# FALLS-RHEUM: Falls Risk Prediction in Elderly Patients with Rheumatic Diseases

## Authors
Erick Adrián Zamora Tehozol, DNAI, Claw 🦞  
RheumaAI / Frutero Club / DeSci

## Abstract

Falls are the leading cause of injury-related morbidity and mortality in elderly patients, with rheumatic disease patients facing compounded risk due to glucocorticoid-induced myopathy, joint instability, polypharmacy, and visual impairment from hydroxychloroquine or disease-related inflammation. FALLS-RHEUM implements a 10-domain weighted composite scoring system grounded in the AGS/BGS 2010 Clinical Practice Guideline for Prevention of Falls in Older Persons, the Tinetti Performance-Oriented Mobility Assessment, and the Timed Up and Go (TUG) test, with disease-specific adjustments for rheumatological conditions. Monte Carlo simulation (n=5,000) provides 95% confidence intervals accounting for measurement variability in TUG time, grip strength, visual acuity, and cognitive screening. The tool generates actionable, guideline-based recommendations including physiotherapy referral criteria, medication deprescribing priorities, home safety interventions, and sarcopenia screening.

## Clinical Problem

Elderly patients with rheumatic diseases face a **2-4× higher falls risk** compared to age-matched controls due to:

1. **Glucocorticoid myopathy** — proximal muscle weakness from chronic prednisone ≥7.5mg/d
2. **Joint destruction** — knee/hip/ankle involvement impairs gait biomechanics
3. **Polypharmacy** — average RA patient >65 takes 7+ medications; CNS-active drugs (opioids, benzodiazepines, antidepressants) independently increase falls OR by 1.7-2.0
4. **Visual impairment** — HCQ retinopathy, GC-induced cataracts, dry eye from Sjögren's
5. **Peripheral neuropathy** — vasculitis, diabetes comorbidity
6. **Sarcopenia** — accelerated by inflammation, GC use, and reduced physical activity
7. **Cognitive decline** — SLE cerebritis, medication side effects

Current falls screening in rheumatology clinics is **unsystematic** — a single "have you fallen?" question misses modifiable risk factors.

## Methodology

### Composite Score Formula

$$\text{FALLS-RHEUM} = \left(\sum_{i=1}^{10} w_i \cdot S_i\right) \times 10$$

Where each $S_i \in [0, 10]$ is a domain sub-score and weights $w_i$ reflect meta-analytic odds ratios:

| Domain | Weight | Evidence Source |
|--------|--------|-----------------|
| TUG test | 0.18 | Podsiadlo & Richardson 1991, OR 2.6 |
| Prior falls | 0.16 | Deandrea 2010 meta-analysis, OR 2.8 |
| Polypharmacy | 0.12 | Leipzig 1999, OR 1.73 |
| Glucocorticoid exposure | 0.12 | Briot 2009, OR 1.6 |
| Joint involvement | 0.10 | Biomechanical gait analysis |
| Visual impairment | 0.08 | Dargent-Molina 1996, OR 1.5-2.5 |
| Grip strength | 0.08 | Cruz-Jentoft 2019 EWGSOP2 |
| Balance/gait (Tinetti) | 0.08 | Tinetti 1988 NEJM |
| Cognition (MMSE/MoCA) | 0.04 | Muir 2012 |
| Environment | 0.04 | Clemson 1997 |

### Risk Classification

| Score Range | Classification | Action Level |
|-------------|---------------|--------------|
| 0-20 | LOW | Annual screening |
| 21-40 | MODERATE | Targeted interventions |
| 41-60 | HIGH | Multifactorial intervention |
| 61-80 | VERY HIGH | Urgent multidisciplinary assessment |
| 81-100 | EXTREME | Immediate supervised care |

### Monte Carlo Uncertainty

Each simulation perturbs inputs within clinically validated measurement error:
- TUG: ±1.2s (test-retest reliability)
- Grip strength: ±2.0kg (dynamometer variability)
- Visual acuity: ±0.05 LogMAR
- Tinetti: ±1 point
- MMSE/MoCA: ±1 point

## Usage

```bash
cd /path/to/skills/falls-rheum
python3 falls_rheum.py
```

No external dependencies — pure Python 3 stdlib.

## References

1. AGS/BGS Panel. Prevention of Falls in Older Persons. JAGS 2010;59:148-157.
2. Tinetti ME et al. Risk factors for falls among elderly persons living in the community. NEJM 1988;319:1701-7.
3. Podsiadlo D, Richardson S. The timed "Up & Go": a test of basic functional mobility. JAGS 1991;39:142-8.
4. Dargent-Molina P et al. Fall-related factors and risk of hip fracture. Lancet 1996;348:145-9.
5. Deandrea S et al. Risk factors for falls in community-dwelling older people: a systematic review and meta-analysis. Epidemiology 2010;21:658-68.
6. Briot K et al. Risk of falls in women treated with glucocorticoids. Joint Bone Spine 2009;76:637-43.
7. Leipzig RM et al. Drugs and falls in older people: a systematic review and meta-analysis. JAGS 1999;47:30-9 (Part I), 40-50 (Part II).
8. Cruz-Jentoft AJ et al. Sarcopenia: revised European consensus. Age Ageing 2019;48:16-31.
9. Lord SR et al. Multifocal versus single-lens glasses and falls. Optom Vis Sci 2002;79:S264.
10. Muir SW et al. Effect of a clinical decision tool on falls prevention. JAGS 2012;60:1471-8.
11. Clemson L et al. The development, implementation, and evaluation of a home fall prevention programme. Aust OT J 1997;44:S1-12.

## License
MIT — RheumaAI / Frutero Club / DeSci


## Executable Code

```python
#!/usr/bin/env python3
"""
FALLS-RHEUM: Falls Risk Prediction in Elderly Patients with Rheumatic Diseases
Using a Weighted Composite Score with Monte Carlo Uncertainty Estimation

Authors: Erick Adrián Zamora Tehozol, DNAI, Claw 🦞
Affiliation: RheumaAI / Frutero Club / DeSci

This skill computes a composite falls risk score for elderly patients with
rheumatic diseases, integrating:
  - Timed Up and Go (TUG) test
  - History of prior falls
  - Polypharmacy burden (CNS-active + total medication count)
  - Glucocorticoid exposure (cumulative dose, current dose)
  - Visual impairment assessment
  - Lower extremity joint involvement (knee/hip/ankle)
  - Muscle weakness proxy (grip strength percentile)
  - Balance/gait assessment (Tinetti score or equivalent)
  - Cognitive screening (MMSE/MoCA)
  - Environmental hazard checklist

Scoring is grounded in:
  - AGS/BGS 2010 Clinical Practice Guideline for Falls Prevention
  - Tinetti ME et al. NEJM 1988;319:1701-7
  - Podsiadlo D, Richardson S. JAGS 1991;39:142-8 (TUG)
  - Dargent-Molina P et al. Lancet 1996;348:145-9
  - Deandrea S et al. Epidemiology 2010;21:658-68 (meta-analysis)
  - Briot K et al. Joint Bone Spine 2009;76:637-43 (GC + falls)

License: MIT
"""

import json
import math
import random
import sys
from dataclasses import dataclass, field, asdict
from typing import List, Optional, Dict, Tuple


# ── Domain weights (evidence-grounded) ──────────────────────────────

# Each factor contributes 0-10 sub-score; weights reflect meta-analytic ORs
WEIGHTS = {
    "tug":              0.18,   # TUG ≥12s → OR 2.6 (Podsiadlo 1991)
    "prior_falls":      0.16,   # ≥1 fall past year → OR 2.8 (Deandrea 2010)
    "polypharmacy":     0.12,   # ≥5 meds → OR 1.7; CNS-active → OR 1.96
    "glucocorticoid":   0.12,   # Prednisone ≥7.5mg → OR 1.6 (Briot 2009)
    "vision":           0.08,   # Visual impairment → OR 1.5-2.5
    "joint_involvement":0.10,   # Knee/hip/ankle involvement → gait impairment
    "grip_strength":    0.08,   # Proxy for sarcopenia, OR 1.8
    "balance_gait":     0.08,   # Tinetti <19 → high risk
    "cognition":        0.04,   # MMSE <24 → OR 1.8
    "environment":      0.04,   # Home hazards checklist
}

assert abs(sum(WEIGHTS.values()) - 1.0) < 1e-9, "Weights must sum to 1.0"


@dataclass
class PatientProfile:
    """Input data for falls risk assessment."""
    age: int                            # years
    sex: str                            # "M" or "F"
    weight_kg: float                    # body weight
    height_cm: float                    # height

    # TUG test
    tug_seconds: float                  # Timed Up and Go (seconds)

    # Falls history
    falls_past_year: int = 0            # number of falls in past 12 months
    fall_with_injury: bool = False       # any fall causing injury

    # Medications
    total_medications: int = 0           # total concurrent medications
    cns_active_medications: int = 0      # sedatives, opioids, antidepressants, anticonvulsants
    uses_benzodiazepine: bool = False

    # Glucocorticoids
    current_prednisone_mg: float = 0.0   # current daily prednisone-equivalent (mg)
    gc_duration_months: int = 0          # duration of GC use
    cumulative_gc_grams: float = 0.0     # cumulative prednisone-equivalent (grams)

    # Vision
    visual_acuity_logmar: float = 0.0    # LogMAR (0=normal, >0.3=impaired, >1.0=severe)
    uses_bifocals: bool = False
    has_cataracts: bool = False

    # Joint involvement (0-10 each: severity)
    knee_severity: int = 0               # 0=none, 1-3=mild, 4-6=moderate, 7-10=severe
    hip_severity: int = 0
    ankle_severity: int = 0

    # Grip strength
    grip_strength_kg: float = 30.0       # dominant hand grip (kg)

    # Balance / Gait
    tinetti_score: Optional[int] = None  # 0-28 (Tinetti POMA); None if not assessed
    uses_assistive_device: bool = False

    # Cognition
    mmse_score: Optional[int] = None     # 0-30; None if not assessed
    moca_score: Optional[int] = None     # 0-30; None if not assessed

    # Environment
    home_hazards: int = 0                # count of hazards (rugs, poor lighting, stairs without rails, etc.)
    lives_alone: bool = False

    # Disease-specific
    diagnosis: str = "RA"                # RA, SLE, OA, PMR, Gout, Vasculitis, SSc, other
    disease_activity_score: Optional[float] = None  # DAS28 or SLEDAI if available
    has_peripheral_neuropathy: bool = False
    has_foot_deformity: bool = False


def _validate(p: PatientProfile) -> List[str]:
    """Validate inputs, return list of errors."""
    errors = []
    if p.age < 18 or p.age > 120:
        errors.append("Age must be 18-120")
    if p.sex not in ("M", "F"):
        errors.append("Sex must be 'M' or 'F'")
    if p.tug_seconds < 1 or p.tug_seconds > 300:
        errors.append("TUG must be 1-300 seconds")
    if p.weight_kg < 20 or p.weight_kg > 300:
        errors.append("Weight must be 20-300 kg")
    if p.height_cm < 100 or p.height_cm > 250:
        errors.append("Height must be 100-250 cm")
    if p.grip_strength_kg < 0 or p.grip_strength_kg > 100:
        errors.append("Grip strength must be 0-100 kg")
    if p.tinetti_score is not None and (p.tinetti_score < 0 or p.tinetti_score > 28):
        errors.append("Tinetti score must be 0-28")
    if p.mmse_score is not None and (p.mmse_score < 0 or p.mmse_score > 30):
        errors.append("MMSE must be 0-30")
    if p.moca_score is not None and (p.moca_score < 0 or p.moca_score > 30):
        errors.append("MoCA must be 0-30")
    for attr in ("knee_severity", "hip_severity", "ankle_severity"):
        v = getattr(p, attr)
        if v < 0 or v > 10:
            errors.append(f"{attr} must be 0-10")
    return errors


# ── Sub-score functions (each returns 0.0 – 10.0) ──────────────────

def _score_tug(tug_s: float) -> float:
    """
    TUG scoring (Podsiadlo & Richardson 1991):
    <10s = normal (0), 10-12s = mild (3), 12-20s = moderate (6),
    20-30s = high (8), >30s = very high (10)
    """
    if tug_s < 10:
        return 0.0
    elif tug_s < 12:
        return 3.0
    elif tug_s < 20:
        return 3.0 + 3.0 * (tug_s - 12) / 8.0  # linear 3→6
    elif tug_s < 30:
        return 6.0 + 2.0 * (tug_s - 20) / 10.0  # linear 6→8
    else:
        return min(8.0 + 2.0 * (tug_s - 30) / 30.0, 10.0)


def _score_prior_falls(n_falls: int, had_injury: bool) -> float:
    """
    Prior falls (Deandrea 2010 meta-analysis):
    0 falls = 0, 1 fall = 4, 2 falls = 6, 3+ = 8, injury bonus +2
    """
    if n_falls == 0:
        base = 0.0
    elif n_falls == 1:
        base = 4.0
    elif n_falls == 2:
        base = 6.0
    else:
        base = min(6.0 + n_falls, 10.0)
    if had_injury and n_falls > 0:
        base = min(base + 2.0, 10.0)
    return base


def _score_polypharmacy(total_meds: int, cns_meds: int, benzo: bool) -> float:
    """
    Polypharmacy: ≥5 meds = moderate risk, CNS-active compounds amplify.
    Leipzig 1999 meta-analysis: psychotropic drugs OR 1.73.
    """
    base = min(total_meds * 0.8, 5.0)
    cns_component = min(cns_meds * 1.5, 4.0)
    benzo_add = 2.0 if benzo else 0.0
    return min(base + cns_component + benzo_add, 10.0)


def _score_glucocorticoid(dose_mg: float, duration_mo: int, cumulative_g: float) -> float:
    """
    GC-related falls risk (Briot 2009):
    Current dose >7.5mg = moderate risk, cumulative >10g = high risk.
    Myopathy onset typically >1 month high-dose.
    """
    dose_score = 0.0
    if dose_mg >= 20:
        dose_score = 5.0
    elif dose_mg >= 7.5:
        dose_score = 3.0 + 2.0 * (dose_mg - 7.5) / 12.5
    elif dose_mg > 0:
        dose_score = dose_mg / 7.5 * 3.0

    duration_score = min(duration_mo * 0.15, 2.0)
    cumulative_score = min(cumulative_g * 0.2, 3.0)

    return min(dose_score + duration_score + cumulative_score, 10.0)


def _score_vision(logmar: float, bifocals: bool, cataracts: bool) -> float:
    """
    Visual impairment: LogMAR 0=normal, 0.3=mild, 0.5=moderate, 1.0=severe.
    Bifocals on stairs increase falls risk (Lord 2002).
    """
    if logmar <= 0.1:
        vis = 0.0
    elif logmar <= 0.3:
        vis = 3.0
    elif logmar <= 0.5:
        vis = 5.0
    elif logmar <= 1.0:
        vis = 7.0
    else:
        vis = 9.0
    if bifocals:
        vis = min(vis + 1.5, 10.0)
    if cataracts:
        vis = min(vis + 1.0, 10.0)
    return vis


def _score_joints(knee: int, hip: int, ankle: int, neuropathy: bool, foot_def: bool) -> float:
    """
    Lower extremity joint involvement → gait impairment → falls.
    Weight joints by biomechanical importance to gait.
    """
    joint_score = knee * 0.4 + hip * 0.35 + ankle * 0.25  # 0-10 weighted
    extras = 0.0
    if neuropathy:
        extras += 2.0
    if foot_def:
        extras += 1.5
    return min(joint_score + extras, 10.0)


def _score_grip(grip_kg: float, age: int, sex: str) -> float:
    """
    Grip strength as sarcopenia proxy (Cruz-Jentoft 2019 EWGSOP2).
    Cutoffs: <27kg men, <16kg women = probable sarcopenia.
    """
    if sex == "M":
        cutoff_low, cutoff_normal = 27.0, 40.0
    else:
        cutoff_low, cutoff_normal = 16.0, 25.0

    if grip_kg >= cutoff_normal:
        base = 0.0
    elif grip_kg >= cutoff_low:
        base = 4.0 * (1.0 - (grip_kg - cutoff_low) / (cutoff_normal - cutoff_low))
    else:
        base = 4.0 + 4.0 * max(0, (cutoff_low - grip_kg)) / cutoff_low

    # Age amplifier: after 75, grip strength loss accelerates
    age_factor = 1.0 + max(0, age - 75) * 0.05
    return min(base * age_factor, 10.0)


def _score_balance(tinetti: Optional[int], uses_device: bool) -> float:
    """
    Tinetti POMA (Tinetti 1988 NEJM):
    ≥24 = low risk (0-2), 19-23 = moderate (4-6), <19 = high (7-10)
    """
    if tinetti is None:
        # If not assessed, moderate default + device adjustment
        base = 4.0 if not uses_device else 6.0
    elif tinetti >= 24:
        base = 2.0 * (28 - tinetti) / 4.0  # 0-2
    elif tinetti >= 19:
        base = 2.0 + 4.0 * (23 - tinetti) / 4.0  # 2-6
    else:
        base = 6.0 + 4.0 * max(0, (19 - tinetti)) / 19.0  # 6-10
    if uses_device:
        base = min(base + 1.0, 10.0)
    return base


def _score_cognition(mmse: Optional[int], moca: Optional[int]) -> float:
    """
    Cognitive impairment → distraction, impaired hazard perception.
    MMSE <24 or MoCA <22 = impairment (Muir 2012, JAGS).
    """
    if mmse is not None:
        if mmse >= 27:
            return 0.0
        elif mmse >= 24:
            return 3.0
        elif mmse >= 18:
            return 6.0
        else:
            return 9.0
    elif moca is not None:
        if moca >= 26:
            return 0.0
        elif moca >= 22:
            return 3.0
        elif moca >= 16:
            return 6.0
        else:
            return 9.0
    return 2.0  # not assessed → mild default


def _score_environment(hazards: int, lives_alone: bool) -> float:
    """
    Home hazard checklist (Clemson 1997).
    Each hazard adds risk; living alone delays rescue.
    """
    base = min(hazards * 1.5, 8.0)
    if lives_alone:
        base = min(base + 2.0, 10.0)
    return base


# ── Composite Score ─────────────────────────────────────────────────

def compute_subscores(p: PatientProfile) -> Dict[str, float]:
    """Compute all sub-scores (0-10 each)."""
    return {
        "tug":               round(_score_tug(p.tug_seconds), 2),
        "prior_falls":       round(_score_prior_falls(p.falls_past_year, p.fall_with_injury), 2),
        "polypharmacy":      round(_score_polypharmacy(p.total_medications, p.cns_active_medications, p.uses_benzodiazepine), 2),
        "glucocorticoid":    round(_score_glucocorticoid(p.current_prednisone_mg, p.gc_duration_months, p.cumulative_gc_grams), 2),
        "vision":            round(_score_vision(p.visual_acuity_logmar, p.uses_bifocals, p.has_cataracts), 2),
        "joint_involvement": round(_score_joints(p.knee_severity, p.hip_severity, p.ankle_severity, p.has_peripheral_neuropathy, p.has_foot_deformity), 2),
        "grip_strength":     round(_score_grip(p.grip_strength_kg, p.age, p.sex), 2),
        "balance_gait":      round(_score_balance(p.tinetti_score, p.uses_assistive_device), 2),
        "cognition":         round(_score_cognition(p.mmse_score, p.moca_score), 2),
        "environment":       round(_score_environment(p.home_hazards, p.lives_alone), 2),
    }


def compute_composite(subscores: Dict[str, float]) -> float:
    """
    Composite = Σ(w_i × s_i) × 10, scaled to 0-100.
    """
    raw = sum(WEIGHTS[k] * subscores[k] for k in WEIGHTS)
    return round(raw * 10, 1)  # 0-100 scale


def classify_risk(score: float) -> Tuple[str, str]:
    """
    Classify composite score:
    0-20: Low risk
    21-40: Moderate risk
    41-60: High risk
    61-80: Very high risk
    81-100: Extreme risk
    """
    if score <= 20:
        return "LOW", "Annual screening, general exercise advice"
    elif score <= 40:
        return "MODERATE", "Targeted interventions: exercise program, medication review, vision check"
    elif score <= 60:
        return "HIGH", "Multifactorial intervention: PT/OT referral, home safety assessment, GC taper evaluation, assistive devices"
    elif score <= 80:
        return "VERY HIGH", "Urgent multidisciplinary falls prevention: comprehensive geriatric assessment, hip protectors, home modification, supervised exercise"
    else:
        return "EXTREME", "Immediate inpatient/supervised care consideration, full geriatric assessment, fall-prevention bundle"


def generate_recommendations(p: PatientProfile, subscores: Dict[str, float]) -> List[str]:
    """Generate specific, actionable clinical recommendations."""
    recs = []

    if subscores["tug"] >= 6:
        recs.append("TUG ≥20s: refer to physiotherapy for gait and balance training (Otago Exercise Program)")
    elif subscores["tug"] >= 3:
        recs.append("TUG 12-20s: consider structured exercise program (tai chi, yoga, or resistance training)")

    if subscores["prior_falls"] >= 4:
        recs.append(f"History of {p.falls_past_year} fall(s): implement multifactorial falls prevention per AGS/BGS guidelines")

    if subscores["polypharmacy"] >= 5:
        recs.append("High polypharmacy burden: pharmacist-led medication review, prioritize deprescribing CNS-active drugs")
    if p.uses_benzodiazepine:
        recs.append("⚠️ Benzodiazepine use: strongly recommend tapering/discontinuation (OR 1.5 for falls)")

    if subscores["glucocorticoid"] >= 5:
        recs.append(f"High GC exposure (prednisone {p.current_prednisone_mg}mg/d × {p.gc_duration_months}mo): evaluate steroid-sparing agents, check vitamin D/calcium, consider bone protection")
    elif p.current_prednisone_mg > 0:
        recs.append("Active GC use: ensure vitamin D ≥800 IU/d + calcium, monitor for proximal myopathy")

    if subscores["vision"] >= 5:
        recs.append("Significant visual impairment: ophthalmology referral, ensure adequate home lighting")
    if p.uses_bifocals:
        recs.append("Bifocal/progressive lenses: consider single-vision distance glasses for outdoor walking (Lord 2002)")

    if subscores["joint_involvement"] >= 5:
        recs.append("Significant lower extremity joint disease: OT assessment for assistive devices, consider joint injection or surgical referral if appropriate")
    if p.has_peripheral_neuropathy:
        recs.append("Peripheral neuropathy present: proprioceptive training, appropriate footwear, fall-risk education")

    if subscores["grip_strength"] >= 5:
        recs.append("Low grip strength (probable sarcopenia): resistance training program, protein intake ≥1.2g/kg/d, consider DEXA body composition")

    if subscores["balance_gait"] >= 6:
        recs.append("Poor balance/gait: supervised balance training, consider hip protectors for fracture prevention")

    if subscores["cognition"] >= 6:
        recs.append("Cognitive impairment: simplify medication regimen, supervised mobility, cognitive-motor dual-task training")

    if subscores["environment"] >= 5:
        recs.append("Home hazards identified: occupational therapy home safety assessment, remove loose rugs, install grab bars, improve lighting")
    if p.lives_alone:
        recs.append("Lives alone: consider personal emergency response system (medical alert), regular check-in schedule")

    # Disease-specific
    if p.diagnosis in ("RA", "OA") and p.disease_activity_score and p.disease_activity_score > 5.1:
        recs.append(f"High disease activity (DAS28={p.disease_activity_score}): optimize disease control — active inflammation increases fall risk via pain and disability")

    if not recs:
        recs.append("Low overall risk: continue annual falls screening, maintain physical activity")

    return recs


# ── Monte Carlo Uncertainty ─────────────────────────────────────────

def monte_carlo_falls_risk(
    p: PatientProfile,
    n_simulations: int = 5000,
    seed: Optional[int] = None
) -> Dict:
    """
    Run Monte Carlo simulation perturbing input parameters within
    clinically plausible measurement error ranges.

    Returns composite score with 95% CI and risk classification.
    """
    errors = _validate(p)
    if errors:
        return {"error": errors}

    rng = random.Random(seed)
    scores = []

    for _ in range(n_simulations):
        # Perturb inputs within measurement error
        sim = PatientProfile(
            age=p.age,
            sex=p.sex,
            weight_kg=p.weight_kg,
            height_cm=p.height_cm,
            tug_seconds=max(1, p.tug_seconds + rng.gauss(0, 1.2)),  # TUG ±1.2s test-retest
            falls_past_year=p.falls_past_year,  # integer, no perturbation
            fall_with_injury=p.fall_with_injury,
            total_medications=p.total_medications,
            cns_active_medications=p.cns_active_medications,
            uses_benzodiazepine=p.uses_benzodiazepine,
            current_prednisone_mg=max(0, p.current_prednisone_mg + rng.gauss(0, 0.5)),
            gc_duration_months=p.gc_duration_months,
            cumulative_gc_grams=max(0, p.cumulative_gc_grams + rng.gauss(0, 0.3)),
            visual_acuity_logmar=max(0, p.visual_acuity_logmar + rng.gauss(0, 0.05)),
            uses_bifocals=p.uses_bifocals,
            has_cataracts=p.has_cataracts,
            knee_severity=max(0, min(10, p.knee_severity + round(rng.gauss(0, 0.5)))),
            hip_severity=max(0, min(10, p.hip_severity + round(rng.gauss(0, 0.5)))),
            ankle_severity=max(0, min(10, p.ankle_severity + round(rng.gauss(0, 0.5)))),
            grip_strength_kg=max(0, p.grip_strength_kg + rng.gauss(0, 2.0)),  # ±2kg test-retest
            tinetti_score=max(0, min(28, p.tinetti_score + round(rng.gauss(0, 1.0)))) if p.tinetti_score is not None else None,
            uses_assistive_device=p.uses_assistive_device,
            mmse_score=max(0, min(30, p.mmse_score + round(rng.gauss(0, 1.0)))) if p.mmse_score is not None else None,
            moca_score=max(0, min(30, p.moca_score + round(rng.gauss(0, 1.0)))) if p.moca_score is not None else None,
            home_hazards=p.home_hazards,
            lives_alone=p.lives_alone,
            diagnosis=p.diagnosis,
            disease_activity_score=p.disease_activity_score,
            has_peripheral_neuropathy=p.has_peripheral_neuropathy,
            has_foot_deformity=p.has_foot_deformity,
        )
        ss = compute_subscores(sim)
        scores.append(compute_composite(ss))

    scores.sort()
    n = len(scores)
    mean_score = sum(scores) / n
    ci_low = scores[int(n * 0.025)]
    ci_high = scores[int(n * 0.975)]
    std_dev = (sum((s - mean_score) ** 2 for s in scores) / n) ** 0.5

    # Deterministic point estimate
    point_subscores = compute_subscores(p)
    point_score = compute_composite(point_subscores)
    risk_class, risk_action = classify_risk(point_score)
    recommendations = generate_recommendations(p, point_subscores)

    return {
        "patient_age": p.age,
        "patient_sex": p.sex,
        "diagnosis": p.diagnosis,
        "composite_score": point_score,
        "risk_classification": risk_class,
        "risk_action": risk_action,
        "subscores": point_subscores,
        "weighted_contributions": {k: round(WEIGHTS[k] * point_subscores[k] * 10, 2) for k in WEIGHTS},
        "monte_carlo": {
            "n_simulations": n_simulations,
            "mean": round(mean_score, 1),
            "std_dev": round(std_dev, 1),
            "ci_95_lower": round(ci_low, 1),
            "ci_95_upper": round(ci_high, 1),
        },
        "recommendations": recommendations,
    }


# ── Demo Scenarios ──────────────────────────────────────────────────

def demo():
    """Run 3 clinical scenarios."""

    print("=" * 80)
    print("FALLS-RHEUM: Falls Risk Prediction in Elderly Rheumatic Patients")
    print("Authors: Erick Adrián Zamora Tehozol, DNAI, Claw 🦞")
    print("RheumaAI / Frutero Club / DeSci")
    print("=" * 80)

    # Scenario 1: Low-risk — 62F RA, well-controlled, no falls history
    s1 = PatientProfile(
        age=62, sex="F", weight_kg=65, height_cm=160,
        tug_seconds=9.0,
        falls_past_year=0,
        total_medications=3, cns_active_medications=0,
        current_prednisone_mg=0, gc_duration_months=0,
        grip_strength_kg=22.0,
        tinetti_score=26,
        mmse_score=29,
        diagnosis="RA",
        disease_activity_score=2.8,
    )

    # Scenario 2: Moderate-risk — 72M RA, prednisone 10mg, 1 fall, mild vision loss
    s2 = PatientProfile(
        age=72, sex="M", weight_kg=78, height_cm=172,
        tug_seconds=14.0,
        falls_past_year=1, fall_with_injury=False,
        total_medications=7, cns_active_medications=1,
        current_prednisone_mg=10.0, gc_duration_months=18, cumulative_gc_grams=5.4,
        visual_acuity_logmar=0.3, uses_bifocals=True,
        knee_severity=5, hip_severity=3,
        grip_strength_kg=28.0,
        tinetti_score=22,
        mmse_score=27,
        home_hazards=2,
        diagnosis="RA",
        disease_activity_score=4.2,
    )

    # Scenario 3: Very high risk — 80F SLE, prednisone 15mg, 3 falls, neuropathy, lives alone
    s3 = PatientProfile(
        age=80, sex="F", weight_kg=55, height_cm=155,
        tug_seconds=25.0,
        falls_past_year=3, fall_with_injury=True,
        total_medications=11, cns_active_medications=3, uses_benzodiazepine=True,
        current_prednisone_mg=15.0, gc_duration_months=36, cumulative_gc_grams=16.2,
        visual_acuity_logmar=0.6, uses_bifocals=True, has_cataracts=True,
        knee_severity=7, hip_severity=6, ankle_severity=4,
        grip_strength_kg=12.0,
        tinetti_score=15,
        uses_assistive_device=True,
        mmse_score=22,
        home_hazards=5, lives_alone=True,
        diagnosis="SLE",
        has_peripheral_neuropathy=True,
        has_foot_deformity=True,
    )

    scenarios = [
        ("Scenario 1: 62F RA, well-controlled, no falls", s1),
        ("Scenario 2: 72M RA, prednisone 10mg, 1 fall, polypharmacy", s2),
        ("Scenario 3: 80F SLE, high-dose GC, 3 falls, neuropathy, lives alone", s3),
    ]

    for label, patient in scenarios:
        print(f"\n{'─' * 70}")
        print(f"  {label}")
        print(f"{'─' * 70}")
        result = monte_carlo_falls_risk(patient, n_simulations=5000, seed=42)

        if "error" in result:
            print(f"  ERROR: {result['error']}")
            continue

        print(f"  Diagnosis: {result['diagnosis']} | Age: {result['patient_age']} | Sex: {result['patient_sex']}")
        print(f"\n  COMPOSITE SCORE: {result['composite_score']}/100 — {result['risk_classification']}")
        mc = result['monte_carlo']
        print(f"  Monte Carlo (n={mc['n_simulations']}): mean={mc['mean']}, SD={mc['std_dev']}, 95% CI [{mc['ci_95_lower']}, {mc['ci_95_upper']}]")

        print(f"\n  Sub-scores (0-10):")
        for k, v in result['subscores'].items():
            w = result['weighted_contributions'][k]
            print(f"    {k:22s}: {v:5.1f}  (weighted contribution: {w:5.1f})")

        print(f"\n  Recommendations:")
        for r in result['recommendations']:
            print(f"    • {r}")

    print(f"\n{'=' * 80}")
    print("All scenarios completed successfully.")
    return True


if __name__ == "__main__":
    demo()

```