Next-Generation Clinical Phenotype Decision Support

Rare diseases
hide in plain sight.

PhenoCDS guides any frontline doctor to see the unseen — turning missed clinical clues into early rare disease diagnoses before irreversible harm occurs. Localized NLP, adaptive disease prevalence mapping, and FHIR-native output, all within a HIPAA-compliant web platform.

Contact Us Learn More
~18K
HPO phenotype nodes covered
HIPAA
Compliant · Zero PHI transmission
R4
HL7 FHIR standard output
0
Patient records leaving your system
Clinical Gap

Three unsolved gaps
in rare disease diagnosis

The rare disease crisis is not a lack of medical knowledge — it's a structural failure to translate clinical observation into standardized, actionable data at the point of care.

01

The Diagnostic Odyssey

Subtle phenotypic features are invisible under high-volume clinical pressure. Non-genetics physicians lack the tools to connect scattered clues — leaving patients in a 5–7 year diagnostic limbo before irreversible harm sets in.

RECOGNITION GAP
02

The Localization Gap

Clinical notes worldwide are written in local languages, mixed terminologies, and informal shorthand (e.g. eye space wide, DMS delay). Existing tools, trained on English-only corpora, fail entirely in non-English clinical environments.

LOCALIZATION GAP
03

The Resource Disconnect

Current tools output cold diagnostic codes and stop there. Patients receive a diagnosis with no bridge to local patient organizations, orphan drug eligibility, social welfare resources, or research enrollment opportunities.

RESOURCE GAP
System Architecture

Four modular layers
from input to FHIR output

Designed as a secure, web-based platform with HIPAA-compliant data handling. All patient phenotype data is processed without transmission of personally identifiable information — zero PHI leaves your environment.

L1 · Knowledge
Multi-dimensional Knowledge BaseKnowledge Graph Layer
HPO hp.json → SQLite OMIM Gene-Phenotype Matrix Orphanet Onset Timeline Local Rare Disease Resource DB Localized HPO Mapping Dictionary
L2 · NLP Engine
Hybrid Inference EngineAdaptive NLP Inference
Rule-based Term Normalization Embedding + Vector Retrieval Onset-Weighted Similarity Algorithm Next-Best-Question Prompting Engine
L3 · Interface
Point-of-Care UIReal-time Clinical Interface
IntelliSense Real-time Suggestions Bilingual Tooltip Design Primary Care Lite / Medical Center Pro One-click Standardized Note Generation
L4 · Output
Standardized Output LayerFHIR Standardization
HL7 FHIR R4 FHIR Observation Bundle HPO → Condition / Observation Seamless HIS Integration
HIPAA-compliant architecture · Zero PHI transmission · All processing within your secure environment
Web-based platform · No local installation required · Accessible from any clinical workstation
Active learning loop · Historical records activate · Dictionary continuously evolves
Adaptive Intelligence

Three self-reinforcing
competitive moats

PhenoCDS is not a static tool — it is a learning platform. Every clinical interaction strengthens three interlocking flywheels that competitors cannot replicate without starting from zero.

01

Adaptive Local Language Mapping

The system continuously tracks which HPO mappings physicians actually select for each local clinical expression. High-frequency selections are automatically promoted in the ranking — making the mapping dictionary increasingly accurate for regional clinical language patterns over time. What takes a competitor years to build, we accumulate through daily clinical use.

LANGUAGE FLYWHEEL
02

Regional Disease Prevalence Calibration

Disease recommendation rankings are dynamically weighted by local population genetics and real-world diagnosis outcomes. As confirmed diagnoses accumulate, the engine automatically recalibrates prior probabilities to reflect true regional prevalence — not just published literature. A rare disease rare in Europe may be far more common in East Asian populations, and our engine learns this.

PREVALENCE FLYWHEEL
03

Regional Phenotype Blind Spot Discovery

By analyzing which phenotypes appear in confirmed diagnoses but were absent from initial clinical notes, PhenoCDS identifies the specific clues that physicians in each region systematically miss. These blind spots are then surfaced as prioritized next-best-question prompts — turning population-level insight into individual clinical guidance.

INSIGHT FLYWHEEL
All three flywheels reinforce each other — better mapping produces cleaner HPO inputs, which improves prevalence calibration, which surfaces more precise blind spots, which improves mapping priority
Competitive Analysis

Why existing tools
fall short

International NGP tools have fundamental gaps across clinical workflow integration, local language support, and privacy architecture — gaps that become critical in non-English healthcare environments.

Dimension Traditional Query Tools
Phenomizer, GDDP
International NLP Models
ClinPhen, Doc2Hpo
Face2Gene (FDNA)
130 countries · 2,000 clinics
DeepRare
Nature 2026 · Academic Engine
PhenoCDS
Clinical workflow integration Manual code entry, post-consultation lookup Retrospective upload, time-lagged Facial photo upload required — consent burden & workflow disruption at bedside Academic tool only, no EHR integration Real-time IntelliSense at point of care
Non-English language support English HPO terms only English-only training corpus English interface; no local clinical language support Primarily English, limited localization Multi-layer bilingual parsing (Rule + Embedding) with adaptive local mapping
Non-facial rare diseases (e.g. XLH) Partial — relies on HPO input only Partial support Completely inapplicable — ~60% of rare diseases have no facial features; Caucasian bias documented in Asian patients Full support via HPO symptom reasoning Full support — symptom-driven, no photo required
Guides doctors to find missing clues No No No — photo-based; cannot prompt for unseen symptoms No — requires complete HPO input upfront Next-best-question engine actively prompts for missing phenotypes
Privacy & HIPAA compliance Public web, uncontrolled data flow Requires upload to foreign cloud servers Facial photo upload raises significant privacy & IRB concerns Cloud-dependent, IRB clearance challenging HIPAA-compliant · Zero PHI transmission · Secure web processing
Regional prevalence calibration Global static database only No regional adjustment No — Caucasian-dominant training; documented accuracy drop in Asian patients Literature-based priors, not locally calibrated Dynamic prevalence weighting — continuously calibrated by real-world diagnosis outcomes
Onset timeline weighting Flat phenotype set, no temporal dimension Not supported Not supported Partial support Congenital / Infantile / Adult dynamic weighting via Orphanet onset data
Post-diagnosis resource linkage None None None None Integrated rare disease foundation, patient support & orphan drug eligibility
FHIR output HPO list only Partial (SNOMED CT primary) No FHIR output Partial FHIR support Native FHIR R4 Observation Bundle output
Academic Positioning

How PhenoCDS advances
Next-Generation Phenotyping

NGP defines the direction. PhenoCDS delivers three critical evolutions for real-world clinical deployment.

Limitations of Traditional NGP Tools
  • Passive extraction: analysis happens after the note is written, creating diagnostic lag and missing real-time intervention windows
  • Language barrier: 100% trained on English corpora — fails completely when encountering mixed-language clinical terminology
  • No temporal dimension: phenotypes treated as a flat set; onset timing excluded from similarity calculations
  • FHIR gap: outputs HPO lists only, not packaged into interoperable health data exchange formats
  • Static intelligence: no learning from real-world usage; regional prevalence and language patterns never improve
PhenoCDS Differentiating Advances
  • Active Copilot: real-time inference at point of care — system suggests associated phenotype checklists from the first input
  • Adaptive Local NLP: dual-layer Rule + Embedding parsing builds a gold-standard local HPO mapping dictionary that improves with use
  • Onset Dynamic Weighting: integrates Orphanet onset timeline data to improve differential diagnosis precision
  • FHIR-native: outputs FHIR R4 Observation Bundle, directly compatible with national health data exchange platforms
  • Living intelligence: regional prevalence and language mapping continuously calibrate through real-world diagnosis feedback
Market Opportunity

Two sides of the same
billion-dollar gap

PhenoCDS addresses two massive, opposing market failures simultaneously — unlocking dormant drug revenue for pharma while eliminating wasteful healthcare spending for payers.

Idle Market · Unlocked Revenue

Orphan drugs sitting on shelves — because patients are never diagnosed

Pharmaceutical companies invest billions developing rare disease treatments. Yet 40–60% of eligible patients remain undiagnosed — meaning drugs with approved indications generate a fraction of their potential revenue. Every year of diagnostic delay is a year of lost treatment.

$258B
Global orphan drug market 2026 · CAGR ~12%
40–60%
Eligible patients currently undiagnosed — idle drug revenue
Taiwan case study — 3 diseases alone:
Disease Cost/patient/yr Idle patients Idle market
XLH NT$5.45M ~1,000 NT$5.45B
Porphyria NT$13.2M ~2,200 NT$29.0B
OI NT$18.6K ~700 NT$13.0M
Combined idle market ~NT$34.5B
Calculation: international literature prevalence ratios × estimated undiagnosed patients × annual drug cost
PhenoCDS
BRIDGES
Wasted Spending · Recovered Value

Healthcare systems burning money on misdiagnosis — for years before the right answer

Every year a rare disease patient spends undiagnosed, they cycle through unnecessary consultations, redundant tests, and ineffective treatments. This is not a clinical failure — it is a systemic failure to connect observable symptoms to the right diagnosis at the first encounter.

4.7 yrs
Average diagnostic odyssey duration — each year burning healthcare budget
NT$1.5–3M
Estimated excess medical cost per patient before correct diagnosis
The compounding cost of delay
Year 1–2 Repeated GP visits, blood panels, imaging — each inconclusive
Year 3–4 Multi-specialist referrals, misdiagnosis treatment costs, medication waste
Year 4.7+ Correct diagnosis — but irreversible complications may now require long-term care
Taiwan-scale impact:
~200,000 undiagnosed patients × NT$1.5–3M excess cost
= NT$300B–600B in preventable healthcare waste
Conservative estimate based on international diagnostic odyssey cost studies. Excludes long-term care costs from irreversible complications.
AI Diagnostic Market
$20B → $194B
Global AI rare disease diagnosis market 2025→2035 · CAGR 29% · Symptom-to-disease matching tools hold 30% market share
Real-World Data Asset
Unique RWD
Prospective phenotype-to-diagnosis pathway data — the rarest dataset in medicine. Licensable to pharma for drug development, RWE studies, and new indication discovery
Clinical Research Network
Physician Atlas
Anonymized map of which physicians hold the highest concentrations of specific rare disease phenotypes — enabling precision clinical trial recruitment for pharma and research centers
Expected Impact

Measurable outcomes
across three dimensions

Clinical: Intelligent Safety Net

Real-time phenotype checklists prevent diagnostic misses at the moment of care. Rare genetics expertise is democratized to every frontline physician — regardless of specialty or institution.

>50% Expected reduction in
phenotype documentation time

Research: Next-Generation Data Asset

Cleaning clinical input at the source produces the first high-quality structured genetic phenotype database aligned to HPO/FHIR standards — a foundation for precision medicine and real-world evidence studies.

10 yrs Historical records
retrospectively activated

Patient: Human-centered Continuity

Shortened diagnostic odyssey. At first suspicion, the system immediately connects families to local patient organizations, orphan drug eligibility, social welfare resources, and clinical trial enrollment opportunities.

Day 0 Resources delivered
at first diagnosis
The Team

Built by the people
who understand the problem

A rare convergence of clinical genetics expertise, AI engineering, and healthcare entrepreneurship — united by a shared mission to end the diagnostic odyssey.

AS
Alexis Syu
CEO · Chief Medical Officer
Attending physician in pediatric genetics at Taipei Medical University Hospital. Deputy Director, Min-Sheng Genomics Center. M.S. in Big Data, Taipei Medical University.
Clinical Genetics Python · AI FITI · Startup Weekend
SC
Steven Chen
CTO
Dual degree in Medicine and Computer Science, National Taiwan University. Datathon Winner. Bridges clinical reasoning and LLM engineering — the rarest combination in medical AI.
NTU Medicine · CS LLM Engineering Datathon Winner
KC
Ken Chang
VP Engineering · Institutional Partnerships
External Executive Advisor, Jin Yu Feng Capital. Former lead on cross-institutional digital health systems for Taiwan's Ministry of Digital Affairs. Generative AI Hackathon Silver Award.
CS · Systems Gov Projects Enterprise Integration
PC
Pierre Chen
Lead Developer
NTU Information Management × Kaohsiung Medical University Medical Chemistry. Founder of SuperB Software. Developer of Dogtor — a clinical scheduling platform used by hospitals across Taiwan.
Full-stack Dev Medical Chemistry Healthcare SaaS
KC
Kathleen Chen
Head of Business Development
NTU International Business × NCKU Life Sciences. iGEM Gold Medal. Former JC Capital. Translates life science innovation into capital market narratives and pharma partnership strategy.
Life Sciences iGEM Gold Medal Finance · BD
AY
Austin Yu
COO · Product
MD, University of Pennsylvania. UIUC CS × UCSF Research. Ph.D. candidate in Biomedical Engineering, NTU. Product designer at Shuigupu. Serial healthcare startup founder — understands both sides of the clinical-commercial divide.
UPenn MD UIUC CS · UCSF Healthcare Founder
Clinical depth no competitor can replicate: Our founder sees these patients every week. "Hairline hidden under a boy's buzz cut" and "bow legs that don't respond to Vitamin D" are not research hypotheses — they are her clinical reality.
Three languages in one team: We speak clinical medicine, AI engineering, and business — the three conversations that must happen simultaneously to build a healthcare platform that actually gets used.
Get In Touch

Let's build the future
of rare disease diagnosis.

Interested in clinical collaboration, research partnership, or investment? We'd love to hear from you.

contact@phenocds.com Back to Top