Daitaro Misawa

Kyoto University

Unclaimed · AI-structured profileOpenAlex

Kyoto University researcher applying deep learning to medical claims data for risk prediction and drug repositioning

papers here

reviewed by Evolve

first-author

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—

Not yet scorable: the underlying data covers only two 2026 arXiv preprints in the Evolve catalog, with no indexed publication count or venue history. This axis would measure cumulative scholarly output and its distribution across venues and years.

Impact—

Not yet scorable: no citation counts, h-index, or citation stream accompany the supplied record, and the two 2026 preprints are too recent for citations to have accrued. This axis would measure citation-based influence once indexed citation data becomes available.

Focus0/100

Focus scores 0/100 because the raw field distribution splits evenly at 0.5/0.5 between 'health informatics (clinical risk prediction with machine learning)' and 'machine learning for healthcare (EHR representation learning and drug repurposing)'. The two catalog papers sit one on each side: claims-based cardiovascular risk prediction versus pretrained medical-code representations for drug repositioning. Rather than scattered work, this reads as a deliberate two-track agenda on the same claims-data substrate.

Collaboration28/100

Collaboration scores 28/100 on a raw base of 3 distinct coauthors across the reviewed record, a notably small network. All recurring collaborators — Yuhei Fujioka, Shingo Fukuma, and Tatsuyoshi Ikenoue — appear on both 2026 preprints, indicating one stable, likely Kyoto-based core team rather than an extended coauthor graph. On neither paper is Misawa the corresponding author, and both list a middle author position, consistent with contributing inside a supervised group.

Activity—

Not yet scorable: the record supplies no year-by-year publication series beyond the two 2026 preprints, so publishing cadence cannot be judged. This axis would measure recent-year productivity and the regularity of research output over time.

Reproducibility20/100

Reproducibility scores 20/100, computed as a mean of 20 over the 2 papers Evolve has assessed, each rated low repro-readiness. The drivers are visible in the verdicts: the drug-repositioning study reports no confidence intervals and baseline comparisons that favor richer inputs, while the MACE model rests on a single private Japanese claims cohort. Neither paper's public code or data availability is established in the reviewed record.

Research summary

Daitaro Misawa is a Kyoto University-affiliated researcher working at the intersection of health informatics and machine learning for healthcare, focused on deep learning over Japanese medical claims and electronic health record data. The Evolve catalog holds two 2026 arXiv preprints on which Misawa appears as a middle author, neither as corresponding author. The first introduces an HSA+PM+CR pretraining scheme for medical codes that reportedly outperforms BEHRT and MED-BERT baselines and is applied to screen Alzheimer's disease drug-repositioning candidates. The second applies cross-attention over diagnosis–treatment links to predict major adverse cardiovascular events, reporting ROC-AUC 0.7720 against 0.7586 for self-attention. Evolve's verdicts are cautiously positive: the pretraining study reports no confidence intervals, while the MACE evidence rests on one private Japanese cohort; both are rated low repro-readiness. Radar scores reflect a small collaborative footprint (3 distinct coauthors; collaboration 28/100), an even 0.5/0.5 split between the two declared fields (focus 0/100), and low reproducibility (20/100 across 2 papers). Output, impact, and activity axes remain unscored for lack of indexed bibliometric data. Frequent coauthors are Yuhei Fujioka, Shingo Fukuma, and Tatsuyoshi Ikenoue.

Collaboration pattern

The collaboration footprint is small and stable: 3 distinct coauthors across the reviewed record, with Yuhei Fujioka, Shingo Fukuma, and Tatsuyoshi Ikenoue recurring on both 2026 preprints. Misawa sits in a middle author position on each paper and is not the corresponding author on either, a pattern consistent with contributing within a single close-knit, likely Kyoto University-based team rather than leading independent projects or an extended coauthor network.

Research keywords

electronic health recordsmedical claims datarepresentation learningBERT pre-trainingcross-attentiontransformerclinical risk predictionMACE predictioncardiovascular riskdrug repositioningAlzheimer's diseaseICD-10

Highlights

Medical-code pretraining pushed toward practical drug-repositioning screening
An HSA+PM+CR pretraining scheme for medical codes reportedly beats BEHRT and MED-BERT baselines and feeds an Alzheimer's disease drug-repositioning candidate screen. Evolve calls the direction promising but notes the comparisons favor richer inputs and no confidence intervals are reported.
Cross-attention lifts MACE prediction on Japanese claims data
Modeling diagnosis–treatment links with cross attention raises ROC-AUC for major adverse cardiovascular events to 0.7720 versus 0.7586 for self-attention. Evolve flags that the evidence rests on one private Japanese cohort.

Timeline

2026Releases an arXiv preprint on HSA+PM+CR medical-code pretraining, benchmarked against BEHRT and MED-BERT and applied to Alzheimer's disease drug-repositioning screening.
2026Releases an arXiv preprint applying cross attention over diagnosis–treatment links to MACE prediction, reaching ROC-AUC 0.7720 versus 0.7586 for self-attention.
2026Both catalog papers appear as middle-author, non-corresponding arXiv preprints rated low repro-readiness, produced within a recurring three-person coauthor team.

Papers in the Evolve catalog

Frequent co-authors

Yuhei Fujioka×2 Shingo Fukuma×2 Tatsuyoshi Ikenoue×1

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 OpenAlex-indexed records only; no citation, h-index, or year-by-year output data were available, so output, impact, and activity are left unscored.
This profile is name-resolved from public bibliographic data; no ORCID or verified identity link is attached, so attribution of the two papers carries some uncertainty.
Both catalog papers are 2026 arXiv preprints; peer-reviewed publication status is not established in the available data.
The reproducibility score rests on only 2 reviewed papers, both rated low repro-readiness; the rating may not generalize to unreviewed work.