Very much appreciate you taking the time to share your thoughts. However, I would push back here. Most would agree one of the areas of greatest promise for AI in medicine is detection. AI-enabled detection basically involves the AI looking at extremely large datasets to detect perturbations in biological data whose very existence (e.g. a very subtle temporal sequence on an ultrasound with large time gaps between the relevant signals) or correlational relationship with disease is too subtle for human detection. So, the acquisition of massive datasets, especially in unique modalities (and this would be the first of its kind), is one of the most critical bottlenecks to realizing any patient benefit in AI detection. We have two options to get those datasets. We can take the slow, bureaucratic, approach of building those datasets that has been the historical norm and which I think you reference. These datasets can be great in certain respects (like UK biobank), and it is often warranted for data that have substantial risk associated with the data acquisition - e.g. radiation from CT, biopsy etc). However, getting through the bureaucratic morass required is insane, massive imaging datasets are even harder, and hardest of all is building an imaging dataset in a modality that might be an AI-rich target but doesn’t have any utility in a pre-AI / human analysis only world so doesn’t exist. On the second point here, there is a reason Enroll-HD has reached 20-30,000 patients (no imaging), but Track-HD is only a few hundred (imaging). A few hundred person dataset is basically useless for AI. Most critical of all is the reality that if there is a viable diagnostic or therapy at the end of a developmental pipeline, that delay costs very real patient lives. At the extreme end of things, that could truly be millions of lives. That’s not an exaggeration. There are about 100,000 cases of Stage III and IV breast cancer in the US per year. If an AI enabled diagnostic method catches them at Stage I (99% 5 year survival), but that diagnostic method is delayed by 30 years, how many lives are lost bc of a lack of development speed? So yes, incidentalomas can have costs. However, so do delays in innovation. It’s just that no one in the broader scientific community is accountable for innovation delays, and it feels abstract to epidemiologists. I’m speaking somewhat first hand here too. I’ve been working on something that requires one of the most harmless data acquisitions from patients possible and deals with a major disease. It has been the most nightmarish process imaginable trying to get through the bureaucracy needed to get that data, to the point that most of my time is now allocated the ideas that I think will be higher yield over a more reasonable time frame for patients and are not as gated by the bureaucracy. The other option is to do what midstream appears to be doing, which is to build their own dataset with speed by scanning a ton of people in the private market. As long as the procedure is safe (ultrasound of course is safe as a modality), there are rigorous in their disclosures to patients, and their informed consent, I don’t see an issue with this. Sure, mid journey could abuse the system, overstate what they can offer, and prey on the fears of patients. If I were to see that, I would have no problem calling foul. It could be that 6 months from now I see them doing something really bad and think they should shut down, or maybe even that they’ve done something already I’m not aware of, someone will point me to it, and I'll yes that's out of bounds. However, I am not going to castigate them a priori for things they *might**do when they’re actually in the arena trying to drive innovation, and I have no issue a priori with a company trying to accelerate the biggest bottleneck to AI advances in medical detection, dataset creation, in a unique modality no less, by going to the private market. Again, as long as there is no misleading marketing and patients are informed appropriately of what the product does and doesn’t do. And then just to hit again on the incidentaloma point, again, if there is a viable diagnostic at the end of the developmental pipeline, slow innovation and development could very well end up costing far more lives than the as of yet hypothetical incidentalomos. Pick your poison, and I don’t agree with the argument that “Do No Harm” applies any less to bureaucratic hurdles that cause harm bc thousands (or millions!) of patients get a life saving diagnostic years or decades later.

This graph is misleading and much of the analysis here is either inaccurate or incomplete. First, the y-axis is showing total reported cases of acute Hep B cases in the US, not infant Hep-B cases. I understand you likely misread the title of the graph but it should have immediately struck you there were never, ever ~26,000 reported infant Hep B cases in the US. The number of annual infant / perinatal cases before the first Hep B vaccine was developed in 1981 and when Hep B was much more endemic was far, far lower. That number really matters when you are talking about "catastrophic consequences". Second, and most importantly, virtually the same decline seen in infant and perinatal Hep B cases in the US from 1985 was seen in many European countries (UK, Norway, etc.) who used a 1988-esque approach through 2017 of screening pregnant mothers and selectively administrating the Hep B vaccine and HBIG to babies whose mothers tested positive for HBsAG, with a few EU countries continuing that approach into today. To put raw numbers on that, the UK had 17 total child Hep B cases (100% of which were perinatal) from March 2014 to April 2022 (gov.uk/government/publicatio…). Even if all those cases were during selective vaccination and therefore from March 2014 to March 2017, that would be 5.6 cases per year. This is particularly notable given the higher percentage of immigrants that come from Hep B endemic countries to countries like the UK and Norway relative to the US. If you scale that number up to fit the 340 million population of the US, you're looking at about 28 cases per year. Third, in a later post in this thread, you point to the lower reported screening rates (~85%) in the US as being a function of the lack of universal healthcare and a reason to have universal vaccination as you presume screening can't be scaled up. You show no real evidence for either of these contentions. It is much more likely that the screening gap is predominantly driven by the fact that the US has had a long-standing universal vaccination policy which reduces the criticality of screening relative to its criticality in countries with longstanding selective vaccination policies. To this point, analyses of Medicaid mothers in the US ( a proxy for universal healthcare) indicate that 16.2% still did not receive Hep B screening, whilst many of the countries that have close to 100% screening rates took the selective screening approach in the recent past (e.g. England and Ireland) or are still taking it (e.g. Denmark, Finland). Further, Estonia does not have universal healthcare but has screening rates above the 90% WHO recommended threshold. Canada does have universal healthcare, but its screening rate is only slightly above the US at 89%, whilst Hep B rates remain very low despite a province specific vaccination protocol. As perhaps the most critical point, it is mechanistically illogical to think that 98% of pregnant women in the US have at least one antenatal care visit (CDC), ~99% of women give birth in hospitals, to get a baby vaccinated at birth you need to to be in a hospital anyways, and yet 15% of pregnant women can't or won't be screened for Hep B under a selective vaccination policy due to of a lack of universal healthcare coverage. Fourth, the graph on a standalone basis does not provide the strong evidence you claim that the universal vaccination policies implemented in 1991 onwards drove significantly differentiated outcomes in Hep-B rates from what would have been achieved if the 1988 selective vaccination policy had simply been maintained. The first Hep B vaccine was FDA approved in 1981. All therapies take time to achieve widespread adoption which in part explains why peak Hep B cases occurred in 1985, before beginning its decline. Post-1988 (and really even through 1988), the graph fails to show a meaningful acceleration in the decline of Hep B cases following the introduction of a new vaccine policy, which would be fairly likely if the new policy was driving meaningful decreases in Hep B cases relative to 1988-esque selective vaccination. Fifth, it is important to note that from 2011 to 2023 only "confirmed cases" were included in this dataset, rather than "confirmed, probable, suspect and unknown" included pre-2011 which likely contributes some to the decline seen. Sixth, a meaningful percentage of the decline in total acute Hep B cases is likely driven due to declining transmission due to changes in adult behavior. So, in totality, yes, universal Hep B vaccination instead of selective screening likely will prevent a very small number of cases per year in the US, particularly in vulnerable groups like immigrants from countries where Hep B is endemic. The no more than 5.6 annual breakthrough cases in the UK is a referent for the scale we are talking about. We can scale that up to 28 cases for the 5x larger US population and even if we 3x that due to the questionable assumption of a lack of universal healthcare driving decreased screening, that only gets us to 112 cases per year out of 3.6M babies born. However, the marginal benefit of universal Hep B vaccination for the vast majority of Americans is extremely limited, you can likely close most of the gap between selective and universal vaccination by combining an additional vulnerability screen with universal maternal testing, and suggesting the precipitous decline in infant Hep B cases was primarily driven by the differentiated impact of a universal Hep B infant vaccination policy over selective vaccination is really just demonstrably wrong.