Predicting Viral Failure in Human Immunodeficiency Virus Perinatally Infected Youth With Persistent Low-Level Viremia on Highly Active Antiretroviral Therapy

Abstract

Background

Less than optimal adherence with antiretroviral therapy occurs commonly among human immunodeficiency virus HIV)-infected youth. In this study, our object was to identify patterns in the prefailure measurement of viral load (VL) that can reliably predict virological failure (VF) in HIV perinatally infected youth on highly active antiretroviral therapy (HAART).

Methods

We conducted a retrospective chart review of HIV-infected youth with low-level viremia (LLV), defined as an HIV VL between the lower limits of detection (20–75 copies/mL) and 1000 copies/mL. All patients were perinatally infected, under 22 years of age, observed for at least 24 months of consecutive follow-up between May 2008 and July 2014, and received their HIV care at the University of Miami Miller School of Medicine. Of the 349 subjects screened, 100 were eligible for analysis. Virological failure was defined as 3 or more consecutive VLs greater than 1000 copies/mL. Multiple logistic regression and receiver operator characteristic curves were used to identify patterns in VL that ultimately resulted in VF.

Results

Fifteen of the 100 patients experienced VF. Higher log₁₀ mean VL, positive slope of the VL (log₁₀ copies/mL per day), and fewer clinic visits were associated with a higher probability of VF. Sensitivity and specificity were .87 and .95, respectively. Resistance was not found in 12 of 15 patients with VF.

Conclusions

Patients with LLV that had fewer clinic visits and a trend toward increasing VLs had an increased risk of VF. Noncompliance seems to be a major component of VF. Physicians should emphasize the critical nature of medication adherence.

The 2016 Department of Health and Human Services (DHHS) Guidelines for the use of antiretroviral therapy (ART) in Pediatric HIV Infection define virological failure (VF) as an incomplete initial response to ART or as viral rebound after viral suppression is achieved [1]. Alternatively, the DHHS guidelines has proposed other definitions of VF. These include children with a viral load (VL) >200 copies/mL after 6 months of therapy or repeated plasma VL greater than lower limit of quantification relative to the particular VL assay. Perinatally infected infants may take longer to achieve viral suppression, and continued close monitoring is recommended [1].

The ability to achieve and maintain an undetectable human immunodeficiency virus (HIV) VL appears to be a reasonable goal among adult patients who are adherent with their highly active ART (HAART) regimen. Many HIV-infected adults with detectable HIV ribonucleic acid (RNA) <200 copies/mL after 6 months of HAART often ultimately achieve viral suppression without regimen change [2–4]. However, approximately 4% to 8% of the HIV-infected adults receiving HAART subsequently experience persistent, detectable, low-level viremia (LLV) that may lead to overt VF if the HAART regimen is not altered [5, 6]. Persistent LLV is also associated with higher immune activation [7]. However, the most recent version of the DHHS Guidelines for HIV-Infected Adults and Adolescents [8] notes that the data on the association between persistently low-level but quantifiable viremia (HIV RNA <200 copies/mL) and viral failure is conflicting with one recent study that showed an increased risk of subsequent failure at this level of viremia that was not observed in other studies [5, 6, 9, 10].

Our anecdotal experience at the University of Miami Miller School of Medicine (UMMSOM) with caring for a large population of perinatally infected children and adolescents over several years does not mirror the adult experience. Episodic and/or continuous nonadherence with HAART continues to be a major problem suspected to be associated with factors such as patient denial, lack of family and social support, immaturity, as well as economic factors including lapses in insurance coverage. Even among our child and adolescent patients with good compliance, LLV commonly occurs. Based upon conversations with colleagues caring for similar populations of HIV-affected youth in large urban areas, our experience is not unique (personal oral communication Dr. Murli Purswani [Bronx-Lebanon and Montefiore Medical Centers, Bronx, NY], Dr. Ram Yogev [Northwestern, Chicago, IL], and Dr. Patricia Flynn [St. Jude’s Research Hospital, Memphis, TN]).

The effect of persistent LLV remains understudied in perinatally infected children, and the significance of persistent LLV in predicting the development of VF in children with vertically acquired HIV infection is not well known, thus the clinical consequences and appropriate management of these patients remains unclear. A major difficulty is attempting to discern the effects of spotty medication compliance, which may result in inconsistent viral control (ie, frequent peaks and dips in VL). Although it is logical to expect that such behavior should increase the risk of VF, it is not known whether it is possible to identify trends in the serial VL from patients with LLV that may put patients at higher risk for eventually developing VF. Therefore, the goal of this study was to identify trends in VL that could reliably predict VF among a population of children and adolescents with LLV.

METHODS

Study Design and Patient Population

We performed a retrospective, observational study involving a population of HIV perinatally infected children and adolescents. Data were collected from May 2008 through July 2014 utilizing the Electronic Health Records available in the Pediatric Special Immunology Clinic at the UMMSOM. The inclusion criteria for this study were as follow: (1) documented perinatally acquired HIV infection; (2) under 22 years of age at presentation; (3) receiving HAART, defined as a regimen consisting of 3 antiretroviral ART agents from 2 or more classes; (4) followed at UMNSOM for at least 24 consecutive months; (5) achieved persistent LLV, defined for purposes of this study as >20–50 copies/mL but <1000 copies/mL persistently while on HAART for at least 24 months; and (6) had regular ambulatory follow-up visits. Patients were excluded from this analyses if they (1) were behaviorally infected, (2) were not taking any HAART because of admitted poor compliance or HAART was not indicated, or (3) had fewer than 2 follow-up visits per year. Data from the patients included in this study were censored either at the end of the study (July 31, 2014), at the time that HAART was discontinued, the patient was lost to follow up or developed viral failure as defined for this study (see definition of quantified variables given below). Because this was an observational and not a fixed design analysis, we were careful to account for temporal spacing of the repeated observations as well as the total number of observations by including these parameters in the statistical model. Data collected produced a continuous time series for each subject. Their initial observation was considered time zero, and they were followed until VF or until the time of their final observation. For those subjects who experienced VF, only data before VF were analyzed. The study protocol was approved by the Institutional Review Board at the UMMSOM.

Demographic and Laboratory Variables

Demographics included age, sex, race, and the Centers for Disease Control and Prevention HIV Clinical and Immunological Classification. Human immunodeficiency virus-associated variables included HIV VL and CD4 cell counts and percentages. Low-level viremia was defined as persistently detectable plasma HIV VL levels greater than the lower limit of detection (20–50 copies/mL) but less than 1000 copies/mL for at least 6 months over at least 3 consecutive clinic visits. Twenty-four months of continuous follow-up at UMMSOM was required as an eligibility criterion for entry into the study. Clinic visits occurred approximately every 3 months. Virological failure was defined as 3 or more consecutive plasma HIV VL levels greater than 1000 copies/mL, in patients who had previously maintained persistent LLV. For purposes of this study, a viral blip (VB) was defined as an isolated HIV VL level >1000 copies/mL but that did not persist in meeting the definition of VF as defined above. The working definition of blips in this study was developed as a reflection of what we have commonly encountered in our clinic population of perinatally infected youth who have less than optimal compliance and VLs that are often less than 1000 but greater than 100–200. Occasional isolated VL >1000 were observed usually due to lapses in ART compliance, but, again, these were not considered to represent VF. The slope of VL (slope VL) was determined by fitting a least squares linear model to each patient’s VL time series across clinic visits.

Human immunodeficiency virus VL quantification was performed using COBAS AmpliPrep/COBAS TaqMan HIV-1 test kit with a prior lower limit of detection of <50 copies/mL and since May 2008 a revised lower limit of detection of 20 copies/mL. CD4 cell count was measured by flow cytometry. All clinical laboratory assays were ordered on a quarterly basis and performed by a certified clinical laboratory (Quest or LabCorp).

Resistance Testing

Testing for the presence of specific antiretroviral agent resistance mutations was ordered at the discretion of the patient’s primary care physician when viral failure occurred as part of their ongoing clinical care. Resistance testing was performed using standardized, commercially available assays as per the manufacturers’ instructions: genotype-based assays (HIV Genotype [Quest]; Trugene HIV-1 [Bayer]) and PhenoSense-based assays (PhenoSense GT [Monogram Biosciences, Virologic Inc.] or HIV Integrase Genotype [Quest Diagnostics]).

Statistical Analysis

Patients were classified into 2 HIV groups: (1) those patients experiencing HIV treatment failure as defined for this study and (2) those patients who experienced no HIV treatment failure. Group mean differences were used to evaluate the univariate relations between HIV treatment failure status and the patients/viral/CD4 count characteristics. Independent variables considered to be possible correlates of HIV treatment failure were age at the time of enrollment, mean VL, slopes of sequential VLs between entry and viral failure, median VL, maximum VL, mean CD4 count, mean CD4%, presence of at least 1 blip, total days observed, number of visits, mean time between visits, and mean number of visit per total days observed. When necessary, logarithmic transformations and square roots were utilized to normalize data.

Two sample t tests were used for comparisons among HIV groups. χ² tests were used for comparison of categorical data. For all statistical tests, a P value of less than .05 was considered statistically significant. Multiple logistic regression was used to evaluate the redundancy between the independent variables that were found to have univariate associations with HIV treatment group classification. Receiver operator characteristic curves (ROCs) were used to tune the results of the logistic regression to maximize sensitivity and specificity.

RESULTS

A total of 349 patients were initially screened for inclusion into this study. One hundred forty-eight patients of the original 349 were identified as having persistent LLV (42.4%). Forty-eight were subsequently excluded for the following reasons: 27 had been followed up for less than 24 months, 8 had behavioral-acquired HIV infection, 4 were not on HAART, 2 had gaps in their follow up greater than 7 months, 4 had unknown mode of transmission, and 3 had incomplete records. This resulted in net total of 100 patients who met the inclusion/exclusion criteria and were the final subjects used in the statistical analysis.

Of the 100 patients who met the inclusion/exclusion criteria, 15 (15%) developed VF as defined for this study, whereas 85 (85%) did not. Table 1 describes the demographics of this patient population by VF. Figure 1 depicts representative graphs of the VL from 3 of the subjects who met the criteria for VF (Figure 1, 2.d–2.f.) and 3 who did not (Figure 1, 2.a.–2.c.). There were no statistically significant differences in any of the demographic or HIV characteristics examined (P > .3161). Even when the more rigorous Cochran-Armitage test for trends was applied to HIV category and stage, no statistical differences were detected (P ≥ .1431). This suggests that there was no confounding effects of sex, race, clinical acquired immune deficiency syndrome, or profound immunosuppression across HIV groups.

Table 1.

Demographics of Patients by Virologic Failure Status

	Failures (Frequency)
Characteristic	No n = 85	Yes n = 15	Failure Rate	χ2	P Value
Sex
Male	44	8	.154	0.0126	.9107
Female	41	7	.146
Race
White	21	3	.125	0.1548	.6939
Non-White	64	12	.158
Human Immunodeficiency Virus (HIV) Category
A	24	3	.111	0.7028	.7037
B	25	4	.138
C	36	8	.182
HIV Stage
1	10	3	.231	2.3029	.3161
2	29	7	.194
3	46	5	.098

Note: Sex and race: degrees of freedom = 1. HIV category and HIV stage: degrees of freedom = 2.

Figure 1.

Representative graphs of 3 study subjects (1.a.–1.c.) who did not meet the criteria for virological failure (VF) and 3 subjects (1.d.–1.f.) who did experience VF as defined in this analysis. The x-axis in each case represents dates at which sequential viral loads (VLs) were drawn, whereas the y-axis represents number of copies of human immunodeficiency virus ribonucleic acid/microliter. The persistence of low-level viremia is apparent in both sets of tracings. Although the 2 sets of longitudinal traces of VL are similar, a pattern of increasing VL is apparent in the latter portion of the tracings of those who experienced VF. Longitudinal analysis of both absolute CD4 and %CD4 values were not shown to be associated with VF in this study.

Table 2 presents the mean ages of patients at time zero and the various measures used to assess the duration and degree of follow up by VF status. Patients with VF were observed for less time with the mean number of total days observed being 1146.38 in the VF group versus 1637.84 in the non-VF group (P = .0006). Similarly, the VF group had fewer number of mean clinic visits than the non-VF group (12.53 versus 17.21, P = .0006). No statistical differences were found in mean age, mean time between visits, and mean number of visits per total days observed (Ps > .2220).

Table 2.

Age at Enrollment and Duration of Follow-up Relative by Virologic Failure Status

	Failure
Time Variables Prefailure	No n = 85 [Mean (SD)]	Yes n = 15 [Mean (SD)]	Difference (No – Yes)	95% Confidence Interval for the Mean Difference	t Ratio	P Value
Age at enrollment (years)	11.74 (5.32)	13.50 (3.56)	−1.75	−3.91 to 0.39	−1.22	.222
Total days observed (days)	1637.84 (511.45)	1146.33 (322.44)	491.51	292.96–690.05	3.58	.0006
Number of visits	17.21 (5.01)	12.53 (2.72)	4.68	2.91–6.44	3.51	.0006
Mean time between visits (days)	102.35 (16.90)	98.68 (11.69)	3.67	−3.34 to 10.68	0.80	.4224
Mean number of visits per total days observed (visits/day)	0.01074 (0.001663)	0.011354 (0.002569)	−0.000614	−0.0002 to 0.0008	−1.21	.2262

Abbreviation: SD, standard deviation.

Note: All mean tests conducted with 98 degrees of freedom.

Table 3 describes the viral and immune parameters by VF status. The prefailure mean and median VLs were higher among patients who subsequently developed VF (Ps < .0034), whereas there were no statistical differences in the maximum VL, the mean CD4 cell count, CD4 cell percentage, or VBs (Ps > .3446). The slopes of the VL were statistically different between the 2 groups (P < .0001). The no Failure Group had mean negative slope versus a positive slope for the Failure Group (−0.000147 versus +0.000384 log₁₀ copies/mL per day).

Table 3.

Virological and Immunological Parameters by Virologic Failure Status

	Failures
Variables Prefailure	NO n = 85 [Mean (SD)]	YES n = 15 [Mean (SD)]	Difference (No – Yes)	95% Confidence Interval for the Mean Difference	t Ratio	P Value
Meana VL (log10 copies/mL)	2.167654 (0.281897)	2.416802 (0.190960)	−0.249148	−0.3642 to −0.134	−3.2852	.0014
Slope VL (log10 copies/mL per day)	−0.000147 (0.000373)	0.000384 (0.000420)	−0.000531	−0.00074 to −0.00032	4.99	<.0001
Median VL (copies/mL)	194.24 (130.84)	307.86 (155.93)	−113.62	−198.34 to −28.90	−3.01	.0034
Maximum VL (copies/mL)	1808.92 (1990.98)	1637.86 (1397.55)	171.06	−663.48 to 1005.60	0.31	.7508
Meana CD4 count (SQRT cells/µL)	28.21 (6.07)	27.96 (7.03)	0.25	−3.58 to 4.08	0.14	.8864
Meana CD4% (percent)	33.35 (7.69)	35.34 (6.01)	−1.99	−5.49 to 1.51	−0.94	.3446
Presence of blips (at least 1) (percent)	0.51 (0.5)b	0.53 (0.5)b	−0.027	−0.32 to 0.26	−0.19	.8448

Abbreviations: SD, standard deviation; SQRT, square root; VL, viral load.

^aMean VL over visits.

^bBinomial approximation.

Note: All mean tests conducted with 98 degrees of freedom.

Table 4 summarizes the multiple logistic regression analysis for the 3 independent variables (mean VL, slope VL, and number of visits) that demonstrated simple univariate relationships with VF. Although median VL and total days observed were also univariately related to VF, they were highly correlated with mean VL and number of visits, respectively (r = .58 and .87). Hence, to avoid the problem of multicollinearity, they were not included in the logistic model. All 3 of the former variables remained statistically significant with VF when evaluated simultaneously by multiple logistic regression analysis (Ps < .0143). Although slope VL’s contribution to the prediction of VF was slightly greater (See Table 4, χ² values), the 3 independent variables explained approximately equal amounts of the variation in VF classification.

Table 4.

Logistic Analysis Parameter Estimates

Term	Estimate	Standard Error	χ2	Prob>χ2
Intercept	−9.9655334	4.5879188	4.72	0.0298
Log10 mean viral load	5.10581811	1.9458227	6.89	0.0087
Number of visits	−0.2656448	0.1084141	6.00	0.0143
Log10 slope	3426.86238	1109.3795	9.54	0.0020

Receiver operator characteristic curve analysis was conducted to optimize the sensitivity and specificity of the logistic model. Figure 2 presents the ROC curve and the confusion matrix for the group prediction. After ROC tuning, sensitivity and specificity were .87 and .95, respectively. Area under the ROC curve was .95. To achieve this level of precision, the cut-point for VF derived from the logistic regression was shifted from .50 to .35 (ie, the probability of VF was made more liberal). This change in cut-point shifted subjects who would have otherwise been classified as false negatives to true positives. Before ROC tuning, sensitivity and specificity were .46 and .96, respectively.

Figure 2.

Receiver operating characteristic curve and confusion matrix for the logistic regression model given in Table 4. AUC, area under the curve.

Resistance Testing

Anti-retroviral drug resistance testing (Genotype and/or Phenotype [PhenoSense GT]) has been done routinely in our Pediatric HIV Clinic for years when the patient’s primary care physician believed that the patient was failing their HAART regimen. A review of the resistance testing performed among the 15 patients with VF while they were failing therapy found that only 3 developed resistance to 1 or more agents in their prior regimen. In the other 12 cases, no evidence was found for resistance as a cause of VF. Continued poor adherence with HAART was suspected as the primary contributing factor to VF in these 12 cases.

DISCUSSION

The goal of ART is to suppress HIV replication to a level below which drug-resistance mutations do not emerge [1, 8]. Although not conclusive, prior data suggests that selection of drug-resistance mutations does not occur in patients with HIV VL persistently suppressed to below the lower level of detection of current assays [9]. Isolated viral “blips” (as defined in the most recent DHHS guidelines) are not usually associated with subsequent viral failure [11]. However, there is controversy regarding the clinical implications of persistent HIV VL between the lower limit of detection and <200 copies/mL in patients on HAART [3–5].

The significance of LLV is particularly difficult to ascertain in a patient population such as ours where compliance with HAART can vary appreciably leaving the caretaker in a quandary, wondering whether they should be aggressive in altering the HAART regimen or wait to see what the effects are of intensive counseling regarding adherence. The ability to accurately identify true VF in perinatally infected children is crucial not only to prevent the development of drug resistance but also to avoid unnecessary switches to second-line therapy. This is important so that viable treatment options can continue to be available during adolescence and adulthood [12, 13].

Our clinical experience over several years with a large number of HAART-treated perinatally infected patients differs from adults in whom viral suppression is usually achieved within 6 months of ART [5]. We observed patients who maintained levels >200 but <1000 copies/mL for years without significant changes in their clinical status, and, for that reason, we decided for this study to adopt a more liberal definition of VF to see what were the variables that would contribute to overt VF.

In our population, the frequency of LLV was 42.4%, which is higher than that described in the adult literature (4.0% [5] and 28.6% [14]). The difference may be explained by the variation in definitions used in the cited studies versus the definitions adopted for this study. In addtion, most studies describe the incidence of the LLV during the first year after initiated HAART in adult naive patients, whereas our population was HAART experienced having been on treatment for at least 5 years [15]. Although somewhat counterintuitive, this observation (ie, higher LLV prevalence in youth) is not unexpected given the continuing problem with adherence in our young perinatally infected population.

We compared a number of variables in the VF versus the non-VF groups in an attempt to identify variables that might reliably predict VF in the presence of persistent LLV. The multivariate logistic regression indicated that the mean VL (log₁₀ copies/mL), number of patient visits, and the slope of the VL (log₁₀ copies/mL per day) were all found to be independently associated with VF. The significance of these findings suggest that for patient populations such as ours who experience persistent LLV <1000, there is a higher risk of VF if they have higher mean VL, fewer patient visits, and increasing slope of sequential VL. Demographic variables (age, sex, race, and HIV status) did not show a statistically significant difference between the 2 groups. Laboratory variables (mean absolute and percent CD4 count, maximum VL, and VB) also did no differ statistically between the 2 groups.

Although our results suggest an inverse association with the number of visits and total days observed, the mean time between visits was similar in both groups, showing that independently of their failure status they had similar time (days) between visits. Similarly, there was no difference in the mean number of visits per total days observed (visits/day). Those children who experienced VF in this analysis were before being censored more likely to be followed more closely and have more frequent visits. This was because of poor medication compliance and the need for them to be seen more frequently to emphasize ART adherence to prevent VF.

There is an increased awareness of the detrimental impact of elevated VLs above the limit of detection of current available assays, but there is a significant challenge in achieving persistent VL suppression as currently defined by the guidelines in perinatally HIV-infected, antiretroviral-experienced youth. Based upon a prior retrospective survey of our population, we found that although 47% of our population had a VL <400, only a small percentage (<15%) of our patient population were able to achieve and maintain persistent viral suppression below the limits of the testing assay (unpublished data, C. D. M., 2007). We believe that the majority of our clinic population experienced persistently detectable viremia because of episodic compliance. A brief review of the results of testing for the presence of specific ART resistance mutations in the 15 subjects who experienced VF bolstered this supposition because only 3 developed resistance to 1 or more agents in their prior regimen. The other 12 had not developed any resistance mutations to the ART agents they were proscribed.

Although our results suggest that there are certain temporal factors that may be useful in identifying patients at risk for VF, we failed to identify any specific demographic characteristics. Despite the fact that we have adopted several of the compliance strategies advocated in the DHHS guidelines, we postulate that ART compliance in adolescents is heavily dependent on “maturity”, which behaviorally is extremely difficult to alter. Therefore, in the case of adolescents, the next best strategy is the early detection of temporal trends that may signal eventual VF.

There were several limitations to this study. Beyond relying upon short-term verbal recall (ie, asking how many doses of medication had been missed during the preceding 2 weeks), we did not have a reliable means of assessing adherence, which precluded us from assessing compliance. Given the recurring nature of poor medication compliance in our population and similar experiences in other centers caring for HIV-affected youth, further research on adherence and how to reliably measure it and how to improve is increasingly becoming a major research priority. Another limitation pertains to the question as to how applicable are these findings to other populations of perinatally infected youth in other urban centers given the definitions of LLV and VF adopted for this study. Our intention was to try to identify some meaningful trends from the clinical data that we have been accumulating since 2008 and use it to identify patients who may experience subsequent VF.

CONCLUSIONS

In conclusion, 3 variables (the log₁₀ mean VL, log₁₀ slope of sequential VLs, and the number of visits) were found upon logistic regression analysis to be independently associated with VF in this study. A subsequent ROC analysis found that a model incorporating all 3 of these elements had a sensitivity of 0.87 and a specificity of 0.95, respectively, for identifying those patients who experienced VF. The results indicate that clinicians should keep a close eye on patients who tend to miss clinical appointments and show an elevated and increasing VL. Further investigations will be necessary to determine how useful these same parameters are for predicting VF in other, similar patient populations (ie, cross-validation).