Analysis code and de-identified weekly aggregated data for: Population-level measles RT-PCR cycle threshold values and their relationships with disease severity, transmission, and incidence metrics (Mexico, 2025-2026)

Analysis code and the de-identified weekly aggregated data needed to reproduce every result reported in the accompanying article, which asks whether the weekly median RT-PCR quantification cycle (Cq) value of notified measles cases precedes the incidence, transmission and severity metrics of an outbreak, and whether any such relationship survives adjustment for the secular trend of the epidemic wave, for the composition of the sampled specimens, and for the outcome's own past. The analysis covers the two largest outbreaks of Mexico's 2025-2026 measles epidemic, in Chihuahua and Jalisco. This deposit contains no individual-level records. Data Three weekly panels aggregated by state and epidemiological week, spanning 2025-02-03 to 2026-07-27: panel_semanal_por_ola.csv (82 rows) - weekly median Cq, case counts, rates and effective reproduction numbers, by state and wave. panel_semanal_gravedad_por_ola.csv (82 rows) - weekly hospitalization and pneumonia counts, rates and proportions. panel_composicion_semanal.csv (80 rows) - weekly demographic composition of the specimens contributing to each median. Everything else the analysis uses is derived from these three panels by running the code, and is deliberately not deposited: a derived table that ships pre-computed invites citation without regeneration. De-identification The source is Mexico's national epidemiological surveillance system, which is name-based. Two rules were applied before deposit. First, small-cell suppression: in any state-week contributing fewer than five Cq determinations, the three demographic composition fields are left empty, because a demographic profile computed over one or two specimens is an individual record wearing a summary's clothes; this affected 4 rows. Second, restriction to the cells the analysis actually consumes, which removes 62 rows whose (state, week) key the analysis discards at its left join, and which are overwhelmingly the historical one-case and two-case weeks. Code 31 R scripts. Eighteen of them run on the deposited data alone, including the entire chain behind the reported results: cross-correlation and lagged regression, the adjusted models, Toda-Yamamoto Granger causality and vector autoregression, the supplementary tables, and the out-of-sample predictive gain analysis. The remaining thirteen read the restricted individual-level extract and will not run from this archive; they are included because they are the record of how the deposited panels were produced. That extract is obtainable by any investigator through Mexico's national framework for transparency and access to public information. Running the documented chain on a clean copy of this archive reproduces every result file it touches. Nineteen come out byte-identical to the versions behind the published tables; the adjusted-regression table agrees to within 1.4e-11 in its worst cell, the precision lost when the composition covariates make a round trip through a text file rather than staying in memory. Every coefficient sign and every significance decision is unchanged. R 4.5.3 with data.table, ggplot2, patchwork, EpiEstim, sandwich, lmtest and tseries. Code comments are in Spanish. MANIFEST.txt carries the SHA-256 of every file. See README.md for the run order, which matters, and for the single documented path change that separates these scripts from the working copies. Licensing Data under CC BY 4.0; code under the MIT license, as stated in LICENSE-CODE.