Journal of Infection 77 (2018) 283–290

Contents lists available at ScienceDirect

Journal of Infection
journal homepage: www.elsevier.com/locate/jinf

Development and validation of a prognostic score to predict

tuberculosis mortality ✩
Duc T. Nguyen, Edward A. Graviss∗
Houston Methodist Research Institute, 6670 Bertner Ave, Houston, TX 77030, USA

a r t i c l e i n f o s u m m a r y

Article history: Objective: To develop and validate a simple prognostic scoring system to predict the mortality risk during
Accepted 19 February 2018 treatment in tuberculosis patients.
Available online 9 April 2018
Methods: Using data from the CDC’s Tuberculosis Genotyping Information Management System of TB pa-
Keywords: tients in Texas reported from 01/2010 to 12/2016, age ≥ 15 years and having an outcome as “completed”
TB risk score or “died”, we developed and validated a prognostic mortality scoring system-based logistic regression
Prognosis beta-coefficients.
Elderly
Meningeal TB Results: The developmental and validation cohorts consisted of 3378 and 3377 patients, respectively. The
TB-CXR score used 9 demographic and clinical characteristics, which are usually available at the patient’s initial
HIV infection visits to a healthcare facility. Prognostic scores were categorized into three groups that predicted mortal-
ity: low-risk (<15 points), medium-risk (15–18 points), and high-risk (>18 points). The model had excel-
lent discrimination and calibration with an area under the receiver operating characteristic curve of 0.82
and 0.80, and a non-significant Hosmer–Lemeshow test P = 0.514 and P = 0.613 in the developmental
and validation cohorts, respectively.
Conclusion: Our validated TB prognostic scoring system, which used demographic and clinical character-
istics available at the patient’s initial visits, can be a practical tool for health care providers to identify TB
patients with high mortality risk so that appropriate treatment, medical supports and follow-up resources
could be appropriately allocated.
© 2018 The Author(s). Published by Elsevier Ltd on behalf of The British Infection Association.
This is an open access article under the CC BY-NC-ND license.
(http://creativecommons.org/licenses/by-nc-nd/4.0/)

Introduction TB rate of 4.9 per 10 0,0 0 0 population in 2015, an increase of 4.3%

Tuberculosis (TB) is one of the diseases with the highest mor-
bidity and mortality. In 2015, approximately 10.4 million new TB
cases with 1.4 million HIV-negative and 0.4 million HIV-positive in-
dividuals died of the disease worldwide.1 Although TB mortality in
the United States (U.S.) has decreased in recent years, the disease
claimed 493 deaths in 2014.2 One of the states having the highest
TB burden in the U.S., Texas, also had TB as the disease having the
highest standardized mortality ratio (SMR) relative to the national
reference between 2001 and 2010 with 679 deaths.3 Texas had a
✩
Funding: None.
∗
Corresponding author. Department of Pathology and Genomic Medicine, Hous-
ton Methodist Research Institute, Mail Station: R6-414, 6670 Bertner, Houston, TX
77030.
E-mail address: eagraviss@houstonmethodist.org (E.A. Graviss).
compared to that in 2014.4
Rapid diagnosis and early treatment, as well as appropriate
prognosis, especially in the patient’s initial visit to a health care
facility, are crucial tools for clinicians and health care workers to
successfully manage individuals who have contracted TB. Many
guidelines have been developed to improve TB diagnosis, prog-
nosis, and treatment outcome.5–8 Multiple risk factors have also
been identified that are associated with TB morbidity and mortal-
ity.1,4,9,10 Even though at least two TB risk scoring systems have
been developed,11,12 there is no currently standardized scoring sys-
tem, validated with large population-based data, to promptly pro-
vide the prognosis for TB patients’ outcome from their initial visits.
The current study’s main objective was to develop and validate a
prognostic scoring system using a large population-based surveil-
lance dataset to predict TB mortality during treatment, which is
applicable prior to microbiologic and or laboratory confirmation.

https://doi.org/10.1016/j.jinf.2018.02.009
0163-4453/© 2018 The Author(s). Published by Elsevier Ltd on behalf of The British Infection Association. This is an open access article under the CC BY-NC-ND license.
(http://creativecommons.org/licenses/by-nc-nd/4.0/)
284 D.T. Nguyen, E.A. Graviss / Journal of Infection 77 (2018) 283–290
Methods
Study population
In this study, we used the de-identified surveillance data on
demographic and clinical characteristics of all confirmed TB pa-
tients from the state of Texas (U.S.) reported to the National TB
Surveillance System. The dataset was downloaded from the Cen-
ters for Disease Control and Prevention (CDC) supported TB Geno-
typing Information Management System (TBGIMS) website. The in-
clusion criteria included: (1) confirmed TB cases in the state of
Texas, U.S. in the National TB Surveillance System (NTSS) from
01/2010 through 12/2016 (based on official case count date); (2)
age ≥15 years old; and (3) had documented treatment outcome
in the dataset as either “completed” or “died”. Patients age un-
der 15 years old or having outcome coding other than as “com-
pleted” or “died” (such as “adverse”, “lost”, “moved”, “other”, “re-
fused”, or “unknown”) were excluded from the analyses. A sensi-
tivity analysis was done to compare the difference in demographic
and clinical characteristics between the excluded and included pa-
tients. TB cases in the dataset were identified and verified by the
local and state TB program staff using the CDC’s TB case defini-
tion,13 which included either positive Mycobacterium tuberculosis
culture (n = 5244), positive nucleic acid amplification (NAA) test
(n = 107), positive acid-fast bacilli (AFB) smear (n = 31), and veri-
fied clinical cases (n = 1373).
Statistical analysis
The dataset was randomly divided into two cohorts with a ratio
of 50/50: developmental and validation cohorts. The developmen-
tal cohort was used to create the prognostic scoring system. The
performance of the scoring system was then assessed in the vali-
dation cohort.
Demographic and clinical data were reported as median and in-
terquartile range (IQR) for continuous variables, and as frequencies
and proportions for categorical variables. Differences across groups
were compared using the Chi-square test. Univariate and multiple
logistic regression models were used to determine the contribu-
tion of potential prognostic variables to the patient outcome. Vari-
ables having a P value <0.2 in univariate logistic regression were
investigated further in multiple logistic regression models using
the Bayesian model averaging (BMA) method to identify significant
Fig. 1. Flowchart of the study population. NTSS, National Tuberculosis Surveillance System.
risk factors for death.14,15 Significant risk factors were assigned
weighted-points that were proportional to their β regression co-
efficient values. The risk scores were calculated for each individ-
ual patient in the cohort. Patients were categorized in deciles of
risk score and then collapsed into three groups which were signif-
icantly distinct in predictive risk for mortality: low (<10% mortal-
ity), medium (10%–20% mortality), and high risk (>20% mortality).
Mortality was then calculated for each risk group. The discrimi-
nation of the predictive model was evaluated by the area under
the receiver operating characteristic (ROC) curve (AUC). Differences
in the AUC between the developmental and validation cohorts
were compared using the Chi-square test. The model’s good cali-
bration (predictive accuracy) was determined by a non-significant
Hosmer–Lemeshow goodness of fit test as well as the similarity
in the mortality differences (delta) between high-risk and low-
risk groups across the cohorts. Multiple logistic regression analyses
with bootstrapped standard errors were also used to compare the
odds of death across different risk groups. All the analyses were
performed on Stata version 14.2 (StataCorp LP, College Station, TX).
A P value of <0.05 was considered statistically significant.
Results
Characteristics of development and validation cohorts
A total of 8421 confirmed TB adult cases from Texas were
reported in the National TB Surveillance System database from
01/2010 through 12/2016. After excluding 1666 patients (1014 with
missing outcome status and 652 with an outcome of other than
“completed” or “died”), 6755 patients were used for the analysis
and randomly divided by Stata’s randomization program into two
cohorts at a ratio of 50/50: the developmental (3378 patients) and
validation (3377 patients) cohorts (Fig. 1). Sensitivity analysis indi-
cated that with the exception of male gender and meningeal TB,
which are in a higher proportion in the 1666 excluded patients,
there was no other significant difference in demographic and clini-
cal characteristics between the excluded and included groups (data
not shown).
The demographic and clinical characteristics of the develop-
mental and validation cohorts are reported in Table 1. There was
no significant difference between the two cohorts in all the char-
acteristics under investigation, which included demographics, med-
ical history, tuberculosis disease site, chest radiograph, specimen
 D.T. Nguyen, E.A. Graviss / Journal of Infection 77 (2018) 283–290 285

Table 1
Demographic and clinical characteristics of the development and validation cohorts.

Development cohort (N = 3378) Validation cohort (N = 3377) P valuea

n (%) n (%)

Age (years) 0.493

15–44 1575 (46.6) 1605 (47.5)
45–64 1237 (36.6) 1190 (35.2)
≥65 566 (16.8) 582 (17.2)
Gender 0.740
Female 1186 (35.1) 1173 (34.7)
Male 2191 (64.9) 2204 (65.3)
Race/Ethnicity 0.291
White 414 (12.3) 420 (12.4)
Black 634 (18.8) 615 (18.2)
Hispanic 1734 (51.3) 1711 (50.7)
Asian 568 (16.8) 614 (18.2)
Other 28 (0.8) 17 (0.5)
U.S.-born 0.778
No 1989 (58.9) 1977 (58.5)
Yes 1389 (41.1) 1400 (41.5)
Resident of long-term care facility 0.097
No 3323 (98.4) 3338 (98.9)
Yes 55 (1.6) 39 (1.2)
Chronic kidney failure 0.380
No 3329 (98.6) 3319 (98.3)
Yes 49 (1.5) 58 (1.7)
Meningeal TB 0.612
No 3331 (98.6) 3325 (98.5)
Yes 47 (1.4) 52 (1.5)
Miliary TB 0.330
No 3228 (97.0) 3243 (97.4)
Yes 99 (3.0) 86 (2.6)
TB-specific abnormality on CXR 0.623
No 378 (11.7) 392 (12.1)
Yes 2843 (88.3) 1839 (87.9)
TB case verified by 0.981
Clinical case definition/provider diagnosis 687 (20.3) 686 (20.3)
Positive culture, NAA or smear 2691 (79.7) 2691 (79.7)
HIV status 0.648
Negative 2793 (82.7) 2800 (82.9)
Positive 234 (6.9) 216 (6.4)
Unknown 351 (10.4) 361 (10.7)
MDR-TB 0.762
No 2539 (75.2) 2560 (75.8)
Yes 19 (0.6) 21 (0.6)
Unknown 820 (24.3) 796 (23.6)
Outcome 0.437
Alive (and completed treatment) 3125 (92.5) 3107 (92.0)
Dead 253 (7.5) 270 (8.0)

Note: CXR: chest radiograph; NAA: Nucleic Acid Amplification; MDR-TB: Multi-drug resistant tuberculosis.
a
Differences across groups were compared using the Chi-square test.

smear and culture, HIV status, multi-drug resistance, and clinical
outcome (Table 1). The data of the 3378 patients in the devel-
opmental cohort was used in developing the prognostic mortality
prognostic scoring system.
Development of the prognostic mortality prognostic score system
chest radiograph, and HIV status) and were used in the develop-
ment of the TB mortality risk score (Table 3).
Weighted points were assigned to each of the final nine risk
factors using the linear transformation of the corresponding re-
gression coefficient (Table 3). A prognostic score was calculated for
individual patients based on the following formula:

Prognostic score = [Age : 15 − 44 years, 0 point; 45 − 64 years,

Univariate logistic regression analyses were used to identify po-
tential risk factors associated with mortality (Table 2). Multiple 6 points; ≥ 65 years, 12 points]
logistic regression models were fitted to determine the indepen- +2 ∗ [US born] + 2 ∗ [Homeless]
dent risk factors. The variable selection process using the Bayesian +4 ∗ [Resident of long-term care facility]
model averaging method suggested ten variables to be of prognos-
tic significance, which were included in the final multiple logis-
+8 ∗ [Chronic kidney failure]
tic regression model: age groups, gender, being U.S.-born, home- +10 ∗ [Meningeal TB] + 4 ∗ [Miliary TB]
lessness, resident of long-term care facility, chronic kidney fail- +6 ∗ [TB-CXR] + [HIV : negative, 0 point;
ure, meningeal TB, miliary TB, TB-specific abnormalities on chest
positive or unknown status, 6 points].
radiograph, and HIV status. Among these variables, only gender
was not significant in the final model. The other nine variables re- Patients in the developmental cohort were divided into ten
mained statistically significant in the final model (age groups, U.S.- groups corresponding to the deciles of their scores. The mortality
born, homelessness, resident of long-term care facility, chronic kid- in each decile was examined to find the best cut-off points for con-
ney failure, meningeal TB, miliary TB, TB-specific abnormalities on solidating patients into three risk groups, which were significantly
286 D.T. Nguyen, E.A. Graviss / Journal of Infection 77 (2018) 283–290

Table 2
Crude association between different potential risk factors and mortality in the development cohort.

N Completed n (%) Deceased n (%) OR (95% CI) P value

3125 (92.5) 253 (7.49)

Age (years)
15–44 1544 (49.4) 31 (12.25) (reference)
45–64 1127 (36) 110 (43.48) 4.86 (3.24, 7.29) <0.001
≥65 454 (14.5) 112 (44.27) 12.29 (8.14, 18.54) <0.001
Male 2014 (64.5) 177 (69.96) 1.28 (0.97, 1.7) 0.079
Race/Ethnicity
White 363 (11.6) 51 (20.16) 2.35 (1.48, 3.73) <0.001
Black 583 (18.7) 51 (20.16) 1.47 (0.93, 2.31) 0.102
Hispanic 1616 (51.7) 118 (46.64) 1.22 (0.82, 1.83) 0.327
Asian 536 (17.2) 32 (12.65) (reference)
Other 27 (0.9) 1 (0.4) 0.62 (0.08, 4.71) 0.644
U.S.-born 1245 (39.8) 144 (56.92) 1.99 (1.54, 2.58) <0.001
Homeless 170 (5.4) 28 (11.07) 2.16 (1.42, 3.3) <0.001
Resident of correction institution 314 (10.9) 4 (1.71) 0.14 (0.05, 0.39) <0.001
Resident of long-term care facility 36 (1.2) 19 (7.51) 6.97 (3.93, 12.34) <0.001
IDU 67 (2.1) 11 (4.35) 2.07 (1.08, 3.98) 0.028
Alcoholism 579 (18.5) 65 (25.69) 1.52 (1.13, 2.04) 0.006
Solid organ transplant 3 (0.1) 6 (2.37) 25.28 (6.28, 101.69) <0.001
History of diabetes 502 (16.1) 42 (16.6) 1.04 (0.74, 1.47) 0.823
Chronic kidney failure 33 (1.1) 16 (6.32) 6.33 (3.43, 11.66) <0.001
Immunosuppression (medical condition or medication) 62 (2) 17 (6.72) 3.56 (2.05, 6.18) <0.001
TB site, pulmonary 2681 (85.8) 227 (89.72) 1.45 (0.95, 2.2) 0.084
TB site, not at cervical lymph 2976 (95.2) 252 (99.6) 12.62 (1.76, 90.54) 0.012
Meningeal TB 32 (1) 15 (5.93) 6.09 (3.25, 11.41) <0.001
Miliary TB 81 (2.6) 18 (7.29) 2.91 (1.72, 4.93) <0.001
TB-specific abnormality on CXR 2615 (87.7) 228 (95) 2.66 (1.47, 4.8) 0.001
Miliary abnormality on CXR 59 (2.3) 12 (5.29) 2.42 (1.28, 4.56) 0.007
Cavitation on CXR 944 (36.1) 68 (29.82) 0.75 (0.56, 1.01) 0.057
Positive specimen AFB smear 1312 (46.3) 116 (62.03) 1.9 (1.4, 2.57) <0.001
Positive specimen Mycobacterium tuberculosis culture 1868 (66.3) 149 (80.11) 2.05 (1.42, 2.96) <0.001
Sputum culture conversion
Converted 1747 (55.9) 41 (16.21) (reference)
Not converted 120 (3.8) 105 (41.5) 37.28 (24.85, 55.93) <0.001
Unknown 1258 (40.3) 107 (42.29) 3.62 (2.51, 5.23) <0.001
TB case verified by:
Positive culture or NAA 2444 (78.2) 233 (92.09) (reference)
Positive smear 13 (0.4) 1 (0.4) 0.81 (0.11, 6.2) 0.837
Clinical case definition 488 (15.6) 11 (4.35) 0.24 (0.13, 0.44) <0.001
Provider diagnosis 180 (576) 8 (316) 0.47 (0.23, 0.96) 0.038
HIV status
Negative 2650 (84.8) 143 (56.52) (reference)
Positive 201 (6.4) 33 (13.04) 3.04 (2.03, 4.56) <0.001
Unknown 274 (8.8) 77 (30.43) 5.21 (3.84, 7.06) <0.001
MDR-TB
No 2327 (74.5) 212 (83.79) (reference)
Yes 18 (0.6) 1 (0.4) 0.61 (0.08, 4.59) 0.631
Unknown 780 (2496) 40 (1581) 0.56 (0.4, 0.8) 0.001
East Asian lineage (L2)
No 1842 (58.9) 162 (64.03) (reference)
Yes 388 (12.4) 35 (13.83) 1.03 (0.7, 1.5) 0.896
Genotype unavailable 895 (2864) 56 (2213) 0.71 (0.52, 0.97) 0.034

Note: IDU: injecting drug user; CXR: chest radiograph; NAA: Nucleic Acid Amplification; MDR-TB: Multi-drug resistant tuberculosis.

distinctive: low-risk group (<10% mortality, 0–14 points), medium-
risk group (10%–20% mortality, 15–18 points), and high-risk group
(>20% mortality, 19–34 points) (Table 4).
Validation of the prognostic score
A similar mean score with a 95% confidence interval (CI) was
found in both the developmental and validation cohorts (11.83
[11.62, 12.04] and 11.78 [11.56, 11.99], P = 0.723) (data not shown).
The classification of patients by risk group was also similar in both
cohorts with a proportion of patients in low-risk, medium-risk, and
high-risk groups of 73.3%, 13.2%, and 13.5% (for the developmen-
tal cohort), and 72.8%, 13.9%, and 13.3% (for the validation cohort),
respectively (Table 4, Fig. 2). The model had excellent discrimina-
tion in both the developmental and validation cohorts with the
area under the ROC curves being (0.82 [95% CI 0.80, 0.85] versus
0.80 [95% CI 0.77, 0.82], P = 0.151) (Table 4, Fig. 3). The model also
had good calibration in both the developmental and validation co-
horts with a non-significant Hosmer–Lemeshow Chi-square of 7.21
(P = 0.514) and 6.30 (P = 0.613), respectively. For both cohorts, the
mortality was significantly different among risk groups (P < 0.001)
with a similar difference in mortality between high-risk and low-
risk groups (23.5% versus 22.1%) (Table 4). In comparison with pa-
tients in the low-risk group, patients in the medium-risk and high-
risk groups were at five and twelve times greater (in the develop-
mental cohort) and four and nine times (in the validation cohort)
greater odds of mortality (Table 5).
Discussion
In this study, we have developed and validated a prognostic
scoring system for TB mortality using 9 demographic and clini-
 D.T. Nguyen, E.A. Graviss / Journal of Infection 77 (2018) 283–290 287

Table 3
Multiple logistic regression model and weighted point assignment in the development cohort.

Variable β coefficient OR (95% CI) P value Weighted Points

Age (years)
15–44 (reference) 0
45–64 1.33 3.80 (2.46, 5.87) <0.001 6
≥65 2.48 11.93 (7.59, 18.76) <0.001 12
U.S.-born 0.40 1.49 (1.10, 2.02) 0.010 2
Homeless 0.49 1.64 (1.01, 2.67) 0.047 2
Resident of long-term care facility 0.86 2.36 (1.22, 4.59) 0.011 4
Chronic kidney failure 1.73 5.66 (2.78, 11.54) <0.001 8
Meningeal TB 2.11 8.22 (3.69, 18.32) <0.001 10
Miliary TB 0.87 2.38 (1.27, 4.48) 0.007 4
TB-CXR 1.04 2.84 (1.49, 5.40) 0.002 6
HIV status
Negative (reference) 0
Positive 1.21 3.36 (2.08, 5.41) <0.001 6
Unknown 1.26 3.51 (2.48, 4.98) <0.001 6

Note: Weighted points of a risk factor were calculated using a linear transformation of the corresponding
β coefficient (was divided by 0.40 [the lowest β value of individual variable, the US born], multiplied by a
constant (2), and rounded to the nearest integer).16 The reference group of categorical variables (age 15–44
years and HIV [–] status) was assigned 0 point, corresponding to a beta coefficient of zero.

Table 4
Mortality by risk group in patients having complete data for all variables of the multiple logistic regression modela .

Development cohort (N = 3219) Validation cohort (N = 3221)

b b
n (%) Mortality % (95% CI) P value n (%) Mortality % (95% CI) P value

Risk group
Low-risk group (<15 points) 2358 (73.3) 2.9 (2.2, 3.6) <0.001 2346 (72.8) 3.6 (2.8, 4.3) <0.001
Medium-risk group (15–18 points) 425 (13.2) 13.4 (10.2, 16.7) 446 (13.9) 13.2 (10.1, 16.4)
High-risk group (>18 points) 436 (13.5) 26.4 (22.2, 30.5) 429 (13.3) 25.6 (21.5, 29.8)
Risk score performance
AUC (95% CI) c 0.82 (0.80, 0.85) 0.80 (0.77, 0.82)
Hosmer–Lemeshow goodness of fit testd Chi-square = 7.21 (P = 0.514) Chi-square = 6.30 (P = 0.613)
Difference in mortality (%)e 23.5 22.1

Note: Comparisons of mortality between risk groups were conducted using Chi-square test.
a
Number (%) of patients did not have complete data for all variables in the development and validation cohorts is 159 (4.7%) and 156 (4.6%), respectively.
b
Overall P value. A P < 0.001 was also found for all pairwise comparisons among groups (i.e. low-risk vs. medium-risk, low-risk vs. high-risk and medium-risk
vs. high-risk groups) in development and validation cohorts.
c
Chi-square P = 0.151 for the comparison of AUC between development and validation cohorts.
d
A non-significant Hosmer–Lemeshow goodness of fit test indicates good calibration.
e
Difference in mortality between high-risk and low-risk groups (%), calculated by [mortality in high-risk group] - [mortality in low-risk group].

Fig. 2. Risk score performance and mortality (95% CI) by risk group in the development cohort versus validation cohort.
288 D.T. Nguyen, E.A. Graviss / Journal of Infection 77 (2018) 283–290

Fig. 3. Comparison of area under the receiver operating characteristic (ROC) curve between development and validation cohorts.

Table 5
Odds for death, by risk group (with bootstrap in estimating coefficient standard errors)

Risk Development cohort (N = 3219) Validation cohort (N = 3211)

group
OR (95% CI) P value OR (95% CI) P value

Low-risk group (<15 points) Reference Reference

Medium-risk group (15–18 points) 5.22 (3.63, 7.51) <0.001 4.11 (2.86, 5.89) <0.001
High-risk group (>18 points) 12.06 (8.65, 16.83) <0.001 9.29 (6.88, 12.53) <0.001

cal characteristics, which are usually available at the patient’s ini-
tial visit to a healthcare facility. Hence, our scoring model, which
had good discrimination and calibration, should promptly provide
the clinicians and health care workers with the patient’s progno-
sis for death without waiting for the biological confirmation. Us-
ing the model with three distinctive risk groups, the clinicians and
health care workers will have a validated, practical tool to allo-
cate the appropriate treatment, medical supports, and follow-up
resources. High-risk patients would need the most attention with
urgent treatment and more aggressive medical supports. Medium-
risk patients could benefit from closer follow-up and prompt in-
tervention if needed to prevent them from falling into aggravated
health conditions. In the meantime, suspected TB patients in the
low-risk group should be treated and managed as per routine
protocol. For example, although HIV antiretroviral therapy is rec-
ommended to be initiated for TB/HIV co-infected patients within
2 weeks of starting TB treatment if CD4 + level <50 cells/mm3
and within 8 weeks of starting TB treatment if CD4 + level ≥50
cells/mm3,17 early initiation of combination antiretroviral therapy
(cART) could be considered in high-risk patients even though the
CD4 + level ≥50 cells/mm3 as the cART has been shown to reduce
up to 68% TB-related deaths in TB/HIV co-infected patients.18 In
addition to having old age, poor living conditions and the presence
of other comorbidities, under-nutritional conditions – a known de-
terminant for TB morality – could contribute in the morality risk
in the individuals living in long-term care facilities.11 More aggres-
sive nutritional support individuals living in long-term care facili-
ties, who are in the high-risk group would be able to improve the
patient survival.
In our model, age greater than 65, TB meningitis, and chronic
kidney disease were the strongest predictors of mortality. These
findings are consistent with the observations of other authors.19–22
HIV co-infection was another strong predictor of TB mortality. As
previously suggested, high mortality has been observed in patients
with TB/HIV co-infection.23,24 Patients in our study, who had pul-
monary abnormalities on chest radiographs, had nearly three times
higher the odds of death. Christensen et al. described a similar
result where pulmonary TB patients have an almost two-fold in-
creased long-term mortality than extrapulmonary TB patients.25
With poor living conditions, it is also obvious that patients who
were homeless or reported being residents of a long-term care fa-
cility had a high rate of mortality. Different than Jung and cowork-
ers’ study that reported a higher mortality rate in the foreign-born
TB patients in the whole U.S.,26 our data showed a higher mortal-
ity rate in U.S.-born TB patients in Texas. Some critical factors may
contribute to this difference. First, the proportion of U.S.-born TB
cases was significantly higher in Texas than in the U.S. (42.7% ver-
sus 33.3% in 2015, respectively, P < 0.001).27 Second, since being
foreign-born has been emphasized as a strong risk factor for TB
disease in the Texas TB control program,4 foreign-born suspected
TB patients might get more attention and receive a more aggres-
sive management than U.S.-born suspected TB patients, especially
those who do not have a recent history of travelling abroad. Addi-
tionally, with a considerably higher number of teaching hospitals
and health facilities at Texas Medical Center in Houston, where al-
most 30% of the population are immigrants,28 foreign-born TB sus-
pects seem to receive a more prompt TB diagnosis and treatment
than in other areas in the U.S. Although the gender inequality issue
has been observed previously with a TB notification rate of twice
as high in men as in women,29,30 the finding that no differences
were found in the TB mortality during treatment across genders
suggests that male and female TB suspects received the same man-
agement when they come to a health facility in Texas.
To our knowledge, this is the first prognostic score for TB
mortality developed and validated using large population-based
surveillance data. In 2013, Horita et al. presented the very first TB
 D.T. Nguyen, E.A. Graviss / Journal of Infection 77 (2018) 283–290
prognostic score using a model of age, respiratory failure requir-
ing oxygen, serum albumin, and activity of daily living.11 However,
their study was conducted in a small sample size of AFB smear
positive hospital inpatients at a single facility in Japan and did
not include patients with HIV co-infection or multi-drug resistant
TB (MDR-TB). Pefura-Yone et al. developed another TB risk score
using a larger sample size. Nevertheless, their study also used
data from a single hospital and did not evaluate many sociode-
mographic (race/ethnicity, homelessness and resident of long-term
care facilities), clinical (comorbidity and different TB types) nor
radiographic characteristics.12 While the Horita et al. and Pefura-
Yone et al. models are more applicable for inpatients in hospital
settings, our TB mortality model is more practical as we used so-
ciodemographic and clinical characteristics which are available at
a patient’s first visit to any healthcare facility. Hence, our TB prog-
nostic score can be used in widely different public health settings.
Our study has some limitations. First, we could not completely
rule out the misclassification bias due to the exclusion of 1014 pa-
tients with missing outcome status and 652 patients with treat-
ment outcome coded other than “completed” or “died”. However,
similarity in most of the demographic and clinical characteristics
in a stratified analysis between the excluded and included patients
suggested that potential misclassification was minimal. Second, the
fact that the time of death was not available in the dataset down-
loaded from TBGIMS prevented us from developing an even more
robust survival model for this population. Third, HIV status was not
available for all patients in our surveillance data. The high mortal-
ity risk found in patients with unknown HIV status in our study
suggested the issue of under-diagnosis or underreporting of the
HIV infection.32 Although the reporting of HIV status has been im-
proving in recent years, the CDC has reported that approximately
13% of the people living with HIV in the U.S. do not know their
HIV status.31 Our model was developed and validated in the U.S.
and might need to be validated further in high TB incidence coun-
tries. Lastly, as Texas has a higher TB burden and morality than
in many other states, and has a more diverse population in which
one third are immigrants,26 the TB population in Texas may not be
representative of the TB population in the U.S. in general, there-
fore, external validation in different settings in the U.S. would be
also necessary.
In spite of the limitations, our study has many strengths such
as using a large dataset obtained from the state’s population-based
surveillance program over a 7 years period of time, all variables
could be collected at the patient’s initial visit and the model has
excellent discrimination and calibration in both developmental and
validation cohorts. In this light, our study findings are highly sig-
nificant to healthcare professionals dealing with TB disease in pro-
viding healthcare providers with a more practical and widely ap-
plicable prognostic scoring system.
Author contributions
Both authors: concept of study, study design, acquisition of
data, data analysis, and writing/revising the manuscript.
Declaration of interests
Both authors: No conflict of interest.
Acknowledgements
The authors acknowledge the selfless work of public health
officials and staff at the City of Houston Bureau of Tuberculosis
Control, Houston Department of Health & Human Services, Harris
County Public Health, TB Elimination Program, Texas Department
289
of State Health Services and the U.S. Centers for Disease Control
that made the data available for use in this analysis.
References
1. World Health Organization. Global tuberculosis report. 2016. Available at http:
//www.who.int/tb/publications/global_report/en/. [Accessed 12 July 2017].
2. Centers for Disease Control and Prevention. Tuberculosis – Data and Statistics.
2017. Available at https://www.cdc.gov/tb/statistics/. [Accessed 14 July 2017].
3. Boscoe F.P., Pradhan E.. The most distinctive causes of death by state,
2001–2010. Prev Chronic Dis 2015;12:E75.
4. Texas Department of State Health Services. Epidemiology and Supplemental
Projects Group Texas TB Surveillance Annual Report 2015. Austin, TX; 2016.
5. Lönnroth K., Corbett E., Golub J., Godfrey-Faussett P., Uplekar M., Weil D.,
et al. Systematic screening for active tuberculosis: rationale, definitions and key
considerations [State of the art series. Active case finding/screening. Number 1
in the series]. Int J Tuberc Lung Dis 2013;17:289–98.
6. World Health Organization. Systematic screening for active tuberculosis: principles
and recommendations, Geneva: World Health Organization; 2013. Available at
www.who.int/ tb/ publications/ Final_TB_Screening _guidelines.pdf . [Accessed 14 July
2017]..
7. Lewinsohn D.M., Leonard M.K., LoBue P.A., Cohn D.L., Daley C.L., Desmond E.,
et al. Official American Thoracic Society/Infectious Diseases Society of Amer-
ica/Centers for Disease Control and Prevention Clinical Practice Guidelines: di-
agnosis of tuberculosis in adults and children. Clin Infect Dis 2017;64 e1–33.
8. Lewinsohn D.M., Leonard M.K., LoBue P.A., Cohn D.L., Daley C.L., Desmond E.,
et al. Official American Thoracic Society/Infectious Diseases Society of Amer-
ica/Centers for Disease Control and Prevention Clinical Practice Guidelines: di-
agnosis of tuberculosis in adults and children. Clin Infect Dis 2017;64:111–15.
9. Dooley K.E., Tang T., Golub J.E., Dorman S.E., Cronin W.. Impact of diabetes mel-
litus on treatment outcomes of patients with active tuberculosis. Am J Trop Med
Hyg 2009;80:634–9.
10. Kourbatova E.V., Borodulin B.E., Borodulina E.A., del Rio C., Blumberg H.M.,
Leonard M.K. Jr. Risk factors for mortality among adult patients with newly di-
agnosed tuberculosis in Samara, Russia. Int J Tuberc Lung Dis 2006;10:1224–30.
11. Horita N., Miyazawa N., Yoshiyama T., Sato T., Yamamoto M., Tomaru K.,
et al. Development and validation of a tuberculosis prognostic score for smear–
positive in-patients in Japan. Int J Tuberc Lung Dis 2013;17:54–60.
12. Pefura-Yone E.W., Balkissou A.D., Poka-Mayap V., Fatime-Abaicho H.K., Enono-E-
dende P.T., Kengne A.P.. Development and validation of a prognostic score dur-
ing tuberculosis treatment. BMC Infect Dis 2017;17:251.
13. Centers for Disease Control and Prevention. Report of Verified Case of Tuber-
culosis (RVCT) Instruction Manual. 2009. Available at https://www.cdc.gov/tb/
programs/rvct/instructionmanual.pdf. [Accessed 20 July 2017].
14. Wasserman L.. Bayesian model selection and model averaging. J Math Psychol
20 0 0;44:92–107.
15. Dunson D.B., Herring A.H.. Bayesian model selection and averaging in additive
and proportional hazards models. Lifetime Data Anal 2005;11:213–32.
16. Rassi A. Jr, Rassi A., Little W.C., Xavier S.S., Rassi S.G., Rassi A.G., et al. Develop-
ment and validation of a risk score for predicting death in Chagas’ heart disease.
N Engl J Med 2006;355:799–808.
17. Panel on Antiretroviral Guidelines for Adults and Adolescents. Guidelines for the
Use of Antiretroviral Agents in Adults and Adolescents Living with HIV. Depart-
ment of Health and Human Services; Available at http://www.aidsinfo.nih.gov/
ContentFiles/AdultandAdolescentGL.pdf. [Accessed 2 May 2017].
18. Podlekareva D.N., Panteleev A.M., Grint D., Post F.A., Miro J.M., Bruyand M.,
et al. Short- and long-term mortality and causes of death in HIV/tuberculosis
patients in Europe. Eur Respir J 2014;43:166–77.
19. Cruz-Hervert L.P., Garcia-Garcia L., Ferreyra-Reyes L., Bobadilla-del-Valle M.,
Cano-Arellano B., Canizales-Quintero S., et al. Tuberculosis in ageing: high rates,
complex diagnosis and poor clinical outcomes. Age Ageing 2012;41:488–95.
20. Negin J., Abimbola S., Marais B.J.. Tuberculosis among older adults–time to take
notice. Int J Infect Dis 2015;32:135–7.
21. Reis-Santos B., Gomes T., Horta B.L., Maciel E.L.N.. The outcome of tuberculo-
sis treatment in subjects with chronic kidney disease in Brazil: a multinomial
analysis. J Bras Pneumol 2013;39:585–94.
22. Lee H.G., William T., Menon J., Ralph A.P., Ooi E.E., Hou Y., et al. Tuberculous
meningitis is a major cause of mortality and morbidity in adults with central
nervous system infections in Kota Kinabalu, Sabah, Malaysia: an observational
study. BMC Infect Dis 2016;16:296.
23. da Silva Escada R.O., Velasque L., Ribeiro S.R., Cardoso S.W., Marins L.M.S., Grin-
sztejn E., et al. Mortality in patients with HIV-1 and tuberculosis co-infection
in Rio de Janeiro, Brazil – associated factors and causes of death. BMC Infect Dis
2017;17:373.
24. Mabunda T.E.South Africa Ramalivhana NJ, Dambisya YM. Mortality associ-
ated with tuberculosis/HIV co-infection among patients on TB treatment in the
Limpopo province. Afr Health Sci 2014;14:849–54.
25. Christensen A.S., Roed C., Andersen P.H., Andersen A.B., Obel N.. Long-term mor-
tality in patients with pulmonary and extrapulmonary tuberculosis: a Danish
nationwide cohort study. Clin Epidemiol 2014;6:405–21.
26. Jung R.S., Bennion J.R., Sorvillo F., Bellomy A.. Trends in tuberculosis mortality
in the United States, 1990–2006: a population-based case-control study. Public
Health Rep 2010;125:389–97.
290 D.T. Nguyen, E.A. Graviss / Journal of Infection 77 (2018) 283–290

27. Centers for Disease Control and Prevention. Reported Tuberculosis in the United
States. 2015. Available at https://www.cdc.gov/tb/statistics/reports/2015/pdfs/
2015_surveillance_report_fullreport.pdf. [Accessed 12 August 2017].
28. Center for Public Policy Priorities. Fact Sheet: Immigrants in Houston 2016.
Available at http://forabettertexas.org/images/EO_2016_Factsheet_Immigrants_
HOU.pdf. [Accessed 10 August 2017].
29. Thorson A., Diwan V.K.. Gender inequalities in tuberculosis: aspects of infection,
notification rates, and compliance. Curr Opin Pulm Med 2001;7:165–9.
30. Neyrolles O., Quintana-Murci L.. Sexual inequality in tuberculosis. PLoS Med
20 09;6 e10 0 0199.
31. Centers for Disease Control and Prevention. HIV and Tuberculosis. 2017. Avail-
able at https://www.cdc.gov/hiv/pdf/library/factsheets/hiv-tb.pdf. [Accessed 1
August 2017].
32. Centers for Disease Control and Prevention. Reported HIV Status of Tuberculosis
Patients – United States, 1993—2005. MMWR Morb Mortal Wkly Rep 2007;56:3.