4 ms·

I agree that they’re quite imperfect. However, as people with Type 1 we don’t have the option of perfect. The alternative to CGMs is finger- pricking. The alter

I agree that they’re quite imperfect. However, as people with Type 1 we don’t have the option of perfect. The alternative to CGMs is finger- pricking. The alternative to pumps is long-acting insulin. A semi-closed loop pump will give you a basal rate that changes through the day, informed by the CGM. A long-acting insulin dose (eg, Lantus) will give you a basal rate that changes through the day, due to absorption rate, temperature, exercise, etc. Now, neither Basal rate will match your bodies needs minute-by-minute. But even with the CGM & looped pump regularly getting it wrong (which it does), it’s still I think getting it much less wrong than inject-and-pray long-acting insulin treatment. As for risks, I had to do two days of training to get my pump and another online course to turn in the hybrid-closed-loop. That’s more than when I was given a glucometer and insulin pen when I was diagnosed, and it’s pretty easy to kill yourself with those.

davidthewatson 4y ago
I don't think you're understanding the problem. Two days of training is incredibly short given the complexity. Killing yourself at your own hand is well-understood culturally, clinically, and legally. The mechanisms of insulin-induced trauma have been well understood since insulin shock therapy: https://en.wikipedia.org/wiki/Insulin_shock_therapy https://en.wikipedia.org/wiki/Insulin_shock_therapy and Klaus von Bulow: https://en.wikipedia.org/wiki/Sunny_von_B%C3%BClow https://en.wikipedia.org/wiki/Sunny_von_B%C3%BClow Being killed by a machine, on the other hand, directly or indirectly may be understood culturally from the long list of science fiction films featuring killer robots, but less so clinically and legally, because we don't have agreement or consensus on accidental death and the role of technology in rising risks in a complex systems environment. We need to take the implications of the design of medical device autonomy seriously as we approach these problems with those solutions, since they often fail a definition of complete autonomy, but may in fact be considered close enough to pass the same standard of risk and care as we'd give to autonomous robots in industrial settings or self-driving cars. Just because the administration of drugs doesn't seem particularly robotic, when that administration is beyond a certain percentage of machine control, I'd argue that it is behaving autonomously even though the control mechanisms fall short of machine learning. It's as if we took Therac 25 and strapped it to a patient's belt with the oncology planning encoded in the radiation device. It may seem as if the complexity is less, but we don't have consensus on that either: https://www.scielo.br/j/rbepid/a/Mn7DTtCYQLL5gkJgBb8XP3q/?lang=en&format=html https://www.scielo.br/j/rbepid/a/Mn7DTtCYQLL5gkJgBb8XP3q/?la... https://en.wikipedia.org/wiki/Therac-25 https://en.wikipedia.org/wiki/Therac-25

davidthewatson 4y ago

I don't think you're understanding the problem. Two days of training is incredibly short given the complexity. Killing yourself at your own hand is well-understood culturally, clinically, and legally. The mechanisms of insulin-induced trauma have been well understood since insulin shock therapy: https://en.wikipedia.org/wiki/Insulin_shock_therapy https://en.wikipedia.org/wiki/Insulin_shock_therapy and Klaus von Bulow: https://en.wikipedia.org/wiki/Sunny_von_B%C3%BClow https://en.wikipedia.org/wiki/Sunny_von_B%C3%BClow Being killed by a machine, on the other hand, directly or indirectly may be understood culturally from the long list of science fiction films featuring killer robots, but less so clinically and legally, because we don't have agreement or consensus on accidental death and the role of technology in rising risks in a complex systems environment. We need to take the implications of the design of medical device autonomy seriously as we approach these problems with those solutions, since they often fail a definition of complete autonomy, but may in fact be considered close enough to pass the same standard of risk and care as we'd give to autonomous robots in industrial settings or self-driving cars. Just because the administration of drugs doesn't seem particularly robotic, when that administration is beyond a certain percentage of machine control, I'd argue that it is behaving autonomously even though the control mechanisms fall short of machine learning. It's as if we took Therac 25 and strapped it to a patient's belt with the oncology planning encoded in the radiation device. It may seem as if the complexity is less, but we don't have consensus on that either: https://www.scielo.br/j/rbepid/a/Mn7DTtCYQLL5gkJgBb8XP3q/?lang=en&format=html https://www.scielo.br/j/rbepid/a/Mn7DTtCYQLL5gkJgBb8XP3q/?la... https://en.wikipedia.org/wiki/Therac-25 https://en.wikipedia.org/wiki/Therac-25