Stat Rad

[Tenk]

Is everyone’s Stat Rad return times abysmal the last few months or is this geographic. We got a fax the other night that said due to staffing they wouldn’t read anything for us that night. Just curious if this is global or just my bad luck.

---

Members don't see this ad.

 

[NYEMMED]

It’s global. That’s private equity in radiology

 

[cyanide12345678]

Your bad luck. The worst I’ve had is 2-3 hours. Normally CT scans are back in 1 hour and xrays are back in 30-40 minutes for me.

Our radiologists are centralized to our entire 20 or so hospital system i believe.

The only painful thing is usually neuroradiology things that can sometimes take 1-2 hours.

 

[Tenk]

Your bad luck. The worst I’ve had is 2-3 hours. Normally CT scans are back in 1 hour and xrays are back in 30-40 minutes for me.

Our radiologists are centralized to our entire 20 or so hospital system i believe.

The only painful thing is usually neuroradiology things that can sometimes take 1-2 hours.

Your night rads read X-rays? Are you sure this is Stat Rad and not some other company? There are multiple.

 

[cyanide12345678]

Your night rads read X-rays? Are you sure this is Stat Rad and not some other company? There are multiple.

I didn’t realize stat rad was a specific company. Our rads reads everything at all times but it has nothing to do with stat rad the company

 

[qwerty89]

I'm an attending radiologist. The radiologist shortage is bad and only going to get worse. I don't think people fully understand how bad this is and why it is different from other specialty shortages.

Historically, similar to anesthesia, there was a substantial reduction in the number of residency positions in the mid 90s. This let to a shortage of rads in the early 2000s. However that coincided with the introduction of PACS which was a one-off event that dramatically increased the productivity of radiologists seemingly overnight. That doesn't happen in any other specialty. You can't just have a group of ER docs or hospitalists or surgeons increase their productivity by that degree that quickly. That allowed the specialty to weather the storm and also let to a huge increase in $$$ for radiologists. Then, to my knowledge, the Deficit Reduction Act of 2005 introduced a variety of cuts to specifically to rads (things like outpatient imaging in particular) which led to rads having to spin the hamster wheel faster and faster for the same pay. That combined with and increasing number of residency positions since the early 2000s and delayed retirements after the financial crisis led to a dearth of jobs from 2008-2013ish. 2 year neurorad fellowships being required for terrible 7 on 7 off 10/12 hour shifts were advertised.

The number of residency positions since that time has gone up only modestly. However, the number of imaging studies being done has increased monumentally. The reimbursement per study has also gone done which incentivizes groups to read more studies to maintain income.

But you can only do that for so long.

Demographic changes including more demand for part-time work and less working hours (particularly from women but men as well) as well as increased focus on quality of life rather than pure $$$ means new grads don't want to grind out a career the way boomers did. Also the incentives have changed. It makes sense to go the extra mile for a referring clinician that you see every week in tumor board when you are a partner in a private practice and you have ownership over something. Going the extra mile when you are another cog in the wheel (see RustedFox's post about Doing the Bare Minimum)? Nope.

There is no midlevel "equivalent" or pretender either. There are some RPAs doing basic fluoro and PAs/NPs doing basic procedures in IR but thats about it. Nothing else on a large scale. You can have more CRNA schools open up to help alleviate a shortage of anesthesiologists or put a bunch of NPs/PAs in the ER that just order every consult and imaging study possible to fill in for a shortage of properly trained ER docs. Doesn't really work for DR (and surgery for that matter).

It will be far worse before it gets better. Hospitals with bad payor mixes that don't want to provide subsidies for their rads will legitimately lose coverage. In the past they would threaten sending out an RFP to get another group to take your place. But today? They will hear crickets. OR they will have RadPartners or some other entity show them a nice PP presentation about how awesome they are (they aren't) and how great their turnaround times are (they aren't) just to get their foot in the door and take the contract. But even then they won't actually be able to service the contract any better than the old group.

I'm going to bed.

 

[RustedFox]

Well, damn.
Good to know AND well-written.

 

[bravotwozero]

I'm an attending radiologist. The radiologist shortage is bad and only going to get worse. I don't think people fully understand how bad this is and why it is different from other specialty shortages.

Historically, similar to anesthesia, there was a substantial reduction in the number of residency positions in the mid 90s. This let to a shortage of rads in the early 2000s. However that coincided with the introduction of PACS which was a one-off event that dramatically increased the productivity of radiologists seemingly overnight. That doesn't happen in any other specialty. You can't just have a group of ER docs or hospitalists or surgeons increase their productivity by that degree that quickly. That allowed the specialty to weather the storm and also let to a huge increase in $$$ for radiologists. Then, to my knowledge, the Deficit Reduction Act of 2005 introduced a variety of cuts to specifically to rads (things like outpatient imaging in particular) which led to rads having to spin the hamster wheel faster and faster for the same pay. That combined with and increasing number of residency positions since the early 2000s and delayed retirements after the financial crisis led to a dearth of jobs from 2008-2013ish. 2 year neurorad fellowships being required for terrible 7 on 7 off 10/12 hour shifts were advertised.

The number of residency positions since that time has gone up only modestly. However, the number of imaging studies being done has increased monumentally. The reimbursement per study has also gone done which incentivizes groups to read more studies to maintain income.

But you can only do that for so long.

Demographic changes including more demand for part-time work and less working hours (particularly from women but men as well) as well as increased focus on quality of life rather than pure $$$ means new grads don't want to grind out a career the way boomers did. Also the incentives have changed. It makes sense to go the extra mile for a referring clinician that you see every week in tumor board when you are a partner in a private practice and you have ownership over something. Going the extra mile when you are another cog in the wheel (see RustedFox's post about Doing the Bare Minimum)? Nope.

There is no midlevel "equivalent" or pretender either. There are some RPAs doing basic fluoro and PAs/NPs doing basic procedures in IR but thats about it. Nothing else on a large scale. You can have more CRNA schools open up to help alleviate a shortage of anesthesiologists or put a bunch of NPs/PAs in the ER that just order every consult and imaging study possible to fill in for a shortage of properly trained ER docs. Doesn't really work for DR (and surgery for that matter).

It will be far worse before it gets better. Hospitals with bad payor mixes that don't want to provide subsidies for their rads will legitimately lose coverage. In the past they would threaten sending out an RFP to get another group to take your place. But today? They will hear crickets. OR they will have RadPartners or some other entity show them a nice PP presentation about how awesome they are (they aren't) and how great their turnaround times are (they aren't) just to get their foot in the door and take the contract. But even then they won't actually be able to service the contract any better than the old group.

I'm going to bed.

You may have unwittingly written one of the best recruiting posts for aspiring radiologists.

Solid job market, reasonable reimbursement, high barrier to entry.

 

[skougess]

I'm an attending radiologist. The radiologist shortage is bad and only going to get worse. I don't think people fully understand how bad this is and why it is different from other specialty shortages.

Historically, similar to anesthesia, there was a substantial reduction in the number of residency positions in the mid 90s. This let to a shortage of rads in the early 2000s. However that coincided with the introduction of PACS which was a one-off event that dramatically increased the productivity of radiologists seemingly overnight. That doesn't happen in any other specialty. You can't just have a group of ER docs or hospitalists or surgeons increase their productivity by that degree that quickly. That allowed the specialty to weather the storm and also let to a huge increase in $$$ for radiologists. Then, to my knowledge, the Deficit Reduction Act of 2005 introduced a variety of cuts to specifically to rads (things like outpatient imaging in particular) which led to rads having to spin the hamster wheel faster and faster for the same pay. That combined with and increasing number of residency positions since the early 2000s and delayed retirements after the financial crisis led to a dearth of jobs from 2008-2013ish. 2 year neurorad fellowships being required for terrible 7 on 7 off 10/12 hour shifts were advertised.

The number of residency positions since that time has gone up only modestly. However, the number of imaging studies being done has increased monumentally. The reimbursement per study has also gone done which incentivizes groups to read more studies to maintain income.

But you can only do that for so long.

Demographic changes including more demand for part-time work and less working hours (particularly from women but men as well) as well as increased focus on quality of life rather than pure $$$ means new grads don't want to grind out a career the way boomers did. Also the incentives have changed. It makes sense to go the extra mile for a referring clinician that you see every week in tumor board when you are a partner in a private practice and you have ownership over something. Going the extra mile when you are another cog in the wheel (see RustedFox's post about Doing the Bare Minimum)? Nope.

There is no midlevel "equivalent" or pretender either. There are some RPAs doing basic fluoro and PAs/NPs doing basic procedures in IR but thats about it. Nothing else on a large scale. You can have more CRNA schools open up to help alleviate a shortage of anesthesiologists or put a bunch of NPs/PAs in the ER that just order every consult and imaging study possible to fill in for a shortage of properly trained ER docs. Doesn't really work for DR (and surgery for that matter).

It will be far worse before it gets better. Hospitals with bad payor mixes that don't want to provide subsidies for their rads will legitimately lose coverage. In the past they would threaten sending out an RFP to get another group to take your place. But today? They will hear crickets. OR they will have RadPartners or some other entity show them a nice PP presentation about how awesome they are (they aren't) and how great their turnaround times are (they aren't) just to get their foot in the door and take the contract. But even then they won't actually be able to service the contract any better than the old group.

I'm going to bed.

I feel that your last paragraph sums up where EM is quickly headed in terms of the CMG collapse.

 

[Pigg-O-Stat]

I'm an attending radiologist. The radiologist shortage is bad and only going to get worse. I don't think people fully understand how bad this is and why it is different from other specialty shortages.

Historically, similar to anesthesia, there was a substantial reduction in the number of residency positions in the mid 90s. This let to a shortage of rads in the early 2000s. However that coincided with the introduction of PACS which was a one-off event that dramatically increased the productivity of radiologists seemingly overnight. That doesn't happen in any other specialty. You can't just have a group of ER docs or hospitalists or surgeons increase their productivity by that degree that quickly. That allowed the specialty to weather the storm and also let to a huge increase in $$$ for radiologists. Then, to my knowledge, the Deficit Reduction Act of 2005 introduced a variety of cuts to specifically to rads (things like outpatient imaging in particular) which led to rads having to spin the hamster wheel faster and faster for the same pay. That combined with and increasing number of residency positions since the early 2000s and delayed retirements after the financial crisis led to a dearth of jobs from 2008-2013ish. 2 year neurorad fellowships being required for terrible 7 on 7 off 10/12 hour shifts were advertised.

The number of residency positions since that time has gone up only modestly. However, the number of imaging studies being done has increased monumentally. The reimbursement per study has also gone done which incentivizes groups to read more studies to maintain income.

But you can only do that for so long.

Demographic changes including more demand for part-time work and less working hours (particularly from women but men as well) as well as increased focus on quality of life rather than pure $$$ means new grads don't want to grind out a career the way boomers did. Also the incentives have changed. It makes sense to go the extra mile for a referring clinician that you see every week in tumor board when you are a partner in a private practice and you have ownership over something. Going the extra mile when you are another cog in the wheel (see RustedFox's post about Doing the Bare Minimum)? Nope.

There is no midlevel "equivalent" or pretender either. There are some RPAs doing basic fluoro and PAs/NPs doing basic procedures in IR but thats about it. Nothing else on a large scale. You can have more CRNA schools open up to help alleviate a shortage of anesthesiologists or put a bunch of NPs/PAs in the ER that just order every consult and imaging study possible to fill in for a shortage of properly trained ER docs. Doesn't really work for DR (and surgery for that matter).

It will be far worse before it gets better. Hospitals with bad payor mixes that don't want to provide subsidies for their rads will legitimately lose coverage. In the past they would threaten sending out an RFP to get another group to take your place. But today? They will hear crickets. OR they will have RadPartners or some other entity show them a nice PP presentation about how awesome they are (they aren't) and how great their turnaround times are (they aren't) just to get their foot in the door and take the contract. But even then they won't actually be able to service the contract any better than the old group.

I'm going to bed.

You forgot to say correlate clinically.

 

[ScepticSkeksis]

AI is likely going to make most radiologist jobs obsolete. We are already in a situation where AI is overreading radiologist studies in some markets. One place I work, the overreading AI actually notifies radiologists of missed critical items. It can do image reading, interpretation, and communication of critical findings at 1000X the speed of radiologists already, and we are in V1.2 of the AI software.

While we have a shortage of radiologists right now, the future is going to be AI reading common studies like CXR, head CT etc and leaving radiologists to review uncommon imaging or studies flagged by the AI as well as procedures. Definitely not a residency I would go into, or recommend anyone else to start.

On the plus side it will improve patient care with radiology interpretations at lightning speed, and with greater accuracy than a human can do.

As an investor in a huge, potentially pardigm-changing AI project (in a non-medical field) the potential for AI to completely change so many volume-based repetitive task careers is still not fully appreciated.

 

[skougess]

AI is likely going to make most radiologist jobs obsolete. We are already in a situation where AI is overreading radiologist studies in some markets. One place I work, the overreading AI actually notifies radiologists of missed critical items. It can do image reading, interpretation, and communication of critical findings at 1000X the speed of radiologists already, and we are in V1.2 of the AI software.

While we have a shortage of radiologists right now, the future is going to be AI reading common studies like CXR, head CT etc and leaving radiologists to review uncommon imaging or studies flagged by the AI as well as procedures. Definitely not a residency I would go into, or recommend anyone else to start.

On the plus side it will improve patient care with radiology interpretations at lightning speed, and with greater accuracy than a human can do.

As an investor in a huge, potentially pardigm-changing AI project (in a non-medical field) the potential for AI to completely change so many volume-based repetitive task careers is still not fully appreciated.

We don’t even trust AI to read EKGs. 2D squiggles on a page.

 

[ScepticSkeksis]

We don’t even trust AI to read EKGs. 2D squiggles on a page.

That's not really AI. EKG computer interpretation is equivalent to an Atari for gaming. With AI, you can literally have it compare Grandma's head CT with a database of millions of other head CTs and use its algorithm to get the best interpretation possible. Modern AIs can even search specific web content, like Getty Images, or Google images and pull data from there. The software is still in its infancy as far as complexity and processing power. Within 5 years, it is going to be able to out-compete radiolgists with ease. It will be easy enough to test. Just pit an AI-radiology program against 5 radiologists. Give them each 100 films which are either normal or have a specific finding. Then compare speed and accuracy. Humans have no chance.

 

[odyssey2]

That's not really AI. EKG computer interpretation is equivalent to an Atari for gaming. With AI, you can literally have it compare Grandma's head CT with a database of millions of other head CTs and use its algorithm to get the best interpretation possible. Modern AIs can even search specific web content, like Getty Images, or Google images and pull data from there. The software is still in its infancy as far as complexity and processing power. Within 5 years, it is going to be able to out-compete radiolgists with ease. It will be easy enough to test. Just pit an AI-radiology program against 5 radiologists. Give them each 100 films which are either normal or have a specific finding. Then compare speed and accuracy. Humans have no chance.

The problem is that most patients don't come in with *a* specific problem. They'll have a series of findings, some of which can only point to a differential (esp on simpler modalities like XR) which require clinical context, which, even if the AI can take in info from the EMR, is often incomplete or inaccurate. Even calling something "normal" is fraught, there can be subtle findings in seemingly clean scans that lead to huge liability later. A radiologist was recently sued for missing a tiny nodule on a CXR which blew up into a lung cancer. You'd need an AI that could take in the imaging findings and the context and synthesize this information into a report which answers the clinician's question, and can also take into account any uncertainties (which the current iteration of AI seems to be really bad at). That's essentially getting to the level of general intelligence, which, if AI ever gets to that point, society will have bigger problems than radiologists being out of work.

 

[ScepticSkeksis]

The problem is that most patients don't come in with *a* specific problem. They'll have a series of findings, some of which can only point to a differential (esp on simpler modalities like XR) which require clinical context, which, even if the AI can take in info from the EMR, is often incomplete or inaccurate. You'd need an AI that could take in the imaging findings and the context and synthesize this information into a report which answers the clinician's question, and can also take into account any uncertainties (which the current iteration of AI seems to be really bad at). That's essentially getting to the level of general intelligence, which, if AI ever gets to that point, society will have bigger problems than radiologists being out of work.

Odyssey, I've seen some of the latest AI software run, and we definitely aren't far from being able to do what's above. An AI could take your indication: "RLQ abdominal pain" then run through the image, find anything "abnormal" (meaning deviating from the 1000's of normal CTs it would compare it to) and then further drill down to whether these are incidental findings, like diverticulosis, or something that needs to be reported in the "Clinical Impression" box, specifically it could comment on the appendix which even human radiologists sometimes neglect to do. In addition it could easily compare the images to past images done for the patient along with past reads for even more detail and accuracy.
Version 3.0 could have access to your labs and notes and say: "Gallbladder duct mildly dilated, but with Total bilirubin of 3.2, it is consistent with early biliary obstruction".

The AI could come back real time and prompt the ED physician for more information: "Has this patient had recent surgery?" "Do they have any blood in the stool" etc.

We could even take a look at the initial read and ask questions back to it: "Can you take another look at the gallbladder and give me more information?" Just try that with human radiologists

 

[odyssey2]

Odyssey, I've seen some of the latest AI software run, and we definitely aren't far from being able to do what's above. An AI could take your indication: "RLQ abdominal pain" then run through the image, find anything "abnormal" (meaning deviating from the 1000's of normal CTs it would compare it to) and then further drill down to whether these are incidental findings, like diverticulosis, or something that needs to be reported in the "Clinical Impression" box, specifically it could comment on the appendix which even human radiologists sometimes neglect to do. In addition it could easily compare the images to past images done for the patient along with past reads for even more detail and accuracy.
Version 3.0 could have access to your labs and notes and say: "Gallbladder duct mildly dilated, but with Total bilirubin of 3.2, it is consistent with early biliary obstruction".

The AI could come back real time and prompt the ED physician for more information: "Has this patient had recent surgery?" "Do they have any blood in the stool" etc.

We could even take a look at the initial read and ask questions back to it: "Can you take another look at the gallbladder and give me more information?" Just try that with human radiologists

Radiology isn't just reading the scans and as it is practiced now there's a lot of human error you need to deal with when it comes to what patients/clinicians say and even how the scans are taken by the techs. Protocols change depending on what the clinican is asking for and what is possible with the patient (physical conditions, psych issues etc). Lab values and prior imaging often isn't available. Clinical indications are incomplete or just wrong or so broad they might as well have been left blank. You need to know what's important and what's not, and what you can't accurately comment on, and reflect that in the report. You also need to prioritize tasks; I don't think the ED doc is going to love having a chatbot ask questions about every potential finding. You need a human to deal with these problems for the forseeable future; everything you mentioned could be absorbed into a radiologist's workflow but cant replace human judgment and communication. And I'm not even saying technology like that isn't possible, but if it ever gets that advanced, most other jobs in the world will have long been replaced too.

In order for radiologists to be kicked off of reading a specific modality, you would need large scale, heterogenous prospective studies showing AI>radiologist for every conceivable pathology with any significant morbidity in that specific modality. Showing it in one center, which uses specific machines and technologists, isn't going to cut it. Even if the technology gets that good it's going to be a long time before that happens.

 

[skougess]

AI will be a fantastic adjunct for radiologists in the very near future. It will be a gut check in the same way the EKG read is for EM physicians. Sometimes I catch things it didn’t. Sometimes it catches things I didn’t.

I’ll eat my shoe if it even eliminates one radiologist job within the next 10 years.

 

[EctopicFetus]

AI will be a fantastic adjunct for radiologists in the very near future. It will be a gut check in the same way the EKG read is for EM physicians. Sometimes I catch things it didn’t. Sometimes it catches things I didn’t.

I’ll eat my shoe if it even eliminates one radiologist job within the next 10 years.

I think it will make them more efficient. Frankly, today we waste so much clinician time with plan films. I review a CXR, so does the Rads doc. Thats a waste of time.

I could see AI calling reads and having a separate queue where the doc needs to review.

While we can point to some items the reality is there is no fatigue for a macnine but humans will make mistakes that are in retrospect obvious. Any ED doc with a functional brain (that number seems to be fewer and fewer) has caught something glaringly obvious on imaging that the Rads missed.

As one of my partners said regarding recruiting today.. “10 years ago every trained EM doc was good, today so many are so very bad”. I think this speaks to the proliferation of trash residencies. We will need AI to help the HCA residency grads see what anyone worth a damn would already know.

 

[CCM-MD]

Integration of AI into radiology is going to be similar to PACS. Except this time the profits from increased productivity are going to go into the pockets of corporations, hospital systems or private equity, rather than the reading physicians.

 

[emergentmd]

2001, finished residency and had Rads as a top 3 choice (anesthesiology, Rads, EM). Still glad I picked EM given my career path.

Rads sounds good but it has become a serious hamster wheel.

 

[ScepticSkeksis]

There really isn't another option. The shortage of radiologist and the wait times to get reads is actually a patient safety issue at this point. Unless radiologists can find a way to do better, AI will be seen as a necessary solution to the problem.

 

[Tenk]

I'm an attending radiologist. The radiologist shortage is bad and only going to get worse. I don't think people fully understand how bad this is and why it is different from other specialty shortages.

Historically, similar to anesthesia, there was a substantial reduction in the number of residency positions in the mid 90s. This let to a shortage of rads in the early 2000s. However that coincided with the introduction of PACS which was a one-off event that dramatically increased the productivity of radiologists seemingly overnight. That doesn't happen in any other specialty. You can't just have a group of ER docs or hospitalists or surgeons increase their productivity by that degree that quickly. That allowed the specialty to weather the storm and also let to a huge increase in $$$ for radiologists. Then, to my knowledge, the Deficit Reduction Act of 2005 introduced a variety of cuts to specifically to rads (things like outpatient imaging in particular) which led to rads having to spin the hamster wheel faster and faster for the same pay. That combined with and increasing number of residency positions since the early 2000s and delayed retirements after the financial crisis led to a dearth of jobs from 2008-2013ish. 2 year neurorad fellowships being required for terrible 7 on 7 off 10/12 hour shifts were advertised.

The number of residency positions since that time has gone up only modestly. However, the number of imaging studies being done has increased monumentally. The reimbursement per study has also gone done which incentivizes groups to read more studies to maintain income.

But you can only do that for so long.

Demographic changes including more demand for part-time work and less working hours (particularly from women but men as well) as well as increased focus on quality of life rather than pure $$$ means new grads don't want to grind out a career the way boomers did. Also the incentives have changed. It makes sense to go the extra mile for a referring clinician that you see every week in tumor board when you are a partner in a private practice and you have ownership over something. Going the extra mile when you are another cog in the wheel (see RustedFox's post about Doing the Bare Minimum)? Nope.

There is no midlevel "equivalent" or pretender either. There are some RPAs doing basic fluoro and PAs/NPs doing basic procedures in IR but thats about it. Nothing else on a large scale. You can have more CRNA schools open up to help alleviate a shortage of anesthesiologists or put a bunch of NPs/PAs in the ER that just order every consult and imaging study possible to fill in for a shortage of properly trained ER docs. Doesn't really work for DR (and surgery for that matter).

It will be far worse before it gets better. Hospitals with bad payor mixes that don't want to provide subsidies for their rads will legitimately lose coverage. In the past they would threaten sending out an RFP to get another group to take your place. But today? They will hear crickets. OR they will have RadPartners or some other entity show them a nice PP presentation about how awesome they are (they aren't) and how great their turnaround times are (they aren't) just to get their foot in the door and take the contract. But even then they won't actually be able to service the contract any better than the old group.

I'm going to bed.

Thanks for this. This is what I feared. The worst part is that when bad outcomes start happening due to delays it will be the radiologists who get blamed, not the crappy system. Then some new protocol for turn around times will be placed on you. This will probably accelerate the burnout, etc.

 

[Tenk]

I didn’t realize stat rad was a specific company. Our rads reads everything at all times but it has nothing to do with stat rad the company

Yeah, they are one of the biggest. There are a few other big ones: Vrad, Nighthawk, etc. They all probably have the same problems based on their system of maximizing profits for someone else. Congrats on having a good group. Hope it lasts for you. Ours is imploding.

 

[thegenius]

That's not really AI. EKG computer interpretation is equivalent to an Atari for gaming. With AI, you can literally have it compare Grandma's head CT with a database of millions of other head CTs and use its algorithm to get the best interpretation possible. Modern AIs can even search specific web content, like Getty Images, or Google images and pull data from there. The software is still in its infancy as far as complexity and processing power. Within 5 years, it is going to be able to out-compete radiolgists with ease. It will be easy enough to test. Just pit an AI-radiology program against 5 radiologists. Give them each 100 films which are either normal or have a specific finding. Then compare speed and accuracy. Humans have no chance.

What is your avatar? Some demon looking death turkey or something?

 

[thegenius]

We don’t even trust AI to read EKGs. 2D squiggles on a page.

I side with you. I think AI will eventually replace radiologists but it's not coming that soon, like in a year or two. or the rollout will be only for select cases. But yea it's coming.

Why can't AI just read 2D squiggles with more accuracy than a Cardiologist or ER doc?

 

[thegenius]

The problem is that most patients don't come in with *a* specific problem. They'll have a series of findings, some of which can only point to a differential (esp on simpler modalities like XR) which require clinical context, which, even if the AI can take in info from the EMR, is often incomplete or inaccurate. Even calling something "normal" is fraught, there can be subtle findings in seemingly clean scans that lead to huge liability later. A radiologist was recently sued for missing a tiny nodule on a CXR which blew up into a lung cancer. You'd need an AI that could take in the imaging findings and the context and synthesize this information into a report which answers the clinician's question, and can also take into account any uncertainties (which the current iteration of AI seems to be really bad at). That's essentially getting to the level of general intelligence, which, if AI ever gets to that point, society will have bigger problems than radiologists being out of work.

Yea that's where Rads interpretation of what's important or not important is helpful.
Imagine if AI writes:
- there are 42 1 mm pulmonary nodules
- there are 15 2 mm pulmonary nodules
- there are 0 3 mm pulmonary nodules
- there are 4 4 mm pulmonary nodules
- there are 1 5 mm pulmonary nodules
- clinically correlate

 

[thegenius]

AI will be a fantastic adjunct for radiologists in the very near future. It will be a gut check in the same way the EKG read is for EM physicians. Sometimes I catch things it didn’t. Sometimes it catches things I didn’t.

I’ll eat my shoe if it even eliminates one radiologist job within the next 10 years.

Yea i can see a radiologist programming his own AI software to read chest rays and have templeted outputs and to have them flag all suspicious findings, and a radiologist just flips through image after image and is more efficient. Just approving most imaging and editing a few that aren't quite right.

But as I remember one radiologist told me...he can interpret most chest xrays in 10-20 seconds but actually dictating them into his broken system is so time consuming. Software never work properly where he works.

 

[dozitgetchahi]

I'm an attending radiologist. The radiologist shortage is bad and only going to get worse. I don't think people fully understand how bad this is and why it is different from other specialty shortages.

Historically, similar to anesthesia, there was a substantial reduction in the number of residency positions in the mid 90s. This let to a shortage of rads in the early 2000s. However that coincided with the introduction of PACS which was a one-off event that dramatically increased the productivity of radiologists seemingly overnight. That doesn't happen in any other specialty. You can't just have a group of ER docs or hospitalists or surgeons increase their productivity by that degree that quickly. That allowed the specialty to weather the storm and also let to a huge increase in $$$ for radiologists. Then, to my knowledge, the Deficit Reduction Act of 2005 introduced a variety of cuts to specifically to rads (things like outpatient imaging in particular) which led to rads having to spin the hamster wheel faster and faster for the same pay. That combined with and increasing number of residency positions since the early 2000s and delayed retirements after the financial crisis led to a dearth of jobs from 2008-2013ish. 2 year neurorad fellowships being required for terrible 7 on 7 off 10/12 hour shifts were advertised.

The number of residency positions since that time has gone up only modestly. However, the number of imaging studies being done has increased monumentally. The reimbursement per study has also gone done which incentivizes groups to read more studies to maintain income.

But you can only do that for so long.

Demographic changes including more demand for part-time work and less working hours (particularly from women but men as well) as well as increased focus on quality of life rather than pure $$$ means new grads don't want to grind out a career the way boomers did. Also the incentives have changed. It makes sense to go the extra mile for a referring clinician that you see every week in tumor board when you are a partner in a private practice and you have ownership over something. Going the extra mile when you are another cog in the wheel (see RustedFox's post about Doing the Bare Minimum)? Nope.

There is no midlevel "equivalent" or pretender either. There are some RPAs doing basic fluoro and PAs/NPs doing basic procedures in IR but thats about it. Nothing else on a large scale. You can have more CRNA schools open up to help alleviate a shortage of anesthesiologists or put a bunch of NPs/PAs in the ER that just order every consult and imaging study possible to fill in for a shortage of properly trained ER docs. Doesn't really work for DR (and surgery for that matter).

It will be far worse before it gets better. Hospitals with bad payor mixes that don't want to provide subsidies for their rads will legitimately lose coverage. In the past they would threaten sending out an RFP to get another group to take your place. But today? They will hear crickets. OR they will have RadPartners or some other entity show them a nice PP presentation about how awesome they are (they aren't) and how great their turnaround times are (they aren't) just to get their foot in the door and take the contract. But even then they won't actually be able to service the contract any better than the old group.

I'm going to bed.

Does this help explain why the quality of most radiology reads seems to have gone to garbage over the last few years?

I’m rheumatology and I’m at the point where I’ve basically quit reading the report for joint films because it’s useless. Gotta look at everything myself. They don’t catch erosions, they don’t catch fractures (!), everything is “osteoarthritis” or “normal” etc.

 

[ScepticSkeksis]

What is your avatar? Some demon looking death turkey or something?

You are essentially correct. Have you seen the Dark Crystal?

 

[ScepticSkeksis]

Yea that's where Rads interpretation of what's important or not important is helpful.
Imagine if AI writes:
- there are 42 1 mm pulmonary nodules
- there are 15 2 mm pulmonary nodules
- there are 0 3 mm pulmonary nodules
- there are 4 4 mm pulmonary nodules
- there are 1 5 mm pulmonary nodules
- clinically correlate

Good AI shouldn't do that. It should know that pulmonary nodules less than 0.5 cm aren't clinically relevant. Just like a radiologist it could include the incidental nonsense on the long-form read, but omit these on the "clinical impression". It's very different than the EKG machine which reads "squiggles". AI for radiology would actually compare the current study with a database of thousands of continually updated studies, and also would theoretically learn and improve over time.

Imagine if it sees a little spot of calcified dura on a brain CT. It could within a few seconds compare that image to previously done scans which both show ICH and calcifications and make a determination as to which one it is.

 

[thegenius]

You are essentially correct. Have you seen the Dark Crystal?

Nope!

 

[thegenius]

Good AI shouldn't do that. It should know that pulmonary nodules less than 0.5 cm aren't clinically relevant. Just like a radiologist it could include the incidental nonsense on the long-form read, but omit these on the "clinical impression". It's very different than the EKG machine which reads "squiggles". AI for radiology would actually compare the current study with a database of thousands of continually updated studies, and also would theoretically learn and improve over time.

Imagine if it sees a little spot of calcified dura on a brain CT. It could within a few seconds compare that image to previously done scans which both show ICH and calcifications and make a determination as to which one it is.

It was partly a tongue-in-cheek post, but I do generally think that both of you guys are correct...AI will change Rads at some point but I don't think it's coming as quickly as you suggest.

And I do submit that there is literally no difference between an .jpg of an EKG vs an ankle xray...to AI.

 

[Franzd'Epinay]

I suspect AI-assisted radiology is just like self-driving cars: “will be ready in 1-2 years” for the foreseeable future. It will eventually finally arrive, though.

Amara’s law: “We overestimate the impact of technology in the short-term and underestimate the effect in the long run”

 

[qwerty89]

I suspect AI-assisted radiology is just like self-driving cars: “will be ready in 1-2 years” for the foreseeable future. It will eventually finally arrive, though.

Amara’s law: “We overestimate the impact of technology in the short-term and underestimate the effect in the long run”

I always think about this when people start talking about projecting technology in the future.

Who can forget this classic from 2016.

 

[ScepticSkeksis]

I always think about this when people start talking about projecting technology in the future.

Who can forget this classic from 2016.

No one can accurately predict the future of course. How laughable is the Jetsons or Star Trek on how badly they got things wrong? If there's one area in medicine that machine-learning or AI could make a positive impact it is radiology. Speedy interpretations could lead to quicker diagnosis of a host of conditions. I'd also say that pharmacy is a much simpler task that could easily be replaced by AI/Automation. Computers can count to 30 much faster than any human......

 

[qwerty89]

Does this help explain why the quality of most radiology reads seems to have gone to garbage over the last few years?

I’m rheumatology and I’m at the point where I’ve basically quit reading the report for joint films because it’s useless. Gotta look at everything myself. They don’t catch erosions, they don’t catch fractures (!), everything is “osteoarthritis” or “normal” etc.

Every radiologist has an internal clock on what an appropriate amount of time is to spend on a study. You get antsy looking at an xray for more than a 1 minute or 2 max knowing that that is way too long given what that exam is worth as an wRVU and what you are expected to read on a normal day.

Chest x-ray 1 view is 0.18 wRVU. You read 100 ICU chest xrays and you are at 18 wRVU. Awesome right? Except you are expected to read 60-90 wRVU/day in most jobs and more in the highest paying ones.

Rheumatology x-rays are some of the worst because the work to income ratio is just downright awful. Up there with thyroid ultrasound and scoliosis xrays.

To be honest x-rays are almost like a courtesy service that radiologists at this point tolerate in order to get cross-sectional imaging (CT/MR). It's essentially a money loser and gets treated as such.

It will probably get worse too since nobody from the younger generation likes reading x-rays. A lot of groups have old boomer rads that mostly read x-rays and sling barium which frees up the others to read everything else.

Back in the day rads didn't talk about wRVUs/day. They just did the work and went home. But with reimbursement cuts and increased corporatization its always one of the first questions people ask when talking about a job. How much does it pay? How much vacation? How much do I have to read everyday?

How many ER docs were concerned with their patients per hour 20-30 years ago? Or the door to doc times? door to admit/discharge times? A good ER doc to an administrator is one that knows to stfu and move the meat. A good radiologist to an administrator (and many surgeons) is one that does what they are told and completes as many studies as possible. Quality? Only measured in terms of wRVUs...

Our group has a couple of old timer rads doing things like outpatient thyroid FNAs and outpatient MSK ultrasound where the radiologist actually scans the patient. Both are money losers and we will not continue doing them after they leave. The endocrinologists can send their "business" elsewhere if they get mad. Doing those things are objectively good patient care but unless its subsidized it doesn't make sense.

You can only take so many cuts.

 

[RustedFox]

I appreciate your perspective, radbro - but you guys are still making absurd bank.

 

[DetroitERDO]

AI is likely going to make most radiologist jobs obsolete. We are already in a situation where AI is overreading radiologist studies in some markets. One place I work, the overreading AI actually notifies radiologists of missed critical items. It can do image reading, interpretation, and communication of critical findings at 1000X the speed of radiologists already, and we are in V1.2 of the AI software.

While we have a shortage of radiologists right now, the future is going to be AI reading common studies like CXR, head CT etc and leaving radiologists to review uncommon imaging or studies flagged by the AI as well as procedures. Definitely not a residency I would go into, or recommend anyone else to start.

On the plus side it will improve patient care with radiology interpretations at lightning speed, and with greater accuracy than a human can do.

As an investor in a huge, potentially pardigm-changing AI project (in a non-medical field) the potential for AI to completely change so many volume-based repetitive task careers is still not fully appreciated.

absolutely... 100% I heard this during med school and stayed away from rads because of AI. And sorry folks I think it is coming that soon.

AI can detect breast cancer as well as two doctors: new study

A new study has proven that AI can screen for breast cancer as well — if not better — than a double reading of mammograms by two highly skilled radiologists.

nypost.com

AI tool can detect heart attack risk years in advance

Fountain Life has developed an AI-enabled coronary artery scan that can detect heart attack risk up to 10 years in advance.

www.beckersasc.com

 

[ScepticSkeksis]

absolutely... 100% I heard this during med school and stayed away from rads because of AI. And sorry folks I think it is coming that soon.

AI can detect breast cancer as well as two doctors: new study

A new study has proven that AI can screen for breast cancer as well — if not better — than a double reading of mammograms by two highly skilled radiologists.

nypost.com

There will still be a need for radiologists, but AI will do the majority of the basic reads, leaving more complex cases and procedures to the humans.

 

[HemeOncHopeful19]

absolutely... 100% I heard this during med school and stayed away from rads because of AI. And sorry folks I think it is coming that soon.

AI can detect breast cancer as well as two doctors: new study

A new study has proven that AI can screen for breast cancer as well — if not better — than a double reading of mammograms by two highly skilled radiologists.

nypost.com

AI tool can detect heart attack risk years in advance

Fountain Life has developed an AI-enabled coronary artery scan that can detect heart attack risk up to 10 years in advance.

www.beckersasc.com

I am not a radiologist so I am just asking some questions that come to mind that I genuinely don’t know the answer to.

Is the current standard to have two Rads read each mammogram ?

What will the cost of the AI read be?

What percentage of the cost of imaging goes to the Radiologist for their report?

I guess I’m just wondering if you guys think this will be a system driven by economic efficiency or if it will be like the EMR where the government is lobbied to mandate its use even if it leads to further exploding healthcare costs.

In a world where physician pay is <10% of the cost of care I’m skeptical that we’re going to see some amazing revolution from reducing that to <5% while paying a bunch of new tech workers to work on the AI

 

[Brigade4Radiant]

If AI takes rads they can take cards and they can also take over primary care

 

[DetroitERDO]

My dad studied Mathematics in the 70s. Back then there were "human computers" working in the labs at his techincal university. He saw the change from large supercomputers to the personal computer and now we have the smart phone which is a computer in our hands.
AI is coming... its going to play an integral role in our daily life that we can only dream of... its just a matter a time

 

[RoyBasch]

I'm an attending radiologist. The radiologist shortage is bad and only going to get worse. I don't think people fully understand how bad this is and why it is different from other specialty shortages.
...

I'm going to bed.

Very interesting. Appreciate the information.

 

[bravotwozero]

Speaking of AI, has anyone noticed the 'DocsGPT' feature on doximity? Seems incredibly useful.

You can use it to produce discharge instructions in different languages (handy if in the middle of a system downtime), create templates for procedure notes you can copy and paste (let's say you have crappy meditech, and have already signed the chart so you can't edit it and use their own template), progress notes templates for the inpatient docs etc.

If you have a doximity account, be sure to take a look.

 

[Apollyon]

Speaking of AI, has anyone noticed the 'DocsGPT' feature on doximity? Seems incredibly useful.

You can use it to produce discharge instructions in different languages (handy if in the middle of a system downtime), create templates for procedure notes you can copy and paste (let's say you have crappy meditech, and have already signed the chart so you can't edit it and use their own template), progress notes templates for the inpatient docs etc.

If you have a doximity account, be sure to take a look.

"But can it core an apple?"

 

[Clean7795]

Is the current standard to have two Rads read each mammogram ?

What will the cost of the AI read be?

I guess I’m just wondering if you guys think this will be a system driven by economic efficiency or if it will be like the EMR where the government is lobbied to mandate its use even if it leads to further exploding healthcare costs.

In a world where physician pay is <10% of the cost of care I’m skeptical that we’re going to see some amazing revolution from reducing that to <5% while paying a bunch of new tech workers to work on the AI

1. In North America a single rad read is standard. In the UK a double read from 2 readers is standard. Hence there is a very strong drive in Europe to find a replacement for this 2nd reader (one of the articles posted above was rad+AI vs rad+rad).

2. Breast imaging is very political. There has been machine learning based AI used in breast imaging for over a decade. It is not useful, and some studies actually show that it decreases the effectiveness of breast screening. Regardless, because of some questionable early research, it's standard to have the current useless AI included in most places. I would imagine if there is a new shiny tool then it would also be mandatory.

 

[ScepticSkeksis]

Here's a question for the community I have never seen addressed. Radiology in my area is routinely taking 3+ hours to read CT scans, sometimes with urgent or critical findings. If we are sued for a bad outcome due to radiology delay, can we personally sue the radiology group for malpractice? If I'm sued due to their negligence, it definitely results in damages to me that they should be liable for.

 

[Apollyon]

Here's a question for the community I have never seen addressed. Radiology in my area is routinely taking 3+ hours to read CT scans, sometimes with urgent or critical findings. If we are sued for a bad outcome due to radiology delay, can we personally sue the radiology group for malpractice? If I'm sued due to their negligence, it definitely results in damages to me that they should be liable for.

I'm not a lawyer (thank heavens), but, I see it as, the imaging is a request for radiology consultation - not just the image, but what it shows. Plain film reading might be delayed, but we are taught in residency to read our own X-rays, and find biggies, not incidentalomas. We are not responsible for CT. So, the delayed CT read could be construed as a "delay in diagnosis". After all, that is what you will be hit with.

 

[HemeOncHopeful19]

Here's a question for the community I have never seen addressed. Radiology in my area is routinely taking 3+ hours to read CT scans, sometimes with urgent or critical findings. If we are sued for a bad outcome due to radiology delay, can we personally sue the radiology group for malpractice? If I'm sued due to their negligence, it definitely results in damages to me that they should be liable for.

You could try but I think it would be complicated.

If you got sued and a jury found against you then by definition they found YOU liable, not just the Radiologist. I think it would be hard to then go to court and sue the Radiologist claiming that you weren’t actually liable (and they were).

Note: I agree with your line of reasoning just thinking how it would look in court

 

[JacobMcCandles]

Here's a question for the community I have never seen addressed. Radiology in my area is routinely taking 3+ hours to read CT scans, sometimes with urgent or critical findings. If we are sued for a bad outcome due to radiology delay, can we personally sue the radiology group for malpractice? If I'm sued due to their negligence, it definitely results in damages to me that they should be liable for.

Everybody is getting sued, anyway.