Routine CD4 count and HIV viral load monitoring is a financial barrier in developing countries.
Routine CD4 count and HIV viral load monitoring is a financial barrier in developing countries.
We assessed factors associated with CD4 counts ≤200 cells/μL and detectable viral load in Thai HIV-infected patients receiving antiretroviral therapy (ART) at the HIV Netherlands Australia Thailand Research Collaboration and the Thai Red Cross AIDS Research Centre (HIV-NAT). Univariate and multivariate Cox proportional hazards models for multiple treatment failures were used to determine factors related to CD4 counts ≤200 cells/μL and detectable viral load. Multivariate Cox proportional hazards models for CD4 counts ≤200 cells/μL were developed with and without viral load in order to build models applicable to contexts in which viral load is not available.
Four hundred and seventeen patients were included in the study. Fifty-four per cent were male, and the median CD4 count and log10 viral load at baseline were 283 cells/μL and 4.3 log10 HIV-1 RNA copies/mL, respectively. Independent factors related to CD4 count ≤200 cells/μL were CD4 count at baseline [hazards ratio (HR) 0.20/100 cells/μL; 95% confidence interval (CI) 0.17–0.23] and changes in CD4 count (HR 0.22/100 cells/μL; 95% CI 0.17–0.28). Factors in multivariate models (in which viral load was considered for inclusion) were CD4 count at baseline (HR 0.21/100 cells/μL; 95% CI 0.18–0.24), changes in CD4 count (HR 0.25/100 cells/μL; 95% CI 0.19–0.32) and detectable viral load (HR 1.94; 95% CI 1.20–3.13). Predictive factors (independent of viral load) were triple ART or highly active antiretroviral therapy (HAART) (HR 0.28; 95% CI 0.22–0.36) and detectable viral load at baseline (HR 2.96; 95% CI 2.24–3.91).
CD4 count at baseline and changes in CD4 count were important in predicting CD4 counts ≤200 cells/μL. Triple ART and detectable viral load at baseline were important in predicting detectable viral load.
The widespread availability of effective combination antiretroviral therapies to treat HIV infection has led to large reductions in HIV-associated morbidity and mortality in developed countries [1–6]. However, in developing countries, the high cost of antiretroviral therapies has limited their availability. Recently, the costs of combination antiretroviral treatment have decreased markedly as a result of technological developments and generic competition . Cheap generic drugs have been developed as a result of the World Trade Organization (WTO) loosening the rules on pharmaceutical intellectual property for developing countries. Consequently, non-government organizations such as Doctors Without Borders (MSF) and Oxfam sought permission from universities in the USA to use generic drugs in developing countries . Canada also amended its legislation of patent act so that manufacturers could obtain licences to make generic drugs and to export them to developing countries [9–13]. Coincident with these developments, The Joint United Nations Programme on HIV/AIDS (UNAIDS) and the World Health Organization (WHO) launched the ‘Treat 3 Million by 2005’ (3 by 5) programme in 2003 to provide antiretroviral treatment to 3 million people infected with HIV in developing countries who were in need. This programme encouraged developing countries, especially 34 countries with high HIV infection prevalences, to cooperate with WHO, UNAIDS and other partners in providing financial and other assistance to enable people with HIV infection to obtain antiretroviral treatment . As a result of innovation and price reductions in the generic drug production industry, HIV-infected patients in some developing countries can access effective combination antiretroviral regimens at much lower cost. For example, patients in Thailand can now access a generic, fixed dose combination of stavudine, lamivudine and nevirapine at a cost of around US$1 a day.
While the costs of antiretroviral treatments have decreased markedly, the costs of the CD4 count and HIV viral load tests that are used to monitor a patient's response to treatment have remained relatively high in developing countries. In Thailand again, for example, the cost of CD4 count and HIV viral load tests are about US$10 and US$75 per person per test (Urichit [BU, ICON Clinical Research Ptd. Ltd., Bangkok, Thailand] and Thongbor [PT, PPD Development (Thailand), Bangkok, Thailand], personal communication). As a typical patient-monitoring scheme would repeat these tests every 3 months [15,16], the costs of the monitoring tests themselves have now become a bigger financial barrier than the costs of the antiretroviral treatments. Furthermore, in some developing countries, particularly in remote regions, the technology required to perform CD4 count and HIV viral load tests may be difficult to access. The development of simplified monitoring schemes, whereby some HIV-infected patients receiving antiretroviral treatments would have fewer CD4 count and HIV viral load tests, could aid antiretroviral treatment availability, if such schemes could be developed without compromising treatment effectiveness and patient outcomes.
In this report we assess factors that predict immunological and virological failure in HIV-infected patients in Thailand receiving antiretroviral treatments. This is the first step in developing statistical models that could be used to identify HIV-infected patients receiving antiretroviral treatments who are at low risk of immunological or virological failure, with the idea that a simplified monitoring scheme could reduce or delay monitoring tests in these patients. On the basis of clinical experience and the results of cohort studies [17–21] we hypothesized that the strongest predictors of immunological and virological success would be the baseline CD4 count and HIV viral load.
Analyses were based on 417 patients enrolled in HIV Netherlands Australia and Thailand Research Collaboration Thai Red Cross AIDS Research Centre (HIV-NAT) clinical trials. HIV-NAT clinical trials in the current database were HIV-NAT001, HIV-NAT002, HIV-NAT003, HIV-NAT005 and HIV-NAT009, with follow up between October 1996 and February 2004.
Details of these HIV-NAT trials are given elsewhere [22–32]; briefly, patients recruited into HIV-NAT001, 002 and 003 were naïve to antiretroviral therapy, while patients in HIV-NAT005 had at least 3 months prior treatment with zidovudine with no previous zidovudine intolerance before receiving new treatments. Patients recruited into HIV-NAT009 had to have fulfilled either of the following two criteria: (1) HIV viral load >5000 HIV-1 RNA copies/mL on combination nucleoside reverse transcriptase inhibitor (NRTI) therapy or (2) documented HIV viral load >1000 copies/mL on combination NRTI therapy, in addition to one or more of the following: (1) CD4 count <baseline CD4 count at commencement of initial combination NRTI therapy, (2) CD4 count <200 cells/μL or (3) clinical treatment failure, with progression to AIDS or the presence of at least two Centers for Disease Control and Prevention (CDC) category B symptoms after a minimum of 3 months on combination NRTI therapy. Once patients were enrolled into HIV-NAT studies most of them were subsequently re-enrolled into a series of roll-over studies, for example 001.1, 001.2, 001.3 and so on. The time-lines of these trials are listed in Duncombe et al. .
Patients were included in the analyses presented here if they had initiated antiretroviral treatment, and had had CD4 count and viral load assessed for at least 12 weeks. The 12-weekly visit scheme was chosen in this study in order to reflect routine clinical practice. Patients were monitored at least once every 3 months in HIV-NAT clinical trials; patients stopping antiretroviral treatment before week 12 were not included in this study. The lower limit of detection of the virological assays varied amongst the HIV-NAT trials (500, 400 and 50 copies/mL). In order to be able to compare across all trials we adopted virological failure at a limit of 500 copies/mL. Baseline detectable viral load was considered for inclusion in model fittings because of protocol changes and because patients could change to a new treatment while their viral loads were still undetectable.
Endpoints for this study were as follows.
Covariates in the analyses included the following.
Since the eventual goal of these analyses was to assess CD4 count and viral load outcomes (monitored using a typical 12-weekly schedule, following antiretroviral treatment), a 12-weekly visit scheme was chosen. CD4 count and viral load were usually measured at least once every 12 weeks for all patients participating in HIV-NAT clinical trials during the study period. For each patient, the CD4 count or viral load measurement closest to the 12-weekly nominal visit time was chosen as the value for that patient at that time-point. If a patient did not have a measurement within±6 weeks of a nominal visit time, that patient's CD4 count or viral load was assumed to be missing at that visit for the purpose of analysis.
HIV-NAT patients could change treatments when they rolled over to other HIV-NAT studies or when they failed treatment. In this study, every patient starting a new treatment was regarded as having a new baseline at commencement of new treatment. Therefore, patients could have more than one failure for each endpoint; however, they could have only one failure per treatment combination. For each endpoint and for each antiretroviral treatment combination, the first failure was selected, and subsequent failures on the same antiretroviral treatment combination were excluded.
Univariate and multivariate Cox proportional hazards models for multiple treatment failures were used to determine factors related to CD4 count ≤200 cells/μL and detectable viral load. In order to build models applicable to situations in which viral load is not available, multivariate Cox proportional hazards models were developed with and without the inclusion of viral load. Multivariate models were built using forward stepwise methods. After fitting univariate models, significant variables from univariate models were then considered for inclusion in multivariate Cox proportional hazards models. Incidence rates per 100 person-years were also stratified by gender, current antiretroviral treatment, detectable viral load at baseline and as a time-dependent variable, and clinical disease stage.
Baseline variables (i.e. current antiretroviral treatment, baseline CD4 count, detectable viral load at baseline, and CDC disease stage at baseline) were updated for each patient at the time at which they commenced a new antiretroviral combination. Changes in CD4 count and detectable viral load were also included as time-dependent variables. All statistical analyses were performed using stata StataCorp, College Station, TX, USA 8.0 software, and all P-values were two-sided. A level of significance of 0.05 was used throughout these analyses.
The HIV-NAT database contained follow-up data for 417 patients. Demographic factors are summarized in Table 2 (see below). The gender distribution was fairly even, with 54% of patients being male. Eighty-nine per cent of patients had heterosexual HIV risk behaviour, while 7% reported having male-to-male sex, and 2% reported both male-to-male sex and heterosexual risk behaviour. Seventy-one per cent of patients were naïve to antiretroviral therapy at entry to the cohort. The mean age was 32.2 years [standard deviation (SD) 7.2 years]. The majority of patients were classified as CDC class A (54%). The median CD4 count at first treatment was 283 cells/μL [interquartile range (IQR) 179–392 cells/μL] and the median log10 viral load was 4.3 log10 copies/mL (IQR 3.7–4.9 log10 copies/mL) (Table 1).
|Predictors||No. of cases||Incidence rate|
(per 100 person-years)
|Univariate analysis||Multivariate analysis with CD4|
|Multivariate analysis with CD4|
count and viral load
|HR (95% CI)||P||HR (95% CI)||P||HR (95% CI)||P|
|Age, per 10 years greater||–||–||1.39 (1.14–1.69)||0.001||1.06 (0.87–1.30)||0.568||1.13 (0.92–1.38)||0.247|
|Female||59||9.63 (7.46–12.43)||0.45 (0.31–0.64)||<0.001||0.76 (0.56–1.04)||0.086||0.76 (0.56–1.02)||0.071|
|Triple/HAART||115||17.69 (14.74–21.24)||1.50 (1.10–2.05)||0.011||0.98 (0.67–1.43)||0.920||1.19 (0.80–1.78)||0.384|
|CD4 count at baseline per 100 cells/μL||–||–||0.30 (0.27–0.34)||<0.001||0.20 (0.17–0.23)||<0.001||0.21 (0.18–0.24)||<0.001|
|Changes in CD4 count per 100 cells/μL||–||–||0.67 (0.56–0.77)||<0.001||0.22 (0.17–0.28)||<0.001||0.25 (0.19–0.32)||<0.001|
|Viral load detectable at baseline|
|Detectable||162||23.85 (20.45–27.82)||4.72 (3.05–7.29)||<0.001||1.04 (0.56–1.92)||0.908|
|Viral load detectable|
|Detectable||152||38.45 (32.80–45.08)||6.08 (4.00–9.24)||<0.001||1.94 (1.20–3.13)||0.007|
|CDC A||57||9.35 (7.21–12.12)||1.00||–||1.00||–||1.00||–|
|CDC B||86||16.95 (13.72–20.95)||1.92 (1.32–2.78)||0.001||1.07 (0.76–1.51)||0.686||1.03 (0.73–1.46)||0.863|
|CDC C||40||74.71 (54.80–101.85)||6.83 (4.35–10.72)||<0.001||1.34 (0.93–1.94)||0.120||1.28 (0.88–1.86)||0.192|
|Male/female [n (%)]||224/193 (53.7/46.3)|
|Mean age (SD) (years)||32.2 (7.2)|
|HIV risk factors [n (%)]|
|Blood transfusion||1 (0.2)|
|Heterosexual and homosexual||10 (2.4)|
|Prior HIV-related illnesses [n (%)]|
|CDC A/B/C||225/158/34 (53.8/38.0/8.2)|
|Naïve/experienced [n (%)]||295/122 (70.7/29.3)|
|Median CD4 cell count (IQR) (cells/μL)||283 (179–392)|
|CD4 category [n (%)]|
|>200 cells/μL||290 (69.54)|
|≤200 cells/μL||127 (30.46)|
|Median log viral load (IQR) (log10 copies/mL)||4.3 (3.7–4.9)|
|Log viral load category|
|>4 log10 copies/mL||263 (63.07)|
|≤4 log10 copies/mL||154 (36.93)|
One hundred and eighty-three treatment failures were observed for this endpoint. Time to CD4 count <200 cells/μL is shown in Fig. 1. In univariate models, the significantly related factors were age per 10 years, gender, treatment, CD4 at baseline per 100 cells/μL, changes in CD4 count per 100 cells/μL, detectable viral load at baseline, and CDC events (Table 2).
In the multivariate Cox proportional hazards model not including detectable viral load, the following were found to be significantly associated factors: CD4 count at baseline per 100 cells/μL and changes in CD4 count per 100 cells/μL (Table 2).
In the multivariate Cox proportional hazards model including both CD4 count and detectable viral load, the following were found to be significantly associated factors: CD4 count at baseline per 100 cells/μL, changes in CD4 count per 100 cells/μL and detectable viral load (Table 2).
Four hundred and five patients with 409 events were observed for this endpoint. Factors that were significantly related to detectable viral load in univariate models were the following: age per 10 years, treatment, CD4 count at baseline per 100 cells/μL and detectable viral load at baseline. Time to detectable viral load is shown in Fig. 2.
In the multivariate Cox proportional hazards model the statistically significantly associated variables were the following: currently on triple ART and detectable HIV load at baseline (Table 3).
|Predictors||No. of cases||Incidence rate|
(per 100 person-years)
|Univariate analysis||Multivariate analysis|
|HR (95% CI)||P||HR (95% CI)||P|
|Age, per 10 years greater||–||–||0.77 (0.65–0.91)||0.002||0.88 (0.76–1.02)||0.097|
|Female||185||38.88 (33.66–44.90)||1.04 (0.84–1.30)||0.713||1.07 (0.87–1.33)||0.516|
|Triple/HAART||129||18.78 (15.81–22.32)||0.26 (0.20–0.33)||<0.001||0.33 (0.25–0.42)||<0.001|
|CD4 count at baseline per 100 cells/μL||–||–||0.92 (0.87–0.96)||0.001||0.96 (0.90–1.03)||0.301|
|Changes in CD4 count per 100 cells/μL||–||–||0.93 (0.86–1.02)||0.116||0.96 (0.88–1.05)||0.372|
|Viral load detectable at baseline|
|Detectable||319||54.81 (49.11–61.17)||3.55 (2.66–4.73)||<0.001||2.50 (1.85–3.39)||<0.001|
|CDC A||189||39.06 (33.87–45.04)||1.00||–||1.00||–|
|CDC B||165||38.48 (33.04–44.83)||1.03 (0.83–1.29)||0.763||1.20 (0.97–1.49)||0.093|
|CDC C||27||29.62 (20.31–43.19)||0.83 (0.50–1.39)||0.482||1.22 (0.73–2.03)||0.446|
In this study we investigated the predictive factors for a number of immunological and virological failure endpoints. Two models for CD4 count ≤200 cells/μL were developed both with and without detectable viral load considered for inclusion. We found that both the CD4 count at baseline and the change in CD4 count were predictors for CD4 count <200 cells/μL. Further, we found that an undetectable viral load at baseline and the use of triple ART consistently predicted virological endpoints. These results were consistent with our expectations (i.e. CD4 count at baseline would be the strongest predictive variable for CD4 count <200 cells/μL outcome and a detectable HIV load at baseline would be the strongest predictive variable for detectable viral load outcome).
Results for CD4 count <200 cells/μL in this study demonstrated that patients with higher baseline CD4 counts were unlikely to fail at these endpoints. This finding is similar to those of Arici et al. and Florence et al., whose studies showed that patients with a high CD4 count at baseline were likely to do better from an immunological point of view [17,21]. The reason for this is likely to be that patients with high CD4 counts usually maintain their CD4 level well above the level of 200 cells/μL while on ART.
In predicting a detectable viral load, previous studies have obtained similar results to ours, with viral load at baseline a predictor of future detectable viral load, as one might expect. Patients with a higher viral load at baseline tended to experience a detectable viral load, while patients with a lower viral load maintained this lower level [35–46].
Our results did not show a clear link between immunological and virological failures. For both CD4 count variables studied we did not find that either predicted virological failure. This is consistent with the results of other studies [44,45,47], which also did not find a definite relationship between immunological and virological failures. However, these findings are in conflict with the findings of other studies which did demonstrate a relationship between immunological and virological failures in the setting of combination ART [19,20,35,37–41,48–50]. It may be that we failed to detect a relationship because of a lack of power in our analyses, which are based on a relatively small sample size. Triple ART was not statistically significant in predicting a drop in CD4 count below 200 cells/μL. This may reflect a lack of power to detect the influence of triple ART on these endpoints or there might not be a strict relationship between triple ART and CD4 cell count. A similar trend was found in the study of Martin et al., who observed a plateau in the proportion of subjects with CD4 counts <200 cells/μL after initiation of triple ART, which was maintained at around 10% over time . Most clinical research findings also suggest that the number of drugs used has an impact on the likelihood of an immunological or virological failure [35,38,40–43,48].
The gender ratio of male and female patients in the HIV-NAT cohort is unlike that described in cohorts in Europe and the USA which tend to consist predominantly of male patients. After treatment initiation, male and female patients in our cohort had similar levels of risk for reaching the endpoints studied. Our results were similar to results from other studies which showed that men and women similarly responded to highly active antiretroviral therapy (HAART) [52–56].
We analysed multiple combination treatments that patients received, rather than just the first or second treatment regimen. This approach maximized the power of the analysis and allowed us to perform analyses that might be broadly generalizable. All analyses in this study were based on findings in Thai patients, and, while there may be subtle differences from other Asian populations, the results are likely to be a better reflection of what occurs in Asian populations than studies in developed countries performed to date.
Data in this study were collected from 1996 onwards. Reflecting the standard treatment of patients with HIV/AIDS in Thailand at that time, physicians frequently treated patients with mono and dual therapy. Dual ART regimens included in this study are not relevant to current practice but we added them to our analyses to improve the statistical power. Analyses based only on triple ART gave very similar results to those for double-ART regimens. The CD4 count at baseline and changes in CD4 count remained predictive factors for every immunological endpoint studied. However, viral load, which significantly predicted CD4 count <200 cells/μL, was not statistically significant when patients only on triple ART were analysed. The presence of a detectable viral load at baseline was an important and independent predictive factor for all detectable viral loads studied (results not shown).
Having to undergo blood tests for CD4 count and viral load every 3 months is a financial burden for HIV-infected patients receiving ART, especially in developing countries. In Thailand, for example, most HIV-infected patients can barely (if at all) fund their own ART, and would be highly unlikely to be able to afford the additional costs associated with routine monitoring (US$85 per time for their CD4 count and viral load tests amounts to an annual sum of US$340 a year, which compares with an average income per capita of $2578 in Thailand ). Most patients in Thailand have family to support and might choose to drop out of therapy and routine follow up once they comprehend the considerable financial burden that the monitoring of their treatment would represent.
Both of these models (CD4 count ≤200 cells/μL and detectable viral load) have been constructed with a view to the construction of parametric models in the future, which may be used to predict failure for each endpoint. If good predictive models can be developed from this data set then it may mean that those patients with a good prognosis could be followed up less frequently and would require less intensive and therefore cheaper monitoring without compromising outcome. This would represent a cost saving not only to the individual patient but to the health system in general, thereby increasing the capacity of any given system to enrol and monitor more patients and allow health care personnel (and the health budget) to concentrate resources on patients with more complicated management problems.