HIV transmission cluster identification shows promise as a tool to help prioritize public health intervention. We assessed cluster-level characteristics associated with temporal cluster growth and incorporated these findings into a predictive model for cluster growth in North Carolina (NC).
HIV-1 pol sequences generated from routine genotypic resistance testing from 11/2010 through 09/2016 in NC (n=8923 persons) were matched to HIV surveillance data and used to identify putative transmission clusters size ≥2 members via pairwise genetic distance differences <1.5%. Of 782 clusters, 275 (35%) were established by 03/2015 and recently active (included any sequences from prior 2 years). Cluster members were categorized as baseline members (sequences prior to 03/2015), hidden members (diagnosed prior to 03/2015 with sequences after 03/2015), and new members (diagnosed after 03/2015) [Figure]. Clusters were retrospectively assessed for growth (any new members) over 18 months (03/2015 – 09/2016). We developed a predictive model for cluster growth incorporating demographic, clinical, and contact tracing characteristics of baseline members and evaluated the model using the area under the receiver operating characteristic curve (ROC AUC).
Of 275 established, recently active clusters (n=1625 persons; size=2-44 persons), 64 (23%) grew over 18 months of follow-up. Growing clusters had a larger median size at baseline than non-growing clusters (6 vs. 2 persons). Persons in growing clusters showed younger median ages (33 vs. 38 years) and were more likely to report male sex (89% vs. 77%), black race (77% vs. 41%), and MSM status (71% vs. 51%) than those in non-growing clusters. Persons in growing clusters had shorter median times to care entry after diagnosis than those in non-growing clusters (45 vs. 71 days) but were less likely to have been in care at the start of follow-up for cluster growth (81% vs. 84%). The final predictive model included terms for cluster size, median time to care entry after diagnosis, median age, and percent with no identified contacts among baseline members, and showed an ROC AUC of 0.85 in the validation sample.
This model has strong predictive ability to forecast new HIV diagnoses in recently active, established genetic clusters in NC and could be adapted to diverse HIV control settings. Identification of HIV transmission clusters that are likely to grow over time could guide prioritization of public health interventions.