A ROC curve shows the relationship between a sensor’s sensitivity and specificity. If the sensor is sensitive but can’t discriminate a true positive signal from a false signal it is useless. That was the problem I have written about on the subject of mid course missile defense and the problem of discrimination of real targets from decoys. Since we could not find a practical approach to discrimination, we could not make mid course intercept work. We expected the interceptor missile would have a terrible ROC curve if the adversary deployed effective decoys.
Let’s say the problem is to deploy a sensor in a coal mine to protect the miners. Since a canary succumbs to CO before the miners are affected, the canary provides a useful warning. The ROC curve for the canary makes it an excellent sensor to detect a poisonous environment. We could instead find a particularly weak miner and use the miner as a warning indicator, but waiting to see if miners show some symptoms would not be a reasonable approach. To improve the miner as a sensor, we could try to create a ROC curve for miners based on if they they have CO in their lungs, are pre symptomatic, slightly sick with a fever, or just becoming unconscious, but the canary is certainly the preferred approach.
So would we use coal miners as sensors and then shut down coal mines if miners die on the job? Instead we could ventilate the coal mines to avoid CO build up and still use the canary, or maybe install a more modern CO sensor and keep the coal mines operating. I admit coal mines are going away, but the metaphor might be useful.
So I wondered, why we are using people as sensors for detecting virus containing aerosols that are inhaled? The ROC curve for the RNA sensor has not been published, but I suspect its performance depends on when the sample is collected relative to time of the exposure of the person, how well the swab is stuck up somebody’s nose, and the sensitivity of the analysis. One article claimed the false negative rate was 25%, but no ROC curve was published. An additional sensor such as fever detection could result in excessive false positives. If just a fever detector is used there would be unacceptable false negatives.
If instead, if we have a closed space, such as an assisted living center, a business, air plane, restaurant, movie theater or factory, it could be ventilated with filtered air, sterilized and an aerosol analysis sensor such as an optical bio sensor could be used as a detector.
The optical sensor might be extremely sensitive and specific to the virus molecule. Except it does not exist as yet, but it is not impossible….I guess since considerable research is under way.
People would enter the protected environment through a revolving door that has its own aerosol sensor to instantly detect if the person entering has contaminated breath or clothing, and if so, the person would be rejected. This suggestion would require a major architecture redesign approach involving reliance on this as yet unavailable canary.
SDI Guy 
Simple breathing is unlikely to produce aerosols. you need sneezing or coughing. you can solve that problem by injecting a sneezing inducing substance (snuff?) into your revolving door space.
You still need to sense a rather specific virus in the aerosol droplet not just the droplet in real time.
LikeLike
The entire detection and response to bio threats would benefit from risk systems engineering. This will be increasingly important as we try to resume a bio threatened society, in some new form.
LikeLike
Sal Alfano and I published a paper in Astronautica Acta on developing ROC curves for satellite collision avoidance approaches. Virtually ALL collision warnings are false. I cannot explain why they keep doing the task this way. Perhaps they do not understand the deficiency, or they have to do something. Many think it works because there have been next no collisions. ROCs are useful in many venues. If we built ROC curves for TSA, would we keep spending on finding hair curlers in old ladies handbags? BTW, we build multidimensional ROC trajectories when there are several discriminants.
LikeLike
Examples of additional
discriminants are a fever and additional tests at various time intervals. A systems engineering approach is needed.
LikeLike
Certified canary protected facilities would be commercially successful.
LikeLike
We need more of these ideas until someone achieves the ultimate out-of-the-box concept that outfoxes the COVID-19 virus which is neither alive or dead and which has only one job to do … get body cells to produce more COVID-19 viruses. More thoughts are needed, unconventional ideas like handwashing was to controlling infection in the operating room, hindsights, insights, reach-arounds. Quickly … as we are in trouble and suffering from this biological insidiousness!
LikeLike
Oved suggestion an induced sneeze to capture a sample. Another friend suggested blow into a tube like a DWI test. Another friend suggested also measure fever and use facial recognition to drive Cell phone GPS record.
LikeLike
The loss of taste and smell is an unusual early COVID-19 sign that people should be on the lookout for, add a whiff of ammonia and track the resulting reaction?
LikeLike
Sir: In the 80s we had the pleasure of sharing our sailboat in the Kwajalein Lagoon with and individual purported to be the SDI Chief Scientist. We were followed by an inflatable boat loaded with armed guards. We’ve lost our records. Could that have been you?
LikeLike
Armed guards seems likely, but SDI Chief Scientist unlikely.
LikeLike