benmaraschino

You (and others) may enjoy going down the rabbit hole of universality. Terence Tao has a nice survey article on this which might be a good place to start: https://direct.mit.edu/daed/article/141/3/23/27037/E-pluribu...

They did. Check out Figure 2 in the paper.

They used a crossover design, so each subject served as their own control. Not a bad choice for trials like this as you gain a lot of statistical power with fewer participants than a parallel-arm, non-crossover design.

This book isn’t specifically about how to best support a blind child, but it might still be of interest.

https://en.wikipedia.org/wiki/Far_from_the_Tree

Either way, I hope for nothing but the best for you, your son, and your family. It’ll be tough, but you’ve got this and your son is lucky to have you as a parent.

Counterpoint: the investigators used a crossover design, where each participant acts as their own control. Since within-subject variation <<< between-subject variation, this makes the effective sample size much, much higher than it would seem at first. In terms of efficiency, their design’s potentially equivalent to a vanilla parallel-arm trial with close to 100 subjects (or more) in each arm. So, you definitely can obtain solid evidence from N=15 (or less), if your study is carefully designed.

Sample size is only a very small part of the overall methodological picture—design matters too.

Yes, this is a part of the problem. This is also why it’s hard to interpret the results of studies like this.

Without knowing more about the methodology, it’s possible that a lot of the apparent variation in cost arises from different ways of paying for the same thing. Take a joint replacement surgery—the surgery itself will generate claims for both the surgeon and the hospital, and maybe also the anesthesiologist. Each of those claims could be counted as a bill. But there’s also the initial clinic visit before the surgery as well as the aftercare and rehab, which also generate their own bills. What’s more, some payment models bundle all of those costs into one bill for the entire episode of care. (See how there’s no range listed for the procedures performed at Kaiser?)

The problem arises when you try try to compare those individual claims to the episode of care bills. You can’t, because they’re basically apples and oranges. But if you’re not aware, that can make it seem like there’s more variation than there really is. Don’t get me wrong, though, there is way too much variation in what we pay for health care, but it’s important we measure it precisely.

There are payment models like this in use already. Look into capitation and accountable care organizations.

The causation arguably goes in the other direction too. You can't make money if you're unhealthy at baseline. Nor can you as easily acquire more education or new skills.

Yup. We don't know the full details, and I don't want to speculate any more than I have to, but it needs to be pointed out that subdural/epidural hematomas caused by head injury can take some time to kill you. Remember Natasha Richardson?

In any case, RIP Officer Sicknick.

Aside from the fact that this demonstration was limited to the semi-public atrium of the Hart Senate Office Building—historically, a location where many such demonstrations have been held, and all peacefully, to my knowledge—I'm willing to bet you that none of those demonstrators are armed. Nor did they arrive with zip ties to take hostages—didn't you see the footage of those armed thugs sweeping the Senate chamber looking for potential hostages? Nor did they rampage through the building, with the intent of targeting specific politicians for violence.

They didn't damage property, either. Nor did they impede the democratic process which we all put so much stock in. There's just absolutely no equivalence between these two events. I don't know if that was your intent or not, but there's absolutely no comparison.

The issue with coatings and lubricants is that you’d need to use a nontrivial amount for them to be effective, and that means the patient is at risk of aspirating the substance. That happens, and boom, you’ve got aspiration pneumonia. Even with some kind of encapsulated or solid gelatin coating you’d have to be very careful to make sure it doesn’t leave any kind of residue or break into pieces that could be aspirated.

One of these "studies" is a 2-page Google Doc with no real data. Another is the discredited Raoult study, which is under investigation by the journal [1]. Another study cited is also by Raoult's team. The Brazil study (ref. 29) doesn't seem to have passed peer review (yet?) and has severe methodological flaws. Like the Raoult studies, the Brazil study was not randomized, but makes no attempt to control for confounding by indication or self-selection, such as with propensity score matching or similar methods commonly used in observational studies. The last study cited isn't a study, but an article from a local news station.

[1] https://retractionwatch.com/2020/04/06/hydroxychlorine-covid...

It’s nowhere near as difficult to make as the article may imply, though. Yes, that first-generation route is quite a doozy, involving some harsh conditions (multiple steps at -78° C, which is actually not as difficult to carry out at the bench as it sounds) and nasty reagents, especially n-BuLi, which explodes into flames upon exposure to air. But it’s fairly easy to carry out on a small scale in a typical med chem lab. If you’re a medicinal chemist and need to make limited amounts, say, maybe in the hundreds of milligrams—which would probably be far more than enough to do preclinical studies—then that route will do. But it definitely can’t be scaled up easily to multi-thousand gallon chemical reactors, although like the article says, that’s where the process chemists come in. The second-generation route looks much better, and I’m sure it’s already being optimized even further.

Unfortunately, this is a really bad analysis, because, for one, the author assumes in his model that the total number of infections won't change as the result of greater social distancing. But distancing will have an effect on the number of infections, and this effect is in fact quite substantial. For example, the results of this model suggest a ~80% reduction in total infection burden with a 50% reduction in the rate of transmission, and nearly 95% with a 75% reduction: https://twitter.com/IDMOD_ORG/status/1237881591784865793

You the nail right on the head. Especially regarding VALI. Personally, I’d also be worried about any residues in the system (e.g. lubricants, grease, etc.) being blown into your lungs. That’d be an excellent recipe for a hellish aspiration pneumonia on top of the underlying illness.

> Owning in a high cost of living geographic area is not a right, it is a luxury obtained mostly by luck

This is kind of a weird take. Whether someone can afford to fulfill as basic of a need as housing shouldn't be about luck, no matter the qualities of the area concerned. It's kind of like saying whether someone can afford to eat on a given day should be based on luck—we would think that's crazy, and for good reason. The housing crisis is currently ripping families and friendships apart, not to mention fraying the very fabric of civic and community life—all so a select few lucky folks can feel like they've made it. House prices in the Bay Area (and in other currently high-priced metros) will revert, sooner or later, but in the meantime, we're willing to cause a lot of pain and suffering in the name of picking winners and losers, if only momentarily. And speaking as someone who owns a home in a desirable part of the Bay Area and is 100% pro-housing, I don't really get it. It's really sad to see a lot of my neighbors supporting regressive anti-housing policy—it's almost like sociopathy on a massive scale.

Likely there would be little to no difference, because as it turns out, outcome risk is an awful predictor of benefit, especially among high-risk patients. Many high-risk patients will be readmitted to the hospital anyway, no matter how much preventative care you throw at them. On the other hand, there are potentially many preventable readmissions among ostensibly low-risk patients, and you can prevent more total readmissions by including them in your targeting as opposed to focusing on a tiny high-risk group.

(A big part of my current PhD research involves building predictive models for patients with complex needs, so I grapple with these issues day in and day out, and often on huge scales—millions of patients. Prediction, at least in the usual supervised ML sense, just doesn't cut it for these kinds of problems—you need causal inference.)

It’s not practical. Patients on ECMO require anticoagulation, as do patients on cardiopulmonary bypass. This is a problem common to all systems which expose blood to foreign surfaces and pass it through pumps. Your hypothetical anticoagulated soldier with such a pump would bleed out lickety-split were they to suffer as much as a minor cut.

Also, such a system would likely require giant cannulae placed in rather inconvenient places, e.g., the neck, thigh, or chest, as in ECMO. That would limit movement significantly.

Agreed. The point of criticism that the authors didn’t include all relevant models is among the lowest of the low-hanging fruit I can imagine one picking. And while the process of peer review isn’t without its flaws, I can also imagine that fruit has long since been picked by the reviewers, who have far deeper knowledge of the published models than I or just about anyone else in this thread, save for the actual climate scientists among us.

Exhibit A: Vimovo. One pill, consisting of two drugs that have been around for decades, and are available over the counter at any pharmacy. If bought separately, a month’s supply of these drugs would run you maybe $40, at most.

Vimovo, on the other hand, is routinely charged to insurers by pharmacies at prices ranging to over $3,000 for a month’s supply. And somehow insurers are willing to pay for it, to the tune of nearly $500 million a year. More details here from ProPublica: https://www.propublica.org/article/horizon-pharma-vimovo-com...

The way many clinical research groups are structured (unfortunately) precludes code review, and to some extent, version control. Often for projects like the one referenced in the OP you'll have just one statistical programmer working with a PhD-level biostatistician in addition to the MD investigators. The biostatistician guides the programmer through the statistical methods to use and steers the overall study design, but otherwise they never see the underlying code. There are some exceptions, but in many cases, the programmer ends up the only person seeing the code. A lot of this has to do with how funding is structured—grants are written assuming one FTE programmer, and something like 0.05 FTE for the biostatistician. It's hard to convince funding agencies that you need more than one FTE programmer in many cases; on top of that, farming out the biostatistican's time in small chunks like that splits their attention between dozens of projects, which precludes them engaging with any one of those projects in depth.

So, there's literally nobody else on these projects who could conduct a code review. This also sort of provides a disincentive towards using VCS even though it's so obviously a good idea if you're the only person contributing code—I've talked with programmers about this before and the response is "why bother?" unfortunately.

Not the OP, but I work on similar problems, albeit in a different setting (healthcare, millions+ of patients). The gist is that you have to bake experimental design into the deployment of your ML model, but in many cases a simple RCT or A/B test just won't cut it. This is largely because when you deploy a model, no matter how sophisticated or accurate, there's no guarantee that it'll actually move the needle in terms of the outcomes you care about—hence you need to run some kind of trial. At the same time, you want to maximize overall utility by not having to allocate more subjects to your control arm (or harmful, or resource-intensive and ineffective treatment arms) than you need to. This latter point is much more of a problem in medicine than it is in other settings, as you can imagine. These considerations point to adaptive designs that balance exploration/exploitation, e.g. those based on multi-armed bandits. Currently working on some cool (in my opinion) variations of MABs that incorporate domain-specific knowledge, so I could talk about this all day!

No way. It's not even close. And I say this as a PhD candidate in one of these fields at Stanford. Grad school is what you make of it, to a large extent. There's definitely toxicity, but it's not as systematic or ritualized to the extent that it is in medicine. In grad school, you have wide leeway over who you work with, and you can distance yourself from the toxic personalities if that's what you want to do without taking a career hit. That's not really possible in medicine.

Well, at the end of the day, you can never really be sure of strongly ignorable treatment assignment/unconfoundedness, no matter what problem you’re working on. Especially if you’re an economist or an epidemiologist working with data that’ve been collected by someone else—you can’t exactly easily go back and measure more predictors of treatment assignment. But if you’re running a website, there are a lot of variables you can measure on the user end and more opportunities to iterate, and so SITA then begins to look more and more like a better bet.

While it’s often an ostensibly appealing point of criticism, it’s important to note that sample size is just one aspect of study design—the investigators used a crossover design, which substantially increases statistical power. In the Methods, they state that 10 subjects are sufficient to achieve 80% power, which (criticisms of NHST aside) is a generally accepted standard, i.e., to obtain NIH funding for a study, you must demonstrate in your grant application that your design achieves that 80% power benchmark.

Not always true. That OOP max often only applies to providers who are considered in-network for your insurance plan. Oftentimes, there's no way to guarantee that all the providers you see during an episode of care will be in-network; you could have a surgery with a surgeon who's in-network, and at a facility that's in-network, but get blindsided by a huge surprise bill from an out-of-network assisting surgeon or nurse or anesthesiologist, or by an out-of-network consult when you're in the ICU recovering after surgery. This is one example: https://www.nytimes.com/2014/09/21/us/drive-by-doctoring-sur...

There's been some progress in states (including CA) to cut down on these sorts of billing practices, thankfully, but it's still a major issue elsewhere, even if you have an ostensibly gold-plated health plan. Many plans are starting to remove caps on out-of-network bills, too: https://khn.org/news/2016-ppo-plans-remove-out-of-network-co...

I assume something similar may have happened to OP, but there are a lot of other ways they can run up the bill. These kinds of billing practices are insane and need to stop.

Denmark's healthcare system tried to implement Epic nationally, and if this article is anything to go by, it didn't go so well: https://www.politico.com/story/2019/06/06/epic-denmark-healt...

Essentially, different healthcare systems have different needs which are reflected in how their electronic health record systems are architected. For example, EHRs in the US are optimized for billing, and if you're moving an US-centric EHR, like Epic, to a country like Denmark with universal healthcare, all that functionality becomes superfluous. Not to mention the language differences, as well, which are touched upon in the article. It's possible that similar concerns played into the NHS's decision to build their own system. Some NHS trusts are moving forward with their own Epic implementations, however. For example: https://www.digitalhealth.net/2018/06/royal-devon-exeter-nhs...

Comparison shopping arguably isn’t the intended purpose of advance care directives—which, FYI, can only go into force if the person is legally incapitated, and being in pain, however extreme, is a grey area as I’m sure you could make the case that decision-making capacity still exists. There’s also the issue of having the agent available when such an event happens, and having them be in possession of the directive.

Anyway, even then, it’s infeasible to enumerate all the possible decisions in advance—not only is it impossible to know what your costs will be ahead of time, there’s a lot of variation in cost. There isn’t going to be a single hospital in your region that’ll have the cheapest prices across the board, so it’s not as simple as “send me to the cheapest hospital”.

I honestly don’t get it either. Medicare takes care of some of the sickest and most complex patients—the elderly and those with ESRD—and manages to do it with an administrative overhead of 2%, compared to 17% for private insurers, and appears to be more efficient overall as well.[1] That’s a simple fact. Expanding Medicare to cover all Americans would substantially increase access and drive down costs for everyone, simply by adding younger and healthier individuals to the risk pool.

[1]: https://www.healthaffairs.org/do/10.1377/hblog20110920.01339...

While Epic produces the database/electronic health record system, the hospital or health system ultimately owns the data. AFAIK patient data don’t sit on Epic’s own servers outside of some research/pilot initiatives.

(Source: my personal experience working extensively with EHRs, including Epic products.)