This proof-of-concept study trialed a somewhat Rube Goldberg auto-injector device which utilizes two 3 axis accelerometers attached to the abdomen to detect central apnea events indicative of overdose. If apneas of more than a preset duration or frequency are detected, the device triggers an injection of naloxone. The team who developed this device may have some useful lessons re: using accelerometers to measure gross body motion and infer respiratory patterns?

Closed-loop wearable naloxone injector system.

Chan et al.

Scientific Reports 2021

https://www.nature.com/articles/s41598-021-01990-0

I agree that an “auto injector” device in thr abdominal area is unlikely to succeed as a clinical intervention. Perhaps someday a transdermal wristband for naloxone. At this stage, I am just doing preliminary research into the feasibility of a wristband that would serve as an early-warning device. It would help if that device could be cheap. Accelerometers are cheap. Do you have the cite of any research or documents available detailing the abdominal naloxone contraption?

Yes, that prototype auto-injector looks very clunky and inconvenient and even with streamlining I suspect it will remain too expensive and bulky to be of any practical utility. I linked to it because the researchers mention that they found they had to use two accelerometers placed independently on the abdomen to gain reliable measurements of respiratory rate and to detect apnea events.

It's possible that a single accelerometer mounted on the wrist may not be effective however good your code is. (I imagine a wristband device with a single accelerometer could easily be set up to measure pulse rate and BP, but for timely intervention in an opioid overdose you would want to be measuring respiratory depression, not pulse rate).

I just found the full text here on RG and this may contain more information on how the device was set up to detect apneas;

"A challenge with using data from a single accelerometer is being able to reliably track diminished respiratory motion, which occurs in the setting of opioid use. An accelerometer measures the sensor’s acceleration relative to the force of gravity. Prior work (14) has shown that performing a double integration of acceleration values to calculate translational motion magnifies noise and error in individual measurements. This is because accelerometer data in practice is not bias-free, and bias increases at a rate of t2 after double integration of the signal where t is the time.

There are two limitations to using a single accelerometer design. First, the result is prone to interference such as involuntary body motion, which introduces an additive time-variant bias that hinders correct respiration detection. Second, even though most surfaces on the chest and abdomen rise and fall during inhalation and exhalation, the single accelerometer may be placed in a position on the abdomen that only exhibits a small amount of rotational change, which the accelerometer is not sensitive enough to detect (see Supplementary Fig. 2). Further, we observe that in some subjects, particularly those with a higher body mass index (BMI), it can be challenging to reliably extract the breathing signal using a single accelerometer. Two reasons account for this. First, the abdominal movement in these subjects can be shallower due to increased effort required during respiration. Second, the amount of breathing motion can vary significantly between persons based on where the sensor is placed on their abdomen. To address this, we used two accelerometers to detect rotation at two points of the abdomen, with each separated by a small distance. This reduces the likelihood that both the two positions suffer from minimized rotational change. Additionally, this allows us to combine the signals across the two accelerometers to more accurately extract breathing motion."

Article Closed-loop wearable naloxone injector system

Devin, I imagine you have already seen the apps that use a smartphone to screen for sleep apneas, and the subsequent efforts to adapt this type of software for detecting overdose events?

Article Screening for obstructive sleep apnea with novel hybrid acou...

Article Opioid overdose detection using smartphones

Article Layperson reversal of opioid overdose supported by smartphon...

Yes I have looked at those, and I agree with you that looking at respiratory depression is key. I am trying, ultimately, someday, to look at respiratory rate (RR) and use that to predict *mortal* levels of declining RR as opposed to merely declining RR (say, sleeping or mere unconsciousness with very little mortality risk). I sent you the link to work done by the MIT Media Lab in 2015 using a wrist accelerometer; they have been able to roughly match hospital grade equipment for RR off the wrist with a single 3-axis accelerometer. We have collected some accelerometer data and just *inferring* the regression algorithm from the paper (in MATLAB) we have been able to get decent results. I also found out that one of the authors (the paper was never *published* except here on ResearchGate) is now at Microsoft and I've reached out to him. I don't have a problem getting the code written from scratch; it would just be nice to know exactly what the MIT regression algorithm is that got them such good results.

Thanks for the 2015 conference paper by Hernandez et al. If the accuracy the MIT team reported is reproduce-able in community settings it sounds like your concept could be implemented relatively cheaply. Following with interest!