Kyoto University researcher applying transformer and pretraining methods to medical claims and EHR risk prediction
2
papers here
2
reviewed by Evolve
2
first-author
0
corresponding
—
works · OpenAlex
—
h-index · OpenAlex
Bibliometric figures come from OpenAlex (2026-10-16) and cover the author's entire public record, not just the papers listed here. Reproducibility signal comes from Evolve's own reviews.
Research profile
Output—
Insufficient data to score: beyond the two cataloged 2026 preprints, no broader publication record is linked. This axis would measure total output, venue mix, and yearly publishing volume once fuller bibliographic records are available.
Impact—
Not scored yet: the snapshot contains no citation counts or venue-level influence signals for these preprints. This axis would measure citations, adoption by downstream work, and field-level influence as citation records accumulate.
Focus0/100
The cataloged portfolio splits evenly between two adjacent subfields — 0.5 health informatics (clinical risk prediction with machine learning) and 0.5 machine learning for healthcare (EHR representation learning and drug repurposing). Both lines share medical claims/EHR sequence data and transformer methods, so the radar's focus score of 0 reflects a narrowly concentrated agenda rather than scattered interests.
Collaboration28/100
The visible co-author network is small: 3 distinct co-authors, with Daitaro Misawa, Shingo Fukuma, and Tatsuyoshi Ikenoue recurring across the reviewed work. Both papers are first-authored and non-corresponding, indicating a lead-contributor position within a compact team. The radar's collaboration score of 28 is consistent with this early-stage, tight-knit pattern; it is computed from only two cataloged papers.
Activity—
Insufficient data to score: the snapshot lacks multi-year publication history, so cadence cannot be established. This axis would measure publishing rhythm and recent engagement, which cannot be inferred from two same-year first-authored preprints.
Reproducibility20/100
The two reviewed papers average a reproducibility score of 20/100 (reproPapers: 2), both rated low readiness. Reviews cite missing confidence intervals in the pretraining comparisons, which favor richer inputs, and MACE results (ROC-AUC 0.7720 vs 0.7586) resting on a single private Japanese cohort; no code or data release is evidenced in the catalog. Reproducibility is the clearest near-term improvement lever.
Research summary
Yuhei Fujioka is a Kyoto University researcher working at the intersection of machine learning and clinical data, publishing first-authored work on medical claims and electronic health record (EHR) modeling. The Evolve catalog holds two 2026 arXiv preprints. The first applies cross-attention over diagnosis–treatment links to predict major adverse cardiovascular events (MACE) from medical claims data, reporting ROC-AUC 0.7720 versus 0.7586 for a self-attention baseline. The second builds a medical-code pretraining scheme (HSA+PM+CR) that outperforms BEHRT and MED-BERT and is applied to screen drug-repositioning candidates for Alzheimer's disease. Field distribution across the catalog is even — 0.5 health informatics (clinical risk prediction with machine learning) and 0.5 machine learning for healthcare (EHR representation learning and drug repurposing) — indicating one coherent program on sequential clinical data rather than separate agendas. The visible co-author network is small (3 distinct co-authors: Daitaro Misawa, Shingo Fukuma, Tatsuyoshi Ikenoue), and both papers carry low reproducibility readiness (mean score 20/100 over 2 papers), with reviews noting absent confidence intervals and reliance on a single private Japanese cohort.
Collaboration pattern
The visible network is compact: 3 distinct co-authors, with Daitaro Misawa, Shingo Fukuma, and Tatsuyoshi Ikenoue the recurring collaborators in the catalog. Both 2026 papers are first-authored but not corresponding, consistent with a lead-contributor role inside a small team rather than independent group leadership. The radar's collaboration score of 28 mirrors this early, tight-knit pattern; the underlying data covers only the two reviewed papers.
Research keywords
Highlights
Cross-attention lifts MACE prediction from medical claims data
Medical-code pretraining repurposed for Alzheimer's drug screening
Timeline
- 2026First cataloged first-authored arXiv preprint: cross-attention over diagnosis–treatment links for MACE prediction, reporting ROC-AUC 0.7720 versus 0.7586 for self-attention.
- 2026Second first-authored preprint: HSA+PM+CR medical-code pretraining outperforming BEHRT/MED-BERT, applied to Alzheimer's drug-repositioning screening.
- 2026Both cataloged works land in the same year, all first-authored, anchoring an early-stage profile centered on medical claims and EHR sequence modeling.
Papers in the Evolve catalog
Frequent co-authors
About this profile
This profile is AI-structured by Evolve.Pub from public bibliographic data (OpenAlex) and Evolve's own paper reviews. It is not written or endorsed by the researcher unless the profile is claimed. Radar dimensions are computed from real bibliometrics with fixed formulas — raw figures are shown alongside; missing dimensions mean insufficient data, never zero. Errors? Claim the profile to correct or hide it.
- Bibliometric scores derive from a limited snapshot; output, impact, and activity axes lack sufficient data and are left unscored.
- The reviewed catalog contains only two 2026 arXiv preprints; no peer-reviewed venue or citation records are available yet.
- ORCID is not linked and the name is resolved from public bibliographic data, so identity disambiguation remains provisional.
- Reproducibility (mean 20/100 over 2 papers) and collaboration (3 distinct co-authors) rest on a very small sample and may shift quickly.