What the 510(k) Evidence Mapping actually produces

INTENDED USE: NovaPulse 3000 — detection and recording of cardiac arrhythmias via 2-lead ECG patch, 14-day wear, Bluetooth cloud transmission, adult patients. Predicate CardioSense Patch II — detection and recording of cardiac arrhythmias via 2-lead ECG patch, 14-day wear, office-based download, adult patients. SAME intended use. TECHNOLOGICAL CHARACTERISTICS (Side-by-Side): | Characteristic | Predicate | NovaPulse | Same/Different | Different-Q Raised? | |---|---|---|---|---| | Lead count | 2-lead ECG | 2-lead ECG | Same | No | | Wear duration | 14 days | 14 days | Same | No | | Electrode material | Wet gel | Dry hydrogel | Different | Possibly — adhesion/skin-irritation risk; bench testing addresses (72h sweat simulation, biocompatibility per ISO 10993; no SAEs in clinical study) | | Data transmission | In-office download via device reader | Real-time Bluetooth cloud | Different | Yes — network security, data integrity, cybersecurity risk. Bench testing shows 0.02% packet loss; Bluetooth immunity per IEC 60601-1-2; clinical study does not assess cybersecurity or wireless-stack failures. | | Arrhythmia classification | None specified (manual review implied) | AI algorithm (cloud-based) | Different | Yes — AI/ML SaMD regulatory expectations apply. Training/validation data, model lock status, subgroup performance, and PCCP (if adaptive) not provided in clinical evidence summary. | | Form factor | 52g | 38g | Different | No — size reduction does not raise safety/effectiveness questions if adhesion, wear comfort, and clinical performance are equivalent. Clinical data shows equivalent sensitivity/specificity. | PERFORMANCE DATA: 120-patient prospective study vs. 12-lead Holter reference: sensitivity 94.2%, specificity 97.1% for AFib. No SAEs. Bench testing addresses electrode adhesion (72h sweat), EMI immunity (per IEC 60601-1-2), battery life (>336h), and Bluetooth data integrity (0.02% packet loss). Biocompatibility per ISO 10993 satisfied. GAP: Clinical validation does not address subgroup performance (age, sex, BMI, skin conditions, atrial fibrillation type/burden) — P0-level for FDA review. AI algorithm validation (training/test data independence, model lock, subgroup performance across intended-use population) not provided.

GAP 1 — SUBGROUP PERFORMANCE: Current evidence: 120-patient study, no subgroup stratification. FDA expectation: sensitivity/specificity by age (categorical), sex, BMI, skin condition, AFib paroxysmal vs. persistent. Recommended next step: Subgroup analysis of existing 120-patient data if sufficient N in each stratum (minimum ~30 per subgroup); if underpowered, plan targeted prospective enrollment (30–50 subjects each subgroup). Owner: Clinical. GAP 2 — AI/ML VALIDATION: Current evidence: None. FDA expectation (per guidance on software modifications and AI/ML SaMD): training data documentation (size, source, labels, curation), validation data independence, model architecture/training algorithm, model-lock status, locked-vs-adaptive determination, PCCP (if adaptive). Recommended next step: Pre-submission meeting (Q-Sub) to confirm FDA's specific expectations for cloud-based AI arrhythmia classification in §510(k) pathway (vs. De Novo or PMA). Owner: Regulatory affairs + Clinical. GAP 3 — WIRELESS SECURITY & DATA INTEGRITY: Current evidence: Bench testing (0.02% packet loss, Bluetooth immunity per IEC 60601-1-2). FDA expectation: cybersecurity risk analysis (NIST framework or equivalent), authentication/encryption scheme, resilience to wireless interference in real-world settings, data-loss and replay-attack scenarios. Recommended next step: Cybersecurity risk summary document (IEC 62304-compliant) and pre-sub meeting confirmation that Bluetooth-stack security is acceptable for this device class. Owner: Engineering. GAP 4 — USABILITY & PATIENT ADHERENCE: Current evidence: None provided. FDA expectation: human-factors analysis (patch wear comfort, removal difficulty, app usability, alert fatigue risk from cloud-based AI alerts). Recommended next step: Usability testing with target population (20–25 subjects, diverse ages/technical comfort). Owner: Clinical/UX. SUBGROUP PERFORMANCE: HIGH PRIORITY. Single sensitivity/specificity figure masks potential performance cliff in subpopulations (e.g., elderly with lower skin impedance; high-BMI patients with variable electrode contact). This is the most likely FDA hold point.

TOPIC 1 — AI/ML CLASSIFICATION PATHWAY: Is cloud-based arrhythmia classification by AI acceptable in §510(k) pathway, or does it trigger De Novo review? If §510(k) is acceptable, what is FDA's specific expectation for model validation (training/test split, external validation data, locked vs. adaptive model, update procedures/PCCP)? TOPIC 2 — SUBGROUP VALIDATION STRATEGY: What is FDA's minimum N per subgroup for sensitivity/specificity claims? Can existing 120-patient data be post-hoc stratified, or must new prospective data be collected? If prospective, what subgroups are mandatory (age, sex, skin condition, AFib burden)? TOPIC 3 — WIRELESS SECURITY ACCEPTANCE: For Bluetooth-enabled wearable cardiac devices, what is the expected cybersecurity documentation standard (NIST, IEC 62304, other)? Is authentication + encryption sufficient, or does FDA expect additional resilience testing (e.g., wireless interference, packet-loss scenarios)? TOPIC 4 — PREDICATE ADEQUACY: Is CardioSense Patch II (K192847) the right predicate for NovaPulse when Bluetooth transmission and AI classification are new? If predicate adequacy is questioned, is De Novo the pathway? Topic 5 — USABILITY/HUMAN FACTORS: Is usability testing required, or can it be waived for a wearable patch (low-interaction device)?