Improving local prevalence estimates of SARS-CoV-2 infections using a causal debiasing framework