Publishing the 3 RECIST Lesion Model

Working specification

Pharmacometrics
What the 2011 ASCO poster on the RECORD-1 three-lesion tumor model holds, the choice of preprint server, what a manuscript needs beyond the poster, and the steps to posting.
Last edited

October 14, 2026

1. Goal

A preprint with a DOI describing the RECORD-1 three-lesion tumor model, written from the 2011 ASCO poster (abstract 4602), with the five poster authors or their stated consent.

Out of scope: refitting the model or any new analysis of RECORD-1 data. The preprint reports what the 2011 analysis found. Section 5 records the decision this rests on.

2. The Poster

Stein AM, Carter A, Hollaender N, Motzer R, Sarr C. Quantifying the effect of everolimus on both tumor growth and new metastases in metastatic renal cell carcinoma (RCC): a dynamic tumor model of the RECORD-1 phase III trial. J Clin Oncol 29, 2011 (suppl; abstr 4602). ASCO Annual Meeting, Chicago, 3–7 June 2011. PDF.

Everything in this section was transcribed on 2026-10-13 from text extracted from the PDF, whose multi-column layout scrambles the order. Each item is ⚠️ until checked against the poster image.

2.1 Data

RECORD-1 (NCT00410124) randomized patients with metastatic RCC who had progressed on sunitinib or sorafenib to everolimus 10 mg daily or placebo. Dose reduction to 5 mg daily was allowed for toxicity. The analysis used local radiology for the 407 patients with a baseline tumor assessment: the sum of longest target lesion diameters (SLD), nontarget lesion progression status, and new lesion appearance.

2.2 Model

Target lesions follow, by nonlinear mixed effects,

\[ \frac{dy}{dt} = (r - E_\text{dose})\,y \]

where \(y\) is SLD, \(r\) the placebo growth rate and \(E_\text{dose}\) the dose-dependent everolimus effect. The extracted text reads dy/dt = r – Edosey, so the bracketing is a guess to check against the image.

Nontarget and new lesions carry no measured size, so the model assumes each grows by the same equation and maps the latent size to a probability of progression.

  • Nontarget lesions. With \(\Delta y\%\) the percentage change from baseline, \[ f(\Delta y\%) = \frac{(100\% + \Delta y\%)^8}{(150\%)^8 + (100\% + \Delta y\%)^8} \] A 50% increase gives a probability of 0.5 and a 100% increase gives 0.91.
  • New lesions. With \(y\) the longest diameter of the new lesion, starting undetectable at 0.1 cm, \[ f(y) = \frac{y^3}{(0.5\ \text{cm})^3 + y^3} \] A lesion at the 0.5 cm imaging resolution is detected with probability 0.5, and one at 2 cm with probability 0.984.

The 5 mg and 10 mg effects were fitted as separate parameters for target lesions. For nontarget and new lesions the data did not support two, and one effect was shared. Correlating the placebo growth rates \(r\) across the three lesion types improved the fit (P < .001). The three lesion predictions were combined by the RECIST rules into a predicted time to progression for each patient.

2.3 Results

Poster results, transcribed ⚠️
Quantity Placebo 5 mg 10 mg
Change in SLD after 1 year, average patient +142.1% ± 98.3% +22.4% ± 17.2% −15.7% ± 11.5%
Nontarget progression after 1 year (90% CI) 95.0% (81.8, 99.5) 64.3% (33.2, 90.0), 5 and 10 mg pooled
New lesion progression after 1 year (90% CI) 90.1% (76.1, 96.8) 52.6% (20.5, 82.4), 5 and 10 mg pooled

Both doses slowed all three lesion types against placebo (P < .0001), and 10 mg slowed target lesions more than 5 mg (P < .0001). Predicted progression-free survival matched observed in both arms (Figure 4 of the poster), with everolimus patients simulated on their actual dosing history, placebo patients simulated without crossover, and deaths from causes other than progression excluded.

2.4 Figures

# Shows
1 Six patients, A–F: SLD, nontarget progression probability and new lesion probability over time, with dose
2 Simulated SLD over 1 year in the average patient, by dose
3 Probability of nontarget (A) and new lesion (B) progression over 1 year
4 Predicted against observed progression-free survival, everolimus (A) and placebo (B)

3. Choice of Server

Candidate servers. The bioRxiv and medRxiv rows are from memory, not checked against either site’s current policy
Server Takes this? DOI Notes
bioRxiv ⚠️ Probably not Yes Directs clinical trial analyses to medRxiv
medRxiv ⚠️ Probably Yes Screens submissions; asks for trial registration, ethics, funding and competing-interest statements, and an author declaration that all authors agree
Zenodo ✅ Yes Yes No screening; posts anything, including the poster PDF itself

Recommendation: medRxiv, with Zenodo as the fallback if medRxiv screening turns it away on the grounds that it reports a 2011 analysis that cannot be reproduced. Posting the poster PDF itself to Zenodo now is a separate, cheaper step that gives it a DOI the day it is done, and does not preclude either.

4. What the Manuscript Needs Beyond the Poster

  1. Methods detailed enough to reimplement: the estimation software, the random effects and their covariance, the residual error models, how the nontarget and new lesion probabilities entered the likelihood, how the dose history entered \(E_\text{dose}\), and parameter estimates with standard errors. None of these are on the poster. They exist only in the 2011 analysis files, if anywhere.
  2. A statement on the prior publications from the same analysis, Stein 2012 and Stein 2013 (Section 6), saying what this one adds, and a related work section drawn from Sections 3, 4 and 9 of the literature review.
  3. Author consent from all five, and Novartis clearance. RECORD-1 is a Novartis trial, and three authors were Novartis employees.
  4. Trial registration, ethics and funding statements, carried over from the RECORD-1 primary publication.
  5. A data availability statement. The individual data are Novartis’s.

5. Decision: Write Up or Refit

Option Needs Gives
A. Write up the 2011 analysis as it stands The poster, and the 2011 analysis files for item 1 of Section 4 A preprint in weeks
B. Refit on RECORD-1 data Data access from Novartis, possibly through a data-sharing platform A reproducible analysis, and months

Recommendation: A. Option B is a new project, and nothing in Section 1 requires it. If the 2011 parameter estimates cannot be found, A shrinks to a methods-and-figures note on Zenodo.

6. Open Questions

  1. ✅ Answered 2026-10-13: the three-lesion model has not appeared in a full paper. Stein 2012 (BMC Cancer 12:311) publishes the target lesion model alone and cites the poster as its reference 22 for the nontarget and new lesion work. Stein 2013 (Eur Urol 64:994) is a Cox analysis of overall survival with no dynamic model. The literature review found no later model that does what the poster does (its Section 9).
  2. Where are the 2011 model files, the control stream and the parameter table?
  3. Are the four co-authors reachable, and who at Novartis clears it?

7. Steps

  1. ✅ Answer open question 1 (done 2026-10-13).
  2. Post the poster PDF to Zenodo for a DOI.
  3. Contact the co-authors and Novartis (open question 3).
  4. Find the analysis files (open question 2).
  5. Check every ⚠️ in Section 2 against the poster image.
  6. Draft the manuscript from Sections 2 and 4.
  7. Submit to medRxiv, or Zenodo per Section 3.
Back to top