• 1

    Crowe S, Turnbull S, Oelrichs R, Dunne A, 2003. Monitoring of human immunodeficiency virus infection in resource-constrained countries. Clin Infect Dis 37 (Suppl 1):S25–S35.

    • Search Google Scholar
    • Export Citation
  • 2

    Kumarasamy N, Mahajan AP, Flanigan TP, Hemalatha R, Mayer KH, Carpenter CC, Thyagarajan S, Solomon S, 2002. Total lymphocyte count (TLC) is a useful tool for the timing of opportunistic infection prophylaxis in India and other resource-constrained countries. JAIDS 31 :378–383.

    • Search Google Scholar
    • Export Citation
  • 3

    Spacek LA, Griswold M, Quinn TC, Moore RD, 2003. Total lymphocyte count and hemoglobin combined in an algorithm to initiate the use of highly active antiretroviral therapy in resource-limited settings. AIDS 17 :1311–1317.

    • Search Google Scholar
    • Export Citation
  • 4

    van der Ryst E, Kotze M, Joubert G, Steyn M, Pieters H, van der Westhuizan M, van Staden M, Venter C, 1998. Correlation among total lymphocyte count, absolute CD4+ count, and CD4+ percentage in a group of HIV-1-infected South African patients. JAIDS Human Retrovirol 19 :238–244.

    • Search Google Scholar
    • Export Citation
  • 5

    WHO, 2004. Scaling up antiretroviral therapy in resource-limited settings: treatment guidelines for a public health approach (2003 revision). Available at http://www.who.int/3by5/publications/en/arv_eng.pdf.

  • 6

    Schreibman T, Friedland G, 2004. Use of total lymphocyte count for monitoring response to antiretroviral therapy. Clin Infect Dis 32 :257–262.

    • Search Google Scholar
    • Export Citation
  • 7

    Le Moing V, Chene G, Carrieri MP, Besnier JM, Masquelier B, Salamon R, Bazin C, Moatti JP, Raffi F, Leport C, APROCO Study Group, 2001. Clinical, biologic, and behavioral predictors of early immunologic and virologic response in HIV-infected patients initiating protease inhibitors. JAIDS 27 :372–376.

    • Search Google Scholar
    • Export Citation
  • 8

    Wanke CA, Silva M, Knox TA, Forrester J, Speigelman D, Gorbach SL, 2000. Weight loss and wasting remain common complications in individuals infected with human immunodeficiency virus in the era of highly active antiretroviral therapy. Clin Infect Dis 31 :803–805.

    • Search Google Scholar
    • Export Citation
  • 9

    Tang AM, Forrester J, Spiegelman D, Knox TA, Tachetgen E, Gorbach SL, 2002. Weight loss and survival in HIV-positive patients in the era of highly active antiretroviral therapy. JAIDS 31 :230–236.

    • Search Google Scholar
    • Export Citation
  • 10

    Silva M, Skolnik PR, Gorbach SL, Spiegelman D, Wilson IB, Fernandez-DiFranco MG, Knox TA, 1998. The effect of protease inhibitors on weight and body composition in HIV-infected patients. AIDS 12 :1645–1651.

    • Search Google Scholar
    • Export Citation
  • 11

    Wilson IB, Roubenoff R, Knox TA, Spiegelman D, Gorbach SL, 2000. Relation of lean body mass to health-related quality of life in persons with HIV. JAIDS 24 :137–146.

    • Search Google Scholar
    • Export Citation
  • 12

    Woods MN, Tang AM, Forrester J, Jones C, Hendricks K, Ding B, 2003. Effect of dietary intake and protease inhibitors on serum vitamin B12 levels in a cohort of human immunodeficiency virus-positive patients. Clin Infect Dis 37 (Suppl 2):S124–S131.

    • Search Google Scholar
    • Export Citation
  • 13

    Wanke CA, Silva M, Ganda A, Fauntleroy J, Spiegelman D, Knox TA, Gorbach SL, 2003. Role of acquired immune deficiency syndrome-defining conditions in human immunodeficiency virus-associated wasting. Clin Infect Dis 37 (Suppl 2):S81–S84.

    • Search Google Scholar
    • Export Citation
  • 14

    Knox TA, Spiegelman D, Skinner SC, Gorbach S, 2000. Diarrhea and abnormalities of gastrointestinal function in a cohort of men and women with HIV infection. Am J Gastroenterol 95 :3482–3489.

    • Search Google Scholar
    • Export Citation
  • 15

    Forrester JE, Woods MN, Knox TA, Spiegelman D, Skinner SC, Gorbach SL, 2000. Body composition and dietary intake in relation to drug abuse in a cohort of HIV-positive persons. JAIDS 25 (Suppl 1):S43–S48.

    • Search Google Scholar
    • Export Citation
  • 16

    Beck EJ, Kupek EJ, Gompels MM, Pinching AJ, 1996. 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 7 :422–428.

    • Search Google Scholar
    • Export Citation
  • 17

    Badri M, Wood R, 2003. Usefulness of total lymphocyte count in monitoring highly active antiretroviral therapy in resource-limited settings. AIDS 17 :541–545.

    • Search Google Scholar
    • Export Citation
  • 18

    Flanigan TP, M.A., Kumarasamy N, et al. 2002. Total lymphocyte count (TLC) as a surrogate for CD4 count to initiate and monitor HAART in resource-limited countries [abstract 456-W]. In 9th Conference on Retroviruses and Opportunistic Infections (Seattle). Seattle: Foundation for Retrovirology and Human Health.

  • 19

    Selker HP, 2001. Cardiac time-insensitive predictive instruments (TIPIs): real-time emergency clinical decision aids and tools for the retrospective assessment of care. Maryland Medicine Suppl :95–97.

    • Search Google Scholar
    • Export Citation
  • 20

    Aufderheide TP, Rowlandson I, Lawrence SW, Kuhn EM, Selker HP, 1996. Test of the acute cardiac ischemia time-insensitive predictive instrument (ACI-TIPI) for prehospital use. Ann Emerg Med 27 :193–198.

    • Search Google Scholar
    • Export Citation

 

 

 

 

PREDICTING CD4 COUNT USING TOTAL LYMPHOCYTE COUNT: A SUSTAINABLE TOOL FOR CLINICAL DECISIONS DURING HAART USE

View More View Less
  • 1 Division of Geographic Medicine and Infectious Diseases, Tufts – New England Medical Center, Boston, Massachusetts; Division of Nutrition and Infection, Department of Family Medicine and Public Health, Tufts University School of Medicine, Boston, Massachusetts

Understanding the total lymphocyte count (TLC)-CD4 count relationship could aide design predictive instruments for making clinical decisions during antiretroviral therapy, especially in underserved resource-poor settings. We performed multiple regression analyses to assess the prediction of CD4 count using TLC on 771 participants with 4,836 visits. In linear and logistic regression TLC, hemoglobin, gender, history of AIDS, and weight predicted CD4 count and CD4 < 200, respectively, before and after highly active antiretroviral therapy (HAART) use. On HAART, the adjusted odds ratios (OR) for TLC < 1500 (optimal TLC cutoff) were 5.1 (95%CI 4.0, 6.5; P < 0.001), and off HAART, 4.6 (95%CI 3.4, 6.2: P < 0.001) with high predictive power. TLC predicts CD4 count and CD4 < 200 cells/μL well during HAART. Including the additional factors improves performance. TLC is simple and inexpensive and can be used in many ways to develop clinical decision-making tools in underserved resource-poor settings during HAART therapy.

INTRODUCTION

The burden of HIV in resource-poor countries is extensive, and a large proportion of HIV patients rely on accessing health care services in rural and underserved areas that do not have the capacity or capability to determine CD4 cell counts. In the absence of viral loads and CD4 counts for monitoring HIV disease, the value of total lymphocyte count (TLC) as a surrogate for CD4 has been argued.14 Viral loads and CD4 counts demand highly skilled laboratory personnel and costly maintenance of sophisticated equipment.1 The focus of research regarding TLC has been on its use in determining when to start therapy.3 Current World Health Organization (WHO) guidelines only commit to using TLC in conjunction with clinical data as a criteria to initiate highly active antiretroviral therapy (HAART) in resource poor settings.5 WHO provides no guidelines on the specific use of TLC in making clinical decisions after patients are started on HAART. TLC is inexpensive and simple to perform and has been shown to be reliable in the HAART-naïve population, but its value may not only be in its ability to identify the need to start HAART but to monitor patients’ progress and immune status during HAART as well.6 Because effective treatment with HAART would tend to stabilize CD4 count and TLC, the smaller variation in these parameters observed during therapy may be insufficient to adequately allow CD4 count prediction using TLC. Understanding TLC could aide designing useful and appropriate predictive instruments for use in making clinical decisions regarding ongoing antiretroviral therapy, especially in rural and underserved areas.

To the best of our knowledge, few studies have compared the utility of TLC as a predictor for CD4 count during and in the absence of HAART or have examined the collective impact of using hemoglobin, weight, presence of an AIDS defining illness (ADI), and gender to improve the TLC’s predictive performance.3,7 There are no large data sets of HAART-experienced populations in resource-poor settings. We used data from the Nutrition for Healthy Living Study (NFHL) to examine the performance of TLC in predicting CD4 counts to demonstrate the possibility and to generate hypotheses for future studies when appropriate data are available in resource-poor settings. A secondary goal was to assess how including data on readily available factors may improve the utility of TLC in monitoring patients before and during antiretroviral therapy.

METHODS

Participants.

Study participants were those enrolled in the ongoing Nutrition for Healthy Living Study (NFHL) in Boston, Massachusetts, and Providence, Rhode Island, between February 1995 and January 2004 and followed biannually.815 The study sample includes HIV-infected adults, 18 years or older, in all stages of the disease regardless of their antiretroviral treatment status. Study exclusion criteria were pregnancy or malignancies other than Kaposi sarcoma at recruitment. The unit of analysis for this study was each participant’s visit. As of January 2004, 884 NFHL participants had been enrolled in the study. We restricted our analysis to those with complete data on gender, history of AIDS defining illness (ADI), hemoglobin (Hb), weight, TLC, CD4 count, and HAART use. The number of visits per individual ranged from 1 to 16. All protocols and procedures were reviewed and approved by the human subjects review committees at each of the participating institutions.

Measurements.

History of ADI was obtained through interviews using a standardized questionnaire. Body weight was measured using a digital scale after a 5-hour fast to the nearest 0.1 kg. Complete blood counts with differential counts were performed using the Sysmex SE analyzer (Long Grove, IL). T-cell subsets, including CD4 cells, were determined using a fluorescent monoclonal antibody labeled cell sorter and expressed as cells/μL. HAART use was defined as receipt of any of the following regimens: 2 protease inhibitors (PI), 1 PI and 2 nucleotide reverse transcriptase inhibitors (NRTI), or 1 non-nucleotide reverse transcriptase inhibitor (NNRTI) and 2 NRTIs.1

Statistical analysis.

Baseline analysis.

We performed comparisons of baseline demographic and clinical characteristics according to HAART use using the Student’s t test for normally distributed variables, Wilcoxon rank sum test for non-normal variables, and χ2 test for categorical variables.

Statistical approach.

The main predictor variable was TLC and the outcome was CD4 count. We performed both linear and logistic regression modeling to explore the versatility of TLC in predicting CD4 count. We used generalized estimation equations (GEE) to adjust for the multiple visits per individual to obtain beta coefficients and odds ratios with robust standard errors. For the linear regression analysis, both TLC and CD4 were continuous variables. For the logistic regression, the outcome was a dichotomous variable representing CD4 < 200 cells/μL. For calibration of the TLC cutoff, we compared a range of cutoffs of TLC between 1,100 and 1,900 cells/μL to the CD4 < 200 cells/μL to obtain the optimal level for dichotomization for each subset analysis.

Multivariate analyses.

We constructed separate multivariable linear and logistic regression models for those visits in which patients were on HAART and those off HAART. We selected factors that would be readily available in under-served areas in resource poor countries. These included gender, body weight, hemoglobin level, and presence of an ADI. We specifically sought to use each of these factors measured or obtained alongside CD4 count in each study visit to develop time-insensitive models. We used the C-statistic to compare the predictive powers of the various models.

Sensitivity analyses testing the influence of a) those categorized as non-HAART but on some form of antiretroviral therapy, and b) categorization of HAART use using data from the two consecutive visits (i.e., HAART use and non-HAART would be those who reported HAART use and those who reported no HAART use at both the study visit and the preceding visit, respectively).

All analyses were performed on SAS version 8.02 SE (SAS, Cary, NC).

RESULTS

There were 771 participants who contributed 4,836 visits, during which 3,177 visits were on HAART and 1,659 were not on HAART.

Baseline characteristics.

Overall, of 771 participants whose average age was 40 years, 200 (26%) of the participants were female, 339 (47%) were non-white, and 389 (51%) were on HAART. Of those 382 (49%) participants categorized as off-HAART, 177 (46%) were not taking any antiretroviral medications, and 205 (54%) were on non-HAART regimens. The mean body weight was 76 kg (77 kg for men and 71 kg for women), and the mean hemoglobin was 13.8 g/dL (14.2 g/dL for men and 12.6 g/dL for women). The median CD4 count was 330 cells/μL, and 28% of the sample (216) had CD4 < 200 cells/μL. The median TLC was 1,692 cells/μL and 288 (37%) had TLC < 1,500 cells/μL. The baseline characteristics of the 771 study participants according to HAART use are shown in Table 1.

Linear regression analyses.

The factors found to predict CD4 count and their associated beta coefficients are shown in Table 2. For those not on HAART, for each cell/μL increase in TLC (P < 0.001), being female (P < 0.001), having a history of ADI (P = 0.06), and every kg increase in body weight (P = 0.02), there was a mean increase in 0.19, 87.1, 12.9, and 1.14 cells/μL in the CD4 count, respectively, after adjusting for hemoglobin, year of recruitment, and non-HAART anti-retroviral usage. R2 for this model was 0.51. For those on HAART, for each cell/μL increase in TLC (P < 0.001), being female (P = 0.09), having a history of ADI (P = 0.84), and body weight (P = 0.008), there was a mean increase in 0.18, 29.4, 0.91, and 1.12 cells/μL in the CD4 count, respectively, after adjusting for hemoglobin and year of recruitment. R2 for this model was 0.44. In both these linear regression models, hemoglobin lost its ability to predict CD4 count when adjusted for TLC.

Calibration for optimal TLC cutoff.

For those not on HAART, the C-statistic peaked at a TLC cutoff (TLC1500) of 1,500 cells/μL with C = 0.778 when the sensitivity and specificity were 78.8 and 76.9, respectively. For those on HAART, the C-statistic also peaked at a TLC cutoff (TLC1500) of 1,500 cells/μL with C = 0.764 when the sensitivity and specificity were 70.6 and 82.3, respectively. In contrast, the C-statistics for predicting CD4 < 200 cells/μL using TLC as a continuous variable were 0.862 and 0.843 for those not on HAART and those on HAART, respectively. The sensitivities, specificities, and the C-statistics for each cutoff from 1,100 to 1,900 cells/μL models and the C-statistics for the continuous-TLC model are shown in Table 3.

Logistic regression analyses.

In the absence of HAART, the associated adjusted odds ratios (OR) were 4.6 (95%CI 3.4, 6.2) for TLC1500, 0.73 (0.67, 0.80) for each g/dL higher in hemoglobin, 0.32 (0.23, 0.47) for female gender, 1.9 (1.4, 2.6) for having history of an ADI, and 0.98 (0.97, 0.99) for each kg heavier in body weight (all P values ≤ 0.02). The C-statistic was 0.840. When the continuous TLC variable was used in this off-HAART model instead of the cutoff of 1,500 cells/μL, the C-statistic was 0.88. During HAART, the TLC < 1,500 was associated with an adjusted odds ratio (OR) of 5.1 (4.0, 6.5) in predicting CD4 count < 200 cells/μL. The C-statistic was equal to 0.823. Each g/dL higher in Hb was associated with an adjusted OR of 0.84 (0.79, 0.91), being female, 0.59 (0.41, 0.84), and having an ADI, 1.3 (1.1, 1.6) (all P ≤ 0.004) in predicting CD4 < 200 cells/μL. Increase in body weight was non-significant (P = 0.14) with an adjusted OR of 0.99 (0.98, 1.002). When the continuous TLC variable was used in this on-HAART model instead of the cutoff of 1,500 cells/μL, the C-statistic was 0.852. The factors found to predict CD4 < 200 cells/μL and their associated odds ratios are shown in Table 4.

Sensitivity analyses.

Adjusting for the use of non-HAART antiretroviral regimens, adjusting for the era or year of recruitment, or using two consecutive visits to categorize HAART use did not change the model outcomes.

DISCUSSION

We found that the total lymphocyte count (TLC) performed well in predicting CD4 count or CD4 < 200 cells/μL. Although better in those off HAART, the performance of TLC was good for both during and in the absence of HAART. The inclusion of hemoglobin, gender, history of AIDS defining illnesses, and weight in the model improved the performance significantly. Our findings suggest TLC, which is relatively inexpensive and available, is a reasonably accurate tool that can be used for monitoring the patients’ immune status during therapy in addition to determining when patients should start antiretroviral therapy. These findings also imply that the possibilities for modifying the models to suit specific needs exist such as employing simple scores, algorithms, or risk calculators for use in clinics in underserved areas.

TLC versus CD4.

Studies have described the relationship between TLC and CD4 count previously that support our study findings.6,16 A relatively high positive correlation has been established between absolute values of TLC and CD4 count16 or between changes in TLC and CD4 cell count.17 A range of TLC cutoffs have been used and reported as predictors of CD4 < 200 cells/μL. These cutoffs range from 1000 cells/μL with a specificity of 98% and a sensitivity of 53% to 1,400 cells/μL with a sensitivity of 73% and a specificity of 88%.2,6 In another study, a TLC cutoff of 1,200 cells/μL was modeled with hemoglobin to improve sensitivity.3 Our study explored the development of a simple time-insensitive model for predicting CD4 count or having a CD4 < 200 cells/μL using TLC and other measures that would be readily available even in underserved resource-poor settings. Our study demonstrated that TLC could be used to predict CD4 count, in various formats, both before and during HAART use.

Calibration to TLC cutoff of 1,500 cells/μL.

In trying to develop a model suitable for use in underserved resource-poor settings, employing simple factors in simple formats is crucial. Using cutoff of TLC simplifies its utility. In this study, TLC < 1,500 cells/μL was consistently the strongest predictor of CD4 < 200 cells/μL. The cutoff level is a balance between sensitivity and specificity in the relationship between TLC and CD4 count. A higher TLC cutoff improves sensitivity at the expense of specificity and a lower TLC cutoff improves specificity at the expense of sensitivity. Using the test with a low sensitivity would result in many patients who have a CD4 < 200 cells/μL being missed and not started on therapy. In contrast, using a test with a low specificity would result in many patients inappropriately started on therapy. In HIV treatment, the implications are that a low sensitivity is a public health concern, and a low specificity is a cost2 and cost-effectiveness3 issue.

Although we present 1,500 cells/μL as the best overall performing cutoff for TLC for these data, a cutoff should be selected depending on the local clinical and/or public health strategy so that the choice of a cutoff level fits the desired goals and is most cost effective. The disadvantages of the desirable simplification of the model by converting TLC into a dichotomous measure, namely loss of precision, can be overcome by the inclusion of the other simple and readily available measures, such as hemoglobin, gender, body weight and history of ADI.

HAART versus non-HAART.

To the best of our knowledge, few studies have examined the utility of TLC in predicting CD4 measures during HAART.17,18 This is the first time the TLC/CD4 relationship has been compared between those on and off HAART in the same sample of patients. The similarity in the performance of TLC in predicting CD4 < 200 cells/μL by HAART use groups was in spite of the significant difference in viral load levels in the two groups. Our findings suggest physicians and auxiliary health care workers may readily be able to monitor patient after initiating HAART without having concerns that HAART use may affect the utility of TLC. Because, in our approach, study predictors (TLC, hemoglobin, ADI, female gender, and weight) and the outcome (CD4 count) were measured at the same time, our models are time-insensitive.19,20 Time-insensitive models are particularly important because, with limitations to health care access in underserved areas and other resource-poor settings, completing patient investigations and making clinical decisions in a single visit minimizes patient loss-to-follow up during care. Often, under these resource-limiting circumstances, only a nurse or clinical assistant manages the clinics. Usually these clinics are the first point of contact or sometimes the only point of care for the HIV/AIDS patient. Our study provides new evidence that for patients on HAART, TLC and other simple measures may be used in making clinical decisions, such as commencing prophylaxis for opportunistic infections, referral to a physician at a district hospital, and possibly even change in HAART regimen depending on the level of care; thus, a potential for improving the access to HIV/ AIDS care and lessening the burden of HIV/AIDS care on secondary and tertiary care facilities.

Possible models for prediction instruments.

One of the implications of our study is that user-friendly prediction instruments may be designed to ensure efficient and accurate exploitation of these predictive models in the field. Discussing the specific possible designs for the various predictive instruments in details is beyond the scope of this paper. However, our model is a demonstration that TLC may be used to predict CD4 count or CD4 < 200 cells/μL before and during HAART. For specific populations (and specific clinical/public health strategies), appropriate calibration with local (or applicable) data to obtain an optimal cutoff for TLC will be required in building the model and developing the predictive instrument. The range of predicted values of CD4 count or probabilities of CD4 < 200 cells/μL derived from the regression models can be converted into an intuitive scale (or a simple algorithm) so that the tools may be used even in primary care clinics in resource-poor settings. To make clinical decisions from such a scale, two strategies may be used in the field depending on availability of resources and population distributions. a) A single cutoff score can be determined by basing calculations on desired (or acceptable) levels of sensitivity and specificity and projected cost effectiveness of implementing such a cutoff. Specific clinical decisions can be made if the score is above or below the selected cutoff, or b) an interval range of scores between two strategically selected cutoff points that would ensure a high specificity below the lower cutoff (albeit with low sensitivity) and a high sensitivity above the upper cutoff (albeit with low specificity). Clinical decisions in the underserved clinic can be made if the score is above or below the selected interval. All patients presenting with scores within this interval, who strategically should represent a small and manageable proportion of all patients, can be referred for CD4 count and/or viral load determination. The performance and cost effectiveness of the strategy in this second example would depend on the level of the each cutoff and the size of the interval between them.

Strengths and limitations.

The NFHL data offered two unique opportunities: the availability of data to permit the comparisons between off-HAART and on-HAART performance of TLC in predicting CD4 < 200 cells/μL, and the use of a large sample size, which also accounts for the within individual variation of measures. The limitations of these models are that the coefficient estimates may not necessarily be applicable to populations in resource-poor settings. Thus, the purpose of this study was to provide evidence that significant relationships between TLC and CD4 count exist during and in the absence of HAART and that this relationship can be exploited elsewhere. Appropriate models for specific populations may be developed using local data.

In summary, this study demonstrates the ability of TLC, whether used as continuous or dichotomous data, to predict CD4 count or CD4 < 200 cells/μL, respectively. TLC, which is an inexpensive and simple test to perform, is valuable for monitoring therapy and making clinical decisions regarding the therapy. The inclusion of readily available factors such as hemoglobin, gender, history of ADI, and body weight significantly improves the predictive performance of TLC. User-friendly predictive instruments based on these models can be developed for use in underserved resource-poor settings and can be tailored to specific HIV/AIDS care needs. With the expansion of the provision of HAART in developing countries, local data can be used to design area and strategy specific predictive models.

Table 1

Baseline characteristics of study participants by HAART use at the time of study visit

CharacteristicsOff HAART (N = 382; 49.5%)On HAART (N = 389; 50.5%)All patients (N = 771)
* P < 0.01.
† Exposures are not mutually exclusive.
P value of comparison between HAART use groups < 0.05.
Age,* mean years (SD)38.9 (7.1)41.4 (7.4)40.2 (7.3)
Female, n (%)106 (27.8)94 (24.2)200 (25.9)
Race/ethnicity, n (%)
    African American128 (33.5)100 (25.7)228 (29.6)
    Hispanic/Latino35 (9.2)37 (9.5)72 (9.3)
    White200 (52.3)232 (59.7)432 (56.0)
    Other19 (5.0)20 (5.1)39 (5.1)
Exposure to HIV transmission,† n (%)
    Homosexual181 (48.5)210 (55.9)391 (52.2)
    Heterosexual79 (21.2)71 (18.9)150 (20.3)
    Injection drug use114 (30.6)99 (26.3)213 (28.4)
Achieved high school education, n (%)304 (79.6)331 (85.1)635 (82.4)
Living below federal poverty line,‡ n (%)251 (65.7)211 (54.2)462 (59.9)
Defined as having AIDS,‡ n (%)135 (35.3)204 (52.4)339 (44.0)
Era of recruitment‡
    < 1997211 (55.2)51 (13.1)262 (34.0)
    1997–1999111 (29.1)206 (53.0)317 (41.1)
    > 199960 (15.7)132 (33.9)192 (24.9)
Total body weight, mean kg (SD)75.4 (16.2)76.3 (14.5)75.9 (15.4)
Hemoglobin, mean g/dL (SD)13.7 (1.7)14.1 (1.8)13.8 (1.7)
White cell count,‡ median cells/μL (Q1, Q3)4,700 (3,800, 5,800)5,100 (4,000, 6,300)4,900 (3,900, 6,100)
Total lymphocyte count, median cells/μL (Q1, Q3)1,668 (1,254, 2,160)1,722 (1,287, 2,214)1,692 (1,258, 2,196)
Total lymphocyte count, proportion < 1500 cells/μL, n (%)148 (38.7)140 (38.0)288 (37.3)
CD4 count, median cells/μL (Q1, Q3)348 (189, 530)311 (180, 509)330 (182, 523)
CD4 count, proportion < 200 cells/μL, n (%)104 (27.2)112 (28.8)216 (28.0)
HIV viral load,‡ median log10 of RNA copies/mL (Q1, Q3)4.2 (3.4, 4.8)2.3 (2.3, 3.7)3.4 (2.3, 4.5)
Table 2

Linear predictors of CD4 count in the absence of HAART and during HAART (n = 4,836 visits)

Off-HAART (N = 1,659; R2 = 0.55)On-HAART (N = 3,177; R2 = 0.44)
EffectCorresponding increase in CD4 count (cell/μL)P valueCorresponding increase in CD4 count (cell/μL)P value
Both models were adjusted for hemoglobin level, non-HAART antiretroviral use, and year of recruitment. The predictive influence of gender and history of ADI during HAART is diminished.
Increase in TLC (cell/μL)0.19< 0.0010.18< 0.001
Being female87.1< 0.00129.40.09
Having history of AIDS defining illness12.90.060.910.84
Increase in body weight (kg)1.140.021.120.008
Table 3

Comparisons of model performances of varying TLC cutoffs between 1,100 and 1,900 cells/mm3 in predicting a CD4 count < 200 cells/μL. We present the sensitivities, specificities, and the C-statistics for each cutoff for the logistic models: Logit (CD4 (< 200)) = α + β(TLC < cutoff), for those on HAART and those not on HAART. Bold face entries show the best performing cutoff model and the linear TLC models

Off-HAART (N = 1,659)On-HAART (N = 3,177)
TLC cutoff (cells/μL)SensitivitySpecificityC-statistics*SensitivitySpecificityC-statistics
* C-statistic, which ranges from 0.5 to 1.0, describes the predictive power of the models. Proximity to 1.0 indicates stronger predictive power, while 0.5 is a 50–50 chance equivalent to tossing a coin.
TLC < 1,10055.394.50.74946.695.80.711
TLC < 1,20061.090.00.75553.693.30.734
TLC < 1,30067.786.70.77260.090.00.750
TLC < 1,40073.181.70.77465.086.80.759
TLC < 1,50078.876.90.77870.682.30.764
TLC < 1,60084.570.90.77775.376.50.759
TLC < 1,70086.664.90.75780.571.10.758
TLC < 1,80011.441.80.73415.834.50.748
TLC < 1,9009.048.90.71212.840.70.733
TLC Continuous0.8620.843
Table 4

Predictors of CD4 < 200 in the absence of HAART and during HAART (N = 4,836 visits)

Off-HAART (N = 1,659; C = 0.840)On-HAART (N = 3,177; C = 0.823)
EffectAdjusted OR [95% confidence interval]P valueAdjusted OR [95% confidence interval]P value
TLC < 1,500 (cells/μL)4.6 (3.4, 6.2)< 0.0015.1 (4.0, 6.5)< 0.001
Increase in hemoglobin (each g/dL)0.73 (0.67, 0.80)< 0.0010.84 (0.79, 0.91)< 0.001
Being female0.32 (0.23, 0.47)< 0.0010.59 (0.41, 0.84)0.004
Having history of AIDS defining illness1.9 (1.4, 2.6)0.0021.3 (1.1, 1.6)0.003
Increase in body weight (each kg)0.98 (0.97, 0.99)0.020.99 (0.98, 1.00)0.14

*

Address correspondence to D. Mkaya Mwamburi, Tufts – New England Medical Center, Division of Geographic Medicine and Infectious Diseases, Boston, MA 02111, E-mail: mmwamburi@tufts-nemc.org

Authors’ addresses: D. Mkaya Mwamburi and Mayurika Ghosh, Tufts – New England Medical Center, Division of Geographic Medicine and Infectious Diseases, Boston, MA, E-mail: mmwamburi@tufts-nemc.org and Mghosh@tufts-nemc.org. Jim Fauntleroy, Sherwood L. Gorbach, and Christine A. Wanke, Tufts University School of Medicine, Division of Nutrition and Infection, Department of Family Medicine and Public Health, Boston, MA, E-mail: jim.fauntleroy@tufts.edu, sherwood.gorbach@tufts.edu, and christine.wanke@tufts.edu

Financial support: This work was supported by NIH grants PO1 DK 45734-07, MO1 RR 00054, CFAR P30-A-142853, K24-AI-055293, and NIH/Forgarty grant D3TW01083A.

REFERENCES

  • 1

    Crowe S, Turnbull S, Oelrichs R, Dunne A, 2003. Monitoring of human immunodeficiency virus infection in resource-constrained countries. Clin Infect Dis 37 (Suppl 1):S25–S35.

    • Search Google Scholar
    • Export Citation
  • 2

    Kumarasamy N, Mahajan AP, Flanigan TP, Hemalatha R, Mayer KH, Carpenter CC, Thyagarajan S, Solomon S, 2002. Total lymphocyte count (TLC) is a useful tool for the timing of opportunistic infection prophylaxis in India and other resource-constrained countries. JAIDS 31 :378–383.

    • Search Google Scholar
    • Export Citation
  • 3

    Spacek LA, Griswold M, Quinn TC, Moore RD, 2003. Total lymphocyte count and hemoglobin combined in an algorithm to initiate the use of highly active antiretroviral therapy in resource-limited settings. AIDS 17 :1311–1317.

    • Search Google Scholar
    • Export Citation
  • 4

    van der Ryst E, Kotze M, Joubert G, Steyn M, Pieters H, van der Westhuizan M, van Staden M, Venter C, 1998. Correlation among total lymphocyte count, absolute CD4+ count, and CD4+ percentage in a group of HIV-1-infected South African patients. JAIDS Human Retrovirol 19 :238–244.

    • Search Google Scholar
    • Export Citation
  • 5

    WHO, 2004. Scaling up antiretroviral therapy in resource-limited settings: treatment guidelines for a public health approach (2003 revision). Available at http://www.who.int/3by5/publications/en/arv_eng.pdf.

  • 6

    Schreibman T, Friedland G, 2004. Use of total lymphocyte count for monitoring response to antiretroviral therapy. Clin Infect Dis 32 :257–262.

    • Search Google Scholar
    • Export Citation
  • 7

    Le Moing V, Chene G, Carrieri MP, Besnier JM, Masquelier B, Salamon R, Bazin C, Moatti JP, Raffi F, Leport C, APROCO Study Group, 2001. Clinical, biologic, and behavioral predictors of early immunologic and virologic response in HIV-infected patients initiating protease inhibitors. JAIDS 27 :372–376.

    • Search Google Scholar
    • Export Citation
  • 8

    Wanke CA, Silva M, Knox TA, Forrester J, Speigelman D, Gorbach SL, 2000. Weight loss and wasting remain common complications in individuals infected with human immunodeficiency virus in the era of highly active antiretroviral therapy. Clin Infect Dis 31 :803–805.

    • Search Google Scholar
    • Export Citation
  • 9

    Tang AM, Forrester J, Spiegelman D, Knox TA, Tachetgen E, Gorbach SL, 2002. Weight loss and survival in HIV-positive patients in the era of highly active antiretroviral therapy. JAIDS 31 :230–236.

    • Search Google Scholar
    • Export Citation
  • 10

    Silva M, Skolnik PR, Gorbach SL, Spiegelman D, Wilson IB, Fernandez-DiFranco MG, Knox TA, 1998. The effect of protease inhibitors on weight and body composition in HIV-infected patients. AIDS 12 :1645–1651.

    • Search Google Scholar
    • Export Citation
  • 11

    Wilson IB, Roubenoff R, Knox TA, Spiegelman D, Gorbach SL, 2000. Relation of lean body mass to health-related quality of life in persons with HIV. JAIDS 24 :137–146.

    • Search Google Scholar
    • Export Citation
  • 12

    Woods MN, Tang AM, Forrester J, Jones C, Hendricks K, Ding B, 2003. Effect of dietary intake and protease inhibitors on serum vitamin B12 levels in a cohort of human immunodeficiency virus-positive patients. Clin Infect Dis 37 (Suppl 2):S124–S131.

    • Search Google Scholar
    • Export Citation
  • 13

    Wanke CA, Silva M, Ganda A, Fauntleroy J, Spiegelman D, Knox TA, Gorbach SL, 2003. Role of acquired immune deficiency syndrome-defining conditions in human immunodeficiency virus-associated wasting. Clin Infect Dis 37 (Suppl 2):S81–S84.

    • Search Google Scholar
    • Export Citation
  • 14

    Knox TA, Spiegelman D, Skinner SC, Gorbach S, 2000. Diarrhea and abnormalities of gastrointestinal function in a cohort of men and women with HIV infection. Am J Gastroenterol 95 :3482–3489.

    • Search Google Scholar
    • Export Citation
  • 15

    Forrester JE, Woods MN, Knox TA, Spiegelman D, Skinner SC, Gorbach SL, 2000. Body composition and dietary intake in relation to drug abuse in a cohort of HIV-positive persons. JAIDS 25 (Suppl 1):S43–S48.

    • Search Google Scholar
    • Export Citation
  • 16

    Beck EJ, Kupek EJ, Gompels MM, Pinching AJ, 1996. 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 7 :422–428.

    • Search Google Scholar
    • Export Citation
  • 17

    Badri M, Wood R, 2003. Usefulness of total lymphocyte count in monitoring highly active antiretroviral therapy in resource-limited settings. AIDS 17 :541–545.

    • Search Google Scholar
    • Export Citation
  • 18

    Flanigan TP, M.A., Kumarasamy N, et al. 2002. Total lymphocyte count (TLC) as a surrogate for CD4 count to initiate and monitor HAART in resource-limited countries [abstract 456-W]. In 9th Conference on Retroviruses and Opportunistic Infections (Seattle). Seattle: Foundation for Retrovirology and Human Health.

  • 19

    Selker HP, 2001. Cardiac time-insensitive predictive instruments (TIPIs): real-time emergency clinical decision aids and tools for the retrospective assessment of care. Maryland Medicine Suppl :95–97.

    • Search Google Scholar
    • Export Citation
  • 20

    Aufderheide TP, Rowlandson I, Lawrence SW, Kuhn EM, Selker HP, 1996. Test of the acute cardiac ischemia time-insensitive predictive instrument (ACI-TIPI) for prehospital use. Ann Emerg Med 27 :193–198.

    • Search Google Scholar
    • Export Citation

Author Notes

Reprint requests: D. Mkaya Mwamburi, MD, Tufts – New England Medical Center, Division of Geographic Medicine and Infectious Diseases, 750 Washington Street, Box 41, Boston, MA 02111, Telephone: 617-636-5562, Fax: 617-636-8525, E-mail: mmwamburi@tufts-nemc.org.
Save