README.md

---
license: cc-by-nc-4.0
task_categories:
  - tabular-classification
  - tabular-regression
language:
  - en
tags:
  - synthetic
  - migraine
  - headache
  - chronic-migraine
  - ichd-3
  - ampp
  - cameo
  - cgrp
  - cgrp-mab
  - erenumab
  - fremanezumab
  - galcanezumab
  - eptinezumab
  - atogepant
  - triptan
  - sumatriptan
  - rimegepant
  - ubrogepant
  - lasmiditan
  - aura
  - midas
  - hit-6
  - neurology
  - clinical-trial
  - tension-headache
  - cluster-headache
  - medication-overuse-headache
pretty_name: "HC-NEU-006 — Migraine & Chronic Headache Dataset (Sample)"
size_categories:
  - 1K<n<10K
---

# HC-NEU-006 — Migraine & Chronic Headache Dataset (Sample)

A schema-identical preview of **HC-NEU-006**, the XpertSystems.ai
synthetic **migraine and chronic headache patient cohort** dataset for
clinical trial research, CGRP-era treatment outcome modeling, ICHD-3
subtype classification ML, AMPP / CaMEO-comparable headache analytics,
and migraine-specific machine learning. The full product covers 10,000
patients; this sample is HF-sized at 3,000 patients.

> **Built by** XpertSystems.ai — Synthetic Data Platform
> **Contact** [pradeep@xpertsystems.ai](mailto:pradeep@xpertsystems.ai) · [xpertsystems.ai](https://xpertsystems.ai)
> **License** CC-BY-NC-4.0 (sample); commercial license available for the full product.

---

## What HC-NEU-006 does — and how it grows the Healthcare/Neurology vertical

HC-NEU-006 is the **sixth Healthcare / Neurology SKU** in the
XpertSystems catalog. After AD, PD, Epilepsy, MS, and Stroke, the
catalog now extends into **chronic episodic neurology** — diseases
managed primarily through pharmacological symptom and prevention
strategies, rather than acute interventional or neuroprotective trials.

| SKU | Disease | US Patients | Annual Pharma R&D | Architecture |
|---|---|---|---|---|
| HC-NEU-001 | Alzheimer's | 6.9M | $8B | Single longitudinal |
| HC-NEU-002 | Parkinson's | 1.0M | $5B | Single longitudinal |
| HC-NEU-003 | Epilepsy | 3.4M | $3B | Cross-sectional |
| HC-NEU-004 | Multiple Sclerosis | 1.0M | $6B | Multi-table relational |
| HC-NEU-005 | Stroke | 7.0M | $3B | Cross-sectional |
| HC-NEU-006 | **Migraine & Headache** | **39M** | **$5B+** | **Cross-sectional** |

**Migraine is the single largest neurology cohort by patient count** —
~39M Americans, with ~9M chronic migraine sufferers. The CGRP era
(Erenumab/Fremanezumab/Galcanezumab/Eptinezumab + Ubrogepant/Rimegepant/
Atogepant) has transformed migraine treatment over the last 6 years,
creating a huge market for clinical research and real-world data.

This is the substrate **migraine pharma R&D teams, CGRP-era market
analytics, headache specialist clinic analytics, and migraine-specific
ML teams** have been waiting for: a coherent cross-sectional dataset
where ICHD-3 subtype × triggers × CGRP biomarkers × acute treatments
× CGRP mAb preventive response × disability outcomes all interact
with **STRIVE / HALO / EVOLVE / PROMISE-2 trial-grade calibration**.

| Buyer Persona | Use Case |
|---|---|
| Migraine Pharma R&D | CGRP mAb comparator modeling, trial design |
| CGRP-Era Market Analytics | Treatment-switching pattern analytics |
| Headache Specialist Clinic | AMPP / CaMEO-comparable benchmarking |
| MIDAS / HIT-6 Modeling | Disability outcome ML training |
| Trigger-Pattern ML | 13-trigger × headache-day prediction |
| Pediatric & Cluster Headache | Subtype-specific cohort enrichment |
| Migraine Digital Therapeutic | Treatment-response wearable ML |
| Real-World Evidence (RWE) | CGRP mAb adherence + discontinuation analytics |
| Migraine Genetics | Aura subtype + family history phenotype ML |

---

## What's inside

**Single cross-sectional dataframe**, one row per patient. 9 clinical
modules concatenated horizontally.

| Output | Rows (sample) | Columns | Size |
|---|---:|---:|---|
| `HC_NEU_006_dataset.csv` | 3,000 | 109 | ~1.5 MB |

Schema provided in `HC_NEU_006_schema.json`.

### Module structure (109 columns total)

| Module | Cols | Coverage |
|---|---:|---|
| Demographics | 8 | sex, age, race/ethnicity, education, region, BMI |
| Comorbidities | 10 | obesity, anxiety, depression, fibromyalgia, sleep, HTN, OCP, psych med, ER, specialist |
| Headache characterization | 15 | subtype, HDM, MDM, duration, pain intensity/location/character, nausea, photophobia, allodynia |
| Prodrome/aura | 9 | flags, duration, symptoms, aura type, visual aura subtype, postdrome |
| Triggers | 17 | 13-class trigger panel + n_triggers + screen time + menstrual cycle |
| Acute treatment | 12 | 11-agent panel, class, dose, time-to-treat, pain-free/relief 2hr, rescue, MOH |
| Preventive treatment | 12 | 11-agent panel, class, dose, MMD reduction, 50/75% responder, adherence, discontinuation |
| Disability/QoL | 14 | MIDAS, HIT-6, SF-12 PCS/MCS, PHQ-9, GAD-7, PSQI, work loss, productivity, cost |
| Biomarkers | 8 | plasma CGRP, ictal flag, serotonin, cortisol, magnesium, inflammatory, autonomic |

---

## Calibration sources

Every distribution is anchored to **named clinical references**. The
headline anchors are **AMPP** (American Migraine Prevalence and Prevention
Study), **CaMEO** (Chronic Migraine Epidemiology and Outcomes Study),
and the four pivotal **CGRP mAb trials** (STRIVE / HALO / EVOLVE /
PROMISE-2). Other anchors:

- **ICHD-3 Diagnostic Criteria (Headache Classification Committee 2018)** —
  6-class headache subtype taxonomy.
- **AMPP Study (Bigal 2008 + Lipton 2007)** — US migraine prevalence,
  triggers, demographics.
- **CaMEO Study (Buse 2013 + Lipton 2014)** — chronic migraine
  epidemiology, longitudinal outcomes.
- **STRIVE Trial (Goadsby 2017 NEJM)** — Erenumab Phase 3, MMD reduction,
  50% responder rate.
- **HALO-EM/CM Trials (Silberstein 2017 NEJM)** — Fremanezumab Phase 3.
- **EVOLVE-1/2 Trials (Stauffer 2018 JAMA)** — Galcanezumab Phase 3.
- **PROMISE-2 Trial (Lipton 2020 Neurology)** — Eptinezumab Phase 3.
- **ACHIEVE-I/II Trials** — Ubrogepant Phase 3 (acute gepant).
- **Bigal 2006 Neurology** — Obesity-migraine bidirectional risk.
- **Edvinsson 2018 Cephalalgia + Goadsby 1990** — Plasma CGRP biomarker
  norms.
- **Cernuda-Morollón 2013** — Chronic migraine CGRP elevation.
- **Mauskop 2012 Headache + Welch 2001** — Magnesium-migraine link.
- **Lipton 2014** — AMPP trigger frequency study.
- **Ferrari 2001 Lancet** — Triptan efficacy meta-analysis.

---

## Validation scorecard

The wrapper ships a 10-metric AMPP/CaMEO/CGRP-trial-anchored scorecard
(`validation_scorecard.json`) that re-scores the dataset on every
generation. Default seed 42 result:

| ID | Metric | Target | Observed | Source |
|---|---|---|---:|---|
| M01 | Chronic Migraine Share | 0.15–0.25 | **0.199** | **ICHD-3 / AMPP** |
| M02 | Episodic Migraine HDM Mean | 3–8 | **4.70** | **CaMEO (Buse 2013)** |
| M03 | Chronic Migraine HDM Mean | 14–22 | **18.39** | **ICHD-3 (≥15)** |
| M04 | Pain Intensity NRS Mean | 5–8 | **6.33** | AMPP + CaMEO |
| M05 | CGRP mAb MMD Reduction | 2.5–4.5 days | **3.71** | **STRIVE / HALO / EVOLVE / PROMISE-2** |
| M06 | Plasma CGRP Mean | 40–110 pg/mL | **54.97** | Edvinsson 2018 + Cernuda-Morollón |
| M07 | Obesity (Chronic Migraine) | 0.20–0.40 | **0.275** | **Bigal 2006 Neurology** |
| M08 | Magnesium Deficiency | 0.30–0.60 | **0.430** | Mauskop 2012 / Welch 2001 |
| M09 | Stress Trigger Reported | 0.60–0.90 | **0.803** | AMPP (Lipton 2014) |
| M10 | Female Patient Share | 0.62–0.82 | **0.690** | AMPP / GBD Migraine 2019 |

**Grade: A+ (100/100). Verified across seeds 42, 7, 123, 2024, 99, 1.**

**Standout calibration**: M01 chronic migraine share lands within
0.13 percentage points of the ICHD-3 / AMPP 20% target. **M05 CGRP mAb
MMD reduction (3.71 days) lands within 0.21 days of the STRIVE / HALO /
EVOLVE / PROMISE-2 pooled mean of 3.5 days** — the exact CGRP-era
clinical efficacy benchmark. M03 chronic migraine HDM (18.39) lands
directly in the ICHD-3 ≥15 range center.

---

## Suggested use cases

- **CGRP-era treatment-response modeling** — patient features +
  preventive class → MMD reduction prediction with STRIVE-calibrated
  CGRP mAb response.
- **ICHD-3 subtype classification** — 6-class headache subtype ML from
  symptom + aura + duration features.
- **Disability outcome forecasting** — MIDAS + HIT-6 prediction from
  baseline features and treatment regimen.
- **Trigger-pattern ML** — 13-trigger feature space × headache_days_per_month
  for trigger-impact modeling.
- **Aura subtype detection** — visual aura sub-typing (scintillating
  scotoma vs fortification spectra vs blurred vision) from prodrome +
  duration features.
- **CGRP biomarker validation** — plasma CGRP × headache_subtype ×
  treatment_class × MMD outcome for biomarker development.
- **Medication overuse headache (MOH) risk modeling** — acute
  medication frequency + class × MOH progression prediction.
- **AMPP / CaMEO comparable cohort analytics** — for healthcare data
  scientists building published-study-comparable models without IRB
  registry access.
- **Health economics / HEOR** — work_days_lost + productivity +
  annual_migraine_cost_usd for migraine cost-effectiveness modeling.
- **Migraine + comorbidity multi-modal** — anxiety / depression /
  fibromyalgia / sleep_disorder co-occurrence ML.

---

## Loading

```python
from datasets import load_dataset

ds = load_dataset(
    "xpertsystems/hcneu006-sample",
    data_files="HC_NEU_006_dataset.csv",
    split="train",
)
```

Or with pandas directly:

```python
import pandas as pd
from huggingface_hub import hf_hub_download

path = hf_hub_download(
    repo_id="xpertsystems/hcneu006-sample",
    filename="HC_NEU_006_dataset.csv",
    repo_type="dataset",
)
df = pd.read_csv(path)
```

The dataset ships with `HC_NEU_006_schema.json` providing per-column
dtypes for pipeline integration:

```python
import json
schema = json.load(open("HC_NEU_006_schema.json"))
# {"patient_id": "object", "headache_subtype": "object", "headache_days_per_month": "int64", ...}
```

Cross-sectional, one row per patient — like HC-NEU-003 (Epilepsy) and
HC-NEU-005 (Stroke). For longitudinal trajectory analysis on individual
patients, use the full product which carries 24-month monthly diary
sub-records.

---

## Schema highlights

**Demographics** — `patient_id`, `headache_subtype` (6-class ICHD-3),
`sex`, `age_at_baseline`, `race_ethnicity`, `education_years`,
`geographic_region`, `bmi`, `obesity_flag`.

**Comorbidities (10 flags)** — anxiety, depression, fibromyalgia,
sleep_disorder, hypertension, OCP_use, psychiatric_medication, prior_er_visit,
headache_specialist.

**Headache characterization** — `headache_days_per_month`,
`migraine_days_per_month`, `headache_duration_hours`, `pain_intensity_nrs`
(0-10), `pain_location`, `pain_character` (Throbbing/Pulsating/Pressing/
Stabbing), `aggravation_by_activity`, `nausea_flag`, `vomiting_flag`,
`photophobia_flag`, `phonophobia_flag`, `osmophobia_flag`,
`allodynia_score_asc12`, `cutaneous_allodynia_flag`,
`chronification_risk_score`.

**Prodrome/aura** — `prodrome_flag`, `prodrome_duration_hours`,
`prodrome_symptoms`, `aura_flag`, `aura_type`, `aura_duration_minutes`,
`visual_aura_subtype` ∈ {Scintillating_Scotoma, Fortification_Spectra,
Blurred_Vision, NaN}, `spreading_depression_proxy`, `postdrome_flag`,
`postdrome_duration_hours`, `postdrome_symptoms`.

**Triggers (13-class)** — stress, sleep_disruption, hormonal,
weather_barometric, skipped_meals, bright_light, strong_odors,
dehydration, alcohol, caffeine_withdrawal, dietary_tyramine,
physical_exertion, screen_time + `n_triggers_reported`,
`trigger_reliability_score`, `menstrual_cycle_day`,
`perimenstrual_attack_flag`.

**Acute treatment** — `acute_treatment_name` (11 agents),
`acute_treatment_class` ∈ {NSAID, Triptan, Gepant, Dittan, Ergotamine,
Analgesic, None}, `acute_dose_mg`, `time_to_treat_hours`,
`pain_free_2hr_flag`, `pain_relief_2hr_flag`,
`most_bothersome_symptom_relief`, `sustained_pain_free_24hr`,
`rescue_medication_flag`, `medication_overuse_days`,
`medication_overuse_headache_flag`, `treatment_satisfaction_score`.

**Preventive treatment** — `preventive_medication` (11 agents),
`preventive_class` ∈ {BetaBlocker, AED, TCA, CGRP_mAb, CGRP_Gepant,
Botox, None}, `preventive_dose_mg`, `botox_units`,
`preventive_duration_months`, `preventive_adherence_pct`,
`monthly_mmd_reduction`, `responder_50pct_flag`, `responder_75pct_flag`,
`cgrp_mechanism_flag`, `preventive_discontinuation_flag`,
`discontinuation_reason`.

**Disability/QoL** — `midas_score`, `midas_grade` ∈ {Grade_I (0-5),
Grade_II (6-10), Grade_III (11-20), Grade_IV (≥21)}, `hit6_score`
(36-78), `promis_pain_interference_t`, `work_days_lost_per_month`,
`presenteeism_days_per_month`, `global_productivity_loss_pct`,
`sf12_pcs`, `sf12_mcs`, `phq9_score` (0-27), `gad7_score` (0-21),
`psqi_score`, `caregiver_burden_score`, `healthcare_visits_per_year`,
`annual_migraine_cost_usd`.

**Biomarkers** — `plasma_cgrp_pg_ml`, `plasma_cgrp_ictal_flag`,
`cgrp_response_index`, `plasma_serotonin_ng_ml`, `cortisol_am_ug_dl`,
`magnesium_serum_mg_dl`, `magnesium_deficiency_flag`,
`inflammatory_index`, `autonomic_dysfunction_score`.

---

## Calibration notes & limitations

In the spirit of honest synthetic data, a few things buyers of the sample
should know:

1. **Generator bug fix applied: missing `obesity_flag` column.** The
   upstream generator's `generate_comorbidities()` does not create
   `obesity_flag`, but `generate_headache_baseline()` references it.
   The wrapper monkey-patches `generate_comorbidities` to add
   `obesity_flag` with Bigal 2006-calibrated prevalence (CM ~30%,
   EM/TTH ~20%). Underlying generator file unmodified. Without this
   patch, the generator crashes with KeyError.

2. **Pain-relief 2hr = 100% and rescue medication = 100% are generator
   quirks.** The upstream `generate_acute_treatment()` module sets
   these flags as constants rather than sampling from the
   `pain_free_2hr` rate dictionary. **Treat the
   `pain_relief_2hr_flag`, `rescue_medication_flag` columns as
   placeholders** until the full product release. The scorecard does
   NOT validate these. For acute treatment efficacy ML, use the
   `acute_treatment_class` field and reference Ferrari 2001 published
   rates externally.

3. **CGRP mAb 50% responder = 100% is a generator quirk.** Same root
   cause as above — `responder_50pct_flag` is set deterministically
   rather than sampled from the trial-anchored response rate. STRIVE /
   HALO / EVOLVE / PROMISE-2 trials report 41-62% 50% responder rates
   for CGRP mAbs vs ~25-30% placebo. The scorecard validates
   `monthly_mmd_reduction` (which IS correctly calibrated, M05) as
   the primary CGRP mAb efficacy metric.

4. **Anxiety = 100% and depression = 0% are generator quirks.** The
   `rng_bool()` helper uses `np.random.random()` (module-level) while
   the broader codebase passes `rng` for seeded reproducibility. This
   creates inconsistent state. **Do not use the
   `anxiety_disorder_flag` or `depression_flag` columns directly for
   comorbidity prevalence work.** For psychiatric comorbidity ML, use
   the `phq9_score` (continuous, 0-27) and `gad7_score` (continuous,
   0-21) columns instead.

5. **Severe HIT-6 (≥60) is 2.8%** vs clinical expectations of 30-50%.
   The HIT-6 distribution in this sample is shifted lower than expected
   for a migraine clinical cohort. For HIT-6 modeling, validate the
   raw distribution before training.

6. **Fibromyalgia flag = 0% is a generator quirk.** Same `rng_bool`
   inconsistency. For migraine-fibromyalgia comorbidity ML, the full
   product fixes this.

7. **Preventive discontinuation = 0% is unrealistic.** Real-world CGRP
   mAb 1-year discontinuation rates are 30-50% (Hepp 2020, Nahas
   2020). Generator does not model discontinuation; the full product
   does.

8. **Plasma CGRP varies bimodally by seed (55 to 96 pg/mL means).**
   The generator's CGRP distribution mixes interictal (~30-50) and
   ictal (~70-110) modes. The scorecard tolerance (35 pg/mL) spans
   both modes; for ictal-only or interictal-only analysis, filter on
   `plasma_cgrp_ictal_flag`.

9. **MIDAS grade label is `Grade_I/II/III/IV`** (not the literature
   convention `I_None/II_Mild/III_Moderate/IV_Severe`). Grade IV =
   MIDAS ≥21 = severe disability per ICHD-3.

10. **Deterministic seeding.** Wrapper passes user-specified seed
    into `CONFIG["seed"]` and `np.random.seed()`. Seed sweep verifies
    Grade A+ across {42, 7, 123, 2024, 99, 1}.

---

## Commercial / full product

The full **HC-NEU-006** product covers 10,000 patients with calibrated
CGRP mAb 50% responder rates per STRIVE/HALO/EVOLVE/PROMISE-2 (not
deterministic 100%), realistic acute-treatment pain-free 2hr response
sampling per Ferrari 2001 meta-analysis, fixed psychiatric comorbidity
sampling, preventive discontinuation modeling per Hepp 2020 / Nahas
2020 real-world data, 24-month monthly diary sub-records for
longitudinal analysis, configurable cohort enrichment (chronic-only,
pediatric, cluster headache, MOH, perimenstrual migraine), and
patient-level outcome modeling. Available under commercial license —
contact [pradeep@xpertsystems.ai](mailto:pradeep@xpertsystems.ai).

XpertSystems.ai also publishes synthetic data products across Oil & Gas
(17 SKUs), Cybersecurity, Insurance & Risk, and Materials & Energy.
Catalog: [huggingface.co/xpertsystems](https://huggingface.co/xpertsystems).