Supreme Court ruling may pave way to identification of Ontario’s top-billing physicians

[Lactic Folly]

1 hour ago, tavenan said:

What are your thoughts on the future of rads in say 10-20 years? Do you think there will still be such a demand in that timespan? I hear a lot of doom and gloom talk about AI as it pertains specifically to radiology. 

The demand for the diagnostic information that imaging provides is ever-increasing. I can't see that humans in radiology are more susceptible to replacement than humans in any other field of medicine (for which there are also AI applications for diagnosis). If there is an error in a machine read, who will take liability? Experiments have shown that AI systems can be hacked, leading to wrong test results. Instead, AI can be better applied throughout all areas of healthcare to help make us more efficient and deal with the increased demand - for example, in radiology, this might include triaging urgent exams and helping pull clinical information.

[NLengr]

7 hours ago, rmorelan said:

ha, well that is predictable

there is also what you mean by appropriate - right now some people on FFS I think reasonable feel they are moving at extreme volume - that isn't about the money, it is about the pressure from long wait lists and clinical demand. Radiology would actually fall into that category - our output per day has at least by informal analysis doubled in the past 15 years and that isn't from the tech advances. You go salary and you better believe people will slow down - in their eyes back to a sane level which is safer for patients. 

I don't expect any government body to see a reduction as positive mind you ha. 

The ones who got flagged were also seeing less volume than the other salaried staff. Things like 4 follow ups a day at 30 minutes a follow up. Just clear abuse of the salary system.

[blah1234]

10 hours ago, Lactic Folly said:

The demand for the diagnostic information that imaging provides is ever-increasing. I can't see that humans in radiology are more susceptible to replacement than humans in any other field of medicine (for which there are also AI applications for diagnosis). If there is an error in a machine read, who will take liability? Experiments have shown that AI systems can be hacked, leading to wrong test results. Instead, AI can be better applied throughout all areas of healthcare to help make us more efficient and deal with the increased demand - for example, in radiology, this might include triaging urgent exams and helping pull clinical information.

I find the liability issue interesting. I think there should be a way to build in the anticipated cost of litigation into the business model and then adjust your product premium accordingly. You can probably even set out terms of liquidated damages in your contract when you sign with hospitals. I think if the current model has physicians or hospitals covering malpractice insurance a larger company should be able to figure out a cost-effective way to insure themselves. 

I think if there is an AI product that can perform at an acceptable level then the business case should be relatively easy in comparison to the R&D. 

[Lactic Folly]

20 hours ago, blah1234 said:

I find the liability issue interesting. I think there should be a way to build in the anticipated cost of litigation into the business model and then adjust your product premium accordingly.

Off-topic: Sure, of course this can be done from the perspective of the manufacturer. I was considering more from the perspective of the party who is making the decision to use AI in patient care, i.e. hospitals. Would they be comfortable employing AI as a diagnostician in its own right (in effect making the hospital responsible for the decision to employ AI in case a worst-case scenario adverse event occurs), or would AI be considered in the realm of equipment (still requiring human input to accept or overrule what AI is doing - thus keeping the liability with the physician). Doesn't seem there is a consensus based on a quick survey of articles out there - legal landscape is still evolving.

[rmorelan]

12 hours ago, Lactic Folly said:

Off-topic: Sure, of course this can be done from the perspective of the manufacturer. I was considering more from the perspective of the party who is making the decision to use AI in patient care, i.e. hospitals. Would they be comfortable employing AI as a diagnostician in its own right (in effect making the hospital responsible for the decision to employ AI in case a worst-case scenario adverse event occurs), or would AI be considered in the realm of equipment (still requiring human input to accept or overrule what AI is doing - thus keeping the liability with the physician). Doesn't seem there is a consensus based on a quick survey of articles out there - legal landscape is still evolving.

most of the current AI commercial products dodge that completely - by saying it is just an augmentation tool. Suing the manufacture in that case would be the same as suing the maker of your PACS because you missed the nodule in the lung despite some tools to help with that (in this case namely the MIP images in lung windows - which help radiologists pick up nodules).

That puts all the liability back on the radiologist, same as it is now. 

That is also how the tools are registered with the FDA for instance. Makes the barrier to entering the market a lot lower and promotes innovation. Ha, I think it will be a while before we take it to the next level. 

[RichardHammond]

Interesting discussion.

Hmm I wonder if there are any examples or lessons from automatic ECG interpretations that one can extend to radiology. Suppose a psychiatrist orders an inpatient ECG to assess qt for an antipsychotic change and the interpretation says normal qt, non-specific ST changes but misses an obscure MI or arrhythmia. Who's at fault? Can you sue the ECG machine manufacturer?

I think it's institution specific regarding if/or when there's a formal read for the ECG but there seems to be a lot of trust in automatic interpretations by non-cardiology people. Of course some squiggly lines are easier for machines to interpret than a million slice CT but who knows, maybe one day the technology will be enough to gain our trust (at least for simple "triaging"). 

[medigeek]

10 hours ago, RichardHammond said:

Interesting discussion.

Hmm I wonder if there are any examples or lessons from automatic ECG interpretations that one can extend to radiology. Suppose a psychiatrist orders an inpatient ECG to assess qt for an antipsychotic change and the interpretation says normal qt, non-specific ST changes but misses an obscure MI or arrhythmia. Who's at fault? Can you sue the ECG machine manufacturer?

I think it's institution specific regarding if/or when there's a formal read for the ECG but there seems to be a lot of trust in automatic interpretations by non-cardiology people. Of course some squiggly lines are easier for machines to interpret than a million slice CT but who knows, maybe one day the technology will be enough to gain our trust (at least for simple "triaging"). 

Machines miss stuff on ECGs all the time. Clinical context is also another important variable. 

[shakeshake]

On 4/17/2019 at 8:29 AM, medigeek said:

Machines miss stuff on ECGs all the time. Clinical context is also another important variable. 

This.

 

The other factor is that diagnosis is actually a moving target in many cases. The criteria we use to make diagnoses based on imaging/histology/clinical information is constantly changing over time based on research into outcomes and knowledge. An AI system is only as good as the data that is put into it. Much like the current ECG interpretations, it can be helpful to catch simple things and streamline simple cases. The reality is medicine is becoming increasingly complex and at least for now AI is likely going to help us become more efficient. 

[ploughboy]

On 4/17/2019 at 11:29 AM, medigeek said:

Machines miss stuff on ECGs all the time. Clinical context is also another important variable. 

Q: Who programs EKG readers?

A: Malpractice lawyers.

[blah1234]

Aren't current ECG machines just measuring waves and applying some rudimentary criteria? I'm not sure there is any element of machine learning in the machines I work with. 

I think the technology will get there one day but you'll still need a physician to act on the interpretation in conjunction with the rest of the clinical picture. Perhaps the savings of replacing physician ECG interpretation isn't enough to justify a medtech company to invest resources towards it when Radiology seems much more lucrative. 

[medigeek]

2 hours ago, blah1234 said:

Aren't current ECG machines just measuring waves and applying some rudimentary criteria? I'm not sure there is any element of machine learning in the machines I work with. 

I think the technology will get there one day but you'll still need a physician to act on the interpretation in conjunction with the rest of the clinical picture. Perhaps the savings of replacing physician ECG interpretation isn't enough to justify a medtech company to invest resources towards it when Radiology seems much more lucrative. 

Cardiologists have to read every EKG, including every single one from the floor/ED. They then bill for each one. In the long run it would be cheaper to have AI reading EKGs.

Also, pretty sure they've improved over time to factor in what cardiologists do to learn from it. Not sure if this applies to every one out there. 

Like anything else, it would take forever to implement it everywhere. 

[blah1234]

1 hour ago, medigeek said:

Cardiologists have to read every EKG, including every single one from the floor/ED. They then bill for each one. In the long run it would be cheaper to have AI reading EKGs.

Also, pretty sure they've improved over time to factor in what cardiologists do to learn from it. Not sure if this applies to every one out there. 

Like anything else, it would take forever to implement it everywhere. 

I get that cardiologists have to review every EKG. I'm wondering if the amount of R+D resources available was calculated to bring a higher ROI in radiology. I'm sure the top companies have modelled this out for what the most lucrative applications are in the next 10-20 years. 

[LostLamb]

psych's best friends in ecg interpretation: on-call cardiology fellow and old ecgs!

 

i'll admit my ecg skills are getting rustier and rustier...but I never hesitate to get a second glance by a real person if I feel there's something and the machine reads it as normal. Similarly, I feel that my non-psych colleagues do the same when requesting a psychiatric perspective on the 'difficult' patient.

I am grateful that at this point I won't have to fear replacement by a machine in the foreseeable future.