Axon Labs
All insights

From the engineering desk

Insight13 min read20 July 2026

Wearable Health Device Development: Feasibility to Certification

Classification, BOM, antennas, batteries and false positives — the engineering decisions that decide whether a health wearable ships.

By Axon Labs Engineering

A health wearable is the hardest class of connected product: it lives on a body, runs on a battery, senses continuously, and answers to regulators. In 2026 the wearable medical device market crosses $68 billion on its way to a projected $330 billion by 2033 — and most of the products chasing that market will fail on the same five constraints. We’ve shipped through all five. This is what the path actually looks like.

Key takeaways

  • The wearable medical device market grows from $68.1B in 2026 to a projected $330.5B by 2033, a 29.5% CAGR (Grand View Research) — home healthcare is already the largest application at 51.8% of revenue.
  • The classification decision — wellness product vs. medical device — shapes architecture, evidence, timeline and budget more than any technical choice. Make it first, deliberately.
  • Field performance, not lab accuracy, is the product: fall-detection systems hit 97.9% sensitivity in validation, but false positives in real homes are what determine whether users keep the device on.

What does it take to develop a wearable health device?

Wearable health development is standard hardware development — the same POC → EVT → DVT → PVT gates from our complete development guide — compressed into the most hostile envelope in consumer electronics. Everything fights everything: sensing wants power, the battery wants space, the body detunes the radio, and the price point caps the BOM.
The demand side is no longer speculative. Home healthcare is the largest application segment at 51.8% of 2025 revenue (Grand View Research), and roughly 69% of healthcare providers adopted wearable monitoring for continuous patient tracking between 2023 and 2025, per Market.us wearable statistics. The market has moved from counting steps to managing patients. The engineering bar moved with it.

Medical device or wellness product?

This is the first decision of the program, and it isn’t a technical one — it’s a claims decision. Say your device “promotes healthy activity” and it’s a general wellness product under the FDA’s General Wellness policy. Say it “detects falls in at-risk patients” and you’ve made a medical claim, with everything that follows. Notably, 76.4% of wearable medical device revenue in 2025 came from consumer-grade devices (Grand View Research) — most of the market lives deliberately on the wellness side of the line.
Wellness productMedical device
ClaimsGeneral health, fitness, awarenessDiagnose, treat, prevent, alert
EvidenceInternal validationClinical data, documented verification
Quality systemGood engineering practiceISO 13485 QMS, design controls, IEC 62304 software
Timeline impactBaselineAdd months to years, by class
Can you upgrade later?Yes — if the architecture anticipated itDowngrading claims is easier than retrofitting rigor
The strategic play we recommend: choose the claim tier deliberately, then engineer one tier stricter than you claim. Shipping a wellness product with medical-grade data integrity, traceability, and verification evidence costs modestly more now — and preserves the option of regulated claims later without a re-architecture. It’s the cheapest optionality in the program.

The five constraints every health wearable fights

When we engineered a connected smart insole for fall risk, we wrote five constraints on the wall on day one: a tiny form factor, a sub-$50 BOM, hybrid AI, always-on sensing, and certification readiness. Every wearable program we’ve seen since fights the same five — usually in a different order, always at the same time.
  1. Form factor. The enclosure is set by the body, not the electronics. Everything else negotiates for the space that remains.
  2. BOM ceiling. Health wearables sell against a price expectation. The ceiling is a day-one architecture input — see why BOM cost is designed in, not negotiated out.
  3. Intelligence placement. Which decisions happen on-device and which in the cloud — the hybrid architecture question — drives silicon, power, and connectivity all at once.
  4. Always-on sensing. Continuous monitoring on a battery means duty-cycled cascades, not brute force.
  5. Certification readiness. Wireless compliance, skin-contact biocompatibility (ISO 10993), and the claims decision above — planned at architecture, not discovered at DVT.

Engineering a wearable to a sub-$50 BOM

A sub-$50 bill of materials for an always-connected, always-sensing health device sounds implausible until you watch where the dollars actually go. Silicon, sensors, radio, battery, and enclosure each claim their share — and each has an architecture-level lever that purchasing can never reach.
  • Right-size the silicon to the duty cycle. A modest MCU running a compressed model beats a premium SoC idling at 1% utilization. On the insole, hybrid AI — light inference on-device, analytics in the cloud — is what kept the compute bill small.
  • Fuse sensors instead of adding them. Pressure plus IMU, fused in software, replaced additional sensing hardware. The model is NRE; a sensor is a per-unit cost, forever.
  • Co-design antenna, enclosure, and board. Separately optimized parts don’t compose in a wearable envelope. Co-design avoided the respin spiral that quietly doubles BOMs.
  • Make DFM a day-one constraint. Part count, standard fasteners, mold-friendly geometry — decided before the first prototype, per the prototype-to-manufacturing playbook.
The result on the insole program: by engineering validation, the BOM conversation was about cents, because the dollars were settled at architecture. Read the full case study for the system-level walkthrough.

Antenna design in impossible form factors

The human body is, electromagnetically speaking, a bag of salt water pressed against your radio. It absorbs RF, detunes resonance, and shifts with every movement. Now shrink the available volume to a shoe insole, add a ground plane crowded by sensors, and hold a cellular link budget. This is routinely the hardest physics problem in a health wearable.
  • Body-load from day one. Simulate and measure against tissue phantoms early — free-space antenna performance is fiction for wearables.
  • Buy the antenna its keep-out first. Reserve the volume and ground-plane clearance before the battery and sensors negotiate; retrofitting clearance means moving everything.
  • Co-design, don’t integrate. Antenna, enclosure material, and board stack-up are one electromagnetic system. CMF choices — metallization, coatings — are RF decisions wearing design clothes.
  • Budget certification headroom. Intentional-radiator testing runs $9,000–$15,000 per attempt (Compliance Testing, 2026); a detuned antenna found at the test lab is a board spin plus a re-queue.

Always-on sensing on a wearable battery

Continuous health monitoring and multi-day battery life sound contradictory. They aren’t — but only if the power budget is architected as a cascade. The always-on tier is a microwatt-class trigger watching cheap features; the expensive tiers — full inference, the radio — wake only on candidate events. “Always sensing” never has to mean “always computing.”
  • Budget by duty cycle, not peak draw. The spreadsheet that matters multiplies each stage’s power by its fraction of the day. Build it before choosing silicon — MCU-class accelerators now run trigger models at under a watt (Promwad, 2026), and far less duty-cycled.
  • Count memory and radio, not just compute. Transmitting a raw sensor stream costs more than inferring on it locally and sending events — on-device AI is a battery strategy as much as a latency one.
  • Design charging behavior, not just capacity. A health device charging nightly monitors nothing overnight. Charge cadence is a clinical-coverage decision disguised as an electrical one.
The deeper treatment of duty-cycled inference and silicon classes is in our edge AI development guide — for wearables, all of it applies with the thermal ceiling of skin contact on top.

BLE, LTE-M, or Wi-Fi: how should a health wearable connect?

The connectivity choice is really a dependence choice: may the product assume a paired phone? For fitness, usually yes — BLE to the user’s phone is cheap, low-power, and good enough. For safety-critical health monitoring of older adults, the assumption collapses: the phone is in another room, unpaired, or doesn’t exist. Standalone products need their own path to the network.
BLE (phone-tethered)LTE-M / NB-IoT (standalone)Wi-Fi (fixed-location)
PowerLowestModerate, manageable with eDRX/PSMHigh for battery wearables
CoverageWherever the phone isCarrier network, roams with the userOne building
BOM & recurring costCheapest, no subscriptionModem + antenna + carrier planLow BOM, no plan
Fails when…Phone absent, unpaired, deadCoverage gaps, plan lapsesUser leaves home
FitsFitness, phone-native usersSafety alerts, senior care, clinical monitoringBedside and home-fixture devices
The insole runs LTE-M precisely because its users are the people least likely to carry a paired smartphone everywhere — a fall alert that requires a phone within ten meters isn’t a safety feature, it’s a demo. The cost is real: a cellular modem, an antenna problem (see above), and a carrier relationship. The feasibility question is whether the product’s core promise survives the phone being absent. If it doesn’t, budget for standalone connectivity from day one.

Fall detection AI and the false-positive problem

The clinical need is enormous: over 14 million US adults 65 and older — 1 in 4 — report falling each year, per the CDC’s falls data, driving nearly 3 million emergency visits and over 38,000 deaths annually, at a cost estimated above $60 billion in 2026.
Detection accuracy looks solved on paper. A 2025 wearable system in Sensors (MDPI) reported 97.9% sensitivity and 99.9% specificity in validation. But run the arithmetic that lab papers don’t: a device sampling continuously evaluates millions of movement windows per year. At 99.9% specificity, that’s a false alarm every few days — and real-world studies of fall detectors consistently find field performance below lab performance, with false positives driving abandonment. Users who get paged for sitting down hard stop wearing the device. Then it detects nothing.
  • Engineer specificity, not just sensitivity. The false-alarm budget — alarms per user-month — is a top-level spec with a number, not an aspiration.
  • Fuse modalities the impostors can’t fake. An IMU alone confuses sitting hard with falling. Pressure plus inertial data — gait context before the event — separates them. This is the sensor-fusion-over-redundancy play from our edge AI guide, applied clinically.
  • Escalate, don’t alarm. On-device inference flags a candidate; cloud analytics weigh it against the user’s baseline; unanswered check-ins escalate. Hybrid architecture is a false-positive filter.
  • Validate on field data at the gates. Lab datasets are the hypothesis. Free-living data from instrumented pilots — worn wrong, half-charged, on real gaits — is the evidence that passes EVT.

Certification-ready from day one

Certification isn’t a phase at the end; it’s a set of constraints that either shaped the architecture or will reshape it expensively. A health wearable stacks several regimes at once: radio compliance (FCC/CE, $3,000–$15,000 per attempt), skin-contact biocompatibility (ISO 10993), the medical claims regime chosen above — and, from September 2026, the EU Cyber Resilience Act’s mandatory vulnerability reporting for every connected product.
Health data raises the security floor further: signed updates, encrypted data at rest and in transit, and a hardware root of trust are table stakes when the payload is someone’s gait, heart rate, or fall history — the full argument is in Secure by Design for Connected Hardware. Our practice: write the certification plan in the feasibility phase, with named standards, test-lab budgets, and the evidence each gate must produce. On the insole, certification readiness was constraint number five — which is why it never became crisis number one.
A wearable that isn’t worn detects nothing. Comfort is a clinical feature.

The bottom line

  • Make the classification decision first and deliberately — then engineer one tier stricter than you claim. It’s the cheapest optionality in the program.
  • The five constraints — form factor, BOM, intelligence placement, always-on power, certification — must be solved simultaneously. Feasibility exists to find the one that breaks.
  • Field performance is the product: false-alarm budgets, body-loaded antennas, and free-living validation data decide adoption, not lab metrics.
  • Standalone safety products need standalone connectivity. If the promise dies when the phone is absent, so does the product.
We’ve carried a health wearable from five constraints on a wall to a production file set and an IP position. If yours is next, start with a discovery & feasibility phase — the five-constraint collision is exactly what it exists to resolve.

Frequently asked questions

How much does it cost to develop a wearable health device?

Wearables sit at the demanding end of connected-device budgets: typically $125K–$250K for a wellness-grade product and $250K–$400K+ once medical claims, clinical evidence, and certification enter (Calcix, 2026 ranges). Miniaturization, antenna work, and always-on power budgets are what push wearables above generic IoT costs.

Is my wearable a medical device?

Your claims decide, not your sensors. Devices promoting general wellness — activity, sleep awareness, fitness — fall under the FDA’s General Wellness policy for low-risk products. Claiming to diagnose, treat, prevent, or alert on a condition makes it a medical device, adding clinical evidence, a quality system (ISO 13485), and regulatory review.

How accurate is wearable fall detection?

Lab results look excellent — a 2025 system in Sensors (MDPI) reported 97.9% sensitivity and 99.9% specificity. Real-world performance runs lower: continuous sensing evaluates millions of windows, so even 99.9% specificity produces regular false alarms, and field studies consistently show false positives driving users to stop wearing devices.

What connectivity should a health wearable use?

Phone-tethered BLE is the default for fitness products: cheapest, lowest power. Safety-critical monitoring should assume the phone is absent — standalone LTE-M or NB-IoT keeps alerts working independently, at the cost of a modem, a harder antenna problem, and a carrier plan. Match the radio to what the product promises.

How do wearables achieve multi-day battery life with continuous sensing?

Cascaded duty cycling: a microwatt-class always-on trigger watches cheap signal features, waking full inference and the radio only on candidate events. MCU-class accelerators run such trigger models under 1 W peak (Promwad, 2026) and far less averaged. Sending inferred events instead of raw streams saves more again.

Sources

If the product has to ship, talk to the team that builds for that outcome.

Senior engineer on the first call. NDA before technical detail. References available under NDA after qualification. Or start with a fixed-fee feasibility study.

Request a conversationFeasibility study