in this drug class, and entry 16 is the intrapatient escalation that a dose-response model cannot be fit to. Twenty minutes each.
• Entry 19 before writing the analysis plan, not before the model. It changes what the analysis is allowed to claim rather than what it estimates.

The general method for reading any of these is on the Reading Papers page: three passes, stopping at the pass that answers the question. An hour is the budget for a paper read that way, and entry 2 is the one that may need two.

Reading queue

1. Rodeghiero et al. 2009 — IWG standardization of terminology and outcome criteria ✅

Blood 2009;113(11):2386-2393. doi:10.1182/blood-2008-07-162503 · ASH · full text PDF

Read on 2026-09-09, and the only ✅ in this file. Section 3 of the specification is transcribed from its Tables 1 to 5. It defines complete response, response and no response; the disease phases; refractory ITP; the trial-adapted primary and secondary endpoints; and, in Table 3, the time to initial and peak response for each ITP agent over its reported dose range.

What it settles. Three things the specification depended on and no longer guesses at. There is no partial response in ITP: the panel considered “partial” and “minimal” and dropped both, because the criteria used for them in the literature are too heterogeneous. Response requires absence of bleeding as well as a platelet threshold, so a platelet-only model predicts an upper bound on the response rate. And rituximab’s time to peak response is 7 to 56 days to initial and 14 to 180 days to peak, which is the prior on \(k_{e0}\) in Section 8.4 with a thirteen-fold spread.

What it does not settle. The thresholds it defines, 30 and 100, are not the thresholds recent trials use. Entry 17 and the table in Section 3.4 carry the divergence.

2. Baltcheva et al. 2018 — model-based dose selection for ianalumab in primary Sjögren’s syndrome ❌

PAGE 2018 abstract. Baltcheva I, Fink M, Petricoul O, Renard D, Novartis. https://www.page-meeting.org/Abstracts/model-based-dose-selection-for-the-phase-2b-study-of-ianalumab-in-primary-sjogrens-syndrome/

Two PK/PD models used together to space three Phase 2b doses for a B-cell activating factor receptor (BAFF-R) antibody.

The question. Whether circulating B cells alone could have selected the doses, or whether the tissue receptor occupancy model was load-bearing. That answer decides whether the specification’s Section 7 is a caution or a blocker. Second question: what the population PK model was, since a two-compartment model with linear clearance and a 9 to 10 day half-life is the starting point for Section 8.1.

What a search summary says, unverified: a two-compartment PK model with linear clearance; B-cell dynamics as an indirect-effect turnover model with a peripheral compartment; a separate hypothesis-driven model of drug-receptor binding in disease-related tissue under quasi-steady-state with competition from soluble BAFF; three dose tiers selected as low (systemic depletion only), medium (both mechanisms) and high (matching proof-of-concept exposure); doses in the Sjögren’s study of 5, 50 and 300 mg; and half-life around 9 to 10 days.

3. Efgartigimod PK/PD simulations for subcutaneous dose selection in ITP ❌

Blood 2021, ASH abstract. https://www.sciencedirect.com/science/article/pii/S0006497121050989

PK/PD simulation used to pick a subcutaneous dose for a Phase 3 ITP trial from the intravenous benchmark.

The question. What the PD anchor was. If it was total immunoglobulin G (IgG) reduction with platelet count treated separately, this is a precedent for model-based dose selection in ITP that stops one link short of the chain in the specification, and the reason it stopped there is the interesting part.

What a search summary says, unverified: 10 mg/kg intravenously weekly for four doses achieves close to maximal IgG reduction, mean reduction up to 63.7%; simulations predicted weekly subcutaneous 1000 mg gives comparable maximum total IgG reduction after the fourth dose and comparable area under the total IgG concentration curve. The Phase 2 report is Newland et al., Am J Hematol 2020, PMID 31821591, and the Phase 3 is ADVANCE IV, Lancet 2023.

4. Translational population target binding model for efgartigimod ❌

J Pharmacokinet Pharmacodyn 2025;52:2. https://d-nb.info/1364699516/34

The published model behind entry 3, and the most complete PK/PD model found for any ITP drug that acts on autoantibody rather than on platelet production.

The question. Whether platelet count appears in it at all. If the model stops at total immunoglobulin G, it is the documented instance of reason 6 in Section 19 of the specification: the sponsor modelled the clean biomarker and left the clinical endpoint to the trial.

What a search summary says, unverified: a population PK/PD model including FcRn binding, built on two healthy-volunteer studies after single and repeated administration, describing serum efgartigimod and total IgG profiles simultaneously across dose groups, with drug-induced FcRn receptor occupancy as the driver of total IgG suppression.

5. Eltrombopag population PK/PD, four-compartment lifespan model ❌

Cancer Chemother Pharmacol 2013. doi:10.1007/s00280-013-2150-9 · PubMed 23564375. Companion in chronic liver disease: Br J Clin Pharmacol, PMC4371540.

The platelet-side structure the specification adopts in Section 8.3.

The question. The parameter values, with units, for the maturation transit rate and the circulating platelet loss rate, and whether the model was fit to ITP patients or only to other thrombocytopenias. Milestone 1’s table is built from this.

What a search summary says, unverified: a linear two-compartment PK model with first-order absorption; a four-compartment lifespan model for platelets, being one precursor production compartment, two transit and maturation compartments, and one blood platelet compartment; drug effect as stimulation of platelet precursor production proportional to plasma concentration.

6. Wang et al. 2010 — romiplostim precursor pool lifespan model ❌

AAPS J. PMC2976980. Pharmacodynamics-mediated drug disposition and precursor pool lifespan model for single-dose romiplostim in healthy subjects.

The age-structured version of entry 5, and the one that also models the drug being consumed by its own pharmacodynamic effect.

The question. Whether the age-compartment structure buys anything the three-transit-compartment version does not, at the resolution of weekly platelet counts. If it does not, Section 8.3 stays as written and this entry is closed as background.

What a search summary says, unverified: megakaryocyte precursor age-compartments generating young platelets, which mature through a series of age-compartments, with circulating platelets the sum over all ages; platelet lifespan in healthy individuals 7 to 10 days.

7. Optimum romiplostim dose profile for platelet control in an ITP patient ❌

IFAC proceedings. https://www.sciencedirect.com/science/article/pii/S1474667016434932

A PK/PD model of romiplostim in ITP used to choose a dose profile, which is the specification’s question with a thrombopoietin receptor agonist in place of a B-cell depleting agent.

The question. What ITP-specific parameters were used, and where they came from. A published ITP platelet turnover parameterization saves Milestone 1 from being assembled out of non-ITP sources.

What a search summary says, unverified: a standard PK/PD model developed and validated on clinical platelet count data after subcutaneous romiplostim, characterizing patient-specific physiological characteristics.

8. Rituximab B-cell depletion PK/PD, target-mediated disposition and CD19 turnover ❌

Population pharmacodynamic modelling of CD19+ suppression, PMC10674351 · Dosing optimization for primary membranous nephropathy, PMC11002205 · B-cell differentiation model in rheumatic diseases, PMC12823323

Three papers, one structure: a target-mediated disposition PK model with a turnover model on circulating CD19+ cells in which the drug increases the loss rate.

The question. The repopulation rate constant, which is the parameter that sets the specification’s \(T_{\rm rep}\), and whether any of the three reports a dose-dependence of duration once depth has saturated. That dose-dependence is the specification’s central claim and this is the cheapest place to check it.

What a search summary says, unverified: systemic clearance 0.54 L/h and half-life 14.7 days in one adult population, half-life 11.6 days in a paediatric one; CD19+ time course described by a turnover model with the drug increasing the degradation rate; the differentiation model includes localization of B-cell subsets in bone marrow and secondary lymphoid organs and the internalization of the CD20-rituximab complex.

9. Rituximab response kinetics in ITP, for the transduction delay prior ❌

Rituximab resistance in ITP and beyond, Front Immunol 2023 · Long-term follow-up in refractory ITP, PMC5451692

Not modelling papers. They are where the median time from dose to platelet response comes from, which is the prior on \(k_{e0}\) in Section 8.4, and where the response-rate ceiling comes from, which is the prior on the long-lived plasma cell floor \(\phi\).

The question. Two numbers with intervals: time to platelet response, and the fraction of patients who never respond. Milestone 3 is one hour and produces those two numbers.

What a search summary says, unverified: up to 60% short-term complete response, 40% durable at 6 to 12 months, 20 to 30% at 5 years; and a mechanism for primary failure in which B-cell depletion reprograms splenic short-lived plasma cells into long-lived ones. That mechanism is the \(\phi\) term in Section 8.3 and if it is dose-dependent the model needs a term it does not currently have.

Trial reports the external checks and the realism constraints come from

These are not modelling papers. They supply the numbers Sections 7, 9 and 16 quote, and each is checked when the claim that rests on it is checked.

10. Fostamatinib and R406 clinical pharmacokinetics and pharmacodynamics ❌

Clin Pharmacokinet 2022. doi:10.1007/s40262-022-01135-0 · PubMed 35781630

The closest thing found to an exposure-response analysis against platelet count for a drug that blocks platelet destruction in ITP, and it came back inconclusive. Fostamatinib inhibits spleen tyrosine kinase, blocking Fc-gamma-receptor-mediated phagocytosis of antibody-coated platelets by macrophages, which is the killing step itself rather than the antibody that triggers it.

The question. What the exposure-response analysis actually did, and whether its inconclusive result came from a flat relationship, from dose titration to effect, or from sample size. That distinction is reason 7 against reason 4 in Section 19 of the specification, and it changes how much the fostamatinib result should discourage this project.

What a search summary says, unverified: responders had higher R406 concentrations than non-responders, but the correlation between R406 exposure and achievement of the platelet endpoint was uncertain in the PK/PD analysis, while R406 exposure correlated significantly with the incidence of hypertension; R406 exposure is approximately dose proportional to 200 mg twice daily; concentration is roughly halved in heavy patients; dosing in ITP starts at 100 mg twice daily and most patients escalate to 150 mg twice daily.

11. Ianalumab Phase 2 in ITP, VAYHIT3 interim ❌

Blood 2024;144(Supplement 1):710. ASH · NCT05885555

Why it is here. It is the sample size the specification is written around. A search summary reports ten patients in the interim, four doses of 9 mg/kg intravenously every four weeks, median best post-baseline platelet count 129 G/L with a range of 3 to 709 G/L, one confirmed response lasting six weeks, and rapid profound B-cell depletion. A range of 3 to 709 in ten patients is the responder mixture of Section 12 in raw form.

The question. Whether B-cell counts and platelet counts are reported per patient over time, and whether more than one dose level was studied anywhere in the ianalumab ITP programme before 3 and 9 mg/kg were carried into Phase 3.

12. Mezagitamab Phase 2b in ITP ❌

TAK-079-1004, NCT04278924. Takeda release · Phase 3 trial-in-progress, Blood 2025

Why it is here. It is the one recent ITP programme found that ran three dose levels against placebo in Phase 2, which is the design the specification is trying to make unnecessary. An anti-CD38 antibody depletes plasmablasts and plasma cells rather than B cells, so it attacks the same chain one step further down.

The question. The dose levels and the response rate at each. A search summary reports three subcutaneous doses weekly for eight weeks, a dose-dependent platelet increase with the greatest response at the highest dose, and 10 of 11 patients responding at 600 mg against 3 of 13 on placebo. If the dose-response was still rising at the top dose, this is the Section 17 case where the plateau lies outside the studied range, observed in a real programme.

13. Rozanolixizumab Phase 2 in ITP ❌

Blood Adv 2020;4(17):4136-4146. PMC7479959 · ASH

Why it is here. A dose-ranging ITP study with both single and multiple dose arms, and paired IgG and platelet time courses. The closest published dataset in shape to what Section 9 asks for, with an FcRn blocker instead of a B-cell depleting agent.

The question. Whether the IgG and platelet time courses are published per arm at a resolution a model could be fit to, and what the lag between minimum IgG and maximum platelet count was. That lag is the transduction delay of Section 8.4 measured directly, for a drug whose action on autoantibody is faster and better characterized.

What a search summary says, unverified: 66 patients, one to five weekly subcutaneous infusions, cumulative doses 15 to 21 mg/kg; platelet count \(\ge 50 \times 10^9\)/L reached at least once in 35.7%, 35.7% and 45.5% of the 5×4, 3×7 and 2×10 mg/kg arms and 66.7% and 54.5% of the single 15 and 20 mg/kg arms; minimum mean IgG and maximum mean platelet count both by day 8 in the single-dose arms, with maximum platelet count from day 11 onward in the multiple-dose arms.

14. Ianalumab plus eltrombopag in ITP, VAYHIT2 Phase 3 ❌

N Engl J Med 2025. doi:10.1056/NEJMoa2515168 · PubMed 41363800

Why it is here. It is external check 1 in Section 16, and the single most important entry for whether the project’s premise holds. Two doses of a B-cell depleting agent, randomized, in ITP, with a clinical endpoint.

The question. The gap between 3 mg/kg and 9 mg/kg, with its interval. A search summary reports 12-month probability of freedom from treatment failure of 51% at 3 mg/kg, 54% at 9 mg/kg and 30% for placebo, and 6-month platelet response of 73.5% at 9 mg/kg against 48% for placebo. If those confidence intervals overlap heavily, the plateau claim is supported and a model that predicted it from escalation data would have saved a Phase 3 arm.

15. Rilzabrutinib LUNA3 Phase 3 in ITP ❌

Blood 2025;145(24):2914. PubMed 40090011 · PMC12824683

Why it is here. The realism constraint on Section 15’s simulated response rates, and the current definition of the durable response endpoint.

The question. The exact endpoint definition, which Section 15 has to simulate. A search summary gives it as platelet count \(\ge 50 \times 10^9\)/L for at least two-thirds of at least 8 of the last 12 of 24 weeks without rescue therapy, achieved by 31 of 133 (23%) against 0 of 69 on placebo. A 23% durable response rate against a 0% placebo rate is a different realism target from the ianalumab numbers and both should be reachable by the generative model.

16. Rilzabrutinib LUNA2 Phase 1/2 Part A, dose finding in ITP ❌

NCT03395210 · N Engl J Med 2022 · Part B, Am J Hematol 2025, PMC11803537

A dose-finding study in ITP for a Bruton tyrosine kinase (BTK) inhibitor, which hits both arms of the mechanism: reduced autoantibody production through the B-cell receptor, and blocked Fc-gamma-receptor-mediated phagocytosis.

The question. Whether the dose escalation was intrapatient. If it was, every patient saw more than one dose level and dose is confounded with time on drug, which is reason 5 in Section 19 of the specification and would explain the absence of a dose-response model from a study that otherwise had the data for one.

What a search summary says, unverified: an adaptive open-label dose-finding design with intrapatient escalation over 24 weeks across 200 mg once daily, 400 mg once daily, 300 mg twice daily and 400 mg twice daily; 60 patients enrolled, 24 met the primary endpoint, and 18 of the 45 who started at 400 mg twice daily did; median time to first platelet count \(\ge 50 \times 10^9\)/L of 11.5 days.

17. Orelabrutinib Phase 2 in ITP, two randomized dose levels ❌

Am J Hematol 2024. doi:10.1002/ajh.27303 · PubMed 38546375 · NCT05232149. Mechanism companion: Br J Haematol, doi:10.1111/bjh.20045

The one randomized dose comparison found in ITP for a drug of this class, and therefore the closest available test of whether a dose-response on the platelet endpoint is visible at Phase 2 sample sizes.

The question. Whether the difference between the two arms survives its confidence interval. A search summary reports 33 patients randomized to 50 mg (n = 15) or 30 mg (n = 18), with the primary endpoint of two consecutive platelet counts \(\ge 50 \times 10^9\)/L within 4 weeks and no rescue reached by 40% at 50 mg and 22.2% at 30 mg. Six responders against four is a difference that 33 patients cannot resolve, which is the point: this is what a dose-response looks like when read off raw counts, and it is the comparator rule D of Section 13 in its natural habitat.

Second question. Whether PK was collected. A randomized two-dose design with concentration data and serial platelet counts is the smallest dataset that could carry the model in Section 8, and if it exists it outranks the rest of this queue.

18. Rituximab dose comparison meta-analyses in ITP ❌

Platelets 2019;30(6), doi:10.1080/09537104.2019.1624706 · Different dosages of rituximab in adult ITP, PMC8514896

Why it is here. External check 2 in Section 16, and the evidence behind Section 7’s claim that depth of depletion carries no dose information.

The question. Whether the low-dose and standard-dose comparisons are within-study randomized or pooled across single-arm studies. Pooled across studies, the comparison is confounded by patient selection and the plateau claim is weaker than it looks.

What a search summary says, unverified: nine studies and 329 patients on 100 mg or 100 mg/m² weekly for four weeks giving pooled overall response 63% (95% CI 0.54 to 0.71) and complete response 44% (95% CI 0.33 to 0.55); standard 375 mg/m² weekly for four weeks giving 69% overall initial response and 35% sustained.

Method sources for the analysis plan

19. Time-dependent clearance and exposure-response confounding ❌

Proctor et al., Clin Transl Sci 2024, PMC10766027 · Clinical study design strategies, CPT Pharmacometrics Syst Pharmacol 2025, doi:10.1002/psp4.13280 · IQ Consortium survey, CPT PSP 2026, doi:10.1002/psp4.70318

Why it is here. Section 11 is built on it. Read before the analysis plan is written.

The question. Which of the mitigations they recommend apply when the target itself is the clearance sink, as B cells are here. A search summary reports the finding that single-dose-level designs have a high probability of a false-positive exposure-response result, and that 88.0% of surveyed industry respondents acknowledge the relevance of the confounding while remaining uncertain about its prevalence outside oncology.

20. Power loss from dichotomizing a continuous endpoint ❌

McMenamin et al., PubMed 32450909, analysis of responder-based endpoints through their continuous components.

Why it is here. Section 12 asserts that fitting the continuous platelet count and deriving the responder endpoint beats fitting the dichotomy. This is the citation for the size of the gain.

The question. Whether the augmented binary method they describe is a better fit for Section 12 than the simulate-the-endpoint-from-the-model approach, since it retains the original endpoint and would be easier to defend to a clinical team.

21. Immature platelet fraction as a pharmacodynamic marker ❌

IPF as predictor in ITP, PMC4928401 · IPF in etiological determination of thrombocytopenia

Why it is here. Requirement 10 in Section 9, and the argument that one extra number on a blood count separates a production effect from a destruction effect.

The question. The typical IPF value in untreated ITP and its within-patient variability, since a marker whose noise exceeds its treatment effect should not be a requirement.

What a search summary says, unverified: IPF consistently elevated above 10% in ITP against normal values in marrow suppression, with a discriminating threshold around 8.45%, and roughly a 200% difference between hyperdestructive and hypoproductive thrombocytopenia.

Regulatory reviews, read in full

The four review documents below were downloaded and read on 2026-09-09. The quotations are from the documents, with page numbers, and the four entries are ✅. Between them they answer the question the journal search could not: every regulatory programme for an ITP drug that blocks platelet destruction attempted a PK-platelet model, and the reviews say what happened to each.

22. FDA integrated review, Wayrilz (rilzabrutinib), NDA 219685, 2025 ✅

219685Orig1s000IntegratedR.pdf, Sections 6.1.1.1 (pp. 26-28) and 14.5.2 (pp. 164-167). The FDA site blocks scripted downloads; a browser user-agent string gets through.

A longitudinal PK-platelet model for a drug that blocks platelet destruction exists, and this is where it is. The Applicant’s exposure-response analysis used “the longitudinal platelet counts … as the primary efficacy endpoint which was based on key aspects of the ITP disease etiology and the presumed mechanisms of action of rilzabrutinib.” The dataset was 86 participants from LUNA 2 and 219 from LUNA 3, “which included 10,652 weekly platelet counts with no missing data or data that were below limit of quantification. An indirect exposure-response model adequately characterized platelet counts, including the magnitude of persistent and sometimes large fluctuations in platelet counts” (p. 164). Covariates: prior response to corticosteroids and concomitant corticosteroids raised the response rate; baseline platelet count below the median lowered it.

What it found, and the two halves disagree in the direction Section 12 of the specification predicts. On the continuous endpoint, “a trend for greater increase in platelet counts from baseline with higher rilzabrutinib exposure (AUC,ss).” On the dichotomy, “response rate as defined by the primary endpoint of LUNA 3, i.e., durable platelet response, did not appear to depend on exposure over the range at 400 mg BID” (p. 26, Table 10 by AUC quartile). Exposure did not predict time to first rescue or time to dropout. The drug-interaction section uses the model quantitatively: “30% rilzabrutinib exposure decrease would lead to a reduction in platelet count response by approximately 15%” (p. 63).

How the Phase 2 dose was chosen, and it was not by this model. Part A of LUNA 2 was “intrapatient dose escalation using a 3+3 design” over 200 mg QD, 400 mg QD, 300 mg BID and 400 mg BID, escalating every 28 days on non-response (p. 60). The 400 mg BID choice rests on three things: mouse BTK occupancy of 91% at the top dose “matching the expected peak occupancies with the maximum 400 mg QD and 400 mg BID doses”; human BTK occupancy reaching a plateau of about 90% at 600 mg/day and above; and the raw Part A observation that 400 mg BID “was also associated with a faster and more pronounced increase in platelet count compared to lower doses” (pp. 60-61). The longitudinal model was built afterwards, on data that were almost all at that one dose, and its stated objective was “to confirm the adequacy of the 400 mg BID regimen” (p. 164).

What this does to the specification. Section 6.3’s question, whether the depletion-to-platelet link is identifiable, has an existence proof at \(n = 305\) pooled across two trials at one dose. It does not have one at \(n \approx 40\) across several doses, and the model here was not asked to choose between doses. The model structure is not in the public review (“indirect exposure-response model” is all it says); the Applicant’s report SAN3001 is cited and not published.

23. FDA multidisciplinary review, Tavalisse (fostamatinib), NDA 209299, 2018 ✅

209299Orig1s000MultidisciplineR.pdf, clinical pharmacology questions (pp. 69-70, 212-213) and pharmacometrics appendix (pp. 216-217).

The one continuous PK-platelet attempt on a destruction-blocking drug that is documented as having failed, in the reviewer’s words. “Due to the small number of subjects and high variability in the platelet count, the relationship between exposure and probability of achieving the target increase at Weeks 14-24 or Week 24 could not be achieved. None of the applicant’s attempted models including semi-mechanistic, direct response, and indirect response model were successful for the platelet count analysis as a continuous variable” (p. 217).

What did work. Logistic regression of response at Week 12 on daily AUC at steady state, 79 subjects (27 placebo, 52 active at 200 or 300 mg/day), from the two Phase 3 studies. Probability of platelet count \(\ge 50 \times 10^9\)/L at Week 12 rose with R406 exposure; no covariate was significant. The reviewer read both 100 mg BID and 150 mg BID as sitting “in the rising part of the E-R curve” and raised a higher starting dose for patients with low baseline counts, noting baseline platelets of about 22,000/µL in those who stayed at 100 mg BID against about 15,000/µL in those escalated to 150 mg BID (p. 70). Weeks 14-24 had \(N = 33\) and Week 24 had \(N = 27\), and neither showed a relationship.

Two numbers the specification can use. Mean daily AUC at steady state 11,192 ± 5,226 ng·h/mL at a mean daily dose of 218 mg, “close to the estimated EC50 for inhibition of Syk signaling in healthy volunteers” (p. 69). And the exposure-safety model for blood pressure: \(E_{\max}\) 10.2 mmHg systolic, EC50 16,300 ng·h/mL, “similar with” the daily AUC at 150 mg BID (p. 217). Efficacy and the one dose-limiting side effect have their half-maximal exposures within a factor of 1.5 of each other, which is the narrow window that a titrate-to- response label expresses.

24. EMA assessment report, Tavlesse (fostamatinib), EMA/CHMP/654949/2019 ✅

tavlesse-epar-public-assessment-report_en.pdf, pp. 52-54 and 76-77.

The same data, read the other way. The FDA reviewer above called the Week 12 relationship “notable” and let it support effectiveness. The CHMP: “The population PK/PD analysis did not demonstrate an exposure-efficacy relationship. Higher exposure on average was evident in responders compared to non-responders only for Week 12, not between Weeks 14 - 24 or Week 24. The variability in response to fostamatinib in the ITP patients is not attributed to PK variability” (p. 76). And in the discussion: “Dose exposure response was addressed by a population PK/PD approach, which is however not considered to add high value to the separate study assessments … Overall, a positive relationship between exposure and efficacy has not been demonstrated” (p. 76).

The sentence the specification’s premise rests on. “It was not possible to fit a continuous PK/PD model for the platelet count-time profiles” (p. 53). The efficacy analysis was linear regression at Week 12 with no inter-individual variability estimable; 102 ITP patients on active drug with three PK samples each, at Weeks 2, 6 and 24; probability of response at Week 12 about 33% at 150 mg BID (p. 53).

Why the two agencies disagree on the same 79 patients. The FDA analysis was cross-sectional at one visit and found a slope; the EMA asked whether the slope survived to the primary endpoint window and it did not. Both readings are of the same fact: exposure separates responders from non-responders early and not late, which is what a mixture with a fast responder subgroup produces. Neither agency had a model that could say so.

25. PMDA review report, Vyvgart for intravenous infusion, primary ITP, 2024 ✅

Report on the Deliberation Results, 6 March 2024, Sections 9.2.4 (pp. 8-9), 6.R.1 (pp. 9-10), and 7.R.5.1 (p. 50). The only regulatory approval of an FcRn blocker in ITP, so the only review of one.

argenx built the IgG-to-platelet model first, at Phase 2, and it did not carry the dose decision. The applicant’s rationale as recorded by PMDA: “A model was created using data from Study 1603 to describe the relationship between the serum total IgG concentration following administration of efgartigimod alfa and change in platelet count. However, there was marked variation in the interindividual platelet count; in addition, it was difficult to fully explain the interindividual variation based on the data from Study 1603. Therefore, there were limitations to select the dosage regimen of efgartigimod alfa for Study 1801 on the basis of the simulation of platelet count change” (p. 9). Study 1603 is the Phase 2 with 5 and 10 mg/kg arms. The Phase 3 dose was chosen on maximum percent reduction in total IgG, and PMDA accepted that as “rational” given the mechanism (p. 10).

The model that was submitted stops at IgG. Population PK/PD in ITP, \(N = 131\) with 2,487 timepoints from Study 1801: three-compartment PK, an effect compartment, and an indirect response model on total IgG with an \(E_{\max}\) function of effect-site concentration on \(k_{\rm out}\). The base model came from healthy subjects and myasthenia gravis. Every covariate in the base model was dropped in the ITP update (p. 8).

One number for Section 19 reason 4. Sustained platelet response over Weeks 16 to 23 by dosing frequency in the efgartigimod arm: 22.8% (13 of 57) on weekly dosing, 90.0% (9 of 10) on every-other-week dosing (p. 50). The every-other-week group is the patients whose platelets allowed the switch, which is titration to response producing an inverted dose-response in the raw data.

Where this leaves Section 7 of the specification. This is the closest documented attempt at the specification’s exact task, one link further down the chain than the specification proposes, with a measured intermediate (total IgG) rather than an unmeasured one, and it failed at Phase 2 sample size for the reason Section 12 names: the platelet response is a mixture, and a model of the mean does not explain the individuals.

PK-platelet models for drugs that raise platelet production

These are the models that exist, and the specification borrows its platelet layer from them. They are for thrombopoietin receptor agonists (TPO-RAs), which act on production; the table says what transfers and what does not.

TPO-RA PK-platelet models and what the specification takes from each

Drug, source	Data	Platelet structure	Drug effect	What transfers to Section 8.3	Reference
Eltrombopag in ITP, entry 26 ✅	88 ITP on drug, 67 on placebo, 6-week Phase 3	Precursor, 2 transit, blood; \(k_{\rm in}\), \(k_T\) fixed to healthy-volunteer values	Linear on \(k_{\rm in}\): 58% per µg/mL	The mixture (19% non-responders with slope 0), the faster \(k_{\rm deg}\) in ITP, the finding that ITP data alone cannot identify the maturation chain	Hayes et al., J Clin Pharmacol 2011;51:1403, doi
Romiplostim in ITP, entry 27 ❌	268 ITP within 12 months of diagnosis, 7 studies, 7,854 counts, 1-15 µg/kg, up to 3 years	Progenitor, 4 transit, circulating	Linear with exposure, plus a term in cumulative dose	The titration simulation; the cumulative-dose term is a warning that a TPO-RA effect is not stationary	Blood 2021;138(Suppl 1):4221, ASH
Romiplostim in healthy subjects, entry 6 ❌	32 subjects, single doses 0.1-10 µg/kg	10 age compartments each for precursors and platelets; receptor-mediated clearance with receptors proportional to platelet count	Hill on \(k_{\rm in}\): \(S_{\max}\) 11.2, SC50 0.052 ng/mL	Precursor lifespan 5.9 d and platelet lifespan 10.5 d; the clearance-depends-on-platelets structure is the TPO-RA version of Section 11’s confounding	Wang et al., AAPS J 2010, PMC2976980
Eltrombopag in chemotherapy thrombocytopenia and liver disease, entry 5 ❌	Non-ITP	Precursor, 2 transit, blood	Linear on production	A second parameter set for the same structure	Cancer Chemother Pharmacol 2013, doi; Br J Clin Pharmacol, PMC4371540
Lusutrombopag, avatrombopag, Related section ❌	Healthy, liver disease, ITP protocols	Same lifespan family	Proportional to concentration	Nothing new	Clin Pharmacokinet 2016, doi

26. Hayes et al. 2011 — eltrombopag population PK/PD in ITP, with response-guided dosing ✅

J Clin Pharmacol 2011;51(10):1403-1417. doi:10.1177/0091270010383019 · PAGE 2009 poster · PAGE 2012 talk

The poster was read in full on 2026-09-09; the numbers below are from its Table 1. The paper itself has not been opened.

This is the model the specification’s Section 8.4 mixture reinvents, and it was published fifteen years ago. 88 ITP patients on eltrombopag and 67 on placebo from a 6-week Phase 3; a lifespan model with one precursor compartment, two transit compartments and blood; eltrombopag concentration stimulates precursor production linearly. “Mixture model was introduced to account for non-responders”: 81.1% of patients respond (P1 = 0.811, 6.1% RSE), with a 58% increase in \(k_{\rm in}\) per µg/mL (SLOP 0.579 mL/µg, 21.1% RSE); “non-responders (19%) had SLOP = 0.”

And it is the evidence for Section 8.3’s decision to fix the platelet parameters. “Sparse and/or variable nature of the PLTC data for ITP patients was not sufficient to estimate KIN and KT uniquely. KIN and KT population estimates were fixed to the estimates from a PK/PD model developed in healthy volunteers, 1.43 Gi/L·hr and 0.0253 hr⁻¹.” The platelet degradation rate was estimated and was higher in ITP than in healthy volunteers. Residual proportional error 44.6%.

Covariates on the slope. Females (×2.42) and older patients (\((\text{age}/50)^{1.27}\)) were more sensitive; East Asian patients had an 8 Gi/L higher baseline. Median baseline 16 Gi/L (range 1-40).

What the model was used for. “90% of steady-state platelet response was achieved by week 2, supporting biweekly individual dose adjustment,” and simulated titration rules that bound the fraction of patients above 200 Gi/L or below 50. Simulated 12.5, 25, 50 and 75 mg daily gave 73%, 120%, 220% and 320% increases from a 15 Gi/L baseline. A dose selection made from a PK-platelet model, in ITP, with a mixture, on 155 patients, for a drug with a two-week response lag.

27. Romiplostim population PK/PD in ITP within 12 months of diagnosis ❌

Blood 2021;138(Supplement 1):4221. ASH · ScienceDirect

Both hosts return 403 to a script; the characterization is from a search summary.

The question. How the cumulative-dose term is written and why it was needed. A drug effect that depends on how much drug has been given, and not only on current concentration, is a departure from the lifespan model that the specification should know about before reusing the structure.

What a search summary says, unverified: 268 adults with ITP within 12 months of diagnosis pooled from 7 studies; 7,854 platelet counts over up to 3 years at 1 to 15 µg/kg weekly; a drug-sensitive progenitor compartment, 4 maturation compartments and a circulating compartment; romiplostim raising precursor production “both linearly with drug exposure and as a function of cumulative administered dose”; used to confirm the label’s dose and response-guided titration for this population.

Animal PK-platelet models for drugs that block destruction

One group, Balthasar’s at Buffalo, built the mechanism this project needs in rodents twenty years ago, for intravenous immunoglobulin (IVIG), whose anti-platelet effect runs through the neonatal Fc receptor. These are the mechanistic ancestors of the efgartigimod model in entry 4, and the only published models found in which a drug that blocks platelet destruction is linked to platelet count by an explicit autoantibody state.

28. Hansen and Balthasar 2003 — IVIG and antiplatelet antibody disposition in a rat model of ITP ❌

J Pharm Sci 2003;92(6):1206-1215. PubMed 12761810

The question. The form of the indirect response model linking 7E3 concentration to platelet count, and its parameters. This is \(A(t)\) driving \(k_{\rm dest}\) in Section 8.3, with \(A\) measured.

What the abstract says. A mechanism-based PK model “based on competition between IgG molecules for occupancy of the protective FcRn receptor” for the effect of IVIG on anti-platelet antibody 7E3 clearance, checked in rat and in wild-type and FcRn-knockout mice; an indirect response model from 7E3 concentration to platelet count; about 50% of IVIG’s protective effect in this acute model attributed to faster antibody disposition.

29. Deng and Balthasar 2007 — IVIG in a murine model of sustained ITP ❌

J Pharm Sci 2007;96(6):1625-1637. doi:10.1002/jps.20828

The question. The same as entry 28, in the sustained model, which is the one that resembles chronic ITP. The companion is Jin, Tayab and Balthasar 2006, AAPS J, PMC2750959, high-dose monoclonal antibody in the rat model, which reports clearance and platelet nadir without a formal PD model.

What the abstract says. MWReg30 induces sustained thrombocytopenia in mice; FcRn-competition PK for the effect of IVIG on MWReg30 clearance; “an indirect response model” for MWReg30 concentration to platelet count; 43 ± 5% of the IVIG effect on thrombocytopenia accounted for by increased MWReg30 clearance, the rest by something else. That remainder is the reason the specification does not treat autoantibody concentration as the whole of the destruction signal.

Published B-cell depletion PK/PD models, for Section 6.2

Found on 2026-09-09 in answer to whether \(EC_{50}\) is ever estimable in an indirect-response model of B-cell depletion. Numbers were extracted from the full papers by a fetch tool rather than read by eye, so each carries ⚠️: the paper is the source, and the transcription has not been checked.

What each model could and could not estimate

Drug, entry	Patients, doses	Drug effect form	\(EC_{50}\) outcome	Reference
Ofatumumab, 32	1,486; 3 to 700 mg, five studies, two arms with partial depletion	Sigmoid \(E_{\max}\) on lysis within TMDD	Estimated, 0.0057 mg/L, 8.5% RSE; \(E_{\max}\) 159	Yu et al., CNS Drugs 2022, doi:10.1007/s40263-021-00895-w
Rituximab paediatric, 31	63; full dose; 281 CD19 counts, 52% below LOQ	\(E_{\max}\) on degradation	Estimated, 0.69 mg, 61% RSE, CI 0.18 to 2.26; \(E_{\max}\) 155	Pharmaceutics 2023;15:2534, doi:10.3390/pharmaceutics15112534
Inebilizumab, 33	213; 0.1 to 10 mg/kg and 30 to 600 mg	Log-linear slope on removal	Abandoned: below PK LLOQ	Yan et al., Br J Clin Pharmacol 2022, doi:10.1111/bcp.15332
Rituximab nephropathy, 34	41; 375 mg/m² × 4, 1000 mg × 2, 100 mg monthly	TMDD, elimination of the drug-CD20 complex	Binding constant 60% RSE	PMC11002205

31. Rituximab CD19+ suppression, population PD in paediatric patients ⚠️

Pharmaceutics 2023;15(11):2534. doi:10.3390/pharmaceutics15112534 · PMC10674351

What it estimated. Turnover model on CD19+ with rituximab amount raising the degradation rate through \(E_{\max} \cdot A / (ED_{50} + A)\); a linear form fit significantly worse and a sigmoid form no better. 63 patients, 281 CD19 measurements, 52% below the quantification limit. \(E_{\max}\) 155 (RSE 23%, CI 153 to 713), \(ED_{50}\) 0.692 mg (RSE 61%, CI 0.177 to 2.259), \(k_{\rm deg}\) 0.004 per day (RSE 22%), CD19 half-life 173 days, rituximab half-life 11.6 days. Nothing fixed. The authors: “the studied dose-range and the highly efficacious doses may have contributed to hampering a precise estimation.”

What it says for Section 6.2. At full dose only, \(E_{\max}\) and \(ED_{50}\) are formally estimable and practically not: a 13-fold interval on \(ED_{50}\) and an upper bound on \(E_{\max}\) four times the point estimate is the ratio being identified and the two factors not.

32. Ofatumumab population PK-B cell model in relapsing MS ⚠️

CNS Drugs 2022. Yu H, Graham G, David OJ, et al. doi:10.1007/s40263-021-00895-w · PMC8927028

What it estimated. Quasi-steady-state TMDD for PK; indirect response on B cells with a sigmoid \(E_{\max}\) stimulation of lysis by total ligand concentration. Five studies, 1,486 patients: 100 to 700 mg IV, the MIRROR Phase 2 at 3, 30 and 60 mg every 12 weeks and 60 mg every 4 weeks, and the 20 mg monthly Phase 3 regimen. \(EC_{50}\) 0.0057 mg/L (RSE 8.5%), \(E_{\max}\) 159 (RSE 3.2%), \(k_{\rm out}\) 0.0124 per day (RSE 5.2%), baseline 194 cells/µL. \(K_D\) and \(k_{\rm off}\) fixed to preclinical values. “The 10-mg dose did not appear to achieve a similar level of depletion as the 20-mg dose until after 6 months.”

What it says for Section 6.2. This is the case where \(EC_{50}\) is identified, and it took two arms with partial depletion, quantifiable PK through the transition, and 1,486 patients. The \(E_{\max}\) of 159 matches entry 31’s 155, which is the number that puts half-maximal depletion two orders of magnitude below \(EC_{50}\).

33. Inebilizumab PD modelling and exposure-response in NMOSD ⚠️

Br J Clin Pharmacol 2022. Yan L, Wang B, She D, et al. doi:10.1111/bcp.15332 · PMC9545531

What it estimated. A haematopoietic transit model: pro-B influx, ageing compartments for CD20+ cells, and inebilizumab accelerating removal from every compartment. 213 subjects across 0.1 to 10 mg/kg and 30 to 600 mg. “The estimated EC50 … was less than the assay lower limit of quantitation due to the high potency of inebilizumab,” so the drug effect became \(k = \text{slope} \cdot \log C\): slope 0.0140 (SE 0.0006), lifespan 391 days (SE 50), baseline 135 cells/µL. Between-patient CV: slope 52%, lifespan 126%, \(k_{\rm CD19}\) 110%; proportional residual error 61%.

What it says for Section 6.2. A 100-fold dose range did not rescue \(EC_{50}\), because the informative concentration band was below the PK assay. The log-linear slope is the form Section 8.2 adopts. The 126% CV on lifespan is the number behind Section 7’s caution on how much dose signal \(T_{\rm rep}\) carries per patient.

34. Rituximab population PK/PD in primary membranous nephropathy ⚠️

PMC11002205

What it estimated. Quasi-steady-state TMDD with CD20 synthesis and degradation and elimination of the drug-CD20 complex standing in for lysis; no separate \(E_{\max}\). 41 patients, 220 CD20+ counts, sampled before each infusion. The binding constant \(k_{ss}\) 6.21 per µmol at 60% RSE. Simulated 100 mg monthly against 375 mg/m² × 4 and 1000 mg × 2, and found comparable depletion with longer duration at a quarter of the cumulative dose.

What it says for Section 6.2. The TMDD route to the same question, with the same result: the potency-like constant is poorly identified at therapeutic doses. The low-dose simulation is the rituximab version of the plateau argument in Section 7.

Assay sources for Section 8.5

30. Anti-platelet antibody assays in ITP: MAIPA and platelet-associated IgG ❌

No source retrieved. The characterization in Section 8.5 of the specification, that MAIPA against GPIIb/IIIa and GPIb/IX is specific for ITP and positive in roughly half to 60% of patients, that it is semi-quantitative, and that platelet-associated IgG is non-specific, is from general knowledge and carries no citation. The entry exists so that it gets one.

The question. Three numbers with sources: MAIPA sensitivity and specificity in a cohort with a clinical diagnosis of ITP; whether any trial of a B-cell depleting agent in ITP measured it serially; and whether a quantitative version exists. The 2019 American Society of Hematology guideline and the IWG bleeding and diagnosis papers are the places to start.

Sources the specification’s numbers come from

Every number the specification quotes from outside itself. The four ✅ rows came from the paper; every other row came from a search summary of the source.

What the specification borrowed

Status	Number	Used in	Source
✅	IWG response criteria: CR \(\ge 100\), R \(\ge 30\) with 2-fold increase, both with absence of bleeding	Section 3.1	Entry 1
✅	Disease phases: newly diagnosed \(<3\) months, persistent 3 to 12, chronic \(>12\)	Section 3.2	Entry 1
✅	IWG trial-adapted primary and secondary endpoints	Section 3.3	Entry 1
✅	Rituximab time to initial response 7 to 56 d, to peak 14 to 180 d	Sections 3.5, 8.4	Entry 1
❌	Endpoint definitions used by LUNA3, ADVANCE IV, VAYHIT2, orelabrutinib and rilzabrutinib Part A	Section 3.4	Entries 15, 3, 14, 17, 16
✅	Rilzabrutinib: 305 patients, 10,652 platelet counts, indirect E-R model fit; continuous endpoint rose with exposure, durable response flat	Sections 6, 12, 19	Entry 22
✅	Fostamatinib: no continuous platelet model fit; Week 12 logistic slope on 79 patients; both doses on the rising part	Section 19	Entries 23, 24
✅	Efgartigimod: Phase 2 IgG-to-platelet model could not explain between-patient variation; dose chosen on IgG	Section 19	Entry 25
✅	Eltrombopag in ITP: 19% non-responders with slope 0; \(k_{\rm in}\) and \(k_T\) not identifiable from ITP data, fixed to healthy-volunteer values	Sections 8.3, 8.4	Entry 26
❌	Romiplostim precursor lifespan 5.9 d, platelet lifespan 10.5 d	Section 8.3	Entry 6
❌	Ianalumab 12-month freedom from treatment failure 51% at 3 mg/kg, 54% at 9 mg/kg, 30% placebo	Sections 7, 16	Entry 14
❌	Rituximab pooled response 63% at 100 mg weekly against 69% at 375 mg/m² weekly	Sections 7, 16	Entry 18
❌	Ianalumab half-life 9 to 10 days; two-compartment linear PK	Section 8.1	Entry 2
❌	Four-compartment platelet lifespan structure: one precursor, two transit, one blood	Section 8.3	Entry 5
❌	Platelet lifespan 7 to 10 days	Section 8.3	Entry 6
❌	Rituximab response 60% short-term, 20 to 30% at five years	Section 8.4, \(\phi\) prior	Entry 9
❌	Single-dose-level designs carry a high probability of false-positive exposure-response	Section 11	Entry 19
❌	IPF elevated above 10% in ITP	Section 9, requirement 10	Entry 21
❌	Ten patients in the VAYHIT3 interim, best platelet count range 3 to 709 G/L	Section 1 sample sizes	Entry 11

Searched for and not found

Recorded so that the search is not repeated, and so that a later find can be dated. Each line is a search that returned clinical results rather than a model.

Searches run on 2026-09-09 and what came back

Looked for	Result
PK-platelet model for any B-cell depleting agent in ITP	Nothing found. This is the project’s gap
PK-platelet model for any FcRn, SYK or BTK inhibitor in ITP, in journals	Nothing found. The efgartigimod model stops at IgG (entry 4)
The same, in regulatory reviews	Found: rilzabrutinib FDA 2025 fit one (entry 22); fostamatinib FDA 2018 and EMA 2019 record the attempt failing (entries 23, 24); efgartigimod PMDA 2024 records the Phase 2 attempt failing (entry 25)
PK-platelet model for a destruction-blocking drug, any species	Found, in rodents: Balthasar group, IVIG via FcRn, 2003-2007 (entries 28, 29)
Population PK/PD analysis of ianalumab in ITP	Nothing found. The Sjögren’s work is entry 2
Obexelimab, povetacicept or telitacicept in ITP	Nothing found. Obinutuzumab in ITP returned only drug-induced thrombocytopenia case reports
Published exposure-response for rilzabrutinib or orelabrutinib in ITP	Nothing found beyond the trial reports in entries 15 to 17
Any exposure-response against platelet count in ITP	Three: fostamatinib (entries 10, 23, 24), rilzabrutinib (entry 22), efgartigimod at Phase 2 (entry 25)
Quantitative systems pharmacology model of ITP	Nothing ITP-specific. B-cell QSP models exist for mouse and for rheumatic disease

Related, not yet placed

• B-cell quantitative systems pharmacology models. QSP of B-cell immune response in mouse, Front Immunol 2026 and translating B-cell and plasma-cell targeting from oncology to autoimmunity, PMC13338978. Out of scope by the specification’s own statement, and the place to go if Section 17’s tissue-depletion failure mode is confirmed.
• T-DM1 thrombocytopenia model, Cancer Chemother Pharmacol 2012, PubMed 22886072. A semimechanistic transit-compartment platelet model for drug-induced thrombocytopenia. The same structure driven in the opposite direction, and a source of a second parameter set for Section 8.3.
• Avatrombopag and lusutrombopag population PK/PD. Two more instances of the entry 5 structure, one of them in healthy subjects, Clin Pharmacokinet 2016. Read only if entry 5’s parameters turn out not to transfer to ITP.
• Single-cell evidence for multiorgan dissemination of autoreactive plasma cells in ITP, bioRxiv 2021. The biology behind Section 7’s tissue argument and Section 17’s first failure mode.
• FDA reviews for the TPO-RAs: Promacta, Nplate, Doptelet, Mulpleta. Entries 22 to 25 covered the destruction-blocking drugs; the production drugs’ reviews would add parameter tables to the models in entries 5, 6, 26 and 27, and nothing to the project’s premise.
• The rilzabrutinib exposure-response report SAN3001, cited in entry 22 and not public. The model structure is there and not in the review.