Predictive Factors of Detectable Viral Load in HIV-Infected Patients

Worldwide ∼38 million individuals are infected with HIV.¹ Evidence suggests better overall management of the disease, notably because of recent therapeutic advances including the advent of integrase inhibitors that are very effective and well tolerated.² Despite better efficacy and tolerability with newer antiretroviral therapy (ART), some individuals still maintain a detectable viral load (VL). In 2017, 70% of people living with HIV were aware of their HIV status, 77% of these patients had access to ART, and 82% of them were maintaining an undetectable VL, whereas the Joint United Nations Program on HIV and AIDS (UNAIDS) 90-90-90 goals set in 2014 target having 90% of HIV-infected people diagnosed, of whom 90% are receiving ART, and of whom 90% have an undetectable VL.^3,4

In Quebec, ∼20,000 people were living with HIV in 2016. In Montreal, the City Without AIDS Action Plan estimated that in 2015 86% of people living with HIV in Montreal knew their status, 97% of these were treated, and 92% of these had an undetectable VL.³

Several authors have studied predictive factors of detectable VL in HIV patients taking ART medication.^5–12 Potential factors included patient characteristics, type of treatments and virus features. Patient characteristics included history of injection drug use, female gender, ethnicity, inappropriate adherence to ART, time since diagnosis, and age <30 years old.^5,7–9,13 Treatment-related predictors included past exposure to ART, history of virologic failure, and boosted protease inhibitor regimens.^6,7 HIV mutations also predicted detectable VL.^10–12

This study aimed to identify predictive factors of detectable VL in HIV-infected patients in the era of highly active ART that are well tolerated.

A total of 2,938 subjects were followed between 2016 and 2018. By applying the inclusion criteria to this cohort, 1,781 patients remained in the data set. After the application of the exclusion criteria, we selected 1,262 potential subjects. Of these, 200 (15.8%) had a detectable VL in 2018. All cases who were available were included. Cases were matched with 200 controls based on a 1:1 ratio (total 400 subjects).

Descriptive statistics of the study sample are given in Table 2. The study sample was mainly composed of males (68%) who were diagnosed with HIV before 2011 (86.5%). The median age was 54 years old. Of all patients, 61% were born in North America. The median CD4 lymphocyte count was 500 cells/μL for the cases and 600 cells/μL for the controls. Cases were less frequently men who have sex with men (35.0%) compared with controls (44.5%). Cases had more substance use disorders (33.0% vs. 21.5%), were more homeless (8.5% vs. 2.5%), and had a greater prevalence of previous or active psychiatric diseases (43.5% vs. 34.5%). Since HIV diagnosis, 34.8% (38.5% for cases vs. 31.0% for controls) of the subjects had received five or more ART treatments. Cases were more covered by a public or governmental insurance medication plan (76.5% vs. 63.0%). Cases had lower adherence to antiretrovirals (34.5% vs. 66.0%). Of all subjects, 41% missed two clinic appointments or more within a 12-month period.

In the conditional multivariable regression, medication discontinuation for more than a week (odds ratio [OR]: 19.00, 95% CI: 5.59–64.57), suboptimal medication adherence (OR: 4.69, 95% CI: 2.03–10.84), health care utilization (OR: 2.22, 95% CI: 1.10–4.48), and seeking asylum (OR: 0.22, 95% CI: 0.06–0.84) were significantly associated with detectable VL (Table 3). The variables strongly predictive of the primary outcome were medication adherence (with two categories, selected in 99% of resampled data sets), asylum seeker status (83%), and seeking health care services defined as a visit to the emergency or hospitalization (76%) (Table 3). Complete results from the variable selection are presented in Figure 1. These results were consistent with the multivariable conditional logistic regression analysis (Table 3).

In the conditional logistic regression analysis using only the selected variables (selection >50%), the asylum seeker status was inversely related to having a detectable VL (OR = 0.31) (Table 3). Stopping antiretrovirals for >1 week (OR = 15.94), suboptimal antiretroviral adherence by forgetting punctual doses (OR = 3.87), and seeking health care services (OR = 1.91) were positive predictive factors of detectable VL.

Of all cases (n = 200), 78 had virologic failure, 86 a viral blip, and 27 persistent low-level viremia. In the conditional logistic regression bivariate analyses of secondary outcomes, only inadequate medication adherence (forgetting punctual doses and stopping for >1 week) was a strong predictive factor of virologic failure (selected >75%) (Supplementary Table S2). When the outcome was viral blip, none of the variables were strong predictive factors (Supplementary Table S3). Owing to the very small sample size for persistent low-level viremia, the imputation models failed to converge when applied to the bootstrapped data sets (Supplementary Table S4). Frequencies of variable selection are not available for this last outcome. The full multivariable model failed to converge for virologic failure, resulting in highly unstable results for viral blip, and was impossible to conduct due to small sample size for persistent low-level viremia. Hence, multivariable results are not available for secondary outcomes.

This study was a retrospective matched case–control study of HIV-infected individuals followed at the CHUM and MUHC clinics in 2018. Of the 2,938 subjects, 200 were considered cases and were matched with 200 controls.

Establishing predictive factors and models of detectable VL is essential for early identification of patients at risk of this outcome. Detectable VL can lead to increased morbidity and mortality.^2,7 Determining causality of the predictive factors of detectable VL was not the goal of this study.

Results from the main analysis showed that being an asylum seeker was inversely related to detectable VL. This could be, in part, explained by the existence of the Canadian Interim Federal Health Program that facilitates the access for asylum seekers to medications and to a systematic HIV screening system. The multidisciplinary approach offered by both clinics also ensures extensive follow-up in this population that may explain the better outcomes observed in our study. Mendelsohn et al.¹⁸ showed similar HIV control between refugees and host community patients. Tiittala et al.¹⁹ found major HIV knowledge gaps in migrants in comparison with the general population but had no data on how these differences could impact the HIV VL.

Seeking health care services was associated with a significant increase in the risk of detectable VL. Requiring more frequent medical assistance could be seen as a surrogate for the complexity of the disease, therefore, implying comorbidities and social precariousness.²⁰ Chow et al.²¹ performed an observational study that revealed a strong connection between hospitalization rate and high VL count.

Stopping ART for >1 week and forgetting punctual doses were found to be highly predictive of detectable VL. Many studies identify nonadherence to antiretrovirals as a plausible cause of detectable VL.^9,22–25 Virologic rebound is expected after cessation of ART.^26–28 This study asserts that establishing patient's adherence history is useful for identifying at-risk patients. Specific causes of inadequate adherence were not explored in this study.

Adherence to ART was found to be the only significant predictive factor of virologic failure. However, the sample sizes for the secondary outcomes were small (n = 200), which limited the number of predictors that could be identified. Results concerning viral blips and persistent low-level viremia were inconclusive since none of the variables were selected in >75% of resampled data sets.

This study has strengths and limitations. The data were collected by medical chart review by trained clinicians allowing for adequate analysis of the information in the chart. The study population was representative of the target population because it included a substantial proportion of the HIV-infected patients in Montreal, Canada. Furthermore, all subjects who fit the inclusion and exclusion criteria for cases were included. However, the definitions of cases included only the patients who had a detectable VL in 2018. Only 200 cases were included, which is a relatively small sample size. This made analysis of secondary outcomes more difficult. Temporal ordering was ensured by doing baseline data collection during a time period (January 2016 to December 2017) before the outcomes (assessed in 2018). However, the data collection period could have differed between subjects depending on start of medical follow-up at clinic and the time of the detectable VL. This could enhance the link between the predictive factors and the detectable VL. In our clinical setting, detectable VL usually triggers more extensive investigation of patient adherence and socioeconomic factors. This could lead to report bias in the medical records. Data collection was limited by information available in subjects' files due to the observational retrospective study design producing missing data for many variables. Multiple imputation was done to limit the impact of the missing data. As gender at birth and date of HIV diagnosis were used as matching variables, their predictive potential of detectable VL could not be evaluated. Gender has previously been identified as influencing VL.⁵ HIV diagnosis year was chosen as a matching criterion because this variable ensures that cases and controls have a history of exposure similar to antiretrovirals and professional care. Indeed, data on time of ART initiation, ART-experienced, and number of ART regimens since diagnosis were similar between cases and controls.

This retrospective matched case–control study demonstrated that inadequate adherence (forgetting punctual doses and stopping antiretroviral for more than a week) to ART and seeking health care services are predictors of detectable VL. Similar findings were obtained by other research teams. In our clinical setting, being an asylum seeker was inversely related to detectable VL. Ascertaining these factors will allow for identification of patients more at risk of detectable VL. When intervening in these at-risk patients, clinicians should use an individualized approach to prevent the occurrence of predictive factors of detectable VL.

The authors thank Stéphanie Matte (CHUM), Dr. Cécile Tremblay (CHUM), Dr. Marina Klein (MUHC), and Costas Pexos (MUHC) for their help in accessing data and the patients followed at each clinic.