---
license: cc-by-nc-4.0
language:
- en
tags:
- healthcare
- respiratory
- infectious-disease
- tuberculosis
- mdr-tb
- xdr-tb
- drug-resistance
- hiv-coinfection
- who-tb
- synthetic-data
- ehr
- clinical
pretty_name: "HC-RES-004 Tuberculosis Synthetic Dataset (Sample)"
size_categories:
- n<1K
task_categories:
- tabular-classification
- tabular-regression
- survival-analysis
---

# HC-RES-004 — Tuberculosis Synthetic Dataset (Sample)

**XpertSystems.ai · Synthetic Data Factory · Respiratory Vertical**

The fourth SKU in the XpertSystems **Respiratory** vertical: a globally-calibrated synthetic
tuberculosis cohort spanning the full resistance spectrum — **drug-susceptible (DS-TB),
multidrug-resistant (MDR-TB), pre-extensively (Pre-XDR-TB), extensively drug-resistant (XDR-TB),
and latent (LTBI)**. It covers diagnosis & molecular testing (GeneXpert, LPA, WGS, smear/culture),
an 11-drug resistance profile with MIC proxies, regimen selection (including BPaL/BPaLM and STREAM
short-course), adherence & DOT, sputum-conversion bacteriology, WHO-defined treatment outcomes,
HIV co-infection, household contact investigation, population PK/PD (NAT2, QTc), and health-system
context.

This repository contains a **500-row, single-seed sample**. The full commercial product scales to
20,000+ patients with CSV / Parquet / JSON delivery.

- **SKU:** HC-RES-004
- **Sample size:** 500 patients × 138 columns
- **License (sample):** CC-BY-NC-4.0 — commercial license available for the full product
- **Contact:** pradeep@xpertsystems.ai · https://xpertsystems.ai

---

## Validation

This sample passes XpertSystems Grade **A+** validation (overall **10.000 / 10**) with
deterministic reproduction across all six canonical seeds `[42, 7, 123, 2024, 99, 1]`.

Validation philosophy: **structural identities over distribution-fit tests** — anchored by the
resistance-severity gradient (treatment success falls and mortality rises monotonically from DS-TB
through XDR-TB) and drug-resistance cascade integrity (DS-TB background isoniazid susceptibility
≥94%). This engine also passes its own built-in ~16-check benchmark suite (WHO type distribution,
type-specific success rates, sputum conversion, LTFU, resistance epidemiology, adherence, HIV).

### Calibration anchors

| Metric | Sample value | Target range | Source |
|---|---|---|---|
| DS-TB prevalence | 59.8% | 55–70% | WHO Global TB Report 2023 (~62%) |
| MDR-TB prevalence | 20.6% | 15–25% | WHO (~20%) |
| DS-TB treatment success | 87.6% | 80–92% | WHO (~85%) |
| MDR-TB treatment success | 49.5% | 48–68% | Lancet MDR-TB meta-analysis (~57%) |
| DS-TB 2-month culture conversion | ~74% | 68–88% | WHO/STREAM |
| Loss-to-follow-up rate | 17.8% | 10–24% | Lancet meta-analysis |
| Mean overall adherence | 74.1% | 70–88% | TB program data |
| HIV-TB co-infection | 8.0% | 4–18% | WHO 2023 |
| Hospitalization rate | 40.0% | 25–55% | TB inpatient utilization |
| INH resistance (whole cohort) | 37.2% | 25–45% | WHO 2022 (incl. MDR/XDR) |
| RIF resistance (whole cohort) | 37.6% | 20–40% | WHO 2022 |
| **DS-TB background INH-susceptibility** | **99.7%** | **≥0.94** | Cascade integrity |
| **Success gradient (DS − XDR)** | **+0.59** | **≥0.20** | Resistance-severity ordering |
| **Mortality gradient (XDR − DS)** | **+0.37** | **≥0.08** | Resistance-severity ordering |
| **Column count** | **138** | **≥132** | Schema completeness (11 modules) |

Observed severity gradient across the full spectrum (success / mortality): DS-TB 0.87 / 0.02 →
MDR-TB 0.55 / 0.10 → Pre-XDR-TB 0.38 / 0.21 → XDR-TB 0.37 / 0.25.

---

## Schema highlights by module (138 columns)

**Demographics.** Region (burden-stratified), ethnicity, sex, age, pediatric flag, SES, household
crowding, ventilation quality.

**Classification & diagnosis.** TB type, anatomical site, lineage, prior-TB, smear grade, culture,
GeneXpert MTB/RIF, line-probe assays (first/second line), WGS, diagnostic & health-system delays,
IGRA/TST, chest X-ray pattern, cavitation, diagnostic platform.

**Drug resistance.** 11-drug susceptibility (INH, RIF, EMB, PZA, FLQ, BDQ, LZD, CFZ, DLD, PA, CS),
WHO resistance category, primary vs acquired, mutation-catalog tier, MIC proxies.

**Treatment.** Type-specific regimen (incl. BPaL/BPaLM, STREAM short-course), planned duration, DOT
type, fixed-dose combination, dose adjustment, weight-based drug doses, initiation delay.

**Adherence.** Overall / intensive / continuation-phase adherence, doses missed, interruptions
(duration & reason), LTFU & timing, pill-count ratio, counseling, support program, adverse-event
severity.

**Bacteriology.** 2-/5-month culture & smear conversion, conversion timing, 6-month smear series,
relapse (timing, reinfection-vs-relapse by WGS), acquired resistance during treatment.

**Clinical outcome.** WHO outcome (Cured/Completed/Failed/LTFU/Died), actual duration, BMI &
malnutrition, hospitalization & ICU, grade 3-4 adverse events (hepatotoxicity, neuropathy, QTc,
visual, myelosuppression, hearing loss), failure/death timing & cause, treatment cost, catastrophic
cost flag.

**HIV co-infection.** Status, CD4, ART status/regimen/timing, cotrimoxazole, TB-IRIS, viral
suppression, opportunistic infection.

**Contact investigation.** Household contacts, screening, secondary attack rate, contact LTBI/active
TB, LTBI treatment cascade, transmission cluster (WGS, SNP distance).

**Pharmacokinetics.** NAT2 acetylator status, rifampicin/isoniazid/bedaquiline/linezolid PK, QTcF
baseline & on-treatment, CYP3A4 interaction, exposure quartile, hepatotoxicity score.

**Health system.** Facility level, TB program type, notification channel, distance & transport cost,
food/social support.

---

## Files

- `hc_res_004_sample.csv` — 500-patient sample (138 columns)
- `generate_sample_dataset_hc_res_004.py` — reproducible generator + validation harness
- `validation_report.json` / `validation_report.md` — full scorecard
- `sweep_summary.json` — 6-seed determinism results

## Loading

```python
import pandas as pd
df = pd.read_csv("hc_res_004_sample.csv")
print(df[["patient_id","tb_disease_type","drug_resistance_category",
          "treatment_regimen","treatment_outcome","hiv_status"]].head())
```

```python
from datasets import load_dataset
ds = load_dataset("csv", data_files="hc_res_004_sample.csv")
```

## Use cases

- TB treatment-outcome prediction across the resistance spectrum
- Drug-resistance classification from molecular diagnostics (GeneXpert / LPA / WGS)
- Adherence / LTFU risk modeling and DOT-strategy evaluation
- HIV-TB co-infection management analytics
- Contact-investigation and transmission-cluster modeling
- Health-economics (catastrophic-cost, program-type) analysis
- ML training where real global TB program data is access-restricted

---

## Honest limitations & disclosed generator behavior

This is a strongly calibrated single-table engine (passes its own ~16-check suite; correct
resistance-severity gradient). Standard caveats and minor specifics:

1. **Cross-sectional episode-level table.** Like the pneumonia SKU (HC-RES-003), each row is one TB
   episode; the monthly smear series is encoded as a pipe-delimited string field rather than a
   separate longitudinal table.
2. **Latent linezolid-neuropathy switch is inert.** In the outcome module, the linezolid-related
   neuropathy escalation checks the disease type string rather than the regimen, so peripheral
   neuropathy is generated at its base rate (~12%) regardless of whether linezolid is in the
   regimen. This does not affect any benchmark but means LZD-specific neurotoxicity is not modeled;
   use the PK `linezolid_trough_ug_ml` field if LZD exposure is needed.
3. **Outcome weights are largely type-conditioned.** For Pre-XDR/XDR/LTBI the WHO outcome
   distribution is applied directly by type; DS/MDR additionally shift on adherence and conversion.
   Individual outcomes are therefore well-calibrated at the cohort level but not fully driven by each
   patient's complete feature vector.
4. **Some flags are marginal draws.** Several adverse-event and contact-cascade fields are drawn at
   fixed prevalences rather than jointly conditioned.

General caveat: cross-field correlations beyond those explicitly modeled may be weaker than in real
cohorts. **Not for clinical decision-making** — research/development use only.

---

## Commercial product comparison

| Capability | This sample | Full HC-RES-004 product |
|---|---|---|
| Patients | 500 | 20,000+ (configurable) |
| Seeds / cohorts | 1 | Multi-seed, reproducible |
| Formats | CSV | CSV + Parquet + JSON |
| Smear trajectory | Pipe-delimited string | Optional longitudinal table |
| LZD neurotoxicity | Base-rate only | Regimen-linked model |
| Outcome model | Type-conditioned | Full feature-driven |
| License | CC-BY-NC-4.0 | Commercial |
| Support & SLA | — | Included |

Full product, custom cohorts, or other respiratory SKUs (asthma, COPD, pneumonia, ILD): **pradeep@xpertsystems.ai**

---

## Citation

```bibtex
@dataset{xpertsystems_hc_res_004_2026,
  title        = {HC-RES-004: Tuberculosis Synthetic Dataset},
  author       = {XpertSystems.ai},
  year         = {2026},
  publisher    = {XpertSystems.ai Synthetic Data Factory},
  url          = {https://xpertsystems.ai},
  note         = {Synthetic; CC-BY-NC-4.0 (sample). Calibrated to: WHO Global TB Report 2023
                  (type distribution, treatment success, HIV co-infection); STREAM trial
                  (short-course MDR-TB); TB-PRACTECAL and ACTG A5279 (BPaL/BPaLM, 1HP);
                  Lancet MDR-TB individual-patient-data meta-analysis (success/LTFU);
                  WHO 2022 mutation catalogue and drug-resistance definitions. Fourth SKU in
                  the XpertSystems Respiratory vertical.}
}
```

*Synthetic data generated by XpertSystems.ai. Not derived from real patient records.
Not for clinical use.*