Abstract
Introduction
CDC and state/local health departments analyze HIV sequences to identify molecular clusters representing rapid transmission. Local analyses can be timelier, but are limited to sequences available in the local surveillance system, and thus may miss all or part of clusters spanning multiple jurisdictions. We assessed the added value of national analyses for clusters missed or only partially detected by local analyses.
Methods
We reviewed national quarterly analyses of HIV sequences (September 2021 – June 2023) that identified priority clusters (defined as ≥5 diagnoses in the past 12 months within a genetic distance of ≤0.005 substitutions per site among people with HIV diagnosed during the most recent three years). For each cluster, we simulated whether and when single-jurisdiction analysis would meet priority threshold and calculated the proportion of cluster members that could be identified locally.
Results
Of 149 clusters detected in the national analysis, 91 (61%) clusters would have been detected concurrently by local analysis, 27 (18%) would have delayed detection (median: 59 days), and 31 (21%) would not have been detected locally. In clusters detected by both approaches, an average of 91% of cluster members identified in the national analysis would have also been identified in the local analysis.
Conclusion
National analyses detected clusters that local analyses missed, identified additional cluster members, and found certain clusters earlier. National cluster detection plays a vital role in enhancing local detection activities.
Keywords: HIV, clusters, molecular epidemiology, national analysis
Introduction
HIV clusters refer to rapid transmission among people in a sexual or drug-using network. HIV cluster detection can identify communities with insufficient access to HIV prevention and care services and inform prioritization of activities to address these gaps, serving as an important tool in the Ending the HIV Epidemic in the US initiative1.
Multiple approaches to cluster detection exist, including time-space analysis, provider notification, and molecular analysis. HIV molecular cluster detection, conducted by the Centers for Disease Control and Prevention (CDC) and local health departments, analyzes HIV sequences to identify groups of closely related sequences that indicate rapid transmission. CDC developed a priority cluster definition to focus on molecular clusters with rapid transmission. Priority clusters are clusters with ≥5 diagnoses in the past 12 months (≥3 diagnoses in low morbidity jurisdictions) that have a genetic distance of ≤0.005 substitutions per site among people with HIV diagnosed during the most recent three years2.
HIV sequences used in molecular cluster detection are generated through routine drug resistance testing and reported to jurisdictions as part of HIV surveillance. These data are stored locally in the Enhanced HIV/AIDS Reporting System (eHARS), along with additional epidemiologic information. Deidentified data are transmitted to CDC for inclusion in the National HIV Surveillance System (NHSS) 3.
Molecular cluster detection analyses are conducted separately using data at national and local levels. CDC conducts quarterly analyses of HIV sequences reported through the NHSS and reports results, including person-level line lists to the jurisdiction with the majority of cluster members. These national datasets can include persons with data reported from multiple jurisdictions, and processing includes sequence deduplication. Limited information is provided when the data are reported by a jurisdiction other than the jurisdiction receiving the line list (e.g., jurisdictions reporting HIV surveillance information, state of residence, whether HIV diagnosis occurred within the past 12 months and when the person was identified as part of the national cluster). States and other separately funded jurisdictions use a CDC-owned application, Secure HIV-TRACE, to run monthly analyses using data in their eHARS3,4. Since local analyses are usually run more frequently and with timelier data than the national analysis, they may detect priority clusters earlier. However, national analyses can detect multijurisdictional clusters. These clusters may be missed or partially detected by local analysis, and some clusters may be detected later in a local analysis after additional cluster members are identified. It is unknown how often multijurisdictional clusters lead to differing results between national and local analyses.
We assessed differences in cluster detection due to the impact of multijurisdictional datasets by recreating priority clusters including and excluding out-of-jurisdiction individuals. We determined cluster size and time of detection under each approach and compared cluster growth and density among clusters not locally detected, detected locally after detection in the national analysis, and detected concurrently in both analyses.
Methods
Data Source and Cluster Detection Analysis
We analyzed preliminary national-level datasets generated each quarter from September 2021–June 2023 to identify new priority clusters. Each cluster was assigned a predominant jurisdiction, defined as the most common jurisdiction of residence at HIV diagnosis among cluster members. For clusters with multiple equally predominant jurisdictions, the jurisdiction with the earliest simulated cluster detection date was treated as the predominant jurisdiction. Simulated detection dates were used to estimate the earliest date jurisdictions could detect clusters as this analysis used national datasets and actual local detection dates were unknown.
Clusters were recreated including all individuals to represent the national analysis (i.e., national clusters), and excluding out-of-jurisdiction individuals to represent the predominant jurisdiction’s analysis (i.e., local clusters). Using methods described previously4, clusters were recreated using the R package igraph5,6. Clusters were reconstructed by serially adding individuals in order of HIV diagnosis date to determine the earliest date a cluster could meet priority criteria in the national analysis. This date was the diagnosis date of the individual who, once added, allowed the cluster to meet priority criteria. Then, we determined if and when the cluster would meet priority criteria in the local analysis. Because priority criteria are based on diagnosis date2, diagnosis date was used in these simulations.
To simulate local detection, we created NHSS subsets that would represent locally available data. Individuals were classified as available in the predominant jurisdiction’s data if they resided in the jurisdiction at the time of diagnosis, resided in the jurisdiction as of June 2023, or had any information in the NHSS reported by the jurisdiction. By identifying the broadest group that could appear in individual jurisdictions’ data, this approach maximizes the likelihood of local cluster detection. We also assumed separately funded cities and counties share data with their respective states. For example, we assumed New York State and New York City would share data, and considered them as a single jurisdiction. We then identified clusters with ≥1 or ≥3 member(s) in a jurisdiction outside the predominant jurisdiction to assess multijurisdictional involvement. For these clusters, we assessed whether inclusion of members outside of the predominant jurisdiction would change the predominant transmission category, age group or race/ethnicity for the cluster.
Clusters were categorized as not locally detected, detected locally after the national analysis, or detected concurrently in both analyses. Concurrently detected clusters would likely be detected earlier by local jurisdictions due to analysis frequency. For clusters detected in the national analysis first, a 60-day delay was added to the national cluster identification date to simulate delays associated with conducting and communicating results from the national analysis. This was done to assess if the cluster would still be detected earlier in the national analysis.
Cluster Growth by Detection Category
Clusters often grow after detection, which can indicate ongoing transmission. To better understand how cluster growth may impact local detection, we classified clusters by detection category and growth. Cluster growth was determined by comparing the size of the cluster when it first met priority criteria to its size as of June 2023; we restricted to clusters with simulated national detection dates before 1/1/2023 to allow for at least six months of follow-up. A Chi square test was conducted comparing growth as a binary outcome between clusters detected concurrently and clusters not detected locally.
Proportion of Cluster Detected by Local Analysis
Among clusters detected by both approaches, we determined the proportion of the national cluster that would be detected by local analysis. To do this, we calculated the percentage of cluster members in each national cluster as of June 2023 that were detected in the local cluster as of the same month.
Cluster Density by Detection Category
Lastly, we determined whether connectivity of sequences within a cluster was associated with detecting the cluster. Cluster density for each national cluster was defined as the number of existing links divided by the total number of possible links in the cluster. An ANOVA using Tukey’s test for post-hoc comparisons was conducted to compare cluster density between cluster detection categories.
Results
Inclusion Criteria and Cluster Detection Categories
Of 149 clusters detected in the national analysis during September 2021–June 2023, 94 (63%) and 40 (27%) had ≥1 member and ≥3 members in a jurisdiction other than the predominant jurisdiction, respectively. Of these 149 clusters, 31 (21%) would not be detected locally due to the inclusion of out-of-jurisdiction individuals in the national analysis. Of 118 clusters detected by both approaches, 91 (77%) would be detected concurrently and 27 (23%) would have delayed local detection. The 27 clusters with delayed local detection had a median delay of 59 days (range: 2–464 days).
When a 60-day delay for the national analysis was applied, 105 (89%) would be detected concurrently and 13 (11%) would have delayed local detection. These 13 clusters had a median delay of 149 days (range: 103–464 days).
Among 94 clusters with ≥1 out-of-jurisdiction member, 9 (10%) had a different predominant transmission category and 22 (23%) had a different predominant age group when out-of-jurisdiction members were included.
Cluster Growth by Detection Category
Among 137 clusters detected before 1/1/2023, 123 (90%) grew after detection in the national analysis. All 27 clusters with delayed local detection grew since delayed local detection implies additional members were needed for the cluster to be detected locally. Of 83 clusters detected concurrently, 77 (93%) grew, and 19 (70%) of 27 clusters not detected by local analysis grew (Figure 1). This difference between the concurrently and never detected clusters was statistically significant (p = 0.0069)
Figure 1.
Cluster detection categories and growth from detection to June 2023. This flowchart shows the number of clusters in each detection category, and of those clusters, how many clusters added new individuals. Information is summarized for all clusters and clusters with simulated national detection dates before 1/1/2023.
Proportion of Cluster Detected by Local Analysis
Local analyses identified all members in 61 (67%) of 91 clusters detected by both analyses. The average local cluster included 91% of members in the corresponding national cluster (range: 48%-100%; Figure 2).
Figure 2.
Distribution of clusters by percent of national cluster members detected in the corresponding local cluster (includes only clusters ever detected locally, N = 118).
Cluster Density by Detection Category
The average cluster density was 64% among clusters detected concurrently, 56% among clusters with delayed local detection, and 51% among clusters not detected locally. The overall ANOVA test was statistically significant (p = 0.015). The only significant group difference was between clusters not detected locally and concurrently detected clusters (p = 0.018).
Discussion
National HIV cluster analysis supports timely detection of priority clusters by identifying clusters and cluster members undetected by local analyses. Approximately 1 in 5 clusters detected in the national analysis would be missed by local analyses. Additionally, 1 in 10 clusters detected locally would have been detected earlier by the national analysis even after accounting for a 60-day delay with the national analysis.
These findings, together with high transmission rates in the follow-up period after cluster detection,7,8 underscore the importance of timely response efforts. Once jurisdictions are alerted to multijurisdictional clusters by CDC, they should promptly begin response efforts with the jurisdiction with the most cases in the cluster generally taking the lead. Secure data sharing consistent with local and CDC data protections can support response to multijurisdictional clusters9.
Clusters detected by both approaches had increased connectivity and were more likely to grow, compared to clusters undetected locally. Denser clusters may be easier to detect as each sequence is more likely to be within the genetic distance threshold of multiple other sequences, and excluding any individual sequence is less likely to change a cluster’s priority classification (Supplementary Figure). Denser clusters could also be due to rapid transmission9,10 or higher sequence completeness in a jurisdiction. Possible reasons for why clusters that were never detected locally were less likely to grow than clusters detected concurrently include: fewer testing or other response activities resulting in reduced likelihood of diagnosis and sequence sampling, and varying sequence completeness across the involved jurisdictions impacting the ability to detect cluster growth.
Continued cluster growth could indicate a larger underlying network not receiving adequate HIV care or prevention services9. Detection of additional cluster members via national analyses changed the predominant transmission category or age group in 10 – 23 % of clusters, suggesting the national analysis can identify important contextual information to further refine response activities.
Our analysis has some limitations. This analysis assumes all jurisdictions conduct monthly analyses and a national analysis will be completed in 60 days. More frequent local analyses would not identify out-of-jurisdiction individuals but could result in earlier identification of in-jurisdiction cluster growth. We anticipate variations in analysis frequency would minimally impact our aggregate findings. In this analysis, sequences were assumed available on HIV diagnosis dates. Although there are delays in sequence reporting with a median of 34 days from sample collection to receipt of the sequence at the jurisdiction3, these delays would equally impact national and local cluster detection. All clusters were followed through June 2023, and clusters detected closer to June 2023 had less time to grow. Despite these limitations, our findings provide important considerations for how jurisdictions and CDC can use the results of national and local analyses.
The national analysis plays a critical role in cluster detection by identifying 30% of clusters that would be missed or detected substantially later by local analysis. Jurisdictions should continue to regularly conduct local analyses and review clusters detected by the national analysis as both are important for timely detection and more complete understanding of clusters.
Supplementary Material
Supplementary Figure. Fictitious example of cluster density impacting cluster detection. Cluster A is a highly dense six-person priority cluster in Georgia that still meets priority criteria in Georgia’s analysis when the person outside of Georgia is excluded. Cluster B is a less dense six-person priority cluster in Georgia that does not meet priority criteria in Georgia’s analysis when the person outside of Georgia is excluded. Priority clusters: Clusters with ≥5 diagnoses in the past 12 months (≥3 diagnoses in low morbidity jurisdictions) that have a genetic distance of ≤0.005 substitutions per site among people with HIV diagnosed during the most recent three years. All persons in this example have HIV diagnosed in the most recent 12 months.
Footnotes
Disclaimer: The findings and conclusions in this report are those of the authors and do not necessarily represent the official position of the Centers for Disease Control and Prevention
Ethics statement: This activity was reviewed by CDC and was conducted consistent with applicable federal law and CDC policy.
References
- [1].Oster AM, Lyss SB, McClung RP, et al. HIV Cluster and Outbreak Detection and Response: The Science and Experience. Am J Prev Med. 2021;61(5 Suppl 1):S130–S142. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [2].Oster AM, France AM, Panneer N, et al. Identifying Clusters of Recent and Rapid HIV Transmission Through Analysis of Molecular Surveillance Data. J Acquir Immune Defic Syndr. 2018;79(5):543–550. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [3].France AM, Hallmark C, Panneer N et al. Nationwide Implementation of HIV Molecular Cluster Detection by Centers for Disease Control and Prevention and State and Local Health Departments, United States. Emerging Infectious Diseases. 2025;31(13):80–88. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [4].Kosakovsky Pond SL, Weaver S, Leigh Brown AJ et al. HIV-TRACE (TRAnsmission Cluster Engine): a Tool for Large Scale Molecular Epidemiology of HIV-1 and Other Rapidly Evolving Pathogens. Mol Biol Evol. 2018. Jul 1;35(7):1812–1819. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [5].Csardi G, Nepusz T The igraph software package for complex network research. InterJournal 1695. 2006. (5) 1–9. [Google Scholar]
- [6].Csárdi G, Nepusz T, Traag V, Horvát S et al. igraph: network analysis and visualization in R. 2024. R package version 1.6.0.9021. [Google Scholar]
- [7].Billock RM, France AM, Saduvala N, et al. Contribution of HIV transmission bursts to future HIV infections. AIDS. 2025;39(10):1403–1412. doi: 10.1097/QAD.0000000000004101 [DOI] [PMC free article] [PubMed] [Google Scholar]
- [8].Panneer N, France A., Schlanger K et al. Clusters contribute disproportionately to future HIV transmission in the United States, in 25th International AIDS Conference. 2024: Munich, Germany; [DOI] [PubMed] [Google Scholar]
- [9].CDC Cluster Detection and Response Guidance for Health Departments https://www.cdc.gov/hivpartners/php/cdr/health-department-guidance.html. November 24, 2025
- [10].Campbell EM, Jia H, Shankar A et al. (2017). Detailed Transmission Network Analysis of a Large Opiate-Driven Outbreak of HIV Infection in the United States. The Journal of infectious diseases, 216(9), 1053–1062. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [11].Wertheim JO, Leigh Brown AJ, Hepler NL et al. (2014). The global transmission network of HIV-1. The Journal of infectious diseases, 209(2), 304–313. [DOI] [PMC free article] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Supplementary Materials
Supplementary Figure. Fictitious example of cluster density impacting cluster detection. Cluster A is a highly dense six-person priority cluster in Georgia that still meets priority criteria in Georgia’s analysis when the person outside of Georgia is excluded. Cluster B is a less dense six-person priority cluster in Georgia that does not meet priority criteria in Georgia’s analysis when the person outside of Georgia is excluded. Priority clusters: Clusters with ≥5 diagnoses in the past 12 months (≥3 diagnoses in low morbidity jurisdictions) that have a genetic distance of ≤0.005 substitutions per site among people with HIV diagnosed during the most recent three years. All persons in this example have HIV diagnosed in the most recent 12 months.