It was involved in at least six accidents between 1985 and 1987, in which patients were given massive overdoses of radiation.:425 Because of concurrent programming errors, it sometimes gave its patients radiation doses that were hundreds of times greater than normal, resulting in death or serious injury. These accidents highlighted the dangers of software control of safety-critical systems, and they have become a standard case study in health informatics and software engineering. Additionally the overconfidence of the engineers:428 and lack of proper due diligence to resolve reported software bugs are highlighted as an extreme case where the engineers' overconfidence in their initial work and failure to believe the end users' claims caused drastic repercussions.
Direct electron-beam therapy, in which a narrow low-current beam of high-energy (5 MeV to 25 MeV) electrons was scanned over the treatment area by magnets;
Megavolt X-ray (or photon) therapy, which delivered a fixed wide beam of X-rays, produced by colliding a narrow 100-times higher current beam of 25 MeV electrons with a target, then passing the emitted X-rays through both a flattening filter and a collimator.
It also included a "Field light" mode, which allowed the patient to be correctly positioned by illuminating the treatment area with visible light.
The six documented accidents occurred when the high-current electron beam generated in X-ray mode was delivered directly to patients. Two software faults were to blame. One, when the operator incorrectly selected X-ray mode before quickly changing to electron mode, which allowed the electron beam to be set for X-ray mode without the X-ray target being in place. A second fault allowed the electron beam to activate during field-light mode, during which no beam scanner was active or target was in place.
Previous models had hardware interlocks to prevent such faults, but the Therac-25 had removed them, depending instead on software checks for safety.
The high-current electron beam struck the patients with approximately 100 times the intended dose of radiation, and over a narrower area, delivering a potentially lethal dose of beta radiation. The feeling was described by patient Ray Cox as "an intense electric shock", causing him to scream and run out of the treatment room. Several days later, radiation burns appeared, and the patients showed the symptoms of radiation poisoning; in three cases, the injured patients later died as a result of the overdose.
A commission concluded that the primary reason should be attributed to the bad software design and development practices, and not explicitly to several coding errors that were found. In particular, the software was designed so that it was realistically impossible to test it in a clean automated way.
Researchers who investigated the accidents found several contributing causes. These included the following institutional causes:
AECL did not have the software code independently reviewed.
AECL did not consider the design of the software during its assessment of how the machine might produce the desired results and what failure modes existed. These form parts of the general techniques known as reliability modeling and risk management.
The system noticed that something was wrong and halted the X-ray beam, but merely displayed the word "MALFUNCTION" followed by a number from 1 to 64. The user manual did not explain or even address the error codes, so the operator pressed the P key to override the warning and proceed anyway.
AECL personnel, as well as machine operators, initially did not believe complaints. This was likely due to overconfidence.:428
AECL had never tested the Therac-25 with the combination of software and hardware until it was assembled at the hospital.
The failure occurred only when a particular nonstandard sequence of keystrokes was entered on the VT-100 terminal which controlled the PDP-11 computer: an "X" to (erroneously) select 25 MeV photon mode followed by "cursor up", "E" to (correctly) select 25 MeV Electron mode, then "Enter", all within eight seconds. This sequence of keystrokes was improbable, and so the problem did not occur often and went unnoticed for a long time.
The design did not have any hardware interlocks to prevent the electron-beam from operating in its high-energy mode without the target in place.
The engineer had reused software from older models. Such methods manifest in so called cargo cult programming where there is blind reliance on previously created code that is poorly understood and may or may not be applicable. These models had hardware interlocks that masked their software defects. Those hardware safeties had no way of reporting that they had been triggered, so there was no indication of the existence of faulty software commands.
The hardware provided no way for the software to verify that sensors were working correctly (see open-loop controller). The table-position system was the first implicated in Therac-25's failures; the manufacturer revised it with redundant switches to cross-check their operation.
The equipment control task did not properly synchronize with the operator interface task, so that race conditions occurred if the operator changed the setup too quickly. This was missed during testing, since it took some practice before operators were able to work quickly enough to trigger this failure mode.
The software set a flag variable by incrementing it, rather than by setting it to a fixed non-zero value. Occasionally an arithmetic overflow occurred, causing the flag to return to zero and the software to bypass safety checks.
The software was written in assembly language that might require more attention for testing and good design. However the choice of language by itself is not listed as a primary cause in the report. The machine also used its own operating system.
Leveson notes that a lesson to be drawn from the incident is to not assume that reused software is safe:
"A naive assumption is often made that reusing software or using commercial off-the-shelf software will increase safety because the software will have been exercised extensively. Reusing software modules does not guarantee safety in the new system to which they are transferred..." This blind faith in poorly understood software coded paradigms is known as cargo cult programming. In response to incidents like those associated with Therac-25, the IEC 62304 standard was created, which introduces development life cycle standards for medical device software and specific guidance on using software of unknown pedigree.