<?xml version="1.0" encoding="UTF-8"?><?xml-stylesheet href="/en/rss-styles.xsl" type="text/xsl"?><rss version="2.0" xmlns:content="http://purl.org/rss/1.0/modules/content/" xmlns:atom="http://www.w3.org/2005/Atom"><channel><title>Hospital AI Lab</title><description>Columns on the regulation of medical AI in the United States: HIPAA, FDA device software, information blocking and new federal rules.</description><link>https://hospital-ai-lab.com/en/</link><language>en</language><atom:link href="https://hospital-ai-lab.com/en/rss.xml" rel="self" type="application/rss+xml"/><lastBuildDate>Tue, 17 Nov 2026 00:00:00 GMT</lastBuildDate><item><title>AI Pilot to Hospital-Wide Use: Five Questions Before Scaling</title><link>https://hospital-ai-lab.com/en/blog/2026-11-17-ai-pilot-to-scale-survey-hospital-questions/</link><guid isPermaLink="true">https://hospital-ai-lab.com/en/blog/2026-11-17-ai-pilot-to-scale-survey-hospital-questions/</guid><description>A Google Cloud survey says few healthcare leaders have scaled advanced AI. Questions administrators and nurse leaders can ask before a pilot grows.</description><pubDate>Tue, 17 Nov 2026 00:00:00 GMT</pubDate><content:encoded>&lt;p&gt;A pilot that works on one unit does not always work across the whole hospital. Healthcare Dive reported on November 16, 2026, on a Google Cloud report about why healthcare organizations struggle to move AI from pilot to everyday use. The numbers come from a vendor-published survey, so read them as one view, not a measurement of the whole industry. Here is what was reported and how administrators and nursing leaders can use it.&lt;/p&gt;
&lt;h2 id=&quot;what-the-survey-reported&quot;&gt;What the survey reported&lt;/h2&gt;
&lt;p&gt;According to Healthcare Dive, Google’s survey covered more than 250 healthcare executives and was run by the consumer insights firm Hypothesis Group. The article does not say when it was fielded. Reported results include:&lt;/p&gt;
&lt;ul&gt;
&lt;li&gt;Nearly 60% faced integration and deployment challenges moving enterprise AI from pilot to live systems.&lt;/li&gt;
&lt;li&gt;40% cited unpredictable costs as a reason for pausing AI projects.&lt;/li&gt;
&lt;li&gt;Only 30% said they had scaled advanced or agentic AI across operations, which Healthcare Dive says is 13 percentage points below the all-industry benchmark.&lt;/li&gt;
&lt;li&gt;More than 80% named cost as a hurdle to relying less on existing AI vendors.&lt;/li&gt;
&lt;li&gt;About 40% said protecting sensitive data is an especially important AI consideration.&lt;/li&gt;
&lt;/ul&gt;
&lt;p&gt;Respondents also said they want full visibility into how an AI tool is governed and the ability to customize it. We did not open the original report, so these figures come from the news article.&lt;/p&gt;
&lt;h2 id=&quot;why-pilots-stall&quot;&gt;Why pilots stall&lt;/h2&gt;
&lt;p&gt;The barriers named are practical, not scientific: connecting the tool to existing systems, costs that change as use grows, more technical complexity, and concern for sensitive data. None of these show up in a demo, which is why they surprise teams after the pilot succeeds.&lt;/p&gt;
&lt;h2 id=&quot;five-questions-to-ask-before-scaling&quot;&gt;Five questions to ask before scaling&lt;/h2&gt;
&lt;p&gt;These are our suggestions, not the survey’s, and they are not legal advice.&lt;/p&gt;
&lt;ol&gt;
&lt;li&gt;&lt;strong&gt;What changes in the workflow?&lt;/strong&gt; Ask nurses on the pilot unit what they stopped doing, started doing, or worked around.&lt;/li&gt;
&lt;li&gt;&lt;strong&gt;What will it cost at ten times the use?&lt;/strong&gt; Ask the vendor how pricing changes with volume, and get it in writing.&lt;/li&gt;
&lt;li&gt;&lt;strong&gt;How does it connect to our systems, and who maintains that link?&lt;/strong&gt;&lt;/li&gt;
&lt;li&gt;&lt;strong&gt;Where does patient data go?&lt;/strong&gt; Confirm with your privacy officer what leaves the hospital and who can see it.&lt;/li&gt;
&lt;li&gt;&lt;strong&gt;How would we switch vendors?&lt;/strong&gt; The survey’s cost-of-switching finding suggests asking this before signing, not after.&lt;/li&gt;
&lt;/ol&gt;
&lt;h2 id=&quot;key-takeaways&quot;&gt;Key takeaways&lt;/h2&gt;
&lt;ul&gt;
&lt;li&gt;A Google Cloud survey, as reported on November 16, 2026, found about 60% hit integration challenges and 30% had scaled advanced AI.&lt;/li&gt;
&lt;li&gt;It is a vendor-run survey of more than 250 executives; the original report was not reviewed here.&lt;/li&gt;
&lt;li&gt;Before scaling a pilot, check workflow changes, cost growth, system connections, data flow, and exit options. This is general information, not legal advice.&lt;/li&gt;
&lt;/ul&gt;
&lt;p&gt;Source: &lt;a href=&quot;https://www.healthcaredive.com/news/healthcare-struggles-to-scale-ai-past-pilot-stage-google-report/874609/&quot;&gt;Healthcare Dive, “Healthcare struggles to scale AI past pilot stage: Google report,” November 16, 2026&lt;/a&gt;. Read on November 17, 2026.&lt;/p&gt;
&lt;hr&gt;
&lt;p&gt;&lt;em&gt;This site’s content is for general information, research, and education. It does not replace medical diagnosis or advice for any specific patient, and it is not legal advice.&lt;/em&gt;&lt;/p&gt;</content:encoded><category>AI Tools</category></item><item><title>Patients Accept AI Far More When a Clinician Can Override It</title><link>https://hospital-ai-lab.com/en/blog/2026-11-16-gallup-patients-ai-comfort-clinician-override/</link><guid isPermaLink="true">https://hospital-ai-lab.com/en/blog/2026-11-16-gallup-patients-ai-comfort-clinician-override/</guid><description>Gallup finds 56% of patients are comfortable with AI diagnosis if a provider can override it, but only 9% without review. What teams can do.</description><pubDate>Mon, 16 Nov 2026 00:00:00 GMT</pubDate><content:encoded>&lt;p&gt;When a vendor says its tool “makes the call,” the next question from a nurse leader is usually, “Who can overrule it?” A new national survey suggests patients ask the same thing. The West Health-Gallup Center on Healthcare in America published a report on November 9, 2026, on how comfortable U.S. patients are with AI in their care, and how well physicians guess that comfort. Here is what it found, in plain terms, and what hospital administrators and nursing leaders can take from it.&lt;/p&gt;
&lt;h2 id=&quot;what-was-surveyed&quot;&gt;What was surveyed&lt;/h2&gt;
&lt;p&gt;Gallup surveyed 5,330 U.S. adults who had seen a primary care provider in the past year, using its web panel from August 12 to 24, 2026. The margin of sampling error is plus or minus 1.7 percentage points. A separate opt-in web survey reached 1,007 licensed, practicing primary care physicians from August 7 to 23, 2026. Because that physician survey is not a probability sample, its numbers describe the people who answered, not all physicians.&lt;/p&gt;
&lt;h2 id=&quot;the-key-number-who-can-say-no&quot;&gt;The key number: who can say no&lt;/h2&gt;
&lt;p&gt;Gallup asked patients how comfortable they would be if AI made decisions, with and without a provider able to override them:&lt;/p&gt;

























&lt;table&gt;&lt;thead&gt;&lt;tr&gt;&lt;th&gt;Use of AI&lt;/th&gt;&lt;th&gt;Provider can override&lt;/th&gt;&lt;th&gt;No provider review&lt;/th&gt;&lt;/tr&gt;&lt;/thead&gt;&lt;tbody&gt;&lt;tr&gt;&lt;td&gt;Diagnosis and treatment&lt;/td&gt;&lt;td&gt;56%&lt;/td&gt;&lt;td&gt;9%&lt;/td&gt;&lt;/tr&gt;&lt;tr&gt;&lt;td&gt;Imaging review&lt;/td&gt;&lt;td&gt;53%&lt;/td&gt;&lt;td&gt;13%&lt;/td&gt;&lt;/tr&gt;&lt;tr&gt;&lt;td&gt;Medication decisions&lt;/td&gt;&lt;td&gt;49%&lt;/td&gt;&lt;td&gt;11%&lt;/td&gt;&lt;/tr&gt;&lt;/tbody&gt;&lt;/table&gt;
&lt;p&gt;When AI only assists and the provider decides, the report gives 38% for diagnosis and treatment, 52% for imaging review, and 45% for medication decisions. In other words, comfort is tied less to how smart the tool is and more to whether a person stays in charge.&lt;/p&gt;
&lt;h2 id=&quot;physicians-guessed-too-low&quot;&gt;Physicians guessed too low&lt;/h2&gt;
&lt;p&gt;Physicians underestimated patient comfort in the “provider can override” setting. For diagnosis and treatment the gap was 20 points. For imaging review, physicians estimated 42% against 53% for patients, and for medication decisions 42% against 49%. When AI only assists, physician estimates were within two points of patients for diagnosis and medication, though they were 11 points low on imaging review.&lt;/p&gt;
&lt;p&gt;Asked what would make them more comfortable with AI acting on its own, no single factor won a majority. The most helpful was the ability to request provider review, chosen by just under half. Privacy protections, evidence of safety and effectiveness, lower costs and faster decisions each drew no more than a third.&lt;/p&gt;
&lt;h2 id=&quot;what-hospital-teams-can-do-with-this&quot;&gt;What hospital teams can do with this&lt;/h2&gt;
&lt;p&gt;Gallup’s own suggestions are to ask patients directly about their preferences and to find out what training and support would help providers use AI well. It also reports that 89% of providers had used AI for at least one work task in the past year, while only 20% had received formal AI training for health care. The steps below are our suggestions, not Gallup’s.&lt;/p&gt;
&lt;ol&gt;
&lt;li&gt;&lt;strong&gt;Write the override into the workflow.&lt;/strong&gt; For any AI that suggests or flags something, state who reviews it, how they can overrule it, and where that is recorded. Nurses should know the path before the first alert.&lt;/li&gt;
&lt;li&gt;&lt;strong&gt;Make the review request easy.&lt;/strong&gt; If patients value being able to ask for a provider’s review, tell front-line staff how to say so and how to route the request.&lt;/li&gt;
&lt;li&gt;&lt;strong&gt;Ask patients.&lt;/strong&gt; A short question at intake or in a patient advisory group can show how your community feels, instead of relying on staff guesses.&lt;/li&gt;
&lt;li&gt;&lt;strong&gt;Plan training.&lt;/strong&gt; Pair each new tool with a short training plan, since formal training is rare in the survey.&lt;/li&gt;
&lt;/ol&gt;
&lt;h2 id=&quot;key-takeaways&quot;&gt;Key takeaways&lt;/h2&gt;
&lt;ul&gt;
&lt;li&gt;Gallup (November 9, 2026): 56% of patients are comfortable with AI diagnosis and treatment if a provider can override it; 9% if no provider reviews it.&lt;/li&gt;
&lt;li&gt;Physicians underestimated patient comfort by 20 points in the override setting, and the physician survey is an opt-in sample.&lt;/li&gt;
&lt;li&gt;Spell out who can override, how requests for review are handled, and what training staff receive. This is general information, not legal advice.&lt;/li&gt;
&lt;/ul&gt;
&lt;p&gt;Source: &lt;a href=&quot;https://news.gallup.com/poll/716261/providers-underestimate-patient-comfort-care.aspx&quot;&gt;Gallup / West Health-Gallup Center, “Providers Underestimate Patient Comfort With AI in Care,” November 9, 2026&lt;/a&gt;. Read on November 16, 2026.&lt;/p&gt;
&lt;hr&gt;
&lt;p&gt;&lt;em&gt;This site’s content is for general information, research, and education. It does not replace medical diagnosis or advice for any specific patient, and it is not legal advice.&lt;/em&gt;&lt;/p&gt;</content:encoded><category>Research Review</category></item><item><title>ML Heart Alert Software: What the FDA Class II Rule Means</title><link>https://hospital-ai-lab.com/en/blog/2026-11-15-ml-heart-alert-software-class-ii-nurses-admins/</link><guid isPermaLink="true">https://hospital-ai-lab.com/en/blog/2026-11-15-ml-heart-alert-software-class-ii-nurses-admins/</guid><description>FDA now has a class II category for machine-learning heart alert software. Questions nurses and administrators can ask before go-live.</description><pubDate>Sun, 15 Nov 2026 00:00:00 GMT</pubDate><content:encoded>&lt;p&gt;An alert that says a patient may have a heart condition changes what a nurse does next. Before software like that reaches a unit, hospital leaders need to know what the FDA has actually required of it. A rule published in the Federal Register on September 11, 2026 puts one kind of this software into a defined category. This post summarizes the rule in plain language. It is general information, not legal advice.&lt;/p&gt;
&lt;h2 id=&quot;what-the-rule-does&quot;&gt;What the rule does&lt;/h2&gt;
&lt;p&gt;According to the &lt;a href=&quot;https://www.federalregister.gov/documents/full_text/text/2026/09/11/2026-18612.txt&quot;&gt;Federal Register notice&lt;/a&gt; (document 2026-18612, effective September 11, 2026), FDA added 21 CFR 870.2380, “cardiovascular machine learning-based notification software,” as a class II device with special controls. The rule describes software that uses machine learning to estimate how likely a cardiovascular disease or condition is, to support referral or diagnostic follow-up. It flags one condition from one or more non-invasive physiological inputs during routine care. It is not meant to give diagnostic-quality output or to detect arrhythmias. The notice also says the device type is not exempt from premarket notification (510(k)).&lt;/p&gt;
&lt;h2 id=&quot;what-the-special-controls-ask-of-the-maker&quot;&gt;What the special controls ask of the maker&lt;/h2&gt;
&lt;p&gt;The notice lists four groups of special controls:&lt;/p&gt;
&lt;ul&gt;
&lt;li&gt;&lt;strong&gt;Clinical performance testing&lt;/strong&gt; on real-world data from a representative population, separate from the training data, including results by subgroup and testing at three or more geographically diverse sites.&lt;/li&gt;
&lt;li&gt;&lt;strong&gt;Software verification and hazard analysis&lt;/strong&gt;, including how sensor hardware affects performance and how poor input quality or user error is handled.&lt;/li&gt;
&lt;li&gt;&lt;strong&gt;Human factors assessment&lt;/strong&gt; of the risk that intended users misread the output.&lt;/li&gt;
&lt;li&gt;&lt;strong&gt;Labeling&lt;/strong&gt; that states known limits, interpretation guidance, and compatible sensor data.&lt;/li&gt;
&lt;/ul&gt;
&lt;p&gt;Two labeling points matter on a ward. The label must warn users not to rely on a negative result to rule out follow-up, and must say the output does not replace a full clinical evaluation.&lt;/p&gt;
&lt;h2 id=&quot;questions-a-hospital-team-can-ask&quot;&gt;Questions a hospital team can ask&lt;/h2&gt;
&lt;p&gt;These come straight from the points above:&lt;/p&gt;
&lt;ul&gt;
&lt;li&gt;Was performance tested on patients like ours, and at sites other than where the model was built?&lt;/li&gt;
&lt;li&gt;What happens when the sensor signal is poor? Who sees that, and what does the alert look like?&lt;/li&gt;
&lt;li&gt;When the software stays silent, how do we make sure staff still follow our usual assessment?&lt;/li&gt;
&lt;li&gt;Who is expected to act on an alert, and how fast?&lt;/li&gt;
&lt;/ul&gt;
&lt;p&gt;Answers to the last two belong in your own workflow plan, not the vendor’s brochure. Confirm details with your compliance staff and clinical leaders.&lt;/p&gt;
&lt;h2 id=&quot;key-takeaways&quot;&gt;Key takeaways&lt;/h2&gt;
&lt;ul&gt;
&lt;li&gt;FDA classified machine learning-based cardiovascular notification software as class II (21 CFR 870.2380), effective September 11, 2026.&lt;/li&gt;
&lt;li&gt;The software suggests a condition may be present so a clinician can follow up. It is not meant to give diagnostic-quality output.&lt;/li&gt;
&lt;li&gt;A negative result should not be used to rule out follow-up, according to the required labeling.&lt;/li&gt;
&lt;li&gt;“Cleared” does not mean “right for our patients”: ask about subgroup results, sensor quality, and who acts on an alert.&lt;/li&gt;
&lt;/ul&gt;
&lt;hr&gt;
&lt;p&gt;&lt;em&gt;This site provides general information for research and education. It is not medical advice and does not replace diagnosis or guidance for any specific patient. It is not legal advice.&lt;/em&gt;&lt;/p&gt;</content:encoded><category>News</category></item><item><title>FDA Rule on MCI Prognosis Software: What Hospitals Check</title><link>https://hospital-ai-lab.com/en/blog/2026-11-14-fda-mci-prognosis-software-class-ii-what-hospitals-check/</link><guid isPermaLink="true">https://hospital-ai-lab.com/en/blog/2026-11-14-fda-mci-prognosis-software-class-ii-what-hospitals-check/</guid><description>What does FDA&apos;s new class II rule for mild cognitive impairment prognosis software mean for a hospital&apos;s AI review? Plain-language summary.</description><pubDate>Sat, 14 Nov 2026 00:00:00 GMT</pubDate><content:encoded>&lt;p&gt;When a vendor says its software is “FDA authorized,” hospital administrators and nurse leaders still need to know what that covers. A rule published in the Federal Register on November 5, 2026 is a useful example. It places software that estimates how a patient with mild cognitive impairment (MCI) may progress into a regulated category. This post summarizes what the rule says and what a hospital team can ask. It is general information, not legal advice.&lt;/p&gt;
&lt;h2 id=&quot;what-the-rule-does&quot;&gt;What the rule does&lt;/h2&gt;
&lt;p&gt;According to the &lt;a href=&quot;https://www.federalregister.gov/documents/full_text/text/2026/11/05/2026-22590.txt&quot;&gt;Federal Register notice&lt;/a&gt; (docket FDA-2026-N-12212, effective November 5, 2026), FDA is classifying “prognostic assessment software of mild cognitive impairment” as a class II device, with special controls. The rule is codified at 21 CFR 882.1454.&lt;/p&gt;
&lt;p&gt;The notice describes the device as prescription-only software that combines information from several clinical data sources. It estimates how likely a patient who already has an MCI diagnosis is to progress to more severe cognitive impairment or dementia within a set timeframe.&lt;/p&gt;
&lt;h2 id=&quot;why-this-matters-beyond-one-product&quot;&gt;Why this matters beyond one product&lt;/h2&gt;
&lt;p&gt;The notice explains that FDA acted on a company’s De Novo request, the pathway for new devices that have no prior classification. FDA found that general controls alone were not enough but that special controls could provide reasonable assurance of safety and effectiveness. It issued its order to that company in January 2024, and this notice writes the classification into the regulations. Because the category now exists, similar future devices can generally use it as a reference and go through the 510(k) process instead of a new De Novo request.&lt;/p&gt;
&lt;p&gt;For a hospital, that means more products may arrive under the same label. “Cleared” does not mean “right for our patients.”&lt;/p&gt;
&lt;h2 id=&quot;what-the-special-controls-and-limits-tell-you&quot;&gt;What the special controls and limits tell you&lt;/h2&gt;
&lt;p&gt;The notice lists special controls that cover clinical performance testing and postmarket surveillance, software verification and hazard analysis, human factors testing, and labeling for clinicians and patients. It also states limits: the software is an adjunct, not a standalone diagnostic or prognostic tool, its output should be read with the full clinical picture, and it may not perform as expected for some subpopulations.&lt;/p&gt;
&lt;p&gt;Questions a team can ask a vendor, drawn from those points:&lt;/p&gt;
&lt;ul&gt;
&lt;li&gt;Which patient groups were the performance data based on, and do they look like ours?&lt;/li&gt;
&lt;li&gt;Who reviews the output, and how is it explained to patients and families?&lt;/li&gt;
&lt;li&gt;What happens after go-live: how are errors and complaints tracked?&lt;/li&gt;
&lt;/ul&gt;
&lt;h2 id=&quot;key-takeaways&quot;&gt;Key takeaways&lt;/h2&gt;
&lt;ul&gt;
&lt;li&gt;FDA classified MCI prognosis software as class II with special controls (21 CFR 882.1454), effective November 5, 2026.&lt;/li&gt;
&lt;li&gt;The software supports clinicians; it is not a standalone diagnostic or prognostic tool.&lt;/li&gt;
&lt;li&gt;Hospital teams can use the stated limits and controls as a checklist for vendor questions. Confirm specifics with your compliance staff.&lt;/li&gt;
&lt;/ul&gt;
&lt;hr&gt;
&lt;p&gt;&lt;em&gt;This site provides general information for research and education. It is not medical advice and does not replace diagnosis or guidance for any specific patient. It is not legal advice.&lt;/em&gt;&lt;/p&gt;</content:encoded><category>News</category></item><item><title>AI that flags lung CT for referral: who acts on the flag?</title><link>https://hospital-ai-lab.com/en/blog/2026-11-09-lung-ct-ai-referral-flag-who-acts/</link><guid isPermaLink="true">https://hospital-ai-lab.com/en/blog/2026-11-09-lung-ct-ai-referral-flag-who-acts/</guid><description>FDA&apos;s new device type for fibrotic lung disease referral software, and the workflow questions it raises for hospital admin and nursing teams.</description><pubDate>Mon, 09 Nov 2026 00:00:00 GMT</pubDate><content:encoded>&lt;p&gt;Some AI tools do not make a diagnosis. They point out a case that may need a closer look and send it down a referral pathway. On October 30, 2026, the FDA published an order creating a device type for this kind of software in lung imaging. The rule itself is about what manufacturers must show. But the way the FDA describes the risks is a useful checklist for hospitals deciding how such a flag fits into daily work. This post is general information, not legal advice.&lt;/p&gt;
&lt;h2 id=&quot;what-the-fda-order-says&quot;&gt;What the FDA order says&lt;/h2&gt;
&lt;p&gt;The &lt;a href=&quot;https://www.federalregister.gov/documents/2026/10/30/2026-22180/medical-devices-radiology-devices-classification-of-the-radiology-software-for-referral-of-findings&quot;&gt;Federal Register order&lt;/a&gt; adds 21 CFR 892.2085, “Radiology software for referral of findings related to fibrotic lung disease,” as a class II device with special controls. The order is effective October 30, 2026; the classification itself was applicable from January 12, 2024, when the FDA granted the original De Novo request.&lt;/p&gt;
&lt;p&gt;According to the &lt;a href=&quot;https://www.federalregister.gov/documents/full_text/text/2026/10/30/2026-22180.txt&quot;&gt;full text&lt;/a&gt;, the device is a prescription image-processing tool. It analyzes CT images to suggest disease or a finding suggestive of disease, and its output is adjunctive information within a referral pathway that is part of the overall diagnostic assessment. In other words, it adds information; it is not the final word.&lt;/p&gt;
&lt;h2 id=&quot;the-risks-the-fda-lists&quot;&gt;The risks the FDA lists&lt;/h2&gt;
&lt;p&gt;The order names the main risks to health:&lt;/p&gt;
&lt;ul&gt;
&lt;li&gt;&lt;strong&gt;False positives&lt;/strong&gt;: a flag that leads to wrong or delayed management of a patient who has a different lung disease.&lt;/li&gt;
&lt;li&gt;&lt;strong&gt;False negatives&lt;/strong&gt;: no flag, so a patient with the disease is managed late or incorrectly.&lt;/li&gt;
&lt;li&gt;&lt;strong&gt;Misuse&lt;/strong&gt;: running the software on the wrong patients, or on images from incompatible scanners or settings.&lt;/li&gt;
&lt;li&gt;&lt;strong&gt;Overreliance&lt;/strong&gt;: misreading the output or leaning on it too heavily.&lt;/li&gt;
&lt;li&gt;&lt;strong&gt;Device failure&lt;/strong&gt;: results that are missing, delayed, or wrong.&lt;/li&gt;
&lt;/ul&gt;
&lt;p&gt;Labeling must describe the clinical context and workflow, compatible imaging hardware and protocols, performance testing results, and a warning to use the device together with other clinical and diagnostic findings.&lt;/p&gt;
&lt;h2 id=&quot;questions-for-your-team-before-go-live&quot;&gt;Questions for your team before go-live&lt;/h2&gt;
&lt;p&gt;These are our suggestions, not FDA requirements. Confirm details with your radiology, compliance, and IT leads.&lt;/p&gt;
&lt;ol&gt;
&lt;li&gt;&lt;strong&gt;Who receives the flag?&lt;/strong&gt; Name the role that sees it, and the backup on nights and weekends.&lt;/li&gt;
&lt;li&gt;&lt;strong&gt;What happens next, and by when?&lt;/strong&gt; A referral flag is only useful if someone schedules the follow-up. Agree who owns that step.&lt;/li&gt;
&lt;li&gt;&lt;strong&gt;Does our scanner setup match the labeling?&lt;/strong&gt; Ask the vendor to confirm compatible hardware and protocols in writing.&lt;/li&gt;
&lt;li&gt;&lt;strong&gt;What if the software is down?&lt;/strong&gt; Plan how staff will know results are missing or late.&lt;/li&gt;
&lt;/ol&gt;
&lt;h2 id=&quot;key-takeaways&quot;&gt;Key takeaways&lt;/h2&gt;
&lt;ul&gt;
&lt;li&gt;The FDA created 21 CFR 892.2085, a class II device type for CT software that suggests fibrotic lung disease findings for referral.&lt;/li&gt;
&lt;li&gt;Its output is adjunctive information in a referral pathway, not a stand-alone answer.&lt;/li&gt;
&lt;li&gt;Listed risks include false positives and negatives, misuse on incompatible images, overreliance, and device failure.&lt;/li&gt;
&lt;li&gt;Before go-live, decide who receives the flag, who owns follow-up, and what to do if the tool fails.&lt;/li&gt;
&lt;/ul&gt;
&lt;p&gt;Source: &lt;a href=&quot;https://www.federalregister.gov/documents/2026/10/30/2026-22180/medical-devices-radiology-devices-classification-of-the-radiology-software-for-referral-of-findings&quot;&gt;Federal Register, October 30, 2026 (FDA-2026-N-12254)&lt;/a&gt;&lt;/p&gt;
&lt;hr&gt;
&lt;p&gt;&lt;em&gt;This site’s content is for general information, research, and education. It does not replace medical diagnosis or advice for any specific patient, and it is not legal advice.&lt;/em&gt;&lt;/p&gt;</content:encoded><category>AI Tools</category></item><item><title>Patients Want to Know When AI Is Used: A Pew Survey</title><link>https://hospital-ai-lab.com/en/blog/2026-11-08-patients-want-to-know-ai-use-pew-survey/</link><guid isPermaLink="true">https://hospital-ai-lab.com/en/blog/2026-11-08-patients-want-to-know-ai-use-pew-survey/</guid><description>72% of U.S. adults want to be told if AI takes notes at a visit. What a 2026 Pew survey means for nursing and admin teams planning AI disclosure.</description><pubDate>Sun, 08 Nov 2026 00:00:00 GMT</pubDate><content:encoded>&lt;p&gt;Most AI planning in hospitals starts with features, cost, and security. A question that often comes later is what to tell patients. A Pew Research Center survey published August 25, 2026 gives useful data on what patients expect. This post summarizes it for administrators and nursing leaders. It is general information, not legal advice.&lt;/p&gt;
&lt;h2 id=&quot;what-the-survey-found&quot;&gt;What the survey found&lt;/h2&gt;
&lt;p&gt;Pew &lt;a href=&quot;https://www.pewresearch.org/short-reads/2026/08/25/americans-want-transparency-when-ai-is-used-in-their-healthcare/&quot;&gt;surveyed 3,488 U.S. adults&lt;/a&gt; from June 22 to 28, 2026. Overall, 72% said it is extremely or very important that a doctor or other provider tells them if AI is being used in their care.&lt;/p&gt;
&lt;p&gt;Asked task by task, the share saying patients should be told was:&lt;/p&gt;
&lt;ul&gt;
&lt;li&gt;Analyze medical scans: 81%&lt;/li&gt;
&lt;li&gt;Make a diagnosis: 81%&lt;/li&gt;
&lt;li&gt;Explain lab results: 80%&lt;/li&gt;
&lt;li&gt;Take notes during an appointment: 72%&lt;/li&gt;
&lt;li&gt;Order prescription refills from a pharmacy: 64%&lt;/li&gt;
&lt;li&gt;Schedule an appointment: 56%&lt;/li&gt;
&lt;/ul&gt;
&lt;p&gt;Also, 63% would like more say over whether AI is used in their care, while 21% are comfortable with the say they have now. Only 16% said their provider has used AI in their care; 37% said no and 46% were not sure.&lt;/p&gt;
&lt;h2 id=&quot;what-stands-out-for-hospital-teams&quot;&gt;What stands out for hospital teams&lt;/h2&gt;
&lt;p&gt;Clinical uses drew the strongest demand for disclosure, as you might expect. But note-taking, which is often treated as a back-office task, still reached 72%. That matters for teams rolling out AI documentation tools, where nurses and clinicians are the people patients will ask. And nearly half of respondents did not know whether AI had been part of their care at all. The survey measures attitudes; it does not say what any law requires.&lt;/p&gt;
&lt;h2 id=&quot;three-things-to-prepare-before-a-rollout&quot;&gt;Three things to prepare before a rollout&lt;/h2&gt;
&lt;p&gt;These are our general suggestions, not survey findings. Check your own approach with compliance, legal, and patient experience staff.&lt;/p&gt;
&lt;ol&gt;
&lt;li&gt;&lt;strong&gt;Make an inventory of AI uses.&lt;/strong&gt; List where AI touches care and operations, such as imaging support, documentation drafts, and scheduling. It is hard to explain what the organization itself has not mapped.&lt;/li&gt;
&lt;li&gt;&lt;strong&gt;Decide who explains, and when.&lt;/strong&gt; For documentation tools used at the bedside or in clinic, agree whether the explanation comes from the clinician, the nurse, or written materials, and use consistent wording.&lt;/li&gt;
&lt;li&gt;&lt;strong&gt;Prepare answers to likely questions.&lt;/strong&gt; Patients may ask who makes the final decision or how their data is used. Have the responsible teams approve those answers ahead of time so front-line staff are not improvising.&lt;/li&gt;
&lt;/ol&gt;
&lt;h2 id=&quot;key-takeaways&quot;&gt;Key takeaways&lt;/h2&gt;
&lt;ul&gt;
&lt;li&gt;In a June 2026 Pew survey of 3,488 U.S. adults, 72% said being told about AI use is extremely or very important.&lt;/li&gt;
&lt;li&gt;81% want to be told when AI analyzes scans or helps diagnose; 72% when it takes notes at a visit.&lt;/li&gt;
&lt;li&gt;46% were not sure whether their provider had used AI at all.&lt;/li&gt;
&lt;li&gt;Hospitals can prepare an AI use inventory, a plan for who explains it, and approved answers to common questions.&lt;/li&gt;
&lt;/ul&gt;
&lt;p&gt;Source: &lt;a href=&quot;https://www.pewresearch.org/short-reads/2026/08/25/americans-want-transparency-when-ai-is-used-in-their-healthcare/&quot;&gt;Pew Research Center, August 25, 2026&lt;/a&gt;&lt;/p&gt;
&lt;hr&gt;
&lt;p&gt;&lt;em&gt;This site’s content is for general information, research, and education. It does not replace medical diagnosis or advice for any specific patient, and it is not legal advice.&lt;/em&gt;&lt;/p&gt;</content:encoded><category>Research Review</category></item><item><title>FDA Classifies Cervical Cytology AI Imaging as Class II</title><link>https://hospital-ai-lab.com/en/blog/2026-11-07-fda-class-ii-cervical-cytology-ai-slide-imaging/</link><guid isPermaLink="true">https://hospital-ai-lab.com/en/blog/2026-11-07-fda-class-ii-cervical-cytology-ai-slide-imaging/</guid><description>A Nov. 5, 2026 final rule puts AI-assisted digital Pap slide imaging in Class II. What its special controls mean for hospital buyers.</description><pubDate>Sat, 07 Nov 2026 00:00:00 GMT</pubDate><content:encoded>&lt;p&gt;When a vendor says an AI tool is “FDA authorized,” the next question is what the FDA actually required. A final rule published on November 5, 2026 gives a concrete example. It sets the rules for a type of AI-assisted device used to review Pap test slides. This post summarizes the rule for administrators and nursing leaders. It is general information, not legal advice.&lt;/p&gt;
&lt;h2 id=&quot;what-the-rule-does&quot;&gt;What the rule does&lt;/h2&gt;
&lt;p&gt;The FDA &lt;a href=&quot;https://www.federalregister.gov/documents/full_text/html/2026/11/05/2026-22593.html&quot;&gt;published a final rule&lt;/a&gt; (docket FDA-2026-N-12384) classifying the “digital cervical cytology slide imaging system with artificial intelligence algorithm” as Class II, which means it is subject to special controls. The rule is codified at 21 CFR 864.3900. The device is a prescription in vitro diagnostic system that creates, displays, and manages digital images of Pap test slides. Its AI algorithm flags areas that may be clinically significant, and the output helps a trained user reach a cytology diagnosis. The order took effect November 5, 2026, and the classification is stated to apply as of January 31, 2024.&lt;/p&gt;
&lt;p&gt;The rule did not exempt these devices from premarket notification. A maker still needs to submit a 510(k) before marketing a device of this type.&lt;/p&gt;
&lt;h2 id=&quot;what-the-special-controls-cover&quot;&gt;What the special controls cover&lt;/h2&gt;
&lt;p&gt;The rule lists required labeling and testing. In plain terms:&lt;/p&gt;
&lt;ul&gt;
&lt;li&gt;&lt;strong&gt;Labeling&lt;/strong&gt; must describe intended use, the specimen types, how AI results are shown, and who the intended users are. It must include step-by-step review instructions, recommended user training, performance data, and limits, such as the need to review all flagged areas before diagnosing and to use the tool alongside standard care and clinical judgment. Cybersecurity, including for remote review, is also addressed.&lt;/li&gt;
&lt;li&gt;&lt;strong&gt;Testing&lt;/strong&gt; must cover each hardware part (for example the scanner and display) and system performance. Studies must show results hold across scanners, operators, and sites, and that reading digital images is non-inferior to a standard microscope, with and without AI. Screening time data and human factors testing are required.&lt;/li&gt;
&lt;/ul&gt;
&lt;h2 id=&quot;why-this-matters-to-buyers&quot;&gt;Why this matters to buyers&lt;/h2&gt;
&lt;p&gt;Digital pathology and cytology projects involve scanners, storage, and workflow changes, not only software. These controls give buyers a checklist for vendor conversations: ask which labeled user training is required, what workload limits the labeling recommends, and how the system fits your quality control process. Whether a particular product has cleared review is something to confirm directly with the vendor and your compliance team.&lt;/p&gt;
&lt;h2 id=&quot;key-takeaways&quot;&gt;Key takeaways&lt;/h2&gt;
&lt;ul&gt;
&lt;li&gt;A November 5, 2026 final rule classifies AI-assisted digital cervical cytology slide imaging systems as Class II with special controls (21 CFR 864.3900).&lt;/li&gt;
&lt;li&gt;Devices of this type still need a 510(k) before marketing.&lt;/li&gt;
&lt;li&gt;Required labeling and testing include user training, workload limits, and non-inferiority to a standard microscope.&lt;/li&gt;
&lt;li&gt;Use the rule’s list as a starting point for vendor questions; confirm any specific product’s status with the vendor and compliance staff.&lt;/li&gt;
&lt;/ul&gt;
&lt;p&gt;Source: &lt;a href=&quot;https://www.federalregister.gov/documents/full_text/html/2026/11/05/2026-22593.html&quot;&gt;Federal Register, document 2026-22593&lt;/a&gt;&lt;/p&gt;
&lt;hr&gt;
&lt;p&gt;&lt;em&gt;This site’s content is for general information, research, and education. It does not replace medical diagnosis or advice for any specific patient, and it is not legal advice.&lt;/em&gt;&lt;/p&gt;</content:encoded><category>News</category></item><item><title>AI vendor business associate agreements: what to write down</title><link>https://hospital-ai-lab.com/en/blog/2026-11-06-business-associate-agreement-ai-vendor/</link><guid isPermaLink="true">https://hospital-ai-lab.com/en/blog/2026-11-06-business-associate-agreement-ai-vendor/</guid><description>HIPAA requires a written agreement before patient data goes to an AI vendor that is a business associate. A plain reading of the rule.</description><pubDate>Fri, 06 Nov 2026 00:00:00 GMT</pubDate><content:encoded>&lt;p&gt;Many AI projects begin with a pilot: send a sample of records to the vendor and see what comes back. Before that first file leaves the building, HIPAA asks one question that is easy to skip: is the vendor a business associate, and is there a written agreement?&lt;/p&gt;
&lt;h2 id=&quot;who-is-a-business-associate&quot;&gt;Who is a business associate&lt;/h2&gt;
&lt;p&gt;The definition in 45 CFR 160.103 covers a person who, on behalf of a covered entity and other than as a member of its workforce, creates, receives, maintains or transmits protected health information for a regulated function. It also covers a person who provides services such as legal, actuarial, accounting, consulting or data aggregation to or for the covered entity where providing the service involves disclosure of that information. Read it against what the vendor actually does with your data, not against how the vendor describes itself.&lt;/p&gt;
&lt;h2 id=&quot;the-agreement-comes-first&quot;&gt;The agreement comes first&lt;/h2&gt;
&lt;p&gt;Under 45 CFR 164.502(e), a covered entity may disclose protected health information to a business associate, and let it create, receive, maintain or transmit it, only if the covered entity obtains satisfactory assurance that the business associate will appropriately safeguard it. Those assurances must be documented in a written contract or other written agreement that meets the requirements of § 164.504(e).&lt;/p&gt;
&lt;h2 id=&quot;what-the-contract-must-contain&quot;&gt;What the contract must contain&lt;/h2&gt;
&lt;p&gt;Section 164.504(e)(2) lists the terms. The contract must establish the permitted uses and disclosures. It may not authorize uses that would violate the Privacy Rule if the hospital did them, apart from two exceptions: the business associate’s own proper management and administration, and data aggregation services for the hospital’s operations. It must also provide that the business associate will:&lt;/p&gt;
&lt;ul&gt;
&lt;li&gt;not use or further disclose the information other than as the contract permits or the law requires;&lt;/li&gt;
&lt;li&gt;use appropriate safeguards and comply with the Security Rule for electronic information;&lt;/li&gt;
&lt;li&gt;report uses or disclosures the contract does not allow, including breaches under § 164.410;&lt;/li&gt;
&lt;li&gt;make sure subcontractors agree to the same restrictions;&lt;/li&gt;
&lt;li&gt;make information available for access, amendment and an accounting of disclosures;&lt;/li&gt;
&lt;li&gt;make its books and records available to HHS; and&lt;/li&gt;
&lt;li&gt;at termination, if feasible, return or destroy the information.&lt;/li&gt;
&lt;/ul&gt;
&lt;p&gt;The contract must also let the hospital terminate if the business associate violates a material term.&lt;/p&gt;
&lt;h2 id=&quot;three-places-ai-contracts-need-extra-attention&quot;&gt;Three places AI contracts need extra attention&lt;/h2&gt;
&lt;ol&gt;
&lt;li&gt;&lt;strong&gt;Use of your data to improve the vendor’s own product.&lt;/strong&gt; The contract may not authorize uses that would violate the Privacy Rule if the hospital did them, with the two exceptions above. Ask counsel to read the vendor’s clause against that text.&lt;/li&gt;
&lt;li&gt;&lt;strong&gt;Subcontractors.&lt;/strong&gt; Cloud and model providers are often subcontractors. The agreement must require the vendor to bind them to the same terms.&lt;/li&gt;
&lt;li&gt;&lt;strong&gt;Return or destruction.&lt;/strong&gt; Ask how the vendor will do this for data embedded in logs, backups or derived datasets.&lt;/li&gt;
&lt;/ol&gt;
&lt;h2 id=&quot;key-takeaways&quot;&gt;Key takeaways&lt;/h2&gt;
&lt;ul&gt;
&lt;li&gt;A vendor that handles PHI on your behalf, or provides listed services involving PHI, is a business associate (45 CFR 160.103).&lt;/li&gt;
&lt;li&gt;PHI may be disclosed only after documented satisfactory assurance in a written agreement meeting § 164.504(e) (§ 164.502(e)).&lt;/li&gt;
&lt;li&gt;The contract must cover permitted uses, safeguards, breach reporting, subcontractors, patient rights support, return or destruction, and termination rights.&lt;/li&gt;
&lt;li&gt;Check the vendor’s use of your data and its subcontractors with counsel before the pilot starts.&lt;/li&gt;
&lt;/ul&gt;
&lt;p&gt;&lt;em&gt;Sources: &lt;a href=&quot;https://www.ecfr.gov/current/title-45/section-160.103&quot;&gt;eCFR, 45 CFR 160.103&lt;/a&gt;, &lt;a href=&quot;https://www.ecfr.gov/current/title-45/section-164.502&quot;&gt;164.502&lt;/a&gt;, &lt;a href=&quot;https://www.ecfr.gov/current/title-45/section-164.504&quot;&gt;164.504&lt;/a&gt;&lt;/em&gt;&lt;/p&gt;</content:encoded><category>Column</category></item><item><title>Is your AI tool a medical device? FDA’s CDS guidance</title><link>https://hospital-ai-lab.com/en/blog/2026-11-06-fda-cds-guidance-vendor-questions/</link><guid isPermaLink="true">https://hospital-ai-lab.com/en/blog/2026-11-06-fda-cds-guidance-vendor-questions/</guid><description>FDA’s January 2026 guidance explains which decision support software falls outside the device definition, and what a hospital buyer should ask a vendor.</description><pubDate>Fri, 06 Nov 2026 00:00:00 GMT</pubDate><content:encoded>&lt;p&gt;A vendor says the product is “just decision support,” so it does not need FDA review. For a hospital administrator, that sentence decides how much checking you need to do yourself. Whether it is true depends on what the software actually does, and FDA has published guidance on exactly this question.&lt;/p&gt;
&lt;h2 id=&quot;what-the-guidance-covers&quot;&gt;What the guidance covers&lt;/h2&gt;
&lt;p&gt;FDA’s page for Clinical Decision Support Software shows a final guidance for industry and FDA staff dated January 2026. It explains how FDA reads the non-device clinical decision support criteria in section 520(o)(1)(E) of the Federal Food, Drug, and Cosmetic Act. That section was added to the statute by section 3060(a) of the 21st Century Cures Act, which excluded certain medical software functions from the device definition.&lt;/p&gt;
&lt;p&gt;The guidance also says that FDA’s existing digital health policies continue to apply to software functions that meet the device definition, including functions meant for patients or caregivers. It includes many examples of both non-device decision support functions and device software functions.&lt;/p&gt;
&lt;h2 id=&quot;why-the-label-is-not-enough&quot;&gt;Why the label is not enough&lt;/h2&gt;
&lt;p&gt;The exclusion is not a category a vendor can claim by name. A software function is outside the device definition only if it meets the statutory criteria, and FDA’s guidance is its explanation of how it applies them. So “decision support” on a brochure does not settle anything. What the software does with its inputs, and who uses its output, does.&lt;/p&gt;
&lt;p&gt;For a hospital, this means the regulatory status of a clinical AI tool is the vendor’s position, not a fact you can read off the box. Ask for that position in writing, with the basis.&lt;/p&gt;
&lt;h2 id=&quot;questions-to-put-to-the-vendor&quot;&gt;Questions to put to the vendor&lt;/h2&gt;
&lt;ul&gt;
&lt;li&gt;Do you treat this product as a medical device, or as non-device clinical decision support? On what basis?&lt;/li&gt;
&lt;li&gt;If it is a device, what is the FDA authorization and the intended use as authorized, and does our planned use match it?&lt;/li&gt;
&lt;li&gt;If you say it is not a device, which of the statutory criteria do you rely on, and what in the product design supports that?&lt;/li&gt;
&lt;li&gt;Does the product have versions or features meant for patients or caregivers, and are they treated differently?&lt;/li&gt;
&lt;/ul&gt;
&lt;p&gt;Keep the written answer with the contract file. If it changes later, you will want to know.&lt;/p&gt;
&lt;h2 id=&quot;key-takeaways&quot;&gt;Key takeaways&lt;/h2&gt;
&lt;ul&gt;
&lt;li&gt;FDA’s clinical decision support guidance is shown as final and dated January 2026 on FDA’s page.&lt;/li&gt;
&lt;li&gt;It explains how FDA reads the non-device criteria in section 520(o)(1)(E) of the FD&amp;#x26;C Act.&lt;/li&gt;
&lt;li&gt;FDA says its digital health policies still apply to software functions that meet the device definition, including those for patients or caregivers.&lt;/li&gt;
&lt;li&gt;A vendor’s “decision support” label does not decide the question; ask for the position and its basis in writing.&lt;/li&gt;
&lt;/ul&gt;
&lt;p&gt;&lt;em&gt;Sources: &lt;a href=&quot;https://www.fda.gov/regulatory-information/search-fda-guidance-documents/clinical-decision-support-software&quot;&gt;FDA, Clinical Decision Support Software, guidance for industry and FDA staff&lt;/a&gt;&lt;/em&gt;&lt;/p&gt;</content:encoded><category>Column</category></item><item><title>The 60-day breach clock and your AI vendor: who is counting?</title><link>https://hospital-ai-lab.com/en/blog/2026-11-06-hipaa-breach-60-day-clock-ai-vendors/</link><guid isPermaLink="true">https://hospital-ai-lab.com/en/blog/2026-11-06-hipaa-breach-60-day-clock-ai-vendors/</guid><description>HIPAA gives a hospital up to 60 calendar days after discovery to notify patients of a breach, and a vendor’s delay can eat into that time.</description><pubDate>Fri, 06 Nov 2026 00:00:00 GMT</pubDate><content:encoded>&lt;p&gt;When an AI vendor holds patient data for you and something goes wrong, the question that matters most in the first week is not technical. It is: when did the clock start? HIPAA’s breach notification rules answer that, and the answer affects what you should ask for in a vendor contract.&lt;/p&gt;
&lt;h2 id=&quot;what-counts-as-a-breach&quot;&gt;What counts as a breach&lt;/h2&gt;
&lt;p&gt;Under 45 CFR 164.402, a breach is an acquisition, access, use or disclosure of protected health information that the Privacy Rule does not permit and that compromises its security or privacy. Such an event is presumed to be a breach unless the covered entity or business associate can show a low probability that the information was compromised. That showing must rest on a risk assessment of at least four factors: the nature and extent of the information, who used it or received it, whether it was actually acquired or viewed, and how far the risk has been reduced. Three narrow exclusions are listed in the definition.&lt;/p&gt;
&lt;h2 id=&quot;the-clock&quot;&gt;The clock&lt;/h2&gt;
&lt;p&gt;A hospital must notify each affected individual without unreasonable delay and no later than 60 calendar days after it discovers the breach (§ 164.404(b)). A breach is treated as discovered on the first day it is known, or would have been known with reasonable diligence, to any workforce member or agent of the hospital other than the person who committed it.&lt;/p&gt;
&lt;p&gt;Two other notices run on the same 60-day outer limit or close to it. For a breach affecting more than 500 residents of a state, the hospital must notify prominent media outlets serving that state (§ 164.406). For 500 or more individuals, HHS must be notified at the same time as the individuals; for fewer than 500, the hospital keeps a log and reports to HHS within 60 days after the end of the calendar year (§ 164.408).&lt;/p&gt;
&lt;h2 id=&quot;where-the-vendor-comes-in&quot;&gt;Where the vendor comes in&lt;/h2&gt;
&lt;p&gt;A business associate must notify the hospital of a breach without unreasonable delay and no later than 60 calendar days after it discovers it (§ 164.410). The text sets that same outer limit for the vendor. It also says that a breach is treated as discovered by the business associate when known to its employees, officers or agents.&lt;/p&gt;
&lt;p&gt;The rules do not say that the hospital’s own clock waits for the vendor’s notice. The hospital’s discovery standard covers its agents. Whether a vendor counts as an agent depends on the federal common law of agency, and that is a question for counsel. The practical point is simple: if a vendor uses most of its 60 days before telling you, you may have little time left.&lt;/p&gt;
&lt;h2 id=&quot;what-to-put-in-the-contract&quot;&gt;What to put in the contract&lt;/h2&gt;
&lt;ul&gt;
&lt;li&gt;A vendor notice period much shorter than 60 days, counted from the vendor’s own discovery.&lt;/li&gt;
&lt;li&gt;A named 24-hour contact and a duty to share the facts you need for the patient notice.&lt;/li&gt;
&lt;li&gt;A list of subcontractors that touch the data, since a subcontractor’s breach reaches you through the vendor.&lt;/li&gt;
&lt;/ul&gt;
&lt;p&gt;These are recommendations, not legal requirements; the rule sets only the outer limit.&lt;/p&gt;
&lt;h2 id=&quot;key-takeaways&quot;&gt;Key takeaways&lt;/h2&gt;
&lt;ul&gt;
&lt;li&gt;A breach is presumed unless a four-factor risk assessment shows a low probability of compromise (45 CFR 164.402).&lt;/li&gt;
&lt;li&gt;Patients must be notified no later than 60 calendar days after discovery; HHS and media notices follow their own thresholds (§§ 164.404–164.408).&lt;/li&gt;
&lt;li&gt;A business associate must notify the hospital within 60 days at most (§ 164.410).&lt;/li&gt;
&lt;li&gt;Ask for a much shorter vendor notice period and a 24-hour contact in the contract.&lt;/li&gt;
&lt;/ul&gt;
&lt;p&gt;&lt;em&gt;Sources: &lt;a href=&quot;https://www.ecfr.gov/current/title-45/section-164.402&quot;&gt;eCFR, 45 CFR 164.402&lt;/a&gt;, &lt;a href=&quot;https://www.ecfr.gov/current/title-45/section-164.404&quot;&gt;164.404&lt;/a&gt;, &lt;a href=&quot;https://www.ecfr.gov/current/title-45/section-164.406&quot;&gt;164.406&lt;/a&gt;, &lt;a href=&quot;https://www.ecfr.gov/current/title-45/section-164.408&quot;&gt;164.408&lt;/a&gt;, &lt;a href=&quot;https://www.ecfr.gov/current/title-45/section-164.410&quot;&gt;164.410&lt;/a&gt;&lt;/em&gt;&lt;/p&gt;</content:encoded><category>Column</category></item><item><title>AI device changes after approval: what a PCCP means</title><link>https://hospital-ai-lab.com/en/blog/2026-11-06-fda-pccp-ai-device-changes/</link><guid isPermaLink="true">https://hospital-ai-lab.com/en/blog/2026-11-06-fda-pccp-ai-device-changes/</guid><description>FDA lists an August 2025 final guidance on predetermined change control plans for AI-enabled devices. What it is, and what a hospital should ask.</description><pubDate>Fri, 06 Nov 2026 00:00:00 GMT</pubDate><content:encoded>&lt;p&gt;Most hospital equipment stays the same after you buy it. AI software may not. A model can be retrained or updated, and then it is no longer exactly the product you reviewed. FDA has a mechanism for authorizing some of those changes in advance, and a hospital buyer should know it exists.&lt;/p&gt;
&lt;h2 id=&quot;what-a-pccp-is&quot;&gt;What a PCCP is&lt;/h2&gt;
&lt;p&gt;FDA’s page on predetermined change control plans (PCCPs) describes a PCCP as a plan that describes certain planned changes that may be made to a device. It also covers how the manufacturer will develop, test and implement those changes and assess their impact. A manufacturer can include a PCCP in a marketing submission so FDA can authorize future changes in advance. The page says those changes can then be made without a new submission for each one.&lt;/p&gt;
&lt;p&gt;The same page lists a final guidance dated August 2025 titled Marketing Submission Recommendations for a Predetermined Change Control Plan for Artificial Intelligence-Enabled Device Software Functions. It also lists a separate draft guidance from August 2024 on PCCPs for medical devices more broadly, which is a draft and not final.&lt;/p&gt;
&lt;h2 id=&quot;what-this-means-on-the-hospital-side&quot;&gt;What this means on the hospital side&lt;/h2&gt;
&lt;p&gt;A PCCP is a tool for manufacturers. It does not tell you whether a given model update is good for your patients. What it does give you is a set of questions that are specific and answerable:&lt;/p&gt;
&lt;ul&gt;
&lt;li&gt;Does this device have an authorized PCCP? If so, which kinds of change does it cover?&lt;/li&gt;
&lt;li&gt;How will we be told when a change is made under the plan?&lt;/li&gt;
&lt;li&gt;How will we check that the tool still performs as expected in our hospital after a change?&lt;/li&gt;
&lt;/ul&gt;
&lt;p&gt;If a vendor cannot say whether a PCCP exists, that is information too. It may mean the product is not an authorized device, or that the vendor has not thought about change management. Either way, put the question in the contract review.&lt;/p&gt;
&lt;h2 id=&quot;do-not-confuse-this-with-a-hospitals-own-monitoring&quot;&gt;Do not confuse this with a hospital’s own monitoring&lt;/h2&gt;
&lt;p&gt;FDA’s mechanism concerns what the manufacturer may change. It does not take over the hospital’s job of watching performance in its own setting. Decide who in your organization looks at the tool’s results after go-live, and what happens when they look worse than expected.&lt;/p&gt;
&lt;h2 id=&quot;key-takeaways&quot;&gt;Key takeaways&lt;/h2&gt;
&lt;ul&gt;
&lt;li&gt;A PCCP describes planned device changes and how they will be developed, tested and implemented, so FDA can authorize them in advance.&lt;/li&gt;
&lt;li&gt;FDA’s page lists a final guidance dated August 2025 for AI-enabled device software functions.&lt;/li&gt;
&lt;li&gt;Ask a vendor whether the device has a PCCP, what it covers, and how you will be told of changes.&lt;/li&gt;
&lt;li&gt;A PCCP does not replace your own monitoring after go-live.&lt;/li&gt;
&lt;/ul&gt;
&lt;p&gt;&lt;em&gt;Sources: &lt;a href=&quot;https://www.fda.gov/medical-devices/premarket-submissions-selecting-and-preparing-correct-submission/predetermined-change-control-plans&quot;&gt;FDA, Predetermined Change Control Plans&lt;/a&gt;&lt;/em&gt;&lt;/p&gt;</content:encoded><category>Column</category></item><item><title>HIPAA Security Rule rewrite is still only a proposal</title><link>https://hospital-ai-lab.com/en/blog/2026-11-06-hipaa-security-rule-proposal-status/</link><guid isPermaLink="true">https://hospital-ai-lab.com/en/blog/2026-11-06-hipaa-security-rule-proposal-status/</guid><description>HHS proposed a HIPAA Security Rule rewrite in January 2025, and a November 2026 search found no final rule. What hospitals must do under the current text.</description><pubDate>Fri, 06 Nov 2026 00:00:00 GMT</pubDate><content:encoded>&lt;p&gt;Articles about a new HIPAA Security Rule often read as if it were already law. As of this writing, it is not. For a hospital that is planning a security budget around an AI project, it helps to separate what is proposed from what is required today.&lt;/p&gt;
&lt;h2 id=&quot;what-happened&quot;&gt;What happened&lt;/h2&gt;
&lt;p&gt;The Federal Register lists a proposed rule titled HIPAA Security Rule To Strengthen the Cybersecurity of Electronic Protected Health Information, published January 6, 2025 at 90 FR 898. Its comment period closed March 7, 2025. When we searched the Federal Register on November 6, 2026, we found no final rule for it. A proposed rule can be finalized, changed or withdrawn, and we do not report predictions about timing.&lt;/p&gt;
&lt;h2 id=&quot;what-applies-today&quot;&gt;What applies today&lt;/h2&gt;
&lt;p&gt;The current Security Rule text in 45 CFR Part 164, Subpart C remains in force. Several of its requirements bear directly on any new AI system that touches electronic protected health information:&lt;/p&gt;
&lt;ul&gt;
&lt;li&gt;&lt;strong&gt;Risk analysis.&lt;/strong&gt; A covered entity or business associate must conduct an accurate and thorough assessment of the risks and vulnerabilities to the confidentiality, integrity and availability of the electronic protected health information it holds, and implement security measures sufficient to reduce those risks to a reasonable and appropriate level (§ 164.308(a)(1)(ii)(A) and (B)). Both are required, not addressable.&lt;/li&gt;
&lt;li&gt;&lt;strong&gt;Activity review.&lt;/strong&gt; Procedures must regularly review records of information system activity, such as audit logs and access reports (§ 164.308(a)(1)(ii)(D)), and information systems that hold or use the data need mechanisms that record and examine activity (§ 164.312(b)).&lt;/li&gt;
&lt;li&gt;&lt;strong&gt;Encryption.&lt;/strong&gt; The current text lists encryption and decryption as an addressable implementation specification under access control (§ 164.312(a)(2)(iv)). Addressable means the entity must assess and document its decision, not that it can ignore it.&lt;/li&gt;
&lt;li&gt;&lt;strong&gt;Documentation.&lt;/strong&gt; Policies and required records must be kept in writing for 6 years from creation or the date last in effect, and reviewed when operations change (§ 164.316(b)).&lt;/li&gt;
&lt;/ul&gt;
&lt;h2 id=&quot;a-practical-approach&quot;&gt;A practical approach&lt;/h2&gt;
&lt;p&gt;Do not wait for a final rule to do the work that is already required. When a new AI tool is added, update the risk analysis to include it, check that logging covers it, and write down your encryption decision. If a final rule arrives and asks for more, you will already have the inventory and the records to build on.&lt;/p&gt;
&lt;h2 id=&quot;key-takeaways&quot;&gt;Key takeaways&lt;/h2&gt;
&lt;ul&gt;
&lt;li&gt;The HIPAA Security Rule rewrite was proposed on January 6, 2025 (90 FR 898); comments closed March 7, 2025; we found no final rule on November 6, 2026.&lt;/li&gt;
&lt;li&gt;The current Security Rule text still applies, including the required risk analysis and risk management.&lt;/li&gt;
&lt;li&gt;Add each new AI system to the risk analysis, logging and documentation you already keep.&lt;/li&gt;
&lt;/ul&gt;
&lt;p&gt;&lt;em&gt;Sources: &lt;a href=&quot;https://www.federalregister.gov/documents/2025/01/06/2024-30983/hipaa-security-rule-to-strengthen-the-cybersecurity-of-electronic-protected-health-information&quot;&gt;Federal Register, 90 FR 898&lt;/a&gt;, &lt;a href=&quot;https://www.ecfr.gov/current/title-45/section-164.308&quot;&gt;eCFR, 45 CFR 164.308&lt;/a&gt;, &lt;a href=&quot;https://www.ecfr.gov/current/title-45/section-164.312&quot;&gt;164.312&lt;/a&gt;, &lt;a href=&quot;https://www.ecfr.gov/current/title-45/section-164.316&quot;&gt;164.316&lt;/a&gt;&lt;/em&gt;&lt;/p&gt;</content:encoded><category>Column</category></item><item><title>HTI-5 final rule: what changes for EHR decision support</title><link>https://hospital-ai-lab.com/en/blog/2026-11-06-hti-5-decision-support-rule/</link><guid isPermaLink="true">https://hospital-ai-lab.com/en/blog/2026-11-06-hti-5-decision-support-rule/</guid><description>ONC’s HTI-5 final rule, published October 28, 2026, removes decision support source attribute and risk-management certification requirements.</description><pubDate>Fri, 06 Nov 2026 00:00:00 GMT</pubDate><content:encoded>&lt;p&gt;If your electronic health record vendor has been telling you its AI-based decision support meets federal transparency requirements, a rule published at the end of October changes what those requirements are. It matters to hospital administrators because it moves some of the burden of vendor review from the certification program to you.&lt;/p&gt;
&lt;h2 id=&quot;what-was-published&quot;&gt;What was published&lt;/h2&gt;
&lt;p&gt;The Federal Register shows a final rule titled Health Data, Technology, and Interoperability: ONC Deregulatory Actions To Unleash Prosperity, published October 28, 2026 at 91 FR 67916. The abstract says it focuses on deregulatory actions on health IT standards, implementation specifications and certification criteria, and on certification programs. It is effective December 28, 2026, except for three amendatory instructions that are effective January 1, 2027.&lt;/p&gt;
&lt;h2 id=&quot;what-it-does-to-decision-support&quot;&gt;What it does to decision support&lt;/h2&gt;
&lt;p&gt;The preamble says the rule reduces the scope of the decision support interventions certification criterion in 45 CFR 170.315(b)(11). It removes paragraphs (b)(11)(iv) through (vi): the source attribute support requirements, the source attribute access and modification requirements, and the intervention risk-management requirements for predictive decision support. The preamble describes the source attribute requirements as the “model card” requirements.&lt;/p&gt;
&lt;p&gt;The rule also removes the older clinical decision support criterion in 170.315(a)(9). ONC points out that this criterion had already expired for certification on January 1, 2025, so its removal has no further practical effect.&lt;/p&gt;
&lt;p&gt;ONC acknowledges comments that the change shifts evaluation and governance work to providers and health systems. It says it will consider in future rulemaking whether similar transparency requirements should be reinstated, and that applying risk management to supplied predictive tools may still be a best practice.&lt;/p&gt;
&lt;h2 id=&quot;what-a-hospital-can-do&quot;&gt;What a hospital can do&lt;/h2&gt;
&lt;p&gt;Certification used to give you a baseline you could ask a vendor to document. That baseline is narrower now, so your own questions matter more:&lt;/p&gt;
&lt;ul&gt;
&lt;li&gt;What data was the predictive tool trained and tested on, and how does that compare with our patients?&lt;/li&gt;
&lt;li&gt;How does the vendor monitor performance and drift, and what will it show us?&lt;/li&gt;
&lt;li&gt;Who at the hospital reviews the tool’s results after go-live?&lt;/li&gt;
&lt;/ul&gt;
&lt;p&gt;None of this is a legal requirement created by the rule. It is the practical consequence of a smaller federal checklist.&lt;/p&gt;
&lt;h2 id=&quot;key-takeaways&quot;&gt;Key takeaways&lt;/h2&gt;
&lt;ul&gt;
&lt;li&gt;The HTI-5 final rule (91 FR 67916) was published October 28, 2026, effective December 28, 2026, with three instructions effective January 1, 2027.&lt;/li&gt;
&lt;li&gt;It removes 45 CFR 170.315(b)(11)(iv) through (vi), covering source attribute and risk-management requirements for decision support.&lt;/li&gt;
&lt;li&gt;ONC says it will consider reinstating similar requirements in future rulemaking.&lt;/li&gt;
&lt;li&gt;Hospital vendor review carries more weight as the certification checklist shrinks.&lt;/li&gt;
&lt;/ul&gt;
&lt;p&gt;&lt;em&gt;Sources: &lt;a href=&quot;https://www.federalregister.gov/documents/2026/10/28/2026-21943/health-data-technology-and-interoperability-onc-deregulatory-actions-to-unleash-prosperity&quot;&gt;Federal Register, HTI-5 final rule, 91 FR 67916&lt;/a&gt;&lt;/em&gt;&lt;/p&gt;</content:encoded><category>Column</category></item></channel></rss>