Joined 13 years ago
To further the learning, can anyone guess which part of the stack is likely generating the error?
I'm 40+ and have a child under 4. When they were born I cut my work back to 1/3 so that I could be a home dad.
It's the best decision I ever made and well worth the (real) financial pain. I can't convey in this text box how fulfilling being a dad has been. This is a time and joy you can never get back. Your kid will be better off from time they spent with you than what the money could have bought.
Raising a kid is a never ending exhausting slog. Two parents that can do everything means you can both get some mental respite. I'm sure our relationship would be worse now if I'd been away working every day. You get to know a lot of other parents and the tension provider/parent puts on relationships is real.
There are downsides. With my wife and I both working part time (nights, naps, 2 days of kinder) we are going backwards financially, but we will survive until school when I'll go back full time. I'm extremely grateful to work in an occupation where that's even possible.
Often he would question something fundamental about human interaction that seemed to me (and I suspect most) to have obvious and quite simple explanations. But he is intelligent and couldn't be missing the forrest for the trees I told myself. Clearly he must be operating many steps ahead. Levels above!
Now I wonder if maybe he just isn't so good with trees.
I do not mean to dismiss the entirety of his work. No doubt much of it has been valuable. He seems to struggle to understand people, in much the same way as one of the alien's he's obsessed might struggle.
It was visually a pleasant read.
The purpose of the program is to watch a large population of COVID patients who are not sick enough to warrant being in hospital. Miserable no doubt, but OK. The purpose of the monitoring is to identify those who are getting sicker before they have become very sick.
The goal is not to have the system calling ambulances. It can (and has) but we should be spotting degradation and reacting before it comes to that. To that end we measure three vitals (temperature, heart rate and oxygen saturation). They are not each equal predictors of the sickness we wish to spot. They are also not linear indicators of sickness.
We have three levels of response to patient vitals. There are some automated suggestions for patient comfort at the bottom end. There is the metcall at the top end. But the most important is the Clinical Review in the middle. Clinical Review connects the patient with a clinician by phone so that we can dig into their situation.
Temperature it turns out is not an especially good marker of the sickness we need to find in the group of people we are monitoring. "[with temperature] there is a U shape association with mortality, but the error bars at any point a super wide". It has some correlation, but high temperatures are not a good enough predictor of what we're trying to spot. Oxygen Saturation is good because of what COVID can cause in the lungs. That can happen without the patient being aware of it so it's critical to what we must spot.
Infection alone wont bring someone up to a temperature that high (we're monitoring people with covid, not amphetamines overdoses or desert marathons). So COVID monitoring is not dependant on spotting patient temperatures that high, and if it was that high it would not be on account of the virus. For those patients this isn't the only healthcare the patient is involved with.
So why do we bother to give the patient a thermometer? It's easy to measure. The devices are cheap and readily available. It's useful at the slightly-elevated end and helps us recommend paracetamol if/when appropriate. Tracking a patient temperature over their monitoring period feeds data back into the ongoing analysis of the virus. We ask a number of supplementary questions of the patients each day. Hopefully there will be trends.
So finally, why is there a temperature threshold at 42 degrees? More an accident of the project history. We'd built the multi-step thresholds for each of the measurements, but it turned out the highest temperature one really isn't applicable in our project. The patient hits Clinical Review well before they get up to those temperatures. It may come out completely in a newer version.
Also they added "Whilst not completely analogous, we draw your attention to the NEWS2 illness scoring system used by the National Health Service (UK) (https://www.mdcalc.com/national-early-warning-score-news-2). The system uses various parameters to calculate a score of 'how sick they are'. Whilst many measurements have a 3+ (add 3 to the score) threshold values, the NHS does not have a temperature threshold value 'worthy' of contributing 3 points to your illness score."
The figures are actually set per-patient. But that 42 as placeholder does seem odd. There are two lower thresholds that trigger before getting to that one. But I’ll flag it with the clinician regardless. Thank you.
https://github.com/rmhcovid/screentool
And we have another project for operating drive-through testing clinics that will hopefully be published in the next few weeks. It has been shared through hospital channels but isn’t up on github yet.
As I understand it there is a lot of communication going between emergency teams and specialists around the world and the odd pre-publication paper. This work has come out of the RMH's emergency department. Full credit goes to Dr Martin Dutch who's idea the whole thing was. As best I know he came to the idea (tracking sats via cheap electronics to spot risky covid patients) quite early, and independently. He also built an initial prototype. I'm a developer who was brought in to build it out into a working system during the first phase of Australia's covid outbreak.
PS: If you're thinking about using the system, it's worth having a read through the redcap_design_overview.md documentation. Building something like this in a research-survey tool was not easy and there's plenty of things in the design that will be head-scratchers when seen for the first time. Now that the model is proven we'll hopefully build out a 'real' version of it at some point.
We're happy to assist where we can to see other hospitals (particularly resource-constrained ones) pick it up. The operating costs are very low (a server + sms gateway cost). The hospital has also open sourced some other covid tools developed this year.
