Planning hospital clinical research capacity: a cross-sectional study of unit staffing and early-phase trial activity in Spain.

Reproducibility repository for the study "Organisational capability and early-phase clinical-trial capacity in Europe: evidence from Spain's national Clinical Trial Unit system". It contains aggregated analytical datasets, a data dictionary, the analysis code (Python), the numeric results (regression, sensitivity and diagnostics as JSON) and the final manuscript figures, for a national observational study of 40 Clinical Trial Units hosted by 38 Spanish hospitals and their host-hospital early-phase clinical-trial activity (2016–2026). Only aggregated datasets derived from publicly available sources are shared; bulk registry downloads are not redistributed. Code is released under the MIT License; the aggregated data under CC BY 4.0. Changes in version 1.1.0 (supersedes 1.0.0, 3 August 2026) The headline result is unchanged: IRR 1.13 (95% CI 1.03–1.25; p=0.013) per additionaldeclared dedicated trial coordinator, on 37 host hospitals. Corrections that change deposited numbers- Functional-form diagnostics (scripts/07_diagnosticos.py, results/diagnosticos.json): the tertile model used cumulative indicators, so the coefficient labelled "T3 vs T1" was in fact T3 vs T2. It is now parameterised with T1 as the reference category, in line with the equivalent model in 04_primary_models.py. The corrected contrast is IRR 2.74 (1.05–7.11; p=0.039); the AIC (368.62) and the conclusion — tertiles do not improve on the linear term — are unchanged.- Supplementary Figure S5 (scripts/make_figureS5.py): rebuilt at host-hospital level (n=38) instead of CTU level, and the oncohaematology share is now computed over classifiable records rather than all records. Previously reported correlations (rho=0.44 absolute, 0.56 percentage) were inflated by title coverage; the corrected values are rho=0.31 (p=0.059) and rho=0.44 (p=0.006). The superseded estimates and the coverage diagnostics are retained in figuras_descriptivas.json. Analyses added or extended- Founding-year-restricted outcome with a log(years active) offset, so numerator and observation time cover the same period (IRR 1.16; p=0.006); the earlier variant is retained as P1-offset-crudo.- Clustering is now run with four algorithms (Ward, k-means, Gaussian mixture, spectral) over k=2–6 with a fixed seed, rather than Ward alone. No configuration is well separated (maximum mean silhouette 0.28, at k=2).- Supplementary Figure S6 redrawn: the two REec queries and the ClinicalTrials.gov export are shown as three parallel extraction streams, none derived from another. Reproducibility and provenance- Figures 2 and S5, previously drawn by hand, now have scripts and read their values from the pipeline; every figure in the article and its supplement is regenerated by a single command.- Figures 3 and 5 rebuilt to journal specification (180 mm width, no titles inside the artwork, p values to two significant figures).- PDF output made byte-reproducible (SOURCE_DATE_EPOCH), so repeated runs are identical.- 11_verificar_texto.py extended: it now checks around 90 figures and strings from the manuscript and supplement against the manifest, and blocks previously corrected wordings from reappearing.- Deposited workbooks cleaned: analisis_exploratorio_ctu.xlsx reduced to the two sheets the pipeline reads plus a deposit index; Ensayos_REec_por_CTU.xlsx relabelled so that full-portfolio columns are no longer described as window-restricted. Changes of this version are in CHANGELOG_v1.1.1.md