@ATieLikeRichardGere
posted slightly silly. sorry about that.
I was interested in how cluster analysis could be done to be meaningful many months ago. So I designed a solution - on paper. It's one of the packed/sealed boxes*. But I remember having a couple of issues that would need to be solved and I don't know if/what data is available:
- location of patient is given. But a cluster would/could form around other than home location. Like, factory, school... so backtracked ttr data would have to be used too to identify common places. and some % of likeliness of catching it there could be devised
- msoa level is too high. Look at Kew - even that is too big, it would have to be split. This is the easy part, many have done it, incl. the royal mail and other logistics companies.
- biggest issue with reliability would be about self confessed places. Many ppl would not be truthful or simply forget.
What you're talking about is slightly different, but interesting. It would have to rely on past (but no longer than, say 10-14 days) of cluster analysis and combine it with movement data of individuals. At this point you have to have the individual's geo data at hand to cross reference if close to any clusters. Geo data is mostly given freely by people simply because they don't turn off google or apple built into the map tracking. So technically it could be possible.
So it would be something like:
Google maps knows that I go to East Sheen Waitrose 2x a week. If there is a cluster there it would send me notification. Or flip it: wherever I want to go I can check if it's green/orange/red on the map.
but.
without almost real-time ttr it's borderline useless. my issue is that as far as i can tell ppl give generic answers to where you were. I might say waitrose, but not mention local grocery store as I forget. Or I don't go to school, but my kid does and even though I go to waitrose 2x a week, it is more likely that my teenage son brought it home from school. And here is where the likelihood % would come into play as the algorithm would have to be trained to know which is more likely. Then again, if my shopping is in a flimsy mask under my nose** than, yes, waitrose will be the culprit more likely.
The solution I looked into (one of the asian ones) actually used individual movement geo data -so app that records movement and time spent in places- and cross checked with covid patient/testing dbs. If I remember correctly.
I think I abandoned this at this point because in western culture this level of individual tracking - even if used on a not identifiable level- would never fly. Which to me is a really sad thing as I think that technology would have been so useful.
Imo the best solution still would be to track individuals anonymize data use datasets. This requires either trust in the gov that is doing the anonymization, or full transparency of the dataset and the anonymization process. The second is much easier to do. But even me, the cynical, anti-gov person would be okay if the second was public info.
*we are moving tomorrow/wed and our life is a mess atm. And kitten is coming on Friday. We didn't think this through properly... :)
**if some baseline things could be taken for granted - like people always wearing high grade masks properly for the entire duration of being out and about.... this would make life easier. In general.