Marco's Morning Report — Wednesday, August 19, 2026

Good morning, Marco. Two things moved yesterday that are worth your five minutes. WHO's emergency committee met and kept the DRC Ebola outbreak at global-emergency status — it has now passed 5,000 cases and become the deadliest in that country's history, driven by a strain with no licensed vaccine. And the FDA opened a public docket on how to regulate generative-AI medical devices, which is the other half of the story you read here yesterday about two-thirds of hospitalists using AI and a third being trained on it. Plus a rounding-structure piece that describes what your nights already look like, and Tatis Jr. taking a grand slam away from Lindor. Coffee, then sleep.

Section A — Top Stories

Update: DRC Ebola passes 5,000 cases, becomes the country's deadliest outbreak — WHO keeps the global emergency in place

What's new WHO's emergency committee met Tuesday and reaffirmed the Democratic Republic of the Congo outbreak as a public health emergency of international concern. Confirmed cases have crossed 5,000 and deaths exceed 2,320, which passes the 2018–2020 outbreak's 2,299 and makes this the deadliest in the DRC's history — its seventeenth. Tedros told the meeting the epidemic is spreading at "unprecedented speed" and is "far from being under control," with transmission chains still unidentified. The outbreak was declared May 17, but sequencing shows it began circulating in February. It has reached a sixth province (Al Jazeera / AFP / Reuters, WHO situation page).

Why it matters to you The clinically important detail is the strain: this is Bundibugyo, for which there is no approved vaccine and no approved therapeutic. The Ervebo/rVSV vaccine and the monoclonal antibodies that shortened the 2018–2020 outbreak are Zaire-strain products and do not apply here. An Oxford-developed candidate dosed its first volunteer in the UK last month, with doses manufactured in India and staged for use if trials read out. Practically, that turns this into a screening and isolation problem rather than a treatment one. Your exposure is low but non-zero: Southwest Florida moves international travelers, and the presentation that matters — fever, myalgias, GI symptoms in someone back from Central or East Africa in the last 21 days — is exactly the undifferentiated picture that lands on a night admission and gets anchored to gastroenteritis or malaria.

Action Worth confirming, once, that your ED triage and admission templates still prompt a travel history with a 21-day window and a named country list, and that you know who the on-call contact is for a viral hemorrhagic fever pathway at 3 a.m. Most hospitals wrote that policy in 2014 and haven't opened it since. Two-minute answer, and you'd rather not be looking for it while the patient is in the hallway.

FDA opens the docket on generative-AI medical devices — and clinicians are explicitly invited

What happened FDA released a discussion paper yesterday on how it might regulate generative-AI-enabled medical devices, covering risk assessment, premarket evaluation, post-market monitoring, foundation models and agentic systems. The proposal sketches a two-axis risk framework and a premarket approach built on competency assessment — non-clinical benchmarking followed by clinical confirmation, rather than a single locked-algorithm clearance. Comments are due Oct. 19 under docket FDA-2026-N-7874, and the agency says it wants input from clinicians and the public, not only manufacturers (FDA, AHA, docket).

Why it matters to you This is the regulatory half of the number in yesterday's report: SHM found 64% of hospitalists using AI clinically and about a third trained on it, with decision support the most common use. January's revised clinical-decision-support guidance moved a large class of single-recommendation tools out of premarket review on the theory that a clinician reviews and finalizes the output — which means the review step is now the safety control, and the reviewer is you. Whatever framework comes out of this docket will decide whether that stays true for generative tools that draft assessments and summarize charts, or whether the agency asks for benchmarked performance before deployment. Nights is the shift where the "a clinician will catch it" assumption is thinnest.

Action If any part of this year annoys you — an AI-generated summary that dropped a code status, a decision-support suggestion nobody could trace — the docket is open until October and comments from practicing clinicians are rare enough to get read. Ten minutes, no institutional sign-off needed.

Split rounding teams get a serious look — and the model is basically your job description

What happened A piece out yesterday walks through hospitals separating the admitting function from the rounding function instead of asking one team to do both. At Mount Sinai, physician assistants staff a dedicated admitting team by day and nocturnists fill that role overnight. Andrew Dunn, MD, chief of hospital medicine for Mount Sinai Health System, makes the efficiency case — "everyone gets to focus on their primary task." Christopher Migliore, MD, at Columbia University Irving Medical Center prefers mixed teams, arguing the split costs continuity (Medscape).

Why it matters to you Nocturnists have run the split model for years without anyone calling it that — you admit, you don't round, and the continuity cost gets paid at handoff. What's changing is that day teams are now adopting it deliberately, which shifts where the seams are. If a daytime admitting team owns the front door until 7 p.m., your admissions arrive later and denser; if a rounding team owns the list, the person receiving your handoff has less context on the patients you admitted than the resident who used to do both. The continuity argument Migliore makes is the one worth tracking, because in a split system it lands almost entirely on the quality of the written handoff.

Action Decent group-meeting item if your service is considering a daytime admitter: ask what happens to the night-to-day handoff structure, since that's the interface that absorbs the change.

Section B — The Debate

Should generative-AI clinical tools reach the bedside without FDA premarket review?

Yesterday's FDA discussion paper reopened a question the agency appeared to have settled in January, when it expanded enforcement discretion for clinical decision support software. The paper now floats competency benchmarking and post-market monitoring for generative tools — partial re-regulation of a category the agency had just stepped back from.

For a lighter touch

Requiring premarket review for every AI tool that suggests a diagnosis or drafts a plan treats software that assists a physician the same as software that replaces one, and the statute never asked for that. The January guidance drew the line at substitution: tools that surface one clinically appropriate recommendation, with the source reasoning available and a licensed clinician reviewing, revising and finalizing the output, stay outside device review — while anything that displaces judgment or drives time-critical care stays in. FDA Commissioner Marty Makary, announcing the change at CES in January, framed it as the agency needing to "adapt with the times" and said it would "promote more innovation with AI in medical devices" (STAT). The practical claim underneath: a review queue that treats a charting assistant like an implantable keeps useful tools out of hospitals for years while the harm it prevents is already handled by the clinician's signature.

For real oversight

The clinician-review backstop only works if the clinician can see what the model did, and that is precisely what is being removed. Michael W. Craige, writing in Penn's The Regulatory Review on Aug. 6, notes that HHS's proposed HTI-5 rule would roll back the decision-support transparency criteria adopted only a year earlier — the first federal requirement that certified EHRs disclose how predictive AI works. He pairs that with two numbers: fewer than 2% of cleared AI devices were supported by randomized trials, and hallucination rates for LLMs used in clinical decision support are estimated at 8–20% (The Regulatory Review). His argument is a governance one rather than a safety-scare one: transparency requirements exist so procurement officers, compliance leads and quality committees can compare vendors and audit deployed models. Strip them and the diligence burden lands on hospitals that have no mechanism to carry it.

Where it might land Comments on the generative-AI framework close Oct. 19; the HTI-5 rule is separately pending at HHS, so the transparency question and the device question are on different clocks.

Section C — On My Radar

Mets 2, Padres 5 (8/18). Fernando Tatis Jr. hit the first pitch of the game 452 feet into the second deck — his 18th career leadoff homer, a Padres record — added a two-run shot in the fifth, and in between reached over the right-field wall to take a grand slam away from Lindor. The loss ends a four-game Mets win streak; they came in 57-69, fifth in the NL East (ESPN). Jonah Tong had won Monday's 2-1 game.

Iran / Hormuz. Parliament speaker Mohammad Bagher Ghalibaf said Tuesday the strait stays closed until Washington meets the conditions of the expired MoU, while Treasury Secretary Scott Bessent signaled a new sanctions wave as early as this week and Tehran said it could shift to offensive operations (Al Jazeera, Al Jazeera).

Puerto Rico: rain finally, rationing still. A tropical wave was expected to bring showers and thunderstorms to Puerto Rico and the USVI late Tuesday into Wednesday, and NOAA's tropics outlook puts development odds in the eastern Atlantic at 20–40% late this month into early September (St. John Source). The 48-on/48-off rationing for 180,000+ customers across San Juan, Carolina, Trujillo Alto, Gurabo, Juncos, Canóvanas and Loíza is still scheduled through month's end; roughly 68% of the island is in drought and 60–65% of treated water is lost to leaks (Yale Climate Connections). One wave won't refill Carraízo.

Puerto Rico, briefly. Year-to-date homicides stand at 281, seven fewer than the same date last year; House Speaker Carlos "Johnny" Méndez listed tax reduction, school modernization, working-class housing, water infrastructure and Roosevelt Roads redevelopment as priorities for the second 2026 legislative session (San Juan Daily Star).

Cuba: nothing new in 48 hours. No fresh mainstream reporting since the sanctions action earlier this week. Background conditions unchanged — six nationwide grid collapses so far in 2026, IMF projecting a 7%+ contraction. Filed in Section H rather than padded here.

Movies & TV. Quiet stretch. Lanterns premiered Sunday on HBO Max to mixed reviews — Forbes's recap headline was "Not That Super And Certainly No 'True Detective'" — and It's Always Sunny in Philadelphia returned for season 18 on Monday. The month's real releases are still ahead: Coyote vs. Acme and Yeon Sang-ho's return to the zombie genre both land Aug. 28 (Rotten Tomatoes).

Golf tech: no new developments in 48 hours — 18th straight day. The pipeline is still back-loaded: Uneekor's AIMY AI-coaching prototype and Rapsodo's PC sim software remain unreleased, and the current value picks are unchanged (Shot Scope LM1 at $200, Blue Tees Rainmaker at $599) (Golf.com).

Section D — Trends to Watch

AI oversight is migrating from the gate to the buyer. January moved single-recommendation tools out of premarket review; HTI-5 would trim the transparency disclosures; yesterday's paper asks the field to help define competency benchmarks. Each step individually is defensible, and the cumulative effect is that the entity responsible for validating a clinical AI tool is the hospital that bought it — an institution with no standing mechanism for that work and, per SHM, a two-thirds-adoption / one-third-trained workforce running the tools in the meantime.

Detection lag, not pathogen, is setting outbreak size. Ebola declared May 17 but circulating since February. Measles running 2,500+ on a curve that started in 2025. Cyclospora at 9,481 before the dashboard caught up. In each case the organism behaved predictably and the surveillance clock is what varied. It's a useful prior when the next "new" outbreak number lands: ask how old the transmission is, not how big the count.

Hospitalist workload is being answered with task specialization rather than headcount. Split admitting/rounding teams, PA-staffed admitters, virtual hospitalists — all restructure who does what rather than adding bodies. That's cheaper and probably more efficient per encounter, and it multiplies the number of handoff interfaces per patient-day. Watch whether continuity metrics get measured alongside the throughput ones.

Section E — Ideas & Opportunities

Comment on the FDA docket before Oct. 19. The framework's weak point is the unsupervised overnight case — decision support used by a solo nocturnist with no second reviewer until morning. That's a scenario almost nobody submitting comments actually works in. FDA-2026-N-7874.

One question to infection prevention: does the current admission template prompt a 21-day travel history, and does the VHF pathway have a named after-hours contact? Pairs with the isolation/testing conversation already on your list from Sunday.

Post-nights watch: nothing new worth queuing this week. Coyote vs. Acme lands Aug. 28 if you want something undemanding on the calendar.

Section H — Background / Already Covered

Older items and previously reported context

Ebola, earlier coverage (8/13): WHO reported the outbreak began spreading in February, deaths approaching 2,000, case fatality 60–70%. Yesterday's items — crossing 5,000 cases, becoming the DRC's deadliest, the emergency committee reaffirmation — are in Section A.
Kindergarten vaccine exemptions (8/17 CDC release): 4.2% of kindergartners with one or more exemptions, fourth consecutive record; MMR coverage 92.4%; increases in 41 states plus D.C. Covered in full yesterday.
COVID indicators: wastewater up 106% August vs. July and classified high nationwide; CDC's Rt model has COVID growing or likely growing in all 50 states as of 8/12; aggregate ARI still reading "very low." No new data release since.
Measles: 2,566 confirmed cases as of 8/13, 38 outbreaks in 2026, 94% outbreak-associated. PAHO review of elimination status still pending later this year.
SHM 2026 Workforce Experience Report: 64% of hospitalists using AI clinically, 73% of those reporting benefit, roughly one-third reporting adequate training. Referenced in Section A and B today.
Cuba: sixth nationwide grid collapse of 2026 occurred Aug. 3 (11th since late 2024); IMF projecting roughly 7% GDP contraction this year; tourism down from 4.7M visitors in 2018 to 1.9M in 2025. Nothing new in mainstream reporting in the last 48 hours.
Puerto Rico water emergency: National Guard distribution mission, FEMA water trailers, DACO price freeze, Carraízo more than 11 feet below full pool. Rationing background for today's Section C item.

An outbreak that had a three-month head start, a comment box that's open until October, and a grand slam that never made it out of the park. Sleep well, Marco.