Predicting Whom to Test is More Important Than More Tests - Modeling the Impact of Testing on the Spread of COVID-19 Virus By True Positive Rate Estimation

Paul F Rodriguez

It has become painfully obvious that the number of COVID-19 cases in New York will still be increasing for some time. A critical strategy is to reduce the reproduction number by reducing contact and movement because the growth factor has an exponential effect. However, at this point in time it may not by itself slow the spread enough unless restrictions become draconian and long lasting. More testing and quarantining will help mitigate the rise, but if testing is going to copy the South Korean numbers, the testing in New York must increase faster than the virus spread, perhaps to get up to 15 to 1 ratio of new cases (as of March 28 th , that would be about ~75000 tests per day). But just as important is to increase efficacy by collecting good data to better score and allocate tests. A good scoring model and allocation strategy, that increases TP rate by a one or two percentage points could have the same impact as increasing the number of tests by tens of thousands. The scoring model perhaps could include contacts and locations, along with demographics, weather, type of places visited, time of day, and anything else that could help build a robust model on top of virus spread models and tracking contacts. In some high profile machine learning competitions, intensive application and modeling efforts have shown improvements of 10 percentage points in accuracy (e.g. Netflix Prize, Wikipedia, Zillow Landing Pages). There should be low technical or material barriers to gathering and using data, and I would argue that health agencies should start doing so and making the anonymized data available. . CC-BY-NC-ND 4.0 International license It is made available under a author/funder, who has granted medRxiv a license to display the preprint in perpetuity.