A Quarter-Century of Synthetic Data in Healthcare: Unveiling Trends with Structural Topic Modeling
Bibliographic record
Abstract
Abstract Data-driven approaches are transforming healthcare, yet acquisition of comprehensive datasets is hindered by high costs, privacy regulations, and ethical concerns. To address these challenges, synthetic data, artificially generated datasets that mimic the statistical properties of real-world data, provides a promising solution. Despite its growing adoption, the thematic landscape of synthetic data research in healthcare remains underexplored. Therefore, we applied structural topic modeling (STM) to map the research landscape of synthetic data in healthcare, revealing prevalent topics and tracking their evolution over time and across geographic locations. PubMed publications from 2000-2024 containing “synthetic data,” “artificial data,” or “simulated data” in the title/abstract were retrieved. After preprocessing the text (lowercasing, punctuation/stopword removal, stemming), structural topic modeling (STM) was performed using year and continent as covariates. The optimal number of topics (K=10) was determined using held-out likelihood and interpretability. Topic prevalence, temporal trends, and inter-topic correlations were analyzed using stacked area charts and network analysis. Analysis of 14,788 PubMed articles (2000-2024) revealed a tenfold increase in publications. Geographically, North America (48.6%) and Europe (33.5%) were primary contributors, but Asia’s share steadily rose from 2.9% to 23.1%. STM identified ten key topics, grouped into Biomedical Imaging & Signal Processing (25.2%), Synthetic Data Applications in Biomedical Research (17.7%), Computational & Statistical Methods (23.9%), and Genomics & Evolutionary Biology (33.2%) themes. We observed gradual declines in initially prominent topics including “Bayesian Modelling” (23.1% to 9.9%), “Neuroimaging” (16.0% to 9.3%), and “Image Simulation” (17.7% to 9.1%), giving ascendancy to “Synthetic Data Generation” (2.2% to 27.1%) and “Disease Modeling and Public Health” (4.8% to 11.9%) by 2024. Synthetic data research in healthcare has experienced increasing interest, marked by shifts in geographic distribution and dynamic evolution of key topics. Realizing the full potential of synthetic data requires fostering cross-disciplinary collaborations, implementing bias mitigation strategies, and establishing equitable partnerships. Author Summary In recent years, synthetic data—artificially generated datasets designed to reflect real-world information—has gained attention as a way to advance healthcare research while addressing concerns around data privacy, costs, and accessibility. Our work explores how this field has evolved over the past 25 years, identifying key research trends and shifts in geographic contributions. By analyzing over 14,000 published studies, I found that synthetic data research has grown nearly tenfold, with increasing contributions from Asia alongside traditional leaders in North America and Europe. The focus of research has also changed: earlier work emphasized medical imaging and statistical modeling, while recent studies highlight synthetic data generation and its use in disease modeling, public health, and clinical trials. Despite this progress, important gaps remain. Areas like drug discovery, mental health, and ethical considerations in artificial intelligence need further attention. By mapping these trends, our work underscores the importance of cross-disciplinary collaboration and equitable global partnerships to maximize the benefits of synthetic data in improving healthcare worldwide.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
How this classification was reachedexpand
Full frame machine prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. The Gemma side is a direct model label for every work in the frame, read from the title-only record. The Codex side is a classifier learned from the 10,348 direct Codex labels and calibrated to design-weighted sample rates; fields without enough sample support carry no Codex call. Candidate is the union of the two sides; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels.
Distilled classifier scores by category (both heads)
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.029 | 0.077 |
| Meta-epidemiology (narrow) | 0.001 | 0.000 |
| Meta-epidemiology (broad) | 0.001 | 0.002 |
| Bibliometrics | 0.006 | 0.007 |
| Science and technology studies | 0.001 | 0.001 |
| Scholarly communication | 0.004 | 0.003 |
| Open science | 0.001 | 0.002 |
| Research integrity | 0.001 | 0.002 |
| Insufficient payload (model declined to judge) | 0.002 | 0.000 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedMachine predicted; a candidate call from one source (direct Gemma or distilled Codex), not a consensus.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".