FDA Reviewer Questions on On-Body Injectors and Large-Volume Wearable Injectors in NDA and BLA Reviews

FDA reviewer questions on on-body injectors and large-volume wearable injectors in NDA and BLA reviews

On-body injectors (OBIs), on-body delivery systems (OBDS) and large-volume wearable infusors get much closer device scrutiny than prefilled syringes or autoinjectors. In the Drugs@FDA review record, CDRH and CDER reviewers treat these products as electromechanical delivery systems, and in at least one case CDRH explicitly classed on-body medication-delivery devices as infusion pumps ⁷¹¹. Their questions fall into five recurring themes: (1) dose and flow-rate accuracy across the full delivery profile, (2) adhesion and skin-contact performance, (3) occlusion, alarm and delivery-failure handling, (4) human factors (HF) validation of error-state recognition and premature removal, and (5) 21 CFR Part 4 combination-product CGMP, design controls and device-constituent facility oversight.

This article summarizes what reviewers actually asked and concluded, drawing mainly on the four programs with the most detailed public device reviews: Neulasta's on-body injector (BLA 125031), Repatha's Automated Mini-Doser/Pushtronex on-body infusor (BLA 125522), the Furoscix on-body infusor (NDA 209988), and the Skyrizi on-body injector (BLA 761105; public device review posted with the BLA 761262 package). Later OBI programs are covered where the public record supports it.

Products and review depth at a glance

Dose accuracy, delivered volume and flow rate

Reviewers want the whole delivery profile specified, not just the total volume

The most consistent device deficiency was a specification that controlled total delivered volume but not how the dose was delivered.

• Neulasta OBI. The device was designed to deliver 0.6 mL, but FDA found no requirement limiting how fast it could do so, such as a maximum flow rate or a minimum delivery time. Reviewers asked for a complete set of requirements needed for safe and effective dosing and evidence that the device reliably met them ⁶⁶⁶. FDA also questioned whether the 59 "Neutime Devices" used in the delivery-time verification report represented the commercial device. It asked Amgen to list and justify every difference and to confirm that delivery-time and delivery-accuracy testing used Neulasta rather than a surrogate ⁶⁶⁶. CDER accepted a residual (redacted) mis-dosing probability, reasoning that it was offset by convenience and better adherence to correctly timed administration ⁶⁶⁵.
• Furoscix infusor. CDRH found the flow-accuracy specification deficient because the only value listed was "10 mL", which the reviewer said describes dose accuracy, not flow accuracy ⁵⁹². Because the biphasic profile mattered clinically, the reviewer called unverified individual rates "potentially a problem." At that stage the dose-delivery accuracy data were judged inadequate and the shipping/transportation evaluation unacceptable ²⁶². An information request asked for the flow-accuracy rating over the entire allowable range, covering both phases and the transition between them, plus minimum and maximum allowed flow rates throughout delivery ⁵⁹⁴. The reviewer also rejected the sponsor's short-term-rate analysis. It did not match the stated acceptance criterion, and the rationale "does not demonstrate the device meets flow accuracy over the entire dose delivery" ⁵⁹⁰.
• Clinical linkage. On Furoscix, the device reviewer disputed the sponsor's position that delivery accuracy would not materially affect over-diuresis risk. The review notes that the CDER clinical division confirmed rapid delivery or cartridge overfill could cause over-diuresis requiring medical intervention ⁵⁹⁶.

The Furoscix deficiencies were resolved in the resubmission. The later device review listed the essential performance specifications as 10 mL dose volume, 5-hour delivery time, and two steady-state flow rates (3.75 mL/hr and 1.56 mL/hr). It found verification and validation acceptable, including accelerated aging and shipping ²⁴⁷. The CMC reviewer did not add "deliverable volume" as a routine drug-product release test, noting that device performance controls already covered total volume and first-hour volume and that CDRH found this acceptable ⁵⁹⁷.

Statistical confidence and sample size

• For Skyrizi, CDRH recorded delivered-volume accuracy and injection time among the recommended essential performance requirements (EPRs) for OBDS injectors. Other EPRs were activation force, extended-needle length, adhesion/peel force and audible/visual feedback ⁶³¹. Delivered-volume verification used a 95% confidence / 97.5% reliability criterion (N=60 and N=124 across the two fill configurations). Delivery time used 95/95 attribute testing with zero failures ²¹⁰. Both were also assessed after vibration, dry heat/cold storage and transportation altitude conditioning ⁶³². CDRH recorded no design-control deficiencies ⁶³¹.
• For Repatha's AMD, FDA asked Amgen to justify or expand sample sizes where only one device appeared to have been tested, including the Delivery Cancellation Onset Time test ⁷⁰⁸. FDA also noted at least one failure in each of the deliverable-volume, injection-time and storage temperature/humidity tests. A root-cause-driven design change had followed, but the devices did not appear to have been retested ⁷⁰⁸.
• For Furoscix, reviewers cited inadequate sample size to establish a 95/95 interval among the design-verification problems ²⁵¹²⁶¹.

Real-time aging, EMI and environmental effects on delivery

• Repatha's approval carried postmarketing commitments for real-time aging studies at 1 year, 2 years plus 31 days, and 3 years plus 31 days. Tests covered device function, deliverable volume, injection time, sterile integrity and adhesive function, using methods equivalent to the accelerated-aging program ^304307716719.
• For Neulasta, the team leader accepted a 36-month device shelf life from accelerated data. Because accelerated data might not fully characterize battery storage, real-time shelf-life results were to be reported in annual reports ²⁰⁶. Reviewers also asked for the device failure rate under air-discharge ESD exposure and why that rate was acceptable ⁶⁶⁶.
• For Furoscix, FDA stated that environmental changes could affect flow and dose accuracy. It requested an IFU warning against use during transit or outside home or clinical settings, testing or justification for fluid and particulate exposure, justification of maximum rated pressure, and EMC limits in the IFU ²⁵⁶. Udenyca's Onbody labeling likewise states that electromagnetic interference may affect dose accuracy and treatment duration ¹⁹⁵¹⁹⁶.

Adhesion and skin contact

Adhesion is treated as a performance requirement, not only as a labeling topic. Later reviews list it explicitly among EPRs: adhesion/peel force for Skyrizi ⁶³¹²¹⁰ and adhesive tack force for Furoscix ²⁴⁶.

• Skyrizi OBDS. A 36-participant healthy-volunteer adhesive study tested abdomen and thigh wear during walking, sitting, reaching and bending. Mean device-surface adherence was 67% on the abdomen and 59% on the upper thigh ⁶⁹⁶. Reviewers noted incomplete edge adhesion in two participants, away from the needle-insertion area. The device still stayed on for the wear period, and the sponsor stated that incomplete adhesion did not cause dislodgement ⁶⁹⁶⁶⁹⁵. The adhesive patch was characterized as a limited-duration, intact-skin surface device (about 14.6 hours of lifetime exposure), tested for cytotoxicity, irritation and sensitization under ISO 10993 ⁶⁹⁵. Reviewers also attributed higher, mild, self-limited injection-site reactions with the OBDS largely to the adhesive and the larger injection volume ⁷⁶².
• Furoscix infusor. The clinical review judged the adhesive "adequate for the proposed use," with three partial dislodgements, none of which interfered with delivery. All subjects had no irritation or only minimal erythema ⁶⁹⁸. The study excluded people with a history of skin reactions to medical adhesives ⁷⁰⁴. On the device side, reviewers flagged inadequate adhesive validation ²⁵¹²⁶¹. They also asked the sponsor to state which components, including skin-contacting ones, were included in each biocompatibility test ⁷⁰⁰. The HF reviewers found the use-related risk analysis incomplete for patients with skin-related risk factors (lotions or oils, prior abdominal surgery, local skin conditions), citing possible skin injury and adhesion failure ⁶⁷⁶.
• Repatha AMD. Complaint tables tracked "Device Application Adhesive Does Not Function as Expected" as a failure category ³⁵⁹. In HF testing, untrained users tore off the whole adhesive or stopped when a small adhesive tab resisted removal. FDA attributed this to users expecting visible tabs to be pulled off and recommended IFU wording such as "Do not pull the skin adhesive backing off" ³⁴⁹.
• Neulasta OBI. When a new device material appeared without biocompatibility data, FDA asked for a detailed ISO 10993-17 toxicological risk assessment. This was to cover all three constituent chemicals, margins of safety, adult and pediatric body weights, reference doses/NOAELs/LOAELs and exposure assessment ⁴³⁴.

Across the approved labeling, adhesive risk controls are consistent. Labels include acrylic-adhesive allergy warnings ^429446389, instructions not to reapply a detached device ⁴¹⁵⁶⁰⁸, and instructions to seek a replacement dose if the device loosens or the adhesive becomes saturated ¹⁸¹¹⁹³.

Occlusion, alarms and delivery failure

Alarm architecture and failure-state completeness

• Furoscix. FDA could not locate several alarms it considered critical for safe operation: occlusion detection, low/empty reservoir, undocking, key-pressed and tone-test failure. The sponsor was asked to add them, justify their absence, or propose alternative mitigations ⁴⁵⁰. Time to occlusion was later listed among the verified design inputs ²⁴⁶²⁴⁷.
• Neulasta OBI. A design requirement said the device must alarm on failure to complete delivery. FDA asked Amgen to identify every cause of that failure state, beyond software defects, occlusion and delivery pressure, with controls and verification for each ³²⁶. Reviewers also flagged an internal inconsistency between two requirements. One required delivery at a set confidence level under constant backpressure; the other required delivery even with an occluded cannula tip. FDA asked Amgen to reconcile them and to confirm that the contract manufacturer's test protocols traced to the Neulasta device requirements ³²⁶. The occlusion and delivery-pressure protocol also did not verify the separate requirement that the device power up when filled below 0.60 mL ³²⁶.

Clinical and bench failure rates

• Repatha AMD. At the July 2014 cutoff there were 70 complaint issues in 828 injections, 53 of them product failures (8.5% and 6.4%) ⁵⁷⁵⁵⁷⁹. By January 2016 the totals were 251 complaints and 76 failures in 4,506 attempts (5.6% and 1.7%). For the modified, commercial-like AMD the figures were 181 and 23 in 3,678 attempts (4.9% and 0.6%) ⁵⁷⁵⁵⁷⁶. "Device not functioning / activation failed" was the leading category (46 of 76 cumulative failures) ⁵⁶⁹. FDA agreed the likely consequence was underdosing: one missed AMD dose equals a missed month of therapy, and repeated misses could impair LDL-C lowering ⁵⁸⁰. In the device-use comparison, complete delivery was achieved in 134 of 144 AMDs (93.1%) ⁵⁸⁷.
• Furoscix. The product design clinical validation study missed its prespecified goal of at least 95% of products free of major system failure (63/67, 94%; 95% CI 85% to 98%). Reported problems included dislodgement, under-delivery, unintended alarms and software issues ⁴⁵². FDA requested a clinical assessment of the worst-case pump-failure scenario causing inadequate furosemide delivery ⁴⁵¹. It stated that observed malfunctions might not be adequately mitigated simply by revising the IFU and running more HF studies ⁴⁵¹⁴⁵².
• Skyrizi OBDS. The bioequivalence bridging study recorded 14 OBDS device failures attributed to use errors, fluid-path obstruction and manufacturing issues. Eleven subjects were excluded from the PK analysis for OBDS dosing failures or related issues ^762763440. The device reviewer found the exclusions higher than expected. The review concluded that the study showed bioequivalence of the drug but raised device safety and effectiveness questions, and asked for a detailed root-cause discussion ⁷⁶². FDA's information request asked for investigation documentation for each dosing failure and mitigations for every failure mode ⁴⁴⁰. Leakage complaints were traced to users not fully pressing the start button, and the IFU was revised to add a "click" cue ⁴⁴⁰. In-study failures also included a damaged cartridge door and incomplete cartridge insertion ⁴⁴².

Alarm confusion and cross-program learning

FDA returned repeatedly to the risk that users would read an error signal as "done" and remove the device early.

• In Repatha HF testing with a pre-triggered hazard alarm, 6 of 45 untrained users did not detect or understand the alarm, and 4 read the flashing red light as completion ⁵⁶⁶. In a later supplemental study, all 45 participants detected the error alert ⁵⁸¹.
• For Skyrizi, the reviewer judged the "not working properly" and "injection complete" sounds too similar. Confusing them could cause premature removal and an incomplete dose, and the sponsor's information-request response did not address the point ⁴⁴⁰. FDA then searched 238 postmarketing reports mentioning beeps or alarms and found no cases of this specific confusion ⁴⁴¹. It also reviewed Repatha Pushtronex FAERS data. Of 59 leakage reports, only 14 said the status light turned red during injection, while 26 confirmed it never turned red despite the leak ⁴⁴¹.

The Neulasta and Udenyca labels both state that missed or partial doses have been reported when the OBI did not perform as intended. They warn of higher risk of neutropenia, febrile neutropenia and infection, and direct patients to contact their healthcare provider about a replacement dose ³³²³²⁵.

Human factors

Critical-task definition

FDA pushes sponsors to classify as critical any task whose failure leads to delayed, partial or omitted therapy.

• For Furoscix, reviewers rejected a narrow critical-task list. They concluded that errors causing delayed, partial or omitted treatment could lead to medical intervention or hospitalization in fragile heart-failure patients, even though the product is not for emergency use ^672675687. FDA also wanted vial-septum disinfection and adapter handling designated critical because of infection risk ⁶⁷⁴. Recognizing that delivery had finished, inspecting for damage, and device reuse were further missing tasks ⁶⁸⁸.
• For Neulasta, the only critical task in the summative plan was responding to an alarm state: recognizing the red light and beeps as an error before disposal and stating the correct contact action ²⁷¹.
• For Repatha's AMD, DMEPA focused on device-unique essential tasks whose failure could cause underdosing or a missed dose, namely cartridge insertion and door closure. It accepted residual risk for familiar injection tasks without added mitigation ⁵⁰⁹.

Study design expectations

• FDA told Amgen that clinical-trial usability data could serve only as formative input and that a simulated-use validation study was required before approval ⁵¹⁶. It recommended at least 15 participants per distinct user group, a trained/untrained split totaling 90 users, and a simulated complete device failure ⁵¹⁰⁵¹⁶. Study conditions were to include challenging environments such as wet or gloved fingers, dim light and noise ⁵¹⁰. The final AMD study enrolled 93 users: 31 patients, 31 caregivers and 31 HCPs ⁵⁰⁹⁵¹⁵.
• For Furoscix, FDA found that the intended-to-market device and user interface had not been tested, because the device and IFU changed after the summative studies ^688689691. It asked for a "train-the-trainer" model using representative nurses and physicians and a training-decay interval of at least five days ⁶⁷⁴. It also cautioned that clinical reliability data could be inflated because trained study staff, not patients, applied and removed the device ⁶⁷².
• For Skyrizi, FDA's pre-submission advice called for an HF validation protocol built on a comprehensive use-related risk analysis. The study was to use intended-to-market labeling and include root-cause analysis of every use error, difficulty and close call ³⁸⁴³⁸⁵.

Observed errors and resulting changes

• Neulasta. Some HCPs had trouble filling the OBI from the syringe on the first attempt. Root cause was unclear instructions on needle insertion angle, which risked needle bending and spillage. Step 6 of the HCP IFU was revised, and DMEPA had no further concerns ²⁷².
• Repatha. Users removed the AMD early after misreading the flashing green light, plunger travel or medicine-window status. FDA recommended adding "Plunger completely fills medicine window" to the IFU and bolding "solid green" ⁵⁶⁸. After a patient failed to fully latch the cartridge door, FDA requested the wording "Apply enough pressure when closing the door and make sure a snap is heard before proceeding" ⁵⁰⁷.
• Furoscix. In one study, 16 users could not tell whether the infusor was properly filled ⁶⁸². Errors persisted after IFU revisions, and FDA found the mitigations ineffective ⁶⁹². DMEPA and CDRH initially concluded the HF data did not show safe and effective use by intended users ^672675692. A later review found the June 2020 validation results acceptable for the device as tested, but not for the post-study modified commercial device without further analysis ⁶⁹¹. A subsequent assessment accepted residual risk for eight tasks with observed errors, including adhesive-liner removal, start-button activation and alarm response ⁶⁶⁸.

When HF validation was not required

For Empaveli's original approval, which used 510(k)-cleared syringe infusion pumps rather than a dedicated wearable, DMEPA accepted a use-related risk analysis and IFU task comparison in place of HF validation data ⁶¹²⁶¹³. The later EMPAVELI on-body Injector was added through a labeling supplement. The approval letter does not include a public review narrative ⁸²⁰.

21 CFR Part 4, design controls and device-constituent oversight

Part 4 CGMP operating model

• For Furoscix, FDA explained the Part 4 options. A manufacturer can comply fully with both 21 CFR 210/211 and Part 820, or use a streamlined approach built on one system plus the Part 4-specified provisions of the other ²⁴⁹. FDA asked for every manufacturing site, the activities at each, and the operating system used ²⁴³. The sponsor chose a drug-CGMP-based streamlined system, and the reviewer found its GMP summary adequate ²⁴⁵.
• For Repatha, where Amgen used drug CGMPs as the primary system, FDA required the Part 820 elements to be demonstrated for the combination product as a whole, not only the device constituent. These were design controls (820.30), purchasing controls (820.50) and CAPA (820.100) ³⁰¹³⁰². FDA also asked that Form FDA 356h identify each facility's device-related activities and the finished-product distribution site ²⁹⁸³⁰².
• For Skyrizi, FDA stated that the Part 4 CGMP and postmarketing safety reporting obligations apply ²¹². It asked AbbVie to identify every facility involved in the OBDS, split final-assembly responsibilities between AbbVie and its supplier, and update Form 356h ²⁰⁸. It also asked for the CAPA documentation and the control strategy for device EPRs ²⁰⁸.

Design controls, EPRs and verification

• For Furoscix, FDA initially required a full account of how 820.30 design controls were applied, from planning through design transfer and the design history file ²⁴³. The first-cycle CDRH review rated the risk analysis, hazard identification, mitigations and verification inadequate ²⁵⁸. FDA said the sponsor had not provided evidence for all expected infusion-pump EPRs, "explicitly including flow-rate accuracy" ²⁴⁴. Test reports lacked clear objectives, acceptance criteria and statistically justified sample sizes ²⁴⁴. The safety assurance case lacked traceability from risks to mitigations ²⁵¹²⁵⁴. In the resubmission, CDRH found the device constituent approvable and the prior-cycle risks addressed ²⁶³.
• For Repatha's AMD, FDA found the software description incomplete and the hazard analysis missing a referenced document. It requested updated software requirements, traceability and verification/validation ³⁰³. CDRH ultimately concluded that the commercial device passed performance, biocompatibility and software validation testing, with acceptable residual risk ³⁸³.
• For Neulasta, FDA asked Amgen to show where specific design requirements were verified in the software and hardware verification reports and to confirm the tested units represented the commercial product ⁶⁶⁶. A CDRH Office of Compliance desk review found management controls, design controls, purchasing controls and CAPA adequate ²⁰⁶.
• For Skyrizi, CDRH found the risk management, EPR definition, verification methods, reliability and traceability acceptable ²⁰⁷. Electrical safety, EMC and MRI labeling gaps were fixed in labeling ²¹¹.

Facilities and inspections

• The Neulasta device manufacturer's September 2014 inspection was classified NAI. FDA considered it sufficient for Part 820 compliance and waived a preapproval inspection ²⁰⁶.
• For Repatha's AMD, firms with recent VAI or NAI device inspections did not need preapproval inspections. Firms with no prior FDA medical-device inspection did ²⁹⁹.

Postmarketing commitments and reporting

• Skyrizi. CDRH recommended a PMC to verify or set acceptance criteria for audible activation feedback (click, timing, relation to dose delivery, decibel level) and to add activation-force release testing as an EPR before commercial distribution ²¹⁶.
• Repatha. Real-time aging PMCs covered deliverable volume, injection time and adhesive function ³⁰⁴³⁰⁷.
• Udenyca. A 2023 supplement approval letter restated PMC 4558-1. This called for IP68 fluid-ingress testing on nine device constituents from three lots, then evaluation of deliverable volume, dose efficiency, dosing time, LED and fill-indicator visual feedback, audio feedback and adhesive force ¹⁶⁹. The Onbody injector is marked IP68 Submersible ⁷⁴⁵.
• Part 4 Subpart B. Approval letters for Neulasta, Repatha, Furoscix, Udenyca and Sarclisa Escena each remind the sponsor that the product is a Part 3 combination product subject to combination-product postmarketing safety reporting under 21 CFR Part 4, Subpart B ^{205311259169438}.

Practical takeaways for OBI and wearable injector submissions

1. Specify delivery kinetics, not just volume. Reviewers expect maximum and minimum flow rate or delivery time, and for multiphase profiles, rate accuracy in each phase and at the transitions, tied to clinical risk ^666592594596.
2. Treat adhesion and time-to-occlusion as EPRs. Recent reviews list adhesion/peel or tack force and occlusion time alongside delivered volume and injection time ⁶³¹²⁴⁶.
3. Map every path to the failure state. FDA asks which conditions trigger an incomplete-delivery alarm, whether key alarms (occlusion, empty reservoir, undocking) exist, and what the clinical consequence of worst-case under-delivery is ^326450451.
4. Design out alarm ambiguity. Error and completion signals that look or sound alike are a recurring HF finding, and FDA mines postmarketing data from similar OBIs when assessing a new one ^566440441.
5. Validate the final configuration with realistic training. Post-summative device or IFU changes, unrepresentative trainers, and clinical data generated by study staff have all undermined HF packages ^688691674672.
6. Explain device failures in clinical and bridging studies. Unexplained device-related exclusions in a PK bridging study prompted CDRH root-cause requests and PMCs ^762440216.
7. Document the Part 4 operating model at the combination-product level. Include facility-by-facility responsibilities on Form 356h and design control, purchasing and CAPA evidence covering the finished product ^243301302208.

Limits of the public record

For several newer OBI presentations (Udenyca Onbody, Sarclisa Escena CirCLIQ, the EMPAVELI Injector and the Ultomiris subcutaneous OBDS), the available Drugs@FDA record mostly contains labeling and action letters, not CDRH or DMEPA review narratives ^820761658. Their device-review findings therefore cannot be characterized in the same depth. Many quantitative acceptance criteria in the detailed reviews (delivery-time limits, dose-accuracy ranges, mis-dosing probabilities) are redacted ^665210632.