[cyberax]

"Worth watching" implies that watching is expensive. It's really not. A full-body MRI scan is about $1k, and it can be even cheaper.

So if you have abnormal findings in 10% of patients that merit follow-up scans, you can trivially do a series of 3-4 scans without affecting the overall cost too much.

Doctors simply need to get out of the headspace where MRIs are extremely scarce tools of last resort and treat them like we treat blood tests.

[jmward01]

I totally agree. US healthcare is broken and costs aren't tied to the reality of how expensive something actually is. I have very high hopes that modern medicine is in for a massive disruptive change where things like full body MRI, along with analysis, could be done very cheap and with no admin overhead. In that model 'we see something we aren't sure of. It is probably nothing but to be sure we want to do follow-ups' is far less of a problem.

A lot of this however is how it is discussed with the patient. Discussions about the likelihood of there being a real issue when something is seen need to be clear and informative without being alarming. 'We did a routine scan and these often show transient artifacts that turn out to be nothing, but in an abundance of caution we want to do a followup' is totally different than 'we saw something we are concerned about and need to do a followup'. How things are messaged really matters.

[lostlogin]

> MRIs are extremely scarce tools of last resort and treat them like we treat blood tests.

How would this work?

I can do a blood test and send it to the lab to be processed in ~5 minutes from the moment I meet the patient. Consumable costs are about $2.

I can also do an MR scan. It took a fair bit of training and the scanner and scan room cost about US$2 million. Service contracts on the scanner, scan room, chillers and required staffing utterly dwarf the cost of the scanner over its lifetime.

The scan takes 20-75 minutes. Then the images get sent for reporting. Unlike a blood test, reporting isn’t automated. Even if it was, how could availability of MR ever be similar to a blood test?

[cyberax]

> I can do a blood test and send it to the lab to be processed in ~5 minutes from the moment I meet the patient. Consumable costs are about $2.

This depends on a blood test. Bacteria cultures or PCR tests still take more time.

A mid-range scanner costs $500k, the room itself indeed might cost more. Just as real estate. Scanners are just not scarce anymore, there are even sites that sell used ones: https://prizmedimaging.com/collections/mri-equipment (I now want one in my backyard...)

You can lease a new top-level device at around $30k per month: https://www.meridianleasing.com/equipment/medical/mri-machin...

So you're looking at maybe $1k a day that you need to pay towards the device cost. The consumables (helium) are pretty negligible.

A full-body scan is about 1 hour. But for a follow-up you will need to focus only on a few areas, reducing that to maybe 20 minutes. So one device can feasibly do 10 primary scans a day and 20-30 follow-ups. So the cost of the device itself becomes on the order of $100 per imaging session.

This is literally in the "blood test" expenses range.

> Unlike a blood test, reporting isn’t automated.

Radiology readings is one thing where AI is already making inroads. And radiologists can be located anywhere, it's a perfectly remotable job.

> Even if it was, how could availability of MR ever be similar to a blood test?

Yeah, indeed. How can we imagine that people will have computers on their _desks_ when even a small IBM takes half a building?

Mass production happened. And this time it has taken the industry completely by surprise.

[lostlogin]

I manage an MR service. Not a large one, but it keeps me busy.

>A mid-range scanner costs $500k, the room itself indeed might cost more. Just as real estate.

The room is about $250k to get it fitted out and the Faraday cage built.

> Scanners are just not scarce anymore, there are even sites that sell used ones: https://prizmedimaging.com/collections/mri-equipment (I now want one in my backyard...) You can lease a new top-level device at around $30k per month: https://www.meridianleasing.com/equipment/medical/mri-machin...

The Skyra is end of life and the ones I know of are getting pulled out. Too old. The generation after it is getting on too. The new generation will appear next year. Likely ‘dry’ magnets - low helium usage. The Siemens 1.5T has already had this upgrade.

> So you're looking at maybe $1k a day that you need to pay towards the device cost. The consumables (helium) are pretty negligible.

A service contract is expensive and also required. I’d expect them to be up to about US$75-100k a year. Coils are US$10-20k each and a scanner has about 10 different ones. We blow about one a month across our scanners and get them replaced. Various things break, and it comes in waves. In recent months: the body coil blew (this is a big deal, $150k?), gradient amp, amp cables, helium expansion vessel, vacuum pump, bore screen, knee coil, another knee coil. The service contract covers all the repairs.

>A full-body scan is about 1 hour. You can do them in 30-45mins, improving the business case.

But for a follow-up you will need to focus only on a few areas, reducing that to maybe 20 minutes.

Doing a real scan of a body site is vastly better, and is about 20 mins per region.

> So one device can feasibly do 10 primary scans a day and 20-30 follow-ups. So the cost of the device itself becomes on the order of $100 per imaging session. This is literally in the "blood test" expenses range.

We get around that number out of our scanners doing real scans and our costs are a lot more than that.

I don’t see staffing mentioned? A good tech will improve throughout, and you need more than one. A good operation will have several per scanner. They reduce the ever present danger and keep the operation running smoothly. I’m biased, I’m a tech.

Another thing missing - liability. I work in a country that isn’t quite as full of lawsuits as the US. Whole body scans are a glorious way of collecting liability. Poor quality images of everything in ‘well’ patients. As a rule in imaging, ask a good question get a good answer. When you’re searching the whole patient for something’s it’s easy to miss something.

> How can we imagine that people will have computers on their _desks_ when even a small IBM takes half a building? Mass production happened. And this time it has taken the industry completely by surprise.

The radiologist needs to be onsite. We give drugs. They inject things, they biopsy, they are doing stuff.

Yes, lots of stuff can be done remotely and AI assisted. Assuming they can all be replaced assumes a role that the radiologist had a long time ago, or maybe never had.

There are a lot of efficiencies that can be made, and radiology needs all the help it can get, but it isn’t a ludicrously inefficient or stuck in the past.

[cyberax]

> I don’t see staffing mentioned? A good tech will improve throughout, and you need more than one. A good operation will have several per scanner. They reduce the ever present danger and keep the operation running smoothly. I’m biased, I’m a tech.

That's why I put the cost of the scan at $1k, to make room for a profit margin and the staffing cost (which will be more than equipment).

Liability is an issue, but not an insurmountable one. If you get poor-quality images, you can just retake them. So biasing radiologists towards that can be a workable solution. There is also a lawsuit in the US going on against one of the full-body scan companies that missed a cancer, so we'll see how it turns out.

> The radiologist needs to be onsite. We give drugs. They inject things, they biopsy, they are doing stuff.

For clinical MRI. Not the diagnostic screening ones, they are done without any contrast.

> There are a lot of efficiencies that can be made, and radiology needs all the help it can get, but it isn’t a ludicrously inefficient or stuck in the past.

Radiology is fine, the AI will help them to do quick screening and point out possible pathologies.

[tptacek]

I think you're missing the point. The psychological cost of a conditional-positive result is nonzero, and can be very significant (I speak from a little bit of experience here). But far more importantly: the physiological cost of invasive followups when you eventually trip the threshold of "time to go explore with a scalpel" is very high, and the missing evidence this story is about is whether you can get to that threshold with an MRI.

Treating MRIs the way we treat blood tests would almost certainly result in huge numbers of needless invasive procedures.

[cyberax]

> Treating MRIs the way we treat blood tests would almost certainly result in huge numbers of needless invasive procedures.

Again, _all_ you need to do is to make a follow-up scan in 1-3 months to see if there are any changes. It's a preventative tool, so unless you have other indications, it's almost always safe to wait for a bit.

And yes, it requires educating patients that sometimes just waiting and doing a follow-up scan is right. And yes, I also have a personal experience with that (I had an "idiopathic lymphadenopathy", aka "we don't know WTF is going on").

[amanaplanacanal]

It's a good thought experiment, but what you really need here is a randomized controlled study to see if your new plan results in better outcomes, before you roll it out to the whole world.

[tptacek]

* Doctors and medical researchers keep saying that routine MRIs for non-symptomatic or low-risk patients is a bad idea, because the outcomes are worse than not scanning.

* There's several clear, understandable mechanistic reasons why this would be the case, including simple applications of the base rate fallacy.

* Nevertheless, here we are, nerds arguing we know better than all these people.

[cyberax]

And I don't buy that. EXACTLY this was also said about prostate cancer screening. Word-for-word. "Overdiagnosing", "people are better off not knowing", "psychological burdens", "invasive procedures", etc.

Now we have multiple longitudinal studies of people receiving aggressive screening and the usual standard of care. The aggressive screening group, unsurprisingly, has better outcomes with less mortality.

> * There's several clear, understandable mechanistic reasons why this would be the case, including simple applications of the base rate fallacy.

Yeah. I guess it's time to stop using fire alarms. People are better off not knowing if a building is on fire, and frequent false alarms have a negative effect on psychological well-being.

Just look at the risk of burning to death. Alarms make no sense at all!

> * Nevertheless, here we are, nerds arguing we know better than all these people.

Yes. Absolutely. And I actually have read (I think) all the studies, and came thoroughly unimpressed. They're utterly sloppy with poor statistical analyses.

It's absolutely an indictment of the medical industry that has become so crusty that it can't be bothered to integrate new diagnostic modality. That is the _only_ way to detect multiple lethal cancers while they are still curable.

[tptacek]

You're not engaging with the logic. Stipulate that it is the only way to detect multiple lethal cancers. If you end up harming more people than you help, the intervention is bad. Right now, you're only looking at one half of the balance sheet.

[cyberax]

> You're not engaging with the logic.

Ditto for you.

> Stipulate that it is the only way to detect multiple lethal cancers.

This is trivially true.

> If you end up harming more people than you help, the intervention is bad.

The only pathway through which diagnostic MRIs can feasibly harm people is aggressive follow-up of uncertain findings. And this is almost completely solved by just doing another scan several weeks/months in the future. With corresponding patient education.

This is literally all what was needed in the case of prostate cancer screening: less aggressive biopsies and bias towards observation rather than action. Yet it took _two_ _decades_ to arrive at this point. And some doctors _still_ refuse to order screening tests out of this misplaced idea of "not knowing is better".

And this is not the only time when "geeks knew better". For example, checklists for surgeries are a no-brainer to anyone with an aviation background. Yet they became standard only two decades ago ( https://pmc.ncbi.nlm.nih.gov/articles/PMC6032919/ )! Over rather strong objections from doctors.

[lostlogin]

This is pretty much what the experts say:

https://www.ranzcr.com/college/document-library/2024-positio...