Access to antiretroviral therapy (ART) in low and middle-income countries has dramatically improved in recent years. At the end of 2012, 9.7 million people living with HIV were receiving ART in low and middle-income countries . Monitoring individuals receiving ART is important to ensure successful treatment, identify adherence problems and individuals at high risk of poor treatment outcomes, and to determine whether antiretroviral regimens should be switched in case of treatment failure .
Clinical trials have shown that viral load monitoring is a more sensitive and early indicator of treatment failure than CD4+ monitoring [3–5]. In the United States , monitoring viral load is therefore recommended every 4–8 weeks until viral suppression is achieved, and every 3–4 months thereafter during the first 2 years of ART. In addition, CD4+ cell count is monitored every 3–6 months. The measurement of viral load requires expensive and sophisticated technologies that are not always feasible or affordable, particularly in resource-poor settings in which the burden of disease is the greatest. When viral load is not routinely available, the 2013 WHO antiretroviral guidelines recommend regular CD4+ cell count monitoring .
Despite highly specific and widely accepted definitions of viral failure, that is, viral load at least 1000 copies/ml (WHO) or at least 200 copies/ml (United States) on two consecutive viral load measurements 3 months apart [2,6], there is no consensus definition for suboptimal immunologic response. Studies have used multiple definitions, most commonly failure to increased CD4+ cell counts above a specific threshold over a specific period (e.g. >200 cells/μl by 6 months on ART), or an increase in CD4+ cell counts above baseline levels by a certain threshold over a given time period (e.g. <0, 0–49, 50–99 and ≥200 cells/μl by 6 months on ART) . The WHO defines immunological failure as a CD4+ cell count at or below baseline level, or persistence of CD4+ cell count below 100 cells/μl . All these definitions ignore the fact that changes in CD4+ cell count are not homogeneous across baseline levels of CD4+ cell count. Furthermore, studies assessing the predictive value of the WHO definition have shown low sensitivity and low positive predictive values for predicting virological failure, particularly in patients starting ART at higher (>350) CD4+ cell count levels .
Defining cut-points for indicators that are continuous in nature (e.g. CD4+ cell count) and for which changes are strongly correlated to values of another continuous indicator (e.g. time on ART) is not uncommon in medicine. One example is that of monitoring changes in weight among children while accounting for variation with age. This is usually addressed by standardizing the distribution of the indicator of interest (e.g. CD4+ cell count) by the continuous cofactor (e.g. time on ART) and using percentile distributions or cut-offs such as −2 SDs [9–14].
The aim of this study was to construct distributions of cumulative CD4+ cell gains from ART initiation to 4, 10, 16, 22, and 28 months post-ART initiation, standardized by baseline CD4+ cell count. We also assessed the correlation of lower centiles (3rd, 10th, 25th, 33rd, and 50th) of the distributions with subsequent mortality and virological failure, and the accuracy of those centiles for early identification of patient at risk of virological failure and death.
Data and data source
The data were collected from the Themba Lethu Clinic Cohort – an open cohort of HIV-infected adults (≥ 18 years) who receive ART at a public ART clinic in Johannesburg, South Africa [15,16]. At baseline, ART eligibility is assessed clinically and by CD4+ cell count. In line with the national guidelines, viral load is not determined at baseline. Patients initiated on ART are monitored by CD4+ cell count and viral load at 4 months after ART initiation and every 6 months thereafter, and at any other time if clinically needed. Patients who fail to achieve viral suppression by month 4 are scheduled for monthly visits until viral suppression is achieved. Patients who miss a clinic visit are actively traced by telephone. Loss to follow-up is defined as being 4 months late for the last scheduled clinic visit. Mortality is ascertained via family, hospital report, and linkage with the South African National Vital Registration Infrastructure Initiative, a system estimated to have 90% sensitivity for adults .
All nonpregnant adults with baseline CD4+ cell count measure below 500 cell/μl who initiated ART between 1 April 2004 and 1 April 2014, and had at least one follow-up CD4+ cell count result were included in the analysis.
Baseline CD4+ was defined as the closest CD4+ cell count measurement collected 6 months before to 7 days after ART initiation.
The change in CD4+ cell count at each time point was determined using the closest measure that was collected within ±6 weeks of the time points of interest (4, 10, 16, 22, and 28 months of ART) and calculated by subtracting the baseline value from the value at the time point of interest. Assuming a linear increase in CD4+ cell count over time within each time interval, we then estimated the CD4+ cell count gain at the time point of interest by dividing the observed change in CD4+ cell count by the real time interval in months and multiplying this by 4 for the 4-month time point or by 6 for all other time points.
Viral failure was defined as the first of two consecutive viral loads above 1000 RNA copies/ml after initial suppression or as not achieving viral suppression after at least 12 months on ART.
Construction of baseline-standardized distributions of CD4+ cell gain at each time point
To obtain the distribution of CD4+ gain at each time point of interest, methods similar to those used by the WHO to construct international growth curves in children  were adapted and used as described previously . The changes in CD4+ cell count for each time point were regressed on baseline values using the generalized additive model for location, scale, and shape, a method that requires a parametric distribution assumption for the response variable (CD4+ gain over time) while allowing the modeling of the distribution parameters as nonparametric (smooth) functions of the explanatory variable (baseline CD4+ cell count) . For CD4+ cell gain at each time point, we assumed a Box-Cox power exponential distribution with four parameters relating to location (μ, median); scale (σ, coefficient of variation); skewness (υ, transformation for symmetry); and kurtosis (τ, power exponential parameter), respectively . The number of degrees of freedom (DF) was determined by a step-down procedure in a training set which consisted of a random sub-sample of 60% of the total data set. We started with the simplest model that included baseline CD4+ cell count and the fitting of μ and σ curves while keeping DF(υ) and DF(τ) fixed at zero. We searched for the DF(μ) and DF(σ) that minimized the global deviance as indicated by the generalized Akaike Information Criterion (with penalty 3 for each DF used). In the next step, using the selected DF(μ) and DF(σ), we sequentially searched for the DF(υ) and DF(τ) that minimized the global deviance. In the last step, Q statistic  and worm plots  were used to fine tune the selected DF(μ), DF(σ), DF(υ), and DF(τ) . To facilitate the convergence of the models and to obtain smoother curves [12,18,22], extreme values of CD4+ cell count, that is, values that looked far apart on visual inspection were set to missing during the construction of the standardized distributions.
The performance of the fitted models was assessed using a validation set. The proportion of sample that fell below the selected centile was used to compare the fit of the training and validated model (see supplemental material, Table 6, http://links.lww.com/QAD/A673) .
Association of lower (3rd, 10th, 25th, 33rd, and 50th) centiles of CD4+ cell gain, the WHO criteria for immunological criteria, and current CD4+ cell count values at months 4, 10, 16, 22, and 28 on antiretroviral therapy with subsequent mortality and virological failure
For the outcome of death, person-time started at initiation of ART (baseline) and was censored at the earliest of recorded date of death or at the date of last visit to the clinic. For the outcome of virological failure, follow-up time started at ART initiation (baseline), with follow-up censored at the earliest of first viral load above 1000 RNA copies/ml, date of death, or date of last visit to the clinic. Cox proportional-hazard models were used to estimate the crude and adjusted hazard ratios and their 95% confidence intervals (CIs) for the association between the lower centiles of CD4+ cell count gain at each time point with each of the two outcomes considered. A Cox proportional model was also used to estimate the strength of the association between WHO definition for immunological failure and subsequent mortality and virological failure at each time point. The proportional hazard assumption was formally evaluated for all baseline covariates (BMI, WHO clinical stage, hemoglobin, CD4+ cell count, and age) using the Kolmogorov-type supremum test . Using a stepwise backward selection procedure and Wald test, all covariates that did not contribute significantly to the fit of each model were dropped. Analyses were done using SAS 9.3 (SAS Institute, Cary, North Carolina, USA). All tests were conducted using a two-sided 0.05 significance level, without correction for multiple comparisons.
Sensitivity, specificity, negative, and positive predictive values of the WHO definition for immunologic failure and CD4+ cell count gain percentile for subsequent mortality and virological failure
We used standard definitions  to compute the sensitivity, specificity, positive, and negative predictive values of the WHO definition for immunological failure and the 3rd, 10th, 25th, 33rd, and 50th percentiles of CD4+ cell count gain for predicting subsequent death and virological failure.
Characteristics of the cohort at antiretroviral therapy initiation
Of the 21 921 nonpregnant adults with baseline CD4+ cell count 500 cells/μl or less, initiating ART between 1 April 2004 and 1 April 2014, 1022 died, 745 were lost to follow-up or transferred out, and 391 were administratively censored before reaching 4 months of ART. Of the 19 980 completing ART for at least 4 months, 14 194 had at least one follow-up CD4+ cell count collected by the seventh month of ART, including 9640 at 4 months ± 6 weeks after ART initiation. A CD4+ cell count result was available at 10, 16, 22, and 28 months ± 6 weeks of ART for 7406, 5577, 4804, and 3446 individuals, respectively.
Among the 9640 included in the analysis of 4 months CD4+ gain, 40.6% were men, median age was 37.2 [interquartile range (IQR) 32.0–43.5] years, 39.4% were in WHO clinical stage III or IV, and almost half (49.7%) had a baseline CD4+ cell count below 100 cells/μl (Table 1). Baseline characteristics of those excluded did not differ substantially from those included in the analysis (Table 4, supplemental material, http://links.lww.com/QAD/A673).
Cumulative CD4+ gains at months 4, 10, 16, 22, and 28 months of antiretroviral therapy and their baseline-standardized distributions
The median cumulative gain in CD4+ cell count increased from 4 to 28 months on ART, with the greatest gain in patients with lowest baseline CD4+ cell counts: from 118 and 36 cells/μl at 4 months to 278 and 148 cells/μl at 28 months for those with baseline CD4+ cell count of 50 and 450 cells/μl, respectively (Fig. 1).
Association between the 3rd, 10th, 25th, 33rd, and 50th percentile of cumulative CD4+ gain and subsequent time to death
Among the 9640 patients initiated on ART, who were retained in the cohort at 4 months of follow-up and had both baseline and the 4-month CD4+ cell count measurements, 843 (8.7%) died after 4 months of ART, with most (48.3%) deaths occurring early (between 4 and 10 months), 436 deaths occurring after 10 months, 244 after 16 months, 203 after 22 months, and 131 after 28 months of ART. Among patients who died, about half had a CD4+ gain below the 33rd percentile: 46.5% at 4 months, 49.8% at 10 months, 50.0% at 16 months, 53.7% at 22 months, and 49.6% at 28 months. Low CD4+ cell count gain at any time point was associated with subsequent death, with strength of association increasing with decreasing centiles of CD4+ gain and increasing length of follow-up (Table 2). For example, the hazard ratios (95% CIs) for subsequent death among patients whose CD4+ gain at 4, 10, 16, 22, and 28 months was below the 3rd percentile, were 2.72 (2.03, 3.65), 3.65 (2.50, 5.33), 4.24 (2.72, 6.61), 4.89 (2.98, 7.96), and 5.73 (3.26, 10.06), respectively.
Association between the 3rd, 10th, 25th, 33rd, and 50th percentile of cumulative CD4+ gain and subsequent time to virological failure
In total, 1101 (11.4%) patients experienced virological failure after 4 months, 812 after 10 months, 662 after 16 months, 633 after 22 months, and 500 after 28 months of ART. Of patients who experienced virological failure, 30.5, 36.1, 43.8, 48.9, and 50.4% had a CD4+ gain below the 33rd percentile at 4, 10, 16, 22, and 28 months, respectively. As with mortality, lower percentiles of CD4+ gain at any time point were associated with subsequent virological failure, with the strength of association increasing with decreasing centiles and increasing time on ART (Table 2). For example, the hazard ratios (95% CIs) for subsequent virological failure for patients whose CD4+ gain at 4, 10, 16, 22, and 28 months was below the 3rd percentile, for instance, were 1.48 (1.01, 2.15), 2.69 (1.91, 3.78), 4.00 (2.91, 5.50), 6.93 (5.17, 9.30), and 4.49 (3.14, 6.43), respectively.
Association between WHO-defined immunologic failure and subsequent mortality and virological failure
The proportion of participants meeting the WHO criteria for immunological failure decreased with time on ART, from 19.8% at 4 months, to 9.8% at 10 months, 6.7% at 16 months, 5.2% at 22 months, and 4.1% at 28 months. There was a U-shaped relationship between the WHO-defined immunologic failure and baseline CD4+ levels, with higher proportion of patients with lowest and highest baseline CD4+ cell count being identified as immunological failure, particularly at earlier follow-up time points (Fig. 2). The strength of the association between WHO-defined immunological failure and subsequent mortality or virological failure increased with increasing time on ART (Table 2). The adjusted hazard ratio (aHR) (95% CI) for subsequent virological failure was 0.84 (0.71, 1.00) at 4 months and increased to 1.78 (1.44, 2.20) at 10 months, 3.30 (2.59, 4.21) at 16 months, 5.54 (4.29, 7.15) at 22 months, and 4.78 (3.46, 6.61) at 28 months. Similarly, the aHR (95% CI) for subsequent death increased from 1.74 (1.48, 2.04) at 4 months to 5.30 (3.07, 9.16) at 28 months.
Association between current CD4+ cell count and subsequent mortality and virological failure
The median (IQR) current CD4+ cell count at 4, 10, 16, 22, and 28 months was 211 (135, 302), 256 (177, 356), 307 (218, 416), 340 (244, 460), and 370 (266, 493) cells/μl, respectively. The aHR (95% CI) for subsequent death for every 50 cells/μl decrease in the current CD4+ cell count was similar at all time points, ranging between 1.16 (1.12, 1.20) at 4 months and 1.23 (1.15, 1.32) at 28 months (Table 2). The aHR (95% CI) for subsequent virological failure ranged from 0.97 (0.94, 1.00) at 4 months to 1.16 (1.12, 1.20) at 28 months.
Sensitivity, specificity, and predictive values of lower centiles of CD4+ gain and WHO-defined immunological failure as early indicators of virological failure and death
As expected, sensitivity decreased and specificity increased with lower CD4+ gain centiles (Table 3). The negative predictive values (NPVs) for mortality were high (>90%) at each time point and for all CD4+ gain centiles below 50. The positive predictive value (PPV) was low and decreased with increasing time on ART. The NPV for virological failure was also high (>83%) for any centiles, but was constant over time. The PPV for virological failure tended to increase with decreasing centiles and with increasing time on ART. The PPV for virological failure was higher than that for mortality, especially at later time points and lower centiles. The NPV of the WHO definition for immunological failure was also high – above 90% for death and above 85% for virological failure. Similarly to low CD4+ gain centiles, the PPV was low for death (<15%) at all time points. The PPV of the WHO definition for immunological failure for subsequent virological failure was lower than that of the third percentile of the CD4+ gain distribution, especially early after ART initiation.
Using data from a large cohort of patients receiving care at an ART clinic in South Africa, we constructed percentile curves for CD4+ cell count gain on ART, standardized by baseline CD4+ cell count. Associations with subsequent death and virological failure showed an increasing strength of association with lower centiles and longer time on ART. The use of CD4+ gain centiles allowed a graded risk assessment of predicted death and virological failure compared to the binary yes/no risk profile associated with the WHO definition for immunological failure, and demonstrated stronger associations with subsequent death and virological failure compared to current CD4+ cell count. The PPV, however, remained low, especially for risk of death.
Although virologic monitoring is the gold standard for ART monitoring, randomized control trials [5,26,27] found routine monitoring of CD4+ cell count to perform similar to virologic monitoring in terms of prediction of disease progression and mortality. The WHO based the preference of viral load over CD4+ cell count monitoring on the poor accuracy of routine CD4+ cell count monitoring for early detection of treatment failure when using the WHO definition for immunological failure . As we demonstrated in this analysis, the poor accuracy may, at least in part, be attributable to the high proportion of patients at the extremes of baseline CD4+ cell count that are misclassified as immunological failure when using the WHO criteria to define immunological failure (Fig. 2). The proposed CD4+ gain percentiles standardized to baseline CD4+ cell count correct this problem and slightly improve the accuracy for early detection of treatment failure. While the use of CD4+ cell count gain percentile curve as an indicator of subsequent risk of treatment failure still suffered from low PPV, especially in the first year of treatment, the CD4+ centile curves offer healthcare workers a tool to grade the risk of subsequent death and virological failure in their patients. As the CD4+ gain drops from the 50th to the 3rd centile, the risk of subsequent death increases from a 1.6 to 2.7-fold hazard at 4 months of ART, and from a 1.9 to 5.7-fold hazard at 28 months of ART. Similarly, as CD4+ cell count gain drops from the 50th to the 3rd centile, the hazard of subsequent virological failure increases from a 1.1 to 2.7-fold hazard at 10 months of ART, and from a 1.8 to 5.6-fold hazard at 22 months of ART. This insight in increasing risk could allow heathcare workers to target interventions such as more frequent clinical follow-up, intensified adherence counselling, and intensified screening for opportunistic infections to those patients at highest risk or patients with increasing risk profiles over time.
Our analysis has many strengths, including a large sample allowing the use of a training and validation dataset, long follow-up (median follow-up time on ART of 54 months), and regular availability of viral load measurement. The exclusion of patients initiating ART due to lack of regular CD4+ cell counts, transfer out or loss to follow-up is a limitation, but the bias introduced may be minimal as baseline characteristics of those excluded from the analysis did not differ from those included. The restriction to patients receiving ART at a single center is another limitation that may limit generalizability. Although the baseline characteristics of patients in our cohort were similar to patients receiving ART in the region , and the median of current CD4+ cell count at each time point was similar to that reported from an analysis of data from 22 sites in Africa, 2 in Asia, and 3 in Latin America , confirmation of the results using data from other facilities in the region or multiple regions would strengthen our findings.
In conclusion, while viral load is the gold standard for monitoring the response to ART,  viral load assays are not yet routinely available in most settings with high burden of HIV. In those settings, the WHO recommends that CD4+ cell count be used. The percentile curves for monitoring CD4+ gain may be a simple tool enabling healthcare workers in resource-limited settings to identify patients at increased risk of virological failure and death, and target interventions at those with highest or increasing risk.
We acknowledge all patients and their families followed at TLC. We also thank the staff of the TLC for their dedication.
Contributors: M.Y. and A.V.R. designed the study. M.M. helped with data collection. M.Y. analyzed the data and wrote the first draft of the report. All authors contributed to the interpretation of the data and read and approved the final manuscript.
M.M. was also supported by Cooperative Agreement AID 674-A-12–00029 from the United States Agency for International Development (USAID). The contents are the responsibility of the authors and do not necessarily reflect the views of USAID or the United States Government. M.Y. is partially supported by a grant from NICHD: R01HD075171 and another from NIAID: U01AI096299. No funding bodies had any role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.
Conflicts of interest
The authors declare no conflicts of interest.
1. UNAIDS/WHO. Global update on HIV treatment 2013: results, impact and opportunities
. Geneva, Switzerland: World Health Organization; 2013. p. 126.
2. WHO. Consolidated guidelines on the use of antiretroviral drugs for treating and preventing HIV infection: recommendations for a public health approach
. Geneva, Switzerland: World Health Organization; 2013. p. 272.
3. Mermin J, Ekwaru JP, Were W, Degerman R, Bunnell R, Kaharuza F, et al. Utility of routine viral load, CD4 cell count, and clinical monitoring among adults with HIV receiving antiretroviral therapy in Uganda: randomised trial
. Br Med J
4. Kekitiinwa A, Cook A, Nathoo K, Mugyenyi P, Nahirya-Ntege P, Bakeera-Kitaka S, et al. Routine versus clinically driven laboratory monitoring and first-line antiretroviral therapy strategies in African children with HIV (ARROW): a 5-year open-label randomised factorial trial
5. Mugyenyi P, Walker AS, Hakim J, Munderi P, Gibb DM, Kityo C, et al. Routine versus clinically driven laboratory monitoring of HIV antiretroviral therapy in Africa (DART): a randomised noninferiority trial
6. Panel on Antiretroviral Guidelines for Adults and Adolescents. Guidelines for use of antiretroviral agents in HIV-1 infected adults and adolescents
. Department of Health and Human Services; 2013. p. 285.
7. Takuva S, Maskew M, Brennan AT, Long L, Sanne I, Fox MP. Poor CD4 recovery and risk of subsequent progression to AIDS or death despite viral suppression in a South African cohort
. J Int AIDS Soc
8. Rutherford GW, Anglemyer A, Easterbrook PJ, Horvath T, Vitoria M, Penazzato M, Doherty MC Predicting treatment failure (TF) in adults and children on antiretroviral therapy (ART): systematic review of the performance characteristics of the 2010 World Health Organization (WHO) criteria for virologic failure
. In 7th IAS Conference on HIV Pathogenesis, Treatment and Prevention. Kuala Lumpur, Malaysia; 2013.
9. Baumgartner RN, Roche AF, Himes JH. Incremental growth tables: supplementary to previously published charts
. Am J Clin Nutr
10. Carey VJ, Yong FH, Frenkel LM, McKinney RE Jr. Pediatric AIDS prognosis using somatic growth velocity
11. WHO Multicentre Growth Reference Study Group. In Press W, editor. WHO child growth standards: length/height-for-age, weight-for-age, weight-for-length, weight-for-height, and body mass index-for-age: methods and development
. Geneva, Switzerland: World Health Organization; 2006.
12. WHO Multicentre Growth Reference Study Group. WHO child growth standards: growth velocity based on weight, length and head circumference: methods and development
. Geneva, Switzerland: World Health Organization; 2009. p. 242.
13. Yotebieng M, Meyers T, Behets F, Davies MA, Keiser O, Ngonyani KZ, et al. Age-specific and sex-specific weight gain norms to monitor antiretroviral therapy in children in low-income and middle-income countries
14. Yotebieng M, Van Rie A, Moultrie H, Meyers T. Six-month gain in weight, height, and CD4 predict subsequent antiretroviral treatment responses in HIV-infected South African children
15. Fox MP, Maskew M, MacPhail AP, Long L, Brennan AT, Westreich D, et al. Cohort profile: the Themba Lethu clinical cohort, Johannesburg, South Africa
. Int J Epidemiol
16. Fox MP, Shearer K, Maskew M, Macleod W, Majuba P, Macphail P, Sanne I. Treatment outcomes after 7 years of public-sector HIV treatment
17. Fox MP, Brennan A, Maskew M, MacPhail P, Sanne I. Using vital registration data to update mortality among patients lost to follow-up from ART programmes: evidence from the Themba Lethu clinic, South Africa
. Trop Med Int Health
18. Rigby RA, Stasinopoulos DM. Generalized additive models for location, scale and shape
. J Roy Stat Soc C
19. Rigby RA, Stasinopoulos DM. Smooth centile curves for skew and kurtotic data modelled using the Box-Cox power exponential distribution
. Stat Med
20. Royston P, Wright EM. Goodness-of-fit statistics for age-specific reference intervals
. Stat Med
21. van Buuren S, Fredriks M. Worm plot: a simple diagnostic device for modelling growth reference curves
. Stat Med
22. Stasinopoulos D, Rigby R, Akantziliotou C. Instructions on how to use the gamlss package
. 2nd ed. 2008.
23. Carey VJ, Yong FH, Frenkel LM, McKinney RM. Growth velocity assessment in paediatric AIDS: smoothing, penalized quantile regression and the definition of growth failure
. Stat Med
24. Lin DY, Wei LJ, Ying Z. Checking the Cox model with cumulative sums of martingale-based residuals
25. Lalkhen AG, McCluskey A. Clinical tests: sensitivity and specificity
. Cont Educ Anaesth Crit Care Pain
26. Mermin J, Ekwaru JP, Were W, Degerman R, Bunnell R, Kaharuza F, et al. Utility of routine viral load, CD4 cell count, and clinical monitoring among adults with HIV receiving antiretroviral therapy in Uganda: randomised trial
27. Jourdain G, Le Coeur S, Ngo-Giang-Huong N, Traisathit P, Cressey TR, Fregonese F, et al. Switching HIV treatment in adults based on CD4 count versus viral load monitoring: a randomized, noninferiority trial in Thailand
. PLoS Med
28. World Health Organization. Consolidated guidelines on the use of antiretroviral drugs for treating and preventing HIV infection: recommendations for a public health approach
. Geneva, Switzerland: World Health Organization; 2013.
29. Nash D, Katyal M, Brinkhof MW, Keiser O, May M, Hughes R, et al. Long-term immunologic response to antiretroviral therapy in low-income countries: a collaborative analysis of prospective studies