Proposed IoT device kit · research-grade, not diagnostic
PRĀṆA–PD Home Research Kit
A small set of connected devices that captures, at home and between clinic visits, the indicators the thesis names for each domain: gait and balance (Vyāna), resting respiration (Prāṇa), voice (Udāna), weight (Samāna), medication timing (context), plus diary entries (Apāna, Samāna, Udāna) through the app. It exists to answer H1 (feasibility) first: can these indicators be collected with acceptable completion, missingness and reliability?
Status: design only. No hardware has been built or tested. Everything below is a proposal. The software side (signed ingest, scoring, alerts) is implemented and tested against a simulated device. Part names are examples for planning, not endorsements; check availability, certification and licensing before buying. Whether the assembled kit counts as a medical device must be settled with the ethics committee and the regulator before any use with people.
Principles
- Prefer validated, off-the-shelf measurement devices (upper-arm cuff, finger oximeter, scale) over building sensors. Build only what does not exist: the gateway firmware, the pillbox, and the gait/voice feature extraction.
- Extract features on the device or phone; send indicators, not raw signals. Voice is processed on the device and discarded: only duration and level leave it. Motion is reduced to gait speed, turn time and a tremor-band ratio. This protects privacy and keeps the server simple.
- Every indicator carries a quality score. Low-quality readings are stored but excluded from scoring and counted in the data-quality strip.
- Guided tasks, not just passive wear. Gait speed, TUG, sustained vowel and resting respiration are collected during short prompted tasks with a known context, because passive free-living numbers are harder to interpret and to compare with clinic tests.
- Record medication state with every motor and voice reading (participant taps ON / OFF / not sure; the pillbox adds timing context). The kit never interprets it as advice.
- The kit is not an alarm. A fall flag or low oxygen reading goes to the review queue; nobody is watching in real time. Participants are told this in plain words and on the packaging.
- Nothing is scored until it is validated. Each indicator passes a pre-specified agreement and repeatability check (below) or stays as context.
Components and what they feed
| Component | Indicators sent | Domain | Build or buy | Main risks |
|---|---|---|---|---|
| Wearable band (wrist or ankle IMU, optional PPG) | gait speed from walking bouts, instrumented TUG, tremor-band ratio, steps, fall candidate; band-derived resting respiratory rate; nocturnal SpO2 only if a validated oximeter module is used | Vyāna; Prāṇa | Buy a research-grade IMU wearable where possible; custom firmware only if no vendor exposes raw data | Wrist-derived gait speed varies with arm swing and tremor; band respiratory rate is the weakest indicator; wrist SpO2 is not clinically interchangeable with a certified oximeter |
| Home station (Bluetooth hub with microphone, tablet or phone dock) | maximum phonation time, calibrated voice level; relays cuff readings | Udāna; Vyāna | Build around a BLE microcontroller or use a phone with a calibrated external microphone | Microphone calibration, background noise, distance from mouth; no audio kept |
| Upper-arm BP cuff (clinically validated, Bluetooth) | seated and standing systolic pressure → standing drop | Vyāna | Buy a cuff validated to the accepted protocol | Standing safety. Orthostatic drops can cause falls; the routine is only enabled by the clinician, with a chair or support at hand and, where advised, another person present |
| Smart scale | weight | Samāna | Buy | Clothing, time of day, fluid shifts; use trend, not a single value |
| Smart pillbox | time each scheduled compartment is opened; missed-dose flag | context only | Build (reed or Hall switch per compartment, BLE) | Opening is not swallowing; do not equate it with dose taken |
| Phone app | diary items, Sākṣī self-ratings, ON/OFF taps, falls, patient goal; also the Bluetooth gateway | Apāna, Samāna, Udāna, felt state | Build (PWA, later native for background Bluetooth) | Phone operating-system limits on background Bluetooth; digital literacy; caregiver assistance |
Prāṇa's olfaction test and any pulmonary function testing are clinic measures in the thesis and are not part of the kit. Apāna has no sensor on purpose: bowel and bladder data are diary-based, and the thesis itself notes that pelvic function is broader than any single variable.
Data flow
[sensor firmware] [phone / station gateway] [edge API] sample IMU, PPG, mic BLE scan + bond (encrypted) POST /api/platform/device/ingest extract features on device queue batches in local storage X-Signature: sha256=HMAC(secret, body) attach quality 0-1 add seq number, retry with back-off validate, de-duplicate (device, seq) discard raw buffers send when online; never edits values store readings, evaluate alert rules
Each batch is a small JSON object: { device_id, seq, fw, readings:[{ ts, metric, value, med_state?, quality? }] }. The sequence number makes retries safe; the per-device key makes forgery and tampering fail. The exact metric list, units and physiological ranges are served by /api/platform/catalog and are what the ingest endpoint enforces.
Guided weekly routine (about 10 minutes)
- Morning, seated rest, 2 minutes — band records resting respiratory rate, with quality score from movement.
- Sustained "aaah" into the station — maximum phonation time and level.
- Standing routine, only if enabled by the clinician — seated cuff reading, stand with support, readings at 1 and 3 minutes.
- Walk-and-turn on a marked short path — gait speed and instrumented Timed-Up-and-Go.
- Diary — bowel count, night-time voids, EAT-10, GI burden, ON/OFF, Sākṣī five ratings.
Daily: scale (alternate days is enough), pillbox, step count, nocturnal oximetry on the days the clinician chooses. Every step has a "skip — not today" button; missingness is data (H1), not failure.
Validation plan: what must be shown before an indicator is scored
| Indicator | Reference | Pre-specified checks (thresholds set in the protocol) |
|---|---|---|
| Gait speed, TUG | Timed 10-metre walk and TUG by a trained rater, same medication state | Agreement (Bland–Altman bias and limits), intraclass correlation, test–retest across two visits |
| Resting respiratory rate | Rater-counted breaths over a full minute | Bias and limits of agreement; fraction of usable recordings. If it fails, remove from Prāṇa scoring. |
| Nocturnal SpO2 | Certified fingertip oximeter during the same night | Mean difference; only a validated oximeter module is allowed to feed the score |
| Orthostatic BP drop | Clinic manual or validated automated protocol | Agreement on the drop, not just on the absolute values |
| Phonation time, voice level | Clinic stopwatch and calibrated sound-level meter | Agreement; effect of background noise; repeatability |
| Weight | Clinic scale | Bias; trend reliability |
| Whole kit | — | Completion, wear time, missingness per domain, participant-reported burden, battery and connection failures: the H1 feasibility set |
Heart-rate variability is deliberately left out of the kit's scored indicators. The thesis treats it as exploratory; a wrist PPG is not suitable for it, and if it is ever studied it needs a validated chest ECG strap and its own protocol.
Reference hardware sketch for the parts that must be built
Pillbox and home-station hub
- Low-power Bluetooth LE microcontroller, with secure storage for the device key.
- Pillbox: one magnetic switch per compartment, real-time clock with battery backup, coin or AAA cell, months of standby; LED and buzzer reminder that follows the clinician's schedule and never suggests a dose.
- Station: MEMS microphone with a known sensitivity, calibrated against a sound-level meter at 30 cm; vowel timing and level computed on the device; no audio stored or sent.
- Over-the-air firmware updates that are signed; version reported with every batch.
Wearable firmware
- 6-axis IMU at 50–100 Hz; walking-bout detection; gait speed estimated from a calibrated step-length model and checked against the timed walk; turn detection for TUG; 3–8 Hz band-power ratio for rest tremor.
- Quality score from wear detection, signal-to-noise and bout length.
- Fall candidate from impact followed by stillness, tuned for few false alarms. Expect false positives and misses; report both rates from the validation study.
- Rechargeable battery with a stated minimum runtime per charge; shows charge state to the participant.
Security and privacy by design
- Per-device secret key; every batch signed; unknown device and wrong signature are indistinguishable to the caller (implemented).
- Bluetooth bonding with encryption; a lost or stolen device is revoked from the console and its batches stop being accepted (implemented).
- No names or contact details on devices or in the research database; the device carries only its own ID and the study code.
- No raw audio or raw motion leaves the device by default. If a validation sub-study needs raw signals, that needs separate consent, a separate store and an expiry date.
- Device keys in the prototype are stored readably in the database. A real deployment must use secure elements on devices and encrypted key storage on the server.
- Data stays in the study's own storage; no analytics beacons in any participant-facing page (the Sākṣī app has none).
Build path
| Step | What | Gate |
|---|---|---|
| 0 · now | Software prototype and simulator on synthetic data | Done; review by a clinician and an engineer |
| 1 · bench | Pillbox and station prototypes; microphone calibration; band firmware on a bought IMU board; signed upload to the API | Reliable upload with phone offline for 48 h; battery life measured |
| 2 · healthy volunteers | Agreement and repeatability against the reference methods above, in people without Parkinson's, under ethics approval | Pre-specified agreement limits met, else indicator dropped or redesigned |
| 3 · Phase I (thesis) | Use alongside clinic measures in people with Parkinson's; feasibility and acceptability (H1, H2, H3) | Thesis stage gates |
| 4 · Phase II | 24-week cohort with kit and recorded disturbances, to support the coordination analysis (H4, H5) | Phase I measurement quality acceptable |
Limits to keep in view
- Consumer-grade sensors have error structure that differs from clinic tests; the indicators are proxies, and the validation steps decide whether any is usable.
- Home measurement adds burden. Older adults with tremor, cognitive change or limited phone use may not manage it; the design assumes caregiver assistance and low-tap flows and tests that assumption in H1.
- Passive and guided data are different things and are labelled differently in the export.
- Nothing in this kit measures Prāṇa as an energy or tests the Upaniṣadic framework. It measures biomedical indicators that the thesis proposes as candidate measures of five functional domains.