<?xml version="1.0" encoding="UTF-8"?><rss xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:content="http://purl.org/rss/1.0/modules/content/" xmlns:atom="http://www.w3.org/2005/Atom" version="2.0" xmlns:cc="http://cyber.law.harvard.edu/rss/creativeCommonsRssModule.html">
    <channel>
        <title><![CDATA[Stories by Lexim AI on Medium]]></title>
        <description><![CDATA[Stories by Lexim AI on Medium]]></description>
        <link>https://medium.com/@marketing.manager_25154?source=rss-24c6ec99c02c------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*3Th9VKShwAhZt00AevpLRw.png</url>
            <title>Stories by Lexim AI on Medium</title>
            <link>https://medium.com/@marketing.manager_25154?source=rss-24c6ec99c02c------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Thu, 19 Nov 2026 22:46:48 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@marketing.manager_25154/feed" rel="self" type="application/rss+xml"/>
        <webMaster><![CDATA[yourfriends@medium.com]]></webMaster>
        <atom:link href="http://medium.superfeedr.com/" rel="hub"/>
        <item>
            <title><![CDATA[Five QMSR Warning Letters, One Pattern: FDA Is Citing Risk, Design, and Suppliers as a Chain.]]></title>
            <link>https://medium.com/@marketing.manager_25154/five-qmsr-warning-letters-one-pattern-fda-is-citing-risk-design-and-suppliers-as-a-chain-93c236010bbd?source=rss-24c6ec99c02c------2</link>
            <guid isPermaLink="false">https://medium.com/p/93c236010bbd</guid>
            <category><![CDATA[pharmaceutical]]></category>
            <category><![CDATA[quality-assurance]]></category>
            <category><![CDATA[medical-devices]]></category>
            <dc:creator><![CDATA[Lexim AI]]></dc:creator>
            <pubDate>Thu, 19 Nov 2026 19:59:28 GMT</pubDate>
            <atom:updated>2026-11-19T19:59:28.073Z</atom:updated>
            <content:encoded><![CDATA[<p>The Quality Management System Regulation has been in effect since February 2, 2026, and enforcement is now visible in numbers. On November 16, Covington &amp; Burling published its quarterly review of device warning letters for July through September 2026. As of that date, FDA had posted nine letters from the quarter. Eight alleged quality system violations. Five of those came from inspections conducted under the QMSR, and three from inspections conducted under the older Quality System Regulation.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*UZsFCz_Q9ztUK4tOcHU63A.png" /></figure><p>The pattern in the five QMSR letters is the useful part. All five cited ISO 13485 Clause 7.4, purchasing. Four cited Clause 7.3, design and development. Four cited Clause 7.1, planning of product realization, which is where ISO 13485 places risk management. The three QSR-era letters looked similar in a different vocabulary: all three cited design controls under 21 CFR 820.30, and one cited purchasing controls under 820.50.</p><p>Nine letters is a small sample, and each letter is FDA’s position on one company, not a rule. But the overlap is specific enough to be worth taking seriously, because the three clauses are not independent. In the two QMSR letters that have been analyzed most closely, FDA cited them as links in a single chain.</p><h3>Why these three clauses travel together</h3><p>Under the QSR, risk management was implicit, a requirement read into design controls. ISO 13485:2016, which the QMSR incorporates by reference, makes it explicit. Clause 7.1 requires an organization to document one or more processes for risk management in product realization and to keep records of the activities. Clause 7.3 covers design and development from planning through transfer, including, in 7.3.9, the control of design changes and the determination of their significance for function, performance, usability, safety, and regulatory requirements. Clause 7.4 requires criteria to evaluate, select, monitor, and re-evaluate suppliers, and verification of what is purchased.</p><p>The FDA Law Blog has argued that the most significant change from QSR to QMSR is its focus on risk management, and that risk-related observations under Clause 7.1, process and record observations combined, have become the most common category on Form 483s, at more than 20% of all citations. Usability, which the QSR left implicit, now appears expressly in the standard as well.</p><p>The practical consequence is that risk management in an ISO 13485 system is not a document in a design history file. It is a process that is supposed to receive input from, and send output to, design changes, suppliers, complaints, corrective actions, and validation. An inspector can pick any of those and trace the line back to the risk file. The warning letters show that is exactly what is happening.</p><h3>Linemaster Switch: risk without a feedback loop</h3><p>FDA’s letter to Linemaster Switch Corporation, issued May 27 after an inspection from February 4 through March 6, concerns foot-pedal accessories used to control medical devices including Class IV medical lasers. It cites five violations. The risk management finding is the thread running through them.</p><p>FDA said the company’s risk management procedure did not define how risk activities are performed and documented, who is responsible for conducting and approving them, when documentation must be updated, or how post-market feedback, including complaints, adverse events, and recalls, is incorporated into risk management. It also said the company lacked a process failure modes and effects analysis for a foot-pedal accessory used to control Class IV lasers.</p><p>The other findings show what an unconnected risk file looks like in practice. FDA described rework that was not documented, a customer corrective-action request with no recorded root cause or corrective action, inadequate environmental control for temperature-related calibration effects, and software validation records missing raw data, acceptance results, a statistical or sample-size rationale, and verification that the software reliably measures and detects nonconformance. Each of those would normally generate a risk input: rework changes process risk, a customer complaint changes occurrence estimates, a calibration drift changes detectability. The letter’s implication is that none of that reached the risk process.</p><h3>Koven Technologies: one product change, four controls</h3><p>The Koven letter, issued July 21 after an inspection from February 2 through 6, concerns Doppler devices and shows the chain in the other direction, starting from a design change.</p><p>FDA cited Clause 7.3.9 for failure to document and assess the design changes involved in adding a fetal application to a Doppler device. It said the company could not provide documentation of the specific changes, supporting validation, an assessment of whether a new 510(k) was required, or the date distribution began for the fetal use. That is a change-control failure, and it is the same kind of record the O.N. Diagnostics letter, covered in an earlier piece, examined for a software change.</p><p>The same change reappears in risk management. FDA said the risk files did not address fetal-specific hazards, including thermal and acoustic exposure, diagnostic accuracy, misinterpretation, delayed care, and false reassurance, even though the devices were labeled or cleared for fetal use. A design change had altered the intended use, and the risk file had not followed.</p><p>Then the supplier. FDA said the contract manufacturer implemented the labeling or indication change without adequate supporting documentation, the quality agreement did not adequately govern design changes beyond labeling, and the company had no documented site assessment for a supplier its own procedure treated as highest or critical impact. That is a Clause 7.4.1 finding. Finally, returned products reporting performance or reliability failures were not evaluated as complaints, which FDA cited under 21 CFR 820.35(a). Returns and service tickets that meet the definition of a complaint are an input to risk management; if they are not recognized as complaints, they never get there.</p><p>Seen together, Koven’s letter reads as a single event, a product change, that was not controlled at four points: design, risk, supplier, and post-market intake.</p><h3>What the QSR-era letters add</h3><p>There is a third group worth knowing about. Letters to IsoTis OrthoBiologics, Longhorn Vaccines and Diagnostics, and Medline Industries, issued after February 2 but based on inspections that predated the QMSR, allege QSR violations. But they also state that corrective actions must be taken under the QMSR. The practical effect is that a company with an open QSR-era observation should plan its remediation to the ISO 13485 clause structure, not the legacy citation, and expect reinspection under QMSR.</p><h3>A way to test your own chain</h3><p>The letters do not prescribe a remediation plan, and no single document name or method is required. But they suggest a short evidence test that follows the chain instead of auditing each clause in isolation.</p><ol><li><strong>Trace a recent change.</strong> Pick a design, labeling, indication, or process change from the last year. Confirm there is a change record, verification and validation, a risk review, a documented decision on whether a new submission was needed, and notice to any supplier who implements it.</li><li><strong>Walk the signals back to the risk file.</strong> Sample complaints, returns, service tickets, adverse events, rework records, and supplier issues. For each, find where the risk file was reviewed or updated, or a reason it was not.</li><li><strong>Read your quality agreements as a change-control document.</strong> Check who controls changes to design, labeling, and process, how the supplier must notify you, and whether your critical suppliers have a documented site assessment.</li><li><strong>Test the risk procedure for the four things FDA named at Linemaster.</strong> Does it say how activities are performed and documented, who approves them, when records are updated, and how post-market data enters the process?</li><li><strong>Check execution evidence.</strong> Choose one promised control, such as rework, corrective action, environmental monitoring, or software validation, and confirm the raw data, acceptance criteria, approvals, and effectiveness checks exist and agree.</li><li><strong>Re-map open QSR-era observations</strong> to the QMSR clauses, and confirm the corrective actions would stand up under the new framework.</li></ol><h3>Keeping it in proportion</h3><p>A warning letter communicates FDA’s position that it found significant violations and gives the company a chance to respond. Covington’s alert summarizes the allegations without drawing conclusions, and the same caution applies here: nine letters across a quarter do not establish a rate of enforcement. What they do show, consistently, is where inspectors have been looking. The questions FDA asked were not exotic. They were whether the risk file knows what the complaint file knows, whether a design change reaches the people who implement it, and whether a supplier operates under terms that keep the manufacturer in control.</p><p>Quality teams that audit these clauses one at a time can pass each audit and still fail the trace. The teams better prepared for the next inspection are the ones that practice the trace.</p><p><em>Sources: Covington &amp; Burling LLP, Quarterly Medical Device Warning Letters Update: July–September 2026 (November 16, 2026); FDA warning letters to Linemaster Switch Corporation (May 27, 2026; CMS 730215) and Koven Technologies, Inc. (July 21, 2026; CMS 734643); FDA Law Blog, The QMSR is Here: FDA’s First QMSR Warning Letters (August 12, 2026); Pure Global, FDA QMSR Warning Letters Focus on Risk and Design Changes (August 13, 2026); FDA warning letters to IsoTis OrthoBiologics, Longhorn Vaccines and Diagnostics, and Medline Industries (February and March 2026); ISO 13485:2016, clauses 7.1, 7.3.9 and 7.4.</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=93c236010bbd" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[FDA Wrote the Playbook for the Next Outbreak Test.]]></title>
            <link>https://medium.com/@marketing.manager_25154/fda-wrote-the-playbook-for-the-next-outbreak-test-b9a6cb5dda79?source=rss-24c6ec99c02c------2</link>
            <guid isPermaLink="false">https://medium.com/p/b9a6cb5dda79</guid>
            <category><![CDATA[medical-devices]]></category>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[fda]]></category>
            <category><![CDATA[regulatory-compliance]]></category>
            <dc:creator><![CDATA[Lexim AI]]></dc:creator>
            <pubDate>Thu, 19 Nov 2026 19:44:15 GMT</pubDate>
            <atom:updated>2026-11-19T19:44:15.227Z</atom:updated>
            <content:encoded><![CDATA[<h3>FDA Wrote the Playbook for the Next Outbreak Test. The Evidence Bar Is More Specific Than Most Teams Expect.</h3><p>On November 10, CDRH finalized <em>Validation of Certain In Vitro Diagnostic Devices for Emerging Pathogens During a Section 564 Declared Emergency</em> (document GUI00007020; Federal Register notice 91 FR 72541; docket FDA-2024-D-2707). It finalizes a January 2025 draft and ties into a companion template for pre-EUA and EUA submissions. The Federal Register notice says the work answers recommendations, made in reviews of the pandemic response, that FDA lay out in advance how diagnostic tests for emerging pathogens should be validated during a declared emergency.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*W8PqXQ3lEVYgURmgEk-fhg.png" /></figure><p>The guidance is nonbinding and applies only in a narrow circumstance: a public health emergency declaration under section 564 of the FD&amp;C Act. But it is the most concrete statement FDA has published about what evidence it will look for first, when specimens are scarce and time is short. For any company that builds infectious disease diagnostics, it is also a planning document for the weeks before an outbreak, when the work that matters most is done.</p><h3>What it covers, and which standard applies</h3><p>The guidance addresses diagnostic IVDs as defined in 21 CFR 809.3 that detect emerging pathogens to aid diagnosis of serious or life-threatening infectious disease, including tests for screening people without symptoms and home-use tests. It also covers home specimen collection kits and certain software used with diagnostic tests. Serology, antibody, and other adaptive immune response tests are outside its scope.</p><p>It applies to three situations: pre-EUA submissions, EUA requests, and tests offered under an applicable enforcement discretion policy. The timing matters. A company can submit a pre-EUA before a declaration exists, which is the point of the exercise. An EUA request, by contrast, requires a current applicable declaration.</p><p>The legal standard is the other thing to keep in mind. An EUA rests on the statutory “may be effective” standard in section 564(c)(2)(A), which is different from the effectiveness standards for clearance and approval under sections 513 and 515. The guidance’s recommendations are calibrated to that lower, emergency-specific bar, and FDA says it has focused them on the most important and least burdensome information to collect during a declared emergency. Evidence that would satisfy a 510(k) is not the target here, and evidence collected for an EUA does not automatically carry over to a later marketing submission.</p><h3>Where the specimens come from</h3><p>Clinical performance is where outbreak testing is hardest, because the infected patients and the reference methods you need are least available when you need them most. The guidance sets out a hierarchy. FDA prefers natural clinical specimens from a prospective study. Early in an outbreak, archived specimens may be a reasonable substitute. If neither is available, FDA may accept contrived samples, and synthetic material may be acceptable when pathogen stocks are not available. If FDA authorizes a test on contrived data, it would likely expect a follow-up study using natural specimens once that becomes feasible.</p><p>For sizing, the guidance gives a typical example rather than a requirement: for an established, high-sensitivity viral nucleic acid test where clinical specimens and comparators are generally available, at least 30 positive and 30 negative specimens. Results are generally reported as positive percent agreement (PPA) and negative percent agreement (NPA) against a highly sensitive comparator, with 95% Wilson score confidence intervals.</p><p>Two terms are worth unpacking. PPA and NPA are used instead of sensitivity and specificity because the comparator is itself imperfect, so the numbers describe agreement with the comparator, not the truth. And the Wilson score interval is a way of expressing uncertainty that behaves better than the simple textbook formula when samples are small or results sit near 0% or 100%, which is exactly the situation in an early outbreak. The practical point is that a 30-positive study produces wide intervals, so FDA is asking for transparency about the uncertainty, not for a precision that cannot exist yet.</p><p>For studies that involve specimens from people, the guidance notes that an IRB may waive or alter informed consent for some minimal-risk investigations under 21 CFR 50.22, and separately points to FDA’s policy on certain leftover specimens that cannot be individually identified.</p><h3>Analytical validation: the numbers</h3><p>The analytical recommendations are more prescriptive, and several are precise enough to build a protocol from.</p><p><strong>Limit of detection (LoD).</strong> Preliminary range-finding with 2- to 3-fold dilutions (3 replicates per level), then at least 20 replicates at the preliminary LoD. The confirmed LoD is the lowest concentration at which at least 95% of replicates are positive, for example 19 of 20. FDA advises against simulated matrix or recombinant antigen for LoD studies.</p><p><strong>Inclusivity and variants.</strong> In silico alignment (100% nucleotide identity for molecular tests, or a risk assessment where there are mismatches with a significant number of published sequences), with wet testing for circulating variants when available. Re-evaluate monthly for clinically significant mutations, and run a risk analysis for any greater than 3-fold reduction in analytical sensitivity, including whether the combined effect pushes clinical performance below FDA’s recommendations.</p><p><strong>Specimen stability.</strong> If applicable CDC specifications cover your storage and transport conditions, additional data are likely unnecessary. Otherwise, test live or inactivated pathogen spiked into negative clinical matrix around the LoD, plus at least 10 negative specimens.</p><p>The LoD definition deserves a second look. Twenty replicates with 19 positive gives a point estimate of 95%, but with so few replicates the true detection rate consistent with that result spans a wide range, roughly three-quarters to nearly 100% at 95% confidence. The LoD is an operational definition that lets two studies be compared, not a proof that the test detects 95% of samples at that level. That is one reason the guidance pairs it with variant monitoring and a later natural-specimen study.</p><p>The monthly inclusivity recommendation is the newest operational burden. A pathogen that mutates can quietly erode a test’s sensitivity after authorization, and FDA’s answer is a standing, scheduled review with a defined trigger (the 3-fold threshold) rather than an ad hoc response.</p><h3>Change protocols replace the draft’s PCCP language</h3><p>The draft used the term PCCP, but the final guidance renamed the concept a Change Protocol, because predetermined change control plans apply to marketing submissions and not to EUAs. The idea is the same: a manufacturer can propose, in the initial EUA request, a protocol for future changes within the test’s intended use, such as adding an instrument or a collection device. A change made under a protocol that FDA reviewed and included in the authorization is considered covered by that authorization, and FDA generally specifies implementation requirements in a condition of authorization. In an outbreak, when an authorized test needs a second instrument platform quickly, having the validation plan agreed in advance can save weeks.</p><h3>Home use and point-of-care testing</h3><p>For home-collection kits, over-the-counter tests, and point-of-care tests intended for CLIA-waived settings, the guidance adds human-factors expectations. It recommends quick reference instructions limited to one or two pages, written at no higher than a 7th-grade reading level, with pictures and diagrams, supported by usability, comprehension, and hazard-based flex studies.</p><p>For point-of-care tests, it describes a near-cutoff evaluation at one or two U.S. sites with four to six operators who have no laboratory training and represent the intended users. Each site tests 10 contrived specimens near the LoD (for example, within 3-fold) and 10 negative specimens, each operator tests at least 3 positives and 3 negatives, and the target is at least 95% positive and negative agreement, or a re-evaluation of the LoD.</p><h3>What to do before the next declaration</h3><p>Because the guidance is only useful if the groundwork exists before an emergency, the practical steps are about preparation:</p><ul><li><strong>Write the validation plan now,</strong> using the specimen hierarchy: what you would collect first, what you would substitute, and what the follow-up natural-specimen study would look like.</li><li><strong>Identify a sensitive comparator</strong> with performance data for your intended specimen type, since PPA and NPA depend on it.</li><li><strong>Decide which foreseeable changes belong in a Change Protocol,</strong> such as additional instruments, collection devices, or sample types, and document the validation and acceptance criteria for each.</li><li><strong>Assign an owner for monthly variant review,</strong> with the 3-fold sensitivity trigger written into a procedure.</li><li><strong>Use the pre-EUA template</strong> to test the plan with FDA before a declaration, and confirm what is covered by a general template versus what needs a sponsor-specific discussion.</li><li><strong>For home and point-of-care products,</strong> plan the instructions and the lay-operator study early, because human-factors work is slow to improvise.</li></ul><p>The guidance will not answer every question about a future outbreak, and its recommendations are explicitly flexible across outbreak stages. But it replaces a good deal of guesswork with a checklist, and the companies that have their plan in place before the declaration will be the ones that submit first.</p><p><em>Sources: FDA, Validation of Certain In Vitro Diagnostic Devices for Emerging Pathogens During a Section 564 Declared Emergency, final guidance GUI00007020 (November 10, 2026); Federal Register, notice of availability, 91 FR 72541 (docket FDA-2024-D-2707); FDA, General IVD EUA/Pre-EUA Template; Pure Global, FDA Finalizes Section 564 Emergency IVD Validation Guidance (November 15, 2026), for section locators and the comparison with the January 2025 draft.</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=b9a6cb5dda79" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[FDA Said It Needed Overall Survival.]]></title>
            <link>https://medium.com/@marketing.manager_25154/fda-said-it-needed-overall-survival-ae137658a0d3?source=rss-24c6ec99c02c------2</link>
            <guid isPermaLink="false">https://medium.com/p/ae137658a0d3</guid>
            <category><![CDATA[fda]]></category>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[medtech]]></category>
            <category><![CDATA[pharmaceutical]]></category>
            <dc:creator><![CDATA[Lexim AI]]></dc:creator>
            <pubDate>Thu, 19 Nov 2026 19:43:25 GMT</pubDate>
            <atom:updated>2026-11-19T19:43:25.403Z</atom:updated>
            <content:encoded><![CDATA[<h3>FDA Said It Needed Overall Survival. Then It Said No. What the Ivonescimab Complete Response Letter Tells Us.</h3><p>On Sunday, November 15, one day after its November 14 goal date, FDA sent Summit Therapeutics a complete response letter (CRL) for ivonescimab plus platinum-containing chemotherapy in EGFR-mutated, locally advanced or metastatic non-small cell lung cancer after prior EGFR tyrosine kinase inhibitor therapy. Summit disclosed it that morning and filed an 8-K the next day. The application cannot be approved in its present form.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*TX5GfiiL6pp_Eu3bXygBdA.png" /></figure><p>Two weeks ago, in anticipation of this decision, we laid out what each possible outcome would teach other sponsors. This is the outcome in which FDA’s written position held, and the reasoning in the letter, as Summit has relayed it, is worth reading closely because it is narrower and more specific than the headline suggests.</p><h3>What FDA said</h3><p>According to Summit’s release, FDA wrote that it cannot approve the application because substantial evidence of effectiveness is not established in the absence of a statistically significant improvement in overall survival (OS). It also said that a progression-free survival (PFS) benefit, without a statistically significant OS improvement, does not support a positive benefit-risk profile for this application.</p><p>Those are two different arguments, and the order matters. The first is about the legal evidentiary standard. For a biologics license, FDA must find substantial evidence of effectiveness from adequate and well-controlled investigations, and the letter says that standard is not met without a significant OS result in this setting. The second is a benefit-risk statement: a gain in PFS, on its own, does not outweigh the risks. In our earlier piece we described a possible path in which a near-miss on OS could be weighed within a benefit-risk judgment. The letter, as described, forecloses that path for this application by resting on the effectiveness standard first.</p><p>The CRL applies to this BLA, not to ivonescimab’s wider development program. Summit’s release said no FDA-approved regimen in this setting has demonstrated a statistically significant OS benefit, and presented that as context for the unmet need. FDA concluded that the context did not make HARMONi’s PFS result sufficient. Summit’s release did not describe what additional evidence FDA requested, a resubmission plan, or timing.</p><h3>The numbers FDA was looking at</h3><p>HARMONi randomized 438 patients 1:1 and had two primary endpoints: PFS by independent review and OS. The PFS result was not close to the margin. Median PFS was 6.8 versus 4.4 months, with a hazard ratio of 0.52 (95% confidence interval 0.41 to 0.66). OS was the other primary endpoint. Median OS was 16.8 versus 14.0 months, a hazard ratio of 0.79 (95% CI 0.62 to 1.01; p=0.057), short of the pre-specified threshold.</p><p>The safety profile adds context for the benefit-risk statement. Grade 3 or higher treatment-related adverse events were 50.0% with ivonescimab versus 42.2% with chemotherapy alone, and serious treatment-related adverse events were 28.0% versus 15.1%. Summit’s description of the letter does not attribute the CRL to safety, but the safety data are part of the balance that a PFS-only argument has to carry.</p><p>Summit pointed to an updated analysis with a June 2026 data cutoff, reporting a hazard ratio of 0.76 in the full population and in both Asian and Western patients. It described this as a post hoc analysis from longer follow-up, and its release did not give confidence intervals or other statistical detail for it. In the primary analysis, the hazard ratio in Western patients had been about 0.98, when their median follow-up was 9.2 months, compared with 32.7 months for Asian patients. In the updated analysis, Western median follow-up reached 23.2 months.</p><h3>Why a later analysis could not rescue the filing</h3><p>It is tempting to read the 0.98 to 0.76 movement as proof that the early number was an artifact of immature data, and that may be true. But a regulator evaluating a pre-specified efficacy endpoint is bound to the analysis the trial was designed around. Three features of the updated analysis limit what it can do.</p><ul><li><strong>It was not the pre-specified test.</strong> The primary OS analysis carried the statistical plan, including how the type I error rate was controlled. A later cut, described as post hoc, does not inherit that control, and looking at a trial repeatedly until a boundary is crossed inflates the chance of a false positive.</li><li><strong>Maturity cuts both ways.</strong> Longer follow-up can reveal a real late separation of the curves, but it can also change which patients and which subsequent therapies drive the result. Without confidence intervals or the full analysis plan, a reviewer cannot tell which is happening.</li><li><strong>A hazard ratio is an average.</strong> If the benefit appears late, the hazard ratio can mask periods when the curves overlap. FDA’s August 2025 draft guidance on assessing OS in oncology trials also flags protocol-specified crossover as something that can complicate interpretation of OS, which is part of why OS is treated with caution in designs that allow it.</li></ul><p>None of this means the updated result is meaningless. It may support further study, and Summit is continuing development. It does mean that it was never going to substitute for a significant primary result, and FDA’s letter, as relayed, treats it that way.</p><h3>Reading the decision against FDA’s other signals</h3><p>FDA’s August 2025 draft guidance, <em>Approaches to Assessment of Overall Survival in Oncology Clinical Trials</em> (docket FDA-2024-D-5850), recommends that randomized oncology trials be designed to assess OS so that potential harm can be evaluated, and emphasizes OS as a pre-specified safety endpoint when it is not the primary endpoint. That guidance sits in the background of this decision, and it points in a consistent direction: FDA wants OS data to carry real weight, even when PFS looks strong.</p><p>HARMONi made OS a primary endpoint, which raised the stakes in one direction. Read as a safety check, the OS result is reassuring, since the point estimate favors ivonescimab and the interval’s upper bound sits at 1.01. Read as a test of the effectiveness requirement FDA had stated to the sponsor in advance, it fell short. The CRL resolves which reading governs for this application. It does not claim that ivonescimab lacks activity, and, as relayed, FDA did not dispute that PFS improved.</p><h3>What happens next</h3><p>Summit can request a Type A meeting to discuss the letter, which FDA generally aims to hold within about 30 days of a request. A resubmission would be classified as either Class 1 (a two-month review goal) or Class 2 (six months), and the content FDA specifies in the letter determines which applies. FDA has published complete response letters since September 2025, so the full text may become available and would answer the question of what evidence FDA says could change its view.</p><p>Separately, Summit and Akeso are running HARMONi-3, comparing ivonescimab plus chemotherapy with pembrolizumab plus chemotherapy in first-line NSCLC. Initial results in the squamous cohort could come by the end of this year, with non-squamous results expected in the first half of 2027. Those trials have different designs and endpoints, and they will be judged on their own data.</p><h3>What other sponsors should take from it</h3><ul><li><strong>A requirement FDA states in writing is the bar.</strong> FDA told Summit that a statistically significant OS improvement was necessary. When the trial missed narrowly, the stated requirement was applied as written. Plan on that basis, and if your results may fall short, raise it with FDA before filing, not after.</li><li><strong>Design for the OS question from the beginning.</strong> When OS is a primary or key endpoint, the number of events, follow-up duration, crossover rules, and regional enrollment pace all determine whether the analysis will be mature enough when you need it.</li><li><strong>Regional maturity is a design choice.</strong> HARMONi enrolled roughly 62% of its patients in Asia and 38% elsewhere, and the two groups matured at very different speeds. A multiregional trial should plan for how the population most relevant to U.S. practice will be followed.</li><li><strong>Unmet need does not lower the evidentiary standard by itself.</strong> Summit’s argument that no regimen in this setting has shown an OS benefit was acknowledged and not accepted as a substitute.</li><li><strong>Treat updated analyses as supportive, not curative.</strong> Share them with FDA, but build the filing on the pre-specified analysis.</li></ul><p>The decision also sends a message to a broader field. Several PD-1 and VEGF bispecific programs are in development, none is approved in the United States, and their sponsors will read this letter closely when they plan registrational trials. What they will take from it is less about the drug class than about the endpoint: where a trial names OS as a primary endpoint, a strong PFS result is not an alternative to it.</p><p><em>Sources: Summit Therapeutics, press release (November 15, 2026) and Form 8-K (filed November 16, 2026); Healio, FDA rejects ivonescimab (November 16, 2026); OncLive and CancerNetwork coverage of the complete response letter (November 16, 2026); BioPharm International (November 16, 2026); Fierce Biotech and pharmaphorum coverage (November 2026); HARMONi primary analysis, Lancet Oncology 2026;27(9):1094–1109; updated HARMONi OS analysis, WCLC 2026 (Passaro et al.); FDA draft guidance, Approaches to Assessment of Overall Survival in Oncology Clinical Trials (August 2025; docket FDA-2024-D-5850); FDA CDER annual report on publication of complete response letters.</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=ae137658a0d3" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Can AI Predict FDA Feedback Before You Submit?]]></title>
            <link>https://medium.com/@marketing.manager_25154/can-ai-predict-fda-feedback-before-you-submit-4b17a3bde5a5?source=rss-24c6ec99c02c------2</link>
            <guid isPermaLink="false">https://medium.com/p/4b17a3bde5a5</guid>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[medical-devices]]></category>
            <category><![CDATA[regulatory-compliance]]></category>
            <category><![CDATA[pharmaceutical]]></category>
            <dc:creator><![CDATA[Lexim AI]]></dc:creator>
            <pubDate>Wed, 18 Nov 2026 23:19:32 GMT</pubDate>
            <atom:updated>2026-11-18T23:19:32.079Z</atom:updated>
            <content:encoded><![CDATA[<p><strong>AI and Regulatory Affairs</strong></p><p>In the world of pharmaceutical and medical device development, FDA approval is more than just a regulatory step — it is a gatekeeper to market access, clinical adoption, and financial sustainability. A single delay in approval can result in thousands of dollars per day in lost revenue, jeopardize funding milestones, and hinder patient access to critical innovations. Despite these stakes, regulatory strategy remains one of the least digitized and most manual components of the product development lifecycle.</p><p>While advances in areas like real-world data, clinical trial optimization, and digital health monitoring have reshaped how life sciences companies operate, regulatory affairs is still largely driven by manual document review, historical precedent, and consultant-driven strategy. For many companies, this means combing through FDA guidance documents, prior approval letters, enforcement actions, and internal documentation — often relying on institutional memory or outdated playbooks. This approach is not only slow but increasingly misaligned with the pace and complexity of today’s innovation.</p><p><strong>FDA Feedback is Unpredictable (Until Now!)</strong></p><p>Traditionally, the industry has turned to regulatory intelligence to improve this process. This field focuses on analyzing publicly available regulatory data — such as warning letters, guidance updates, advisory committee transcripts, and inspection trends — to anticipate how the FDA might respond to a new submission. However, the process is labor-intensive and difficult to scale. Large pharmaceutical companies can afford to dedicate specialized teams to this work or hire consultants to produce strategic assessments, but smaller firms are often left with limited insight and high uncertainty.</p><p>In recent years, artificial intelligence — particularly large language models (LLMs) — has opened the door to a fundamentally different approach. LLMs are AI systems trained on vast amounts of textual, image, video, and data, enabling them to understand, interpret, and generate new information with a high degree of accuracy. When applied to regulatory content, these models can analyze complex guidance documents, compare language across previous versions, analyze submissions, and simulate the likely response from regulatory reviewers based on historical patterns.</p><p>This capability is more than theoretical. Research published in 2025 demonstrated that LLMs trained on labeled clinical and regulatory text could accurately classify document types and flag inconsistencies. Another study out of Stanford showed that transformer-based models could perform question-answering and classification tasks on unstructured clinical notes with performance close to domain experts (Sun et al., 2025). These findings suggest that regulatory language — while complex — is highly amenable to machine interpretation, especially when models are trained on the right corpus of structured and unstructured data.</p><p><a href="http://data.ai/"><strong>AI</strong></a><strong> to the Rescue</strong></p><p>One of the most promising applications of this technology is predicting regulatory feedback. In practice, this means taking a draft submission — such as a 510(k) notification, IND, or clinical study protocol — and running it through an AI model trained on prior FDA interactions. The model can flag areas of risk, identify language that has historically triggered questions or rejections, and suggest revisions that better align with agency precedent. In effect, the model serves as a proxy reviewer, helping teams strengthen their submissions before engaging with the FDA.</p><p>This can offer significant strategic and financial benefits. A 2025 Deloitte report on regulatory trends found that delays in approval were one of the top contributors to underperformance in new product launches. Delays often stem not from inadequate science but from unclear or inconsistent submissions that fail to anticipate regulator concerns. Even a single unexpected request for additional information can push timelines back by months, with cascading effects on trials, commercial plans, and investor confidence. If AI can help anticipate and prevent even a fraction of these issues, the return on investment is substantial.</p><p>At the same time, it’s important to note the limitations. AI is not a replacement for regulatory experts, nor can it fully replicate the discretion and context that FDA reviewers bring to each decision. Rather, it functions as an augmentation layer — providing faster, broader pattern recognition, and surfacing insights that might take human reviewers hours or days to uncover. It also allows teams to work more iteratively, reviewing documents continuously during development rather than waiting for a late-stage review to uncover major issues.</p><p>The potential impact of AI in regulatory strategy is even more significant as the FDA itself explores more data-driven approaches. Through initiatives such as the Digital Health Center of Excellence and efforts to modernize submission formats (e.g., electronic common technical documents, or eCTDs, and eSTAR), the agency is signaling a shift toward structured, searchable, and machine-readable regulatory interactions. As that infrastructure evolves, companies that already leverage AI internally will be better prepared to align with the FDA’s digital expectations.</p><p><a href="https://lexim.ai/"><strong>Lexim AI </strong></a><strong>Can Help</strong></p><p>For organizations looking to operationalize these capabilities, Lexim AI offers a practical path forward. Built specifically for the life sciences, Lexim’s products are trained on regulatory data to analyze draft submissions in real-time. By reviewing language, structure, and precedent, it can flag potential red flags, suggest improvements, and help teams better align with likely FDA expectations. Lexim, through the use of AI, enhances productivity and provides insights to regulatory professionals, reducing the time spent on manual document reviews, eliminating blind spots, and enabling faster and more confident decision-making.</p><p>In an industry where the cost of delay is measured in both dollars and patient outcomes, the ability to preempt regulatory feedback is a competitive advantage. Applied thoughtfully, AI transforms regulatory strategy from a reactive compliance task into a proactive engine for faster approvals. As tools like Lexim gain traction, they may not just improve the quality of submissions — but reshape how companies approach regulatory thinking altogether.</p><p>To learn more or get help addressing your needs, please contact <a href="https://www.lexim.ai/">Lexim AI</a>.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=4b17a3bde5a5" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>