---
license: cc-by-4.0
task_categories:
- tabular-classification
- tabular-regression
language:
- en
tags:
- synthetic
- healthcare
- hiv
- antiretroviral-therapy
- treatment-cascade
- 90-90-90
- viral-load
- cd4
- pmtct
- sub-saharan-africa
- lmic
- unaids
pretty_name: HIV/ART Treatment Cascade Dataset (Cascade Stage, VL, CD4, Adherence,
  PMTCT)
size_categories:
- 10K<n<100K
configs:
- config_name: high_performing
  data_files: data/hiv_art_high_performing.csv
- config_name: moderate_performing
  data_files: data/hiv_art_moderate_performing.csv
  default: true
- config_name: low_performing
  data_files: data/hiv_art_low_performing.csv
data_type: synthetic
---

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

# HIV/ART Treatment Cascade Dataset

## Abstract

This dataset provides **30,000 synthetic records** (10,000 per scenario) of people living with HIV (PLHIV) aged 15–65 years across the treatment cascade in sub-Saharan African ART programmes. Each record contains 20 variables spanning cascade stage (undiagnosed / diagnosed not on ART / on ART), ART regimen and line, CD4 count at diagnosis and current, viral load, adherence category, WHO clinical stage, TB co-infection, pregnancy/PMTCT status, VL monitoring, and outcome (alive in care / LTFU / transferred / died). Three programme-performance scenarios range from high-performing (59% population-level VL suppression, 4.2% mortality) to low-performing (25% suppression, 12.1% mortality), reflecting real-world disparities across sub-Saharan Africa.

## 1. Introduction

Sub-Saharan Africa bears the largest HIV burden globally, with 25.6 million people on ART (UNAIDS 2023). The UNAIDS 90-90-90 targets aim for 90% diagnosed, 90% of diagnosed on ART, and 90% of those on ART virally suppressed. A systematic review of 92 studies (Hakim et al., JIAS 2020) found substantial heterogeneity: youth aged 15–24 achieved only 60-49-81 across the cascade, while men achieved 66-72-85 compared to women at 79-76-89. Dolutegravir-based regimens have improved retention and suppression rates (Dzangare et al., 2023).

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

## 2. Methodology

### 2.1 Epidemiological Parameterization

| Parameter | Value | Source |
| --- | --- | --- |
| PLHIV on ART in E/S Africa | 25.6 million | UNAIDS 2023 |
| Cascade (youth 15–24) | 60-49-81 | Hakim et al., JIAS 2020 (92 studies) |
| Cascade (men) | 66-72-85 | Hakim et al., 2020 |
| Cascade (women) | 79-76-89 | Hakim et al., 2020 |
| CD4 at ART initiation (SSA) | Median 200–350; late presenters 30–50% | IeDEA Consortium |
| CD4 recovery year 1 | ~120 cells/µL/yr | IeDEA |
| VL suppression on ART | 85–92% | Bock et al., Lancet HIV 2022 |
| 12-month retention | 75–85% (good programmes) | Onoya et al., 2021 |
| DTG vs EFV retention | aRR 1.09 (95% CI 1.04–1.14) | Dzangare et al., 2023 |
| PMTCT ART coverage | 82% globally | UNAIDS 2022 |
| TB co-infection (HIV+) | 8–18% depending on setting | WHO TB/HIV 2022 |
| Pre-treatment NNRTI resistance | 10–15% | Gupta et al., Lancet ID 2018 |

### 2.2 Scenario Design

| Scenario | Setting | Diagnosed | On ART | VL Suppressed | Pop VL Supp | Deaths |
| --- | --- | --- | --- | --- | --- | --- |
| High performing | Botswana, Rwanda | 90% | 89% | 92% | 59.3% | 4.2% |
| Moderate performing | Kenya, Tanzania | 80% | 78% | 88% | 43.2% | 7.1% |
| Low performing | Nigeria, DRC | 65% | 62% | 80% | 25.3% | 12.1% |

## 3. Schema

| Column | Type | Description |
| --- | --- | --- |
| id | int | Unique identifier |
| age_years | int | Age (15–65) |
| sex | categorical | M / F |
| cascade_stage | categorical | undiagnosed / diagnosed_not_on_ART / on_ART |
| years_on_art | float | Duration on ART (years; 0 if not on ART) |
| art_regimen | categorical | TDF/3TC/DTG, TDF/3TC/EFV, AZT/3TC/ATV-r, DRV-r/DTG/AZT/3TC, none |
| regimen_line | int | 0=none, 1=first-line, 2=second-line, 3=third-line |
| ever_switched_regimen | binary | Ever switched ART regimen |
| cd4_at_diagnosis | int | CD4 count at HIV diagnosis (cells/µL) |
| cd4_current | int | Current CD4 count (cells/µL) |
| adherence_category | categorical | good / moderate / poor / unknown |
| adherence_pct | int | Self-reported adherence (%) |
| viral_load_copies_ml | int | Most recent viral load (copies/mL) |
| vl_suppressed | binary | VL <1000 copies/mL (WHO threshold) |
| vl_test_done_12mo | binary | VL test done in last 12 months |
| who_clinical_stage | int | WHO clinical stage (1–4) |
| tb_coinfection | binary | Active TB co-infection |
| pregnant | binary | Currently pregnant (females only) |
| on_pmtct | binary | Receiving PMTCT services |
| outcome | categorical | alive_in_care / ltfu / transferred / died / undiagnosed / pre_art_care |

## 4. Validation

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

Key validation checks:

- **Cascade monotonicity**: On-ART proportion decreases with programme weakness ✓
- **VL suppression by adherence**: Good 90%+ → moderate ~65% → poor ~25% ✓
- **CD4 recovery**: On-ART CD4 higher than not-on-ART ✓
- **Age-sex gap**: Men and youth have lower cascade achievement ✓
- **DTG dominance**: First-line DTG-based regimen most common ✓
- **Population VL suppression**: 59% → 43% → 25% across scenarios ✓

## 5. Usage

```python
from datasets import load_dataset
dataset = load_dataset("electricsheepafrica/hiv-art-treatment-cascade", "moderate_performing")
df = dataset["train"].to_pandas()
```

```python
import pandas as pd
df = pd.read_csv("data/hiv_art_moderate_performing.csv")
```

Regenerate:

```bash
python generate_dataset.py --scenario low_performing --n 20000 --seed 123
python generate_dataset.py --all-scenarios --n 10000 --seed 42
```

## 6. Limitations

- **Synthetic**: Not derived from real patient records.
- **Cross-sectional**: Single time-point snapshot; no longitudinal CD4/VL trajectories.
- **No resistance genotyping**: Drug resistance mutations not modelled.
- **Simplified adherence**: Three-category model; real adherence is continuous and time-varying.
- **No geographic specificity**: Parameters are regional averages, not country-specific.
- **No opportunistic infections**: WHO staging simplified to CD4-driven; no specific OI modelling.
- **No paediatric HIV**: Only adults ≥15 years; PMTCT modelled for pregnant women only.

## 7. References

1. UNAIDS (2023). Global AIDS Update. 25.6M on ART in E/S Africa.
2. Hakim J, et al. (2020). Sociodemographic heterogeneity across HIV treatment cascade in SSA. *JIAS*, 23(2):e25470. PMID: 32153117
3. WHO (2019). Updated recommendations on first-line and second-line ART.
4. IeDEA Consortium. CD4 at ART initiation in SSA.
5. Bock P, et al. (2022). Viral load monitoring cascade in SSA. *Lancet HIV*.
6. Onoya D, et al. (2021). 12-month retention in care in SSA ART programmes.
7. WHO (2007/2014). WHO clinical staging of HIV/AIDS.
8. Gupta RK, et al. (2018). HIV drug resistance in SSA. *Lancet Infect Dis*, 18(12):1378–1390.
9. Dzangare J, et al. (2023). Implementation of DTG-based first-line regimens. *Lancet HIV*. PMID: 37001536
10. Shroufi A, et al. (2021). TLD as second-line. *JAIDS*. PMID: 33973876
11. UNAIDS (2022). PMTCT: 82% coverage globally.

## Citation

```bibtex
@dataset{esa_hiv_art_cascade_2025,
  title={HIV/ART Treatment Cascade Dataset},
  author={Electric Sheep Africa},
  year={2025},
  publisher={Hugging Face},
  url={https://huggingface.co/datasets/electricsheepafrica/hiv-art-treatment-cascade}
}
```

## License

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