Robust Automated Harmonization of Heterogeneous Data Through Ensemble Machine Learning: Algorithm Development and Validation Study
Bibliographic record
Abstract
Background: Cohort studies contain rich clinical data across large and diverse patient populations and are a common source of observational data for clinical research. Because large scale cohort studies are both time and resource intensive, one alternative is to harmonize data from existing cohorts through multicohort studies. However, given differences in variable encoding, accurate variable harmonization is difficult. Objective: We propose SONAR (Semantic and Distribution-Based Harmonization) as a method for harmonizing variables across cohort studies to facilitate multicohort studies. Methods: SONAR used semantic learning from variable descriptions and distribution learning from study participant data. Our method learned an embedding vector for each variable and used pairwise cosine similarity to score the similarity between variables. This approach was built off 3 National Institutes of Health cohorts, including the Cardiovascular Health Study, the Multi-Ethnic Study of Atherosclerosis, and the Women's Health Initiative. We also used gold standard labels to further refine the embeddings in a supervised manner. Results: The method was evaluated using manually curated gold standard labels from the 3 National Institutes of Health cohorts. We evaluated both the intracohort and intercohort variable harmonization performance. The supervised SONAR method outperformed existing benchmark methods for almost all intracohort and intercohort comparisons using area under the curve and top-k accuracy metrics. Notably, SONAR was able to significantly improve harmonization of concepts that were difficult for existing semantic methods to harmonize. Conclusions: SONAR achieves accurate variable harmonization within and between cohort studies by harnessing the complementary strengths of semantic learning and variable distribution learning.
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 | 0.011 | 0.017 |
| Meta-epidemiology (narrow) | 0.002 | 0.001 |
| Meta-epidemiology (broad) | 0.002 | 0.002 |
| Bibliometrics | 0.002 | 0.002 |
| Science and technology studies | 0.001 | 0.001 |
| Scholarly communication | 0.001 | 0.002 |
| Open science | 0.002 | 0.003 |
| Research integrity | 0.002 | 0.003 |
| Insufficient payload (model declined to judge) | 0.002 | 0.001 |
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".