| HN Mirror

    > Guess you deleted this part.

It detracted from the overall point. But yes, I do look forward to the day artificial intelligence mostly replaces humans in this industry. I think the benefits are self-evident.

    > If a better system existed given the current set of parameters, I believe it would have come into existence.

One of these "parameters," as your call them, includes the AMA and doctors restricting the number of new doctors who can be trained in the past https://usatoday30.usatoday.com/news/health/2005-03-02-docto... , actively fighting against allowing Nurse Practitioners and other trained providers from providing routine services that would lessen the requirement of doctors https://blog.petrieflom.law.harvard.edu/2022/03/15/ama-scope... , fighting against collection of misdiagnosis data (as mentioned above), lobbying against banning pharmaceutical companies from giving money or gifts to doctors (in fact, the majority of doctors say that accepting these gifts is OK, despite research repeatedly showing that these nudges change behavior and lead to practises like overprescribing opiates — https://www.statnews.com/2020/12/04/drug-companies-payments-... ), refusing to adopt practises that reduce mortality (see: the checklist example from above) and on and on.

This trend is not unique to the US. Similar efforts exist everywhere doctors do. The profession was associated with status in the past (and still is). Doctors are fighting and have fought tooth and nail to preserve the benefits they accrue from this status.

As you're a physician, I think you should take a moment to put yourself in the shoes of a third party. Imagine there's a Profession X. Members of this Profession X, and organizations that represent Profession X, have fought to make outcomes of people who interact with them worse in measurable ways. What would you think if a member of Profession X made an equivalent claim? Would you give it equal weight?

    > Your AI system would only be accurate at diagnosis if it has all the pertinent data and unlimited resources

No, it doesn't need unlimited resources. Just better sensors and more data. We'll get there sooner or later. It's an inevitability. There are too many smart people working towards this end because of the reasons outlined above.

    > Patients often provide an unclear history, their symptoms are not common for their underlying condition, and there are not enough resources to do a full workup for every complaint.

Medicine is the only profession I know of where its practitioners fight against gathering more data points. The argument goes that the more you look, the more you find and that's bad etc.

If you take some time to reflect on it, these fairly common claims make no sense. The more data we collect, the more data we have to understand what the true distribution of human biology looks like. The more data we have, the more information we can gather to better understand how to treat when needed, and distinguish between things that are critical and aren't.

As to the claim that there aren't "enough resources," the truth is that we can create the resources. Lateral flow tests, for example, have dropped significantly in price and what we can test with them has greatly increased. We can also automate how these tests are read — an effort I've been peripherally involved in. We can build labs on chips and mass produce the silicon, https://en.wikipedia.org/wiki/Lab-on-a-chip . We can take commodity sensors and use better algorithms to detect subtle things — for example, using IMUs on a patient's bed in an ER setting as a type of ballistocardiograph to passively monitor their heart function.

We have the technology to do all of this and more. We also have the capability to invent new things that do more. But we choose not to. This status quo is a choice.

    > Regarding your second link regarding misdiagnosis in emergency rooms, you just need to look at the PERC rule or the HEART score that ED physicians use for evaluating for pulmonary embolus or acute coronary syndrome to get a somewhat clearer picture of the deviation. Even with scores of zero, there will be a substantial amount of missed diagnoses. Not everyone who goes to the emergency room is going get a full cardiac evaluation or a CT angiogram of the chest if their symptoms don't suggest the illness. There are not enough resources to do this.
    
    > 250 * 9 = 2250 preventable deaths over a 9 year period.

    > Far less alarming than the data you present of 198,534 missed diagnoses.

With all due respect, I feel like you've validated my argument in these paragraphs. Medicine is a safety critical industry that doesn't behave like it's a safety critical industry.

If a process had a 33% rate of failure, for any other safety critical system, there would be a sustained effort to research and develop a better alternative.

Perhaps we need better sensors — maybe we could embed a dozen or so IMUs around the patient and measure the forces to see the performance of their heart https://journals.biologists.com/jeb/article/225/10/jeb243872.... Perhaps we need to expand the metrics we look at. Perhaps the way things are measured at the bedside should be changed. There are many things we can do to reduce the absurdly high process failure rate of 33%.

Detecting and treating disease early leads to better outcomes. 198k people were denied this better outcome.

    > 250 * 9 = 2250 preventable deaths over a 9 year period.

The data is for the UK. As the population is roughly a fifth of that of the US, let's do some rough math and multiply that number by 4 (instead of 5, to be conservative). That's 1,000 lives per year, or a death every 8 hours.

What does this rate look like when compared to other safety critical systems? The one I'm most familiar with is aerospace. Within civilian aerospace, parts are rated to 1 failure in 1 billion hours of operation. Or, more broadly, safety critical systems are designed to hit the goal of 1 death per 1 billion hours. https://dl.acm.org/doi/10.1145/332051.332078

    > Of course other civilian professions don't have death rates this high. They don't deal with dying people every day. Again, I'm not claiming that the medical field is perfect as it is. I just believe you are mischaracterizing data for which there are many intricacies.

All of the statistics I've given were for people killed via mistake. These are people who would have otherwise lived. They died because a medical professional made a mistake. That's an important distinction.

Furthermore, although medicine deals with the sick and the dying, it's not the only industry where a very small mistake could equal death.

Small mistakes, like a bolt not being tightened correctly, on an airplane can lead to catastrophic failure (and have!). And yet, an upward of 2.3 billion person trips occur safely every year via airplanes. Between 2012 and 2021, in seven of these ten years, no airliners crashed. No catastrophic failures occurred.

By contrast, the total number of ER visits is 131.3 million. As I've stated, tens of thousands of these visits lead to death for the patient via error. Per year.

Why are these fields so different?

I believe that the difference is in the margins. One has a culture of excellence. The other has the culture of shrugging. The smallest change in outcome probabilities adds up for processes. Improving a process by 0.5% to 0.05% per case doesn't seem like much, but it adds up.

Interestingly enough, great doctors recognize this. The very best doctors fight in these margins,

    > “Let’s look at the numbers,” he said to me, ignoring Janelle. He went to a little blackboard he had on the wall. It appeared to be well used. “A person’s daily risk of getting a bad lung illness with CF is 0.5 per cent.” He wrote the number down. Janelle rolled her eyes. She began tapping her foot. “The daily risk of getting a bad lung illness with CF plus treatment is 0.05 per cent,” he went on, and he wrote that number down. “So when you experiment you’re looking at the difference between a 99.95-per-cent chance of staying well and a 99.5-per-cent chance of staying well. Seems hardly any difference, right? On any given day, you have basically a one-hundred-per-cent chance of being well. But”—he paused and took a step toward me—“it is a big difference.” He chalked out the calculations. “Sum it up over a year, and it is the difference between an eighty-three-per-cent chance of making it through 2004 without getting sick and only a sixteen-per-cent chance.”

From the cystic fibrosis piece.

Medicine has failed to broadly adopt this culture.

> You are also throwing out a lot of strawmen regarding race and sex which bears little relevance to your initial claim.

They aren't treated seriously by their doctors. That leads to poor outcomes.

Ideally, medicine should take every possible cause of process failure seriously, especially if that cause is observed to be this common.