Ryan's Journal

My undergraduate education included case studies of really, really bad engineering mistakes, like the Therac-25 incident (also in Wikipedia) wherein poor software design lead to a radiation therapy machine administering lethal or nearly lethal doses to patients who, insult to injury, were already suffering some form of cancer.

Apparently this lesson isn't taught as widely as it needs to be. According to the L.A. Times, it took eighteen months for Cedars-Sinai hospital to realize it was giving radiation overdoses with a diagnostic machine to some two hundred patients. They're blaming it on computer programming errors (edit: but not any more; now it's an error in "resetting the machine.") While it likely isn't possible to kill a patient in one go with a machine built for diagnosis, one might still consider having a "misunderstanding about an embedded default" a dramatic failure when the system in question affects actual living people. The manufacturer of the machine, G.E., claims "no malfunctions or defects" in the hardware, but failing to be clear in documenting the purpose of a default isn't so easy to dismiss. And hardware that can operate at overdose level is quite possibly malfunctioning right out of the gate.

Therac-25 should be required reading for anybody who works on medical computer systems. The inability to learn from the past reeks here as they doom others (to no more than patchy hair, thankfully) while they repeat it.

You must login to leave a comment