Can synthetic data reproduce real-world findings in epidemiology? A replication study using adversarial random forests

This repository contains the code which produced the results of the manuscript “Can synthetic data reproduce real-world findings in epidemiology? A replication study using adversarial random forests” by Kapar et al. (2026).

Summary

The paper proposes the use of adversarial random forests (ARF) as a competitive, fast, and user-friendly method for synthesizing tabular epidemiological data. To evaluate its performance, statistical analyses from six epidemiological publications were replicated and the results on synthetic data compared to original findings. Additionally, the impact of dimensionality and variable complexity on synthesis quality was assessed by limiting datasets to variables relevant to individual analysis tasks, including necessary derivations.

Supplementary analyses included a comparison of the ARF the effects of the parameter min_node_size on synthesis quality, and a comparison of the ARF to other popular synthesis methods for tabular data. The trade-off between utility and privacy (measured by membership and attribute inference attacks) was assessed as well as the generalisation to unseen test data and the synthesis runtime.

Replicated original publications

The replicated original publications cover blood pressure, anthropometry, myocardial infarction, accelerometry, loneliness, and diabetes, and are based on data from the German National Cohort (NAKO Gesundheitsstudie), the Bremen STEMI Registry U45 Study (BSR-U45), and the Guelph Family Health Study (GFHS).

Publication list

  • Tamara Schikowski, Claudia Wigmann, Kateryna B Fuks, et al. Blutdruckmessung in der NAKO—methodische Unterschiede, Blutdruckverteilung und Bekanntheit der Hypertonie im Vergleich zu anderen bevölkerungsbezogenen Studien in Deutschland [Blood pressure measurement in the NAKO German National Cohort (GNC)—differences in methods, distribution of blood pressure values, and awareness of hypertension compared to other population-based studies in Germany]. Bundesgesundheitsblatt Gesundheitsforschung Gesundheitsschutz 2020;63(4):452–64. https://doi.org/10.1007/s00103-020-03109-8

  • Beate Fischer, Anja M Sedlmeier, Saskia Hartwig, et al. Anthropometrische Messungen in der NAKO Gesundheitsstudie—mehr als nur Größe und Gewicht [Anthropometric measures in the German National Cohort—more than weight and height]. Bundesgesundheitsblatt, Gesundheitsforschung, Gesundheitsschutz 2020;63(3):290–300. https://doi.org/10.1007/s00103-020-03096-w

  • Harm Wienbergen, Daniel Boakye, Kathrin Günther, et al. Lifestyle and metabolic risk factors in patients with early-onset myocardial infarction: a case-control study. Eur J Prev Cardiol 2022;29(16):2076–87. https://doi.org/10.1093/eurjpc/zwac132

  • Becky Breau, Hannah J Coyle-Asbil, Jess Haines, David WL Ma, and Lori Ann Vallis. ActiGraph cutpoints impact physical activity and sedentary behavior outcomes in young children. J Meas Phys Behav 2022;5(2):85–96. https://doi.org/10.1123/jmpb.2021-0042

  • Klaus Berger, Steffi Riedel-Heller, Alexander Pabst, Marcella Rietschel, Dirk Richter, and NAKO-Konsortium. Einsamkeit während der ersten Welle der SARS-CoV-2-Pandemie—Ergebnisse der NAKO-Gesundheitsstudie [Loneliness during the first wave of the SARS-CoV-2 pandemic—results of the German National Cohort (NAKO)]. Bundesgesundheitsblatt, Gesundheitsforschung, Gesundheitsschutz 2021;64(9):1157–64. https://doi.org/10.1007/s00103-021-03393-y

  • Justine Tanoey, Christina Baechle, Hermann Brenner, et al. Birth order, caesarean section, or daycare attendance in relation to child-and adult-onset type 1 diabetes: results from the German National Cohort. Int J Environ Res Public Health 2022;19(17):e10880. https://doi.org/10.3390/ijerph191710880

Data availablity

None of the data used in this work are openly available. Hence, this repository does not contain any data files for reproduction.

The NAKO data are not openly available due to data protection measures. However, scientists can apply for data access following the official usage regulations and upon formal request to the NAKO use and access committee (https://transfer.nako.de).

Participant data from the Bremen STEMI Registry U45 Study (BSR-U45) are not publicly available.

Due to University of Guelph Research Ethics Board restrictions and participant confidentiality, no GFHS participant data are publicly available. The GFHS welcomes outside collaborators. Interested investigators can contact GFHS investigators to explore this option, which preserves participant confidentiality and meets the requirements of the University of Guelph Research Ethics Board, to protect human subjects.

Repoistory organisation

  • Publication-specific code is organised within subfolders of /NAKO and /GFHS folders
  • Creation of full and task-specific original datasets: create_train.R (calls publication-specific code in publication folders)
  • Data synthesis with R: create_syn.R (global for all publicaitons)
  • Analyses replication and comparison: run_analyses.R (calls publication-specific code analysis.R and analysis_taskspec.R in publication folders)
  • Results are stored in /figures and /tables subfolders of publication folders
  • Supplementary analyses are stored in /Supplement

Requirements

Required R logo packages

For synthesis:

  • arf

For statistical analyses:

  • survival
  • zscorer
  • scales

For visualization:

  • ggplot2
  • ggh4x
  • xtable

For parallel processing:

  • doParallel
  • foreach

For data processing and console output:

  • data.table

For supplement:

  • synthpop
  • ddpm(PyPI)
  • sdv (PyPI)
  • ranger
  • pROC
  • caret
  • approxOT
  • philentropy
  • binom
  • microbenchmark