National transmitted HIV-1 drug resistance (TDR) rates are typically monitored through HIV surveillance data and this represents the gold-standard for determining TDR rates in various populations. However, it often requires several years to prepare data for presentation and publication resulting in long lag times between data accumulation and result dissemination. Here we present a computational method using de-identified clinical test results from a large commercial database to infer TDR rates in near real-time.
A HIPAA-compliant unique Patient ID was assigned to each test result in the LabCorp Clinical Database. New cases of HIV diagnosis were inferred using the linked results from HIV diagnosis, viral load and drug resistance tests. We developed a novel inference model which employs a successive series of filters to derive a subset of resistance assays that were likely to have been performed at the time of initial HIV diagnosis. We used the 2009 surveillance drug resistance mutations (SDRM) list from the Stanford University HIV Drug Resistance Database to determine mutational frequencies. We validated the model by comparing the predicted 2008-2011 NRTI, NNRTI, and PI frequencies to ranges established from prior TDR surveillance studies. We then used our model to calculate rates of TDR for 2012-2014 per drug class, including INIs. This model allows us to assess prevalence rates according to age group, gender, and geography.
We identified 8612 putative baseline resistance tests from 2008-2011. The rates for NRTI (7.6%), NNRTI (11.2%), and PI (2.9%) SDRMs compared favorably with previously reported TDR rates, as did the overall rates for any SDRM (17%). For 2012-2014, we analyzed 14,325 putative baseline resistance tests and observed an overall rate for TDR mutations of 17.2%. We observed prevalence rates of 6.3%, 12.3%, 2.7%, and 0.6% for the NRTIs, NNRTIs, PIs, and INIs, respectively.
Our model demonstrates a novel real-time method for inferring the prevalence of TDR by drug class. This model suggests that national TDR rates for NRTI, NNRTI, and PIs from 2012-2014 have not decreased relative to prior years and that INI TDR rates were surprisingly low, possibly reflective of the efficacy of current antiretroviral regimens which incorporate this class. Near real-time TDR data can facilitate early detection of new drug resistance strains should they arise, such as in response to PrEP. Analysis of 2015 rates will be included for presentation.