Proof of Concept · Real Vitals + PhysioNet Open Data · API v1.3

Real physiological data.
Real API output.

We ran PhysioSim's API against two data types: real daily vitals from actual individuals (smartwatch-style readings), and a high-frequency ECG/waveform dataset from PhysioNet — 242,815 observations. Different formats, same engine. Below is the actual output.

242,815
observations processed
4
physiological signals
v1.3
proprietary algorithm + tiered routing
<3s
API response time
This proof of concept uses de-identified open-source data from PhysioNet (MIMIC). No real patient data. Not diagnostic. All outputs are illustrative of API capability only. PhysioSim does not diagnose and does not make clinical decisions.
// Dataset A — Real daily vitals · Two individuals

PhysioSim on real smartwatch readings

PhysioSim was run against actual daily vital sign readings collected from two real individuals — the kind of data a smartwatch or consumer health device produces. Two people, two datasets, different signals. Same API, same individualized baseline engine.

⚠ Data collected with consent for development purposes. De-identified. For research and pilot use only.

Dataset — CA148D39
Individual A · 4 signals · 23 observations
MEDIUM Risk level

API output — rationale

Repeated or multi-signal deviation detected across 3 concurrent signals, 4 multi-signal deviation days. Trend acceleration flagged in systolic and diastolic BP. SpO₂ deviation detected — monitoring for rising oxygen requirement indicated. Pattern warrants monitoring; routed to Tier 2 remote monitoring command center within 24h.

Observations
23
total analyzed
Deviations
9
39.1% of readings
Signals
4
HR · SpO₂ · SBP · DBP
Routing
Tier 2
within 24h
Signal Individual Median IQR Mean Std Dev Triggers Accelerating?
Resting Heart Rate 70.0 bpm 6.0 bpm 72.6 bpm 4.5 bpm 1 No
SpO₂ 98.0% 1.5% 97.6% 1.5% 3 No
Systolic BP 123.0 mmHg 6.5 mmHg 119.3 mmHg 7.2 mmHg 4 YES ↑
Diastolic BP 84.0 mmHg 6.0 mmHg 80.0 mmHg 6.9 mmHg 6 YES ↑
Signal trends vs. individual baseline — Dataset CA148D39
Resting Heart Rate (bpm) — baseline 70.0
SpO₂ (%) — baseline 98.0 · 3 deviations
Systolic BP (mmHg) — baseline 123.0 · 4 deviations · accelerating ↑
Diastolic BP (mmHg) — baseline 84.0 · 6 deviations · accelerating ↑

Green line = observed · Dashed = individual baseline median · Shaded = IQR · Amber/red dots = flagged deviations

Dataset — 97DBAAA9
Individual B · 1 signal · 25 observations
MEDIUM Risk level

API output — rationale

Repeated deviation pattern detected in resting heart rate despite low trigger count. Trend acceleration flagged — rate of change in HR is increasing. Tachycardia is nonspecific; HR elevation may reflect pain, anxiety, fever, dehydration, or stress. Routed to Tier 2 remote monitoring command center for review within 24h.

Observations
25
total analyzed
Deviations
1
4.0% of readings
Signal
1
Resting HR
Routing
Tier 2
within 24h
Signal Individual Median IQR Mean Std Dev Triggers Accelerating?
Resting Heart Rate 73.0 bpm 3.5 bpm 73.0 bpm 4.6 bpm 1 YES ↑
Signal trend vs. individual baseline — Dataset 97DBAAA9
HR (bpm) — baseline 73.0 · acceleration detected ↑

Green line = observed · Dashed = individual baseline (73.0 bpm) · Shaded = IQR · Amber dot = flagged deviation

What these datasets demonstrate

Two real individuals. Two different signal sets. The same individualized baseline engine assessed each against their own physiological normal — not a population average. Sparse daily readings, real deviation patterns, real alert routing. Below, the same engine processes 242,815 high-frequency waveform observations.

// Dataset B — PhysioNet ECG waveform · Step 1 of 5 — Input data

High-frequency waveform — what we fed the API

Data source
PhysioNet — MIMIC Waveform Database
Open-source, de-identified ICU physiological waveform data. Freely available at physionet.org. No credentials required for demo dataset.
Format
CSV — 242,815 rows × 4 signal columns
Signals: ecg, pleth, temperature, accelerometer
High-frequency waveform data sampled at 125Hz. Signal type and data mode auto-detected by the API.
API call
POST /v1/patients/analyze — uploaded via multipart/form-data
No pre-processing required. PhysioSim auto-detected signal columns, data mode, and baseline window.
Processing time
< 3 seconds for 242,815 observations including proprietary frequency analysis across all 4 signals.
// Dataset B · Step 2 — Individualized baseline

What PhysioSim learned about this individual

Before flagging anything, PhysioSim builds a statistical baseline from an initial longitudinal window of this individual's own data — specific to them. This is what makes deviation detection meaningful: deviation from their normal, not a population average.

Signal Median (individual) IQR Mean Std Dev Observations Unit
ECG 31,849 3,900 32,624 5,839 242,815 ADC units*
Plethysmogram 64,766 558 64,790 400 242,815 ADC units
Temperature 32.94 0.38 33.00 0.24 242,815 °C
Accelerometer 4.57 4.81 5.17 3.03 242,815 g

* PhysioNet MIMIC ECG values are raw ADC (analog-to-digital converter) units, not millivolts. To convert to mV, divide by the gain factor (typically ~200 ADC/mV). PhysioSim correctly labels these as ADC units in v1.3+.

// Dataset B · Step 3 — Signal trend vs baseline

Temperature signal — observed vs baseline

The chart below shows actual temperature values from the dataset versus the individualized baseline median (dashed) and IQR band (shaded). The distinctive U-shaped pattern shows the patient's temperature dipping during the mid-recording period and recovering — a real longitudinal physiological signal, not synthetic data.

Temperature (°C) — 242,815 observations · Individual baseline overlaid
33.75°C 33.00 (baseline) 32.50°C 0 50,000 100,000 150,000 200,000+

Green line = observed values · Dashed = individual baseline median · Shaded = baseline IQR · Reproduced from actual API output

// Dataset B · Step 4 — Proprietary frequency analysis

Frequency-domain analysis — what the proprietary algorithm found

PhysioSim applies a proprietary frequency-domain algorithm to detect when signal rhythms shift — not just whether they did. This provides clinically precise time-localized detection for non-stationary physiological waveforms.

Signal Baseline Pattern Recent Pattern Pattern Shift Spectral Score Deviated? Clinical Significance
ECG Stable Shifted Significant ↑ Elevated YES Meaningful shift in cardiac rhythm pattern
Plethysmogram Stable Reduced Moderate ↓ Reduced YES Energy reduction in PPG waveform pattern
Temperature Stable Shifted Moderate Reduced YES Spectral pattern shift vs baseline window
Accelerometer Stable Minimal shift Minimal Reduced YES Minimal pattern shift, spectral change detected

Note: PhysioSim's proprietary algorithm flags both increases and decreases in spectral pattern energy as warranting clinical review — reduced activity patterns (e.g., sedation, sleep, reduced movement) are as clinically meaningful as elevated ones.

// Dataset B · Step 5 — Risk stratification output

What the API returned

HIGH Risk level

Rationale (actual API output)

Sustained multi-signal deviation detected across 4 signals. Pattern is persistent versus individual baseline and consistent across signals — not consistent with short-term noise. Proprietary frequency-domain analysis additionally flagged spectral pattern shifts in ECG, plethysmogram, temperature, and accelerometer. Trend velocity analysis shows accelerating deviation trajectory.

ACTIVATED
Tier 1

EHR / Epic Direct

High-confidence alert routes directly into clinical workflow. Nurse review within 4–8h. Do not route through intermediate layers — bedside team awareness required.

Tier 2

Monitoring Center

Remote monitoring command center. Escalate to EHR if pattern worsens over 24h.

Tier 3

Routine Queue

Routine monitoring queue. No immediate outreach required. Re-evaluate if signals continue trending.

// What this demonstrates

What this proof of concept actually proves

⚙️

The API works

242,815 observations processed, baselined, analyzed with PhysioSim's proprietary algorithm, and risk-stratified in under 3 seconds. The engineering is real and deployed on Railway.

Individualized baselines work

The baseline is built from this individual's own longitudinal data — not a population average. Every deviation threshold is calibrated to their specific physiological normal range.

Proprietary frequency analysis works

PhysioSim's proprietary algorithm detected spectral pattern shifts across all 4 signals including ECG and PPG waveforms — with time-localized precision that standard frequency methods cannot provide.

🔀

Tiered routing works

HIGH risk correctly routed to Tier 1 (EHR/Epic direct) — bypassing remote monitoring intermediate layers for high-confidence sustained multi-signal alerts.

📄

PDF report output works

The API generated a clinician-readable PDF report with baseline tables, trend charts, frequency analysis, and flagged events — downloadable, no dashboard required.

Zero hardware overhead

Input: a CSV file. Output: structured JSON + PDF report. No device required. No proprietary format. Any data source that can export CSV works with PhysioSim.

Important context: This waveform dataset (125Hz ICU data) represents the most technically demanding input PhysioSim handles. The same engine works equally well on sparse daily vitals — as shown above with the smartwatch datasets. The individualized baseline approach applies regardless of data density or signal type.

Interested in what PhysioSim finds
in your patient population?

Request pilot access → De-identified retrospective data sufficient.
4–8 week evaluation. No dashboard required.