Clinically significant Candida in adults at a Mexican tertiary hospital, 2016–2025: de-identified dataset and analysis code

This deposit supports the article Decade-Long Dynamics of Clinically Significant Candida in Adults at a Mexican Tertiary University Hospital, 2016–2025: Incidence, Species Distribution, and Pandemic-Era Trends.

It contains de-identified, isolate-level records of clinically significant Candida disease, together with monthly aggregate counts, the monthly hospital denominators, the regional COVID-19 series used in the time-series analyses, and the full R analysis pipeline (code/). The data and code reproduce the species-distribution, fluconazole-susceptibility, incidence, time-series, and mortality results reported in the article.

Setting

Single centre: Antiguo Hospital Civil de Guadalajara "Fray Antonio Alcalde," Jalisco, Mexico. Isolates were collected from 2016 through early 2026; the longitudinal analyses cover 2016–2025. The data are retrospective and were collected during routine clinical care.

Files

File	Description
candida_invasive_isolates_deidentified.csv	One row per Candida isolate (3,505 isolates, 39 variables).
monthly_significant_counts.csv	Monthly counts of significant isolates by age stratum (for incidence/time-series).
monthly_hospital_denominators.csv	Monthly admissions and patient-days (total / adult / pediatric).
monthly_covid_jalisco.csv	Monthly laboratory-confirmed COVID-19 activity for Jalisco.
DATA_DICTIONARY.md	Variable-by-variable codebook for all files.
CITATION.cff	Machine-readable citation metadata.
CHECKSUMS.txt	SHA-256 checksums and expected dimensions for integrity checking.
LICENSE	Licence terms (CC BY 4.0).
code/	Full R analysis pipeline, helper functions, classification catalogs, and manuscript build scripts.

Code

The code/ folder documents the complete analytical pipeline:

• code/R/scripts/ — the numbered R pipeline (00–71) that runs import, cleaning, syndrome and cohort classification, incidence and time-series modelling, the COVID-coupling battery, the susceptibility shift-share decomposition, the mortality models, and figure/table generation.

• code/R/functions/ — shared helper functions.

• code/config/ — service- and specimen-classification catalogs used by the pipeline.

• code/scripts/ — manuscript and supplement build scripts.

The pipeline was executed on the institutional source records (raw line lists with direct identifiers), which are not released: under the ethics approval only the de-identified extract in this archive is shared. Paths in the scripts are relative to the project root; the scripts are provided for methodological transparency and review rather than as a turnkey re-run against the de-identified file. No generative-AI code-authoring tools were used to produce the analysis.

Cohorts and key counts

Each level of the funnel can be reconstructed from the variables in the main file:

• All isolates: 3,505.

• Clinically significant invasive disease (clinically_significant == TRUE): candidemia, other sterile-site invasive disease, or catheter tip; colonizing isolates excluded.

• Primary cohort — significant adults (clinically_significant == TRUE & age_group == "adult"): 1,386 (candidemia 935, other sterile-site invasive 327, catheter tip 124). Non-albicans species account for 61.5%; fluconazole non-susceptibility is 23.0% among the 1,280 isolates with a CLSI interpretation.

• Pediatric stratum (clinically_significant == TRUE & age_group == "pediatric"): 748 (742 within 2016–2025).

De-identification

The data carry no direct identifiers. Patient names and record numbers were removed and replaced by a synthetic record_id. To reduce the risk of re-identification in a single-centre dataset, quasi-identifiers were generalised: exact dates were reduced to calendar year (monthly detail is released only as aggregate counts); age was banded into decades with ages 80 and over top-coded; free-text specimen descriptions were replaced by controlled anatomical categories; species with fewer than 10 isolates were grouped; diagnosis and cause-of-death codes were limited to ICD-10 three-character roots, with rare roots suppressed; and exact length of stay and time-to-death were not released (30- and 90-day mortality indicators are provided instead). Demographic fields that had been reconstructed from patient names were dropped.

Residual re-identification risk from combinations of quasi-identifiers cannot be entirely eliminated for row-level, single-centre clinical data. Reuse is permitted on the condition that no attempt is made to identify individuals.

Reading the data

In R:

d <- read.csv("candida_invasive_isolates_deidentified.csv")
adults <- subset(d, clinically_significant & age_group == "adult")
nrow(adults)                                  # 1386
round(100 * mean(adults$non_albicans), 1)     # 61.5

Monthly incidence (adults), using the aggregate counts and denominators:

counts <- read.csv("monthly_significant_counts.csv")
den    <- read.csv("monthly_hospital_denominators.csv")
adult  <- merge(subset(counts, age_group == "adult"),
                subset(den, stratum == "adult"),
                by.x = "month_start", by.y = "month")
adult$candidemia_per_1000_pd <- 1000 * adult$n_candidemia / adult$patient_days