CLINICAL PRACTICE

Absolute Lymphocyte Count in the EmergencyDepartment Predicts a Low CD4 Count inAdmitted HIV-positive PatientsAnthony M. Napoli, MD, Christopher M. Fischer, MD, Jesse M. Pines, MD, MBA, MSCE,Hahn Soe-lin, Munish Goyal, MD, and David Milzman, MD

AbstractObjectives: This study sought to determine if the automated absolute lymphocyte count (ALC) predicts a‘‘low’’ (<200 · 106 cells ⁄ lL) CD4 count in patients with known human immunodeficiency virus (HIV+)who are admitted to the hospital from the emergency department (ED).

Methods: This retrospective cohort study over an 8-year period was performed in a single, urban aca-demic tertiary care hospital with over 85,000 annual ED visits. Included were patients who were knownto be HIV+ and admitted from the ED, who had an ALC measured in the ED and a CD4 count measuredwithin 24 hours of admission. Back-translated means and confidence intervals (CIs) were used todescribe CD4 and ALC levels. The primary outcome was to determine the utility of an ALC threshold forpredicting a CD4 count of <200 · 106 cells ⁄ lL by assessing the strength of association between log-transformed ALC and CD4 counts using a Pearson correlation coefficient. In addition, area under thereceiver operator curve (AUC) and a decision plot analysis were used to calculate the sensitivity, speci-ficity, and the positive and negative likelihood ratios to identify prespecified optimal clinical thresholdsof a likelihood ratio of <0.1 and >10.

Results: A total of 866 patients (mean age 42 years, 40% female) met inclusion criteria. The transformedmeans (95% CIs) for CD4 and ALC were 34 (31–38) and 654 (618–691), respectively. There was a signifi-cant relationship between the two measures, r = 0.74 (95% CI = 0.71 to 0.77, p < 0.01). The AUC was 0.92(95% CI = 0.90 to 0.94, p < 0.001). An ALC of <1700 · 106 cells ⁄ lL had a sensitivity of 95% (95%CI = 93% to 96%), specificity of 52% (95% CI = 43% to 62%), and negative likelihood ratio of 0.09 (95%CI = 0.05 to 0.2) for a CD4 count of <200 · 106 cells ⁄ lL. An ALC of <950 · 106 cells ⁄ lL has a sensitivityof 76% (95% CI = 73% to 79%), specificity of 93% (95% CI = 87% to 96%), and positive likelihood ratioof 10.1 (95% CI = 8.2 to 14) for a CD4 count of <200 · 106 cells ⁄ lL.

Conclusions: Absolute lymphocyte count was predictive of a CD4 count of <200 · 106 cells ⁄ lL in HIV+patients who present to the ED, necessitating hospital admission. A CD4 count of <200 · 106 cells ⁄ lL isvery likely if the ED ALC is <950 · 106 cells ⁄ lL and less likely if the ALC is >1,700 · 106 cells ⁄ lL. Depend-ing on pretest probability, clinical use of this relationship may help emergency physicians predict thelikelihood of susceptibility to opportunistic infections and may help identify patients who should receivedefinitive CD4 testing.

ACADEMIC EMERGENCY MEDICINE 2011; 18:385–389 ª 2011 by the Society for Academic EmergencyMedicine

ª 2011 by the Society for Academic Emergency Medicine ISSN 1069-6563doi: 10.1111/j.1553-2712.2011.01031.x PII ISSN 1069-6563583 385

From the Department of Emergency Medicine, Brown University Medical School (AMN), Providence, RI; the Department ofEmergency Medicine, Beth Israel Deaconess Medical Center (CMF), Boston, MA; the Department of Emergency Medicine (JMP,MG, DM), George Washington University School of Medicine; the Department of Health Policy, George Washington UniversitySchool of Public Health and Health Sciences (JMP), Washington, DC; and the Georgetown University School of Medicine (HS),Washington, DC.Received March 25, 2010; revisions received July 29 and August 17, 2010; accepted September 18, 2010.Presented at the Society for Academic Emergency Medicine annual meeting, Phoenix, AZ, June 2010.The authors have no relevant financial information or potential conflicts of interest to disclose.Supervising Editor: Richard E. Rothman, MD, PhD.Address for correspondence and reprints: Anthony M. Napoli, MD; e-mail: anapoli@lifespan.org.

T he estimated prevalence of known human immu-nodeficiency virus (HIV) is as much as 3% in someurban environments, and the prevalence of undi-

agnosed HIV in the emergency department (ED) rangesfrom 0.7% to 16%.1–4 The CD4 count is a useful predictorof the susceptibility of HIV patients to certain types of ill-nesses; a CD4 count of <200 · 106 cells ⁄ lL signifies agreater risk for opportunistic infections.5 Emergencycare for patients with HIV can be challenging becausethe differential diagnoses and treatment for acute illnessmay vary based on the level of immune suppression.

The CD4 count is often unknown in the ED becausetesting is not rapidly available and patient recollectionor medical record reviews are sometimes not helpful.The inability to accurately assess immunosuppressionin HIV patients may lead emergency physicians toeither overuse or underuse resources aimed at specificopportunistic infections, such as antibiotics for Pneu-mocystis jiroveci. Several studies have shown that theabsolute lymphocyte count (ALC), a commonly availableED test, is a reliable predictor of a low CD4 count.6–10

However, these studies have included U.S. clinics,6,9

underserved regions of the world,7 or a mixture ofinpatient, outpatient, and ED patients.10 To our knowl-edge, no study has assessed the relationship betweenALC and CD4 count in a cohort of only ED patients.

We sought to determine the utility of the ALC in pre-dicting CD4 counts of HIV patients presenting to theED. Our primary hypothesis was that the ED ALCwould predict the CD4 count in HIV+ ED patients beingadmitted to the hospital. Our secondary hypothesis wasthat this relationship would be strong enough such thatone could use the ALC as a surrogate for the CD4 inclinical decision-making.

METHODS

Study DesignThis was a retrospective, cohort study of consecutiveHIV+ patients admitted through the ED. The institu-tional review board approved this study with waiver ofwritten informed consent.

Study Setting and PopulationThe study was conducted at an urban academic Level Itrauma center (annual ED census >85,000) betweenNovember 30, 2001, and November 30, 2009. Patientswere included if they 1) were admitted to the hospitalthrough the ED, 2) had a known diagnosis of HIV byInternational Classification of Diseases (ICD-9) code, 3)had an ALC measured in the ED, and 4) had a CD4count measured within 24 hours of the ALC. The mainoutcome was a ‘‘low’’ CD4 count, defined as <200 · 106

cells ⁄ lL.5

Study ProtocolThe chart review of demographic, laboratory, and clini-cal data used an electronic medical record (Amalga,Redmond, WA). The primary hypothesis, inclusion andexclusion criteria, and desired data variables were alldefined prior to data abstraction. A trained chart abs-tractor with extensive experience using the electronicmedical record, who was blinded to the hypothesis of

the study, extracted the data based on the predefinedinclusion and exclusion criteria. Inter-rater reliabilitywas not assessed because all data extracted were objec-tive data and did not require any interpretation otherthan application of the search filter within the elec-tronic medical record. One of the authors of the study(not blinded to the hypothesis) confirmed that the cor-rect filter had been applied prior to data analysis. In thestudy hospital, the ALC was measured using a FACS-Calibur (Becton Dickson Immunocytometry Systems,San Jose, CA). CD4 lymphocyte counts were analyzedutilizing a Sysmex XT-1800 (Sysmex, Inc., Mundelein,IL). Data were extracted into a standard Excel database(Microsoft Corp., Redmond, WA).

Data AnalysisPatient age was normally distributed and is reported asmean and standard deviation (SD). To reduce the effectsof skewness in the data distribution, data were log-transformed for analysis. Back-transformed means and95% confidence intervals (CIs) were reported for theCD4 and ALC. Because some CD4 counts were zero,log transformation of the CD4 was done after addingone to each CD4 count to avoid undefined numbers. Ap-value of <0.05 was considered significant. A Pearsoncorrelation coefficient was used to assess if a significantrelationship exists between ALC and CD4, the primarystudy hypothesis. Using an expected correlation ofr = 0.75 and a null correlation of r = 0.7 (a = 0.05,b = 0.9), the estimated necessary sample size wasapproximately 944 patients. Based on a single-yearsampling, and excluding repeated analyses of a singlepatient, we estimated this would require 8 years ofrecord review. To test our secondary hypothesis, thearea under the receiver operator curve (AUC) was cal-culated using the Wilcoxon method. Using decision plotanalysis, we calculated the sensitivity, specificity, andpositive and negative likelihood ratios across all ALCfor a CD4 count of <200 · 106 cells ⁄ lL. Bayesian theorysuggests that a positive likelihood ratio of >10 and neg-ative likelihood ratio of <0.1 for a CD4 count <200 · 106

cells ⁄ lL would significantly affect clinical decision-making, so we identified ALC cutoffs that correspondedto these values from the decision plot and present themost salient points as proportions.

RESULTS

A total of 1,546 patients were identified over an 8-yearperiod who were treated in the ED with an ICD-9–coded diagnosis of HIV. A portion of those were admit-ted (82.4%) leaving a total of 1,274 records in 936unique patients. The first ED admission was included inthe case of duplicate patient admissions. Of the 936unique patients, 14 left against medical advice, and twoexpired prior to obtaining a CD4 count. Fifty-fourpatients were excluded because an ALC or CD4 countwas not drawn. The remaining 866 unique patients fullymet inclusion criteria during the study period. Themean (±SD) age was 42 (±10) years, 40% were female,and 91% were African American. The transformedmean for CD4 was 34 (95% CI = 31–38) and for ALCwas 654 (95% CI = 618–691); the raw ranges were

386 Napoli et al. • ABSOLUTE LYMPHOCYTE COUNT IN HIV+ PATIENTS

0–1116 and 12–7725, respectively. A CD4 count of<200 · 106 cells ⁄ lL was present in 84.6% (95%CI = 82.2% to 87.0%) of the patients.

There was a significant relationship between the ALCand CD4, r = 0.74 (95% CI = 0.71 to 0.77). A linearregression through the log-transformed data demon-strates a linear relationship; this is apparent in the scat-ter plot in Figure 1. The AUC was 0.92 (95% CI = 0.90 to0.94, p < 0.001; Figure 2). The sensitivity, specificity,positive likelihood ratio, and negative likelihood ratiosat certain ALC are shown in Table 1. An ALC of<1,700 · 106 cells ⁄ lL has a sensitivity of 95% (95%CI = 93% to 96%), specificity of 52% (95% CI = 43% to62%), and negative likelihood ratio of 0.09 (95%CI = 0.05 to 0.2) for a CD4 count of <200 · 106 cells ⁄ lL.Of all patients with a CD4 count <200 · 106 cells ⁄ lL,95% (95% CI = 93% to 96%) had an ALC of <1,700 · 106

cells ⁄ lL (Table 2). An ALC <950 · 106 cells ⁄ lL had a sen-

sitivity of 76% (95% CI = 73% to 79%) and specificity of93% (95% CI = 87% to 96%), with a positive likelihoodratio of 10.1 (95% CI = 8.2% to 14%) for a CD4 count of<200 · 106 cells ⁄ lL. Of all patients with a CD4 count of>200 · 106 cells ⁄ lL, 93% (95% CI = 87% to 96%) had anALC of >950 · 106 cells ⁄ lL (Table 3).

DISCUSSION

Rapid diagnostic testing for the presence of HIV in EDpatients is becoming common; however, rapid assess-ment of the risk of opportunistic infections by measur-ing CD4 counts in the ED is limited because the testcan take hours to run, and most laboratories do nothave the resources to immediately run it. Early andappropriate treatment of opportunistic infections is akey aspect of the care of the HIV patient in the ED.We have found a simple, readily available measure to

Figure 1. Scatterplot of CD4 count as a function of ALC.ALC = absolute lymphocyte count.

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

True

pos

itive

rate

(Sen

sitiv

ity)

False positive rate (1-Specificity)

Figure 2. The ROC for ALC as a predictor of a CD4 count lessthan 200 · 106 cells ⁄ lL (ROC = 0.92, 95% CI = 0.9 to 0.94,p £ 0.0001). ALC = absolute lymphocyte count; ROC = receiveroperating curve.

Table 1Sensitivity, Specificity, and Likelihood Ratios (LR) for ALCPredicting a CD4 Count of <200 · 106 cells ⁄ lL

ALC (·106

Cells ⁄ lL)Sensitivity

(%)Specificity

(%)Positive

LRNegative

LR

750 63 97 28.3 0.37950 76 93 10.1 0.26

1000 78 90 7.4 0.241250 87 75 3.5 0.161500 93 62 2.4 0.121700 95 52 2.0 0.092000 97 36 1.5 0.07

Sensitivity = true positives ⁄ (true positives + false negatives);specificity = true negatives ⁄ (true negatives + false positives);positive likelihood ratio = sensitivity ⁄ (1 – specificity); nega-tive likelihood ratio = (1 – sensitivity) ⁄ specificity.ALC = absolute lymphocyte count.

Table 22 · 2 Table for Diagnostic Test Results for a Cutoff Likelihoodratio < 0.1

CD4 < 200 CD4 ‡ 200 Total

ALC < 1700 696 64 760ALC ‡ 1700 37 69 106Total 733 133 866

ALC = absolute lymphocyte count.

Table 32 · 2 Table for Diagnostic Test Results for a Cutoff LikelihoodRatio >10

CD4 < 200 CD4 ‡ 200 Total

ALC < 950 557 10 567ALC ‡ 950 176 123 299Total 733 133 866

ALC = absolute lymphocyte count.

ACAD EMERG MED • April 2011, Vol. 18, No. 4 • www.aemj.org 387

estimate a CD4 count in HIV+ ED patients based on theALC. This relationship has been reported previously,although not in a cohort of only ED patients. Priorstudies have demonstrated this relationship in cohortswithout acute illness or in mixed populations. Multiplefactors, particularly the acuity of illness and acutecomorbidities, may affect the lymphocyte count and theaccuracy of this relationship. Thus, demonstrating aclinically useful relationship in ED patients with acuteillness is important before using certain thresholds inclinical decision-making. Because clinicians cannot cur-rently obtain a CD4 count, and patients may often beunaware, it is important to validate this relationship ina select group of patients where clinical use of the rela-tionship may be important.

By only studying the relationship between the ALCand CD4 count in this group of patients, we hoped totarget the spectrum of patients in which clinician uncer-tainty led to ordering the CD4 within the first 24 hours.Discharged patients, or admitted HIV+ patients with arecent CD4 count, would not likely undergo testing ofthe CD4 count early in hospitalization. Augmentation ofclinical treatment of acutely ill patients requiring admis-sion who have an unknown CD4 count is likely a moreimportant population to study because knowledge ofthe risk of opportunistic infections may alter the earlycourse of hospital care.

Our results demonstrate a strong relationshipbetween ALC and CD4 in HIV+ patients, with a highercorrelation (r = 0.74) than prior studies.7,8,10 Our cutoffsare similar to prior studies where ALC cutoffs ofapproximately 1,100 · 106 to 1,300 · 106 and 2,000 · 106

cells ⁄ lL have been proposed to rule in and rule out a‘‘low’’ CD4, although the results should be viewed aspart of a continuum.6,7,10 These results can be used as atool in conjunction with Bayesian decision-making andconsideration of pretest probability. It is generally rec-ognized that likelihood ratios of >10 or <0.1 may signifi-cantly affect clinical decision-making, particularly withintermediate pretest probability.11 For example, a likeli-hood ratio of 0.1 means an ALC greater than that valuewould be 10 times less likely to come from an individualwith a CD4 count <200 · 106 cells ⁄ lL. An ALC of>1,700 · 106 cells ⁄ lL has a negative likelihood ratio of0.1, and an ALC of <950 · 106 cells ⁄ lL has a positivelikelihood ratio of 10.1. Because of the high prevalenceof a low CD4 in our cohort, one cannot use a cutoff ofALC >1,700 · 106 cells ⁄ lL to rule out disease despite itshigh sensitivity; the posttest probability would still be34%. An ALC of <950 · 106 cells ⁄ lL, however, wouldhave a 98% posttest probability of disease and confirma low CD4 when suspected.

Prior studies have documented various aspects of therelationship between the ALC and the CD4 count.6–10

The reported prevalence of a CD4 count <200 · 106

cells ⁄ lL in outpatient or mixed patient populationsranges from 16% to 40%.6,10 The similar cutoffs acrossstudies suggest that there is little spectrum bias. Thus,in ED cohorts where the prevalence of a low CD4 countis <40%, an ALC of >1,700 · 106 cells ⁄ lL would reducethe probability of a CD4 count of <200 · 106 cells ⁄ lL to5%. It remains to be seen what the general prevalenceof a low CD4 in HIV+ patients is across multiple EDs. If

the prevalence is as high as our cohort of admittedpatients, then the ALC would likely be useful to con-firm, but not exclude a low CD4.

The 2 · 2 tables allow for reporting of the sensitivityand specificity for the two chosen cutoffs in a moresimplified proportional approach. The sensitivity repre-sents the proportion of patients with a CD4 count<200 · 106 cells ⁄ lL (sum of first column) who also hadan ALC of <1,700 · 106 cells ⁄ lL. In Table 2, this is 696of 733, or 95%. The specificity represents the propor-tion of patients with CD4 count of ‡200 (sum of secondcolumn) who also had ALC of ‡1,700 · 106 cells ⁄ lL. InTable 2, this is 69 of 133, or 52%. This same processcan be applied at a different cutoff as in Table 3, whereone finds that the proportion of patients with aCD4 count of >200 · 106 cells ⁄ lL (sum of second col-umn) that had an ALC of >950 · 106 cells ⁄ lL would be123 of 133, or 93%. Taken as a whole this approachsuggests it is unlikely to have a CD4 count of <200 · 106

cells ⁄ lL if the ALC is greater than 1,700 · 106 cells ⁄ lL,but likely if it is less than 950 · 106 cells ⁄ lL.

Clinicians may choose to use this information in con-junction with the clinical scenario to empirically treatfor opportunistic infections in HIV+ patients. Oneshould not interpret these findings to mean thatpatients with a low ALC in the ED should be empiri-cally treated for an opportunistic infection, unless thereis a high suspicion of HIV, since all studies done todate, including this one, have been in known HIVpatients. Studies in undifferentiated ED patients with alow ALC would be helpful to assess the value of ALC inpredicting a patient with HIV and a low CD4 count.However, these results may be useful in guiding thetreatment of patients with suspected HIV when rapidtesting is available.

LIMITATIONS

This study is limited by its retrospective design, poten-tial for selection bias, and single-center nature. Study-ing only known HIV+ admitted ED patients with CD4counts could lead to substantial selection bias. We envi-sion three potential selection biases. First, dischargedpatients were excluded because 24-hour CD4 countswere unavailable. Second, we also excluded knownHIV+ admitted patients in whom a CD4 count was notordered, presumably because of previous knowledgeon the part of the clinician or patient of a recent CD4count. Third, patients with an unknown or unsuspectedHIV+ serostatus at the time of admission could not beassessed using this study design and were thereforeexcluded. However, we believe known HIV+ patientswith unknown CD4 counts represent the patients forwhom the clinician would most likely use the ALC andCD4 count relationship in decision-making. The factthat our results are similar to those reported by otherstudies suggests selection bias is not a major factorinfluencing our results. Broad inclusion of all HIV+patients, regardless of their disposition or knowledgeof a recent CD4 count, would likely lead to spectrumbias by including these patients in determining a clinicalthreshold that would only be used in situations ofclinical uncertainty.

388 Napoli et al. • ABSOLUTE LYMPHOCYTE COUNT IN HIV+ PATIENTS

Although the CD4 count was measured within24 hours of presentation, it is unclear whether it maybe partly affected by the acuity of illness, the type of ill-ness, or the therapy that ensued in the first 24 hours.Demographic factors (age, sex, weight, presence ofhighly active antiretroviral therapy) have been shownto correlate with CD4 counts and may affect the accu-racy of this prediction instrument.7,9 Because our popu-lation consisted primarily of African Americans, thismay limit extension of our results to other ethnicities.However, previous work has demonstrated these cova-riates had a very modest affect on the ability to predicta ‘‘low’’ CD4.9 We believe that these differences wereminimal and chose to evaluate this relationship inde-pendent of these factors, as a clinician would at thebedside. Last, generalizing these results to other EDsmay be limited, as 85% of the patients had a CD4 of<200 · 106 cells ⁄ lL, although this may be partiallyreflective of the acuity of illness in an admitted HIV+ED population.

CONCLUSIONS

Acute lymphocyte count was predictive of a CD4 countof <200 · 106 cells ⁄ lL in human immunodeficiencyvirus–positive patients who present to the ED necessi-tating hospital admission. A CD4 count of <200 · 106

cells ⁄ lL is very likely if the ED absolute lymphocytecount is <950 · 106 cells ⁄ lL and less likely if the abso-lute lymphocyte count is >1,700 · 106 cells ⁄ lL. Depend-ing on pretest probability, clinical use of thisrelationship may help emergency physicians predictthe likelihood of susceptibility to opportunistic infec-tions and may help identify patients who shouldreceive definitive CD4 testing.

The authors are indebted to Dr. Mark Smith for his developmentof the software necessary for completion of this work. His visionand continued support made this work possible.

References

1. Government of the District of Columbia. District ofColumbia HIV ⁄ AIDS Epidemiology Update 2008.Available at: http://doh.dc.gov/doh/lib/doh/pdf/dc_hiv-aids_2008_updatereport.pdf. Accessed Jan 20,2011.

2. Lyss SB, Branson BM, Kroc KA, Couture EF, New-man DR, Weinstein RA. Detecting unsuspected HIVinfection with a rapid whole-blood HIV test in anurban emergency department. J Acquir ImmuneDefic Syndr. 2007; 44:435–42.

3. Lyons MS, Lindsell CJ, Ledyard HK, Frame PT,Trott AT. Health department collaboration withemergency departments as a model for publichealth programs among at-risk populations. PublicHealth Reports. 2005; 120:259–65.

4. Kelen GD, Hexter DA, Hansen KN, et al. Feasibilityof an emergency department-based, risk-targetedvoluntary HIV screening program. Ann EmergMed. 1996; 27:687–92.

5. Centers for Disease Control and Prevention. UnitedStates Public Health Service ⁄ Infectious DiseaseSociety of America guidelines for the preventionand treatment of opportunistic infections in HIV-infected adults and adolescents. MMWR RecommRep. 2009; 58:1–207.

6. Blatt SP, Lucey C, Butzin CA, Hendrix C, Lucey D.Total lymphocyte count as a predictor of absoluteCD4 count and CD4 percentage in HIV-infected per-sons. JAMA. 1993; 269:622–6.

7. Kumarasamy N, Mahajan AP, Flanigan TP, et al.Total lymphocyte count (TLC) is a useful tool for thetiming of opportunistic infection prophylaxis inIndia and other resource-constrained countries.J AIDS. 2002; 31:378–83.

8. Beck EJ, Kupek EJ, Gompels MM, Pinching AJ.Correlation between total and CD4 lymphocytecounts in HIV infection: not making the good anenemy of the not so perfect. Int J STD AIDS. 1996;7:422–8.

9. Mwamburi DM, Ghosh M, Fauntleroy J, GorbachSL, Wanke CA. Predicting CD4 count using totallymphocyte count: a sustainable tool for clinicaldecisions during HAART use. Am J Trop Med Hyg.2005; 73:58–62.

10. Shapiro N, Karras DJ, Leech SH, Heilpern KL.Absolute lymphocyte count as a predictor of CD4count. Ann Emerg Med. 1998; 32:323–8.

11. Gallagher EJ. Clinical utility of likelihood ratios.Ann Emerg Med. 1998; 31:391–7.

ACAD EMERG MED • April 2011, Vol. 18, No. 4 • www.aemj.org 389