This article was downloaded by:[Universitetsparken]On: 12 January 2008Access Details: [subscription number 787828591]Publisher: Informa HealthcareInforma Ltd Registered in England and Wales Registered Number: 1072954Registered office: Mortimer House, 37-41 Mortimer Street, London W1T 3JH, UK

Scandinavian Journal of InfectiousDiseasesPublication details, including instructions for authors and subscription information:http://www.informaworld.com/smpp/title~content=t713690438

TBscore: Signs and symptoms from tuberculosispatients in a low-resource setting have predictive valueand may be used to assess clinical courseChristian Wejse ab; Per Gustafson c; Jens Nielsen a; Victor Francisco Gomes a;Peter Aaby a; Paul Lehm Andersen b; Morten Sodemann ada From the Bandim Health Project, Statens Serum Institute, Bissau, Guinea Bissaub Infectious Disease Research Unit, Skejby, Aarhus University Hospital, Denmarkc Infectious Diseases Research Group, Department of Clinical Sciences, LundUniversity, Malmö, Swedend Department of Infectious Diseases, Copenhagen University Hospital, Denmark

First Published on: 03 August 2007To cite this Article: Wejse, Christian, Gustafson, Per, Nielsen, Jens, Gomes, Victor Francisco, Aaby, Peter, Andersen,Paul Lehm and Sodemann, Morten (2007) 'TBscore: Signs and symptoms from tuberculosis patients in a low-resourcesetting have predictive value and may be used to assess clinical course', Scandinavian Journal of Infectious Diseases,40:2, 111 - 120To link to this article: DOI: 10.1080/00365540701558698URL: http://dx.doi.org/10.1080/00365540701558698

PLEASE SCROLL DOWN FOR ARTICLE

Full terms and conditions of use: http://www.informaworld.com/terms-and-conditions-of-access.pdf

This article maybe used for research, teaching and private study purposes. Any substantial or systematic reproduction,re-distribution, re-selling, loan or sub-licensing, systematic supply or distribution in any form to anyone is expresslyforbidden.

The publisher does not give any warranty express or implied or make any representation that the contents will becomplete or accurate or up to date. The accuracy of any instructions, formulae and drug doses should beindependently verified with primary sources. The publisher shall not be liable for any loss, actions, claims, proceedings,demand or costs or damages whatsoever or howsoever caused arising directly or indirectly in connection with orarising out of the use of this material.

Dow

nloa

ded

By:

[Uni

vers

itets

park

en] A

t: 18

:46

12 J

anua

ry 2

008

ORIGINAL ARTICLE

TBscore: Signs and symptoms from tuberculosis patients in alow-resource setting have predictive value and may be used to assessclinical course

CHRISTIAN WEJSE1,2, PER GUSTAFSON3, JENS NIELSEN1,

VICTOR FRANCISCO GOMES1, PETER AABY1, PAUL LEHM ANDERSEN2 &

MORTEN SODEMANN1,4

From the 1Bandim Health Project, Statens Serum Institute, Bissau, Guinea Bissau, 2Infectious Disease Research Unit,

Skejby, Aarhus University Hospital, Denmark, 3Infectious Diseases Research Group, Department of Clinical Sciences, Lund

University, Malmo, Sweden, and 4Department of Infectious Diseases, Copenhagen University Hospital, Denmark

AbstractWe developed a clinical score to monitor tuberculosis patients in treatment and to assess clinical outcome. We used theWHO clinical manual to choose signs and symptoms, including cough, haemoptysis, dyspnoea, chest pain, night sweating,anaemia, tachycardia, lung-auscultation finding, fever, low body-mass index, low mid-upper arm circumference givingpatients a TBscore from 0 to 13. We validated the score with data from a cohort of 698 TB patients, assessing sensitivity tochange and ability to predict mortality. The TBscore declined for 96% of the surviving patients from initiation to end oftreatment, and declined with a similar pattern in HIV-infected and HIV-uninfected patients, as well as in smear negative andsmear positive patients. The risk of dying during treatment increased with higher TBscore at inclusion. For patients with aTBscore of �8 at inclusion, mortality during the 8 months treatment was 21% (45/218) versus 11% (55/480) for TBscoreB8 (pB0.001). TBscore assessed at end of treatment also strongly predicted subsequent mortality. The TBscore is a simpleand low-cost tool for clinical monitoring of tuberculosis patients in low-resource settings and may be used to predictmortality risk. Low TBscore or fall in TBscore at treatment completion may be used as a measure of improvement.

Introduction

Clinical outcome for TB patients is difficult to

measure, especially for smear negative patients.

The World Health Organization (WHO) defines

TB outcomes as Cure, Treatment completed,

Death, Failed, Defaulted, Transferred out, Not

accounted for [1], but the definition for cure only

applies to the 40%�70% of TB patients with a

positive microscopy before treatment initiation.

Weight gain, sedimentation rate, Karnofsky score

and respiratory symptoms have been used as out-

comes in trials [2,3] and recently a symptom count

rate was suggested to evaluate treatment response

[4], but a simple clinical measure of TB patients for

settings in which laboratory access is scarce does not

exist. A number of score systems are available but

only for diagnostic purposes [5,6]. A specific TB

clinical scale may be a useful tool for assessing

prognosis and for measuring outcome in clinical

trials.

The purpose of the present study was to develop a

simple clinical score for repeated clinical status

evaluation of TB patients during treatment without

using advanced technical equipment. We hypothe-

sized that scoring symptoms, signs and anthropo-

metric values of the patients would describe patients

sufficiently to monitor change during TB treatment

and predict mortality.

Study population and methods

Patients

For empirical validation of the TBscore, we used

a cohort of 698 pulmonary TB patients from

Correspondence: C. Wejse, Department of Infectious Diseases, Aarhus University Hospital, Brendstrupgaardsvej, 8200 Aarhus N, Denmark. Tel: �45

89498492. Fax: �45 89498490. E-mail: wejse@dadlnet.dk

Scandinavian Journal of Infectious Diseases, 2008; 40: 111�120

(Received 29 January 2007; accepted 30 June 2007)

ISSN 0036-5548 print/ISSN 1651-1980 online # 2008 Informa UK Ltd. (Informa Healthcare, Taylor & Francis As)

DOI: 10.1080/00365540701558698

Dow

nloa

ded

By:

[Uni

vers

itets

park

en] A

t: 18

:46

12 J

anua

ry 2

008 an epidemiological study in Guinea Bissau in

1996�2001, as described elsewhere [7�9]. The study

district has been followed demographically through a

census system since 1978 [10]. The population

currently consists of 92,000 individuals.

Direct microscopy of 3 morning sputum samples

and chest X-rays were performed. Cases having 2 or 3

sputum smears positive for acid-fast bacilli (AFB) in

direct microscopy were regarded as smear positive

tuberculosis. In 27 cases, only 1 sputum was positive

for AFB; all of these patients had X-ray changes

compatible with TB and they were recorded as smear

positive. The positive smears were graded with (�),

(��) or (���) according to the visual presenta-

tion of number of AFB in the smear by standard

methods [11]. If clinical signs, symptoms and X-ray

changes were compatible with active pulmonary

tuberculosis, and no improvement was seen following

treatment with antibiotics, patients were diagnosed

with smear negative tuberculosis. Six per cent

(42/698) could not produce a sputum sample; they

were coded as smear negative tuberculosis. There is a

high frequency of smear negative patients in this

material (57%), which is comparable to other high-

incidence, high HIV-prevalence areas [12]. Further-

more, a civil war in the area in 1998�99 destroyed TB

culture and other diagnostic facilities [8].

A 4-month intensive phase of daily directly

observed treatment with ethambutol, isoniazid,

rifampicin, and pyrazinamide was followed by a

4-month continuation phase with isoniazid and

ethambutol collected at the health centre twice per

month by the patient. This treatment regimen was

recommended for HIV-infected individuals by the

national TB programme in Guinea-Bissau when the

research project was initiated in 1996. For reasons of

confidentiality and comparability, HIV-infected and

uninfected individuals received the same treatment.

Adherence to treatment was verified by pill count

by the nurses supplying medication and an isoniazid

urine test was performed at the hospital at 2, 5 and 8

months of follow-up. Patients who did not come to

collect their medicine were visited by the nurse and

encouraged to continue treatment.

Patients were interviewed using standardized

questionnaires and clinical examination, as de-

scribed elsewhere [7,9]. Symptoms were registered

after the decision to treat for TB.

Human immunodeficiency virus (HIV) status was

obtained for 676 of the patients [9]. No anti-

retroviral treatment was available in Bissau during

the study period.

Patients were followed from initiation of treatment

until death, loss to follow-up or at the end of 8

months treatment, whichever came first. Addition-

ally, patients were followed for mortality 18 months

after treatment initiation. Field assistants registered

mortality status at later home visits even outside the

study area if the patient had moved, the last visits

being conducted ultimo 2003. If the patient had died

the death date was noted, if alive the last visit date

was registered; hence a mortality status was available

at 8 and 18 months after treatment initiation. Verbal

autopsies (an interview of the relatives by a physician

to establish the likely cause of death) were conducted

and none of the deaths was due to violence [8].

Definitions

We assessed the following outcomes [13,14]:

. Sensitivity to change: ability of the TBscore to

detect a change in clinical state during treat-

ment

. Mortality: ability to predict death

. Improvement: ability to detect better clinical

status at end of treatment

Improvement was defined as completing treatment

and having less than 5 points (SC-I) or dropped

�25% in score compared with inclusion. The basis

for this definition was the clinical judgement that if

few symptoms were present at the end of treatment

or a substantial reduction in initial symptoms was

seen, the patient would be valued to be in good

health.

The following clinical findings comprised the

TBscore:

Symptoms

Self-reported cough, dyspnoea, night sweats, hae-

moptysis and chest pain: patients were asked in the

local language by a trained trial nurse for these

symptoms at the clinical examination.

Signs

Anaemia: paleness of conjunctivae at eye-examina-

tion. Tachycardia: pulse �90/min. Positive finding

at lung auscultation: any 1 of the following findings

present � crepitation, rhonci, subdued or complete

absence of respiratory sounds. Axillary temperature:

temperature �37.08C [15] measured by an electro-

nic thermometer in a closed axillary fold. Body mass

index (BMI): height measured by a meter scale and

at each visit weight was measured using the same

weight scale. BMI�weight/(height)2.

Mid upper arm circumference (MUAC): mea-

sured over biceps of the non-dominant arm by a

non-stretchable measuring tape (TALC, Guildford)

[16].

112 C. Wejse et al.

Dow

nloa

ded

By:

[Uni

vers

itets

park

en] A

t: 18

:46

12 J

anua

ry 2

008 Score

We based the final decision regarding items to

include in the TBscore on the following considera-

tions [14]:

A. Parameters should be repeatedly measurable in

a low-resource setting and subject to change

over time. Hence, known risk factors for

increased mortality, such as gender and HIV

status, were not included.

B. The WHO clinical manual for TB and HIV was

used to select items relevant for the clinical

score [11]. The manual lists: cough �2 weeks,

haemoptysis, sputum production, weight loss,

chest pain, shortness of breath, fever, night

sweating, fatigue, loss of appetite, tachycardia,

clubbing and auscultory findings. We had not

collected data on sputum production, loss of

appetite and presence of fatigue, clubbing was a

very rare event in this cohort and was not

considered. We decided not to use reported

weight loss in the TBscore as nearly all � 675

out of 688 � reported weight loss at the primary

examination. Furthermore, at 2 months, con-

trol 74/87 (85%) of those indicating a contin-

ued weight loss had in fact gained weight when

measured on a scale.

C. Instead of using reported weight loss and loss of

appetite, we decided to use the objective

measures, Body Mass Index (BMI) and Mid

Upper Arm Circumference (MUAC). Whereas

the other variables by default were categorical

variables, BMI and MUAC were chosen to be

ordinal to be able to discriminate more clearly

the occurrence of weight gain. Decreasing BMI

is known to be associated with increased

mortality [17�19], therefore a patient could

obtain 2 points if BMI or MUAC were very

low, allowing for patients being able to change

score when gaining weight. Cut-offs for BMI

were chosen according to van der Sande and

Zachariah [18,19]. Cut-off for MUAC was set

at wasting B220 mm according to Villamor

[20] and we further defined severe wasting as

MUAC B200 mm.

The score was therefore based on the 11 clinical

variables shown in Table I, each contributing 1

point, and maximum score was 13. For all variables

examined and used in the score, missing values were

assumed to be normal (i.e. zero). Missing values

occurred 2�10 times for each parameter, except

BMI where 31 patients had missing values at

inclusion and an average of 15 patients missing at

later time points, as some hospitalized patients were

unable to stand for weighing. At baseline for all 698

patients there was a total of 57 missing values for the

11 parameters. We estimated the bias of assuming

zero for missing by assuming missing�1 and found

a rise in mean TBscore of 0.1�0.2 at all time points

(3�5% of the score), evenly distributed regardless of

HIV and smear status.

Severity classes

By dividing the patients in thirds based on initial

score, we distributed the patients into 3 severity

classes (SC): SC-I, a TBscore of 0�5; SC-II, a

TBscore of 6�7; and SC-III when TBscore was ]

8 points. The basis for this was primarily numerical;

to be able to gather the 13 categories into fewer

groups, the patients were divided into 3 groups using

the 33rd and 66th centiles 5 and 7 points at

treatment initiation.

Validation

The TBscore was validated with regard to content

validity and criterion validity as well as ability to

measure change [21]. Content validity signifies that

the items used have been judged relevant by an

expert panel; this holds true for the items of the

TBscore as they have been taken from WHO guide-

lines.

Criterion validity signifies the correlation of the

scale with a gold standard or another measure used,

and although no other directly comparable measure

is available, we have assessed criterion validity

through comparison with WHO outcomes and

assessed if high TBscore predicts mortality and

smear grade.

Table I. Parameters used for the Tbscore.

Parameters Points assigned

Self-reported

Cough 1

Haemoptysis 1

Dyspnoea 1

Chest pain 1

Night sweating 1

Anaemic conjunctivae 1

Tachycardia 1

Positive finding at lung auscultation 1

Axillary temperature �37.08C 1

BMI*B18 1

BMIB16 1

MUAC*B220 1

MUACB200 1

*BMI: Body Mass Index; MUAC: Middle Upper Arm Circum-

ference.

TBscore 113

Dow

nloa

ded

By:

[Uni

vers

itets

park

en] A

t: 18

:46

12 J

anua

ry 2

008 Finally, we have assessed the sensitivity to change

of the scale by looking at change in TBscore over

time as a measure of any degree of change over time

in association with antibiotic treatment. We have

defined the responsiveness of the scale as clinically

important change according to Liang [22]. One-half

a standard deviation (SD) was considered a good

approximation of the Minimally Clinical Important

Difference (MCID).

Ethics

Pre- and post-test counselling regarding HIV infec-

tion and tuberculosis was available. The patients

were informed in writing in Portuguese and verbally

in the common language Creole, and gave consent

by signature or fingerprint. The epidemiological

studies collecting the data were approved by the

Ministry of Public Health in Guinea-Bissau, and by

the Central Ethical Committee of Denmark.

Statistics

Analyses were performed using Stata Statistical

Software v.9 (Stata Corporation, College Station,

TX, USA). Pearson x2 test was used to assess

statistical differences in proportions between groups,

t-test to assess differences in means between 2

groups with normal distribution, and Mann-

Whitney tests for non-normal distributed values. A

2-sided pB0.05 was considered significant. Mortal-

ity rates were assessed by dividing numbers of deaths

with the time at risk. Differences in survival were

examined using Kaplan-Meier survival graphs. Po-

tential confounding of the effects of baseline

TBscore on mortality rates was assessed by Cox

regression analysis.

Cohen’s effect size is the ratio of the mean

difference to the SD of baseline scores [21]; an

effect size of 0.8 is considered a large effect, 0.5 a

medium effect and 0.2 a small effect.

Receiver operating characteristic (ROC) curve

establishes the optimal cut-point and describes the

discriminating ability of a test [21].

Results

We included 816 adult patients (�15 y) diagnosed

with tuberculosis according to WHO criteria for TB

[11]. After excluding 10 extrapulmonary TB pa-

tients, 5 patients not living in study area, 92 patients

who were included �7 d after initiation of anti-

tuberculosis treatment, and 11 patients with no

clinical data record at baseline, 698 patients

(433M, 265F) were available for analysis of clinical

data. Figure 1 describes patient flow.

816 TB patients included Patients excluded:10 with extrapulmonaryTB5 living outside study area103 were scored after 7 daysof treatmentBaseline: Treatment initiation

698 TB patients scored

Clinical evaluation at 2 months651 alive, 516 scored

5 lost to follow-up42 died

2 lost to follow-up37 died

21 died

38 lost to follow-up72 died

18 months follow-up481 alive in total

295 of these were seen at 8 months

Clinical evaluation at 5 months612 alive, 375 scored

Clinical evaluation at 8 months591 alive, 351 scored

Figure 1. Flow chart of patients.

114 C. Wejse et al.

Dow

nloa

ded

By:

[Uni

vers

itets

park

en] A

t: 18

:46

12 J

anua

ry 2

008 At 2 months of treatment 516 patients showed up

for clinical control, at 5 months 375 patients, and at

8 months 351 patients were seen, hence 59% (351/

591) of those alive at 8 months had repeated clinical

follow-up data. We conducted mortality follow-up

and registered 100 patients as dead and 7 were lost

to follow-up; all remaining were registered as alive at

some point after 8 months of treatment. Clinical

examination at 8 months was available on 351

patients because 100 patients had died, 83 defaulted,

7 were lost to follow-up and the remaining 157

patients did not show for clinical evaluation at 8

months but 8-month mortality was assessed. Addi-

tionally 38 patients were lost to follow-up during the

8�18-month follow-up; hence 18-months mortality

follow-up was available for 93% (645/698). Of the

351 patients with clinical data at 8 months, 331 had

complete mortality follow-up. A total of 172

deaths were recorded, 100 between inclusion and

8-months follow-up, the remaining 72 from 8�18-

months follow-up. 36 of these deaths were recorded

among patients seen at 8 months, the remaining 36

among patients with no clinical follow-up at 8

months. 21 of these deaths were among patients

abandoning TB treatment.

Distribution of the TBscore in various groups

Table II provides descriptive statistics detailing the

spread of scores at the various time points. Scores

show normal distribution at treatment start and are

skewed towards a lower score during treatment.

Hospitalized patients. The score detected severe

clinical cases. Among hospitalized patients 49%

(107/215) were in SC-III whereas only 23%

(109/483) of non-hospitalized patients were in

SC-III at treatment initiation (pB0.001).

HIV-infected. The mean score differed by HIV status

(Table II). At treatment initiation, HIV-infected

(240 patients comprising both HIV-1 and HIV-2)

had a mean score of 6.75 (95% CI 6.12�7.04)

whereas HIV-uninfected (436 patients) had a mean

score of 6.33 (95% CI 6.12�6.53) (p�0.03). At 8

months of treatment, 110 HIV-infected patients had

a mean score of 2.39 (95% CI 1.95�2.83) compared

with 239 HIV-uninfected patients having a mean

score of 1.85 (95% CI 1.61�2.10) (p�0.08).

Smear grade. At treatment initiation, smear positive

patients had a mean score of 6.75 (95% CI 6.48�6.98) whereas smear negative patients had a mean

score of 6.31 (95% CI 6.1�6.54) (p�0.02). At

treatment initiation we assessed TBscore and pro-

portion in high severity class among smear positives

according to smear grade and we found an increasing

TBscore and proportion in SC-III with smear

positivity grade as shown in Table III. There was

only a significant difference between TBscore among

smear negatives and smear positives grade 3, but a

clear trend was observed towards higher score and

higher proportion in SC-III with higher smear grade.

Sensitivity to change

The score declined for 96% (340/351) of the

patients from initiation to end of treatment. The

score declined with a similar pattern in HIV-infected

Table II. Descriptive statistics detailing the spread of scores and background factors on the TBscore at the measured time points.

Inclusion 2 months 5 months 8 months

Range of scores 0�12 0�12 0�8 0�11

Number of patients 698 516 375 351

Mean (SD*) (%) 6.50 (2.2) 3.21 (2.3) 2.28 (1.9) 2.03 (2.0)

HIV-1 infected§ (%) 131/676 (19) 98/508 (19) 60/369 (16) 53/349 (16)

HIV-2 infected (%) 109/676 (16) 80/508 (16) 61/369 (17) 57/349 (16)

HIV uninfected 436/676 (65) 330/508 (65) 248/369 (67) 239/349 (68)

Mean score HIV-infected/HIV-uninfected (SD*) (%) 6.8 (2.3)/6.3 (2.2) 3.6 (2.5)/3.0 (2.3) 2.5 (2.0)/2.2 (1.9) 2.4 (2.3)/1.9 (1.9)

Smear positive 297/698 (43) 226/516 (44) 167/375 (45) 163/351 (46)

Mean score smear positive/smear negative (SD*) 6.7 (2.2)/6.3 (2.2) 3.3 (2.3)/3.2 (2.3) 2.1 (1.7)/2.4 (2.1) 1.6 (1.3)/2.3 (2.2)

Gender �/�(% males) 433/265 (62) 309/207 (60) 217/158 (58) 204/147 (58)

Mean score males/females (SD*) 6.3 (2.2)/6.7 (2.3) 3.0 (2.3)/3.5 (2.4) 2.0 (1.7)/2.7 (2.1) 1.7 (1.9)/2.5 (2.3)

Age (SD*) 39.42 (5.4) 39.0 (5.2) 38.73 (4.8) 38.83 (5.0)

Distribution of patients according to SC* (%)

I 249/698 (36) 434/516 (84) 347/375 (93) 321/351 (91)

II 231/698 (32) 53/516 (10) 24/375 (6) 23/351 (7)

III 218/698 (31) 29/516 (6) 4/375 (1) 7/351 (2)

*: SD: standard deviation; SC: severity class; §: including HIV-1�2 double infected.

TBscore 115

Dow

nloa

ded

By:

[Uni

vers

itets

park

en] A

t: 18

:46

12 J

anua

ry 2

008

and HIV-uninfected patients, as well as in smear

negative and smear positive patients, as shown in

Table II. The table displays mean scores for all

patients seen at the various time points. Minimally

Clinical Important Difference of ½SD was 1.1

point.

Cohen’s effect size was 0.73 comparing baseline

scores with 8-month scores. The chosen MCID of 1

point, i.e. a change from 6.5 to 5.4 in baseline scores,

gave a Cohen’s effect size of 0.25. A 25% change of

baselines scores gave a Cohen’s effect size of 0.37.

To estimate survival bias we analysed TBscore

separately for the 351 patients with complete

8-months clinical data. Figure 2 shows mean scores

for patients with complete clinical follow-up. We also

assessed TBscore at treatment initiation for this

group compared with the 157 alive patients com-

pleting treatment but not seen at 8 months. Mean

initial TBscore among patients compliant for clinical

follow-up was 6.4 (95% CI 6.2�6.6) and 6.2 (95%

CI 5.8�6.5) for those not seen.

Prediction of mortality

We assessed number of deaths at 8 and 18 months

after inclusion (Table IV).

Eight-months follow-up. Patients with an initial score

�8 (SC-III) had a higher mortality, 21% (45/216)

died whereas among patients in SC I-II only 13%

(55/475) died (pB0.001). Figure 3 shows Kaplan-

Meier estimates for patients in the 3 severity classes.

SC-III at initiation of treatment predicted mortality

at 8-months follow-up with sensitivity of 0.45,

specificity of 0.75 and predictive value of 0.22;

area under curve (AUC) was 0.65 (95% CI

0.6�0.7). The curve showed 8 points to be the best

cut-off for 8-months mortality.

Stratified for HIV status, increased mortality in

higher severity class persisted. At inclusion, risk of

mortality for SC-III HIV-infected patients was 31%

(26/83); in SC-I-II the risk was 20% (31/157) (p�0.04). HIV-uninfected patients in SC-III had a 12%

(15/126) risk of mortality, patients in SC-I-II a 6%

(19/304) risk at 8-months follow-up (p�0.05). A

Cox regression analysis of 8-months mortality risk

for 670 patients adjusted for gender, smear status,

HIV infection (all categorical variables) and age

(continuous), showed that the Hazard Ratio (HR)

for individuals in SC-III was 2.0 (95% CI 1.3�3.0),

for HIV-infected individuals 3.1 (95% CI 2.0�4.8),

for smear positive patients 0.84 (95% CI 0.54�1.3),

and for female patients 0.65 (95% CI 0.4�1.0).

Figure 3 provides Kaplan-Meier estimates of mor-

tality in different severity classes, log rank test

pB0.001.

18-months follow-up. The TBscore at end of treatment

after 8 months of treatment predicted mortality

strongly. A SC-III patient at 8 months had an all-

cause mortality of 71% (5/7) versus 9% (31/346) if in

SC I-II (pB0.001). At 8 months, SC-III had a

sensitivity of 0.14, a specificity of 0.99 and a

predictive value of 0.71 for later mortality, the AUC

being 0.75 (95% CI 0.66�0.84). The ROC curve

showed 3 points at 8 months to be the optimal cut-off

for estimating 18-month mortality with a sensitivity of

0.56 and specificity of 0.94 for predicting mortality if

TBscore was ]3 points, the predictive value being

0.27. Cox regression analysis of 18-months mortality

risk for 325 patients seen at 8 months with complete

follow-up, adjusted for age, gender, and HIV infec-

tion, showed that the HR for patients in SC-III

was 7.9 (95% CI 2.7�23.0), and for HIV-infected

patients 3.5 (95% CI 1.7�7.2). The other variables

in the models showed no significance.

Estimation of improvement at end of treatment

At 8 months of treatment the outcomes of the

patients according to WHO definitions were as

follows: cured 169 (24%), completed treatment

Table III. Mean TBscore and severity class according to the grade

of smear positivity for 297 smear positive patients at inclusion.

Smear

positivity

Mean TBscore

(95% CI)

Proportion in

SC-III (n/N)

Smear negative 6.3 (6.1�6.5) 28% (113/401)

Grade 1� 6.5 (6.1�6.8) 32% (42/131)

Grade 2� 6.8 (6.3�7.3) 35% (28/88)

Grade 3� 7.1 (6.6�7.5) 41% (35/86)

Time in months

86420

Mea

n T

Bsc

ore

with

SE

0

1

2

3

4

5

6

7

HIV infectedHIV uninfected

Figure 2. Mean TBscore by HIV status over time in 351 patients

seen at all clinical controls.

116 C. Wejse et al.

Dow

nloa

ded

By:

[Uni

vers

itets

park

en] A

t: 18

:46

12 J

anua

ry 2

008 310 (44%), died 100 (14%), defaulted 83 (12%) and

transferred out 36 (5%) (on some of which we

obtained mortality follow-up, but not treatment

outcome).

At treatment initiation, 36% (249/698) were in

SC-I and after completed treatment, 91% of those

seen (321/351) were in SC-I, the remaining 30 in SC

II-III (Table II). These patients all completed treat-

ment; not all could produce sputum but 283

provided sputum samples at 8 months, which were

all negative. Given that only 169 of these provided

a positive smear before treatment initiation, only

169/698 (24%) were cured by standard definitions.

Of those seen at the health centre at the end of

treatment, 35% (169/479) were cured.

Compared to this criterion, we defined improve-

ment as being in SC-I or decreasing TBscore �25%

at end of treatment and this provided a different

estimate of effect of treatment. The patients seen at 8

months had changes in score from �2 to 11 points,

but among the 30 patients still in SC II-III at end of

treatment, most of the 18-month mortality was seen

among those with an increase in score or a decrease

525%. Six of these patients had an increase in

score; they were all HIV-infected. Among the 30

patients in SC II-III at the end of treatment, 55%

(11/20) of patients with an increase or a decrease in

score 525% died within 18-months follow-up, and

20% (2/10) of those with a change in score �25%

points died.

Hence, in this population 94% (331/351) of all

completing anti-tuberculous treatment improved

according to our definition and we identified a

remnant with a very high mortality of 55% in the

following 10 months.

A cut-off of 3 points must also be considered as

indicator of improvement as only 7% (49/698) of

patients had more than 3 points at treatment

initiation and 3 point-giving signs and symptoms

on the scale may be seen as a better definition of

good health. If using 3 points as cut-off, only 79%

(278/351) were in the category indicating good

health at treatment completion, but among the 73

patients with TBscore �3 points, 41 had a decrease

in TBscore �25%, leaving a total of 32/351 (9%) as

not improved by this definition.

Discussion

We developed a simple score to measure TB

patients’ clinical status at repeated visits. The score

has advantages in measuring outcome of TB com-

pared to WHO standard outcomes as well as the

various measures others have used [2�4,20]. In

particular, the score makes it possible to assess the

outcome of treatment for smear negative patients. InTable

IV.

Mort

ality

by

seve

rity

class

at

dif

fere

nt

tim

epoin

ts.

Incl

usi

on

score

an

d8-m

on

ths

mort

ality

Sco

reat

8m

on

ths

an

d8-1

8m

on

ths

mort

ality

(su

stain

ab

ilit

y)

Sev

erit

ycl

ass

Alive

at

8m

on

ths

(%)

Dea

d(%

)

Follow

-up

tim

ein

per

son

y

MR

*p

er100

per

son

y(9

5%

CI)

Alive

at

18

mon

ths

(%)

Dea

d(%

)

Follow

-up

tim

ein

per

son

y

MR

per

100

per

son

y(9

5%

CI)

I229

(93)

16

(7)

158.2

10

[6�1

6]

279

(92)

23

(8)

222.5

10

[7�1

5]

II191

(83)

39

(17)

138.9

28

[20�3

8]

14

(64)

8(3

6)

13.9

57

[25�1

12]

III

171

(79)

45

(21)

125.6

36

[26�4

8]

2(2

9)

5(7

1)

2.4

250

[81�5

83]

Tota

l591

(86)

100

(14)

422.7

24

[19�2

9]

295

(90)

36

(10)

238.8

15

[11�2

1]

*M

R:

mort

ality

rate

.

TBscore 117

Dow

nloa

ded

By:

[Uni

vers

itets

park

en] A

t: 18

:46

12 J

anua

ry 2

008

this study, however, the smear negative patients were

not confirmed by culture.

The score was sensitive to change during treat-

ment, showed good responsiveness and had predic-

tive capacity. Minimally Important Clinical

Difference was found to be just above 1 point, i.e.

1 sign/symptom or a change in BMI/MUAC. How-

ever, although a change of 1 point is relevant when

comparing overall scores or groups of patients, we

have found that an anchor based approach of 25%

change in baseline scores was more practical in terms

of defining clinical improvement.

The empirical validation showed strong associa-

tion with mortality and grade of smear positivity.

Others have found strong association between BMI/

MUAC and TB mortality [19,20] and these para-

meters have strong weight in the score. An associa-

tion between grade of smear positivity and clinical

feature have previously been demonstrated in Sudan

[23], which supports that the TBscore describes

clinical status.

Survival bias is likely to have affected the fall in

mean TBscore over time, but there was a decline in

score for 96% of the patients on an individual basis,

so the drop in score was also present in the 351

patients with complete 8-month follow-up data

(Figure 2). There was a substantial loss to clinical

follow-up, and we do not know if those patients not

showing for later consultations were different at 8

months from those compliant with all examinations,

but their initial scores were lower, although not

significantly. Mortality at 18-month follow-up was

comparable for the non-defaulted patients not seen

at 8-months control (10% (15/157)) and the patients

seen at 8 months (10% (36/351)). In recent years we

have managed to increase the fraction of patients

seen at 8 months to 77% of those initially seen and

still alive at 8 months (unpublished data). However,

a substantial loss to follow-up will always occur in

highly mobile urban populations. Our mortality

follow-up was excellent due to an efficient demo-

graphic surveillance set-up and to patients being

tracked even outside the study area. Therefore the

data allow for strong conclusions regarding the

empirical validation of the ability of the TBscore at

treatment initiation to predict 8-month mortality.

The AUC of 0.65 for predicting 8-month mortal-

ity and 0.75 for predicting 18-month mortality at 8

months were not high values for a diagnostic test,

but we have used the method to predict mortality as

a way of showing that there is a correlation between

the TBscore and subsequent mortality. The AUC

value is comparable to what others have found when

using clinically based scores to predict mortality

[24,25]. We are not aware of any other measure that

can more accurately predict mortality in TB pa-

tients.

The clinical presentation of HIV and TB can

be difficult to distinguish, but the score declined

during TB treatment irrespective of HIV status (cf.

Figure 2), and the differences found by HIV status

were smaller than 1 point and thus not clinically

important by our definition. Likewise, a comparable

clinical change towards fewer symptoms was found

among both smear positive and smear negative

individuals also with differences smaller than

MCID at all time points. This suggests that our

definition for improvement is valid for both smear

positive and smear negative patients.

The TBscore at the end of specific TB treatment

was only available among patients with a success-

ful treatment outcome, i.e. cured or completed

0.75

0.80

0.85

0.90

0.95

1.00

Sur

viva

l pro

babi

lity

0 2 4 6 8Analysis time (months)

Severity Class ISeverity Class IISeverity Class III

Kaplan Meier 8 month survival estimates, by Severity Class

Figure 3. Kaplan-Meier survival estimates based on mortality status at 8-months follow-up, by severity class at inclusion.

118 C. Wejse et al.

Dow

nloa

ded

By:

[Uni

vers

itets

park

en] A

t: 18

:46

12 J

anua

ry 2

008 treatment. However, it provides important informa-

tion on the subsequent health status. Clinical cure or

complete improvement from TB is ideally to have no

symptoms, yet hardly achievable in a population

with high-untreated HIV prevalence and damaged

lung parenchyma following severe TB. However,

having few symptoms, i.e. being in SC-I, may be a

measure of good health or improvement at the end of

treatment. Clinical improvement defined as the

proportion in SC-I or the fall in score is readily

usable as an endpoint in a clinical trial, whereas

completion of treatment cannot be used to discrimi-

nate between patients, as most will complete treat-

ment in a trial with good follow-up. Hence, the

suggested definition for improvement may be a

useful surrogate marker, especially in smear negative

patients for whom no marker is currently available.

The score gives valuable information on high-risk

patients. A high TBscore provides the clinician with

information that the patient has a much higher risk

of dying. This could guide the clinician in decisions

on ‘special care’, such as hospitalization or more

intensive follow-up. The increased subsequent mor-

tality in high SCs suggests that intervention is

needed when a patient�despite treatment�is still in

a high SC. The clinical assessment is possible at a

health centre by a physician equipped with a stetho-

scope, weighing scale, height- and arm circumfer-

ence measuring tape and thermometer without

expensive diagnostics. Many of the measurements

are subjective, in particular anaemia and lung

auscultation, and therefore subject to inter-personal

variation. When grouped in severity classes these

signs and symptoms construct a robust index usable

in settings in which advanced laboratory measure-

ments are not available. We were unable to assess

inter-observer variability with the current data set,

but such data are currently being collected.

The study has several important limitations, such

as the lack of diagnostic facilities at the time of the

study, including TB culture that was impossible due

to the destruction of the National Health laboratory

during a civil war in 1998�1999. Another limitation

is the substantial loss to clinical follow-up as

previously discussed, and the failure to assess inter-

observer agreement. Furthermore, the score cannot

be generalized to HIV-positive patients treated with

antiretroviral therapy. However, in resource-poor

settings, a large part of the world’s concomitantly

TB- and HIV-infected patients still face conditions

as in the present study. A rapid clinical low-cost

assessment may be of value for such patients.

Further studies are warranted to examine the

score in other TB patient cohorts with assessment

of inter-observer variation and testing of ‘special

care’ interventions for high- risk patients.

In conclusion, the TBscore is a useful clinical

index sensitive to change during treatment and with

ability to predict mortality. We propose the TBscore

as a supplementary index for determining improve-

ment and as an outcome measure for patients in

clinical trials.

Acknowledgements

We thank field workers, nurses and physicians at the

Bandim Health Project, in particular data manage-

ment assistant Adulai Gomes Rodrigues for the

substantial work in data entry and checking these

data.

The first author holds a PhD scholarship from

Aarhus University Hospital. The initial data collec-

tion was funded by the Agence Nationale de

Recherche sur le SIDA/Projet SIDAK in France,

European Union grant ERBIC18CT980375, The

Danish International Development Assistance (DA-

NIDA) and Sida/SAREC (Swedish International

Development Cooperation Agency/Department for

Research Cooperation).

References

[1] WHO. Global tuberculosis control: surveillance, planning,

financing. WHO report 2006. Geneva, World HealthOrga-

nization (WHO/HTM/TB/2006.362).

[2] Schon T, Elias D, Moges F, Melese E, Tessema T, Stendahl

O, et al. Arginine as an adjuvant to chemotherapy improves

clinical outcome in active tuberculosis. Eur Respir J 2003;/21:/

483�8.

[3] Madebo T, Lindtjorn B, Aukrust P, Berge RK. Circulating

antioxidants and lipid peroxidation products in untreated

tuberculosis patients in Ethiopia. Am J Clin Nutr 2003;/78:/

117�22.

[4] Wilson D, Nachega J, Morroni C, Chaisson R, Maartens G.

Diagnosing smear negative tuberculosis using case defini-

tions and treatment response in HIV-infected adults. Int J

Tuberc Lung Dis 2006;/10:/31�8.

[5] Kanaya AM, Glidden DV, Chambers HF. Identifying

pulmonary tuberculosis in patients with negative sputum

smear results. Chest 2001;/120:/349�55.

[6] Bruchfeld J, Aderaye G, Palme IB, Bjorvatn B, Britton S,

Feleke Y, et al. Evaluation of outpatients with suspected

pulmonary tuberculosis in a high HIV prevalence setting in

Ethiopia: clinical, diagnostic and epidemiological character-

istics. Scand J Infect Dis 2002;/34:/331�7.

[7] Gustafson P, Gomes VF, Vieira CS, Rabna P, Seng R,

Johansson P, et al. Tuberculosis in Bissau: incidence and risk

factors in an urban community in sub-Saharan Africa. Int J

Epidemiol 2004;/33:/163�72.

[8] Gustafson P, Gomes VF, Vieira CS, Jensen H, Seng R,

Norberg R, et al. Tuberculosis mortality during a civil war in

Guinea-Bissau. JAMA 2001;/286:/599�603.

[9] Gustafson P, Gomes VF, Vieira CS, Samb B, Naucler A,

Aaby P, Lisse I. Clinical predictors for death in HIV-positive

and HIV-negative tuberculosis patients in Guinea-Bissau.

Infection 2007;/35:/69�80.

[10] Aaby P. Bandim: an unplanned longitudinal study. In: Das

Gupta M, Aaby P, Pison G, Garenne M, editors. Prospective

TBscore 119

Dow

nloa

ded

By:

[Uni

vers

itets

park

en] A

t: 18

:46

12 J

anua

ry 2

008 Community Studies in Developing Countries. Oxford:

Clarendon; 1997. p. 276�96.

[11] Harries AD, Maher D, Graham S. TB/HIV: A clinical

manual, 2nd edn. WHO/HTM/TB 2004;329:50,106.

[12] Siddiqi K, Lambert ML, Walley J. Clinical diagnosis of

smear negative pulmonary tuberculosis in low-income coun-

tries: the current evidence. Lancet Infect Dis 2003;/3:/

288�96.

[13] Bossuyt PM, Reitsma JB, Bruns DE, Gatsonis CA, Glasziou

PP, Irwig LM, et al. The STARD statement for reporting

studies of diagnostic accuracy: explanation and elaboration.

Ann Intern Med 2003;/138:/1�12.

[14] Laupacis A, Sekar N, Stiell IG. Clinical prediction rules. A

review and suggested modifications of methodological stan-

dards. JAMA 1997;/277:/488�94.

[15] Mackowiak PA. Concepts of fever. Arch Intern Med 1998;/

158:/1870�81.

[16] Powell-Tuck J, Hennessy EM. A comparison of mid upper

arm circumference, body mass index and weight loss as

indices of undernutrition in acutely hospitalized patients.

Clin Nutr 2003;/22:/307�12.

[17] van Lettow M, Harries AD, Kumwenda JJ, Zijlstra EE, Clark

TD, Taha TE, et al. Micronutrient malnutrition and wasting

in adults with pulmonary tuberculosis with and without HIV

coinfection in Malawi. BMC Infect Dis 2004;/4:/61.

[18] van der Sande MA, Schim van der Loeff MF, Aveika AA,

Sabally S, Togun T, Sarge-Njie R, et al. Body mass index at

time of HIV diagnosis: a strong and independent predictor of

survival. J Acquir Immune Defic Syndr 2004;/37:/1288�94.

[19] Zachariah R, Spielmann MP, Harries AD, Salaniponi FM.

Moderate to severe malnutrition in patients with tuberculosis

is a risk factor associated with early death. Trans R Soc Trop

Med Hyg 2002;/96:/291�4.

[20] Villamor E, Saathoff E, Mugusi F, Bosch RJ, Urassa W,

Fawzi WW. Wasting and body composition of adults with

pulmonary tuberculosis in relation to HIV-1 coinfection,

socioeconomic status, and severity of tuberculosis. Eur J Clin

Nutr 2006;/60:/163�71.

[21] Streiner DL, Norman GR. Health measurement scales: a

guide to their development and use. New York: Oxford

University Press; 2003.

[22] Liang MH. Longitudinal construct validity: establishment of

clinical meaning in patient evaluative instruments. Med Care

2000;/38:/84�90.

[23] El-Sony A, Enarson D, Khamis A, Baraka O, Bjune G.

Relation of grading of sputum smears with clinical features of

tuberculosis patients in routine practice in Sudan. Int J

Tuberc Lung Dis 2002;/6:/91�7.

[24] Larsen SH, Pedersen J, Jacobsen J, Johnsen SP, Hansen OK,

Hjortdal V. The RACHS-1 risk categories reflect mortality

and length of stay in a Danish population of children

operated for congenital heart disease. Eur J Cardiothorac

Surg 2005;/28:/877�81.

[25] Ferjani AM, Griffin D, Stallard N, Wong LS. A newly

devised scoring system for prediction of mortality in patients

with colorectal cancer: a prospective study. Lancet Oncol

2007;/8:/317�22.

120 C. Wejse et al.