Tropical Medicine and International Health doi:10.1111/j.1365-3156.2011.02870.

volume 16 no 11 pp 1372–1379 november 2011

Total lymphocyte count is a good marker for HIV-related

mortality and can be used as a tool for starting HIV treatment
in a resource-limited setting
Helena P. W. Oudenhoven1, Hinta Meijerink1, Rudi Wisaksana3, Suryani Oetojo2, Agnes Indrati3,
Andre J. A. M. van der Ven1, Henri A. G. H. van Asten1, Bachti Alisjahbana2 and Reinout van Crevel1

1 Department of Internal Medicine, Radboud University Nijmegen Medical Centre, Nijmegen, The Netherlands
2 Internal Medicine Department at Medical Faculty, Padjadjaran University ⁄ Hasan Sadikin Hospital, Bandung, Indonesia
3 Department of Clinical Pathology, Medical Faculty, Padjadjaran University ⁄ Hasan Sadikin Hospital, Bandung, Indonesia

Summary objectives Total lymphocyte counts (TLC) may be used as an alternative for CD4 cell counts to monitor
HIV infection in resource-limited settings, where CD4 cell counts are too expensive or not available.
methods We used prospectively collected patient data from an urban HIV clinic in Indonesia.
Predictors of mortality were identified via Cox regression, and the relation between TLC and CD4 cell
counts was calculated by linear regression. Receiver operating characteristics (ROC) curves were used to
choose the cut-off values of TLC corresponding with CD4 cell counts <200 and £350 cells ⁄ ll. Based
on these analyses, we designed TLC-based treatment algorithms.
results Of 889 antiretroviral treatment (ART)-naı̈ve subjects included, 66% had CD4 cell counts
<200 and 81% had 350 £ cells ⁄ ll at baseline. TLC and CD4 cell count were equally strong predictors of
mortality in our population, where ART was started based on CD4 cell count criteria. The correlation
coefficient (R) between TLC and CD4 was 0.70. Optimal cut-off values for TLC to identify patients
with CD4 cell counts <200 and £350 cells ⁄ ll were 1500 and 1700 cells ⁄ ll, respectively. Treatment
algorithms based on a combination of TLC, gender, oral thrush, anaemia and body mass index per-
formed better in terms of predictive value than WHO staging or TLC alone. In our cohort, such an
algorithm would on average have saved $14.05 per patient.
conclusion Total lymphocyte counts is a good marker for HIV-associated mortality. Simple
algorithms including TLC can prioritize patients for HIV treatment in a resource-limited setting, until
affordable CD4 cell counts will be universally available.

keywords total lymphocyte count, CD4 cell count, clinical algorithm, resource-limited settings

Several studies have investigated the ability of TLC to

Introduction
Ideally, decisions about starting antiretroviral treatment
(ART) for patients infected with human immunodeficiency
virus (HIV) are based on CD4 cell count criteria, combined
with clinical staging. The latest WHO guidelines advise
to start ART at a CD4 count £350 cells ⁄ ll (WHO 2009),
but the former guidelines that use a CD4 cell count of
200 cells ⁄ ll as a cut-off are still being used in many
countries (Raizes et al. 2008). Unfortunately, CD4 cell
counts are not available or too expensive in many resource-
limited settings. Under these circumstances, ART is mostly
initiated based on disease stage according to WHO criteria
(WHO 2007). As an alternative, decisions may use the
total lymphocyte count (TLC), which is more widely
accessible and cheaper than CD4 cell counts.
predict CD4 cell counts. However, the correlation between
TLC and CD4 cell counts appears to be only moderate
(Akinola et al. 2004; Kamya et al. 2004; Liotta et al. 2004;
Schreibman & Friedland 2004; Angelo et al. 2007; Gitura
et al. 2007; Daka & Loha 2008). Combining the TLC with
other simple markers, like haemoglobin and body mass
index (BMI), can increase its accuracy (Kumarasamy et al.
2002; Spacek et al. 2003; Chen et al. 2007). In addition,
some studies have compared TLC-based treatment algo-
rithms with WHO staging criteria for starting ART
(Morpeth et al. 2007). However, only few studies evalu-
ated the prognostic value of TLC for mortality among
patients infected with HIV (May et al. 2010). We inves-
tigated the ability of TLC to predict mortality and CD4 cell
counts, and identified the most important covariates that

1372 ª 2011 Blackwell Publishing Ltd


Tropical Medicine and International Health
H. P. W. Oudenhoven et al. Lymphocyte count as a marker for HIV-related mortality
influence the relation between TLC and CD4 in patients
infected with HIV in Indonesia, which has one of the most
rapidly growing HIV epidemics (National Aids Commis-
sion 2008). The use of TLC as a surrogate for CD4 cell
counts would be of great value in Indonesia, as CD4 cell
measurement is available in only 25 hospitals in the whole
of Indonesia (240 million people). Based on these results,
we designed a clinical algorithm to predict CD4 cell counts
<200 and £350 cells ⁄ ll to identify patients who are eligible
for ART in resource-limited settings.
Methods
Setting and study population
This study was embedded in a 5-year programme called
IMPACT, aimed at better prevention, control and treat-
ment of HIV in the context of injecting drug use (IDU) in
West Java, Indonesia (Alisjahbana et al. 2009). IMPACT is
supporting HIV-related patient care, to both people with
and without a history of IDU, at Hasan Sadikin Hospital in
Bandung, the top-referral hospital for West Java (40
million people). Subjects who were at risk for HIV
infection or who presented with signs and symptoms
suggesting HIV ⁄ AIDS were counselled and tested for HIV.
All testing was voluntary, and informed consent was
obtained from all individuals. Standard questionnaires
were used to obtain information about demographical
factors such as age, gender, IDU and history of ART. A
physician scored the stage of disease using WHO criteria
and performed a physical examination (WHO 2007). Anti-
hepatitis C antibodies (optical density) were measured by
ECLIA (Roche diagnostic, Mannheim, Germany), with
external quality control showing a 100% accuracy (Na-
tional Serology Reference Laboratory, Australia). CD4 cell
measurements (cells ⁄ ll) were taken using FACS-count flow
cytometry technology (BD Biosciences, Jakarta, Indonesia);
TLC (cells ⁄ ll) and haemoglobin concentrations (g ⁄ dl) were
measured using the Cell Dyne 3000 (Abbott, Jakarta,
Indonesia). During follow-up, death was recorded from
medical records, reports from community organizations or
telephone interviews from the clinic. If patients were lost to
follow-up, we used the most recent date at which the status
of a patient was known as an endpoint. This study was
approved by the ethical committee of the Faculty of
Medicine of Padjadjaran University in Bandung, Indonesia.
Data analysis
In this study, we included all adult patients (‡16 years)
diagnosed with HIV infection between August 2007 and
February 2010. Patients were seen every 3 months if CD4
ª 2011 Blackwell Publishing Ltd
13653156, 2011, 11, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/j.1365-3156.2011.02870.x by Nat Prov Indonesia, Wiley Online Library on [28/08/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
volume 16 no 11 pp 1372–1379 november 2011
cell count £350 cells ⁄ ll, every 6 months if CD4 cell count
>350 cells ⁄ ll and every month when started on ART
independent of CD4 cell count. All patients were censored
at the date of last visit before February 2011. We excluded
all patients who had previously been exposed to ART or
who were using ART at time of enrolment. All variables
used in the analyses were selected based on previous
literature (Lau et al. 2003; Schreibman & Friedland 2004;
Nowicki et al. 2007), as they have a relation with CD4 cell
counts, TLC, or mortality.
To identify independent predictors of mortality, we used
Cox proportional hazard models. We performed univariate
Cox regression analysis for the baseline variables TLC,
CD4 cell count, haemoglobin, gender, oral thrush, BMI,
IDU, hepatitis C infection (HCV), age and ART initiation.
ART initiation was evaluated as a time-dependent covar-
iate. Using backward selection based on P-values <0.05,
variables were selected for two different multivariate
models, one based on TLC and the other based on CD4 cell
counts. These multivariate models were compared using
the Akaike Information Criterion (AIC) and used to
calculate hazard ratios (HRs) of independent predictors of
mortality. To prevent potential bias caused by missing
data, we included a separate group for missing data in the
mortality analysis (for example, IDU: yes, no or missing).
We used univariate and multivariate linear regression
analyses to examine the correlation between the indepen-
dent variable TLC and the dependent variable CD4 cell
count. Gender, age, haemoglobin, BMI, IDU, HCV and
oral thrush were evaluated as co-variables. These variables
were selected based on previous research (Moore et al.
2007; Gautam et al. 2010) and because they are easy to
measure in resource-limited settings. To prevent systematic
error, WHO staging was not entered in the linear model, as
oral thrush, haemoglobin and weight are included in the
WHO staging system. The square root of the dependent
variable (CD4 cell count) was used to meet all criteria for
linear regression.
Receiver operating characteristics (ROC) were used to
determine the sensitivity and specificity of different cut-off
values of TLC to predict CD4 cell counts <200 and
£350 cells ⁄ ll. The area under the curve (AUC) of each cut-
off point was used to determine the optimum cut-off value
of TLC within the study population. All variables with a
predictive effect on CD4 cell counts, according to the
multivariate regression analyses, were used to design
clinical algorithms, predicting CD4 cell counts <200 and
£350 cells ⁄ ll. In these algorithms, we first included vari-
ables that could be used to identify patients who could be
started on ART without measuring the TLC. We selected
those variables based on clinical experience in the field,
previous studies and the results of our own analyses. As a
1373
 13653156, 2011, 11, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/j.1365-3156.2011.02870.x by Nat Prov Indonesia, Wiley Online Library on [28/08/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
Tropical Medicine and International Health volume 16 no 11 pp 1372–1379 november 2011

H. P. W. Oudenhoven et al. Lymphocyte count as a marker for HIV-related mortality

final test in the algorithms, we included TLC to determine
whether ART should be initiated. In addition, we looked
whether different TLC cut-off points were needed for
subcategories using ROC and AUC. Test characteristics of
these algorithms were compared with test characteristics
of TLC alone and with WHO staging to predict CD4 cell
counts. Sensitivity analysis was carried out to calculate
the predictive values of the two algorithms after a
hypothetical 50% reduction in the prevalence of patients
with CD4 cell counts <200 and £350 cells ⁄ ll, under the
assumption that the sensitivity and specificity will not
change. SPSS version 16.0 (SPSS Inc., Chicago, IL, USA)
was used for all statistic analyses.
Results
We included 889 HIV-positive ART-naı̈ve patients in
this study. The majority of patients had advanced disease:
590 subjects (66%) had CD4 cell counts <200 cells ⁄ ll
and 723 (81%) had CD4 cell counts £350 cells ⁄ ll. The
median CD4 cell count in our population was 74 cells ⁄ ll.
The characteristics of our study population are listed in
Table 1. The median follow-up period was 496 days
(interquartile range [IQR] 108–744 days), 94 patients
(10.5%) died during follow-up, and the median time to
death was 82 days (IQR 16–220 days). A total of 366
patients (41%) dropped out of the study; those patients
had a higher median CD4 count (165 cells ⁄ ll) than
patients who died (13 cells ⁄ ll) or did return for follow-up
(70 cells ⁄ ll).
TLC, CD4 cell count, haemoglobin, oral thrush, gender,
BMI, ART and age had a significant relationship with
mortality (P < 0.05) in univariate Cox regression analysis
(Table 2). There were missing values for hepatitis C
(n = 83), age (n = 1), oral thrush (n = 130), haemoglobin
(n = 1) or BMI (n = 220); 150 subjects had missing data on
multiple variables. Four factors appeared as significant
covariates in the multivariate proportional hazard model
that was based on TLC, namely haemoglobin (P = 0.005),
gender (P = 0.009), ART (P < 0.001) and missing data on
oral thrush (P < 0.001). In the multivariate model based on
CD4 cell count, only haemoglobin (P = 0.001), ART
(P < 0.001) and missing data on oral thrush (P < 0.001)
were significant covariates (Table 2). The AIC of the model
based on TLC was 1022.8, the AIC of the model based on
CD4 was 1019.2, and this small difference indicates that
the predictive value of both models is comparable. In
addition, TLC and CD4 cell count had comparable hazard
ratios (HRs) for mortality in these two proportional
hazard models (Table 2).
There was a good correlation between TLC and CD4
count in univariate linear regression (n = 889, R = 0.700).
The test characteristics for different cut-off values of the
TLC to predict a CD4 cell count <200 and £350 cells ⁄ ll

Table 1 Population characteristics

Male Female All

Number of subjects 595 294 889

Male gender (%) n.a. n.a. 66.9
Median age, years (IQR) 30 (27–33) 26 (24–30) 29 (26–32)
Injecting drug use (%) 76.0 15.6 56.4
Oral thrush (%) 40.6 18.4 33.5
Hepatitis C infected (%) 73.8 22.3 56.9
WHOStage (%)
I 19.3 54.0 30.6
II 6.4 8.0 6.9
III 24.4 12.8 20.6
IV 49.9 25.2 41.8
Median CD4, cells ⁄ ll (IQR) 42 (10–197) 211 (57–366) 74 (16–302)
Median TLC, cells ⁄ ll (IQR) 1240 (791–1860) 1468 (918–1943) 1290 (814–1883)
Median haemoglobin, g ⁄ dl (IQR) 12.9 (11.0–14.7) 11.8 (10.7–12.9) 12.4 (10.8–14.1)
Median body mass index (kg ⁄ m2) (IQR) 18.8 (17.0–21.7) 19.7 (18.0–21.9) 19.1 (17.2–21.7)
Mortality during follow-up (%) 12.3 6.8 10.5
Median follow-up, days (IQR) 458 (88–748) 496 (171–733) 491 (108–744)
Median time to death, days (IQR) 87 (18–245) 66 (9–190) 82 (16–220)

Data were missing for age (n = 1), BMI (n = 220), injecting drug use (n = 82), hepatitis C (n = 83), WHO stage (n = 196), oral thrush
(n = 130) and haemoglobin (n = 1). Of the subjects, 16.9% (n = 150) had missing data on multiple (>1) variables.
IQR, interquartile range; CD4, CD4 cell count; TLC, total lymphocyte count; WHO stage, clinical stage according to World Health
Organization criteria.

1374 ª 2011 Blackwell Publishing Ltd

 13653156, 2011, 11, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/j.1365-3156.2011.02870.x by Nat Prov Indonesia, Wiley Online Library on [28/08/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
Tropical Medicine and International Health volume 16 no 11 pp 1372–1379 november 2011

H. P. W. Oudenhoven et al. Lymphocyte count as a marker for HIV-related mortality

Table 2 Hazard ratios of baseline variables for mortality, calculated with Cox proportional hazard models

Univariate Adjusted model based on TLC Adjusted model based on CD4

Baseline Variables HR (95% CI) P HR (95% CI) P HR (95% CI) P

TLC (cells ⁄ ll) 0.998 (0.998–0.999) <0.001 0.999 (0.998–0.999) <0.001 n.a. n.a.
CD4 (cells ⁄ ll) 0.988 (0.984–0.993) <0.001 n.a. n.a. 0.991 (0.987–0.995) <0.001
Haemoglobin(g ⁄ dl) 0.739 (0.681–0.803) <0.001 0.866 (0.784–0.957) 0.005 0.845 (0.767–0.930) 0.001
ART 0.175 (0.059–0.515) 0.002 0.127 (0.043–0.378) <0.001 0.099 (0.033–0.293) <0.001
Gender
Female 1 1 1
Male 1.848 (1.124–3.038) 0.015 1.992 (1.186–3.346) 0.009 1.310 (0.799–2.242) 0.324
BMI (kg ⁄ m2) 0.869 (0.787–0.957) 0.005 – 0.776 – 0.418
BMI missing 3.484 (2.279–5.326) <0.001 – 0.471 – 0.316
Oral thrush
No 1 1 1
Yes 3.036 (1.806–5.104) <0.001 1.215 (0.692–2.131) 0.498 1.032 (0.592–1.798) 0.912
Missing 5.859 (3.438–9.986) <0.001 4.026 (2.330–6.956) <0.001 3.597 (2.080–6.220) <0.001
Age (years) 1.044 (1.012–1.077) 0.007 – 0.709 – 0.489
Hepatitis C Virus – 0.516 – 0.448
No 1
Yes 0.911 (0.585–1.419) 0.681
Missing 1.380 (0.702–2.715) 0.351
IDU – 0.578 – 0.407
No 1
Yes 1.034 (0.658–1.624) 0.884
Missing 2.259 (1.188–4.296) 0.013

The Akaike Information Criterion (AIC) for the model based on TLC was 1022.836, and the AIC for the model based on CD4 was
1019.247. Data were missing for age (n = 1), BMI (n = 220), injecting drug use (n = 82), hepatitis C (n = 83), oral thrush (n = 130) and
haemoglobin (n = 1). H, hazard ratio; CI, confidence interval; TLC, total lymphocyte count; CD4, CD4 cell count; BMI, body mass
index; ART, antiretroviral therapy; n.a., not applicable; –, excluded from model based on P > 0.05.
Time-dependent initiation of ART equal to 0 for those never on ART and equal to 1 at the time of ART initiation.

are shown in Table 3. Based on the combined sensitivity
and specificity, the most accurate TLC cut-off value was
1500 cells ⁄ ll to predict CD4 cell counts <200 cells ⁄ ll and
1700 cells ⁄ ll to predict CD4 cell counts £350 cells ⁄ ll. The
negative predictive value (NPV) of these cut-off values
were 0.67 and 0.55; this means that 33 patients with a
TLC > 1500 cells ⁄ ll would actually have a CD4 cell count
<200 cells ⁄ ll and 49 patients with a TLC > 1700 cells ⁄ ll
would have a CD4 cell count £350 cells ⁄ ll. In multivariate
regression analysis, gender, oral thrush, haemoglobin, BMI
and age were significant covariates for CD4 cell counts
(Table 4). Adding these covariates to the model resulted in
a significant change in R2 from 0.47 to 0.62 (P < 0.001).
The AUC of the ROC curve of TLC was 0.80 (95% CI:
0.76–0.83) to predict CD4 cell counts <200 cells ⁄ ll and
0.83 (95% CI: 0.79–0.87) to predict CD4 cell counts
£350 cells ⁄ ll.
On the basis of previous analyses, two clinical algo-
rithms were designed to predict CD4 cell counts <200 and
£350 cells ⁄ ll. According to these algorithms, all patients
with oral thrush and all patients with both haemoglo-
bin < 12 g ⁄ dl and a BMI < 18.5 kg ⁄ m2 should commence
ART irrespective of TLC. For all other patients, TLC
should be measured. We calculated TLC cut-off values for
various subgroups and found that different cut-off values
should be used for men and women. To predict a CD4 cell
count <200 cells ⁄ ll, cut-off values of TLC <1500 cells ⁄ ll
for female patients and <1700 cells ⁄ ll for male patients
should be used (Figure 1). In this cohort of patients, use of
this algorithm would have obviated the need for CD4 cell
measurements in all patients and measurement of TLC in
37% of patients (n = 332). Assuming a cost of US$16.70
per CD4 cell count, CD4 cell measurements in all 889
patients would have cost US$14 846, whereas measure-
ment of TLC according to the algorithm in 567 patients
would have cost US$2211 (assuming a cost of US$3.90 per
TLC). Consequently, the use of this algorithm would have
saved US$12 635 in our cohort, which is US$14.05 per
patient. This ignores any costs associated with errors of
inclusion (over-diagnoses) and exclusion (under-diagnoses)
of the algorithm compared to the use of CD4 cell count.
A similar algorithm was made to predict a CD4 cell count

ª 2011 Blackwell Publishing Ltd 1375

 13653156, 2011, 11, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/j.1365-3156.2011.02870.x by Nat Prov Indonesia, Wiley Online Library on [28/08/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
Tropical Medicine and International Health volume 16 no 11 pp 1372–1379 november 2011

H. P. W. Oudenhoven et al. Lymphocyte count as a marker for HIV-related mortality

Table 3 Test characteristics of different cut-off values of the Table 4 Results of multivariate regression analysis with
total lymphocyte count to predict CD4 cell counts <200 and dependent variable (CD4 cell count)
£350 cells ⁄ ll
B 95% CI b P
TLC cut-off
value (cells ⁄ ll) Sens. Spec. PPV NPV AUC (95% CI) Step 1
Constant 2.293 1.364–3.222 <0.001
CD4 < 200 TLC (cells ⁄ ll) 0.006 0.006–0.007 0.683 <0.001
1000 0.51 0.97 0.97 0.50 0.74 (0.71–0.77) Step 2
1200 0.64 0.93 0.95 0.57 0.78 (0.75–0.82) Constant 1.222 )2.086–4.530 0.468
1400 0.75 0.85 0.91 0.63 0.80 (0.77–0.83) TLC (cells ⁄ ll) 0.005 0.004–0.005 0.502 <0.001
1500 0.80 0.79 0.88 0.67 0.80 (0.76–0.83) Male gender )3.768 )4.648 to )2.888 )0.241 <0.001
1600 0.83 0.74 0.86 0.69 0.78 (0.75–0.82) Oral thrush )3.142 )4.075 to )2.208 )0.188 <0.001
1700 0.87 0.67 0.84 0.72 0.77 (0.73–0.80) BMI 0.189 0.072–0.313 0.094 0.002
1800 0.89 0.67 0.82 0.74 0.75 (0.71–0.79) Haemoglobin 0.494 0.293–0.696 0.152 <0.001
1900 0.92 0.55 0.80 0.77 0.73 (0.67–0.77) Age )0.109 )0.181 to )0.038 )0.079 0.003
2000 0.94 0.48 0.78 0.79 0.71 (0.70–0.75) IDU – – – 0.851
2500 0.98 0.24 0.72 0.84 0.61 (0.57–0.65) Hepatitis C – – – 0.472
CD4 £ 350
1000 0.42 0.98 0.99 0.28 0.70 (0.66–0.74) R = 0.683 for step 1; R = 0.790 for step 2 (P < 0.001).
1200 0.55 0.97 0.99 0.33 0.76 (0.73–0.79) B, unstandardized linear regression coefficient; b, standardized
1400 0.66 0.94 0.98 0.39 0.80 (0.77–0.83) regression coefficient (b = B*sx ⁄ sy) (sx, standard deviation inde-
1500 0.72 0.90 0.97 0.42 0.81 (0.78–0.84) pendent variable; sy, standard deviation dependent variable); CI,
1600 0.76 0.88 0.97 0.46 0.82 (0.79–0.85) confidence interval; TLC, total lymphocyte count; BMI, body mass
1700 0.81 0.85 0.96 0.51 0.83 (0.79–0.87) index; IDU, injecting drug users.
1800 0.85 0.81 0.95 0.55 0.83 (0.79–0.86)
1900 0.87 0.73 0.93 0.57 0.80 (0.76–0.84)
2000 0.90 0.66 0.92 0.60 0.78 (0.73–0.82)
2500 0.97 0.38 0.87 0.72 0.67 (0.62–0.73) (PPV) of the TLC algorithm would be 0.58 (95% CI: 0.53–
0.62) compared to 0.88 (95% CI: 0.83–0.92) for WHO
TLC, total lymphocyte count; CD4, CD4 cell count; Sens.,
staging. The NPV of the TLC algorithm and WHO staging
sensitivity; Spec., specificity; PPV, positive predictive value; NPV,
negative predictive value; AUC, area under the curve; CI, would be 0.95 (95% CI: 0.92–0.97) and 0.83 (95% CI:
confidence interval. 0.80–0.85). When looking at CD4 cell count £350 cells ⁄ ll
(prevalence 40%), the TLC algorithm would have a PPV of
0.71 (95% CI: 0.67–0.75) compared with 0.75 (95% CI:
£350 cells ⁄ ll, using values of <1700 cells ⁄ ll for female 0.70–0.79) for WHO staging. The NPV would be 0.90
patients and <1800 cells ⁄ ll for male patients. (95% CI: 0.88–0.93) for TLC algorithm and 0.86 (95%
Algorithms combining TLC with oral thrush, BMI and CI: 0.83–0.89) for WHO staging. This shows that even
haemoglobin were compared with WHO staging or TLC when the average CD4 cell counts would be higher in this
alone (Table 5). When predicting CD4 cell counts setting, the TLC-based algorithms would still outperform
£350 cells ⁄ ll, only 88 out of 889 patients (9.9%, median WHO staging.
CD4 cell count: 254 cells ⁄ ll) would be denied ART using
the TLC-based algorithm, compared with 284 out of 693
Discussion
patients (41.0%, median CD4 cells: 182 cells ⁄ ll) when
using WHO staging. ART would be started prematurely in This study shows that the TLC is a useful tool to monitor
35 patients (3.9%, median CD4 cells: 425 cells ⁄ ll) using HIV infection and to decide who is eligible for starting
the TLC-based algorithm and five patients (0.7%, median ART in a resource-limited setting. The TLC was a strong
CD4 cells: 422 cells ⁄ ll) when using WHO staging. predictor of mortality and corresponded reasonably well
The previous results were based on the actual proportion with CD4 cell counts in this setting. Optimal cut-off values
of CD4 cell counts <200 and £350 cells ⁄ ll in our study of the TLC to predict CD4 cell counts <200 and
population (66% and 81%, respectively). To determine the £350 cells ⁄ ll in our population were much higher than
effect of these proportions on the predictive values of the 1200 cell ⁄ ll, which is often reported, and higher for men
tests, we hypothetically reduced these proportions by 50%, than for women. A clinical algorithm that includes TLC,
assuming that sensitivity and specificity of our tests oral thrush, BMI, haemoglobin and gender more reliably
remained the same. If only 33% of patients would have a identified patients qualifying for ART than clinical staging
CD4 cell count <200 cells ⁄ ll, the positive predictive value based on WHO criteria.

1376 ª 2011 Blackwell Publishing Ltd

 13653156, 2011, 11, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/j.1365-3156.2011.02870.x by Nat Prov Indonesia, Wiley Online Library on [28/08/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
Tropical Medicine and International Health volume 16 no 11 pp 1372–1379 november 2011

H. P. W. Oudenhoven et al. Lymphocyte count as a marker for HIV-related mortality

Oral thrush
Yes No

N = 635
ART
254 (241)* Hb and BMI
Hb < 12 AND
Other
BMI < 18,5

ART N = 557
78 (63)*
TLC
Female ≤ 1500 Female > 1500
100 (66)* 114 (13)*
Male ≤ 1700 Male > 1700
205 (171)* 138 (36)*

ART No ART
305 (237)* 252 (49)*

Figure 1 Clinical algorithm to decide which patients should start antiretroviral treatment (ART), based on CD4 cell counts <200 cells ⁄ ll.
All patients with oral thrush and all patients with both Hb < 12 g ⁄ dl and BMI < 18.5 kg ⁄ m2 should commence ART. For all other
patients, TLC is measured. To predict CD4 cell count <200 cells ⁄ ll, cut-off values for TLC are £1500 cells ⁄ ll for female and
£1700 cells ⁄ ll for male patients. The same algorithm can be adjusted to predict CD4 cell counts £350 cells ⁄ ll, by changing the cut-off
value of TLC to £1700 cells ⁄ ll for female patients and to £1800 cells ⁄ ll for male patients. *Total number of patients in this group
according to algorithm. In parenthesis: the number of patients with CD4 cell counts <200 cells ⁄ ll. Hb, haemoglobin concentration; BMI,
body mass index; TLC, total lymphocyte count.

Table 5 Test characteristics of different

tests to predict CD4+ cell count <200 and Sens. Spec. PPV NPV AUC (95% CI)
£350 cells ⁄ ll
Test to predict CD4 < 200
Algorithm 1 0.92 0.67 0.85 0.81 0.80 (0.76–0.84)
TLC alone (cut-off 1500) 0.80 0.77 0.88 0.67 0.79 (0.75–0.82)
WHO stage III and IV 0.59 0.96 0.97 0.52 0.77 (0.74–0.81)
Test to predict CD4 £ 350
Algorithm 2 0.88 0.76 0.95 0.60 0.82 (0.77–0.86)
TLC alone (cut-off 1700) 0.80 0.82 0.98 0.55 0.73 (0.69–0.77)
WHO stage III and IV 0.80 0.82 0.95 0.30 0.81 (0.77–0.85)

CD4, CD4+ cell count; TLC, total lymphocyte count; Sens., sensitivity; Spec., specificity;
PPV, positive predictive value; NPV, negative predictive value; AUC, area under responder
operating characteristics curve; 95% CI, 95% confidence interval.

Optimal cut-off values of TLC to predict CD4 cell
counts as described by others show great variation – values
range from 1000 (WHO 2006; Liu et al. 2008) to
2250 cells ⁄ ll (Moore et al. 2007) for a CD4 cell count
<200 cells ⁄ ll. This variation can be explained by the
influence of certain patient characteristics on the relation
between TLC and CD4 and by the distribution of CD4 cell
counts within a certain population. Our finding that TLC
and haemoglobin are strong predictors of mortality in
patients infected with HIV is in concurrence with previous
research (Post et al. 1996; Anastos et al. 2004; Harris et al.
2008).
Algorithms that include TLC and simple markers like
haemoglobin and BMI have been designed before, pre-
dicting CD4 cell counts with moderate accuracy (Spacek
et al. 2003; Chen et al. 2007; Moore et al. 2007; Morpeth
et al. 2007; Gautam et al. 2010). We improved this
approach by adjusting the cut-off values of the TLC for

ª 2011 Blackwell Publishing Ltd 1377


Tropical Medicine and International Health
H. P. W. Oudenhoven et al. Lymphocyte count as a marker for HIV-related mortality
gender, because gender influenced the correlation between
TLC and CD4. The use of these algorithms will obviate
the need for laboratory testing for many and thereby save
costs. In this study population, 37% (332 out of 889
patients) would have been offered ART based on the
clinical parameters in the algorithm, without use of TLC
measurement.
An important limitation of our study is that median
CD4 cell counts in our population were relatively low.
Test characteristics, especially the predictive values, are
affected by the high proportion of patients with low CD4
cell counts. In our study, 66% of all subjects had CD4 cell
counts <200 cells ⁄ ll and 81% had CD4 cell counts
£350 cells ⁄ ll. In Indonesia, HIV patients generally present
with advanced disease; therefore, our study population is
representative of this population (Wisaksana et al. 2010).
When we hypothetically increased the average CD4 cell
count, the algorithms still outperformed WHO staging.
Even though more people would receive ART prematurely
when using the TLC algorithm, with WHO staging
significantly more patients would unintentionally be
denied ART, and this might have severe consequences for
HIV-related morbidity and mortality. The TLC algorithm
is also much more practical than using WHO staging.
In resource-limited settings, it is often quite difficult to
determine whether a patient has WHO stage III or IV,
whereas the variables used in our algorithm are easy to
collect.
The TLC algorithm could also be a useful tool for
screening in settings where a CD4 cell count is available
but too expensive for routine care. In this case, especially
the high sensitivity of the algorithm is important. In these
settings, the algorithm would be used to prioritize patients
for CD4 cell testing, which would be used to initiate
ART. When using the algorithm for this purpose, one
might consider a higher TLC cut-off value to increase
sensitivity.
We did not yet evaluate the usefulness of TLC to
monitor treatment success after starting ART. Monitoring
treatment is ideally performed with HIV viral load and
CD4 cell counts. HIV viral load measurements are even
more expensive to take and less widely available than CD4
cell counts. Therefore, a cheaper alternative for resource-
limited settings would be useful. Some studies have already
looked at the use of TLC and simple algorithms to monitor
HIV treatment, with varying results (Badri & Wood 2003;
Mahajan et al. 2004).
In summary, we demonstrated that the TLC is a good
marker for HIV-related mortality and that an algorithm
based on simple patient characteristics can improve its
predictive value, outperforming WHO staging when it
comes to prioritization of patients for ART.
1378
13653156, 2011, 11, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/j.1365-3156.2011.02870.x by Nat Prov Indonesia, Wiley Online Library on [28/08/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
volume 16 no 11 pp 1372–1379 november 2011
Acknowledgement
We would like to thank Dr Bayu Wahyudi, Director of
Hasan Sadikin General Hospital, and Prof. Tri Hanggono
Achmad, Dean of the Medical Faculty, Padjadjaran Uni-
versity, for encouraging and accommodating research at
their institutions. Everyone working at the HIV clinic in the
hospital is thanked for providing HIV care and collecting
the data used for this study. This study was supported
by IMPACT (Integrated Management of Prevention And
Care and Treatment of HIV ⁄ AIDS), a collaborative
research and implementation programme of Padjadjaran
University, Bandung, Indonesia; Maastricht University
and Radboud University Nijmegen, the Netherlands; and
Antwerpen University, Belgium. IMPACT is funded by the
European Commission (SANTE ⁄ 2005 ⁄ 105-033).
References
Akinola NO, Olasode O, Adediran IA et al. (2004) The search for
a predictor of CD4 cell count continues: total lymphocyte count
is not a substitute for CD4 cell count in the management of HIV-
infected individuals in a resource-limited setting. Clinical
Infectious Diseases 39, 579–581.
Alisjahbana B, Susanto H, Roesli R et al. (2009) Prevention, control
and treatment of HIV-AIDS among injecting drug use in Ban-
dung, Indonesia. Acta Medica Indonesia 41 (Suppl. 1), 65–69.
Anastos K, Shi Q, French AL et al. (2004) Total lymphocyte
count, hemoglobin, and delayed-type hypersensitivity as
predictors of death and AIDS illness in HIV-1-infected women
receiving highly active antiretroviral therapy. Journal of
Acquired Immune Deficiency Syndromes 35, 383–392.
Angelo AL, Angelo CD, Torres AJ et al. (2007) Evaluating total
lymphocyte counts as a substitute for CD4 counts in the follow
up of AIDS patients. Brazilian Journal of Infectious Diseases 11,
466–470.
Badri M & Wood R (2003) Usefulness of total lymphocyte count
in monitoring highly active antiretroviral therapy in resource-
limited settings. AIDS 17, 541–545.
Chen RY, Westfall AO, Hardin JM et al. (2007) Complete blood
cell count as a surrogate CD4 cell marker for HIV monitoring
in resource-limited settings. Journal of Acquired Immune
Deficiency Syndromes 44, 525–530.
Daka D & Loha E (2008) Relationship between total lymphocyte
count (TLC) and CD4 count among peoples living with HIV,
Southern Ethiopia: a retrospective evaluation. AIDS Research
and Therapy 5, 26–30.
Gautam H, Saini S, Bhalla P & Singh T (2010) Use of total
lymphocyte count to predict absolute CD4 count in HIV-
seropositive cases. Journal of the International Association of
Physicians in AIDS Care 9, 292–295.
Gitura B, Joshi MD, Lule GN & Anzala O (2007) Total lym-
phocyte count as a surrogate marker for CD4+ t cell count in
initiating antiretroviral therapy at Kenyatta National Hospital,
Nairobi. East African Medical Journal 84, 466–472.
ª 2011 Blackwell Publishing Ltd

Tropical Medicine and International Health
H. P. W. Oudenhoven et al. Lymphocyte count as a marker for HIV-related mortality
Harris RJ, Sterne JA, Abgrall S et al. (2008) Prognostic impor-
tance of anaemia in HIV type-1-infected patients starting
antiretroviral therapy: collaborative analysis of prospective
cohort studies. Antiviral Therapy 13, 959–967.
Kamya MR, Semitala FC, Quinn TC et al. (2004) Total lympho-
cyte count of 1200 is not a sensitive predictor of CD4 lym-
phocyte count among patients with HIV disease in Kampala,
Uganda. African Health Sciences 4, 94–101.
Kumarasamy N, Mahajan AP, Flanigan TP et al. (2002) Total
lymphocyte count (TLC) is a useful tool for the timing of
opportunistic infection prophylaxis in India and other resource-
constrained countries. Journal of Acquired Immune Deficiency
Syndromes 31, 378–383.
Lau B, Gange SJ, Phair JP, Riddler SA, Detels R & Margolick JB
(2003) Rapid declines in total lymphocyte counts and hemo-
globin concentration prior to AIDS among HIV-1-infected men.
AIDS 17, 2035–2044.
Liotta G, Perno CF, Ceffa S et al. (2004) Is total lymphocyte count
a reliable predictor of the CD4 lymphocyte cell count in
resource-limited settings? AIDS 18, 1082–1083.
Liu FR, Guo F, Ye JJ et al. (2008) Correlation analysis on
total lymphocyte count and CD4 count of HIV-infected
patients. International Journal of Clinical Practice 62,
955–960.
Mahajan AP, Hogan JW, Snyder B et al. (2004) Changes in total
lymphocyte count as a surrogate for changes in CD4 count
following initiation of HAART: implications for monitoring in
resource-limited settings. Journal of Acquired Immune
Deficiency Syndromes 36, 567–575.
May M, Boulle A, Phiri S et al. (2010) Prognosis of patients with
HIV-1 infection starting antiretroviral therapy in sub-Saharan
Africa: a collaborative analysis of scale-up programmes. The
Lancet 376, 449–457.
Moore DM, Awor A, Downing RS et al. (2007) Determining
eligibility for antiretroviral therapy in resource-limited settings
using total lymphocyte counts, hemoglobin and body mass
index. AIDS Research and Therapy 4, 1.
Corresponding Author Reinout van Crevel, Department of Internal Medicine, Radboud University Nijmegen Medical Centre,
Route 456, PO Box 9101, 6500 HB Nijmegen, The Netherlands. Tel.: +31 24 361 69 80; Fax: +31 24 356 63 36;
E-mail: R.vanCrevel@aig.umcn.nl
ª 2011 Blackwell Publishing Ltd
13653156, 2011, 11, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/j.1365-3156.2011.02870.x by Nat Prov Indonesia, Wiley Online Library on [28/08/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
volume 16 no 11 pp 1372–1379 november 2011
Morpeth SC, Crump JA, Shao HJ et al. (2007) Predicting CD4
lymphocyte count <200 cells ⁄ mm(3) in an HIV type 1-infected
African population. AIDS Research and Human Retroviruses
23, 1230–1236.
National Aids Commission (2008) Country report on the follow
up to the declaration of commitment on HIV ⁄ AIDS. UNGASS
report 2006–2007.
Nowicki MJ, Karim R, Mack WJ et al. (2007) Correlates of CD4+
and CD8+ lymphocyte counts in high-risk immunodeficiency
virus (HIV)-seronegative women enrolled in the women’s
interagency HIV study (WIHS). Human Immunology 68,
342–349.
Post FA, Wood R & Maartens G (1996) CD4 and total lympho-
cyte counts as predictors of HIV disease progression. QJM 89,
505–508.
Raizes E, Lindegren M, Ellerbrock T, Ferris R & Marston B
(2008) A Comparison of the adult national antiretroviral
therapy (ART) guidelines of 15 PEPFAR focus countries. 2008.
Schreibman T & Friedland G (2004) Use of total lymphocyte
count for monitoring response to antiretroviral therapy. Clinical
Infectious Diseases 38, 257–262.
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.
WHO (2006) Antiretroviral therapy for HIV infection in adults
and adolescents: recommendations for a public health approach.
2006 revision.
WHO (2007) WHO case definitions of HIV for surveillance and
revised clinical staging and immunological classification of
HIV-related disease in adults and children.
WHO (2009) Rapid advice, antiretroviral therapy for HIV
infection in adults and adolescents.
Wisaksana R, Indrati AK, Fibriani A et al. (2010) Response to
first-line antiretroviral treatment among human immunodefi-
ciency virus-infected patients with and without a history of
injecting drug use in Indonesia. Addiction 105, 1055–1061.
1379