Upload folder using huggingface_hub

README.md

@@ -0,0 +1,65 @@
+---
+license: cc-by-4.0
+task_categories:
+  - tabular-classification
+language:
+  - en
+tags:
+  - laboratory
+  - HIV
+  - viral-load
+  - CD4
+  - ART-monitoring
+  - synthetic
+  - sub-saharan-africa
+pretty_name: HIV Viral Load & CD4 Testing
+size_categories:
+  - 10K<n<100K
+configs:
+  - config_name: vl_accessible
+    data_files: data/hiv_vl_cd4_vl_accessible.csv
+  - config_name: vl_limited
+    data_files: data/hiv_vl_cd4_vl_limited.csv
+    default: true
+  - config_name: no_vl_access
+    data_files: data/hiv_vl_cd4_no_vl_access.csv
+---
+
+# HIV Viral Load & CD4 Testing
+
+## Abstract
+
+Synthetic dataset modeling HIV viral load and CD4 laboratory services across three SSA scenarios. Captures VL/CD4 test availability, platforms (conventional/POC/DBS), results, turnaround times, clinical actions (EAC, regimen switch), and quality indicators. Parameterized from WHO/UNAIDS guidelines and SSA implementation research.
+
+## Parameterization Evidence
+
+| Parameter | Value | Source | Year |
+| --- | --- | --- | --- |
+| VL coverage SSA | 17% to 71% (2015-2019) | Lecher et al. MMWR | 2021 |
+| VL suppression on ART | 76% SSA | UNAIDS | 2023 |
+| Conventional VL TAT | 2-6 weeks | Jani et al. Lancet HIV | 2016 |
+| POC VL TAT | 1-3 hours | Jani et al. | 2016 |
+| SSA PLHIV on ART | 21M / 25.6M | UNAIDS | 2023 |
+
+## Validation
+
+![Validation Report](validation_report.png)
+
+## Usage
+
+```python
+from datasets import load_dataset
+ds = load_dataset("electricsheepafrica/hiv-viral-load-cd4", name="vl_limited")
+df = ds['train'].to_pandas()
+```
+
+## References
+
+1. WHO (2023). HIV treatment guidelines
+2. UNAIDS (2023). Global HIV statistics
+3. Lecher SL et al. (2021). VL monitoring scale-up. *MMWR*. DOI: 10.15585/mmwr.mm7034a2
+4. Jani IV et al. (2016). POC CD4 and VL. *Lancet HIV*. DOI: 10.1016/S2352-3018(15)00236-X
+
+## License
+
+CC-BY-4.0

generate_dataset.py

@@ -0,0 +1,176 @@
+#!/usr/bin/env python3
+"""
+HIV Viral Load & CD4 Testing Dataset
+========================================
+Each record = ONE HIV viral load or CD4 test request/result.
+
+Literature-Grounded Parameterization:
+[1] WHO (2023). HIV treatment guidelines.
+    - VL preferred monitoring; target <1000 copies/mL
+    - CD4 for staging and OI prophylaxis
+    - VL coverage SSA: ~70% of ART patients
+
+[2] UNAIDS (2023). Global HIV statistics.
+    - SSA: 25.6M PLHIV, 21M on ART
+    - VL suppression: 76% of those on ART
+
+[3] Lecher SL et al. (2021). Scale-up HIV VL monitoring.
+    MMWR. DOI: 10.15585/mmwr.mm7034a2
+    - VL coverage 17% to 71% (2015-2019)
+    - Centralized testing: TAT weeks
+
+[4] Jani IV et al. (2016). POC CD4 and VL testing.
+    Lancet HIV. DOI: 10.1016/S2352-3018(15)00236-X
+    - POC reduces loss to follow-up
+    - Conventional VL TAT: 2-6 weeks
+    - POC VL TAT: 1-3 hours
+    - DBS for sample transport
+"""
+import numpy as np, pandas as pd, argparse, os
+
+SCENARIOS = {
+    'vl_accessible': {
+        'exemplar': 'South Africa/Botswana/Kenya (urban)',
+        'vl_coverage': 0.75,
+        'cd4_available': 0.85,
+        'poc_vl_available': 0.20,
+        'dbs_used': 0.15,
+        'conventional_vl': 0.65,
+        'vl_suppression_rate': 0.80,
+        'cd4_median_on_art': 450,
+        'tat_vl_days': 7,
+        'tat_cd4_days': 1,
+        'eac_for_unsuppressed': 0.60,
+        'result_returned_patient': 0.70,
+        'reagent_stockout': 0.10,
+    },
+    'vl_limited': {
+        'exemplar': 'Kenya (district)/Ghana/Tanzania/Ethiopia',
+        'vl_coverage': 0.35,
+        'cd4_available': 0.50,
+        'poc_vl_available': 0.05,
+        'dbs_used': 0.30,
+        'conventional_vl': 0.25,
+        'vl_suppression_rate': 0.72,
+        'cd4_median_on_art': 380,
+        'tat_vl_days': 21,
+        'tat_cd4_days': 3,
+        'eac_for_unsuppressed': 0.30,
+        'result_returned_patient': 0.40,
+        'reagent_stockout': 0.30,
+    },
+    'no_vl_access': {
+        'exemplar': 'DRC/CAR/Sierra Leone/Niger (rural)',
+        'vl_coverage': 0.08,
+        'cd4_available': 0.15,
+        'poc_vl_available': 0.01,
+        'dbs_used': 0.05,
+        'conventional_vl': 0.05,
+        'vl_suppression_rate': 0.60,
+        'cd4_median_on_art': 300,
+        'tat_vl_days': 45,
+        'tat_cd4_days': 14,
+        'eac_for_unsuppressed': 0.10,
+        'result_returned_patient': 0.15,
+        'reagent_stockout': 0.55,
+    },
+}
+
+def generate_dataset(n=10000, seed=42, scenario='vl_limited'):
+    rng = np.random.default_rng(seed)
+    sc = SCENARIOS[scenario]
+    records = []
+    for idx in range(n):
+        rec = {'id': idx + 1}
+        rec['age'] = int(np.clip(rng.normal(38, 12), 1, 80))
+        rec['sex'] = rng.choice(['male','female'], p=[0.38, 0.62])
+        rec['pregnant'] = 1 if (rec['sex']=='female' and 15<=rec['age']<=45 and rng.random()<0.08) else 0
+        rec['facility_level'] = rng.choice(['tertiary','district','health_centre','clinic'],
+            p=[0.10, 0.25, 0.35, 0.30])
+        rec['on_art'] = 1 if rng.random() < 0.85 else 0
+        rec['art_duration_months'] = int(rng.exponential(36)) if rec['on_art'] else 0
+        rec['art_regimen'] = rng.choice(['TLD','TLE','AZT_based','PI_based','other'],
+            p=[0.40, 0.25, 0.15, 0.10, 0.10]) if rec['on_art'] else 'none'
+
+        # Test type ordered
+        rec['vl_ordered'] = 1 if rng.random() < sc['vl_coverage'] else 0
+        rec['cd4_ordered'] = 1 if rng.random() < sc['cd4_available'] else 0
+
+        # VL testing
+        tier = 1.3 if rec['facility_level'] in ['tertiary','district'] else 0.7
+        if rec['vl_ordered']:
+            rec['vl_platform'] = rng.choice(['conventional_central','poc_vl','dbs_referred'],
+                p=[sc['conventional_vl']*tier/(sc['conventional_vl']*tier+sc['poc_vl_available']+sc['dbs_used']+0.01),
+                   sc['poc_vl_available']/(sc['conventional_vl']*tier+sc['poc_vl_available']+sc['dbs_used']+0.01),
+                   (sc['dbs_used']+0.01)/(sc['conventional_vl']*tier+sc['poc_vl_available']+sc['dbs_used']+0.01)])
+            if rec['on_art']:
+                suppressed = rng.random() < sc['vl_suppression_rate']
+                if suppressed:
+                    rec['vl_result'] = int(np.clip(rng.exponential(50), 0, 999))
+                else:
+                    rec['vl_result'] = int(np.clip(rng.lognormal(np.log(5000), 1.5), 1000, 500000))
+            else:
+                rec['vl_result'] = int(np.clip(rng.lognormal(np.log(50000), 1.2), 1000, 1000000))
+            rec['vl_suppressed'] = 1 if rec['vl_result'] < 1000 else 0
+            rec['vl_tat_days'] = round(max(0.1, rng.lognormal(
+                np.log(0.1 if rec['vl_platform']=='poc_vl' else sc['tat_vl_days']), 0.5)), 1)
+        else:
+            rec['vl_platform'] = 'not_done'
+            rec['vl_result'] = np.nan
+            rec['vl_suppressed'] = np.nan
+            rec['vl_tat_days'] = np.nan
+
+        # CD4 testing
+        if rec['cd4_ordered']:
+            rec['cd4_platform'] = rng.choice(['flow_cytometry','poc_cd4','manual'],
+                p=[0.40, 0.30, 0.30] if sc['cd4_available']>0.5 else [0.15, 0.25, 0.60])
+            if rec['on_art']:
+                rec['cd4_result'] = int(np.clip(rng.normal(sc['cd4_median_on_art'], 180), 10, 1500))
+            else:
+                rec['cd4_result'] = int(np.clip(rng.lognormal(np.log(200), 0.7), 5, 800))
+            rec['cd4_below_200'] = 1 if rec['cd4_result'] < 200 else 0
+            rec['cd4_tat_days'] = round(max(0.1, rng.lognormal(np.log(sc['tat_cd4_days']), 0.5)), 1)
+        else:
+            rec['cd4_platform'] = 'not_done'
+            rec['cd4_result'] = np.nan
+            rec['cd4_below_200'] = np.nan
+            rec['cd4_tat_days'] = np.nan
+
+        # Clinical actions
+        rec['result_returned'] = 1 if rng.random() < sc['result_returned_patient'] else 0
+        rec['eac_provided'] = 1 if (rec['vl_ordered'] and rec.get('vl_suppressed')==0 and rng.random() < sc['eac_for_unsuppressed']) else 0
+        rec['regimen_switch'] = 1 if (rec['vl_ordered'] and rec.get('vl_suppressed')==0 and rng.random() < 0.15) else 0
+        rec['oi_prophylaxis'] = 1 if (rec['cd4_ordered'] and rec.get('cd4_below_200')==1 and rng.random() < 0.65) else 0
+
+        # Quality
+        rec['reagent_stockout'] = 1 if rng.random() < sc['reagent_stockout'] else 0
+        rec['sample_rejected'] = 1 if rng.random() < 0.04 else 0
+        rec['test_not_done_stockout'] = 1 if (rec['reagent_stockout'] and rng.random() < 0.40) else 0
+        rec['specimen_transport_issue'] = 1 if (rec.get('vl_platform') in ['dbs_referred','conventional_central'] and rng.random() < 0.12) else 0
+
+        rec['year'] = rng.choice([2019,2020,2021,2022,2023], p=[0.12,0.18,0.20,0.25,0.25])
+        records.append(rec)
+
+    df = pd.DataFrame(records)
+    print(f"\n{'='*60}\nHIV Viral Load & CD4 — {scenario} ({sc['exemplar']})")
+    print(f"  VL ordered: {df['vl_ordered'].mean()*100:.1f}% | CD4 ordered: {df['cd4_ordered'].mean()*100:.1f}%")
+    vl_done = df[df['vl_ordered']==1]
+    if len(vl_done)>0: print(f"  VL suppressed: {vl_done['vl_suppressed'].mean()*100:.1f}% | VL TAT median: {vl_done['vl_tat_days'].median():.1f}d")
+    print(f"  Result returned: {df['result_returned'].mean()*100:.1f}% | Stockout: {df['reagent_stockout'].mean()*100:.1f}%")
+    return df
+
+if __name__ == '__main__':
+    parser = argparse.ArgumentParser()
+    parser.add_argument('--all-scenarios', action='store_true')
+    parser.add_argument('--n', type=int, default=10000)
+    parser.add_argument('--seed', type=int, default=42)
+    args = parser.parse_args()
+    os.makedirs('data', exist_ok=True)
+    if args.all_scenarios:
+        for sc in SCENARIOS:
+            df = generate_dataset(n=args.n, seed=args.seed, scenario=sc)
+            df.to_csv(os.path.join('data', f'hiv_vl_cd4_{sc}.csv'), index=False)
+            print(f"  -> Saved\n")
+    else:
+        df = generate_dataset(n=args.n, seed=args.seed)
+        df.to_csv(os.path.join('data', 'hiv_vl_cd4_vl_limited.csv'), index=False)

requirements.txt

@@ -0,0 +1,3 @@
+numpy>=1.24
+pandas>=2.0
+matplotlib>=3.7

validate_dataset.py

@@ -0,0 +1,83 @@
+#!/usr/bin/env python3
+"""Validation for HIV Viral Load & CD4 Testing Dataset."""
+import pandas as pd, numpy as np, matplotlib.pyplot as plt, os, glob
+
+def load_scenarios(data_dir='data'):
+    dfs = {}
+    for f in sorted(glob.glob(os.path.join(data_dir, 'hiv_vl_cd4_*.csv'))):
+        name = os.path.basename(f).replace('.csv', '')[11:]
+        dfs[name] = pd.read_csv(f)
+    return dfs
+
+def main():
+    dfs = load_scenarios()
+    if not dfs: return
+    all_df = pd.concat([df.assign(scenario=n) for n, df in dfs.items()], ignore_index=True)
+    fig, axes = plt.subplots(4, 2, figsize=(16, 20))
+    fig.suptitle('HIV Viral Load & CD4 Testing — Validation Report', fontsize=14, fontweight='bold', y=0.98)
+    colors = {'vl_accessible': '#2ecc71', 'vl_limited': '#f39c12', 'no_vl_access': '#e74c3c'}
+    labels = {'vl_accessible': 'VL Access (SA/BW)', 'vl_limited': 'Limited (KE/GH/TZ)', 'no_vl_access': 'None (DRC/SLE)'}
+    scenarios = list(dfs.keys())
+
+    ax = axes[0, 0]
+    metrics = ['VL Ordered %', 'CD4 Ordered %', 'Result\nReturned %', 'EAC %', 'Stockout %']
+    for i, s in enumerate(scenarios):
+        d = dfs[s]; vals = [d['vl_ordered'].mean()*100, d['cd4_ordered'].mean()*100, d['result_returned'].mean()*100, d['eac_provided'].mean()*100, d['reagent_stockout'].mean()*100]
+        ax.bar(np.arange(len(metrics))+i*0.25, vals, 0.25, label=labels.get(s,s), color=colors[s], alpha=0.8)
+    ax.set_xticks(np.arange(len(metrics))+0.25); ax.set_xticklabels(metrics, fontsize=6); ax.set_ylabel('%'); ax.set_title('Panel 1: Key Metrics'); ax.legend(fontsize=6)
+
+    ax = axes[0, 1]
+    for s in scenarios:
+        vl = dfs[s][dfs[s]['vl_ordered']==1]['vl_result'].dropna()
+        if len(vl) > 0: ax.hist(np.log10(vl.clip(lower=1)), bins=25, alpha=0.5, label=labels.get(s,s), color=colors[s], density=True)
+    ax.axvline(3, color='black', linestyle='--', alpha=0.5, label='1000 cp/mL')
+    ax.set_xlabel('Log10 VL (copies/mL)'); ax.set_title('Panel 2: Viral Load Distribution'); ax.legend(fontsize=7)
+
+    ax = axes[1, 0]
+    for s in scenarios:
+        cd4 = dfs[s][dfs[s]['cd4_ordered']==1]['cd4_result'].dropna()
+        if len(cd4) > 0: ax.hist(cd4.clip(upper=1000), bins=25, alpha=0.5, label=labels.get(s,s), color=colors[s], density=True)
+    ax.axvline(200, color='black', linestyle='--', alpha=0.5, label='200 cells')
+    ax.set_xlabel('CD4 (cells/uL)'); ax.set_title('Panel 3: CD4 Distribution'); ax.legend(fontsize=7)
+
+    ax = axes[1, 1]
+    for s in scenarios:
+        vl_tat = dfs[s][dfs[s]['vl_ordered']==1]['vl_tat_days'].dropna()
+        if len(vl_tat) > 0: ax.hist(vl_tat.clip(upper=60), bins=25, alpha=0.5, label=labels.get(s,s), color=colors[s], density=True)
+    ax.set_xlabel('VL TAT (days)'); ax.set_title('Panel 4: VL Turnaround Time'); ax.legend(fontsize=7)
+
+    ax = axes[2, 0]
+    plats = ['conventional_central','poc_vl','dbs_referred','not_done']
+    for i, s in enumerate(scenarios):
+        vals = [dfs[s]['vl_platform'].value_counts(normalize=True).get(p,0)*100 for p in plats]
+        ax.bar(np.arange(len(plats))+i*0.25, vals, 0.25, label=labels.get(s,s), color=colors[s], alpha=0.8)
+    ax.set_xticks(np.arange(len(plats))+0.25); ax.set_xticklabels([p.replace('_','\n') for p in plats], fontsize=5); ax.set_ylabel('%'); ax.set_title('Panel 5: VL Platform'); ax.legend(fontsize=6)
+
+    ax = axes[2, 1]
+    acts = ['EAC', 'Regimen\nSwitch', 'OI\nProphylaxis', 'Transport\nIssue', 'Sample\nRejected']
+    for i, s in enumerate(scenarios):
+        d = dfs[s]; vals = [d['eac_provided'].mean()*100, d['regimen_switch'].mean()*100, d['oi_prophylaxis'].mean()*100, d['specimen_transport_issue'].mean()*100, d['sample_rejected'].mean()*100]
+        ax.bar(np.arange(len(acts))+i*0.25, vals, 0.25, label=labels.get(s,s), color=colors[s], alpha=0.8)
+    ax.set_xticks(np.arange(len(acts))+0.25); ax.set_xticklabels(acts, fontsize=5); ax.set_ylabel('%'); ax.set_title('Panel 6: Clinical Actions & Issues'); ax.legend(fontsize=6)
+
+    ax = axes[3, 0]
+    regs = ['TLD','TLE','AZT_based','PI_based','none']
+    for i, s in enumerate(scenarios):
+        vals = [dfs[s]['art_regimen'].value_counts(normalize=True).get(r,0)*100 for r in regs]
+        ax.bar(np.arange(len(regs))+i*0.20, vals, 0.20, label=labels.get(s,s), color=colors[s], alpha=0.8)
+    ax.set_xticks(np.arange(len(regs))+0.20); ax.set_xticklabels(regs, fontsize=6); ax.set_ylabel('%'); ax.set_title('Panel 7: ART Regimen'); ax.legend(fontsize=6)
+
+    ax = axes[3, 1]
+    num_cols = ['vl_ordered','cd4_ordered','result_returned','reagent_stockout','eac_provided','on_art']
+    corr = all_df[num_cols].corr()
+    im = ax.imshow(corr, cmap='RdBu_r', vmin=-1, vmax=1, aspect='auto')
+    ax.set_xticks(range(len(num_cols))); ax.set_yticks(range(len(num_cols)))
+    ax.set_xticklabels([c.replace('_','\n') for c in num_cols], fontsize=5, rotation=45, ha='right')
+    ax.set_yticklabels([c.replace('_','\n') for c in num_cols], fontsize=5)
+    ax.set_title('Panel 8: Correlation Heatmap'); fig.colorbar(im, ax=ax, fraction=0.046)
+
+    plt.tight_layout(rect=[0,0,1,0.96]); plt.savefig('validation_report.png', dpi=150, bbox_inches='tight'); plt.close()
+    print("Saved validation_report.png")
+
+if __name__ == '__main__':
+    main()