Slashdot Mirror
Radiation Therapy Mistakes Cost Lives
jmtpi recommends a long NY Times investigative report about how powerful medical linear accelerators have contributed to at least two deaths in the New York area. Although the mistakes were largely due to human error, buggy software also played a role. "...the records described 621 mistakes from 2001 to 2008... most were minor... The Times found that on 133 occasions, devices used to shape or modulate radiation beams... were left out, wrongly positioned, or otherwise misused. On 284 occasions, radiation missed all or part of its intended target or treated the wrong body part entirely. ... Another patient with stomach cancer was treated for prostate cancer. Fifty patients received radiation intended for someone else, including one brain cancer patient who received radiation intended for breast cancer."
People make mistakes with technology which results in unintended consequences. Giving someone treatment for the wrong disease may have adverse side effects.
Basically this only proves that people are stupid in general. I don't see anything wrong with this technology.
Sewage Treatment Facilities - "Our duty is clear."
This is scary indeed when you are under the assumption that you are going into a hospital or clinic for a treatment like this, and assuming he/she is well trained and know what they are doing to your body... to read something like this makes one wonder just what, if any training they get to operate these machines?
How did one guy above me put it... Highly trained morons? I have to agree!!!
You must master your joystick like a fisherman masters bait! - Gimpy
Therac-25 is only the most prominent medical radiation incident from the past 20 years or so. The IEEE linked at the bottom explores problems with replacing hardware interlocks (mostly literal interlocks) with software interlocks, which fell prey to memory errors, bugs and human intervention. Tools like this require constant diligence and skepticism, which is nearly impossible to maintain when faced with incentives to update, promote and distribute new technology. I suspect this will devolve into some meta-discussion about regulation, but look closely at the allegations regarding cover-ups in the Therac-25 case and this article--market response presupposes that customers and investors are informed about errors in products. Where companies downplay or obfuscate errors of this magnitude, public choice fails. Regulatory bodies won't work perfectly, but I suspect that their intervention in the market would reduce these errors at some high but acceptable cost (in either monetary terms or terms of new technologies forgone due to the cost of compliance).
People did learn...
And then they got laid off, and replaced with outsourced development companies from India, who haven't learned yet, or just don't care as much.
These numbers don't mean anything unless we know how many procedures are conducted in total. It could be that the probability of a fatal complication, defined as (# of fatal complications) / (# of procedures) is quite low.
I took a course on clinical decision analysis last semester. Every intervention, even diagnostic ones, carry a risk. The risk needs to be weighed versus its benefit to determine its overall efficacy. If the patient is very ill and has a short life expectancy or very low quality of life, then even dangerous procedures become acceptable.
One can conduct analyses based on expected life expectancy, QALYs (Quality-Adjusted Life Years), QOL (Quality of Life), or from a purely economic point of view. How much is a patient's life worth? Is a 5-year-old's life worth more than an 85-year-old's life? What about a 45-year-old? This can get quite philosophical. One could even conduct an analysis against a combination of outcomes, though how we choose to weigh the different outcomes is arbitrary.
Bayesian probabilities figure heavily into these analyses, and they can give quite counter-intuitive results. For example, if a test for AIDS is 99% "accurate" (in terms of sensitivity and specificity), it can still have a very high false positive rate (if AIDS is rare in the general population). In this sense, the AIDS test carries a toll of emotional devastation for the false positives. It can be a challenge to convince the general public, even your average physician, of the validity of a model. A good model will have conducted sensitivity analyses to allow for the possibility that a given procedure may have a higher (or lower) risk than expected.
Alternate title: Radiation Therapy Success Saves Lives
Not to put too fine a point on it, but buggy software is also human error.
What do you mean they cut the power? How can they cut the power, man? They're animals!
Blaming software isn't the answer(outside of specific software bugs); but blaming humans, while fun and morally satisfying, is also dubiously useful from the perspective of the system as a whole(this does not, of course, mean that you should feel any compunction about sacking egregious cases).
For instance: The radiation shield/guide setup. Yeah, the nurse should have installed it, and she fucked up. However, it is a basic fact of humans that all of them fuck up from time to time, some more than others, and more under some conditions than others. Unless that particular nurse has an atypically bad record for forgetting, it is unlikely that firing her will improve the quality of the system as a whole very much. Instead, such safety critical systems should be designed to take human error into account. Routine use of checklists, for instance, has been demonstrated to reduce human error. Or, for the more high tech approach, the Radiotherapy machine could have a few extra sensors(RFID and optointerrupters) and the shield and guide units could be RFID tagged. If the machine does not detect the presence of the correct guides in the correct locations, it alerts the operators and refuses to provide a beam.
Humans are flawed, often annoyingly so; but they are what we have to work with. Luckily, it is possible to systematically characterize the form of flawedness exhibited by humans(eg. limits of short and long term memory, probability of making an error on a procedure of given complexity as a function of experience, and so forth) and design systems that, as much as possible, are resistant to those errors. This requires a combination of organizational changes(eg. control of working hours, verification of nonimpairment for critical staff, enforced use of checklists and procedures, firing atypically unreliable staff) and technological changes(substitution of highly reliable barcodes/RFIDs for unreliable handwriting, automated sanity checking, marking patients before surgery, machines that refuse to operate unless their interlock conditions are met, etc.)
Some of this is just a matter of time, some of it will piss off doctors, and some of it will probably piss off patients; but building reliable systems is possible.
While, as nerds, everyone here leaps to 'computer error'(And everyone mentions that Therac-25 disaster we all learned about in comp sci 101.), computers aren't really responsible for a brain cancer patient getting treatment for breast cancer.
A computer might say where to aim the machine, but someone who was even slightly familiar with the case would say 'Um...the breasts? No, that can't be right.'.
What is responsible is the constant reduction in the amount of staff at medical facilities, and consequently, the inability for any actual checking or familiarity with patients.
Read the horrific description of what happened to Jerome-Parks, please notice that it was people trying program crashing machines, machines that were obviously screwed up, and no one bothering to actually look at the result. And then doing it twice more because no one bothered to look into the obvious mistake.
Essentially, the problem here isn't the Therac-25 one, where a shitty user interface resulted in the screen saying one thing and doing another. Note that in every described situation, the machine clearly described what it was doing. It wasn't 'doing something else besides what it said', it was doing what it had, incorrectly, been told to do. It said it was doing it, it did it. The machine worked perfectly.
It is equivalent of being a newspaper reporter, and Word crashes while I save my article...but I submit it anyway, and the front page of the newspaper is filled with gibberish. You know whose fault that is? Sure as hell not Word. It's my fault, it's the editor's fault, it's the guy doing the final check before the print run. If I were to claim the solution to this constantly happening was 'crash-proof software', I'd get laughed out of society.
Oh, but newspapers actually, you know, pay people to check that before spending thousands of dollars doing a print run. If only someone's life was worth more than that.
Yes, we can argue the machine should have fail safes to stop them from working in obvious stupid situations, but this just stops obviously stupid situations, and only overdoses. What is that is a perfectly reasonable dose...aimed at entirely the wrong spot, for someone with an entire different type of cancer?What if it's 100x what you should be getting, but still within the bounds of reasonable for certain extreme types of cancer? What if that is, in fact, practically no dose at all, so you die of a fucking treatable cancer because you got not treatment?
More to the point, why are we worried about this, when drug errors kill ten thousand times as many people? (Because machines often do have failsafes, unlike prescriptions.)
If only we had a system where all the money wasn't sucked out of the system by insurance companies, one where we actually paid to have competent medical staff who could actually watch what was going on, instead of spending ten damn seconds a patient.
If corporations are people, aren't stockholders guilty of slavery?
Heck, Therac-25 is the freakin' case study that people use to learn about the possible consequences of bad software design. You'd think somebody at the FDA would have heard of it and made some sort of link to the work they were doing before approving the successor to the Therac-25.
outsourced development companies from India, who haven't learned yet, or just don't care as much.
Unfortunately, it's the latter and not the former. And by the way, it's not "India" that is the problem, but "outsourcing company". I have worked with some fantastic Indian developers, but they don't work for outsourcing companies. "Cheap" outsourcing companies are not good at developing software, they are experts in sending out invoices.
Science advances one funeral at a time- Max Planck
People survive just fine with one arm? People survive just fine that are blind? People survive just fine who can't hear? People survive just fine with a permanent colostomy bag? People survive just fine with their tongue cut out? People survive just fine with one leg? People survive just fine with their face melted off with acid or burned off in fire? People survive just fine with their testicles mistakenly removed? People survive just fine with their dick cut off? People survive just fine without a bladder of any sorts?
You're welcome to accept any of these conditions any time you want jack-ass!
Over-the-top Response Guy! Giving "Over-the-Top Responses" since 1970.
CHECKLISTS! Pilots have an easy job and they need them. huge benefits resulted from giving them checklists.
Doctors and medical workers must be forced to use checklists. period.
Democracy Now! - uncensored, anti-establishment news
These "Highly Trained Morons" are working on killing my wife. She went in for a Hysterectomy and ended up with her ureter sutured or cauterized shut resulting in her kidney backing up and shutting down. Now she has a tube out her back to keep her kidney alive and in a few weeks they'll go in an cut her ureter above the blockage and reattach it to her bladder. All for the low, low, price of $$$$$$$$$$$$. Meanwhile, the nursing staff and E/R staff have done everything in their power to see how much additional damage they can do. No one has any common sense or care that I can see. I'm fit to be tied!
If you survive a hospital stay for anything serious, it's either luck or because you had reasonably intelligent friends and family looking out for you the whole time. Heaven help anyone without such a network of support. It helps if they're taking notes -- keeping their own charts, as it were. Twist all the arms you can, call in all your chips, and good luck.
I am not a crackpot.
So why does public health care also work in Australia, Canada, the UK, France, Germany, Brazil, Chile, Ireland, Belgium, Japan, Italy, Sweden, Switzerland, Finland, Israel, Taiwan, even in the USA under Medicare, Medicaid, military and veteran health care, OMG AMERICA IS SOCIALIST!
"Indeed, most of these errors don't sound like Therac-25 type errors, more like PEBKAC errors."
Actually, a lot of these sound like Therac-25 type errors - not to mention that mission-critical software that can cause harm when used incorrectly should be designed in a way that recognizes PEBKAC errors exist.
The article talks about the Varian software a little. For one of the machines, the procedure appears to be:
1. set treatment parameters, click "save"
2. set the part of the body to scan, click "save"
3. set some other parameters, click "save" to finalize the plan
The operator did 1), and finished 2) but the program crashed when she hit "save". She then restarted the program and did 3), thinking that it in fact saved part 2) (but it didn't). From a software point of view:
1. the program shouldn't crash. It certainly shouldn't crash with the frequency observed (it crashed two or three more times during the treatment involved in this article).
2. the program should never have let treatment proceed without getting proper answers for part 2).
3. the software design should do all within its power to validate a treatment before allowing use on a patient. It could involve running simulations (eg. with all the requested parameters, is any body part exposed beyond safe limits?), or it could involve a physical test (the article says that it was common practice to do a test run first, but not required - it was skipped due to understaffing). Require the test, and _don't_ proceed until the operator does it.
All three of these are exactly Therac-25 problems (eg. program crashed due to a buggy key input routine, allowed the operator to proceed to the next screen without setting values, and neglected to double-check parameters for safety).
Fewer PEBKAC errors would have helped them catch the error before it became a problem, sure. But a design that doesn't take all reasonable precautions to eliminate human error - or, worse, that counts on the user to fix software errors - is flawed.
I am also astounded by the poor follow-up. The guy's wife complains to his doctors that something is wrong, and (instead of suspending therapy) they have her talk to a psychiatrist and give him another session of the botched radiation exposure. You could describe this as PEBKAC, but I'd say it's really a matter of procedure ("human" software). Sometimes, things just go wrong even though everyone did the right thing. But, based on the article, this really isn't one of those cases.