---
license: cc-by-4.0
task_categories:
- tabular-classification
- tabular-regression
language:
- en
tags:
- synthetic
- healthcare
- tuberculosis
- tb
- genexpert
- smear-microscopy
- chest-xray
- hiv-tb
- mdr-tb
- who-guidelines
- lmic
pretty_name: Synthetic TB Screening & Symptom Dataset (GeneXpert, Smear, CXR)
size_categories:
- 10K<n<100K
configs:
- config_name: low_tb_burden
  data_files: data/tb_low_tb_burden.csv
- config_name: moderate_tb_burden
  data_files: data/tb_moderate_tb_burden.csv
  default: true
- config_name: high_tb_burden
  data_files: data/tb_high_tb_burden.csv
data_type: synthetic
---

> ⚠️ **Synthetic dataset** — Parameterized from published SSA literature, not real observations. Not suitable for empirical analysis or policy inference.

# Synthetic TB Screening & Symptom Dataset (GeneXpert, Smear, CXR)

## Abstract

This dataset provides **30,000 synthetic records** (10,000 per scenario) of patients undergoing TB screening at LMIC health facilities. Each record contains 25 variables including demographics, HIV status, BMI, 8 symptom indicators, WHO symptom screen score, diagnostic results (smear microscopy, GeneXpert MTB/RIF with rifampicin resistance, chest X-ray), TB classification, and treatment outcome. Diagnostic test performance (sensitivity, specificity) is modelled from Cochrane reviews and WHO product evaluations. Three burden scenarios (low, moderate, high) span TB prevalence from 6% to 35% among screened patients, with HIV co-infection rates from 10% to 46%.

**This dataset is entirely synthetic. It must not be used for clinical decision-making.**

## 2. Methodology

### 2.1 Diagnostic Test Performance

| Test | Sensitivity | Specificity | Source |
| --- | --- | --- | --- |
| Smear microscopy | 60% (40% if HIV+) | 98% | Getahun et al., Lancet ID 2007 |
| GeneXpert (smear+) | 98% | 98% | Steingart et al., Cochrane 2014 |
| GeneXpert (smear-) | 68% (79% HIV+) | 98% | Steingart et al., 2014 |
| GeneXpert RIF resistance | 95% sensitivity | 98% specificity | WHO 2021 |
| Chest X-ray | 87% | 70% | van't Hoog et al., PLoS One 2012 |

### 2.2 Scenario Design

| Scenario | TB Prevalence | HIV+ | TB-HIV Co-infection | MDR-TB | Smear+ |
| --- | --- | --- | --- | --- | --- |
| Low burden | 6.4% | 4.1% | 10% of TB cases | 3.1% | 5.4% |
| Moderate burden | 17.2% | 10.8% | 26% of TB cases | 4.8% | 10.4% |
| High burden | 35.1% | 27.5% | 46% of TB cases | 8.1% | 16.8% |

## 3. Schema

| Column | Type | Description |
| --- | --- | --- |
| age_years | int | Age (15-85) |
| sex | categorical | M/F (M:F ratio ~1.4:1 per WHO) |
| bmi | float | Body mass index |
| hiv_status | binary | HIV serostatus |
| cough_2weeks, fever, night_sweats, weight_loss | binary | WHO symptom screen |
| hemoptysis, chest_pain, fatigue, loss_of_appetite | binary | Additional symptoms |
| cough_duration_weeks | int | Duration of cough |
| who_symptom_screen_count | int (0-4) | Number of WHO screening symptoms |
| smear_result | binary | Sputum smear microscopy |
| xpert_mtb_detected | binary | GeneXpert MTB detected |
| xpert_rif_resistance | categorical | detected / not_detected / N/A |
| cxr_result | categorical | normal / infiltrate / cavity / miliary / pleural_effusion / old_tb_scar / other_pathology |
| cxr_abnormal | binary | Any CXR abnormality |
| true_tb_status | binary | Ground truth TB status |
| tb_type | categorical | none / pulmonary / extrapulmonary |
| tb_classification | categorical | not_tb / smear_positive_ptb / smear_negative_ptb / extrapulmonary_tb |
| rifampicin_resistant | binary | Drug resistance status |
| treatment_outcome | categorical | cured / completed / failed / died / lost_to_followup / not_applicable |

## 4. Validation

<p align="center">
  <img src="validation_report.png" alt="Validation Report" width="100%">
</p>

## 5. Usage

```python
from datasets import load_dataset
dataset = load_dataset("electricsheepafrica/synthetic-tb-screening-symptoms-genexpert-WHO", "moderate_tb_burden")
df = dataset["train"].to_pandas()
```

## 6. Limitations

- **Synthetic**: Not for clinical use or diagnostic algorithm validation.
- **No paediatric TB**: Only adults ≥15 years; paediatric TB has different presentation.
- **Simplified CXR**: CXR findings are categorical; no image data.
- **No treatment timeline**: Outcomes are assigned, not modelled over time.
- **Single sputum sample**: Real diagnosis often requires multiple specimens.

## 7. References

1. WHO (2023). Global Tuberculosis Report 2023. Geneva.
2. WHO (2021). Consolidated guidelines on TB: Rapid diagnostics. Geneva.
3. Corbett EL, et al. (2003). TB and the HIV epidemic. *Arch Intern Med*, 163:1009.
4. Steingart KR, et al. (2014). Xpert MTB/RIF for pulmonary TB. *Cochrane*, CD009593.
5. Getahun H, et al. (2007). Smear-negative pulmonary TB in HIV. *Lancet ID*, 7(4):238-246.
6. WHO (2013). Systematic screening for active TB. Geneva.
7. van't Hoog AH, et al. (2012). Screening strategies for TB. *PLoS One*, 7(5):e36392.

## Citation

```bibtex
@dataset{esa_tb_2025,
  title={Synthetic TB Screening and Symptom Dataset},
  author={Electric Sheep Africa},
  year={2025},
  publisher={Hugging Face},
  url={https://huggingface.co/datasets/electricsheepafrica/synthetic-tb-screening-symptoms-genexpert-WHO}
}
```

## License

[CC-BY-4.0](https://creativecommons.org/licenses/by/4.0/)