Correct my misconception

[doctalaughs]

Question for the pathologists from a dumb dermatologist with a limited understanding of immunohistochemistry:

Reading path is obviously very complex and can’t be automated.

That being said from my (limited) experience it seems IHC stains are either positive or negative and could be read by a machine with limited AI compared to controls in a standardized way. Could a pathologist not choose the stains needed and get an automated read?

Obviously would be bad for path financially but why do payors continue to pay for IHC reads? Enlighten me.

---

Members don't see this ad.

 

[bauber]

If we still have to make the slide why feed it to a machine when I can just put it on my microscope? Why add complexity? Same reason why regular H&E hasn't all gone digital...still gotta make the slide so I'll just put it on my scope.

 

[Spikebd]

Question for the pathologists from a dumb dermatologist with a limited understanding of immunohistochemistry:

Reading path is obviously very complex and can’t be automated.

That being said from my (limited) experience it seems IHC stains are either positive or negative and could be read by a machine with limited AI compared to controls in a standardized way. Could a pathologist not choose the stains needed and get an automated read?

Obviously would be bad for path financially but why do payors continue to pay for IHC reads? Enlighten me.

IHC interpretation is nuanced, it isn’t always just yes/no.

Most specimens come with background normal cells, which need to be accounted for. Some IHC markers will stain those scattered background cells. This needs to be taken into account.

Other stains have a specific pattern that needs to be assessed, and isn’t just a positive or negative question. For example, in Hodgkin lymphoma nodular lymphocyte predominant type, the PD-1 stain has a characteristic rosette formation around the LP cells that has to be looked for. Many other examples like this exist. Derm is actually full of cases like this.

 

[SunBakedTrash]

HMB45 staining in different subtypes of nevi and melanoma is a good example of how nuanced interpretation can be. Or S100 in scar vs residual melanoma.

 

[Nilf]

It's a misconception that the stains are 'positive' or 'negative'. As Spikebd said, it's more nuanced than that.

It all depends on the case. In most cases basic melanocytic markers rarely make or break the diagnosis. In pigmented lesion they are helpful but There is no 'melanoma stain'. In poorly differentiated tumors or mets they are more helpful, like in a poorly differentiated neoplasm which lights up with cytokeratin and nothing else. In cutaneous lymphomas immunostains are indispensable, but the interpretation is very very complex and a whole battery of stains are needed, often in addition to molecular studies, i.e. is this cluster of CD30+ cells on the tiny shave biopsy truly a LyP or worse, or just an artefact? Are the CD4+ cells in the epidermis lymphocytes, or Langerhans cells? Is CD5 positive in the CD20 cells, or in the background CD3 lymphocytes. No AI will be able to handle this.

 

[doctalaughs]

It's a misconception that the stains are 'positive' or 'negative'. As Spikebd said, it's more nuanced than that.

It all depends on the case. In most cases basic melanocytic markers rarely make or break the diagnosis. In pigmented lesion they are helpful but There is no 'melanoma stain'. In poorly differentiated tumors or mets they are more helpful, like in a poorly differentiated neoplasm which lights up with cytokeratin and nothing else. In cutaneous lymphomas immunostains are indispensable, but the interpretation is very very complex and a whole battery of stains are needed, often in addition to molecular studies, i.e. is this cluster of CD30+ cells on the tiny shave biopsy truly a LyP or worse, or just an artefact? Are the CD4+ cells in the epidermis lymphocytes, or Langerhans cells? Is CD5 positive in the CD20 cells, or in the background CD3 lymphocytes. No AI will be able to handle this.

Totally understand. Makes sense. I was actually thinking of cutaneous lymphoma- as I read reports on these all the time and they need an obscene number of stains. I just remember sitting with dermpaths many years ago and most of the time they would look at the immunostain for literally 1 (maybe 1.5) seconds each. But it makes sense that you need to correlate which cells are actually picking up the stains and then make a call and would be very hard for a computer to do so.

 

[gbwillner]

I agree that it is very nuanced (background staining, correct organelles/nuclear/cytoplasmic staining, controls, etc), BUT automation tools are pretty good already and can already account for a lot of this. I agree IHC and H&E requires professional interpretation today BUT the tech is being developed and will likely perform as well as or better than humans at some point.

 

[LADoc00]

Question for the pathologists from a dumb dermatologist with a limited understanding of immunohistochemistry:

From my experience any dermatologist who claims the moniker of "dumb" is not only not dumb, but very likely an attractive female.

I know this because the things I have done for "dumb dermatologists" are so humiliating I have told no one..ever...clearly I learned nothing from interacting with girls in high school and college.

 

[Nilf]

From my experience any dermatologist who claims the moniker of "dumb" is not only not dumb, but very likely an attractive female.

I know this because the things I have done for "dumb dermatologists" are so humiliating I have told no one..ever...clearly I learned nothing from interacting with girls in high school and college.

Go to AAD meeting one day. See for yourself the difference between the female attendees at AAD and female attendees at USCAP. ESPECIALLY between the Latin/South American female dermatologists and FMGs in pathology.

 

[KeratinPearls]

Go to AAD meeting one day. See for yourself the difference between the female attendees at AAD and female attendees at USCAP. ESPECIALLY between the Latin/South American female dermatologists and FMGs in pathology.

AAD meetings are like Hollywood meets Medicine 😂 . Been once. Got myself a crapload of free expensive creams, soaps, skin care products. 😂 😆

 

[HeyDalaron]

Go to AAD meeting one day. See for yourself the difference between the female attendees at AAD and female attendees at USCAP. ESPECIALLY between the Latin/South American female dermatologists and FMGs in pathology.

Looks matter, probably more than anything else.

 

[Dral]

Totally understand. Makes sense. I was actually thinking of cutaneous lymphoma- as I read reports on these all the time and they need an obscene number of stains. I just remember sitting with dermpaths many years ago and most of the time they would look at the immunostain for literally 1 (maybe 1.5) seconds each. But it makes sense that you need to correlate which cells are actually picking up the stains and then make a call and would be very hard for a computer to do so.

There are some lymphoma stains that you can look at for a few seconds just to see if they are positive or negative in general. However, even looking at CTCL stains, a lot of cases take a long time to interpret. I would love to have an automated system to have them subtract CD1a stain away from what CD4 shows for when I compare to CD8, but other than that, I don't think automated learning would help a ton (at least at this point). There can be subtle lining up of CD8 at the DE junction with CD4 loss, follicular patterns of CTCL, CTCL that mimics spong derm, etc. I guess that could be learned, but I think it's too nuanced.

There are some CTCL cases where we'll not be able to make a clear top line decision, especially on early cases. These types of cases are the ones that I feel would be too difficult for an automated system to handle. The early evolving forms don't have enough material to push through T cell receptor studies. So we're left with "Possible early MF/CTCL, continue to monitor the patient clinically and resample as clinically appropriate" (or something to that effect). The translation of the H/E, stains, clinical info into a sensible path report is a lot of times not straight forward.

Immunostains alone can't even reveal a final answer lots of times. It's a catch 22. Until we are at the point where we can get accurate molecular signatures for tumors, we can't even really teach an automated system since we can't really validate a case as truly being a certain tumor all the time. Think cellular DF for example. Are the ones that metastasized really cellular DF or are they a tumor misdiagnosed as cellular DF. A molecular signature would be able to help parse that apart hopefully. By the time we can do accurate molecular signatures, I don't think we'll need an automated histologic interpreter anymore anyway. For awhile there, molecular pathologists will be worth their weight in gold probably though.

 

[LADoc00]

Looks matter, probably more than anything else.

Strangely you have no idea how correct this is..

 

[gbwillner]

Sorry to derail the likely "staging/grading DPs physical appearance" thread that was likely about to develop, but having seen a lot of the image analysis technology, here is how I think it will play out.

1. Image analysis support systems. These will not be systems that replace pathologists, rather, ones that pathologists will license to increase the speed with which they can render a DX. For example, do you hate spending 20 min on a 12 needle core prostate Bx? What if a WSI system and analytic software package can pre-screen the 12 Bx plus 3 levels and identify the most likely areas with cancer for you? BTW, this already exists.

2. "proprietary" IHC analysis. We know that sometimes pathologists are pretty bad at reproducibility in interpreting stains, unless it is really straight forward or there is a complex process with a rugged required training (like for PD-L1). Even then... not so great. However, computers can (and are being trained) to identify the correct organelle and consistent straining to improve reproducibility. This analysis can be added on top of the Dx process, wherein the IHC analysis can be considered its own "test" and likely will be more than 1 IHC stain. Basically, think gene expression profiling but with IHC analysis. This also is already in development.

3. I think the next step in the future is along the lines of #1 above, where the software reviews the material, identifies likely malignant areas, and even renders a DX. However, this will require physician oversight and review, so that instead of pushing glass, you are reviewing the findings of the system as a qualified lab director. This does not yet exist.

4. One day it will be noted that the cost of the pathologist is not worth it since the error rates for #3 are too low to be worthwhile, and the review is passed to a qualified tech. The lab director will supervise the techs.

5. Star Trek Tricorder.

 

[Spikebd]

Sorry to derail the likely "staging/grading DPs physical appearance thread that was likely about to develop, but having seen a lot of the image analysis technology, here is how I think it will play out.

1. Image analysis support systems. These will not be systems that replace pathologists, rather, ones that pathologists will license to increase the speed with which they can render a DX. For example, do you hate spending 20 min on a 12 needle core prostate Bx? What if a WSI system and analytic software package can pre-screen the 12 Bx plus 3 levels and identify the most likely areas with cancer for you? BTW, this already exists.

2. "proprietary" IHC analysis. We know that sometimes pathologists are pretty bad at reproducibility in interpreting stains, unless it is really straight forward or there is a complex process with a rugged required training (like for PD-L1). Even then... not so great. However, computers can (and are being trained) to identify the correct organelle and consistent straining to improve reproducibility. This analysis can be added on top of the Dx process, wherein the IHC analysis can be considered its own "test" and likely will be more than 1 IHC stain. Basically, think gene expression profiling but with IHC analysis. This also is already in development.

3. I think the next step in the future is along the lines of #1 above, where the software reviews the material, identifies likely malignant areas, and even renders a DX. However, this will require physician oversight and review, so that instead of pushing glass, you are reviewing the findings of the system as a qualified lab director. This does not yet exist.

4. One day it will be noted that the cost of the pathologist is not worth it since the error rates for #3 are too low to be worthwhile, and the review is passed to a qualified tech. The lab director will supervise the techs.

5. Star Trek Tricorder.

I’m sure DARPA is actively working on development as we speak.

 

[Doormat]

Question for the pathologists from a dumb dermatologist with a limited understanding of immunohistochemistry:

Reading path is obviously very complex and can’t be automated.

That being said from my (limited) experience it seems IHC stains are either positive or negative and could be read by a machine with limited AI compared to controls in a standardized way. Could a pathologist not choose the stains needed and get an automated read?

Obviously would be bad for path financially but why do payors continue to pay for IHC reads? Enlighten me.

This is an unusual post. Why do payors pay for IHC? Bad for path financially? First off, binary interpretation of an IHC stain is rare. Automated interpretation of IHC is not even on the radar for PathAI, the digital path and machine learning start-up in Cambridge, Mass. 1% positive cells can be interpreted as positive for some IHC markers.

My question to the doctalaughs is why can't a whole-body digital camera setup with AI machine-learning interpretation replace the dermatologist? Obviously that would be bad for derm financially, but why do payors continue to pay for skin exams performed by a human? Educate me.

 

[gbwillner]

This is an unusual post. Why do payors pay for IHC? Bad for path financially? First off, binary interpretation of an IHC stain is rare. Automated interpretation of IHC is not even on the radar for PathAI, the digital path and machine learning start-up in Cambridge, Mass. 1% positive cells can be interpreted as positive for some IHC markers.

My question to the doctalaughs is why can't a whole-body digital camera setup with AI machine-learning interpretation replace the dermatologist? Obviously that would be bad for derm financially, but why do payors continue to pay for skin exams performed by a human? Educate me.

You re looking at this all wrong. The physical interaction cannot be replaced until we make androids. MORE likely is this scenario (BTW, it is ALREADY in process):

1. new diagnostic tools developed to R/O melanoma without the need of a biopsy (already exist!);
2. Same or similar tools refined to diagnose melanoma;
3. similar tech is deployed for other skin conditions that require intervention;
4. Need for Derms decrease as GPs can use the above tools as well;
5. Growth of new techs that deploy said tests and surgical proceduralists that perform Bxs when needed. Fewer and fewer Derms exist for rarer conditions that don't fit the paradigm.

This is the future of Derm, whether they know it or not. But it is true for many of today's specialties.

 

[Allegri]

So will the median attractiveness of dermatologists shift over to the GP population, or will they leave medicine altogether? Please try to stay on topic!

 

[Guest8]

So will the median attractiveness of dermatologists shift over to the GP population, or will they leave medicine altogether? Please try to stay on topic!

They'll still be in derm, they'll just be doing all the cosmetic nonsense - fillers, lip puffing snake oil, Botox, "peels", etc. That's what made the field attractive to the attractive in the first place.

 

[cmz]

They'll still be in derm, they'll just be doing all the cosmetic nonsense - fillers, lip puffing snake oil, Botox, "peels", etc. That's what made the field attractive to the attractive in the first place.

You can do this without being derm.... once weed becomes legal in TX, I am going to sharpen my pen and script pad and dust of my DEA license.

 

[WEBB PINKERTON]

Remember that crazy pathologist who came here and claimed he knew of pathologists doing hair restoration due to the bad job market?

 

[Zarniwoop]

I don't know if this is still a thing, but docs in all specialties were also operating those "Low T" centers.

 

[gbwillner]

could a pathologist do this? i assume you would need some extra training or certificate, but I'm not sure exactly what that entails

Extra training besides going to medical school?
You just need to keep your DEA registration (if it is still considered a controlled substance). Most of us just let it lapse.

 

[Guest8]

Extra training besides going to medical school?
You just need to keep your DEA registration (if it is still considered a controlled substance). Most of us just let it lapse.

Seems like recreational weed is being legalized just as fast as medicinal, if not faster. Not sure there's a need for docs to write scripts for it once you can just pop in the local weed shop and buy whatever you want.

 

[Med Director New England]

Seems like recreational weed is being legalized just as fast as medicinal, if not faster. Not sure there's a need for docs to write scripts for it once you can just pop in the local weed shop and buy whatever you want.

Still pretty high demand for medicinal in states that have both medical and Recreation use.

much cheaper & not taxed if medical. In New England where it is easy to get recreational plenty of burnt out docs are leaving their practice starting clinics that issue med cards.
Not accepting insurance - all out of pocket by patient. They pay for one 15 minute office annually (~ 200 bucks) save thousands per year buying medical weed. MD can collect 15 K / day in billing.

 

[cmz]

Still pretty high demand for medicinal in states that have both medical and Recreation use.

much cheaper & not taxed if medical. In New England where it is easy to get recreational plenty of burnt out docs are leaving their practice starting clinics that issue med cards.
Not accepting insurance - all out of pocket by patient. They pay for one 15 minute office annually (~ 200 bucks) save thousands per year buying medical weed. MD can collect 15 K / day in billing.

That's my dream plan. I not only want to write the scripts, sell the weed, etc but I also want to do weed testing with HPLC, Mass spec etc. All the weed sellers locally must come to me to get their official certification stamp saying their stuff is legit good. Since I am in good with agilent since I own an Omnis or three, I am sure I can get a nice discount on one of their weed analyzers. (This is a gross oversimplification of my end-game but I have a sound plan in mind in case anyone wants to invest. I guesstimate it would be a 3-4M start-up cost for a moderately-sized lab).

 

[LADoc00]

That's my dream plan. I not only want to write the scripts, sell the weed, etc but I also want to do weed testing with HPLC, Mass spec etc. All the weed sellers locally must come to me to get their official certification stamp saying their stuff is legit good. Since I am in good with agilent since I own an Omnis or three, I am sure I can get a nice discount on one of their weed analyzers. (This is a gross oversimplification of my end-game but I have a sound plan in mind in case anyone wants to invest. I guesstimate it would be a 3-4M start-up cost for a moderately-sized lab).

Cannabis QC testing is totally a bust. I did exhaustive research into this when we legalized. I went to Emerald Scientific and talked to 20+ labs already in the space, talked to government inspectors, processors, growers etc.

The start up costs are nowhere near 3M haha. More like 400-600K.

The problem is the growers and processors who are forced by government regs to get tested will just lab shop if you actually call them on pesticide contamination or fungi. The entire process is totally corrupt and the profit is nowhere near where you might think it is. One poor guy who ran a lab said that when he sent a fungal contamination report to one grower they actually showed up his residence and threatened to set his house on fire in front of his kids.

These guys will grow 2m in crop and whine they are paying 400 bucks for sample test. They are the proverbial dirt bag POS.

RETAIL. It is all about RETAIL. Sell at retail. Make the shop cool, place where people want to hang out. You can make crazy money at retail. Everything else in the cannabis business: growing, processing, packaging, testing is trash tier.

I would love to do something in Texas if someone out there wants to get a strip club/dispensary/taco joint up. I think that is the trifecta.

 

[Zarniwoop]

I love how after all those years of education, time in labs, etc, our gut reaction on how to make money today hasn't evolved from our high school days, and still circles around the themes of strippers, weed, and cheap Mexican food.

 

[AZpath]

" guesstimate it would be a 3-4M start-up cost for a moderately-sized lab"

I guess it depends on what kind of lab.
Entry into a pathology lab is much lower.
The current problem with a new pathology lab is: 1) insurance contracts and 2) getting a client base.
1) is a giant barrier. It is not worth starting a new independent lab these days unless you have unique test that gets cash or you can somehow exist on MC dollars alone.
20 years ago insurance contracts were much easier to negotiate. If you are a new lab they don't want to talk with you

 

[cmz]

Cannabis QC testing is totally a bust. I did exhaustive research into this when we legalized. I went to Emerald Scientific and talked to 20+ labs already in the space, talked to government inspectors, processors, growers etc.

The start up costs are nowhere near 3M haha. More like 400-600K.

The problem is the growers and processors who are forced by government regs to get tested will just lab shop if you actually call them on pesticide contamination or fungi. The entire process is totally corrupt and the profit is nowhere near where you might think it is. One poor guy who ran a lab said that when he sent a fungal contamination report to one grower they actually showed up his residence and threatened to set his house on fire in front of his kids.

These guys will grow 2m in crop and whine they are paying 400 bucks for sample test. They are the proverbial dirt bag POS.

RETAIL. It is all about RETAIL. Sell at retail. Make the shop cool, place where people want to hang out. You can make crazy money at retail. Everything else in the cannabis business: growing, processing, packaging, testing is trash tier.

I would love to do something in Texas if someone out there wants to get a strip club/dispensary/taco joint up. I think that is the trifecta.

A "small lab" with full automation has some big upfront costs. 400-600K sounds like a pipe dream. It's not unreasonable to estimate 1-1.5M for a small lab and upwards of 3-4M for a medium-sized lab.

It's kind of scary to think that someone would get so riled up if their report came back with something abnormal. If you get a bad result like that, you have to pretty much destroy all of your crop (or, at the very least, the area that you're testing). The fact remains that if you wish to sell THC as a regulated commodity. then it has to be given an A-OK by a licensed lab.

I am down to do the testing, handing out the scripts and growing the product. You can set up a pan dulce shop next door with a taco truck to the side. I would be afraid of opening up a strip club, though...

 

[cmz]

I guess it depends on what kind of lab.
Entry into a pathology lab is much lower.
The current problem with a new pathology lab is: 1) insurance contracts and 2) getting a client base.
1) is a giant barrier. It is not worth starting a new independent lab these days unless you have unique test that gets cash or you can somehow exist on MC dollars alone.
20 years ago insurance contracts were much easier to negotiate. If you are a new lab they don't want to talk with you

Yeah I'm talking about a THC testing lab....