justwantaccount

I thought this article was going to talk about GloVe, which actually performs better than Google's word2vec without a neural network according to its paper, but I guess not.

I'm in the healthcare industry, so I try my best to pay attention to it through news and blogs and such, though I don't claim to be an expert in the field.

My impression is that price transparency is the least of the healthcare industry's problems. On aggregate, Medicare pays out about 25% of what the hospital or physician asks for, and private insurances apparently about 50%. The pricing for the uninsured is the initial price that hospitals/physicians set to get the maximum they can out of different insurance companies, and are not meant to be the charge for uninsured patients. There's been studies talking about how price transparency doesn't change physician behavior in hospitals that tried it out. Some patients report trying to compare prices for procedures across physicians and hospitals, but didn't get the quality of treatment due to picking the cheapest one.

Instead, the #1 concern you hear about all the time is EHR (electronic health records). It's the #1 reason physicians give for feeling burned out, according to some surveys. The popular opinion on electronic healthcare records seem to be that they're glorified cash registers for keeping track of claims data to bill insurance, but doesn't contain much clinically relevant data - for example, I hear physicians complaining online about how they can find when a patient went to an ER and through what means, but they have trouble finding what clinical reasons the patient went to the ER for. Granted, apparently a few percentage of physicians think that their EHR contain clinically relevant data, but most EHR systems that people are using right now aren't that sophisticated, usually due to cost and to the stupid time it would take to update to the new system and train people to use it properly. Word is that it takes years for a practitioner to learn how to use the EHR system (EPIC, Cerner, etc.). Different hospitals use different systems, which makes interoperability impossible, since there aren't good standards for recording EHRs, or they're over-complicated, or something according to people who develop EHRs. I also hear hospital C-level people talking about how expensive EHR systems are - one hospital reported upgrading to a new one cost about half a billion dollars, which was a third of their net patient revenue. Independent physicians are opting for direct primary care models (where you directly pay the physician some monthly fee, circumventing insurance) or to be employed by larger hospitals or physician groups, because the federal government requires that you try to implement some level of EHR, which is too expensive for independent physicians that can't afford hiring IT people on top of everyone else. Some surverys say that physicians spend half their time on looking at / putting data into EHRs, which can be twice as much time as they spend on seeing patients face-to-face.

Health insurance billing being extremely complicated doesn't help, either. You have to learn CPT/ICD-9/ICD-10 codes to bill correctly and get reimbursed, and it's so complicated that apparently some people get paid ~$40,000 a year just to convert diagnoses/procedures to CPT/ICD-9/ICD-10 codes. Hospitals/physicians end up needing to pay for those jobs as well, on top of IT people for maitaining/upgrading EHRs. In fact, there's a lot of consultation groups out there that advise you on how to bill insurance / submit required documents for Medicare / set up EHR.

There's a lot of other administrative overhead costs that contribute to healthcare costs seems to need higher priority, to me.

I learned growing up in Japan in the 90's that the Ainu are indigenous people of Hokkaido, the northernmost island in Japan. Okinawa and Hokkaido weren't really Japanese until recently. I think I remember learning that Hokkaido was like Australia, where the mainland sent Japanese prisoners/criminals to or something.

Yes, you can conduct NHST without the central limit theorem. However, it's used very widely in NHST. Was there anything wrong with what I said about what a typical t-test usually looked like? My lab would use that approach to do molecular biology.

You don't seem to understand my argument, so let me rephrase: The example about Native Hawaiians was meant to highlight the nature of conditional probabilities, and p-values are conditional probabilities. Just because p-values are below some threshold doesn't necessarily mean that the null hypothesis is incorrect and therefore should be rejected. Just because p-values values are high doesn't mean that the null hypothesis should fail to be rejected. P-values do not theoretically give that information. It doesn't even represent the probability of observing that value, since it's a conditional probability - as in, the probability of observing that value given that the null hypothesis is true, not the probability of observing that value. If the p-value is below 0.005, can you scientifically, theoretically conclude that the null hypothesis should be rejected? The probability of seeing a Native Hawaiian person given that the person is a US citizen is below the threshold of 0.005, but does that mean the conditioned part (the US citizen thing) should be rejected? Granted, it's hard to relate that example to actual experiments, but my argument is that p-values don't theoretically give any conclusions either way, and trying to make it "scientific" to draw conclusions by introducing thresholds to a conditional probability, no matter how strict, seems inherently flawed. Using it as a single metric among many, to use it as a tool for exploration makes sense to me. Even to make strong suggestions, sure, especially with all the controls RCTs put in. But to make hard conclusions, as in NHST? The approach itself doesn't have the theoretical power to do so.

Maybe my example was contrived, but my main argument was that p-values represent P(D|H), or probability seeing your data conditioned on assuming that the hypothesis is correct, and that approach seems inherently flawed to me. If you're doing an experiment, p-values don't say anything about if the hypothesis is true or false, in fact the approach can't theoretically give that information. If you're doing a basic t-test, it goes something like this:

1. You have a null hypothesis you assume to be true

2. Based on this null hypothesis and the central limit theorem, if you theoretically conduct many experiments you expect some summary statistic to have a specific distribution around the 'true' population summary statistic and 'true' population variance, which you don't have so you assume it to be the null hypothesis

3. You compare your experiment's summary statistic to the hypothesized distribution. If the probability of seeing your summary statistic is below some threshold, you say it's unlikely to see this summary statistic again.

But the probability of seeing that summary statistic again actually depends on P(H), which NHST doesn't provide any information on. My examples were meant to highlight the nature of conditional probabilities, rather than how real life experiments are conducted.

I personally find NHST suspicious, even if the p-value is less than 0.005. It means that, ASSUMING that the null hypothesis is correct, the probability of observing the data is less than 0.5%. That's still not zero, though. For example, if I try to decide that native Hawaiians are US citizens, and the null hypothesis is that they are, but since only ~0.2% of total US population is native Hawaiian, NHST would conclude that native Hawaiians aren't US citizens. Conversely, the p-value being high isn't really meaningful either. For example, if I try to decide if a white person is from South Africa, and the null hypothesis is that they are, and since ~9% of the population in South Africa is white, I would fail to reject the null hypothesis, even if the person's actually from Europe or literally any other country in the world. p-values are dependent on the sample size, too, and become more and more sensitive to smaller differences in the ASSUMED population distribution(s), when the effect size can in fact be small. NSHT seems highly dependent on quality null hypothesis that's correct, which you can't really establish since that's what you're trying to find out. So the approach doesn't really conclude anything either way, and just really seems weak as an approach to scientific testing overall.