Study population
Data for this study was obtained from the Washington DC Metropolitan WIHS site which is housed at Georgetown University in Washington DC with sub-sites in Montgomery County Maryland and in northern Virginia. WIHS is an ongoing prospective cohort study of HIV infection in women. WIHS recruited women from six sites (Bronx and Brooklyn, New York; Chicago, Illinois; Los Angeles and San Francisco, California; and, Washington, DC) during 3 phases (1994–1995; 2000–2001; 2012–2013) [17]. Data from all three waves of the Washington DC site were used for this analysis. Details of recruitment and enrollment for WIHS have been described previously [17, 18]. The DC WIHS recruited women through community outreach and among care providers within DC, and is not a clinic-based cohort. Within this non-intervention observational cohort, health outcomes reflect the local treatment practices and health-seeking behaviors of the participants [17, 18]. For this study, DC WIHS HIV-positive participants who contributed at least four visits over the course of the study were included in this analysis.
Outcomes
Laboratory collection methods and measurements of viral load (plasma HIV-1 RNA) and CD4+ cell count included isothermal nucleic acid sequence-based amplification and standard flow cytometric protocols, respectively, and have been previously described.[19] HIV RNA detection levels over time were set at the level below assay detection (which varied from <80 copies/mL in semi-annual visits 1–28; <48 copies/mL in visits 29–33; and, < 20 copies/mL for visits 34–36).
Viral load suppression for an individual at a particular visit was defined as viral load less than or equal to the detection limit at the time of the assay. For each visit for which there were data, individuals were assigned 1 for suppression variable if viral suppression was achieved, or 0 if not. Cumulative viral load suppression-year was defined for each individual at each visit by summing suppression values for the current and all prior visits and dividing by two.
Time of report of death during study for any cause was used as the mortality outcome. Ascertainment and classification of deaths in WIHS have been previously described [20, 21].
Covariates
Demographic variables were recorded with survey questionnaires. Covariates from this questionnaire include the constants race (black defined as non-Hispanic black, Hispanic defined as Hispanic of any race, other defined as non-white, non-black, non-Hispanic, and, referent of white defined as non-Hispanic white) and education (> = 12 years of school with <12 years as referent), and the time-varying variables housing (reporting having own home or, apartment with the following included in the referent of non-housed: living in a parent’s house; someone else’s house, or apartment; a rooming, boarding, or halfway house; a shelter or welfare hotel; the street; jail or correctional facility; a residential drug or alcohol treatment facility; other place; or no report), depressive symptoms (Center for Epidemiologic Studies Depression Scale; CES-D > =16 at visit with CES-D < 16 as no depressive symptoms referent), illicit drug abuse (reported use of at least one of the following since the last visit: marijuana or hash; crack; cocaine; heroine; illicit methadone; methamphetamines; amphetamines, narcotics, hallucinogens, and other drugs; injected drugs; or non-injected drugs. Alcohol was not included), and alcohol use (use defined as > = 7 reported drinks per week since last visit).[22] HIV medication use was also reported with a questionnaire using time-varying variables adherence to treatment (taking HIV drugs > =95 % of the time with <95 % use as referent) and therapy (ART, including mono therapy or combination therapy; cART, including at least three antiretrovirals from at least two drug classes based on the Department of Health and Human Services 2008 guidelines;[23] and no ART/cART treatment as referent). Age (in years) was used as constant (age at participant’s baseline visit in study) or time-varying (age at visit) depending on the analysis.
Statistical methods
Descriptive statistics using data from the baseline visit were generated. Group-based trajectories were modeled using a logistic trajectory model as a function of visits (PROC TRAJ, available online: http://www.andrew.cmu.edu/user/bjones/) with HIV RNA detection as a binary variable. The optimal number of trajectories was selected based on the Bayesian information criteria (BIC); the model with the lowest BIC value representing the statistically optimal number of latent groups. Group characteristics were explored with generalized linear modeling with generalized estimating equations for repeated measures using PROC GENMOD. Variables from univariate analyses with P < 0.1 were included in multivariate models. The overall lost to follow up rate in this specific study group was calculated using the number of individuals classified as “missing” and “disenrolled” in the DC WIHS. Overall mortality in this specific study group was calculated using the number of deceased recorded by DC WIHS. Median viral load and the interquartile range were calculated using viral load reports over time in the study group.
A graph depicting cross sectional proportion of DC WIHS women with viremia and those without viremia was plotted to compare HIV care continuum with results from longitudinal group-based trajectory analyses. Data used included only observations with recorded values and excluded deaths and missing data.
Mean value of viral suppression cumulative years of each visit for each of the three HIV treatment careers was calculated and plotted. To probe the relationship between HIV treatment careers and structural, biographical and clinical factors, both univariate and multivariate multinomial logistic regression analyses were conducted. For time-varying predictors, random effects for subjects were included to account for repeated measures.
Kaplan-Meier survival analysis was performed to identify differences in mortality between the trajectory groups, and univariate and multivariate Cox proportional hazards modeling was conducted to identify predictors of survival. For Cox regression, race, age at entry to study, education, and group variables were treated as constant whereas remaining predictors were treated as time varying.
All analyses were performed in SAS 9.4 64-bit and statistical significance was defined as P <0.05.
Role of the funding sources
The National Institute of Allergy and Infectious Diseases (NIAID) (UO1-AI-34994; PI: Mary A. Young) and the National Cancer Institute, the National Institute on Drug Abuse and the Eunice Kennedy Shriver National Institute of Child Health and Human Development funded the collection of data for this study from the Washington DC Metropolitan site of the Women’s Interagency HIV Study (WIHS), and the Office of the Senior Vice President for Research at Georgetown University funded additional analytical support.