Impacts on community transmission and disease burden of a clinical prediction tool to prioritize limited COVID testing

Community spread of coronavirus disease 2019 (COVID-19) continues to be high in many areas, likely due, in part, to insufficient testing and contact tracing. As regional test kit shortages are likely to continue with increased transmission, it is important that available testing capacity be used effectively. To date, testing for COVID-19 has largely been restricted to persons reporting symptoms, with no additional criteria being systematically employed to select who is tested. In situations when testing capacity is limited, we propose the use of a clinical prediction rule to allow for prioritized testing of people who are most likely to test positive for COVID-19. Using data from the University of Utah Health system, we developed a robust, deployable clinical prediction rule which incorporates data on demographics and clinical characteristics to predict which patients are most likely to test positive. We then incorporated prioritized testing into a stochastic SEIR model for COVID-19 to measure changes in disease burden compared to a model with indiscriminate testing. Our best performing clinical prediction rule achieved an AUC of 0.7. When incorporated into the SEIR model, prioritized testing resulted in a delay in the timing of the infection peak, a meaningful reduction in both the total number of infected individuals and the peak height of the infection curve, and thus a reduction in the excess demand on local hospital resources. These effects were strongest for lower values of Rt and higher proportions of infected individuals seeking testing.