Shingo Fukuma

Hiroshima University

Unclaimed · AI-structured profileOpenAlex

Hiroshima University researcher using transformers on 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—

Output data is insufficient: the catalog holds only the two Evolve-reviewed 2026 arXiv preprints, with no full publication count or venue history. This axis would measure cumulative research output and its growth over time.

Impact—

Impact cannot be assessed: no citation counts, h-index, or downstream adoption data are present in the current snapshot, which rests on two recent 2026 preprints. This axis would measure citation influence and uptake of the work by other researchers.

Focus0/100

The focus score sits at 0 because the raw field distribution splits evenly: 0.5 to health informatics (clinical risk prediction with machine learning) and 0.5 to machine learning for healthcare (EHR representation learning and drug repurposing). In practice the two catalog papers are closely related — both apply transformer-style models to Japanese medical claims data — so the split reads more as two labels for one coherent ML-for-healthcare agenda than genuine dispersion.

Collaboration28/100

Collaboration scores 28 from a raw count of 3 distinct co-authors across the two catalog papers — Yuhei Fujioka, Daitaro Misawa, and Tatsuyoshi Ikenoue. This is a compact team profile: small author lists, consistent partners, and no evidence yet of large consortium work or external network breadth. Fukuma appears as last author on both papers, suggesting a senior or supervisory role within this small group.

Activity—

Activity level is not measurable yet: the snapshot spans only 2026 and lacks yearly publication counts or submission records. This axis would track research cadence — papers per year and continuity of output across recent years.

Reproducibility20/100

Reproducibility scores 20, an average computed across 2 scored papers, both rated low readiness by Evolve. Neither 2026 preprint ships with confidence intervals, and the cross-attention MACE study rests on a single private Japanese claims cohort while the drug-repositioning screen uses richer inputs than its baselines. Releasing code, cohort documentation, and uncertainty estimates would lift this axis most directly.

Research summary

Shingo Fukuma is a researcher at Hiroshima University whose reviewed work sits squarely in machine learning for healthcare, with two 2026 arXiv preprints in the Evolve catalog: one on clinical risk prediction from medical claims data, the other on EHR representation learning and drug repositioning. On both papers Fukuma appears as last author without corresponding-author designation, working with a compact team: Yuhei Fujioka, Daitaro Misawa, and Tatsuyoshi Ikenoue are the recurring co-authors, and only 3 distinct co-authors appear across the cataloged output. The cardiovascular paper proposes cross-attention over diagnosis–treatment links, raising MACE ROC-AUC to 0.7720 versus 0.7586 for self-attention, though Evolve notes the evidence rests on a single private Japanese cohort. The second introduces HSA+PM+CR medical-code pretraining that outperforms BEHRT and MED-BERT and feeds an Alzheimer's disease drug-repositioning screen, with comparisons that favor richer inputs and no confidence intervals. The overall picture is a focused, small-team research agenda applying transformer architectures to Japanese claims data: methodologically promising, but with low reproducibility readiness (scored 20 across 2 papers) and a collaboration footprint of just 3 co-authors. Because 'Shingo Fukuma' is a fairly common Japanese name and no ORCID is linked, this profile is name-resolved from public bibliographic data and should be read as early-stage.

Collaboration pattern

Collaboration is compact: only 3 distinct co-authors (Yuhei Fujioka, Daitaro Misawa, Tatsuyoshi Ikenoue) appear across the two cataloged papers, both with Fukuma as last author. This suggests a stable, small in-house team — likely a supervisory lab setup — rather than wide external networks; the collaboration dimension scores 28. Repeated pairing with the same three co-authors indicates an ongoing group, and last-author placement suggests Fukuma leads or supervises the work.

Research keywords

machine learning for healthcareelectronic health recordsmedical claims datarepresentation learningBERT pre-trainingdrug repositioningAlzheimer's diseasecross-attentiontransformerMACE predictioncardiovascular riskICD-10

Highlights

Cross-attention lifts MACE prediction on Japanese medical claims data
As last author, Fukuma's team applies cross-attention over diagnosis–treatment links in medical claims data, reaching MACE ROC-AUC 0.7720 versus 0.7586 for self-attention; Evolve notes the evidence rests on one private Japanese cohort.
Medical-code pretraining screens Alzheimer's drug repositioning candidates
The team's HSA+PM+CR pretraining of medical codes outperforms BEHRT and MED-BERT and enables a practical screen of Alzheimer's disease repositioning candidates; Evolve flags comparisons that favor richer inputs and no confidence intervals.

Timeline

2026Posted a last-author arXiv preprint applying cross-attention to medical claims data for MACE prediction, with ROC-AUC 0.7720 versus 0.7586 for self-attention.
2026Posted a second last-author arXiv preprint introducing HSA+PM+CR medical-code pretraining that outperforms BEHRT and MED-BERT and drives an Alzheimer's disease drug-repositioning screen.
2026Both cataloged papers appeared as 2026 arXiv preprints with low reproducibility readiness, anchoring an early-stage profile centered on Japanese medical claims data.

Papers in the Evolve catalog

Frequent co-authors

Yuhei Fujioka×2 Daitaro Misawa×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.

Bibliometrics and radar scores come from a limited OpenAlex-based snapshot; the output, impact, and activity dimensions lack sufficient data and are not scored.
Only two 2026 arXiv preprints are in the reviewed catalog; no journal publications, citation counts, or venue history are included.
No ORCID or country record is linked, and 'Shingo Fukuma' is a fairly common Japanese name, so this profile is name-resolved from public bibliographic data.
Both papers are rated low reproducibility readiness (reproMean 20 across 2 papers), resting on private Japanese cohorts without confidence intervals or documented released code.