What the PCCP Scope Audit actually produces

PCCP SECTION 1 (Description of Modifications) — Current scope: retraining on adult fundus datasets from new clinical sites, minor threshold adjustments (±5% of cleared performance), software version updates not altering intended use or patient population. Proposed change: expand training dataset to include pediatric fundus images (ages 10-17) and update thresholds for pediatric population. ASSESSMENT: The proposed change materially EXCEEDS the existing PCCP scope. The current PCCP explicitly states modifications do not alter 'patient population,' and expanding from adults to pediatric (ages 10-17) is a clear patient-population change. This does not fit Section 1 as written. PCCP SECTION 2 (Modification Protocol) — Current scope: specifies verification/validation procedures for retraining on new adult datasets within ±5% performance adjustment. Proposed change verification needs: (a) pediatric dataset representativeness validation, (b) pediatric-specific threshold optimization, (c) clinical study in pediatric population (if claimed efficacy in pediatric cohort). ASSESSMENT: The existing modification protocol assumes adult-dataset retraining and adult-centered performance metrics (sensitivity/specificity on adult fundus images). Pediatric images may differ in quality, anatomic landmarks, and disease presentation. The protocol does not address pediatric-specific verification (e.g., handling of variable pupil dilation, media opacity common in pediatric exams). DOES NOT FIT existing protocol. PCCP SECTION 3 (Impact Assessment) — Current scope: analysis of risk for adult-population retraining. Proposed change impact: pediatric population introduces new failure modes (e.g., poorer performance on non-dilated pediatric exams, sensitivity to anatomic variants in younger patients) not contemplated in adult-focused risk assessment. ASSESSMENT: The existing impact assessment does not cover pediatric patient subgroup. DOES NOT FIT. CONCLUSION: Proposed change does NOT fit existing PCCP across all three sections. Requires amendment or new submission.

IF PCCP AMENDMENT IS PURSUED: (1) SECTION 1 (Description of Modifications) — AMENDMENT REQUIRED. Current language: 'Modifications do not alter intended use or patient population.' Proposed amendment: 'Modifications include retraining on pediatric fundus image datasets (ages 10-17) to optimize performance in pediatric Type 1 diabetes screening context. Intended use expands to include pediatric patients ages 10-17. Threshold adjustments up to ±10% of cleared adult-population performance baseline permitted to optimize pediatric sensitivity.' Rationale: Making patient-population expansion explicit and setting pediatric-specific performance threshold range. (2) SECTION 2 (Modification Protocol) — AMENDMENT REQUIRED. Current protocol does not specify pediatric verification. Proposed addition: 'For pediatric datasets: (a) Clinical validation on minimum 500 pediatric fundus images from ≥2 independent clinical sites; (b) demographic stratification by age group (10-13, 14-17), gender, ethnicity; (c) sensitivity/specificity targets for pediatric cohort [specify target, e.g., ≥95% sensitivity matching adult predicate]; (d) image-quality assessment (dilated vs non-dilated, media opacity grading) across pediatric cohort; (e) threshold optimization via pediatric validation cohort, with results reported separately from adult performance.' (3) SECTION 3 (Impact Assessment) — AMENDMENT REQUIRED. Current assessment does not address pediatric-specific risks. Proposed addition: 'Pediatric patient subgroup failure modes: (a) Reduced image quality in pediatric exams may increase false-negative rate; mitigation: image-quality flagging alerts clinician to manual review; (b) anatomic/physiologic differences (pupil size, lens density) may affect algorithm performance; mitigation: age-stratified validation and threshold adjustment per Section 2; (c) low population incidence of advanced DR in pediatric Type 1 diabetes (most are early stages) may bias model toward sensitivity at cost of specificity; mitigation: performance targets emphasize sensitivity >95% to minimize missed cases. New risk: use of adult-trained thresholds on pediatric images may yield higher false-positive rate, potentially driving unnecessary referrals; mitigation: pediatric threshold optimization and clinical audit of false-positive rate in real-world deployment.' Expected regulatory timeline for amendment: 4-6 weeks for regulatory affairs + engineering collaboration to assemble evidence; submission to FDA for pre-submission meeting (Type C) recommended before formal amendment.

RECOMMENDATION: The proposed change REQUIRES A PCCP AMENDMENT (not a full 510(k)). However, a PCCP amendment is appropriate only if (a) the pediatric validation data are robust (≥500 images from ≥2 sites, independent of adult training), and (b) the performance targets for pediatric cohort are clearly defined and met. If either condition is not met, a NEW 510(k) submission (substantial equivalence to adult predicate) is necessary. CRITICAL CAVEAT: If the proposed change is driven by post-market signal (e.g., pediatric patients presenting with symptoms of advanced DR being missed by the adult model), this may trigger MDR-adjacent obligations under 21 CFR 803 even if the modification mechanically fits a PCCP amendment. QUESTIONS FOR REGULATORY AFFAIRS LEAD: (1) Do we have existing pediatric fundus-image data (500+ images, ≥2 clinical sites)? If yes, what is the data-maturity status (raw, labeled, validated)? Timeline to finalize validation? (2) What performance targets are we claiming for the pediatric cohort, and how do they compare to the adult predicate (IDx-DR)? If pediatric sensitivity is lower, what is the clinical/regulatory rationale? (3) Is there a post-market signal (complaints, failed detections) driving this change, or is it a proactive opportunity expansion? If signal-driven, what is the incident frequency and severity? (4) Have we engaged the FDA in a Type C pre-submission meeting for the PCCP amendment, or should we do so before investing in validation? QUESTIONS FOR CLINICAL LEAD: (1) What is the clinical justification for expanding to ages 10-17? Is there documented unmet need (e.g., rising Type 1 diabetes incidence in this age group, screening gaps in pediatric endocrinology clinics)? (2) Are the pediatric fundus images representative of intended real-world use (e.g., mix of dilated and non-dilated images, typical image quality from pediatric clinics)? (3) What is the expected sensitivity/specificity in pediatric cohort, and is it clinically acceptable for screening (e.g., ≥95% sensitivity to avoid missed cases of moderate-to-severe DR)? QUESTIONS FOR EXTERNAL REGULATORY COUNSEL: (1) Does FDA guidance on pediatric AI/ML modifications favor PCCP amendment (if data-robust) or new submission? Has FDA issued specific PCCP amendment precedents for patient-population expansion? (2) If this is signal-driven (post-market issue in pediatric cohort), does 21 CFR 803 MDR reporting apply, and should we prepare adverse-event documentation alongside the amendment submission? (3) What is the anticipated FDA review timeline for a PCCP amendment vs. a full 510(k) for this modification?