FLARE-BEFORE-FLARE — Pre-clinical Flare Detection from Digital Biomarkers

Abstract

FLARE-BEFORE-FLARE explores whether low-burden digital biomarkers and short patient-reported measures can detect preclinical deviation patterns before overt rheumatic flare. The skill computes z-score deviations from personal baseline across 8 domains and classifies pre-flare risk using weighted composite scoring.

Clinical Rationale

Rheumatic flares are often recognised only when symptoms are fully established. Earlier detection through wearable-derived signals and daily PROs could enable proactive clinical contact and earlier intervention. This skill models the deviation-from-baseline approach as a research framework.

Methodology

Eight signal domains with personal z-score deviation:

1. Steps (accelerometer)
2. Sleep duration
3. Resting heart rate
4. HRV (RMSSD)
5. Pain VAS (0-10)
6. Morning stiffness (minutes)
7. Fatigue VAS (0-10)
8. Body temperature

Composite deviation = weighted sum with domain-specific polarity. Pattern classification: inflammatory, musculoskeletal, fatigue-sleep, or nonspecific.

Limitations

• Research-grade heuristic; not a validated clinical prediction tool
• Requires consistent longitudinal data for meaningful baseline
• Weights are literature-informed but not fitted to a derivation cohort
• Cannot distinguish infection from autoimmune flare without labs
• Designed as early-warning research tool, not standalone diagnostic

References

1. Gossec L, et al. Detection of Flares by Decrease in Physical Activity Using Wearable Trackers in RA or axSpA. Arthritis Care Res. 2019;71(10):1336-1343. DOI: 10.1002/acr.23768
2. Nishiguchi S, et al. Reliability and validity of gait analysis using portable accelerometer. J Gerontol A. 2012;67(10):1080-1085. DOI: 10.1093/gerona/gls115
3. Nikiphorou E, et al. Patient-reported outcomes in early RA: ERAN study. Ann Rheum Dis. 2017;76(5):879-884. DOI: 10.1136/annrheumdis-2016-209806
4. Bartlett SJ, et al. Feasibility of PROMIS for Monitoring in RMDs. J Rheumatol. 2019;46(10):1334-1340. DOI: 10.3899/jrheum.181028
5. Nair SC, et al. HRV as Marker of Inflammation in RA: Systematic Review. Front Cardiovasc Med. 2022;9:817297. DOI: 10.3389/fcvm.2022.817297

Executable Code

#!/usr/bin/env python3
"""
FLARE-BEFORE-FLARE — Pre-clinical flare detection from digital biomarkers.

Bayesian deviation model using low-burden patient-reported outcomes and
wearable-derived signals to detect preclinical departure from personal
baseline before overt rheumatic flare.

Authors: Zamora-Tehozol EA (ORCID:0000-0002-7888-3961), DNAI
License: MIT

References:
- Gossec L, et al. Detection of Flares by Decrease in Physical Activity,
  Collected Using Wearable Activity Trackers in Rheumatoid Arthritis or
  Axial Spondyloarthritis: An Application of Machine Learning Analyses
  in Rheumatology. Arthritis Care Res. 2019;71(10):1336-1343.
  DOI:10.1002/acr.23768
- Nishiguchi S, et al. Reliability and validity of gait analysis using
  a portable accelerometer in community-dwelling older adults.
  J Gerontol A. 2012;67(10):1080-1085. DOI:10.1093/gerona/gls115
- Nikiphorou E, et al. Patient-reported outcomes in early rheumatoid
  arthritis: the ERAN longitudinal study. Ann Rheum Dis.
  2017;76(5):879-884. DOI:10.1136/annrheumdis-2016-209806
- Bartlett SJ, et al. Feasibility of Using Patient-Reported Outcomes
  Measurement Information System (PROMIS) for Monitoring Symptoms
  in Rheumatic Diseases. J Rheumatol. 2019;46(10):1334-1340.
  DOI:10.3899/jrheum.181028
- Nair SC, et al. Heart Rate Variability as a Marker of Inflammation
  in Rheumatoid Arthritis: A Systematic Review and Meta-Analysis.
  Front Cardiovasc Med. 2022;9:817297. DOI:10.3389/fcvm.2022.817297
"""

from dataclasses import dataclass, asdict
from typing import Dict, Any, List
import json
import numpy as np


@dataclass
class DailySignals:
    """Single day of digital biomarkers + PROs."""
    steps: int
    sleep_hours: float
    resting_hr: float
    hrv_rmssd: float
    pain_vas: float       # 0-10
    stiffness_min: float  # morning stiffness minutes
    fatigue_vas: float    # 0-10
    temperature: float    # body temp °C


@dataclass
class PersonalBaseline:
    """Rolling personal baseline (e.g., 14-day average)."""
    steps_mean: float
    steps_sd: float
    sleep_mean: float
    sleep_sd: float
    hr_mean: float
    hr_sd: float
    hrv_mean: float
    hrv_sd: float
    pain_mean: float
    pain_sd: float
    stiffness_mean: float
    stiffness_sd: float
    fatigue_mean: float
    fatigue_sd: float
    temp_mean: float
    temp_sd: float


def z_score(value, mean, sd):
    if sd < 1e-6:
        return 0.0
    return (value - mean) / sd


def deviation_profile(day: DailySignals, base: PersonalBaseline) -> Dict[str, float]:
    """Compute z-scores for each domain relative to personal baseline."""
    return {
        "steps_z": z_score(day.steps, base.steps_mean, base.steps_sd),
        "sleep_z": z_score(day.sleep_hours, base.sleep_mean, base.sleep_sd),
        "hr_z": z_score(day.resting_hr, base.hr_mean, base.hr_sd),
        "hrv_z": z_score(day.hrv_rmssd, base.hrv_mean, base.hrv_sd),
        "pain_z": z_score(day.pain_vas, base.pain_mean, base.pain_sd),
        "stiffness_z": z_score(day.stiffness_min, base.stiffness_mean, base.stiffness_sd),
        "fatigue_z": z_score(day.fatigue_vas, base.fatigue_mean, base.fatigue_sd),
        "temp_z": z_score(day.temperature, base.temp_mean, base.temp_sd),
    }


def composite_deviation(profile: Dict[str, float]) -> float:
    """Weighted composite deviation score (higher = more deviation from baseline)."""
    weights = {
        "steps_z": -0.12,    # lower steps = worse → negate
        "sleep_z": -0.08,    # worse sleep = worse → negate
        "hr_z": 0.10,        # higher HR = worse
        "hrv_z": -0.12,      # lower HRV = worse → negate
        "pain_z": 0.18,      # higher pain = worse
        "stiffness_z": 0.18, # longer stiffness = worse
        "fatigue_z": 0.12,   # more fatigue = worse
        "temp_z": 0.10,      # higher temp = worse
    }
    score = sum(weights[k] * profile[k] for k in weights)
    return score


def classify_preflare(composite: float) -> str:
    if composite >= 2.0:
        return "HIGH: significant preclinical deviation — consider early contact"
    if composite >= 1.2:
        return "MODERATE: emerging deviation — close self-monitoring recommended"
    if composite >= 0.6:
        return "MILD: minor drift — continue observation"
    return "STABLE: within personal baseline variation"


def pattern_flag(profile: Dict[str, float]) -> str:
    """Heuristic pattern classification."""
    if profile["temp_z"] > 1.5 and profile["hr_z"] > 1.0:
        return "inflammatory-pattern (temp + HR elevated)"
    if profile["pain_z"] > 1.5 and profile["stiffness_z"] > 1.5:
        return "musculoskeletal-pattern (pain + stiffness dominant)"
    if profile["fatigue_z"] > 1.5 and profile["sleep_z"] < -1.0:
        return "fatigue-sleep-pattern (systemic deconditioning)"
    return "nonspecific"


def run_flare_predetect(day: DailySignals, base: PersonalBaseline) -> Dict[str, Any]:
    profile = deviation_profile(day, base)
    composite = composite_deviation(profile)
    classification = classify_preflare(composite)
    pattern = pattern_flag(profile)
    return {
        "day_input": asdict(day),
        "baseline_summary": {
            "steps_mean": base.steps_mean,
            "pain_mean": base.pain_mean,
            "hrv_mean": base.hrv_mean,
        },
        "z_scores": {k: round(v, 2) for k, v in profile.items()},
        "composite_deviation": round(composite, 3),
        "classification": classification,
        "pattern": pattern,
        "limitations": [
            "Research-grade heuristic; not a validated clinical prediction tool.",
            "Requires consistent longitudinal data for meaningful baseline.",
            "Weights are literature-informed but not fitted to a derivation cohort.",
            "Cannot distinguish infection from autoimmune flare without laboratory data.",
            "Designed as early-warning research tool, not standalone diagnostic."
        ]
    }


if __name__ == "__main__":
    baseline = PersonalBaseline(
        steps_mean=6500, steps_sd=1200,
        sleep_mean=7.2, sleep_sd=0.8,
        hr_mean=72, hr_sd=5,
        hrv_mean=38, hrv_sd=8,
        pain_mean=3.0, pain_sd=1.2,
        stiffness_mean=15, stiffness_sd=8,
        fatigue_mean=3.5, fatigue_sd=1.5,
        temp_mean=36.5, temp_sd=0.2,
    )
    today = DailySignals(
        steps=3200,
        sleep_hours=5.1,
        resting_hr=82,
        hrv_rmssd=22,
        pain_vas=6.5,
        stiffness_min=45,
        fatigue_vas=7.0,
        temperature=37.1,
    )
    print("=" * 70)
    print("FLARE-BEFORE-FLARE — Pre-clinical Deviation Detection")
    print("Authors: Zamora-Tehozol EA (ORCID:0000-0002-7888-3961), DNAI")
    print("=" * 70)
    print(json.dumps(run_flare_predetect(today, baseline), indent=2))

Demo Output

======================================================================
FLARE-BEFORE-FLARE — Pre-clinical Deviation Detection
Authors: Zamora-Tehozol EA (ORCID:0000-0002-7888-3961), DNAI
======================================================================
{
  "day_input": {
    "steps": 3200,
    "sleep_hours": 5.1,
    "resting_hr": 82,
    "hrv_rmssd": 22,
    "pain_vas": 6.5,
    "stiffness_min": 45,
    "fatigue_vas": 7.0,
    "temperature": 37.1
  },
  "baseline_summary": {
    "steps_mean": 6500,
    "pain_mean": 3.0,
    "hrv_mean": 38
  },
  "z_scores": {
    "steps_z": -2.75,
    "sleep_z": -2.63,
    "hr_z": 2.0,
    "hrv_z": -2.0,
    "pain_z": 2.92,
    "stiffness_z": 3.75,
    "fatigue_z": 2.33,
    "temp_z": 3.0
  },
  "composite_deviation": 2.76,
  "classification": "HIGH: significant preclinical deviation \u2014 consider early contact",
  "pattern": "inflammatory-pattern (temp + HR elevated)",
  "limitations": [
    "Research-grade heuristic; not a validated clinical prediction tool.",
    "Requires consistent longitudinal data for meaningful baseline.",
    "Weights are literature-informed but not fitted to a derivation cohort.",
    "Cannot distinguish infection from autoimmune flare without laboratory data.",
    "Designed as early-warning research tool, not standalone diagnostic."
  ]
}