PATIENT SAFETY
FALL DETECTION
MONITORING
Clinically validated wearable fall detection, sub-8-second wandering alerts, and FHIR-integrated nurse call for hospitals, memory care, and home telecare.
The Engineering Problem
Patient Safety Hardware Fails When It Matters Most
Fall detection hardware fails in one of two modes: sensitivity so high that nursing staff become desensitised to alarms, or specificity so conservative that genuine falls go undetected. Most off-the-shelf pendants use fixed-threshold accelerometry tuned in laboratory conditions — a single threshold cannot capture the distinct fall kinematics of post-stroke, Parkinson's, and hip-replacement cohorts.
Wandering prevention demands tight alert latency — a 15-second door delay is unacceptable at a fire exit during shift handover. BLE-only systems face multipath fading in dense masonry buildings; UWB time-difference-of-arrival resolves sub-metre ambiguity but requires anchor placement and infrastructure investment that most care home operators have not budgeted for.
CQC registration and MHRA MDR 2002 classification are baseline requirements, not differentiators. The differentiators are FHIR R4 AlarmEvent EPR integration, EN 60849 nurse call compliance, BS 8521-2 telecare conformance, and documented MCA best-interests evidence — the combination that determines whether a device gains clinical adoption or sits unused.
Wearable Fall Detection
Wandering Prevention & BLE/UWB RTLS
EN 60849 Nurse Call Integration
BS 8521-2 Telecare & Home Safety
Continuous Vital Signs Wearables
FHIR R4 AlarmEvent & EPR Integration
CQC & MHRA MDR Compliance
Memory Care & MCA Compliance
Multi-Axis IMU Fusion & Fall Classification
6-DOF IMU quaternion fusion with population-trained fall classifier. Threshold-free ML model trained on labelled clinical fall data — post-stroke, Parkinson's, and hip-replacement cohorts included.
BLE 5.3 / UWB IEEE 802.15.4z RTLS Stack
Dual-radio architecture: BLE 5.3 for alarm relay at room granularity; UWB time-difference-of-arrival for sub-50 cm corridor accuracy. Coexistence managed via dynamic channel scheduling.
On-Device Edge-ML Inference
TinyML CMSIS-NN quantised model on ARM Cortex-M running at under 1 mW average. Fall event classified in <200 ms from impact onset — no cloud round-trip in the alarm critical path.
EN 60849 Nurse Call Protocol Stack
Full EN 60849 hardwired and IP nurse call stack — call priority routing, staff acknowledge handshake, escalation timer, and integration with legacy analog nurse call buses via protocol bridge.
MHRA MDR 2002 / EU MDR 2017 Classification
Class I and IIa device classification, QMS (ISO 13485), clinical evaluation report, IEC 62133 battery safety, and IEC 60601-1 electrical safety for patient-worn devices.
FHIR R4 AlarmEvent & AuditEvent
Device firmware generates HL7 FHIR R4 AlarmEvent and AuditEvent resources at the edge. SMART on FHIR OAuth 2.0 authentication for direct EPR integration without middleware server dependency.
False Alarm Rate Optimisation
ROC curve calibration with per-ward operating point selection. False alarm rate profiled against ward type, patient mobility class, and time-of-day — not a single factory default.
MCA 2005 & GDPR Article 9 Compliance
On-device inference architecture keeps biometric data on-chip — no raw IMU streaming to cloud. MCA best-interests documentation, DoLS/LPS alignment, and GDPR Article 9 DPIA for every deployment.
Chipsets & Platforms
Tested silicon and proven stacks — no experimental platform dependencies.
Clinical-Grade Fall Detection: Sensitivity Without Alarm Fatigue
Sensitivity vs. Specificity: Why Both Numbers Matter
Mannequin drop tests do not replicate the stochastic kinematics of a frail elderly patient losing balance in a ward corridor. The fall classifier is trained on labelled IMU data from prospective ward trials across post-stroke, Parkinson's, and hip-replacement cohorts, and tuned to a sensitivity/false alarm rate operating point agreed with the clinical team before production build. Reporting sensitivity alone is a red flag — clinical deployments require both numbers, measured on real patients and stratified by mobility class and ward environment.
Perimeter-Aware Wandering Prevention With MCA Compliance
MCA 2005, DoLS & Liberty Protection Safeguards
Electronic monitoring of a patient who lacks capacity engages the MCA 2005 best-interests framework and, where it constitutes a deprivation of liberty, a DoLS authorisation or LPS approval. Systems include configurable monitoring modes documented against the least-restrictive option principle, with documentation templates reviewed against both the current DoLS framework and the forthcoming LPS transition so procurement teams can make defensible decisions.
BS 8521-2 Telecare & the PSTN Digital Switchover
PSTN Sunset & Digital Switchover Readiness
BT's PSTN retirement by 2027 renders legacy alarm transmitters obsolete. Telecare hardware uses 4G LTE primary with Wi-Fi fallback, integrating with IP-based alarm receiving centres to TSA's ADASS interoperability standard — alarm delivery is maintained even when one path fails. All hardware ships with PSTN sunset compliance documentation for local authority procurement frameworks, and on-device GDPR Article 9 inference ensures no biometric data leaves the device without consent.
Clinical Validation From Day One
Ward staff, tissue viability nurses, and falls prevention leads are engaged before design freeze — not during a post-production clinical evaluation that finds problems too late to fix. Sensitivity and false alarm rate targets are agreed with clinical stakeholders and documented before any firmware architecture decision.
Regulatory Pathway Without the Surprises
MHRA MDR 2002 classification, CQC registration evidence, IEC 60601-1 electrical safety, and BS 8521-2 telecare compliance are built into the development process — not appended as a certification exercise at the end.
Algorithm Calibration for Real Populations
Fall detection models are trained and validated on clinical ward data — post-stroke, Parkinson's, hip-replacement, and frailty cohorts — not laboratory mannequins. The resulting specificity at any given sensitivity target is meaningfully better in deployment.
Full-Stack FHIR Integration Delivered
The firmware-to-EPR alarm pipeline is delivered as a tested, integrated system — FHIR R4 AlarmEvent generation at the device edge, SMART on FHIR authentication, and validated routing to EMIS, SystmOne, and Epic. No middleware server to maintain.
80-Bed Acute Stroke Unit — Fall Detection Deployment
Prospective 6-month ward trial on an 80-bed stroke unit. Existing pendant system was generating over 4 false alarms per patient per day — nursing staff had learned to ignore the alarm panel.
Our population-trained ML classifier was calibrated specifically for post-stroke kinematics and the ward's physical environment. Post-deployment audit showed 94.7% sensitivity with 0.9 false alarms per patient per day — a level the ward manager described as clinically manageable.
- 94.7% detection sensitivity achieved on prospective ward trial — not laboratory mannequin testing
- False alarm rate reduced from 4.8 to 0.9 per patient per day — rated clinically manageable by ward manager
- EN 60849 nurse call integration with priority-routed escalation deployed alongside existing infrastructure
- Edge-ML on-device inference in <200ms — no cloud dependency in the alarm critical path
48-Bed Memory Care Unit — MCA-Compliant Wandering Prevention
A 48-bed dementia care unit required wandering prevention that met MCA best-interests requirements and CQC inspection evidence standards.
UWB anchors were placed at all exit routes and high-risk corridors; monitoring modes were documented against the least-restrictive option principle with individual resident capacity assessments. Fourteen months of operational data showed zero unauthorised exits and a 63% reduction in physical redirection incidents — with CQC inspectors citing the system as evidence of good practice.
- Zero unauthorised exits across 14 months of operational deployment
- 63% reduction in physical redirection incidents versus pre-deployment baseline
- CQC inspectors cited the system as evidence of good practice in their inspection report
- MCA Section 5 best-interests documentation and DoLS alignment evidence pack included in CQC registration bundle
Building patient safety electronics for healthcare? Let's protect your patients.
From fall detection algorithm calibration to CQC-compliant fleet deployment and FHIR EPR integration, we deliver clinically validated patient safety electronics.
Quote Your Patient Safety Project