O International Epidemiological Association 1998 Printed in Great Britain International Journal ojEpidemiology 1998:27:703-712

CD4+ lymphocytes and tuberculin skin test assurvival predictors in pulmonary tuberculosisHIV-infected patientsCarles Serrat,a Guadalupe G6mez,a Patricia Garcia de Olallab and Joan A Caylab

Background We analyse whether the tuberculin skin test is a good survival marker in a cohortof pulmonary tuberculosis patients with HTV infection (PTB/fflV). In all, 494PTB/HTV patients were enrolled in Barcelona (Spain) between January 1992 andDecember 1994 in the Tuberculosis Program of Barcelona. The main data prob-lem was the large proportion of missing values in the covariates percentage ofT CD4+ lymphocytes and the tuberculin test results: only 157 patients (31.8%)had both covariates recorded.

Methods Patients were dichotomized into two groups according to their level of immuno-suppression (=S14 and >14% T CD4+ cells). First, we carried out the semipara-metric and parametric complete case analysis. After this, we analysed the dataassuming a missing at random non-response pattern. We developed a bootstrapapproach where missing data in the markers are imputed via a two-way linearmodel. Using Weibull regression estimation, we used a multiple imputation schemeto estimate the parameters of interest.

Results We found significative differences for the most immunosuppressed group whencomparing positive tuberculin patients with those who were tuberculin negative.From a complete case approach and through a multivariate Cox analysis, weobtained a significant relative hazard of 0.3657 (95% CI : 0.13-1.02; P = 0.054).When a Weibull model was fitted, we estimated a constant relative percentilevalue of pR = 4.1329 (95% CI : 0.97-17.59). From a missing data approach, weobtain a higher constant relative percentile 5.48 (P = 0.079).

Conclusions The imputation method allows us to assess the protective character of positivityfor the tuberculin test for the lowest CD4+ level. These findings strongly suggestthe value of the tuberculin skin test as a qualitative measure of the immuno-logical response and its interest for developing countries where specific laboratorytests are not affordable.

Keywords Acquired immunodeficiency syndrome, CD4, missing data, survival analysis,tuberculin skin test, tuberculosis

Accepted 8 October 1997

In recent years, the human immunodeficiency virus (HIV) hashad a considerable influence on the incidence of tuberculosis(TB) in both developed and developing countries.1'2 In fart, theWorld Health Organization estimates that of the eight millionnew cases recorded in 1990, 4% were HIV-infected, a percentagethat is expected to rise to 15% in the year 2000.3

' Universitat Politecnica de Catalunya, Barcelona, Spain..b Servel d'Epidemiologia, Institut Municipal de la Salut, Unitat Investigad6

de Tuberculosis, Barcelona, Spain.Reprint requests: Cailes Serrat, Universitat Politecnica de Catalunya, EscolaUniversitaria Politecnica de Barcelona, Dept. de Matematlca Aplicada I. Avgda.Doctor Gregorio Marafi6n, 44-50, 08028-Barcelona, Spain.

The high mortality rate among TB HIV-infected patients(TB/HIV), compared with those without HTV infection, suggeststhat immunodeficiency due to HTV is the cause of such figures.4

Thus it is estimated that in 1996 alone some 226 000 TB/HIVpatients will have died worldwide, and one-third of the HIVfatalities can be directly attributed to TB.5

It has been demonstrated that in TB/HIV patients, thedepletion of T CD4+ lymphocytes constitutes a good marker forthe progression of the disease,6 which seems to gain groundmore rapidly among those individuals with high replicationrates of the virus.7'8

Furthermore, many developing countries with high HIV andMycobacterium tuberculosis coinfection rates9 lack the necessary

703

at NorthW

est Norm

al University on July 13, 2011

ije.oxfordjournals.orgD

ownloaded from

704 INTERNATIONAL JOURNAL OF EPIDEMIOLOGY

resources to carry out laboratory tests such as the determinationof lymphocyte subsets or the viral load. This is one of thereasons why the use of clinical parameters, with inexpensiveand simple tests (e.g. the tuberculin skin test) or straightforwardand practical laboratory markers, would be desirable for estim-ating survival probability.

Survival studies on TB/HTV patients are few and based on com-plete data; that is to say, without taking into account missingdata . 4 ' 6 1 0 However, when the problem one wishes to analysedepends on the behaviour of covariates with missing data, thetreatment of such data is crucial both for conclusions that maybe reached and for decisions that must be taken based on thoseconclusions. The aim of this work, therefore, is to carry out asurvival study by analysing the predictive role played by vari-ables such as the tuberculin skin test and the precentage ofT CD4+ lymphocytes, which both present a large amount ofmissing data. To this end, bootstrap methodology" and amultiple imputation scheme12 '13 will be applied as an initialapproach to the methods of treating variables with missing data.

MethodsIn Barcelona, both acquired immunodeficiency syndrome (AIDS)and TB are monitored actively by the Epidemiology Service ofthe Municipal Institute of Health. This service gathers dataprovided by doctors, as well as results from microbiological ana-lysis, hospital discharges, mortality data and linkage betweenAIDS and TB Registers.

DefinitionsTuberculosisA patient was considered to be tuberculous if he or she hadbeen receiving chemotherapy treatment for TB, and if suchtreatment had been administered for a pre-established period oftime, unless it had been interrupted by either the appearance ofsecondary effects or by death of the patient. In cases of pleuralTB, anatomical, pathological and/or enzymatic criteria14'15 werealso considered.

Tuberculin skin testThis was performed according to the Mantoux technique, ap-plying 2 TU of PPD RT-23 intradermally, and a reading takenbetween 48 to 72 h. Indurations of s«5 mm were consideredpositive.14

Human Immunodeficiency VirusA patient was considered HIV-infected if the ELISA (enzyme-linked immunosorbent assay) test was positive at the time of TBdiagnosis, and if this result was confirmed by Western blot.16

DataThe study population was HIV-infected patients with pulmonarytuberculosis (PTB), with or without extrapulmonary infection,resident in Barcelona city, and diagnosed between 1 January1992 and 31 December 1994. If the culture identified anatypical mycobacteria, the patient was excluded from the study.Patients with a medical record of TB were included only if theyhad received no treatment against TB for more than a year.

The survival time for each patient was established as thenumber of days between the date of TB diagnosis and death, or31 December 1994 for those patients who were still alive on this

date. Consequently, those patients alive when the study wasconcluded, or when loss to follow-up occurred, are consideredto be right censored because death has not been observed.

Study variablesIn addition to the survival time and vital status, other variableswere recorded for each patient: sex, age at the TB diagnosis,district of residence, prison history, HIV transmission group,alcohol addiction, homelessness, result of tuberculin test (PPD:negative = 0, positive =1) , percentage of lymphocyte subsets(CD4and CD8), microbiological results, radiological pattern andsite of TB.

The CD4 and CD8 variables, measured as percentages, werechosen rather than the respective absolute counts because theyhave lower variability and also because it has been shownthat a high correlation exists between the absolute number ofT CD4+ and percentage of T CD4+.17

Missing dataThe percentage of missing data was in general less than 10%.However, there was a large amount of missing data in both thevariable CD4 (38.8%) and in the PPD variable (50.4%). Only31.8% of the data were complete for these two variables. Themethodology we set forth in the following subsection enabledus to make full use of the information found in the other68.2%, where missing data were present.

Statistical analysisAnalysis was carried out in two stages:18 (a) univariate andmultivariate analysis of the complete data, and (b) the treat-ment of missing data and the parametric estimate by multipleimputation.

(a) In the complete data univariate analysis, the Kaplan-Meier method was used to estimate the survival curves, andthe Cox regression model was used to analyse the role of thevarious categories within the covariates. For each variable, themodel was validated by studying the significance of the inter-action of the variable with time, and with the logarithm of time,assuming the hypothesis of proportional hazards if suchinteractions are not significant. For the multivariate analysis,the Cox regression model was set with those covariates thatwere significant from the univariate analysis.

We defined the observed subsample as that consisting ofthe 157 (31.8%) patients for whom the CD4 and PPD valueswere available. We dichotomized the CD4 variable as 0 for thosepatients with a percentage of T CD4+ cells =S14% (correspond-ing to 200 T CD4+ cells/mm3, which is the index below whicha patient is judged to have AIDS);16 and 1 otherwise.

To obtain more predse estimates, we fit a Weibull regressionmodel for each immunosuppression level. The hazard functionat moment t for a patient with PPD value in the tuberculin testis given by the expression

h(t, PPD) = —{e-Ptfcjt

with p = Po + Pt PPD and a = ao + al PPD, where Po and a 0 arethe values for the reference group (negative tuberculin). Theparameters for the positive tuberculin group are p 0 + Pj ando 0 + O[, respectively.

The fitting, graphically validated, provides, among others,the relative n-percentile (0 =£ n =* 1) estimate for a positive

at NorthW

est Norm

al University on July 13, 2011

ije.oxfordjournals.orgD

ownloaded from

SURVIVAL IN PULMONARY TB HIV-INFECTED PATIENTS 705

0 7 5 -

0.60 -

0.25-

0.00-

y v

A,

CO4* ni.|M>»j 14 4SSB> t4_^ 1200

' - , (a)

I

i

i i 1 10 200 400 BOO WO 1000 1200 0 200 400 OO MO 1000 1200

Tin 11M

Figure 1 Kaplan-Meier estimates of survival function for pulmonary tuberculosis-human immunodeficiency virus infected patients, Barcelona(1992-1994), for which CD4+ lymphocytes % and tuberculin test are available (N = 157), stratified by. a) CD4+ lymphocytes %, and b) tuberculintest result

m/n, the proportion of deaths

tuberculin patient with respect to the reference group, pR(Jr;PPD = 1) = e^i [-log(l -Tt)]°i, as a measure of expansion of time(dilation or contraction) between categories.

(b) It is assumed that the response process of the covariatesCD4 and PPD is Missing At Random (MAR),12 i.e. that theprobability of observing the values of CD4 and PPD does notdepend on the missing values of these variables. This assump-tion has been graphically validated.

Let XQ denote the initial incomplete sample, and 9 the relev-ant parameter. To estimate 0, we proceed according to the fol-lowing scheme:

bootstrap imputation estimation

which is repeated 200 times. Each time b, b = 1, .... 200, weobtain a bootstrap replica of the initial sample XQ, which wedenote by X^, then, by means of the bilinear imputationmethod," we get a completed sample, X^ and, finally, using aWeibull regression model, we obtain the corresponding estimateof 0, Qb. This procedure enabled us to obtain 200 estimates(3l---"02OO- The estimator we proposed by multiple imputation is

A _ _ L V 200 -o - 200 ^b-\ w i

with a variance composed of two parts: the /nfravariance due tothe imputation itself and the irtfrrvariance due to the differentimputations. ' So, we obtained confidence intervals for therelative percentiles between groups.

For the basic statistical analysis, SPSS-PC20 and EGRET21

software were used. The bootstrap sampling and the bilinearimputation method have been implemented in PASCAL langu-age, and carried out on a Sun SPARC work station in a UNIXenvironment.

ResultsComplete data analysisThe cohort under study was made up of 494 TB/HIV-infectedpatients. The median age was 32 years (range 17-66 years) and

413 (83.6%) were men. Some 151 (30.8%) came from the leastaffluent district and 146 (29.6%) were or had previously beenin prison at the beginning of the study. By HTV transmissioncategory, 358 (72.5%) were exclusively IVDU, only 5 (1.0%)exclusively homosexual, 11 (2.2%) heterosexual, 94 (19.0%)others and 26 (5.3%) unknown. With regard to TB site, 377(76.3%) were cases of PTB and the rest were pulmonary withextrapulmonary involvement. In all, 102 patients (20.6%) hada history of TB. The radiological pattern was normal in 30 cases(6.2%), abnormal non-cavitary in 357 (73.9%), and cavitary in96 cases (19.9%). The microbiological result was negative in 82(18.5%) patients, smear positive in 193 (43.6%), and theculture exclusively positive in 168 (37.9%). The tuberculin skintest was available for 245 (49.6%) cases; of these, 96 (39.2%)were positive. The percentage of T CD4+ cells was available for302 cases (61.1%). The median CD4+ percentage was 11%,(range 1-86%), the first and third quartile were 5% and 18%respectively, and the mean was 14% (SD: 13.1%).

Overall 63.8% (315 cases) were censored and, among them,only 23 (4.7%) were lost to follow-up. The median follow-uptime for the censored patients was 455 days.

Figure 1 shows the Kaplan-Meier survival estimates accord-ing to immunosuppression level and the result of the tuberculinskin test for the observed subsample. Note that while the firstquartile and the median can be evaluated for the most immuno-suppressed group and for the negative tuberculin group, for theother two groups the median cannot be explicitly derived.

Table 1 presents the results of the univariate Cox model. Itcan be observed that only the T CD4+ percentage, the result ofthe tuberculin test and age are significant.

For the multivariate analysis in the observed subsample (N =157), we only considered the covariates CD4 (continuous andcategorized), PPD and age, and only the dichotomized CD4 vari-able was found to be significant (Table 2). The estimated relativehazard was 0.3198 (95% CI: 0.17-0.60) for the least immuno-suppressed group with respect to the most immunosuppressedgroup.

Using the same model, and based on the 302 patients whohad an observed %CD4 value, independently of the observation

at NorthW

est Norm

al University on July 13, 2011

ije.oxfordjournals.orgD

ownloaded from

7 0 6 INTERNATIONAL JOURNAL OF EPIDEMIOLOGY

Table 1 Estimated relative hazards3 in pulmonary tuberculosis-human immunodeficiency virus infeaed patients, Barcelona (1992-1994)

Variable Observed cases Relative hazard* 9J% CI P-valueb

Sex

Female

Male

Age (years)

Continuous variable

Age (years)

*28

29-32

33-36

»37

District of residence

Others

Inner city

Transmission group

No IVDU*

rvDU

Alcohol addiction

No

Yes

Homelessness

No

Yes

Tuberculin test result

Negative

Positive

CD4+ lymphocytes %

Continuous variable

CD4+ lymphocytes %

494

494

494

CD8+ lymphocytes %

Continuous variable

Smear

Negative

Positive

Positive culture only

Radiological pattern

Normal

Cavitary

Non cavitary

SlteTB

Pulmonary

Mixed

491

468

494

494

245

302

302

299

443

483

494

1

0.9603

1 035

1

1.137

0.9368

1.671

1

1.049

1

0.7962

1

1 107

1

0 7014

1

0 4064

0.9493

1

0.3537

0.9929

1

0.9220

1.324

0.8605

1.359

1.051

0.66-1.41

1.02-1.05

0.75-1.73

0.59-1.48

1.12-2.48

0.76-1.44

0.57-1 12

0 82-1.49

0.33-1.49

0.24-0.68

0.93-0.97

0.23-0.55

0.98-1.01

0.60-1.42

0.86-2 03

0.39-1.89

0.67-2.77

0.75-1.47

0.836

<0.001

0.546

0 778

0.011

0.768

0.188

0.508

0.358

<0.001

<0.001

<0.001

0.273

0.712

0.197

0.709

0.399

0.771

' Relative hazards are obtained using a validated unlvariate Cox proportional hazards model for each variable.b Wald test.c Intravenous drug users.

of the PPD covariate, we were able to estimate a relative hazardsimilar to that mentioned above (RH = 0.3537; 95% CI : 0.23-0.55).

If we divided the observed subsample into rwo groups accord-ing to the immunosuppression level, and if in each group weanalysed the effect of the PPD, CD4 and age covariates, we

observed that in the most immunosuppressed group there wasa significant relative hazard (0.3657, 95% CI : 0.13-1.02; P =

0.054) for patients with positive tuberculin, compared to thosewith negative tuberculin. However, in the least immunosuppres-sed group, and after adjusting for all the mentioned covariates,there was no significant difference between the cases (Table 3).

at NorthW

est Norm

al University on July 13, 2011

ije.oxfordjournals.orgD

ownloaded from

SURVIVAL IN PULMONARY TB HIV-INFECTED PATIENTS 7 0 7

Table 2 P-values on fitting a multivariate Cox proportional hazardsmodel (N = 157) to the pulmonary tuberculosis-humanimmunodeficiency virus infected patients, Barcelona (1992-1994)

Variable

Age (years)

Continuous variable

Age (years)

< 2 8

29-32

33-36

> 3 7

Tuberculin test result

Negative

Positive

CD4+ lymphocytes %

Continuous variable

CD4+ lymphocytes %

> U

Variables

NoneP-valueb

0.140

0.586c

-

0.710

0.733

0.315

0.004c

-

0.004

0.002

<0.001c

<0.001

In the model

Dichotomized CD4+ %a

F-valueb

0.657

0.732c

-

0.363

0.706

0.826

0.129c

-

0.129

0.468

-

_

a This variable is selected in the multivariate Cox model because it has thelowest P-value.

b Wald test.c Overall score test on all degrees of freedom of the variable.

After fitting a Weibull model, that had been graphically valid-ated, for the whole sample the maximum likelihood estimate ofP and a, before including the covariates in the model, is 7.364and 1.411, respectively. Analogously, for the observed subsamplewe obtained values of 7.318 and 1.326, respectively. Table 3 alsoshows the P and a estimates according to the level of immuno-suppression, both before and after recording the effect of thePPD covariate. For the most immunosuppressed group we veri-fied the protective effect of positivity in the tuberculin skin testby estimating the PPD coefficient of P, P, =8.100-6.681 = 1.419(SE: 0.739, P = 0.055) and a constant relative percentile value

of pR = 4.1329 (95% CI : 0.97-17.59), whereas the PPD covari-ate was not significant in the least immunosuppressed group.The corresponding percentile estimates, expressed in days, forthe different relevant subsamples can be seen in Table 4. So, forexample, the 20% percentile was estimated at 190 days for thewhole sample, and at 206 days for the observed subsample,before any covariate adjustments are made. Nevertheless, if weconsider the most immunosuppressed group, and we adjust forthe tuberculin test, we estimate the 20% percentile at 101 daysfor the negative tuberculin patients, and at 417 days for thosewho were positive. Finally, in the least immunosuppressedgroup, the information furnished by the tuberculin test was notsignificant, and we were able to estimate this percentile at 611days. Figure 2 shows both the estimate of the survival functionfor each of the different relevant subsamples, and the timeelongation between categories; one may observe that the groupshowing greatest survival is that made up of the most immuno-suppressed patients with positive tuberculin tests.

Missing data analysisFigure 3(a) shows the survival curves according to whether ornot the CD4+ percentage has been observed. Analogously,Figure 3(b) shows the survival curves according to the observedvalue of the PPD variable.

After obtaining 200 completed bootstrap replicas (N = 494)and estimating the coefficients (Pt and a{) of the PPD variablefor each one, and for each immunosuppressed level, whilethe scale parameter, P t, is found to be significant for most ofthe replicas in the most immunosuppressed group, the shapeparameter, a,, is not. It was therefore decided to fit a Weibullmodel dependent on a scale Pj parameter for each of the 200samples. In this way, 200 P, estimates were obtained with theaverage proposed as estimator. An analogous process wascarried out for the least immunosuppressed group. In this casethe shape parameter Oj was the parameter of greatest interest.Likewise, the average of the 200 a] estimates are proposed asthe shape parameter estimator. Table 5 shows the results of thismethodology.

Table 3 Estimated relative hazards3 and parameters estimates'5 on fitting a Weibull model to the pulmonary tuberculosis-humanimmunodeficiency virus infected patients, Barcelona (1992-1994)

Variable Relative hazard195% CI P-valuec

Observed subsampled (N = 157)

CD4+ lymphocytes %

CD4+ % «S14 subsample (N = 98)

Tuberculin test result

Negative

Positive

CD4+ % > 14 subsample (N = 59)

Tuberculin test result

Negative

Positive

1

0.3198

0.3657

1

1.3032

0.17-0.60

0.13-1.02

0.39^1.33

<0.001

0.054

0.666

6.8727.737

6.6S1

8.100

7 737

7.737

1.3830.8816

1.378

1.378

0.8816

0.8816

' Relative hazards estimates after adjusting a multivariate Cox proportional hazards model.b p and a Weibull maximum likelihood estimates.c Wald test.d The observed subsample Is Integrated by the patients lor whom CD4+ lymphocytes % and tuberculin test are available.

at NorthW

est Norm

al University on July 13, 2011

ije.oxfordjournals.orgD

ownloaded from

7 0 8 INTERNATIONAL JOURNAL OF EPIDEMIOLOGY

Table 4 Estimated percentiles (in days) of the distributions of survival times on fitting a WeibuU model to the pulmonary tuberculosis-humanimmunodeficiency virus infected patients, Barcelona (1992-1994)

CD4 % «S14 subsample

Percentlle (%)(100 JI)

All thesample

Observedsubsample

Negativetuberculin test

Positivetuberculin test

CD4 % >14subsample

10

20

30

40

50

60

70"

66

190

368

612

941

1395

2051

76

206

384

618

927

1342

1928

36

101

193

316

481

707

1029

148

417

796

1306

1988

2921

4255

315

611

923

1268

1659

2122

2699

1 The Weibull regression model has been only validated until the 70th percentlle of suvival time

SurvIvaI

Prob•bIIIt

y

1

0.8

0.6

04

0.2

\"V

V

0 500 1000 1600 2000 2500 3000

Tlnw (days)

Figure 2 Estimated survival functions for pulmonary tuberculosis-human immunodeficiency virus infected patiems, Barcelona(1992-1994), on fitting a Weibull model to: a) all the sample(N = 494), b) cases for whom CD4+ % and tuberculin test are available(N = 157), c) cases with CD4+ % <14 and negative tuberculin(N = 80), d) cases with CD4+ % "S14 and positive tuberculin (N = 18),and e) cases with CD4+ % >14 (N = 59)

Specifically, the P| estimate differs significantly from 0 forthe most immunosuppressed patients, and we can estimate aconstant relative percentile 5.48 (P = 0.079) for the positivetuberculin group compared with the negative tuberculin group.Furthermore, the O] estimate does not differ significantly from0 for the least immunosuppressed patients. Finally, Figure 4shows the confidence intervals for the aforementioned relativepercentiles. The results obtained for the two immunosuppres-sion levels are presented separately in Table 6.

DiscussionMissing data constitute a frequent and on occasions unavoid-able problem in epidemiological studies, and therefore recoursemust be made to sophisticated methodologies. That is the case

in this study. For this reason it is worthwhile consideringthe following two elements: the use of the MAR hypothesis,and the precision of the corresponding estimators.

In our study, if we analyse survival from the point of view ofthe absence or presence of the CD4 variable, Figure 3a) sug-gests, graphically (and validated according to a Cox model) thatthe distribution of patients for whom the CD4 variable is avail-able is basically the same as that for those whom it is unavail-able. In this way we would thus obtain a MCAR response pattern.

If the survival study is carried out assuming that the tuber-culin test is positive, negative or even unavailable, Figure 3b)and the corresponding Cox modelling show that there are nosignificant differences between the negative tuberculin groupand the not available tuberculin group. However, differencesdo exist between these two groups and the positive tuberculingroup. In this case, the PPD observed values are not a randomsubsample of the potential PPD values, although they arerandom within the categories defined by the survival times. Inother words, since the missing data distribution in the PPDvariable depends on the survival time, the response probabilityin the PPD variable is the same for all patients for whom thesame survival time has been observed. In this case, the responsemechanism can be considered ignorable.12 Thus, inferencesbased directly on the observed values of the PPD variable can-not be made since they are not representative. Nevertheless,inferences can be made from the conditional probabilities ofthe PPD variable, given the observed variables. In particular,methods based on maximum likelihood, which is the methodused in this work, are valid, and the said inferences take intoaccount the information referring to the non-observation of thePPD variable at some levels of the survival time variable.

If we consider the precision of the estimators resulting fromthe new methodology, two points are worthy of note; on onehand, at each step of the imputation we recover the size of theinitial sample (N = 494), and this facilitates greater precision inthe estimates. However, the added variance due to the method-ology in each bootstrap replica, and that due to the multipleimputation process itself, must be taken into account.

In recent years the relation existing between a series of demo-graphical, clinical and immunological factors in AIDS patientprognosis has become increasingly dear.22 Nevertheless, studiesbased on TB/HFV are scarce, and up till now somewhat incon-clusive, since the number of cases available for study has beensmall.

at NorthW

est Norm

al University on July 13, 2011

ije.oxfordjournals.orgD

ownloaded from

SURVIVAL IN PULMONARY TB HIV-INFECTED PATIENTS 7 0 9

0 7! -

0.50-

0 . 3 -

0.00-

T-- 1 nlilllluHo 1M/X2

Ym- . iima,

(a)

^ r ~ — r

, 1 , 1

MlJ»l I—ti/U*

o 200 400 no too 1000 1200 0 200 400 no too 1000 1200Tim Hn

Figure 3 Kaplan-Meier estimates of survival function for pulmonary tuberculosis-human immunodeficiency virus infeaed patients, Barcelona(1992-1994), stratified by: a) whether or not CD4+ lymphocytes % is available, and b) tuberculin test result

m/n, the proportion of deaths

Table 5 Parameters estimate8 and estimated relative percentiles of the distributions of survival times on fitting a WeibuU model to thepulmonary tuberculosis-human immunodeficiency virus infected patients, Barcelona (1992-1994)

SubsampleParameter"

Relative percentlleb Estimate ± SE 95% CI P-valuec

CD4+ % *U4

CD4+ % >14

pRd (constant)

pRd 100 It = 30

100 It = 50

100 It = 70e

1.7012 ± 0.9680

5.48

-0.6658 ± 0 5527

1.99

1.28

0.88

-0.20 to 3.60

0.82-36.55

-1.75 to 0.42

0.65-6.07

0.86-1.90

0.72-1.08

0 079

0.228

a Parameters estimates are obtained via multiple imputation using a WeibuU regression model for each subsample in each imputated bootstrap replica.b Relative percentiles are for a positive tuberculin patient with respect to a negative onec Wald test.d Relauve percentiie.c The Weibuil regression model has been only validated until the 70th percentiie of survival time

In this study, it is important to point out the high number ofpatients with PTB studied and the survival time (a median of28.8 months)—greater than that recorded in other studies(median between 12 and 24 months).22 This difference couldwell be accounted for by the fact that TB was detected in theearly stages of HTV infection,23 which would represent a leadtime between 6 and 9 months in AIDS diagnosis.24'25 Moreover,younger AIDS patients have a greater chance of survival.22'26'27

It is well worth noting that more than 75% of all patientsstudied were rVDU, and it is well known that IVDU patients areoften younger than people belonging to other HIV transmissiongroups.2*"30

Furthermore, after fitting different variables such as HIV trans-mission group, presence of opportunistic diseases and therapyagainst HTV, TB is an accelerating factor in the progression ofHIV/AIDS, and patients with TB have a shorter survival thanHIV patients who do not have TB.31 However, TB preventivetherapy delays the appearance of other illnesses in HIV-infectedpatients, while at the same time prolonging survival.32 Con-sequently, it is important to note that early diagnosis and treat-ment of TB is not only beneficial for the community at large,

preventing further cases of infection, but also improves thequality of life and extends the survival time of patients alreadyinfected. To this end, in communities where HIV/Af. tuberculosiscoinfection is high, such as the IVDU group,33 it is necessary topromote programmes of chemoprophylaxis and chemotherapytogether with the administration of methadone to ensure effect-ive treatment and to bring down the incidence of TB and AIDS,as well as to reduce the contagiousness of TB patients.34"36

Previous survival studies show that patients with a highpercentage of T CD4+ have a better prognosis.6'37 In this study,a positive result in the tuberculin test proved to be a bettermarker for survival prediction than the CD4+ percentage inpatients with a T CD4+ lymphocyte percentage lower than14%, which would indicate that the activity of T CD4+ lympho-cytes, evaluated indirectly by means the tuberculin test, is moreimportant than the T CD4+ percentage. In this regard, variousauthors have suggested that the presence of lymphocytes withThl cytokines pattern (H.-2 and IFN-y, among others) of pro-duction would increase chances of survival for HTV positivepatients, since these Thl lymphocytes are the ones involvedin immunity-delay phenomena such as the tuberculin test.38

at NorthW

est Norm

al University on July 13, 2011

ije.oxfordjournals.orgD

ownloaded from

710 INTERNATIONAL JOURNAL OF EPIDEMIOLOGY

pR

30 - -26.94

10

5.48

1.11

5 - -

1 --

- (c)

(b) • - •

A±1

vh

f'-• (a)

0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9

Figure 4 90% confidence intervals for the relative percentiles estimates for a positive tuberculinpatient with respect to a negative one in pulmonary tuberculosis-human immunodeficiency virusinfected patients. Barcelona (1992-1994): a) negative tuberculin group (reference group), b) cases withCD4+ % <14 and positive tuberculin, and c) cases with CD4+ % >14 and positive tuberculin. Relativepercentiles estimates were obtained by multiple imputation using a Weibuli regression model for eachsubsampie in each imputated bootstrap replica

p, percentile; pR, relative percentile.

TSble 6 Comparative study between complete data analysis and the new methodology3 in pulmonary tuberculosis-human immunodefioencyvirus infected patients. Barcelona (1992-1994)

CD4+ % «14

CD4+ % >14

Parameter8

Relative percentlleb

P, .pRd (constant)

Complete data analysis(N = 157)

Estimate (95% CI)

1.419 (-0.03 to 2.87)

4.1329 (0.97-17.59)

No significant differences

VCC %

52.08

New methodology:bootstrap 8- imputation (N

Estimate (95% CI)

1.7012 (-0.20 to 3.60)

5.48 (0.82-36.55)

No significant differences

= 494)

VCC%

56.90

a New methodology is based in multiple imputation using a WeibuU regression model for each subsampie in each imputated bootstrap replicab Relative percentiles are for a positive tuberculin patient respect to a negative onec Variation coefficient.d Relative percentile.

Before the HIV epidemic, 90% of PTB had a positive tuberculintest.39 This suggests that the high prevalence of negative tuber-culin test in TB HIV-infected patients observed in our study isdue to the HIV immunodepression. Therefore, a negative tuber-culin test without CD4 data is indicative of poor prognosis inthese patients (Figure 3b).

The results presented here concur with those of other studieswhich point to the possibility of carrying out an estimate ofsurvival probability in TB patients without necessarily using TCD4+ lymphocyte counts.4 '40 As can be seen from this study,the tuberculin skin test can play an important role, particularlywhen lymphocyte subsets or viral load, for example, are still

at NorthW

est Norm

al University on July 13, 2011

ije.oxfordjournals.orgD

ownloaded from

SURVIVAL IN PULMONARY TB HIV-INFECTED PATIENTS 7 1 1

unavailable, as is the case in countries with the highest levels ofHTV/Af. tuberculosis coinfection; countries which, moreover, dueto their economic situation, are unable to draw upon the tech-nical resources necessary for specific laboratory tests.

AcknowledgementsThis work has been partially supported by the 'Ministerio deEducaci6n y Ciencia' DGICYT project PB95-0776 and by the'Fondo de Investigaciones Sanitarias de la Seguridad Social'grant 92/0528. The authors wish to thank Dr Alvaro Mufiozfrom the School of Hygiene and Public Health of Johns HopkinsUniversity of Baltimore for his suggestions, the GRASS group('Grup de Recerca en Analisi eStadistica de Superviventia':Maria Luz Calle, Anna Espinal and Josep Ginebra) for theircontribution in the discussion phases, the AIDS-TB group fromBarcelona and the Public Health nurses for their collaborationand the anonymous referees and the editor for their helpfulsuggestions. The AIDS-TB group from Barcelona investigators atthe time of this analysis were the following: Barcelona: HospitalValle Hebr6n: Alguero C and Vidal R. Hospital del Mar: L6pez-Colom^s JL. Hospital de Sant Pau: Barrio JL. Hospital Clinic:Soriano E. Servei d'Epidemiologia, Institut Municipal de laSalut: Botia AM, Diez E, Galdos H, Pafiella H, Plasencia A,Brugal T, Miranda P and Jansa JM. Badalona: Hospital GermansTrias i Pujol: Clotet B. L'Hospitalet: Ciutat Sanitaria de Bellvitge:Corominas M and Podzamczer D.

References1 Raviglione M, Snider DE, Kochi A. Global epidemiology of

tuberculosis. JAMA 1995:273:220-26.2 Tuberculosis and AIDS. Statement on AIDS and tuberculosis. Geneve

1989. Global Programme on AIDS and Tuberculosis Programme,World Health Organization, in collaboration with the InternationalUnion Against Tuberculosis and Lung Disease. Bull Int Union TubercLung Dis 1989:64:8-11.

3 World Health Organization TB A Global Emergency. Geneve:WHO/TB/94. 177. 1994.

4 Richter C, Koelemay MJW, Swai ABM et al. Predictive markers ofsurvival in HIV-seropositive and HTV-seronegative Tanzanian patientswith extrapulmonary tuberculosis. Tuberc Lung Dis 1995:76:510-17.

5World Health Organization. TB Groups at Risk. Geneve. WHO, 1996.6 Jones B, Young S, Antoniskls D eta!. Relationship of the manilestations

of tuberculosis to CD4 cell counts in patients with human immuno-deficiency virus infection. Am Rev Respir Dis 1993,148:1292-97.

7 Bravo R, Soriano V, Martin R et al. Cuantificaci6n de la viremia por elvirus de la inmunodefidenda humana en padentes con progresi6nrapida y lenta de la enfermedad. Mcd Clm (Bare) 1995:104:530-34.

8 Cao Y, Qin L, Zhang L et al. Virologic and immunologic character-itzation of long-term survivors of human immunodefidency virustype 1 infection. N Engl J Med 1995:332:201-08.

9De Cock KM. Interaction HTV/TB. Conference on Global Lung Healthand the 1995 Annual Meeting on the IUATLD/UICTMR, Paris. TubercLung Dis 1995;76(SupplJ):5-6.

10 Cayla JA, Jansa JM, Artacoz L et al. Predidors of AIDS in a cohort ofHIV-infeded patients with pulmonary or pleural tuberculosis. TubercLung Dis 1993:74:113-20.

" Efron B. Missing data, imputation and the bootstrap (with discuss-ion). JASA 1994:89:463-79.

1 2 Little RJA, Rubin DB. Statistical Analysts with Missing Data. New York:John Wiley, 1987.

1 3 Rubin DB. Multiple Imputation for Nonresponse in Surveys. New York:John Wiley, 1987.

14Grupo de Trabajo sobre Tuberculosis (FIS-MSC). Consenso nationalpara el control de la tuberculosis en Espafia. Med Clm (Bare) 1992;98:24-31.

Ocafia I, Ribera E, Martinez Vazquez JM. Diagnostic thoracocentesisand pleural biopsy. Ann Intern Med 1996:105:297-98.

16 Centers for Disease Control. 1993 Revised classification system forHIV infedion and expanded surveillance case definition for AIDSamong adolescents and adults. MMWR 1993;41(No RR-17).

17 Phair J, Mufloz A, Detels R et al. The risk of Pneumocystis carinii pneu-monia among men infeded with Human Immunodefidency VirusType l.NEnglJMcd 1990:322:161-65.

1 8Serrat C, G6mez G. EshmatiS de Parametres amb Dades Absents en unEstudi d'Analisi de Supervivencia. Dept d'Estadlstica i InvestigadoOperativa. Unlversitat Politecnica de Catalunya 1995. DR 95/10.

1 9 Collett D Modelling Survival Data in Medical Research. Texts in StatisticalSdence. London. Chapman & Hall, 1994.

20Norussis MJ. SPSS/PV+for the IBM PC/XT/AT. Chicago- SPSS Inc., 1986.2 1 Epidem'wlogical Graphics, Estimation and Testing Package (EGRET) Seattle,

Washington: Statistics and Epidemiology Researdi Corporation, 1990.2 2 Macroft A, Johnson M, Philips A. Fadors affeding survival in patients

with the acquired immunodefidency syndrome. AIDS 1996:10:1057-67.

2 3 Stoneburrer R, Laroche E, Prevots R et al. Survival in a cohort ofhuman immunodefidency virus infeded tuberculosis patients inNew York City. Implications for expansion or the AIDS case definition.Arch Intern Med 1992:152:2033-37.

2 4Ellner J. Tuberculosis in the time of AIDS. Chest 1990:98:1051-52.2 5 Poznasky MC, Coker R, Skinner C et al. HIV positive patients first

presenting with an AIDS defining illness: charaderistics and survival.BrMedJ 1995:311:156-58.

2 6 Darby S, Ewart, Giangrande P et al Importance of age at infedionwith HIV-1 for survival and development of AIDS in UK haemophiliapopulation. Lancet 1996:347:1573-79.

2 7 Vella S, Giuliano M, Florida M et al. Effed of sex, age and transmissioncategory on the progression to AIDS and survival of zidovudine-treated symptomatic patients. AIDS 1995:9:51-55.

2 8 Cayla JA, Jansa JM, Iglesias B et al. Epidemiologfa del sfndrome deinmunodefidenda adquirida en Barcelona (1981—1991 )<I). Estudiodescriptivo y de tendendas temporales. MedClin 1993;101:2'86-93.

2 9 Marco A, Guerrero RA, Rodriguez AM etal. Caraderfsticas de los casosde sida detectados en una prisi6n de Barcelona (1991-1993). AtenadnPrimana 1995,15:487-90.

3 0 Cayla JA, Marcos A, Bedoya A et al Differential charaderistics of AIDSpatients with a history of imprisonment. Int J Epidemiol 1995:24:1188-96.

3 1 Whalen C, Horsburch CR, Horn D etal Accelerated course of humanimmunodefidency virus infedion after tuberculosis. Am J Respir CritCare Med 1995,151:129-35.

3 2 Pape J, Jean J, Ho L et al. Effed of isonlazid prophylaxis on inddenceof active tuberculosis and progression of HIV infection. Lancet 1993;342:268-72.

3 3 de Maron P, Garria A. La evolud6n de la infeeddn VIH/SIDA en lospafses desarrollados. Impado de la tuberculosis. Med Clin (Bare) 1993;100:187-93.

3 4 Curtis R, Friedman S, Nealgus A et al. Implication of directly observedtherapy in tuberculosis control measures among IDUs. Public HealthRep 1994:109:321-27.

3 5 Gourevitch M, Wasserman W, Panero Metal Successful adherence toobserved prophylaxis and treatment of tuberculosis among drug usersin a methadone program. J Addictive Dis 1996:15:93-104.

at NorthW

est Norm

al University on July 13, 2011

ije.oxfordjournals.orgD

ownloaded from

7 1 2 INTERNATIONAL JOURNAL OF EPIDEMIOLOGY

36Raistrick D. Hay A, Wolff K. Methadone maintenance and tubercu- 3 9 Femandez-Nogues F, Garau J, Gudiol F el al. Inddencia y diagnosticolosis treatment. Br Med J 1996;313:925-26. de la tuberculosis pulmonar en un serviao de medidna intema. Med

3 7 Milles G, Jones D. Relation between CD4 lymphocyte count and AIDS ain (Bar^ 1983:81:651-54.

mortality, 1986-91 AIDS 1993;7:1383-«6. '"'whalen CC, Okwera A, Johnson J « al Prognosis and survival inLevy JA. HTV pathogenesis and long-term survival AIDS 1993;7: HTV-infected individuals tuberculosis patients: comparison of US and1401-10. Uganda cohorts. Tubcrc Lung Dis 1994;75(SuppI.l):66.

at NorthW

est Norm

al University on July 13, 2011

ije.oxfordjournals.orgD

ownloaded from