Reconstructed IPD durability benchmark in relapsed/refractory follicular lymphoma

Reconstructed IPD durability benchmark in relapsed/refractory follicular lymphoma

Reproducibility materials (digitized data, reconstructed individual patient data, analysis code, and outputs) for:

Burke OJ, Peruzzo N, Tul Ain Khan N, Kancharla P. Durability Benchmarking of Contemporary Second-Line and Later Therapies for Relapsed or Refractory Follicular Lymphoma Using Reconstructed Individual Patient Data From Published Kaplan-Meier Curves. Cureus. (year, volume, and DOI to be added on publication.)

Corresponding author: Oscar J. Burke ([e-mail]).

What this is

This repository contains everything needed to inspect and reproduce the durability benchmarking analysis reported in the article. No individual patient records were used. All inputs were digitized from the published Kaplan-Meier (KM) figures and numbers-at-risk tables of the cited pivotal trials, so only published, aggregate, de-identified data are included here.

The workflow has three stages:

Digitize each published KM curve with WebPlotDigitizer (Ankit Rohatgi; https://automeris.io/WebPlotDigitizer), exporting closely spaced (time, survival) coordinates after calibrating the plot axes to the published ranges. These exports are in curves/.

Reconstruct individual patient data (time, event/censoring indicator) from the digitized coordinates and the reported numbers-at-risk using the Guyot et al. algorithm implemented in the IPDfromKM R package. Outputs are in ripd/.

Analyze: compute landmark survival and restricted mean survival time (RMST), pool within modality classes and overall by random-effects meta-analysis, and fit a Cox model to the reconstructed ROSEWOOD arms as an internal validity check. Outputs are in results/.

Folder structure

FL_rIPD_reproducibility/
├── README.md
├── 01_reconstruct_ipd.R # stage 2: curves/ -> ripd/
├── run_ripd_benchmark.R # stage 3: ripd/ -> results/
├── curves/ # digitized KM coordinates (inputs)
├── ripd/ # reconstructed individual patient data
└── results/ # analysis outputs and figures
└── figures/

curves/ (digitized KM coordinates)

Two columns, no analysis performed yet:

time — time in months

surv — survival probability on a 0 to 100 scale

These are the WebPlotDigitizer exports for each trial, arm, and endpoint (PFS, OS, DoR, TTNT). File names indicate trial/regimen and endpoint, for example rosewood_pfs_zo_ripd.csv is the ROSEWOOD progression-free survival curve for the zanubrutinib-plus-obinutuzumab arm, and zuma_figB_PFS_ripd.csv is the ZUMA-5 PFS curve. Note: transcend_pfs_ripd_old.csv carries an "old" label inherited from the working project; confirm it is the intended TRANSCEND FL PFS digitization before relying on it (see "Notes" below).

ripd/ (reconstructed individual patient data)

Two columns, one row per reconstructed patient:

time — time to event or censoring, in months

status — 1 = event (progression and/or death, per the endpoint), 0 = censored

Reconstructed sample sizes match the reported analyzed cohorts (for example, transcend_pfs_ripd.csv has 101 rows for the TRANSCEND FL third-line-and-later cohort; zuma_pfs_ripd.csv and zuma_os_ripd.csv have 86 rows for ZUMA-5). The files provided here are representative outputs; the complete ripd/ set for every curve regenerates by running 01_reconstruct_ipd.R on curves/ (the script contains the corresponding numbers-at-risk and cohort sizes).

results/ (analysis outputs)

landmarks_by_regimen.csv — landmark survival at 12, 24, and 36 months per regimen and endpoint. Columns: t (landmark month), surv, se, lower, upper (0 to 1 scale), regimen, class, trial, endpoint, max_fu_m (maximum follow-up), n, and the same survival/CI values on a percentage scale (surv_pct, lower_pct, upper_pct).

rmst_by_regimen.csv — restricted mean survival time. Columns: regimen, class, trial, endpoint, max_fu_m, n, RMST_24, RMST_36 (months; NA where follow-up did not reach the horizon).

pooled_landmarks_overall.csv and pooled_landmarks_by_class.csv — random-effects pooled landmark estimates. Columns: endpoint, t, k (number of cohorts pooled), pooled_surv, pooled_low, pooled_high, tau2, I2, group. These are descriptive summaries and are not comparative effect estimates.

rosewood_internal_validation_hr.csv — reconstructed Cox hazard ratios for zanubrutinib plus obinutuzumab versus obinutuzumab in ROSEWOOD (internal validity check). Columns: endpoint, HR_ZO_vs_O, low95, high95, p, n.

figures/ — landmark PFS and OS forest plots at 12 and 24 months (forest_pfs_12m.png, forest_pfs_24m.png, forest_os_12m.png, forest_os_24m.png).

How to reproduce

Requirements: R (version 4.x recommended). The scripts auto-install the packages they use:

01_reconstruct_ipd.R uses IPDfromKM, readr, dplyr, tibble, purrr.

run_ripd_benchmark.R uses survival, metafor, ggplot2, dplyr, readr, tibble, scales.

From the repository root:

r

# 1) reconstruct IPD from the digitized curves
source("01_reconstruct_ipd.R") # reads curves/, writes ripd/

# 2) run the durability benchmark
source("run_ripd_benchmark.R") # reads ripd/, writes results/

Both scripts use paths relative to the working directory (curves/, ripd/, results/), so run them with the repository root as the working directory.

Source trials

Digitized from the published KM figures and numbers-at-risk tables of: ZUMA-5 (axicabtagene ciloleucel), ELARA (tisagenlecleucel), TRANSCEND FL (lisocabtagene maraleucel), the mosunetuzumab phase 2 study (GO29781), EPCORE NHL-1 (epcoritamab, follicular cohort), and ROSEWOOD (zanubrutinib plus obinutuzumab versus obinutuzumab). Full references are in the article.

Reconstruction method

N. Liu, Y. Zhou, J.J. Lee. IPDfromKM: reconstruct individual patient data from published Kaplan-Meier survival curves. BMC Med Res Methodol. 2021;21:111, implementing the algorithm of P. Guyot, A.E. Ades, M.J. Ouwens, N.J. Welton. Enhanced secondary analysis of survival data: reconstructing the data from published Kaplan-Meier survival curves. BMC Med Res Methodol. 2012;12:9.

Notes

These materials were assembled from the analysis project workspace. Before final deposition, confirm the set is complete against your local project, in particular that every reconstructed ripd/ file you intend to share is present (or regenerate the full set by running 01_reconstruct_ipd.R), and verify which TRANSCEND FL PFS curve file (transcend_pfs_ripd_old.csv or a newer digitization) produced the published values.

Only published, aggregate, de-identified data are included; no individual patient records were accessed, and institutional review board approval was not required.