Mahajan, Anish P. MD, MPH*; Hogan, Joseph W. ScD†; Snyder, Brad MS†; Kumarasamy, N. MD‡; Mehta, Kalindi MD*; Solomon, Suniti MD‡; Carpenter, Charles C. J. MD*; Mayer, Kenneth H. MD*; Flanigan, Timothy P. MD*
The cost of highly active antiretroviral therapy (HAART) is dropping dramatically for low-income countries through various international initiatives. As availability of antiretroviral medications improves, it is important to develop feasible strategies for the clinical management of antiretroviral therapies in resource-limited settings. 1 One of the major obstacles to the administration of HAART is the absence of sophisticated and expensive laboratory equipment and infrastructure required for monitoring the efficacy of therapy. 1–4 In industrialized nations, changes in CD4 count and plasma viral load are used to determine the response of the virus to antiretroviral therapy. Standard methods of CD4 count and plasma viral load enumeration require highly trained personnel and tens of thousands of dollars of initial investment in laboratory instrumentation. 5 In the few resource-limited countries where such laboratory facilities exist, they are often concentrated in major cities and the assays are too expensive for the majority of patients to use for routine monitoring of HAART.
With 40 million people living with HIV in low-income countries and an estimated 6 million people in these countries currently requiring life-sustaining HAART, the World Health Organization (WHO) and UNAIDS has issued the “3 by 5 Initiative,” an effort to get 3 million people in resource-limited countries on antiretroviral therapy by the end of 2005. WHO offers technical recommendations to achieve this goal in their recently revised guidelines for scaling up antiretroviral therapy in resource-limited settings. 2 Citing the urgency of providing therapy on a wide scale and the financial and technological constraints to drastically upgrading laboratory facilities, the “Monitoring” section of these guidelines stipulates that CD4 count testing is “desirable” but not essential for antiretroviral use in resource-limited settings. 2 Where CD4 count testing is not available or too expensive for routine use, WHO recommends the use of total lymphocyte count (TLC) to monitor the immune response to HAART. 2 TLC is an inexpensive and widely available laboratory parameter. TLC is easily obtained from the routine complete blood count (CBC) with differential by multiplying percentage lymphocytes by leukocyte count. In southern India, for example, the cost of a single TLC from a CBC is <$1 (US) whereas a single CD4 count by flow cytometry is approximately $30 (US). In India, where the average annual income is <$350 and annual per capita spending on health by the government is $133, 6 the cumulative cost of monitoring HAART becomes a significant financial challenge. When computing cost on a annual basis, quarterly monitoring with TLC amounts to only $4 per year while that of CD4 count testing is $120 per year.
According to UNAIDS, even with the dramatic price reductions on HAART regimens to $300–$400 (US) per year, administration of antiretroviral therapy will remain unaffordable for many resource-limited countries. 7 The extent of funding required for the wide range of activities that make up anti-retroviral therapy programs in addition to monitoring of therapy, including drug procurement, drug distribution, clinical management of complications, monitoring of therapy, assessment and support of drug adherence, necessitates cost-effective solutions where appropriate and achievable. In light of its low cost and widespread availability, TLC has already been a useful tool in low-income countries for predicting im-munosuppression 8–10 and triggering opportunistic infection prophylaxis. 11–14 Recent studies have also demonstrated that TLC alone and in combination with hemoglobin may be useful in determining when to initiate antiretroviral therapy. 15–18 However, there are fewer studies examining the change of TLC in patients on antiretroviral therapy. 19
The purpose of this paper is to assess the capability and clinical utility of TLC change to serve as a surrogate marker for CD4 count change in monitoring patients on HAART, which has important implications for resource-limited settings. TLC and CD4 count changes—as opposed to cross-sectional values—are used because the standard quantification of immunologic response to therapy is change in CD4 count.
Data Collection and Study Setting
Data for this study were retrospectively derived from HIV-positive patients attending The Miriam Hospital Immunology Center, Brown University School of Medicine, from 1996–2001. Inclusion criteria for the clinical database of all patients attending the Center were as follows: age at least 18 years; antiretroviral-naive or -experienced patient initiating a new HAART regimen; CD4 count <250 cells/mm3 prior to the initiation of a new HAART regimen; not on antiretroviral therapy in the 3 months prior to starting the new HAART regimen; and at least 6 months’ duration on the new HAART regimen. A new HAART regimen was defined as ≥3 antiretroviral drugs in combination that included at least 1 agent that the patient has not been on before. Only patients with CD4 count <250 cells/mm3 were included in this study since patients in resource-limited settings usually initiate therapy at later stages of HIV infection when they are already severely immunosuppressed. Patients from the clinical database meeting these criteria were included in the analysis irrespective of adherence to simulate actual clinical management.
For each patient, the clinic visit prior to initiating a new HAART regimen served as the baseline. CD4 count, TLC, and plasma viral load from the baseline visit were recorded. Follow-up while on HAART generally occurred at 2- to 3-month intervals. The CD4 count, TLC, and plasma viral load measured on the same day from each of the follow-up visits were recorded.
Potential surrogate markers can be evaluated by ≥1 of several criteria. Our focus is on individual-level surrogacy, 20,21 or the ability of the surrogate (change in TLC) to predict the primary endpoint (change in CD4 count). Our analyses used data from a retrospective cohort of 126 patients treated at the Miriam Hospital Immunology Center, Providence, RI, who met the eligibility criteria described here. Patient data were recorded on average every 13 weeks, leading to 896 total data points over a 69-month time span.
We evaluated 2 specific strategies for using TLC change as a diagnostic marker for CD4 change. First, we quantify sensitivity, specificity, and both positive and negative predictive values (PPV and NPV) for using direction of TLC change as a marker for the direction of CD4 change over the same interval. These analyses include graphical representations of correlation between TLC and CD4 changes over 6-month intervals during the 24-month follow-up. We also present graphical summaries of TLC change corresponding to concomitant changes in HIV RNA (plasma viral load).
In the second part of the analysis, we evaluated the diagnostic capability of absolute change in TLC as a marker of 6-month CD4 change of ≥50 and 12-month CD4 change of ≥100. These benchmarks were defined by WHO in their 2002 provisional guidelines for scaling up antiretroviral therapy in resource-limited settings. 2 We compute receiver-operator characteristic (ROC) curves for change in TLC as a marker of 6-month CD4 count increase ≥50 cells/mm3 and 12-month CD4 count increase ≥100 cells/mm3. The area under the ROC curve summarizes diagnostic ability of a marker; in the context of the 6-month analysis, it can be interpreted as follows: if X is the change in TLC for an individual with CD4 increase of ≥50 cells/mm3, and Y is the change in TLC for an individual with CD4 increase of <50 cells/mm3, then the area under the ROC curve is the probability that X is greater than Y; i.e., that TLC will increase more for those whose CD4 changes by ≥50 cells/mm3 than for those with CD4 increase of <50 cells/mm3.
To complement the evaluation of diagnostic utility, a random-effects regression model for longitudinal data 22 is used to characterize longitudinal co-variation between the 2 markers and to estimate within-subject variability of change in TLC. The model is used to summarize the average change in TLC per unit change in CD4 count, investigate whether the association is constant over time since HAART initiation, and whether the association varies from individual to individual. Because the model summarizes both within- and between-subject variations in TLC change at fixed values of CD4 count change, it can be used to suggest ways of improving the diagnostic capability of TLC change.
Summary Statistics and Graphs
As indicated above, we used data on 126 patients followed for 24 months and evaluated about every 13 weeks. The cohort provides 896 pairwise measures of TLC and CD4 over a 69-month time span. Across individuals, the median number of visits with data available on CD4 count, TLC, and plasma viral load is 8 (range 3–9). Cumulative person-time of follow-up on the new HAART regimen was 209.6 years.
Regarding baseline characteristics, the majority of patients were male (77.8%). The mean CD4 count and mean TLC at baseline were 97 ± 70 cells/mm3 and 1020 ± 540 cells/mm3,respectively. Median plasma viral load (log10) at baseline was 5.0, with a range of 2.1–5.8.
Figure 1A shows box plots of change in TLC corresponding to categories of change in CD4 count, with data aggregated over all visits. A strong positive relationship is evident. For example, for occasions where CD4 count change from baseline is +100 to +150 cells/mm3, the median of the corresponding change in TLC was +600 cells/mm3, whereas for occasions where CD4 count improved by ≥+150, the median concomitant change in TLC was +1000 cells/mm3. In general, decreases in CD4 were also associated with decreases in TLC. For example, for occasions where CD4 count change from baseline is 0 to -50 cells/mm3, the median of the corresponding change in TLC was -300 cells/mm3.
Figure 1B is similar to 1A and shows change in TLC from baseline corresponding to change in plasma viral load from baseline, again aggregated over all visits. The left-most box plot refers to observations where plasma viral load changed from a detectable value to a value below the lower detection limit for the viral load assay (<500 copies/mL). Median change in TLC from baseline exhibited an inverse relationship with change in viral load from baseline, but with more variability than the TLC-CD4 relationship. For example, on occasions where change in log-10 viral load ranges between -1 and -0.5, the median change in TLC is +300 but a significant proportion of TLC changes are negative.
Direction of TLC Change as a Marker for Direction of CD4 Change Over 24 Months of HAART
To determine whether direction of TLC changes corresponds to direction of CD4 changes while on HAART, we grouped data into 6-month intervals. Figure 2 shows scatter plots of the changes in CD4 count vs. corresponding changes in TLC at 6-month intervals for the first 24 months on antiretroviral therapy. Each point on the plot represents a CD4 count and TLC pair. Some patients were evaluated more than once during the 6-month interval and therefore contribute more than one observation (e.g., the plot for the first 6-month interval shows 249 data points from 117 individuals). Figure 2 shows a strong positive (but possibly nonlinear) association between change in TLC and change in CD4 count during each 6-month interval over 24 months of HAART. Spearman nonparametric correlation coefficients range from 0.67–0.80. Each graph is divided into 4 quadrants, labeled A through D. Points in quadrant A represent an increase in TLC corresponding to a decrease in CD4, where increase is defined as a change ≥0 and decrease is defined as a change <0; points in quadrant B represent an increase in both markers; quadrant C contains points where both markers decrease, and quadrant D shows TLC decreases that correspond to CD4 increases.
We evaluated the diagnostic utility of direction of TLC change as a marker for direction of CD4 change using the usual metrics of sensitivity (proportion of those with TLC increase among those with CD4 increase), specificity (proportion of those with TLC decrease among those with CD4 decrease), and predictive values. During the first 6 months of treatment, sensitivity is 0.94 (210/223) and specificity is 0.85 (22/26). For a given sensitivity and specificity, PPV, or proportion of TLC increases that correspond to a true CD4 increase) and NPV, or proportion of TLC decreases that correspond to true CD4 decrease, depend on the proportion of occasions at which CD4 increases, which for our sample is 0.90 (223/249). In populations with the same prevalence of CD4 increase, the estimated PPV is 0.98 and the estimated NPV is 0.63. Summaries of sensitivity, specificity, and predictive values at other treatment intervals (6–12 months, 12–18 months, and 18– 24 months of HAART) are shown in Table 1. In summary, positive TLC change is a sensitive and relatively specific marker of positive CD4 change. Predictive values are directly relevant to clinical monitoring of therapy in practice. For settings where the probability of positive CD4 change is high, a TLC increase is almost perfectly predictive of a CD4 increase, but a TLC decrease is only moderately predictive of a true CD4 decrease. One implication for monitoring, therefore, is that observed decreases in TLC may represent a situation where confirmatory measures are needed (either repeated TLC measures, or where possible, ascertainment of the true CD4 count).
Use of TLC as a Marker of Benchmark Changes in CD4 Count Following HAART
We evaluated changes in TLC as a diagnostic monitoring marker of benchmark changes in CD4 count that indicate favorable response to HAART. We considered absolute TLC change as a marker of 2 different benchmark clinical events: change in CD4 count of ≥+50 cells/mm3 at 6 months of HAART and change in CD4 count of ≥+100 cells/mm3 at 12 months of HAART. Figure 3 depicts the distribution of TLC changes in patients with a CD4 count change of ≥+50 cells/mm3 (median change in TLC +600) vs. <+50 cells/mm3 (median change in TLC +50) at 6 months of therapy. The area under the corresponding receiver operator curve (ROC) for change in TLC as a marker for 6-month change in CD4 count of ≥50 cells/mm3 is 0.78 (SE = 0.05), indicating strong diagnostic capability. Figure 3 also depicts the distribution of TLC changes in patients with a CD4 count change of ≥+100 cells/mm3 (median TLC change = +766) vs. <+100 cells/mm3 (median TLC change = +100) at 12 months of therapy. The area under the corresponding ROC for change in TLC as a marker for a 12-month CD4 increase of ≥100 is 0.89 (SE = 0.04), again suggesting the high diagnostic utility of TLC change as a marker for clinically meaningful CD4 changes in response to HAART.
Table 2 provides sensitivity and specificity estimates corresponding to several threshold values of 6- and 12-month TLC change, but these should be interpreted with caution for 2 reasons. First, measurement instruments and associated measurement error may vary from location to location; this is easily overcome by applying our analysis to a training sample derived from the target population. Second, and perhaps more important, determination of a cutoff will depend both on the proportion of 6- (12)-month changes in CD4 that exceed 50 (100) and on the potential cost of making an erroneous prediction of whether the true change exceeds 50 (100). Cost considerations include the risk of elaborating resistant HIV vs. the risk of toxicity or higher financial burden of alternative anti-retroviral regimens. These costs will dictate whether false positives or false negatives are least desirable, and estimates of the costs can be used to develop appropriate cut-points.
Model for Individual TLC Changes as a Function of CD4 Count Changes
A mixed-effects regression model 22 for longitudinal data was used to characterize the within-subject variability of changes in TLC as a function of corresponding change in CD4, adjusting for baseline CD4, baseline TLC, and underlying temporal trend in TLC. Specifically, we assumed that rate of change in TLC follows the model
where t represents months since initiation of HAART, L (t) and C (t) are, respectively, TLC and CD4 at t months from initiation of HAART (and hence L (0) and C (0) are the corresponding baseline values), e (t) is a normally distributed error term with constant variance over time, and the β’s are regression coefficients. Time was scaled using months, so that dL/ dt represents TLC change per month (scaling has no bearing on P values). The model was fitted to 841 data points of paired, nonmissing CD4 count and TLC from 126 patients. The model assumes that average change in TLC varies linearly with corresponding change in CD4, and we allow the coefficients β0 and&beta2 to vary from person to person, reflecting the fact that underlying rate of change and the association between TLC and CD4 is not the same across individuals. The model was fit using Proc Mixed in SAS Version 8.2.
Directly following initiation of a new HAART regimen, estimated mean change in TLC per 1 CD4 cell/mm3 change was 7.3 cells/mm3 (SE 1.2, P < 0.001). Because the regression model uses multiple measures on the same individual, the standard deviation of the error terms can be used to provides an estimate of within-subject variability of 1-month change in TLC for fixed values of change in CD4 count. The error terms correspond to measurement error in 1-month TLC change under the assumption that measurement error does not change over time within a specific person; their estimated SD is 73.6. By contrast, the standard deviation of the intercept termscaptures between-subject variation in 1-month TLC change after accounting for variation in CD4 count change; the estimated between-subject SD is 73.5. This suggests that the variation of TLC measurements on the same individual varies as much as or more than measurements on different individuals and suggests that a single TLC measure, while highly associated with CD4 count on average, could be improved upon as a marker for CD4 change simply by taking replicate measurements to reduce measurement error. This applies particularly to predicting decreases in CD4.
This model is meant to illustrate that within-subject variation of TLC change is appreciable, but it most likely is overestimating the measurement error because the TLC measures are not taken at the same time and including other within-subject covariates would explain more of the within-subject variability. Nevertheless, studies designed to estimate TLC measurement error could prove valuable for informing the design of monitoring schemes with better diagnostic properties than those based on single TLC measures. For example, the average of 3 TLC changes has measurement error SD = 73.6/√– 3 = 42.5, a 73% reduction in measurement error relative to a single TLC change.
In resource-limited countries, widespread and routine use of CD4 count and plasma viral load in the management of HIV infection has not been possible. 22 Traditional methods of CD4 count measurement, such as immunophenotyping by flow cytometry or labeling with monoclonal antibodies, require expensive laboratory equipment and expertise. 23 Even where such facilities have been established at centralized laboratories, specimens typically require processing within 48 hours of sample collection, 5 which can be difficult to ensure in settings with poor transportation and communication infrastructure. Plasma viral load testing, with its reliance on sophisticated virology laboratories and patented polymerase chain reaction test kits, has also been extremely challenging to scale up in resource-limited settings. In addition to the scarcity of laboratory technologies, the high prices of these tests vis-a-vis the antiretroviral medications preclude regular monitoring of HAART. For example, in Chennai, the largest city in southern India, the annual cost of quarterly monitoring with CD4 count and plasma viral load is at least $500 (US) while a year’s supply of a generic 3-drug antiretroviral regimen made can be obtained for only $350.
As a result, a number of alternative methods for monitoring antiretroviral therapy in resource-limited settings have been proposed, such as HAART administration by directly observed therapy programs, 23,24 enumeration of CD4 count by simpler and less expensive methodologies, 25–27 and the development of a low-cost technique to track p24 antigen as a surrogate marker for viral load. 28,29 Although these and other strategies have promise, WHO recognized the need for a low-cost monitoring system that is immediately feasible on a wide scale. In December 2003, WHO released guidelines recommending that the monitoring of HAART be based on clinical indicators of response to therapy in conjunction with basic laboratory tests such as hemoglobin and TLC where CD4 count testing is not available. 2 While the guidelines strongly encourage further development of affordable and locally usable technologies for CD4 count enumeration, they recommend that this work should be pursued in parallel with a planned scale-up of antiretroviral therapy. 2 Scale-up of anti-retroviral therapy programs in resource-limited settings will require a structured framework for clinical management based upon careful validation of monitoring strategies relying on TLC. Further monitoring strategies to minimize the emergence of drug resistant HIV are needed. 30,31
In an earlier report by Badri and Wood, 19 median changes in TLC on HAART were shown to correlate well with median changes in CD4 count, indicating the potential usefulness of TLC in monitoring antiretroviral therapy. In our study, we offer further insight to clinicians on how to interpret change in TLC in patients on HAART. The regression model highlights the variability in TLC change within and between patients.
In this study, patients with baseline CD4 counts <250 cells/mm3 who had increases in TLC during the first 2 years on HAART had corresponding increases in CD4 count >95% of the time. According to these data, clinicians seeing a positive trend in TLC in patients on HAART can be almost certain that the CD4 count change is also positive. However, if TLC decreases, the direction of CD4 count change is not easily predicted. In this study, patients who had decreases in TLC during the first 2 years on HAART had corresponding decreases in CD4 count only 43–63% of the time. Since decreases in TLC are less accurate predictors of decreases in CD4 count, clinicians would need to further evaluate immune status by correlating the TLC drop with other clinical indications of deteriorating immune status (such as development of an opportunistic infection) or measuring the actual CD4 count. Another strategy requiring further study is whether prediction of true CD4 count change in the face of a decrease in TLC can be improved with an average of repeated TLC measures, which would at least reduce the effect of high TLC variability and thereby may improve overall accuracy.
In addition to correlating direction of change between TLC and CD4 count, this study also examined the average change in TLC per unit change in CD4 count. In this study population, the average individual-specific mean change in TLC per 1 CD4 cell/mm3 was 7.3 cells/mm3, but individual-level variation was substantial. Thus, while a single TLC measure has good prognostic properties for predicting direction of change in CD4, it may be less reliable as a proxy for the actual CD4 change. Monitoring strategies that use an average of multiple TLC measurements at a single time point may reduce measurement error in TLC and may offer a more precise prediction of actual CD4 change.
Results of this study demonstrate that single measurements of TLC at quarterly intervals are reliable to confirm a positive CD4 count response to therapy; prediction of poor immunologic response is more limited. The development of monitoring strategies incorporating TLC would benefit from future studies of replicate measures of TLC at quarterly intervals to determine if prediction of poor responses to therapy can be improved. Until such studies are conducted, the specific monitoring strategy adopted in a given setting will depend on resources available to both the providers and users of medical care. For example, in some settings, a combination of TLC and occasional CD4 counts may be possible for patients. In these instances, clinicians can follow patients with periodic TLC and reserve the use of CD4 count only when TLC significantly drops. In settings with very limited resources, monitoring strategies would rely on periodic TLC correlated with clinical indications of response, such as fluctuations in hemoglobin, 32 weight, or opportunistic infections including tuberculosis.
Limitations of this study necessitate further investigations. As this was an observational study of subjects in the United States, more assessments of TLC using data from the target countries are needed. Conditions specific to resource-limited settings, such as the higher prevalence of leukocytosis due to intercurrent infections, may influence the interpretability of changes in TLC. Further, as has been shown with CD4 count, 33,34 absolute TLC values as well as the dynamics of the measure may vary across different ethnicities. Region-specific validation of CD4 and TLC changes on HAART are needed. Also, TLC measured by CBC may not be possible in very low income countries or remote regions. Therefore, the evaluation of TLC measured by light microscopy and manual cell counting may be beneficial. Further study of TLC monitoring strategies that incorporate a low-cost surrogate marker for plasma viral load, such as heat-denatured and signal-amplified p24 antigen, would enable the development of a comprehensive monitoring system for resource-limited settings.
We are grateful to Kim Perry, The Miriam Hospital Immunology Center, Providence, RI, for her assistance in this the study. We also thank the Lifespan-Tufts-Brown Center for AIDS Research (CFAR-Grant # P30-AI 42853), NIH Grant R01 5U01AI046381, and the AIDS International Research and Training Program of the Fogarty International Center of the National Institutes of Health, USA (grant #TW00237) for their financial support and generous facilitation of this study. Work by Joseph Hogan was partially supported by grant R01-AI-50505 from NIH/NIAID.
1. Rabkin M, El-Sadr W, Katzenstein DA, et al. Antiretroviral treatment in resource-poor settings: clinical research priorities. Lancet. 2002;360: 1503–1505.
3. Stephenson J. Cheaper HIV drugs for poor nations bring a new challenge: monitoring treatment. JAMA. 2002;288:151–153.
4. Kumarasamy N, Flanigan TP, Mahajan AP, et al. Monitoring HIV treatment in the developing world. Lancet Infect Dis. 2002;2:656–657.
5. Joint United Nations Programme on HIV/AIDS (UNAIDS) and the World Health Organization. Guidance Modules on Antiretroviral Treatments. UNAIDS and World Health Organization; 1998. WHO/ASD/98.1 UNAIDS/97.7.
7. Cohen J. The high cost of poverty. Science. 2002;296:2324.
8. Montaner JSG, Le TN, Le N, et al. Application of the World Health Organization system for HIV infection in a cohort of homosexual men in developing a prognostically meaningful staging system. AIDS. 1992;6: 719–724.
9. Schechter M, Zajdenverg R, Machado LL, et al. Predicting CD4 counts in HIV-infected Brazilian individuals: a model based on the World Health Organization staging system. J Acquir Immune Defic Syndr. 1994;7:163–168.
10. Post FA, Wood R, Maartens G. CD4 and total lymphocyte counts as predictors of HIV disease progression. Q J Med. 1996;89:505–508.
11. Blatt SP, Lucey CR, Butzin CA, et al. Total lymphocyte count as a predictor of absolute CD4 count and CD4 percentage in HIV infected persons. JAMA. 1993;269:622–626.
12. Beck EJ, Kupek EJ, Gompels MM, et al. Correlation between total and CD4 lymphocyte counts in HIV infection: not making the good an enemy of the not so perfect. Int J STD AIDS. 1996;7:422–428.
13. Badri M, Ehrlich R, Wood R, et al. Initiating cotrimoxazole prophylaxis in HIV-infected patients in Africa: an evaluation of the provisional WHO/UNAIDS recommendations. AIDS. 2001;15:1143–1148.
14. Kumarasamy N, Mahajan AP, Flanigan TP, et al. The total lymphocyte count (TLC) is a useful tool for the timing of opportunistic infection prophylaxis in India and other resource-constrained countries. J Acquir Immune Defic Syndr. 2002;31:378–383.
15. Gange SJ, Lau B, Phair J, et al. Rapid declines in total lymphocyte count and hemoglobin in HIV infection begin at CD4 lymphocyte counts that justify antiretroviral therapy. AIDS. 2003;17:119–121.
16. Jacobson MA, Liu L, Khayam-Bashi H, et al. Absolute or total lymphocyte count as a marker for the CD4 T lymphocyte criterion for initiating antiretroviral therapy. AIDS. 2003;17:917–919.
17. Spacek LA, Griswold M, Quinn TC, et al. Total lymphocyte count and hemoglobin combined in an algorithm to initiate the use of highly active antiretroviral therapy in resource-limited settings. AIDS. 2003;17:1311–1317.
18. Lau B, Gange SJ, Phair JP, et al. Rapid declines in total lymphocyte counts and hemoglobin concentration prior to AIDS among HIV-1-infected men. AIDS. 2003;17:2035–2044.
19. Badri M, Wood R. Usefulness of total lymphocyte count in monitoring highly active antiretroviral therapy in resource-limited settings. AIDS. 2003;17:541–545.
20. Buyse M, Molenberghs G. Criteria for the validation of surrogate endpoints in randomized experiments. Biometrics. 1998;54:1014–1029.
21. Leung DHY. Statistical methods for clinical studies in the presence of surrogate end points. J R Stat Soc Ser A Stat Soc. 2001;164:485–503, part 3.
22. Laird NM, Ware JH. Random-effects models for longitudinal data. Biometrics. 1982;38:963–974.
23. Farmer P, Leandre F, Mukheree JS, et al. Community-based approaches to HIV treatment in resource-poor settings. Lancet. 2001;358:404–409.
24. Mitty JA, Stone VE, Sands M, et al. Directly observed therapy for the treatment of people with human immunodeficiency virus infection: a work in progress. Clin Infect Dis. 2002;34:984–990.
25. Lyamuya EF, Kagoma C, Mbena EC, et al. Evaluation of the FACScount, TRAx CD4 and Dynabeads methods for CD4 lymphocyte determination. J Immunol Methods. 1996;195:103–112.
26. Sherring A, Bergeron M, Mandy F. Comparison of two alternative CD4 T-cell enumerating methods suitable for laboratories with low volume immunophenotyping in resource poor settings. Paper presented at: XIV International AIDS Conference; July 2002; Barcelona.
27. Diagbouga S, Chazallon C, Kazatchkine MD, et al. Successful implementation of a low cost method for enumeration of CD4+ T lymphocytes in resource limited settings: the ANRS 12-26 study. AIDS. 2003;17:2201–2208.
28. Ledergerber B, Flepp M, Boni J, et al. Human immunodeficiency virus type 1 p24 concentration measured by boosted ELISA of heat-denatured plasma correlates with decline CD4 cells, progression to AIDS, and survival: comparison with viral RNA measurement. J Infect Dis. 2000;181: 1280–1287.
29. Pascual A, Cachafeiro A, Funk ML, et al. Comparison of an assay using signal amplification of the heat denatured p24 antigen with Roche monitor human immunodeficiency virus RNA assay. J Clin Microbiol. 2002;40: 2472–2475.
30. Harries AD, Nyangulu DS, Hargreaves NJ, et al. Preventing antiretroviral anarchy in sub-Saharan Africa. Lancet. 2001;358:410–414.
31. Weidle PJ, Mastro TD, Grant AD, et al. HIV/AIDS treatment and HIV vaccines for Africa. Lancet. 2002;359:2261–2267.
32. Semba RD, Shah N, Klein RS, et al. Highly active antiretroviral therapy associated with improved anemia among HIV-infected women. Aids Patient Care STDS. 2001;15:473–480.
33. Anglaret X, Diagbouga S, Mortier E, et al. CD4+T lymphocyte counts in HIV infection: are European standards applicable to African patients?J Acquir Immune Defic Syndr. 1997;14:361–367.
34. Kam KM, Wong KH, Lee SS. Intepretation of CD4+ T lymphocyte values in different HIV-infected populations. J Acquir Immune Defic Syndr. 1998;17:185–186.
© 2004 Lippincott Williams & Wilkins, Inc.