Development and validation of a novel diagnostic nomogram to differentiate between

Development and Validation of a Novel Diagnostic
Nomogram to Differentiate Between Intestinal
Tuberculosis and Crohn’s Disease: A 6-year Prospective
Multicenter Study
Yao He, MD, PhD1
, Zhenhua Zhu, MD, PhD2
, Yujun Chen, MD1
, Fang Chen, MD1
, Yufang Wang, MD, PhD3
, Chunhui Ouyang, MD, PhD4
,
Hong Yang, MD, PhD5
, Meifang Huang, MD, PhD6
, Xiaodong Zhuang, MD, PhD7
, Ren Mao, MD, PhD1
, Shomron Ben-Horin, MD1,8
,
Xiaoping Wu, MD4
, Qin Ouyang, MD3
, Jiaming Qian, MD, PhD5
, Nonghua Lu, MD2
, Pinjing Hu, MD1
and Minhu Chen, MD, PhD1
OBJECTIVES: Differentiating Crohn’s disease (CD) from intestinal tuberculosis (ITB) remains a diagnostic challenge.
Misdiagnosis carries potential grave implications. We aimed to develop and validate a novel diagnostic
nomogram for differentiating them.
METHODS: In total, 310 eligible patients were recruited from 6 tertiary inflammatory bowel disease centers. Among
them, 212 consecutive patients (143 CD and 69 ITB) were used in the derivation cohort for the
establishment of diagnostic equation and nomogram; 7 investigative modalities including clinical
manifestations, laboratory results, endoscopic findings, computed tomography enterography features,
and histology results were used to derive the diagnostic model and nomogram. Ninety-eight consecutive
patients (76 CD and 22 ITB) were included for validation of the diagnostic model.
RESULTS: Eight out of total 79 parameters were identified as valuable parameters used for establishing diagnostic
equations. Two regression models were built based on 7 differential variables: age, transverse ulcer,
rectum involvement, skipped involvement of the small bowel, target sign, comb sign, and interferon-
gamma release assays (for model 1) or purified protein derivative (for model 2), respectively.
Accordingly, 2 nomograms of the above 2 models were developed for clinical practical use, respectively.
Further validation test verified the efficacy of the nomogram 1 with 90.9% specificity, 86.8%
sensitivity, 97.1% PPV, 66.7% negative predictive value (NPV), and 87.8% accuracy for identifying
CD, and the efficacy of the nomogram 2 with 100% specificity, 84.2% sensitivity, 100% positive
predictive value, 64.7% NPV, and 87.8% accuracy for diagnosing CD.
CONCLUSIONS: The derivation and validation cohorts identified and validated 2 highly accurate and practical diagnostic
nomograms for differentiating CD from ITB. These diagnostic nomograms can be conveniently used to
identify some difficult CD or ITB cases, allowing for decision-making in a clinical setting.
SUPPLEMENTARY MATERIAL accompanies this paper at http://links.lww.com/AJG/A22.
Am J Gastroenterol 2019;114:490–499. https://doi.org/10.14309/ajg.0000000000000064
INTRODUCTION
The confusing similarities between Crohn’s disease (CD) and
intestinal tuberculosis (ITB) make the diﬀerential diagnosis
between these 2 diseases a challenging issue. Arriving at the
correct diagnosis is particularly important in this era when bio-
logics are increasingly used for the treatment of CD, which may
1
Department of Gastroenterology, The First Affiliated Hospital of Sun Yat-sen University, Guangzhou, People’s Republic of China; 2
Department of Gastroenterology,
The First Affiliated Hospital of Nanchang University, Nanchang, People’s Republic of China; 3
Department of Gastroenterology, West China Hospital, Sichuan
University, Chengdu, Sichuan Province, People’s Republic of China; 4
Department of Gastroenterology, The Second Xiangya Hospital of Central South University,
Changsha, Hunan Province, People’s Republic of China; 5
Department of Gastroenterology, Peking Union Medical College Hospital, Peking Union Medical College,
Chinese Academy of Medical Sciences, Beijing, People’s Republic of China; 6
Department of Gastroenterology, Zhongnan Hospital, Wuhan University School of
Medicine, Wuhan, Hubei Province, People’s Republic of China; 7
Department of Cardiology, The First Affiliated Hospital of Sun Yat-sen University, Guangzhou,
People’s Republic of China; 8
Department of Gastroenterology, IBD Service, Sheba Medical Center and Sackler School of Medicine, Tel-Aviv University, Tel
Hashomer, Israel. Correspondence: Minhu Chen, MD, PhD. E-mail: chenminhu@mail.sysu.edu.cn. Pinjing Hu, MD. E-mail: pjhumd@vip.163.com
Received May 8, 2018; accepted November 8, 2018; published online February 7, 2019
ARTICLE490
The American Journal of GASTROENTEROLOGY VOLUME 114 | MARCH 2019 www.amjgastro.com
Copyright © 2019 by The American College of Gastroenterology. Unauthorized reproduction of this article is prohibited.
INFLAMMATORYBOWELDISEASE

portend fatal outcomes if used in intestinal tuberculosis (TB)
misdiagnosed as CD. Moreover, the rapidly increasing incidence
of CD on the one hand and likewise increasing incidence of TB on
the other in China (1,2) make the need for early and accurate
differential diagnosis between these 2 diseases even more perti-
nent. Although there is gold standard diagnostic, namely the
caseating granulomatous necrosis for ITB, the low detection rate
limited its value as a single diagnostic criterion (3).
Previously, a series of studies on this issue have been carried
out, but a small sample size, retrospective design, lack of definite
disease diagnostic standard, including only parts of investigative
modalities, and inconvenience in clinical use were major caveats
in these studies (4–9). Recently, we performed a prospective
study and reported that adding computed tomography enter-
ography (CTE) findings to the colonoscopic model improved
the diagnostic accuracy of diagnosing either CD or ITB from
66.7% (70/105) to 95.2% (5). However, when increasing
the sample size to 93 patients (CD/ITB: 60/33) to revalidate the
above model, the accuracy, specificity, and sensitivity for di-
agnosing CD were only 78.5%, 75.8%, and 80%, respectively (see
Supplementary Materials, Supplementary Digital Content 1,
http://links.lww.com/AJG/A22).
Thus, we performed this multicenter, prospective study and
established a novel diagnostic prediction model displayed as
nomogram to improve the accuracy and practicality in the
differential diagnosis of ITB and CD. Valuable indices from
clinical manifestations, laboratory examinations, special tests
(interferon-gamma release assays [IGRAs] and purified pro-
tein derivative [PPD] test), endoscopic findings, CTE features,
histological results, and polymerase chain reaction (PCR)
results were included in the analysis. To the best of our
knowledge, this was the first externally validated model that
prospectively combined together relative overall investigative
modalities from a multicenter with a large sample size, and the
result was displayed directly by nomogram, which was easily
applied to clinical practice.
METHODS
Study design and patients
A prospective, multicenter, observational Chinese study of ITB
and CD cases from July 2011 to May 2017 was conducted. The
study patients were prespecified to be divided into 2 cohorts:
a derivation cohort comprising all patients recruited since
study launching until April 2015 and an independent validation
cohort consisting of consecutive patients who were sub-
sequently recruited from that date onward until study termi-
nation. During the validation cohort, all investigators were
blinded to the results of the diagnostic model developed in the
derivation cohort. The study was approved by the institutional
review board, and all patients gave written informed consent to
participate.
We enrolled patients suspected of CD or ITB based on
clinical, laboratorial, and special tests (PPD and interferon-g
release assay); endoscopic, radiologic, and histological findings;
and PCR results from departments of gastroenterology of 6
tertiary hospitals in China. The criteria for enrolling patients
were as follows: consecutive cases with chronic intestinal in-
flammation on endoscopy, suspected as CD or ITB clinically,
plus (i) no empiric antituberculosis therapy used in suspected
ITB cases; (ii) no infliximab or azathioprine (AZA)/6-mercap-
topurine (MP)/methotreexate (MTX)/thalidomide used in
suspected CD cases, or if AZA/6-MP/MTX/thalidomide was
once used, it should be less than 12 weeks and stopped for at least
3 months.
The confirmed diagnostic criteria for ITB required one of
the following: (i) pathologic diagnosis: caseating granuloma
detected in intestinal biopsy, surgical specimen, or mesenteric
lymph nodes; (ii) clinical diagnosis: complete clinical recovery
accompanied by endoscopic mucosal healing with 6 months or
longer standard antituberculosis therapy and no recurrence
9–12 months after the initiation of antituberculosis therapy in
the follow-up. The confirmed diagnostic criteria for CD re-
quired one of the following: (i) pathologic diagnosis: pathology
of surgical specimen revealing the diagnosis of CD, with no
caseating granuloma detected in intestine or mesenteric lymph
nodes; (ii) clinical diagnosis: a favorable response to CD ther-
apy evaluated by clinical manifestations, laboratory examina-
tion, and endoscopy or CTE, and the clinical course compatible
with CD during at least 1 year of follow-up, with mucosal
healing being required at least one time during the follow-up
colonoscopy.
Endoscopic criteria of the efficacy of antituberculosis therapy
were detailed as follows (see Supplementary Materials, Sup-
plementary Digital Content 2, http://links.lww.com/AJG/A22).
Colonoscopies were performed at baseline, 3 months, and
9–12 months after the initiation of antituberculosis therapy for all
recruited cases. Endoscopic features including ulcers, mucosal
nodules, pseudopolyps, and luminal strictures, as well as their dis-
tribution and range were evaluated. Complete mucosal healing was
defined as disappearance of all ulcers, with/without polypoid
changes. Obvious endoscopic improvement was defined as a mu-
cosal ulcerated area reduced at least half or more when compared to
baseline. No improvement was defined as mucosal ulcerated area
reduced less than half, unchanged, or aggravated when compared to
baseline. The response to antituberculosis therapy in 3 months de-
termined the next therapeutic strategy. Antituberculosis therapy
would be continued in patients achieving complete mucosal healing
or obvious endoscopic improvement; otherwise, the patients would
be treated as having CD. Complete mucosal healing was required for
the confirmed diagnosis of ITB in 9–12 months of follow-up.
Data capture and analysis
Eight predefined data-recording forms were filled out by
pretrained specialists in each center. These included clinical
manifestations, laboratory examinations, special tests (PPD
test and interferon-g release assay), PCR results, endoscopic
findings, CTE and histological features, and response to different
therapeutic strategies (see Supplementary Materials, Supple-
mentary Digital Content 3, http://links.lww.com/AJG/A22). The
specialists responsible for recording endoscopic, CTE, and histo-
logical features and evaluating the results of empiric antitubercu-
losis therapy were trained before the initiation of this study to
ensureconformance to the standards. A length ofulcer $ 4 cm was
required to diagnose a longitudinal ulcer, and transverse ulcer was
defined as a circular ulcer if it occupied $ 1/2 intestinal circum-
ference (see Supplementary Materials, Supplementary Digital
Content 4, http://links.lww.com/AJG/A22). Scleroma more than
20 mm or local blisters, hemorrhage, necrosis, and lymphangitis
were defined as PPD test strong positive. Seventy-nine variables
derived from the 7 designated investigative modalities were first
analyzed for their discriminatory power between CD and ITB in
the derivation cohort.
© 2019 by The American College of Gastroenterology The American Journal of GASTROENTEROLOGY
Differentiation Between Intestinal Tuberculosis and… 491

Informative variables’ screening and diagnostic prediction
model building
The prediction model was developed using a 2-step approach.
First, a nonparametric, data-driven approach was used to identify
potentially informative differential variables based on random
forest from the primary data set. Second, a nomogram including
informativevariablesidentified fromstep 1was built onthebasis of
multivariable logistic regression analysis to provide the clinician
with an intuitive and quantitative tool to predict diagnosis of CD
or ITB.
Details of the modeling process are as follows. In step 1,
a random forest regression model was applied to rank
the informative differential variables according to their
permutation importance, and variables with a negative im-
portance score were excluded from further analysis (Ran-
domForestSRC package in R). Random forest is an ensemble
classification method that determines a consensus prediction
for each observation by averaging the results of many
individual recursive partitioning tree models. Each of the in-
dividual trees is fitted to a randomly selected subset of the
observations and uses a random subset of the available pre-
dictors at each node as candidates for splitting. In step 2,
multivariable logistic regression analysis began with the in-
formative differential variables identified from step 1, which
were applied to develop a diagnostic model for differentiating
CD from ITB using the derivation cohort. Backward stepwise
selection was applied using the likelihood ratio test with
Akaike’s information criterion as the stopping rule. The
results were displayed as nomogram to provide clinician with
an intuitive and quantitative tool to predict the probability of
CD or ITB.
Validation of nomogram
Prediction model formed in derivation cohort was applied to an
independent cohort of patients to validate and evaluate the pre-
diction efficacy.
Figure 1. Flow diagram of the final diagnosis of patient enrollment.
Table 1. Multivariate regression analysis including IGRAs
Variables Estimate s.e. OR 95% CI (lower) 95% CI (upper) P value
Age 20.1067 0.0315 0.8988 0.8450 0.9560 7e-040
IGRAs 25.3784 0.9324 0.0046 7e-040 0.0287 0.0000
Comb sign 1.6029 0.8018 4.9676 1.0319 23.9139 0.0456
Rectum involvement 1.6741 0.9002 5.3342 0.9137 31.141 0.0629
Transverse ulcer 23.6674 1.1619 0.0255 0.0026 0.2491 0.0016
Skipped involvement of the small bowel 1.1812 0.7590 3.2583 0.7361 14.4224 0.1196
Constant 6.1982 1.5005 491.3589 25.9529 9,302.7432 0.0000
CI, confidence interval; OR, odds ratio.
INFLAMMATORYBOWELDISEASE He et al.492

Statistical analysis
Statistical analyses were conducted using R software, version 3.1.3
(The R Foundation for Statistical Computing, www. R-project.
org), and SAS, version 9.2 (SAS Institute, Cary, NC). The reported
statistical significance levels were all 2-sided, with statistical sig-
nificance set at 0.05.
RESULTS
Patient recruitment
Three hundred seventy-two patients were screened, and 62
patients were excluded because of refusal to study enrollment,
inadequate data collection, and wrong or indefinite diagnosis. In
total, 310 cases that met the inclusion criteria were included in the
Table 2. Multivariate regression analysis including PPD test
Variables Estimate s.e. OR 95% CI (lower) 95% CI (upper) P value
Age 20.0444 0.0179 0.9566 0.9236 0.9907 0.0132
Target sign 0.9534 0.5686 2.5946 0.8513 7.9075 0.0936
Rectum involvement 1.9344 0.6758 6.9198 1.8401 26.0224 0.0042
Transverse ulcer 22.2955 0.5831 0.1007 0.0321 0.3158 1e-040
Skipped involvement of the small bowel 2.0544 0.5025 7.8025 2.9138 20.8936 0.0000
PPD test 23.3814 1.1288 0.0340 0.0037 0.3107 0.0027
Constant 1.50474 0.7001 4.5030 1.1417 17.7604 0.0316
CI, confidence interval; OR, odds ratio.
Figure 2. Nomogram 1 for model 1.

study. Of these, 212 patients were included in the derivation co-
hort (143 CD and 69 ITB) and 98 in the validation cohort (76 CD
and 22 ITB) (Figure 1).
Informative variables identified based on random forest analysis
In step 1 of the diagnosis prediction model development,
the most important 8 informative differential variables were
identified based on random forest analysis (see Supplementary
Materials, Supplementary Digital Content 5, http://links.lww.
com/AJG/A22). The remaining variables were excluded since
including more variables had very little influence on the
diagnostic efficacy (see Supplementary Materials, Supplemen-
tary Digital Content 5, http://links.lww.com/AJG/A22).
Nomogram construction
Diagnostic equation was built on the basis of multivariable lo-
gistic regression analysis as follows: LogitP 5 7.12531 2
0.10900 * AGE 2 7.47109 * (IGRAs 5 1) 1 0.31906 * (Target
sign-CTE 5 1) 1 2.14456 * (Comb sign-CTE 5 1) 1 3.10319 *
(Rectum involvement 5 1) 2 4.48391 * (Transverse ulcer 5 1)
1 2.19195 * (Skipped involvement of the small bowel 5 1) 2
7.05513 * (PPD test strong positive 5 1). Age, IGRAs, Comb
sign, rectum involvement, transverse ulcer and skipped
involvement of the small bower were included in model 1, and
age, target sign, rectum involvement, transverse ulcer, skipped
involvement of the small bower, and PPD test were included in
model 2. Due to the high cost of IGRAs which limited its use in
China and the similar value of IGRAs and PPD test in the di-
agnosis of TB, we constructed 2 regression models including
either IGRAs (model 1, Table 1) or PPD test (model 2, Table 2
and other variables. The diagnostic equation built for model 1
was LogitP 5 6.1972 2 0.1067 * age 2 5.3784 * IGRAs 1 1.6029
* Comb sign 1 1.6741 * Rectum involvement 2 3.6674 *
Transverse ulcer 1 1.1812 * Skipped involvement of the small
bowel, and the result was displayed as nomogram shown in
Figure 2 The diagnostic equation built for model 2 was LogitP 5
1.50474 2 0.0444 * age 1 0.9534 * Target sign 1 1.9344 *
Rectum involvement 2 2.2955 * Transverse ulcer 1 2.0544 *
Skipped involvement of the small bowel 2 3.3814 * PPD test
strong positive, and the result was displayed as nomogram
shown in Figure 3. Comparison between model 1 and model 2
showed that the specificity, sensitivity, and accuracy of model 1
are superior to those of model 2 (Table 3). Receiver operating
curve analysis shows that the area under the curve of model 1
was 0.977, which was superior to 0.930 in model 2 (P , 0.05)
(Figure 4).
Figure 3. Nomogram 2 for model 2.

Validation of the nomogram
Among the 98 consecutive eligible patients recruited to the vali-
dation cohort, 76 had CD and 22 ITB. Efficacy of nomogram 1 for
diagnosingCD at thecutoff point P 5 0.5 showed90.9% specificity,
86.8% sensitivity, 97.1% positive predictive value (PPV), 66.7%
negative predictive value (NPV), and 87.8% diagnostic accuracy;
and nomogram 2 showed 100% specificity, 84.2% sensitivity, 100%
PPV, 64.7% NPV, and 87.8% diagnostic accuracy.
Construction of algorithm
As shown in Figures 2 and 3, one total point corresponds to one
linear predictor which demonstrated the possibility of diagnosis
of CD. Based on the above results, we proposed an algorithm for
the differential diagnosis between CD and ITB (Figure 5a).
Chronic intestinal inflammation was suspected as CD or ITB
without caseaous granulomas or acid-fast bacilli on mucosal bi-
opsy, no extra-intestinal active TB was recognized either, so the
nomogram was applied to the case. Crohn’s disease is diagnosed
and treated with strategy for CD when the total point is $170 for
nomogram 1 (Figure 5b) and $292 for nomogram 2, namely the
possibility of diagnosis of CD $90%; otherwise, a 3-month em-
piric antituberculosis therapy should be initiated for avoiding the
dissemination of TB in the suspected case. Evaluation of the en-
doscopy finding decided the next therapeutic strategy. Mucosal
healing or obvious improvement in endoscopy findings shows
a possible ITB diagnosis, whereas nonimproved or aggravated
endoscopy results show a possible CD diagnosis. When the al-
gorithm to 98 cases from validation cohort was applied, none
of the ITB patient was misdiagnosed as having CD, while 23.7%
(18/76) of CD patients needed empiric anti-TB therapy with
nomogram 1 and 32.9% (25/76) CD patients needed empiric anti-
TB therapy with nomogram 2.
DISCUSSION
Given the many similarities between CD and ITB in clinical
manifestations, laboratory results, endoscopic findings, CTE/
magnetic resonance enterography (MRE) features, and even
histological features, there are a dire need and challenges for
developing better diagnostic tools to differentiate between CD
and ITB in TB endemic area, which is experiencing an increase in
the incidence of CD and embracing more and more biologics
(10–13). Several studies from Asian countries with a high ITB
prevalence rate reported that the rate of misdiagnosis of CD as
ITB was as high as 21%–65%, indicating the difficulty in dis-
tinguishing between these 2 diseases even for experienced
physicians (14–17). The caseaous granulomatous necrosis was
considered a gold standard for diagnosis of ITB, but it is un-
commonly found on histological examination and was reported
to be 4%–40% in ITB in studies retrospectively, albeit including
only a small number of cases (3,10,18). Mycobacterium culture
was another gold standard and reported to achieve 44.1% sensi-
tivity in identifying ITB in a retrospective study (19), but the strict
requirements for the specialized media, culture condition, and
a long incubation time limited its use in clinical practice, which
was not included in our study. And there is still no gold standard
for CD.
Some studies reported that identification of valuable param-
eters from different investigative modalities such as clinical
manifestations, endoscopic features, CTE/MRE findings, serol-
ogy, interferon-g assay, or histological features was helpful for the
differential diagnosis between CD and ITB, and a few studies
attempted to construct diagnostic models with multiple valuable
parameters and showed a modest diagnostic yield (3,8,10,20,21).
We previously performed a multicenter retrospective study in-
cluding clinical and endoscopic parameters and established
a scoring model for the differential diagnosis between CD and
ITB, in which 100% specificity in diagnosing CD or ITB and
39.7% differential diagnostic accuracy were achieved. A study
from Korea established a diagnostic model with colonoscopy
parameters and showed a positive predictive value for CD as
94.4%, a positive predictive value for ITB as 88.9%, and accuracy
95.5% (22). Zhao et al. (8) also established models based on
clinical and CTE features, which showed diagnostic accuracy ofFigure 4. Receiver operating curve of model 1 and model 2.
Table 3. Comparison between model 1 and model 2
Test Model 1 with
IGRAs
Model 2 with
PPD test
P value
ROC area (AUC) 0.9770 0.9300 0.0208
95% CI lower 0.9459 0.8851 —
95% CI upper 0.9921 0.9621 —
Best threshold 0.4756 1.1783 —
Specificity 0.9242 0.9091 —
Sensitivity 0.9580 0.8252 —
Accuracy 0.9474 0.8517 —
Positive LLR 12.6462 9.0769 —
Negative LLR 0.0454 0.1923 —
Diagnose OR 278.5667 47.2 —
N for diagnose 1.1334 1.3619 —
Positive pv 0.9648 0.9516 —
Negative pv 0.9104 0.7059 —
AUC, area under the curve; CI, confidence interval; LLR, log-likelihood ratio;
OR, odds ratio; ROC, receiver operating curve.

91.0% and 95.7%, respectively. All above models were established
based on single or limited examination tools, and no validation
was performed. A study from India recruited parameters from
clinical manifestation, endoscopic findings, and pathologic fea-
tures and established a diagnostic model with 83% sensitivity,
79.2% specificity, and 81.1% diagnostic accuracy (9). A recent
meta-analytic Bayesian model for differentiating ITB from CD
showed 90.9% sensitivity, 92.6% specificity, and 91.8% accuracy
for diagnosis of ITB in the validation cohort (23). The above
studies reported the diagnostic yield of differential diagnosis
with big variations, and most of them either were retrospective
studies, or had small sample size, or included only one or some
examination tools. Recently, we performed a prospective study
and reported that adding CTE findings to a Korean-built
colonoscopy model improved the diagnostic accuracy of di-
agnosing CD (5). Recently, Bae JH et al. developed and
Figure 5. (a) Algorithm for differential diagnosis between CD and intestinal tuberculosis. (b) CD possibility demonstrated by nomogram 1. Total point $170
corresponding to the possibility of CD $90% (see the cross of the red line).

validated a prediction scoring model integrating colonoscopy,
laboratory, and radiologic parameters from 117 patients and
reported an increased diagnostic accuracy from 81.2% to 96.3%
by including more investigative tools (24). The results of the
above 2 studies hinted that diagnostic model including valuable
parameters extracted from more examination tools might bring
better results.
The present study used a derivation–validation methodo-
logical investigation to develop and validate a diagnostic
prediction model and to display as nomogram for differenti-
ating CD from ITB. The model was built on the basis of 79
candidate indices prospectively collected from clinical mani-
festations, laboratory examinations, special tests, endoscopic
findings, CTE features, histological results, and PCR results
from 6 tertiary inflammatory bowel disease (IBD) centers,
which makes the model more systematic and robust than
previous models which included only parts of investigative
modalities (5,22,24). Eight potential informative predictors
Figure 6. Example of adopting nomogram in the differential diagnosis.
Figure 7. Example of the effect of antituberculosis therapy.

were identified by random forest analysis for the preparation
of model establishment. The use of random forest analysis to
identify informative variables according to their permutation
importance from the primary data set could avoid the statis-
tical defects of overfitting when using univariate analysis in
the situation of too many variables with insufficient sample
size.
The present study incorporated variables from clinical, labo-
ratory, IGRAs/PPD, endoscopic, histological, imaging, and PCR
features into 2 nomograms, which were proven to be of excellent
performance in external validation cohort. The intuitive, easy-to-
use, and high-accuracy characteristics of the nomogram will
facilitate its wide application and improve its efficacy in the dif-
ferential diagnosis of CD and ITB in clinical practice. As shown in
Figures 2 and 3, the score of each variable can be easily acquired;
after calculating the total scores of the 6 variables, the possibility
of diagnosing a patient as having CD was intuitively demon-
strated. For example, a 47-year-old male presented with diarrhea,
fever, and erythema in lower limb for 4 months; laboratory
findings were as follows: erythrocyte sedimentation rate 106 mm/
h, C-reactive protein 81.3 mg/L, platelet 434 3 109
/L, PPD test
positive, IGRAs positive, chest x-ray negative, CTE skipped in-
volvement of the small bowel. Endoscopy showed multiple ir-
regular ulcers distributed through ileum terminal to rectum.
Some ulcers presented as transverse ulcers, and histological
testing showed chronic inflammation with ulcer and multiple
epitheloid granulomas, some merged up to 400 mm in diameter.
Atrial fibrillation staining was negative. No gold standard was
discovered. We adopted nomogram 1 to this case, and the result
was as follows: Age 47 5 42.5, IGRAs1 5 0, Comb sign2 5 0,
Rectum involvement1 5 21, Transverse ulcer 1 5 0, Skipped
involvement of the small bowel1 5 14.5, calculating the sum of
above scores reaching a total point of 78 which showing that the
possibility of diagnosing CD was below 10% (Figure 6). There-
fore, we treated this patient with antituberculosis therapy (2
HRZE/6 HR, H: isoniazid; R: rifampicin; Z: pyrazinamide; E:
ethambutol) and achieved mucosal healing in 9 months
(Figure 7).
In the present study, we constructed 2 models including ei-
ther IGRAs or PPD test considering the similar significance of
them in evaluating TB and the limitation of widely using IGRAs
in China due to its high cost. The comparison of these 2 models
showed that although both models achieved high accuracy,
which was 0.9474 in model 1 and 0.8517 in model 2, the model
including IGRAs was superior to the one including PPD test.
Therefore, IGRAs should be recommended as priority selection
for the differential diagnosis of CD and ITB.
The present multicenter, prospective study had several
advantages. First, the study had a large sample size and rep-
resentative results. Six tertiary IBD centers from cities located
in the east, south, west, north, and central parts of China
participated in the study; therefore, the results were relatively
representative of the situation in China. Second, we performed
a prospective study. This allowed us to recruit study candidates
on the basis of strict preset diagnostic criteria, recruiting
and excluding criteria. Third, informative variables were
identified using random forest analysis which could avoid
overfitting phenomenon in many models established on the
basis of univariate analysis due to limited sample size. Fourth,
the model was displayed as nomogram which was intuitive
and easy to use in clinical practice. Fifth, external validation
study was carried out and verified the high efficacy of these
models in the differential diagnosis of CD and ITB. Finally,
we proposed an algorithm for the differential diagnosis be-
tween CD and ITB, which could minimize the misdiagnosis of
ITB as CD, especially in high-incidence areas of TB such as
China.
ACKNOWLEDGEMENTS
This work was supported by Health Research and Special Projects
Grant of China (Grant No.201002020), National Natural Science
Foundation of China (Grant No. 81670607), and Guangzhou Science
Technology and Innovation Commission (Grant No.
201707010091).
Study Highlights
WHAT IS KNOWN
3 Given similarities between CD and ITB in clinical
manifestations, laboratory results, endoscopic findings, CTE/
MRE features, and even histological features, differentiating
CD from ITB remains adiagnostic challenge in many endemic
regions of the world.
3 Misdiagnosis carries potential grave implications, especially
with the increasing use of anti-TNF agents for CD.
3 The low sensitivity of the caseaous granulomatous necrosis
and mycobacterium culture, gold standard diagnostics for
ITB, limited their value in clinical use.
3 Although diagnostic models with multiple valuable
parameters were constructed for differential diagnosis in
many previous studies, no easy-to-use diagnostic model was
built in clinical practice.
WHAT IS NEW HERE
3 This prospective, observational, multicenter study was
conducted in 6 tertiary IBD centers from cities located
in the east, south, west, north, and central parts of
China and included the largest sample size to our
knowledge.
3 Diagnostic models were built with multiple valuable
parameters extracted from the most comprehensive
investigative modalities so far to our knowledge and displayed
as nomogram which was intuitive and easy to use in clinical
practice.
3 External validation study was carried out and verified the high
efficacy of these models in the differential diagnosis of CD and
ITB.
3 An algorithm for the differential diagnosis between CD and
ITB was built based on nomogram.
3 These 2 diagnostic models and the algorithm can improve the
differential diagnosis rate between CD and ITB, especially for
the inexperienced clinicians.
3 Although tremendous efforts to control TB have been
undertaken at the global and national levels over the
past decades, almost 2 million patients still die every year due
to drug resistance. Empiric antituberculosis therapy is one of
the causes for developing multidrug resistance strains. Our
effective and easy-to-use differential diagnostic nomograms
can help avoid overusing empiric antituberculosis therapy as
a differential diagnosis measure.

CONFLICTS OF INTEREST
Guarantor of the article: Minhu Chen, MD, PhD.
Specific author contributions: Yao He and Zhenhua Zhu
contributed equally to this work. M.C., S.B.-H., X.W., Q.O., J.Q., N.L.,
and P.H. contributed to the study concept and design. Y.H., Z.Z.,
Y.C., F.C., Y.W., C.O., H.Y., M.H., and R.M. collected the data. M.C.,
P.H., Y.H., Z.Z., and X.Z. analyzed and interpreted the data. M.C.,
Y.H., and Z.Z. drafted the manuscript, and all authors critically re-
vised the manuscript for important intellectual content. M.C. and
P.H. obtained the funding to conduct the study. All authors approved
the draft submitted.
Financial support: This work was supported by Health Research and
Special Projects Grant of China (Grant No.201002020), National
Natural Science Foundation of China (Grant No. 81670607), and
Guangzhou Science Technology and Innovation Commission (Grant
No. 201707010091).
Potential competing interests: None.
REFERENCES
1. Ng SC, Tang W, Ching JY, et al. Incidence and phenotype of
inflammatory bowel disease based on results from the Asia-pacific
Crohn’s and colitis epidemiology study. Gastroenterology 2013;145:
158–615.
2. National Center for Tuberculosis Control and Prevention, China CDC.
http://www.chinatb.org/. Accessed on July 1, 2018.
3. Kirsch R, Pentecost M, Hall Pde M, et al. Role of colonoscopic biopsy in
distinguishing between Crohn’s disease and intestinal tuberculosis. J Clin
Pathol 2006;59:840–4.
4. Jung Y, Hwangbo Y, Yoon SM, et al. Predictive factors for differentiating
between Crohn’s disease and intestinal tuberculosis in Koreans. Am J
Gastroenterol 2016;111:1156–64.
5. Mao R, Liao WD, He Y, et al. Crohn’s disease from intestinal tuberculosis:
A potential diagnostic algorithm computed tomographic enterography
adds value to colonoscopy in differentiating. Endoscopy 2015;47:322–9.
6. Huang X, Liao WD, Yu C, et al. Differences in clinical features of Crohn’s
disease and intestinal tuberculosis. World J Gastroenterol 2015;21:3650–6.
7. Kedia S, Sharma R, Nagi B, et al. Crohn’s disease from intestinal
tuberculosis computerized tomography-based predictive model for
differentiation of Crohn’s disease from intestinal tuberculosis. Indian
J Gastroenterol 2015;34:135–43.
8. Zhao XS, Wang ZT, Wu ZY, et al. Differentiation of Crohn’s disease from
intestinal tuberculosis by clinical and CT enterographic models. Inflamm
Bowel Dis 2014;20:916–25.
9. Makharia GK, Srivastava S, Das P, et al. Crohn’s disease and intestinal
tuberculosis clinical, endoscopic, and histological differentiations
between. Am J Gastroenterol 2010;105:642–51.
10. Epstein D, Watermeyer G, Kirsch R. Review article: The diagnosis and
management of Crohn’s disease in populations with high-risk rates for
tuberculosis. Aliment Pharmacol Ther 2007;25:1373–88.
11. Yao T, Matsui T, Hiwatashi N, et al. Crohn’s disease in Japan: Diagnostic
criteria and epidemiology. Dis Colon Rectum 2000;43(10 Suppl):S85–93.
12. Shin DH, Sinn DH, Kim YH, et al. Increasing incidence of inflammatory
bowel disease among young men in Korea between 2003 and 2008. Dig
Dis Sci 2011;56:1154–9.
13. Leong RW, Lau JY, Sung JJ. The epidemiology and phenotype of Crohn’s
disease in the Chinese population. Inflamm Bowel Dis 2004;10:646–51.
14. Liu TH. [Crohn’s disease: Differential diagnosis with intestinal
tuberculosis (Author’s Transl)]. Zhong hua Nei Ke Za Zhi 1981;20:211–5.
Chinese.
15. Isbister WH, Hubler M. Inflammatory bowel disease in Saudi Arabia:
Presentation and initial management. J Gastroenterol Hepatol 1998;13:
1119–24.
16. Amarapurkar DN, Patel ND, Rane PS. Diagnosis of Crohn’s disease in India
wheretuberculosisiswidelyprevalent.WorldJGastroenterol2008;14:741–6.
17. Das K, Ghoshal UC, Dhali GK, et al. Crohn’s disease in India: A
multicenter study from a country where tuberculosis is endemic. Dig Dis
Sci 2009;54:1099–107.
18. Gu Q, Ouyang Q, Zhang WY, et al. A comparison of clinical and
pathologic characteristics between Crohn’s disease and intestinal
tuberculosis. Zhonghua Nei Ke Za Zhi 2009;48:291–4. Chinese.
19. Ye BD, Yang SK, Kim D, et al. Diagnostic sensitivity of culture and drug
resistance patterns in Korean patients with intestinal tuberculosis. Int J
Tuberc Lung Dis 2012;16:799–804.
20. Ng SC, Hirai HW, Tsoi KK, et al. Systematic review with meta-analysis:
Accuracy of interferon-gamma releasing assay and anti-Saccharomyces
cerevisiae antibody in differentiating intestinal tuberculosis from Crohn’s
disease in Asians. J Gastroenterol Hepatol 2014;29:1664–70.
21. Park YH, Chung WS, Lim JS, et al. Diagnostic role of computed
tomographic enterography differentiating Crohn disease
from intestinal tuberculosis. J Comput Assist Tomogr 2013;37:
834–9.
22. Lee YJ, Yang SK, Byeon JS, et al. Analysis of colonoscopic findings in the
differential diagnosis between intestinal tuberculosis and Crohn’s disease.
Endoscopy 2006;38:592–7.
23. Limsrivilai J, Shreiner AB, Pongpaibul A, et al. Meta-analytic bayesian
model for differentiatingintestinal tuberculosis from Crohn’s disease. Am
J Gastroenterol 2017;112:415–27.
24. Bae JH, Park SH, Ye BD. Development and validation of a novel
prediction model for differential diagnosis between Crohn’s disease and
intestinal tuberculosis. Inflamm Bowel Dis 2017;23:1614–23.

Development and validation of a novel diagnostic nomogram to differentiate between

Recommended

Recommended

More Related Content

What's hot

What's hot (19)

Similar to Development and validation of a novel diagnostic nomogram to differentiate between

Similar to Development and validation of a novel diagnostic nomogram to differentiate between (20)

Recently uploaded

Recently uploaded (20)

Development and validation of a novel diagnostic nomogram to differentiate between