Prediction of HIV Transmission Cluster Growth With Statewide Surveillance Data

Abstract

Background: Prediction of HIV transmission cluster growth may help guide public health action. We developed a predictive model for cluster growth in North Carolina (NC) using routine HIV surveillance data. Methods:We identified putative transmission clusters with ≥2 members via pairwise genetic distances ≤1.5% from HIV-1 pol sequences sampled 11/2010-12/2017 in NC. Clusters established by a baseline of 01/2015 with any sequences sampled within two years before baseline were assessed for growth (new diagnoses) over 18 months. We developed a predictive model for cluster growth incorporating demographic, clinical, temporal, and contact tracing characteristics of baseline cluster members. We internally and temporally externally validated the final model in the periods 01/2015-06/2016 and 07/2016-12/2017. Results:Cluster growth was predicted by larger baseline cluster size, shorter time between diagnosis and HIV care entry, younger age, shorter time since the most recent HIV diagnosis, higher proportion with no named contacts, and higher proportion with HIV viremia. The model showed areas under the receiver operating characteristic curves of 0.82 and 0.83 in the internal and temporal external validation samples. Conclusion:The predictive model developed and validated here is a novel means of identifying HIV transmission clusters that may benefit from targeted HIV control resources.

Characterization of Molecular Cluster Detection and Evaluation of Cluster Investigation Criteria Using Machine Learning Methods and Statewide Surveillance Data in Washington State

“…This demonstrates the limitations in predictive power of HIV core surveillance and molecular cluster detection in isolation. Previous successful prediction efforts have used partner services data to supplement these information sources, but some jurisdictions may have limited access to such data through structural or legal barriers [12]. In addition to partner services data, information about cluster topology and data from prior cluster and data for care investigations could be useful in prediction of cluster growth.…”

“…For cluster detection and response to be most effective it should incorporate robust core surveillance including an ongoing understanding of changes in HIV transmission trends and complete case and lab reporting, robust partner services activities with disease investigators that are familiar with the profile of their local cases, and efficient and complete molecular data collection and analysis [10,12,23]. There should also be transparency and dialogue with the community about the activities being undertaken and why they are important.…”

“…It would also include only data that are readily accessible to surveillance programs to minimize the burden of prediction. Previous efforts to predict cluster growth have been successful in a number of jurisdictions, though some have demonstrated that predictors of cluster growth change in performance over time, which indicates that static models may not be suitable tools for describing a dynamic transmission environment [11,12]. Random forest modeling is a supervised learning algorithm that requires minimal input and has demonstrated considerable predictive success in other public health settings.…”

HIV-1 Transmission linkages among persons with incident infection to inform public health surveillance

“…Persons ≥ 13 years old with HIV-infection residing in NC with at least one HIV pol sequence reported to the NC Division of Public Health (NC DPH) were evaluated. HIV partial pol sequences are generated by reference laboratories through clinical genotypic resistance testing and are routinely reported to NC DPH, with samples available since 2010 [10] . We assessed persons with recent versus prior HIV diagnoses; recent diagnose were persons with diagnosis dates recorded between January 1, 2014 to December 31, 2018 and prior diagnoses categorized based on diagnosis dates <2014 ( Fig.…”

Natural selection favoring more transmissible HIV detected in United States molecular transmission network

“…Each state has primary authority for determining whether their laws and regulations permit data submission to GenBank or other open databases. State and local health departments also have ability to determine whether and when these data can be shared with other researchers, as has occurred for previous studies on HIV surveillance data 41,52,79–82 .…”