Asking AI about AI in Medicine

12.06.22 // Medicine, Radiology

I had a brief chat with OpenAI’s ChatGPT about noise, radiology, and AI in medicine:

Human: Summarize the book Noise by Daniel Kahneman
AI: Noise by Daniel Kahneman is a book that examines how our minds work in a noisy, uncertain world. It dives into the cognitive biases and heuristics we use to make decisions with limited information, and explains how they can lead us astray. The book covers topics such as perception, intuition, and decision-making, and explores how noise and confusion can shape our lives and our thinking.

Human: How does the book argue that noise impacts radiology?
AI: In Noise, Daniel Kahneman argues that noise can have a major impact on radiology. He explains that radiologists are often subject to cognitive biases like confirmation bias and anchoring bias, which can lead them to make decisions that are not in the best interest of their patients. Additionally, he shows how noise can affect diagnostic accuracy and decision-making, making it harder for radiologists to provide accurate treatments and diagnoses.

Human: How could radiology address those limitations?
AI: Radiologists can address the limitations of noise in radiology by taking steps to reduce cognitive biases and increase accuracy. These steps include utilizing evidence-based decision-making, double-checking results for accuracy, and employing advanced imaging technologies such as computer-aided detection systems. Additionally, radiologists should strive to keep up with advances in medical technology and imaging techniques so that they can identify any noise or errors in their images.

Human: What are the current limitations in utilizing AI in radiology?
AI: Currently, one of the biggest limitations in utilizing AI in radiology is the potential for errors and false positives. While AI systems have the potential to be more accurate than human radiologists, there is still a risk that the AI system will miss diagnoses or misinterpret images. Additionally, AI systems can be costly to implement, and require significant computing power, which may not be feasible in many medical settings.

Human: What would be the easiest type of medicine to outsource to AI?
AI: The easiest type of medicine to outsource to AI would be administrative tasks such as scheduling appointments and completing insurance forms. AI can quickly and accurately complete these types of tasks, saving time and resources for medical professionals.

Even the machines think there’s some low-hanging fruit out there.

Hindsight in the Valley of the Normal

11.16.22 // Miscellany

From Nobel-laureate Daniel Kahneman’s Noise: A Flaw in Human Judgement:

More broadly, our sense of understanding the world depends on our extraordinary ability to construct narratives that explain the events we observe. The search for causes is almost always successful because causes can be drawn from an unlimited reservoir of facts and beliefs about the world.

[…]

This continuous causal interpretation of reality is how we “understand” the world. Our sense of understanding life as it unfolds consists of the steady flow of hindsight in the valley of the normal. This sense is fundamentally causal: new events, once known, eliminate alternatives, and the narrative leaves little room for uncertainty. As we know from classic research on hindsight, even when subjective uncertainty does exist for a while, memories of it are largely erased when the uncertainty is resolved.

Noise is a good book. It’s not an exciting book–though neither is Thinking, Fast and Slow for that matter–but it’s an important book. Bias is important, but human noisiness and our imperfect and very noisy systems are also important: Judgments (predictions, decisions) are hard to get right, and so much of the world’s punditry is just storytelling.

Doctors make dozens if not hundreds of little and big decisions every day. And I know that not only I am noisy, but I’m noisy in a variety of ways. A big part of my deliberate practice as it pertains to radiology is to try to limit how much noise distorts my decision-making and the quality of my reports.

We want to be not just good but consistently so.

ABR soliciting nominations for its Board of Governors

11.12.22 // Radiology

The bylaws of the American Board of Radiology are interesting. I’ve written about them here, and you should read that post. Go ahead, I’ll wait.

Relevent to an email blast the ABR sent out this week is section 4.3:

Section 4.3. Election of Governors. Nominees shall be solicited from the Board of Trustees and Board of Governors, and may be solicited from any appropriate professional organization. Professional organizations shall provide such nominations in writing. An affirmative vote of at least three-fourths (3/4ths) of the entire Board of Governors shall be necessary for the election of any nominee to the Board of Governors.

The Board of Governors is the main decision-making body of the ABR. Historically, as you might surmise, it’s been a pretty insular group. The people already in charge nominate their friends and colleagues, most of whom have already put in the time and effort to earn a place in this upper echelon through years of service to the organization.

As I wrote in that post:

If you didn’t know, there are currently 8 Governors, and they basically run the show. Lincoln’s famous “team of rivals” approach this is not. The current people in power shall nominate their replacements and other organizations may, but the key for any hopeful member is making sure that you fit in with the cool kids, essentially guaranteeing that no one with substantially differing views would ever make it to the upper echelon.

On November 7, the ABR sent an email out to all diplomates (people who are board-certified):

The American Board of Radiology (ABR) is seeking interested candidates to serve on its Board of Governors. The Board of Governors discharges the fiduciary duties of the Board through its oversight of the business and affairs of the ABR.

The ABR recognizes the benefits inherent in a diverse community and seeks individuals with varied experiences, perspectives, and backgrounds. Interested candidates must be board certified by the ABR and willing to participate in Continuing Certification (MOC). Those from nonacademic practice environments are especially encouraged to apply.

And the nomination process for those interested candidates?

Nomination Process: Interested candidates should send a curriculum vitae and a letter of interest before November 15, 2022, to Brent Wagner, MD, ABR Executive Director (bwagner@theabr.org).

How about that.

Now presumably the actually election process is still the same (as is the impressive Conflict of Interest policy), but I believe this is the first time the ABR has publicly asked for all-comers. Now that doesn’t mean they’ll actually take any of those candidates seriously or that such candidates would ever comprise more than a single spot on the board, but—nonetheless—I would encourage anyone interested to apply.

Improving my Figs

11.10.22 // Medicine

I finally broke down and bought a pair of Figs so I could try to fit in with young folk.

But I had to do a little surgery with my seam ripper:

Just because I want to pretend I can pull off squeezing into some joggers doesn’t mean I want four(!) separate branding labels on a pair of scrubs of all things.

Now they’re closer to the fantasy world where the generic hospital scrub pants have regular pockets.

Explanations for the 2022 Official Step 3 Practice Questions

11.02.22 // Medicine

Here are my explanations for the August 2022 update of the official practice materials. (As of 2024, the PDF of these questions is no longer available, but the 2020 set is still archived and nearly identical to the 2022 PDF.)

The asterisks (*) signify one of the only two new questions compared with the prior set.

My explanations for the old 2020 set are here and the 2018/2019 set are here. There were 71 new questions in 2020 vs 2019, so going through that older set may still be worth your time. The one before that, which I explained here, was revised in November 2017.

You can find my thoughts on preparing for Step 3 here. Since writing that post, the main substantive change in the exam has been the ability to schedule CCS on a nonconsecutive day. In short, I think the free materials and UWorld should be enough for most folks. If you want book recs, they’re in that post. If you need another question source, I haven’t tried any of them, but you can get 10% off the popular BoardVitals if you’re interested by using code BW10.

As for this free 137-question practice exam, Blocks 1 and 2 are “Foundations of Independent Practice” (FIP). These should take up to 1 hour each. Blocks 3 and 4 are “Advanced Clinical Medicine” (ACM). These should take up to 45 minutes each. Total practice time should be no more than 3:30 if taken under test-day conditions.Continue Reading

Perceptions of Radiology MOC

10.21.22 // Radiology

In August, the results of a large ACR survey about radiologists’ opinions concerning MOC were released. The summary:

Similar proportions judged the existing program as excellent or very good (36%), or fair or poor (35%), with 27% neutral. MOC–CC was perceived more often as excellent or very good by those who were grandfathered yet still participating in MOC, were in academic practice, were in an urban setting, were older, or had a role with the ABR. In contrast, MOC–CC was more often judged as fair or poor by those who were not grandfathered, were in private practice, were in a rural setting, or were younger.

It’s a pretty sharp divide. Perhaps it is no great surprise that ABR volunteers and grandfathered academics are among those who view the ABR’s offering most favorably. The whole paper is worth a read, and the survey construction itself was very involved.

I’m not personally involved in any of this work, but the story behind why the survey even occurred (which I’m relaying secondhand) is perhaps the most interesting part.

If you recall, there was an ACR Taskforce on Certification in Radiology that was initially authorized in 2019 and concluded in 2020. You can read my highlights and analysis of their work here.

You also might not recall said task force, because their work apparently marks the only time in history that the ACR Board of Chancellors voted against authorizing a task force to submit their findings as a paper to the JACR. What could have been a paper shared with the broader radiology community was instead buried in a lonely random corner of the ACR website.

This is politics at work, of course.

Behind the scenes, the executive committee asked the task force to water down their language and conclusions, remove certain points, and generally “be nice.” The ACR, trying to repair some historically sour relationships with other radiology societies, didn’t want to be mean to the ABR. It probably doesn’t help when inbred leadership positions across multiple societies read like a game of musical chairs. It was apparently after multiple rounds of softening edits that the task force report was eventually buried anyway.

As a consolation, the board did permit a next-step survey in order to ascertain the true feelings of the radiology community (and not just the task force’s presumably squeaky wheels). The ACR used an outside consultant to help generate a fair survey, and then at the subsequent request of leadership, all “irrelevant” questions concerning the ongoing lawsuit, handling of COVID-19/testing delays, and the kerfuffle over the MOC agreement, etc were excised.

The survey results paper was initially submitted to JACR in 2021 and was–as you may have surmised–also rejected (though please note that the JACR is editorially independent). Much back and forth ensued–largely in order to limit perceived “bias against the ABR”–and the paper you see was finally published a year later.

In the end, thanks to editorial assistance, the limitations section is longer than the neutered discussion.

Joining and Leaving Private Equity: A Radiologist’s Story

10.19.22 // Radiology

Previously in the PE series, we spoke with someone who joined a practice that had previously been purchased (before eventually leaving). In this entry, we’re hearing from someone who joined an independent practice and was an associate in the work-up when the group sold.

Just like last time, I’ve sanitized names and some details. This case study is food for thought, not an indictment of a specific group or corporate entity.

Forgiveness App is Live

10.17.22 // Finance

The official application for the Biden student loan forgiveness is now live. Impossible to say if any legal action will block it in the end, but for now the safest thing to do if you qualify is to apply as soon as possible. It’ll take about 2 minutes.

Apply here.

Envision: The PE Healthcare Harbinger

10.08.22 // Finance

From Bloomberg, some crazy machinations involving the restructuring of Envision, the massive medical staffing company, in order to pay off debts.

This is because Envision had $7 billion debt from its 2018 leveraged buyout by KKR, and KKR needed to find ways to exploit the paper details in order to restructure the company without getting taken to court and blocked.

“Loose documents have become the norm rather than the exception,” says Damian Schaible, co-head of restructuring at Davis Polk & Wardwell. “If we go into a real recession, we are going to see more and more borrowers and sponsors seeking to exploit document loopholes to create leverage against and among their creditors.”

Here’s the quick summary, which invokes the playbook of spinning off the valuable assets in order to essentially dump the crappy ones:

Envision, which also explored a consensual debt exchange that would have raised less funding, ultimately opted for what is considered one of the most controversial and coercive out-of-court restructurings to date. The [initial] deal…would prove to be just the beginning of a series of maneuvers that eventually allowed the company to restructure the vast majority of its debt but forced creditors to turn against one another.

The strategy rested on two pillars. The first was a drop-down transaction, in which a company’s most valuable assets are moved away from existing creditors and used as collateral for new debt. The second was a series of debt repurchases and exchanges that gave certain creditors priority over others and pushed anyone who declined to participate to the end of the line for repayment.

We always talk about stocks being risky and bonds being relatively safe, but corporate debt is its own beast in this world where overleveraged companies can successfully exploit loopholes to screw over current creditors in order to obtain new financing.

Envision first designated [their profitable ambulatory surgery business] Amsurg as a so-called unrestricted subsidiary, effectively moving it out of reach of existing creditors without violating provisions in the credit agreement that prohibited moving or transferring the asset. The Amsurg assets would then be used as collateral to borrow $1.3 billion from Angelo Gordon and Centerbridge, who’d effectively be stepping ahead of everyone else in the repayment waterfall. Envision could then use the cash raised from the hedge funds to boost liquidity and to repurchase some of its existing debt at steep discounts.

Ultimately, only $153 million of the original loan was left outstanding, as owners of 96% of the debt had exchanged their holdings and waived their rights to litigate the transaction in the future.

Likely a harbinger of things to come. The “bad guys” largely win. And for creditors, better to get pennies on the dollar than nothing.

The years of low-interest easy money did a lot of emboldening and overleveraging, and everyone was eager to deploy capital on dubious deals:

There was a bitter irony in the way most of the company’s creditors ended up competing for crumbs. Four years earlier, when Credit Suisse Group AG sold the debt that financed KKR’s purchase of Envision, demand was so high that a salesperson teased investors with a picture of cake crumbs on a plate. The message then: Hurry up and grab it before it’s all gone.

The Reward of Writing

10.04.22 // Writing

Anne Lamott from her lovely book on writing, Bird by Bird:

Writing has so much to give, so much to teach, so many surprises. That thing you had to force yourself to do—the actual act of writing—turns out to be the best part. It’s like discovering that while you thought you needed the tea ceremony for the caffeine, what you really needed was the tea ceremony. The act of writing turns out to be its own reward.”

I gave a talk about the benefits of writing last year at WCICON21. I didn’t know about this passage back then, but if I had, I would have included it at the beginning.

As the absurdly prolific Brandon Sanderson repeats in The Way of Kings as one of the ideals of the Knights Radiant: “Journey Before Destination.”

Process > outcome.

A reader asked me recently about starting a website, something that happens every so often and a topic I’ve written about before. And my answer in that post, which Lamott so beautifully captures, is that the writing has to be worth it in and of itself.

You think you want to teach others, get a following, sell something, or influence (as a verb). And of course you do! You’re human. I do too. But most people can’t just write for that and keep it going very long. Because while the output can be awesome, the outcome is unknowable. The process is the only guarantee.