Abstract
Background
Public health action combating human immunodeficiency virus (HIV) includes facilitating navigation through the HIV continuum of care: timely diagnosis followed by linkage to care and initiation of antiretroviral therapy to suppress viral replication. Molecular epidemiology can identify rapidly growing HIV genetic transmission clusters. How progression through the care continuum relates to transmission clusters has not been previously characterized.
Methods
We performed a retrospective study on HIV surveillance data from 5226 adult cases in Los Angeles County diagnosed from 2010 through 2014. Genetic transmission clusters were constructed using HIV-TRACE. Cox proportional hazard models were used to estimate the impact of transmission cluster growth on the time intervals between care continuum events. Gamma frailty models incorporated the effect of heterogeneity associated with genetic transmission clusters.
Results
In contrast to our expectations, there were no differences in time to the care continuum events among individuals in clusters with different growth dynamics. However, upon achieving viral suppression, individuals in high growth clusters were slower to experience viral rebound (hazard ratio 0.83, P = .011) compared with individuals in low growth clusters. Heterogeneity associated with cluster membership in the timing to each event in the care continuum was highly significant (P < .001), with and without adjustment for transmission risk and demographics.
Conclusions
Individuals within the same transmission cluster have more similar trajectories through the HIV care continuum than those across transmission clusters. These findings suggest molecular epidemiology can assist public health officials in identifying clusters of individuals who may benefit from assistance in navigating the HIV care continuum.
Keywords: HIV, care continuum, molecular epidemiology, transmission network, cluster, Cox proportional hazard, gamma frailty
Members of the same HIV genetic transmission clusters progress at similar rates to events in the HIV care continuum. Individuals in rapidly growing clusters at the time of HIV diagnosis took longer to experience viral rebound.
Key to successful control of human immunodeficiency virus (HIV) is ensuring that individuals newly diagnosed with HIV are linked to care, achieve viral suppression via antiretroviral therapy (ART), and remain in care to prevent recurrence of viremia. Successful navigation through this HIV continuum of care [1, 2] ensures that people living with HIV have greatly reduced risk of transmitting the virus [3, 4]. Rapid linkage to care and adherence to ART improves individual long-term health prognosis [5], and achieving viral suppression virtually eliminates risk of onward viral transmission [6, 7]. These tenets guide HIV public health activities worldwide [5, 8–10].
Molecular epidemiology can identify and characterize genetic clusters of HIV transmission [11, 12], which can assist in prioritizing cases for public health services (e.g., linkage or re-linkage to care to help individuals achieve sustained suppression of viral replication) [13–15]. These clusters are comprised of individuals whose HIV is sufficiently genetically similar as to imply direct or indirect epidemiological connections among their members [16, 17]. The frequency of clustering is most strongly associated with reported transmission risk and demography [17–20]; however, other factors such as sampling density and time between infection and diagnosis can substantially influence clustering propensity [21–23]. Nonetheless, in the US surveillance setting, HIV genetic clusters with evidence of recent transmission are disproportionately likely to give rise to subsequent cases of HIV [18, 19] and can have transmission rates 8 times the national average [24]. Furthermore, people in these growing transmission clusters with unsuppressed viral replication are associated with future incident HIV infections [25]. We hypothesize that individuals most likely to be associated with future cluster growth (i.e., those in rapidly growing clusters at the time of HIV diagnosis) will take longer to progress through the HIV care continuum.
Here we incorporate molecular HIV surveillance data from Los Angeles County (LAC) in analyses of predictors of time to events in the care continuum employing both semiparametric and parametric frailty survival models.
METHODS
Data Source
The Division of HIV and STD Programs (DHSP) in LAC Department of Public Health (DPH) has been conducting mandatory named-based HIV surveillance since 2006, which includes reporting results of viral load (VL) and CD4+ T-cell count laboratory tests, as well as human immunodeficiency virus type 1 protease and reverse transcriptase (pol) genetic sequences generated during routine antiretroviral drug resistance testing [26]. All reported HIV cases in LAC are stored in the Enhanced HIV/AIDS Reporting System (eHARS); approximately 10–14% of people living with HIV in LAC are undiagnosed [27]. Additional case-level metadata in eHARS include date, age, and ZIP code at diagnosis, current ZIP code, vital status, race/ethnicity, sex at birth, and transmission risk factor. HIV transmission risk categories were assigned in a hierarchical order: people who inject drugs (PWID), men who have sex with men (MSM), heterosexual contact, and unidentified risk. MSM who reported injection drug use were grouped with PWID. Cases with no reported risk were classified as “unknown risk” [28]. Stage of infection at HIV diagnosis was defined using the first CD4+ count postdiagnosis: ≥500 cells/mL, 200–499 cells/mL, and < 200 cells/mL. Socioeconomic status was assessed at the level of residential ZIP code at diagnosis: the fraction of the population living below the 100% federal poverty level [29].
Inclusion Criteria
We performed a population-based retrospective study using HIV/AIDS surveillance data from LAC DPH on 22 398 individuals who resided or received HIV care in LAC and had a reported HIV pol sequence. We restricted our statistical analysis to the 5226 cases diagnosed from January 2010 through December 2014, representing 75% of all cases diagnosed during these years (Supplementary Table 1). This time-period was chosen to minimize gaps or delays in reporting prior to 2010 and to ensure at least 2 years of follow-up laboratory data. It also predates the era of cluster-based investigations in LAC. Included cases also met the following criteria: (i) a reported HIV pol genotype; (ii) diagnosed in and current resident of LAC; (iii) ≥ 13 years of age at HIV diagnosis; (iv) detectable first VL (≥200 copies/mL); and (v) complete month and year reported for time of diagnosis and VL/CD4+ lab results. Our final data set was enriched for older individuals, Latinos, MSM, and individuals with later disease stage at diagnosis (Supplementary Table 1).
Transmission Network
A molecular transmission network was constructed using HIV-TRACE [30] from 22 398 individuals whose HIV pol sequences were reported to LAC DPH between 2006 through 2016; characteristics of this network were previously described [31]. To build the network, cases were linked to each other in the transmission network if their pairwise genetic distance was ≤ 0.015 substitutions/site. This threshold is in accordance with the expected divergence between sequences within an individual [32] and the genetic distance inferred between named HIV risk partners [33]. Linked cases were combined into clusters, wherein all members of a cluster are linked to at least 1 other member of that cluster.
Cluster Growth
For each individual, on the date of their diagnosis, we determined the size of their cluster (i.e., only including members linked through cases diagnosed at or prior to that date). Cluster growth was measured by calculating the number of newly diagnosed cases linked in a cluster in the previous year divided by the square root of the cluster size. People with a reported HIV sequence but unclustered in the network had a cluster growth value of 0. This cluster square root growth metric, referred to as cluster growth for the remainder of this article, is a reliable predictor of future cluster growth in a US HIV surveillance setting [18]. Cluster growth at diagnosis was categorized as high (≥75th percentile), medium (>50th to < 75th percentile), and low (≤50th percentile).
Outcome Measures
We assessed time to each event in the HIV care continuum and potential failure in treatment adherence at the individual level using the date of laboratory test as a proxy for medical office visit for the following endpoints: (i) months from diagnosis to receipt of care (i.e., first VL or CD4+ after diagnosis); (ii) months from care to viral suppression (i.e., first VL < 200 copies/mL); (iii) months from diagnosis to viral suppression; and (iv) months from viral suppression to viral rebound (i.e., first postsuppression VL ≥ 1500 copies/mL). Inclusion in statistical analysis required having a reported HIV sequence and at least 1 follow-up VL/CD4+ measurement after reaching a previous continuum event.
Statistical Analyses
Ordinal logistic regression was used to compare cases diagnosed in high growth clusters with cases in medium and low growth clusters. Parametric and semiparametric models were constructed to adjust for covariates in STATA 15.1, College Station, TX: StataCorp LP.
Semiparametric Modeling
Separate Cox proportional hazard models [34] with robust standard errors were built to analyze times between events in the care continuum. All cases were followed until death or the administrative censoring date of December 2016. To investigate the impact of delayed reporting or interruptions in engagement in care, we performed sensitivity analyses that restricted consideration to only individuals who had at least one follow-up visit within 1 year after initial linkage to care. Viral suppression can occur any time between last VL ≥ 200 copies/mL and first VL < 200 copies/mL, and viral rebound can occur anytime between last VL < 200 copies/mL and first subsequent VL ≥ 1500 copies/mL. Therefore, we repeated the restricted analysis using methods that accommodate interval censoring [35] to account for our inability to observe exact event dates of achieving viral suppression or viral rebound; only the interval in which the event occurred is observed. To evaluate whether the time interval relationship between cluster growth and time to viral suppression was modified by time to linkage to care, we tested for interaction between cluster growth and linkage to care within 2 months after diagnosis. Model fit for Cox models were assessed using Schoenfeld and Cox-Snell residuals.
Parametric Shared Frailty Model
To identify the appropriate survival model and account for heterogeneity between transmission clusters, we conducted additional parametric analyses using exponential, Weibull, log-logistic, and log-normal baseline hazard functions in a shared gamma frailty [36, 37] framework. The log-normal distribution provided the best fit according to Akaike’s information criterion (AIC) (Supplementary Table 2). Sensitivity analysis of the frailty variance to exclusion of covariates and nonclustered individuals were also conducted.
RESULTS
Individuals included in our analysis of the care continuum were predominantly male (91%), Hispanic/Latino (51%), and MSM (71%) (Table 1). The adjusted odds of being in a high growth cluster (≥75th percentile) at date of diagnosis, compared with being in a medium or low growth cluster, were higher for males and cases with CD4+ ≥500 cells/mL at diagnosis (Table 1). Adjusted odds of being in a high growth cluster was lower for individuals ≥ 30 years old at diagnosis, blacks compared with whites, and cases with unidentified transmission risk compared with MSM.
Table 1.
Characteristics of Cases in the Los Angeles County HIV Transmission Network Diagnosed in High, Medium, and Low Growth Clusters Using Multivariable Ordinal Logistic Regressiona
| Attributes | Total n (%)2 | Low Cluster Growth n (%)b | Medium Cluster Growth n (%) b | High Cluster Growth n (%) b | AORc | 95% CI |
|---|---|---|---|---|---|---|
| Total population | 5226 (100) | 2621 (100) | 1301 (100) | 1304 (100) | – | – |
| Birth sex | ||||||
| Male | 4749 (91) | 2317 (88) | 1203 (92) | 1229 (94) | 1.30 | 1.02–1.67 |
| Female | 477 (9) | 304 (12) | 98 (8) | 75 (6) | Ref. | – |
| Age at diagnosis | ||||||
| 13–19 | 177 (3) | 59 (2) | 58 (4) | 60 (5) | 1.25 | .94–1.66 |
| 20–29 | 1813 (35) | 730 (28) | 482 (37) | 501 (46) | Ref. | – |
| 30–39 | 1493 (29) | 772 (29) | 370 (28) | 351 (27) | 0.65 | .57–.74 |
| 40–49 | 1100 (21) | 645 (25) | 266 (21) | 189 (15) | 0.52 | .45–.61 |
| 50–59 | 521 (10) | 329 (13) | 102 (8) | 88 (7) | 0.49 | .40–.60 |
| 60+ | 122 (2) | 86 (3) | 21 (2) | 15 (1) | 0.35 | .24–.53 |
| Race/ethnicity | ||||||
| White | 1124(21) | 557 (21) | 288 (22) | 279 (21) | Ref. | – |
| Black/African American | 1040 (20) | 574 (22) | 237 (18) | 229 (18) | 0.79 | .67–.95 |
| Hispanic/Latino | 2655 (51) | 1264 (48) | 695 (53) | 696 (53) | 1.07 | .92–1.23 |
| Asian/Pacific Islander | 240 (5) | 134 (5) | 46 (4) | 60 (5) | 0.80 | .61–1.06 |
| Otherd | 167 (3) | 92 (4) | 35 (3) | 40 (3) | 0.75 | .55–1.04 |
| Poverty level at ZIP code of diagnosise | ||||||
| 25th percentile | 487 (9) | 249 (10) | 115 (9) | 123 (9) | Ref. | – |
| 50th percentile | 989 (19) | 474 (18) | 246 (19) | 269 (21) | 1.00 | .81–1.24 |
| 75th percentile | 1332 (26) | 711 (27) | 316 (24) | 305 (23) | 0.83 | .68–1.02 |
| >75th percentile | 2418 (46) | 1187 (45) | 624 (48) | 607 (47) | 0.97 | .80–1.18 |
| CD4+ count at diagnosis | ||||||
| ≥500 cells/mm3 | 1561 (30) | 681 (26) | 438 (34) | 442 (34) | 1.88 | 1.62–2.18 |
| 200–499 cells/mm3 | 2247 (43) | 1030 (39) | 587 (45) | 630 (48) | 1.80 | 1.57–2.06 |
| <200 cells/mm3 | 1412 (27) | 908 (35) | 274 (21) | 230 (18) | Ref. | – |
| Missing | 6 (<1) | 2 (<1) | 2 (<1) | 2 (<1) | 2.36 | .55–10.07 |
| Transmission risk | ||||||
| MSM | 3729 (71) | 1741 (66) | 979 (75) | 1009 (77) | Ref. | – |
| PWIDf | 224 (4) | 113 (4) | 59 (5) | 52 (4) | 0.96 | .74–1.25 |
| Heterosexual sexual contact | 155 (3) | 95 (4) | 34 (3) | 26 (2) | 0.86 | .59–1.28 |
| Unidentified | 1118 (22) | 672 (26) | 229 (17) | 217 (17) | 0.77 | .66–.89 |
Abbreviations: AOR, adjusted odds ratio; CI, confidence interval; HIV, human immunodeficiency virus; MSM, men who have sex with men; PWID, people who inject drugs.
aLikelihood ratio test of proportional odds assumption was performed to assess whether the associated log odds is equal to that comparing high and medium growth clusters to low growth clusters (P = .187).
bColumn percentages.
cAdjusted odds ratio comparing cases in high growth clusters at diagnosis compared with cases in low and medium growth clusters at diagnosis.
dIncludes American Indian/Alaskan Native, multiracial, or unknown race/ethnicity.
eFraction of population in ZIP code of diagnosis living below 100% of the federal poverty level.
fPWID includes MSM who reported injection drug use.
We investigated the time between events in the HIV care continuum: (i) diagnosis to linkage to care, (ii) linkage to care to viral suppression, and (iii) viral suppression to viral rebound. Of cases linked to care, 92% achieved viral suppression and 26% experienced viral rebound postsuppression (Table 2); Median time from diagnosis to care was < 1 month, care to suppression was 5 months, and diagnosis to suppression was 6 months. The majority of individuals achieved suppression within 2 visits after receiving care (Supplementary Figure 1). The greater number of consecutive visits prior to viral suppression is associated with a longer time to suppression (Supplementary Figure 2), supporting the belief that longer observed times to suppression are not driven primarily by the visit schedule.
Table 2.
Proportion of Individuals That Progressed to Each Event in the Care Continuum and the Median Survival Time to Each Event
| Continuum Interval | All Cases | Restricted to Cases With Follow-up Visit Within 1 year After Initial Linkage to Care | ||||||
|---|---|---|---|---|---|---|---|---|
| Total Cases at Starta | Total Cases at Endb | % Reached Outcome | Median Months to Outcome (IQR)c | Total Cases at Starta | Total Cases at Endb | % Reached Outcome | Median Months to Outcome (IQR)c | |
| Diagnosis to care | 5,226 | 5,225 | 100% | 0 (0–1) | 4,944 | 4,944 | 100% | 0 (0–1) |
| Care to suppression | 5,204 | 4,791 | 92% | 5 (3–10) | 4,944 | 4,604 | 93% | 5 (3–9) |
| Diagnosis to suppression | 5,226 | 4,791 | 92% | 6 (4–13) | 4,944 | 4,604 | 93% | 6 (4–12) |
| Suppression to rebound | 4,646 | 1,223 | 26% | 15 (7–28) | 4,482 | 1,176 | 26% | 15 (7–28) |
Abbreviation: IQR, interquartile range.
aCases with at least 1 viral load after the beginning of each interval are eligible.
bCases not meeting the end point are censored.
cMedian survival times.
Cluster Growth and Time to Care Continuum Events
Investigation of the effect of cluster growth at HIV diagnosis on time to events in the care continuum revealed no significant differences in the time to each event between cases in medium and low growth clusters at time of diagnosis (Cox proportional hazards; Figure 1 and Table 3). However, in contrast to our initial hypothesis, cases in high growth clusters at diagnosis were more rapidly linked to care (adjusted hazard ratio [HR] 1.06; P = .028) and achieved suppression more rapidly after diagnosis (HR 1.08; P = .034), compared with cases in low growth clusters. Upon achieving viral suppression, cases in high growth clusters were slower to rebound (HR 0.83; P = .011), compared with cases in low growth clusters.
Figure 1.
Flow diagram of the human immunodefiency virus care continuum based on cluster growth at diagnosis. Values indiciate adjusted hazard ratios, with low cluster growth at diagnosis serving as the reference category. Arrows show path between care continuum events (i.e., diagnosis to care). Thick arrows denote significantly increased rate of progression between care continuum events; thin arrows denote significantly slower rate of progression between care continuum events. Asterisks indicate significance at P < .05 in the Cox proportional hazards model.
Table 3.
Adjusted Hazard Ratios for Months to Events in the Continuum of Care by Cluster Growth
| Cox Proportional Hazard | |||||||
|---|---|---|---|---|---|---|---|
| All Cases | Restricted to Cases With Follow-up Visit Within 1 year After Initial Linkage to Care | ||||||
| Continuum Interval | Cluster Growth | HRa | 95% CI | P | HRa | 95% CI | P |
| Diagnosis to care | Low | Ref. | – | – | Ref. | – | – |
| Medium | 1.04 | .99–1.10 | .084 | 1.05 | .99–1.11 | .066 | |
| High | 1.06 | 1.01–1.12 | .028 | 1.07 | 1.01–1.12 | .024 | |
| Care to suppression | Low | Ref. | – | – | Ref. | – | – |
| Medium | 0.98 | .91–1.05 | .489 | 0.99 | .93–1.07 | .934 | |
| High | 1.03 | .96–1.11 | .357 | 0.98 | .92–1.06 | .661 | |
| Diagnosis to suppression | Low | Ref. | – | – | Ref. | – | – |
| Medium | 1.01 | .94–1.08 | .749 | 1.04 | .97–1.11 | .318 | |
| High | 1.08 | 1.01–1.16 | .034 | 1.04 | .97–1.11 | .302 | |
| Suppression to rebound | Low | Ref. | – | – | Ref. | – | – |
| Medium | 0.98 | .85–1.13 | .806 | 0.99 | .86–1.13 | .837 | |
| High | 0.83 | .72–.96 | .011 | 0.84 | .73–.97 | .018 |
Abbreviations: CI, confidence interval; HR, hazard ratio.
aAdjusted hazard ratio adjusted for age at diagnosis, race/ethnicity, poverty level at ZIP code of diagnosis, transmission risk category, birth sex, and CD4+ count at diagnosis.
The time from diagnosis to suppression was modified by an indicator of linkage to care within 2 months of diagnosis (Cox proportional hazards model; P < .05). Among the 988 cases (19%) not linked to care within 2 months, cases in the medium (HR 1.26, P = .005) and high (HR 1.29, P = .003) growth clusters achieved suppression faster than cases in low growth clusters (Figure 2). However, there was no significant difference in the time to viral suppression between individuals who were linked to care within 2 months of diagnosis. Therefore, the faster average time from diagnosis to suppression for cases in high growth clusters is likely attributable to the small fraction of cases not linked to care within 2 months of diagnosis (Figure 1).
Figure 2.
Kaplan-Meier survival estimates and adjusted HRs of time from diagnosis to viral suppression by linked to care in 2 months. Median months from diagnosis to suppression is 5 (IQR 3–11), 6 (IQR 4–14), and 6 (IQR 4–12) if linked to care in 2 months for low, medium and high growing clusters, respectively. Median months from diagnosis to suppression is 19 (IQR 9–36), 16 (IQR 8–28), and 15 (IQR 8–29) if linked to care >2 months after diagnosis for low, medium and high growing clusters, respectively. There was no significant difference in hazards for cases linked to care within 2 months. For cases linked to care in >2 months after diagnosis, medium and high growing clusters reached suppression 26–29% faster (P < .01) compared with low growing clusters, after adjusting for age at diagnosis, race/ethnicity, poverty level at ZIP code diagnosis, transmission risk category, birth sex, and CD4+ at diagnosis. Abbreviations: HIV, human immunodeficiency virus; HR, hazard ratio; IQR, interquartile range.
To assess sensitivity of results to analytical method, we compared results from a Cox regression with those from a multivariable logistic regression of the odds of being linked to care in 2, 6, and 12 months. Results of these models were consistent; among cases not linked to care within 2 months of diagnosis, those in high growth clusters at diagnosis had increased odds of being linked to care in 6 (P = .007) and 12 (P = .004) months compared with cases in low growth clusters at diagnosis (Supplementary Table 3).
Excluding Out-of-Care Cases and Interval Censoring
To ensure these results were not primarily driven by cases who fell out of care, we repeated the Cox regression but including only cases with a follow-up visit within 1 year after being linked to care (Table 3). In the restricted analyses, the relationships between cluster growth at diagnosis and (i) diagnosis to care, (ii) care to suppression, and (iii) suppression to rebound were unchanged. However, the time from diagnosis to suppression was no longer significant (P = .30). This result suggests that the difference in time from diagnosis to suppression may be driven by cases diagnosed in low growth clusters who disproportionately fell out of care. The same pattern was observed when we restricted this analysis to only cases linked to care within a year of being diagnosed. Furthermore, to account for the inherent difference between event time (i.e., viral suppression or rebound) and visit time (i.e., the date of laboratory test), the analysis was repeated using methods that accommodate interval censoring; results of these restricted models were unchanged.
Cluster Membership Predicts Time to Care Continuum Event
We then explored the relationship between cluster growth and outcomes in the care continuum using parametric models. The times from diagnosis to care, care to suppression and diagnosis to suppression were, once again, were not significantly different among cases diagnosed in low, medium, and high growth clusters in the log-normal gamma frailty model (Table 4). However, time from suppression to rebound remained significantly longer for cases diagnosed in high growth clusters (adjusted time ratio [TR] 1.38; P < .001). These results were robust to inclusion only of individuals with at least one follow-up visit within 1 year after first linkage to care (TR 1.36; P = .002). For comparison, the frailty model was also conducted with the commonly used Weibull distribution [38], and the time ratio results were qualitatively similar to those obtained from a log-normal baseline hazard (Supplementary Table 4).
Table 4.
Adjusted Time Ratios for Months to Events in the Continuum of Care by Cluster Growth Accounting for Genetic Heterogeneity (i.e., Membership in Same Transmission Cluster) Using Gamma Frailty Model
| Log Normal Gamma Shared Frailty | |||||||||
|---|---|---|---|---|---|---|---|---|---|
| All Cases | Restricted to Cases With Follow-up Visit Within 1 year After Initial Linkage to Care | ||||||||
| Continuum Interval | Cluster Growth | TRa | 95% CI | θ2 | P | TRa | 95% CI | θb | P |
| Diagnosis to care | Low | Ref. | … | Ref. | … | ||||
| Medium | 0.98 | .85–1.14 | .848 | 0.98 | .84–1.14 | .770 | |||
| High | 0.91 | .77–1.07 | .252 | 0.93 | .79–1.09 | .355 | |||
| 0.11 | <.001 | 0.11 | <.001 | ||||||
| Care to suppression | Low | Ref. | … | Ref. | … | ||||
| Medium | 1.07 | .99–1.14 | .061 | 1.05 | .98–1.12 | .162 | |||
| High | 1.04 | .96–1.11 | .349 | 1.07 | .99–1.15 | .053 | |||
| 0.28 | <.001 | 0.28 | <.001 | ||||||
| Diagnosis to suppression | Low | Ref. | … | Ref. | … | ||||
| Medium | 1.04 | .97–1.11 | .229 | 1.02 | .96–1.09 | .473 | |||
| High | 0.99 | .92–1.07 | .873 | 1.02 | .95–1.10 | .519 | |||
| 0.22 | <.001 | 0.25 | <.001 | ||||||
| Suppression to rebound | Low | Ref. | … | Ref. | … | ||||
| Medium | 1.10 | .91–1.31 | .321 | 1.09 | .91–1.32 | .345 | |||
| High | 1.38 | 1.14–1.68 | .001 | 1.36 | 1.12–1.66 | .002 | |||
| 0.13 | .005 | 0.12 | .007 |
Abbreviations: CI, confidence interval; TR, time ratio.
aTime ratios > 1 indicate slower time to event, whereas time ratios < 1 indicates accelerated time to event.
bTheta is the variability of the gamma frailty across groups of cluster estimated from the data with a mean of 1.
We then investigated the role of heterogeneity among genetic clusters in time to events in the care continuum (i.e., treating cluster membership as a random effect) using a shared gamma frailty model. Along every event interval in the care continuum, the variance of the cluster random effect was highly significant (Table 4). This finding was also observed when we restricted the analysis to individuals with at least 1 follow-up visit within a year of first linkage to care.
To ensure the variance of cluster random effect was not driven by the 2035 cases (39%) that did not cluster in the HIV transmission network, we performed additional sensitivity analyses for the outcome of time from diagnosis to suppression. These analyses included only cases linked genetically to at least one other case diagnosed from 2010 through 2014 (i.e., cluster size ≥ 2). The variance remained highly significant using both log normal and Weibull baseline hazard models (P < .001; Supplementary Table 5).
Even after adjusting for individual level confounders (i.e., birth sex, age at diagnosis, race/ethnicity, CD4+ count at diagnosis, and transmission risk), the variance of the cluster random effect remained significant (P < .001); the magnitude of the estimate effect was reduced by a third (Supplementary Table 5). For all events in the HIV care continuum, event times were more similar within than across transmission clusters (Supplementary Figure 3).
Discussion
We characterized the relationship between the HIV transmission network and the care continuum using HIV surveillance data from individuals diagnosed and residing in Los Angeles County. The timing of progression to events in the HIV care continuum was more similar among individuals who were members of the same transmission cluster than among individuals in different transmission clusters. For example, individuals in clusters with members that take longer to achieve viral suppression are themselves more likely to take longer to achieve suppression. This effect extends beyond what is captured by demographic, laboratory, and transmission risk surveillance data, which themselves tend to be assortative across the transmission network [39]. Therefore, epidemiological connections informed by genetic linkage can potentially inform future progression through the care continuum. We suggest that public health services related to linkage-to-care, viremia support, and return-to-care could be prioritized for individuals in clusters whose other members have failed to timely reach events in the care continuum or have fallen out-of-care.
Our results do not support the hypothesis that continued expansion of rapidly growing clusters is associated with prolonged viremia due to slower progression through the care continuum. In fact, we estimated that individuals in rapidly growing clusters at diagnosis took 17% longer to experience postsuppression viral rebound. Furthermore, among individuals who took longer than 2 months after diagnosis to be linked to care, those in rapidly growing clusters at diagnosis became virally suppressed 29% faster. LAC DPH did not initiate cluster investigation and response until after 2016, so these findings are not influenced by cluster-based interventions. We acknowledge that cluster growth is influenced by access to and engagement with care, including HIV testing: a higher proportion of cases in low growth clusters had CD4+ count < 200 copies/mm3 at diagnosis. The shorter time to viral rebound for cases in low growing clusters may be related to this decreased engagement with care.
The HIV care continuum is typically evaluated at a population level over a fixed time period (e.g., 51% of people living with HIV in the US were virally suppressed in 2015 [1]). However, the HIV transmission network comprises individuals who may or may not be linked to care or virally suppressed at any given time point. This discrepancy informed our study design: our approach focused on the time from linkage to care to virologic suppression, rather than on the proportion of people who successfully transition from linkage to suppression in the next year. Furthermore, our study design did not permit analysis of retention in care, which relies on reported laboratory results whose completeness is around 80% [40]. Hence, missing lab results are an unreliable indicator of care status. In addition, relocation outside of LAC results in erroneous designation of out-of-care status. Therefore, the relationship between out-of-care status and the transmission network would be better assessed at the state or federal level.
Our conclusions are limited to individuals with reported laboratory data. Cluster growth cannot be assessed for individuals without a reported HIV sequence, and characterizing progression through the care continuum relies on reported laboratory data. Nonetheless, among observable cases, our results demonstrate the underlying relationship between shared membership in an HIV transmission cluster and progression through the care continuum.
We described the relationship between the HIV transmission network and the rate of progression through the HIV continuum of care in Los Angeles County. Planned efforts to reduce HIV incidence through the Ending the HIV Epidemic initiative [10] would benefit from further study of this relationship.
Supplementary Data
Supplementary materials are available at Clinical Infectious Diseases online. Consisting of data provided by the authors to benefit the reader, the posted materials are not copyedited and are the sole responsibility of the authors, so questions or comments should be addressed to the corresponding author.
Notes
Acknowledgments. The authors are grateful to their partners at the Los Angeles County Department of Public Health (LAC DPH) Division of Human Immunodeficiency Virus (HIV) and sexually transmitted disease (STD) Programs for their collaboration on this research. The study was approved by the University of California (UC) San Diego and LAC DPH Institutional Review Boards. All data were collected through standard surveillance protocols. Per the terms of a Data Use Agreement between LAC DPH and UC San Diego, all data were deidentified prior to sharing with UC San Diego researchers. To further preserve individual privacy, only month and year for the date of diagnosis and lab tests were shared.
Financial support. This work was supported by an International Design Excellence Award from the California HIV Research Program (CHRP) ID15-SD-052, National Institutes of Health (NIH)-National Institute of Allergy and Infectious Diseases (NIAID) Career Development Award K01AI110181, NIH-NIAID R01 AI135992, and NIH NIAID R37 51164.
Potential conflicts of interest. J. O. W. receives grants and contracts through his institution from Gilead Sciences and the Centers for Disease Control and Prevention (CDC). All other authors report no potential conflicts. All authors have submitted the ICMJE Form for Disclosure of Potential Conflicts of Interest. Conflicts that the editors consider relevant to the content of the manuscript have been disclosed.
References
- 1. Centers for Disease Control and Prevention. Understanding the HIV care continuum Available at: https://www.cdc.gov/hiv/pdf/library/factsheets/cdc-hiv-care-continuum.pdf. Accessed 11 June 2019.
- 2. Kay ES, Batey DS, Mugavero MJ. The HIV treatment cascade and care continuum: updates, goals, and recommendations for the future. AIDS Res Ther 2016; 13:35. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 3. Li ZPD, Sansom SL, Hayes D, Hall HI. Vital signs: HIV transmission along the continuum of care—United States, 2016. MMWR Morb Mortal Wkly Rep 2019; 68:267–72. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 4. Cohen MS, Chen YQ, McCauley M, et al. ; HPTN 052 Study Team Prevention of HIV-1 infection with early antiretroviral therapy. N Engl J Med 2011; 365:493–505. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 5. Cohen MS, Chen YQ, McCauley M, et al. ; HPTN 052 Study Team Antiretroviral therapy for the prevention of HIV-1 transmission. N Engl J Med 2016; 375:830–9. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 6. Quinn TC, Wawer MJ, Sewankambo N, et al. Viral load and heterosexual transmission of human immunodeficiency virus type 1. Rakai Project Study Group. N Engl J Med 2000; 342:921–9. [DOI] [PubMed] [Google Scholar]
- 7. Rodger AJ, Cambiano V, Bruun T, et al. ; PARTNER Study Group Risk of HIV transmission through condomless sex in serodifferent gay couples with the HIV-positive partner taking suppressive antiretroviral therapy (PARTNER): final results of a multicentre, prospective, observational study. Lancet 2019; 393:2428–38. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 8. World Health Organization. Policy brief: consolidated guidelines on HIV prevention, diagnosis, treatment and care for key populations, 2016. update. Available at: http://www.who.int/hiv/pub/guidelines/keypopulations/. Accessed 11 June 2019.
- 9. Granich RM, Gilks CF, Dye C, De Cock KM, Williams BG. Universal voluntary HIV testing with immediate antiretroviral therapy as a strategy for elimination of HIV transmission: a mathematical model. Lancet 2009; 373:48–57. [DOI] [PubMed] [Google Scholar]
- 10. Fauci AS, Redfield RR, Sigounas G, Weahkee MD, Giroir BP. Ending the HIV epidemic: a plan for the United States. JAMA 2019; 321:844–5. [DOI] [PubMed] [Google Scholar]
- 11. Hué S, Pillay D, Clewley JP, Pybus OG. Genetic analysis reveals the complex structure of HIV-1 transmission within defined risk groups. Proc Natl Acad Sci U S A 2005; 102:4425–9. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 12. Lewis F, Hughes GJ, Rambaut A, Pozniak A, Leigh Brown AJ. Episodic sexual transmission of HIV revealed by molecular phylodynamics. PLoS Med 2008; 5:e50. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 13. Smith DM, May SJ, Tweeten S, et al. A public health model for the molecular surveillance of HIV transmission in San Diego, California. AIDS 2009; 23:225–32. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 14. Poon AF, Gustafson R, Daly P, et al. Near real-time monitoring of HIV transmission hotspots from routine HIV genotyping: an implementation case study. Lancet HIV 2016; 3:e231–8. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 15. Oster AM, France AM, Mermin J. Molecular epidemiology and the transformation of HIV prevention. JAMA 2018; 319:1657–8. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 16. Wertheim JO, Chato C, Poon AFY. Comparative analysis of HIV sequences in real time for public health. Curr Opin HIV AIDS 2019; 14:213–20. [DOI] [PubMed] [Google Scholar]
- 17. Poon AF, Joy JB, Woods CK, et al. The impact of clinical, demographic and risk factors on rates of HIV transmission: a population-based phylogenetic analysis in British Columbia, Canada. J Infect Dis 2015; 211:926–35. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 18. Wertheim JO, Murrell B, Mehta SR, et al. Growth of HIV-1 molecular transmission clusters in New York City. J Infect Dis 2018; 218:1943–53. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 19. Dennis AM, Hue S, Billock R, et al. HIV-1 phylodynamics to detect and characterize active transmission clusters in North Carolina. J Infect Dis 2019. doi:10.1093/infdis/jiz176 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 20. Kouyos RD, von Wyl V, Yerly S, et al. Molecular epidemiology reveals long-term changes in HIV type 1 subtype B transmission in Switzerland. J Infect Dis 2010; 201:1488–97. [DOI] [PubMed] [Google Scholar]
- 21. Volz EM, Koopman JS, Ward MJ, Brown AL, Frost SD. Simple epidemiological dynamics explain phylogenetic clustering of HIV from patients with recent infection. PLoS Comput Biol 2012; 8:e1002552. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 22. Wertheim JO, Oster AM, Switzer WM, et al. Natural selection favoring more transmissible HIV detected in United States molecular transmission network. Nat Commun 2019; 10:5788. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 23. Dasgupta S, France AM, Brandt MG, et al. Estimating effects of HIV sequencing data completeness on transmission network patterns and detection of growing HIV transmission clusters. AIDS Res Hum Retroviruses 2019; 35:368–75. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 24. 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; 79:543–50. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 25. Wertheim JO, Panneer N, France AM, Saduvala N, Oster AM. Incident infection in high-priority HIV molecular transmission clusters. Conference on Retroviruses and Opportunistic Infections. Seattle, Washington, 2019. Accessed 11 June 2019. [Google Scholar]
- 26. Division of HIV and STD Programs. 2017 Annual HIV Surveillance Report. Los Angeles: Los Angeles County Department of Public Health, 2017. [Google Scholar]
- 27. Los Angeles County Department of Public Health. 2016 COH Annual Planning Meeting Recap. Available at: publichealth.lacounty.gov/dhsp/Presentations/DPH_PRESENTATION_7.13.17_FINAL.pdf. Accessed 6 Jan 2019.
- 28. Ragonnet-Cronin M, Hué S, Hodcroft EB, et al. ; UK HIV Drug Resistance Database Non-disclosed men who have sex with men in UK HIV transmission networks: phylogenetic analysis of surveillance data. Lancet HIV 2018; 5:e309–16. [DOI] [PubMed] [Google Scholar]
- 29. United States Census Bureau. 2016 American community survey 1-year estimate - poverty status in the past 12 months by sex by age. Available at: https://factfinder.census.gov/faces/tableservices/jsf/pages/productview.xhtml?pid=ACS_17_1YR_B17001&prodType=table. Accessed 9 September 2019.
- 30. Kosakovsky Pond SL, Weaver S, Leigh Brown AJ, Wertheim JO. HIV-TRACE (TRAnsmission Cluster Engine): a tool for large scale molecular epidemiology of HIV-1 and other rapidly evolving pathogens. Mol Biol Evol 2018; 35:1812–9. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 31. Ragonnet-Cronin M, Hu YW, Morris SR, Sheng Z, Poortinga K, Wertheim JO. HIV transmission networks among transgender women in Los Angeles County, CA, USA: a phylogenetic analysis of surveillance data. Lancet HIV 2019; 6:e164–72. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 32. Hué S, Clewley JP, Cane PA, Pillay D. HIV-1 pol gene variation is sufficient for reconstruction of transmissions in the era of antiretroviral therapy. AIDS 2004; 18:719–28. [DOI] [PubMed] [Google Scholar]
- 33. Wertheim JO, Kosakovsky Pond SL, Forgione LA, et al. Social and genetic networks of HIV-1 transmission in New York City. PLoS Pathog 2017; 13:e1006000. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 34. Cox DR, Oakes D. Analysis of Survival Data Boca Raton: Chapman & Hall/CRC Press; 2018. Available at: https://www.taylorfrancis.com/books/9781315137438 Accessed 4 March 2019. [Google Scholar]
- 35. Lindsey JK A study of interval censoring in parametric regression models. Lifetime Data Analysis 1998; 4:329–54. [DOI] [PubMed] [Google Scholar]
- 36. Gutierrez RG Parametric frailty and shared frailty survival models. Stata Journal 2002; 2:22–44. [Google Scholar]
- 37. Hougaard P Frailty models for survival data. Lifetime Data Analysis 1995; 1:255–73. [DOI] [PubMed] [Google Scholar]
- 38. Aalen OO Effects of frailty in survival analysis. Stat Methods Med Res 1994; 3:227–43. [DOI] [PubMed] [Google Scholar]
- 39. Oster AM, Wertheim JO, Hernandez AL, Ocfemia MC, Saduvala N, Hall HI. Using molecular HIV surveillance data to understand transmission between subpopulations in the United States. J Acquir Immune Defic Syndr 2015; 70:444–51. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 40. Karch DL, Chen M, Tang T. Evaluation of the national human immunodeficiency virus surveillance system for the 2011 diagnosis year. J Public Health Manag Pract 2014; 20:598–607. [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.