Clinical Trial Details
— Status: Not yet recruiting
Administrative data
| NCT number |
NCT06113861 |
| Other study ID # |
R01AI173021 |
| Secondary ID |
|
| Status |
Not yet recruiting |
| Phase |
|
| First received |
|
| Last updated |
|
| Start date |
May 2024 |
| Est. completion date |
October 2028 |
Study information
| Verified date |
February 2024 |
| Source |
Tulane University |
| Contact |
Richard Oberhelman, MD |
| Phone |
001-504-952-0964 |
| Email |
oberhel[@]tulane.edu |
| Is FDA regulated |
No |
| Health authority |
|
| Study type |
Observational [Patient Registry]
|
Clinical Trial Summary
Pediatric tuberculosis (TB) continues to pose diagnostic challenges in low- and middle-income
countries with high rates of TB disease, due to the well-described impact of paucibacillary
disease in children, and current TB culture and polymerase-chain reaction tests are of
limited usefulness due to cost, restricted availability, and poor sensitivity in specimens
available from younger children. Our team of experts from Tulane, Johns Hopkins University,
Universidad Peruana Cayetano Heredia, and Asociación Benéfica Prisma have confronted all of
these challenges through more than 25 years of collaboration in Peru and Bolivia. Our goal is
to directly address the challenges of TB in children by evaluating a new diagnostic approach
developed by MPI Tony Hu at Tulane University using a CRISPR-mediated TB assay (CRISPR-TB)
optimized to detect circulating Mycobacterium tuberculosis cell-free DNA (Mtb-cfDNA), and
used to analyze cryopreserved serum in pilot studies from adults and children with
presumptive TB, their asymptomatic household contacts, and a cohort of symptomatic children
living with HIV (CLHIV) at high risk for TB. Results from symptomatic adult cohorts yielded a
pooled sensitivity of 93%; specificity of 93%; positive predictive value of 95%; and negative
predictive value of 92%. In limited pilot studies in CLHIV CRISPR-TBD results accurately
identified all confirmed TB (13/13) and most children with unconfirmed TB (80%; 52/65). We
propose to enroll 200 presumptive TB cases and an equal number of well control subjects in
each of 2 study populations (test population and validation population) identified through
clinics associated with the "Dr. Mario Ortiz Suarez" Children's Hospital in Santa Cruz,
Bolivia. We will determine the distribution of cfDNA concentrations in peripheral blood in a
"test population" composed of two age-based groups of children (2 months-6 years, 7-14 years)
with respiratory disease grouped by likelihood of TB based on the NIH consensus case
definitions (confirmed TB, unconfirmed TB, and unlikely TB) and in age-matched controls
grouped by presence of latent TB infection (LTBI), with cfDNA measured serially in time among
TB cases receiving antibiotic therapy. We will also validate standard ranges of quantitative
cfDNA established for clinical subgroups of children with TB disease or LTBI in an
independent validation cohort. An additional aim will determine the correlation between
quantitative cfDNA and quantitative imaging-based TB scores based on evidence of disease in
the lung, the primary target organ in TB disease, by (1) chest radiograph, measured by
computer-aided analysis using the CAD4TB v7 system, and by (2) lung ultrasound, performed
with a portable/low-cost probe assisted by machine learning algorithms for automatic
interpretation. These biomarkers will be tested as potential cofactors that may be combined
with cfDNA levels in peripheral blood, to improve the detection of TB disease in children.
The results of this study will be the first step in a process to find a path to allow
detection of the many "unconfirmed" TB cases and ideally make the diagnosis of pediatric TB
in reach for low resource settings where it is so critically needed.
Description:
Pediatric subject populations, enrollment, and follow-up: During the first two years of the
study, a "test population" will be recruited for 18 months to establish quantitative cfDNA
standards and ranges for each clinical outcome group, to assess predictive values for cfDNA
levels as a biomarker of clinical outcome. We will characterize the dynamics of cfDNA levels
in peripheral blood in two age-based groups of children (2 months-6 years ["younger
children"], 7-14 years ["older children"]). A stratified analysis with these age-based
subgroups is logical because the diagnostic test yield and clinical presentations are
different in these groups.
For Aim 2 during years 3-4 of the study, a "validation population" will be recruited for 18
months to validate quantitative cfDNA standards and ranges for each clinical outcome group.
Subject recruitment, informed consent, data collection, and study groups for analysis will be
the same as for the "test population" in Specific Aim #1. Subjects in each study group will
be stratified into age-based subgroups (2 months-6 years, 7-14 years). Serial levels of cfDNA
will be assessed and characterized for children in the confirmed and unconfirmed TB groups,
to validate normalization of cfDNA values with effective anti-TB therapy.
Table 1. Project Timeline
Initial enrollment and data/specimen collection will be done prior to initiation of TB
treatment in the hospital or clinic setting. Inclusion criteria: Children ages 2 months to 14
years identified through the clinics and hospital wards of the "Dr. Mario Ortiz Suarez"
Children's Hospital in Santa Cruz and presenting for evaluation for symptomatic respiratory
disease and suspicion of tuberculosis will be eligible for enrollment (inclusion criteria
based on Bolivian Ministry of Health guidelines for suspect cases of tuberculosis in children
). After screening for exclusion criteria (prior treatment for TB within the past year,
current treatment for prevention of TB, weight < 2.5 kg., or clinical instability, positive
COVID-19 diagnostic test) study staff will present the study verbally to the parents and
provide a brochure with more detailed information designed for both parents and older.
Parental informed consent and pediatric participant assent will be obtained. The study will
include collection of specimens for diagnostic testing and clinical data as outlined below,
but decisions on treatment for tuberculosis will be made by the attending physician who is
not involved in the study. As the study groups for data analysis are determined in part based
on the results of diagnostic tests performed and on clinical response to treatment, subjects
will not be assigned to study groups on enrollment. Study group assignment (confirmed TB,
unconfirmed TB, and unlikely TB) will be determined by the project biostatistician after the
subject completes all study activities (see D.5. and D.7. study group assignment for criteria
and sample size by group). As the Bolivian guidelines for pediatric TB allow for clinical
evaluation of children with respiratory disease and only a few TB-related criteria, this
recruitment strategy will enroll a population of suspect TB patients that will allow us to
compare outcomes in "TB cases" (confirmed TB and unconfirmed TB) and in ill patients who do
not meet NIH case definitions for unconfirmed TB (i.e., the unlikely TB group, see D7). This
unlikely TB group will serve as "ill/respiratory disease controls", separate from the "well
controls" group. On a weekly basis, well control subjects without respiratory symptoms and
age-matched (+ 2 years) to suspected TB cases will also be recruited from community health
clinics. Well controls will only have a single set of specimens, and no invasive specimens.
D.4. Bolivia study staff: Recruitment of study participants and specimen collection
activities will be managed by a physician study coordinator, supervised by co-investigator
Dr. Ramiro Cabrera, pediatric pulmonologist and regional consultant for pediatric
tuberculosis in Santa Cruz. Clinical and patient related activities will be further overseen
by PIs Richard Oberhelman (pediatric infectious diseases specialist) and Robert Gilman (ID
specialist), as well as by ID specialists Jeffrey Tornheim and Lima-based Prisma site
director Carlton Evans.
D.5. Primary Outcomes, Statistical Power, and Sample Size The primary outcome for Aim 1 is
cfDNA levels for a) children with confirmed and unconfirmed TB ("TB cases"), b) children with
unlikely TB (UTB), and c) well children (WC). Based on preliminary data, the mean cfDNA
levels (and standard deviations) for these groups were 4.2 (4.0), 2.0 (3.2), and 1.1 (0.03),
respectively. We hypothesize that TB children will have significantly higher serum
concentrations of cfDNA than the UTB and WC groups. Statistical power and sample size: With a
two-sided type I error is set at 0.05, statistical power set at 80%, 42 subjects per group
are required to detect a significant difference between the TB and UTB children (based on the
formulae by Hulley et al. ). Comparisons between the TB and WC group requires 13 children per
comparison group. Based on data from current studies and local Ministry of Health (see D.7.
below) we anticipate 135 TB cases per year (55 in 2 mo-6 yrs.; 80 in 7-14 yrs.), or
approximately 200 TB cases per 18-month study period (for test population and validation
population). We also anticipate 80 unlikely TB cases per year (40 in 2 mo-6 yrs.: 40 in 7-14
yrs.), or approximately 105 unlikely TB cases per 18-month study period (18 months each for
test population and validation population recruitment). Subjects will be stratified into two
age groups (2 months-6 years, 7-14 years). With 42 required TB cases per age group, our
proposed sample exceeds the minimum requirements by 50% or more per age group. The enrollment
targets here will provide flexibility in meeting recruitment goals and non-response rates.
D.6. Procedures and specimen collection. For hospitalized patients with suspected
tuberculosis, enrollment, clinical evaluation, and sample collection are performed in the
same room; for outpatients the process of clinical evaluation, gastric aspirate sampling will
be performed in the consulting room of Pneumology clinic at the pediatric hospital and blood
sampling will be collected in the clinical laboratory phlebotomy room. Follow-up specimen
collection will be done through community clinics where patients receive their weekly
allotment of TB medicines, with follow-ups at 1- and 2-weeks post treatment initiation. A
2-month follow-up visit in conjunction with a routine physician visit is scheduled, to allow
for a later follow-up cfDNA biomarker specimen.
All participants under evaluation for TB will have procedures at the time of enrollment
including:
- Medical history and clinical evaluation required to accurately assess each criterion
that contributes to their subsequent assignment to TB outcome groups, based on the 2015
revised consensus criteria.
- Quantiferon-TB Gold In-Tube testing to detect immunologic evidence of tuberculosis
(including latent infection [LTBI]) will be performed in our laboratory at Universidad
Catolica, based on standard protocols
- HIV serology unless known to be HIV+, by Abbott Alere Determine HIV 1/2 (See D.8 Data
analysis for analytical adjustments by HIV serostatus)
- COVID-19 PCR or antigen test-- suspect TB patients only. (If positive these subjects are
excluded)
- Chest radiograph-- suspect TB patients only; not performed on well controls. AP and
lateral views. Digital images processed in DICOM will be transmitted to Dr. Zimic's lab
for analysis.
- Lung ultrasound-- suspect TB patients only; Not performed on well controls. Ultrasound
examination is performed based on a standard protocol supervised by Dr. Fentress,
imaging the chest in perpendicular planes in the midclavicular line anteriorly and
posteriorly from apices to diaphragm, and in the midaxillary line from axilla to
diaphragm, for a total of 12 views per participant. , ,
- Specimen collection for TB microbiology and PCR-- suspect TB patients only; not
performed on well controls. All subjects with a work-up for suspected TB will have at
least three sputum-equivalent specimens collected for TB microbiology and PCR analysis
by Xpert MTB/RIF. Acceptable specimens include (1) gastric aspirates, (2) expectorated
sputum (for older subjects who can produce sputum), or (3) string test specimens,
conducted by our standard protocol. Specimens will be collected in the early morning
prior to eating, and specimen collection times will be separated by at least 24 hours.
In cases where parents or attending physicians only agree to two specimens, a third
specimen will not be collected. Anti-TB treatment will be initiated when indicated by
the physician following collection of the last specimen for TB microbiology.
- Microbiologic analysis for TB isolation and detection will be performed on specimens
collected from suspect TB patients, including (1) Ziehl-Neelsen AFB smear, (2) TB
culture, and (3) and TB PCR detection by Xpert MTB/RIF. TB cultures will use the MODS
technique that we developed in Peru 8, 10
- Serum for cfDNA collected by venipuncture. We ship all serum cfDNA samples to Tulane by
World Courier, on dry ice, and we aliquot all samples so half will be shipped and the
other half stored at -80 C.
Participation by well control subjects will end after enrollment specimen collection and
evaluations. TB treatment will be initiated for all subjects with presumptive TB disease,
based on the attending physician's clinical opinion and based on Bolivian national TB
treatment standards. Study participants with compatible disease and on anti-TB therapy will
have the follow-up procedures listed at the times specified:
- Clinical evaluation (1 week, 2 weeks, and 2 months post enrollment and start of
treatment)
- Serum for cfDNA and whole blood for CBC (1 week, 2 weeks, and 2 months post enrollment)
- For subjects on antituberculous therapy and not demonstrating improvement-repeat
evaluation (per clinician decision at 2 weeks or 2 months post enrollment). Minimum 2
samples for smear, culture, and Xpert MTB/RIF (Gastric aspirates [by intubation/string
test] or sputum)
TB culture techniques. Specimens collected will be cultured for M. tuberculosis using the
Microscopic-Observation Drug-Susceptibility (MODS) Method.8 Standard volumes of each
decontaminated specimen will be inoculated into modified Middlebrook 7H9 media and cultured
in a sterile 24-well plate. Plates are placed in a plastic resealable bag, incubated at 37°C,
and examined every other day for up to 30 days by inverted light microscopy. Presumptive TB
isolates with cording are reported as positive and confirmed by IS 6110 PCR.
CRISPR-TBD assays: The circulating cfDNA will be extracted with Quick-cfDNA Serum & Plasma at
the Hu lab at Tulane. CRISPR-TBDB requires an PCR-based target amplification prior to
CRISPR-mediated fluorescent signal production, following procedures recently published by
Huang et. al. 23 The CRISPR-TBDB data will be evaluated in silico analysis using SnapGene
software (version 5.0.8) and by triplicate CRISPR-TBDB assays.
Chest radiograph and lung ultrasound studies and data processing. See Specific Aim #3 below.
D.7. Study group assignment. We will employ an observational study design with 5 study groups
of children ages 2 months to 14 years, as detailed below. These are A) confirmed pulmonary
TB, B) unconfirmed pulmonary TB, C) unlikely TB, D) well age-matched controls (which will be
divided into 2 groups based on presence or absence of LTBI). Most children evaluated for TB
and started on treatment as determined by the attending (non-study) physician will fall into
groups A (approximately 10%) and B (60-65%), based on recent data. Groups A and B together
are referred to here as "TB cases". Approximately 25-30% of subjects who meet entry criteria
based on Bolivian guidelines but fail to meet the 2015 NIH guidance definition for confirmed
or unconfirmed will be included as group C (unlikely TB), serving as an "other respiratory
disease" comparison group. The estimated eligible population is based on data from the
ongoing R21 study (for children aged 0-5 years) and on data from the Ministry of Health.
Criteria for each study group and estimated number per group to be recruited per year during
subject enrollment periods are:
A. Confirmed TB- estimated 65 subjects per year (15 in 2 mo. - 6 yr.; 50 in 7 - 14 yr.),
defined as:
- Symptomatic at initial enrollment, defined based on the clinical case definition from
the 2015 NIH Consensus Expert Panel 22 (Quantiferon Gold positive or negative).
- Positive M. tuberculosis culture or positive Xpert MTB/RIF from at least one biological
sample.
B. Unconfirmed TB-estimated 80 subjects per year (40 in 2 mo. - 6 yr.; 30 in 7 - 14 yr.)
defined as:
- Symptomatic at enrollment and clinical/radiographic evaluation consistent with TB, based
on the clinical case definition from the 2015 NIH Consensus Expert Panel 22 (Quantiferon
Gold positive or negative).
- Negative M. tuberculosis culture and negative Xpert MTB/RIF from all biological samples.
C. Unlikely TB--- 80 subjects per year (40 in 2 mo. - 6 yr.; 40 in 7 - 14 yr.), defined as:
- Symptomatic at enrollment BUT clinical/radiographic evaluation does NOT meet criteria
for Unconfirmed TB, per clinical definition-2015 Expert Panel 22 (Quantiferon Gold
positive or negative).
- Negative M. tuberculosis culture and negative Xpert MTB/RIF from all specimens. D. Well
Children Control Group-One for each child in groups A-C above, age matched for paired
case.
- Well children presenting for reasons other than respiratory diseases, and not
symptomatic for TB.
- Clinical and radiographic evaluation (if performed) NOT suggestive of TB, and NOT
diagnosed or treated for TB during the follow-up period (minimum 8 weeks).
- Divided into 2 subgroups based on presence or absence of LTBI (by Quantiferon Gold
assay).
D.8. Data analysis. Subjects will be sorted at the time of data analysis based on diagnostic
test results into "confirmed," "unconfirmed," and "unlikely" TB based on diagnostic test
results and clinical evolution. Since most children under age 7 with TB disease do not have
culture or Xpert confirmation, and unconfirmed TB cases are considered true cases for
epidemiologic purposes, these groups will also be combined as "TB cases" for analysis.
Subjects in each study group will be stratified into age-based subgroups (2 months-6 years,
7-12 years). For exploratory analyses, additional subgroups based on HIV serostatus will be
conducted both including and excluding HIV+ subjects (HIV seroprevalence is extremely low in
children in this population, so the impact of HIV status will likely be minimal). Sex will be
considered as an independent variable in all analyses. Additionally, sex will be included as
an interaction term against all primary predictor variables (time of observation, cfDNA
level). Serial levels of cfDNA will be assessed and characterized for children in the
confirmed and unconfirmed TB groups. Replicate cfDNA values will be standardized by modeling
their average values against their corresponding concentrations using four-parameter logistic
regression models. Mann Whitney U tests will be used for testing hypotheses comparing the TB
case group with the three comparison groups (children with unlikely TB, well children with
LTBI, and well children without LTBI). Exploratory analyses will compare outcomes (cfDNA
level, TB risk scores) between the "gold standard" confirmed TB (culture or Xpert MTB-RIF
positive) group and the unconfirmed (culture and Xpert negative) group. A Bonferroni
correction factor will be applied to account for multiple testing over the age group strata.
For response to treatment analysis (Specific Aim 2), a positive response is defined as per
2015 clinical case definitions consensus paper, measured at 1 wk., 2 wks., and 2 mo. post
enrollment, as either:
a) Response to antituberculosis therapy: clinical features suggestive of tuberculosis disease
that were present at baseline have improved, and there is no new clinical feature suggestive
of tuberculosis; OR (b) No response to antituberculosis therapy: clinical features suggestive
of tuberculosis disease that were present at baseline have not improved or have worsened.
The analytical objective is to compare response differences in cfDNA levels over time.
Generalized estimating equations (GEE) will be used to model response to treatment against
time of observation and cfDNA level. An interaction term will be included in the GEE model to
test whether temporal pattens in cfDNA levels differ by response type. Pairwise comparisons
will be performed to compare cfDNA levels between the response groups at each time point.
Weighted GEE models will be used to account for missing follow-up data. To assess time to
positive treatment response, Kaplan Meier curves will be applied, including treatment
response as a censoring variable and plotted against cfDNA levels (classified into tertiles)
over time.
D.9 Specific Aim #3: Determine the correlation between cfDNA levels and quantitative TB
disease risk score biomarkers determined by (1) chest radiograph, measured by computer-aided
analysis, and by (2) lung ultrasound, performed with a portable/low-cost probe and assisted
by machine learning algorithms for automatic interpretation without requiring a radiologist.
These biomarkers will be tested as potential cofactors that may combine with cfDNA levels in
peripheral blood, to improve the detection of TB disease in children. As a secondary goal, we
will examine chest radiograph and lung ultrasound findings using artificial intelligence
algorithms for interpretation, to explore potential associations between cfDNA and imaging
biomarkers. The fact that imaging findings are the most concrete evidence of disease
available among the many "soft" diagnostic criteria validates this exploratory analysis in
spite of limited sensitivity and specificity. If presence of lung pathology by imaging among
children with TB correlates with cfDNA level, this would be a finding with immense clinical
utility for health professionals in LMICs. However, interpretation of chest radiographs is
arguably the most challenging component of TB diagnostic criteria since it requires
sophisticated, expensive equipment and expert staff, both technicians for data collection and
health professionals for interpretation. We propose a multifaceted approach to the analysis
of imaging data, utilizing two unconventional approaches that are supported in the medical
literature, i.e. 1) lung ultrasound as a radiation-free alternative to chest radiograph that
is generally more portable, with several low-cost, portable platforms; and 2) utilization of
automated interpretation of both chest radiographs and LUS, using machine learning
computer-aided design (CAD) technology. Of several CAD technologies marketed for
interpretation of chest radiographs in patients with tuberculosis ages 15 and higher, only
one is extensively validated and CE certified in the European Union for use in children ages
4 and up-the Delft Imaging CAD4TB platform. , , CAD4TB processes DICOM images by chest
radiograph to produce a score from 0 to 100 to assess probability of active TB, as well as a
heat map image to indicate sections of the lung with TB-related pathology (see Fig. 7). The
CAD4TB software and technical support for this study will be donated by Delft Imaging (see
letter of support). In addition, Dr. Mirko Zimic will use machine learning approaches used to
develop algorithms for pediatric pneumonia by LUS to develop a similar strategy for pediatric
TB. Both the CAD4TB platform and the Zimic LUS algorithm will produce a TB risk score that
can be compared to radiologist's interpretations, and both the AI-based risk scores and
traditional risk scores will be correlated with cfDNA levels.
All ultrasound scans will be performed by trained staff using a General Electric VIVID i BT12
with a high frequency linear probe. The sonographer will record the presence and location of
abnormal ultrasound findings as per Fentress et. al. 16 We will record at least one video
from each sonographic view to allow for external review after scanning, for at least 12
ultrasound videos per participant.
D. 10. Imaging interpretation and analysis: Chest radiograph outcomes will be recorded in two
formats:
1. Radiographs will be reviewed independently by two pediatric radiologists who will note
the presence or absence of specific findings suggestive of PTB, as defined by the 2015
expert panel. 22 These findings include lymphadenopathy, air space opacification,
nodules, and pleural effusions. Reports from the radiologists will be compared to
determine congruency with "consistent with tuberculosis" as per 2015 clinical criteria.
In cases where the determination is discordant, a third radiologist will make a
determination. Radiologists will record findings and severity based on a standard scale,
to produce a risk score for analysis.
2. DICOM images will be evaluated using the CAD4TB v7 program, using published procedures,
to generate a TB risk score and heat map image of the lungs.
Conventional chest X ray scores are determined by the number of findings checked as positive
(out of a total n =8 possible findings) based on the NIH consensus paper CXR template. CAD4TB
is trained on independent annotated datasets to recognize features of TB, and it outputs a
score 0-100 which indicates the probability of TB. An abnormality heatmap is also generated
to indicate regions of the lung detected as abnormal. The analysis plan includes three
factors: Conventional X-ray or CAD4TB score (ordinal response variable), time of observation,
and cfDNA level. The analytical objective is to test for significant differences between
CAD4TB score values and cfDNA levels over time. GEE will be used to model CAD4TB scores
against time and cfDNA level. An interaction term will be used to test whether temporal
trends in cfDNA levels differ by CAD4TB score. Pairwise comparisons will be used to test for
significant differences between cfDNA levels and CAD4TB scores at each time point. Weighted
GEE models will be used to account for missing follow-up data. 31
Lung ultrasound (LUS) outcomes will similarly be evaluated by both conventional and automated
machine learning approaches. All saved ultrasound images will be reviewed by a lung
ultrasound expert blinded to the participant's clinical data. When there is disagreement
between the field sonographer's interpretation and the blinded LUS expert, this will be
adjudicated by a second blinded ultrasound expert. LUS images will be visually assessed and
classified as positive or negative for TB based on 2015 expert panel-defined lung findings
for chest radiographs applied to LUS. An artificial convolutional neural network (CNN) based
on lung ultrasound images will be trained for prediction of TB by methods previously
published by Zimic et. al.,22, 23 according to the gold standard classification based on NIH
criteria and chest radiographs. In previous studies, we developed a CNN for TB recognition in
MODS cultures which achieved 96% accuracy, 96% sensitivity and 96% specificity, demonstrating
that CNNs could assist or replace personnel for the automated diagnosis of TB.22, 23 The
training of the proposed CNN will be done using the Caffe and Tensorflow frameworks because
their speed and modularity. The system will be developed in Python using OpenCV libraries in
the detection step. Every frame of each LUS video will be labelled manually by the expert, to
determine if it has evidence of TB lesions. We will use a LeNet CNN variant model, relying on
two convolutional layers, each one followed of pooling operations like sub-sampling that will
account for translation invariance. The combination of the aforementioned layers is expected
to give a total of 60K learnable parameters. With approximately 280 children expected to be
enrolled in the first two years, each with 1-minute videos extracted from LUS at 30 frames
per second (1,800 frames), we will have a total of 504,000 images. Using standard data
augmentation, we expect to at least have 4x106 images for an adequate CNN training. Multiple
logistic regression analyses will be used to model the gold-standard TB classification TB,
including as predictors the cfDNA level and ultrasound CNN score, adjusting for age and sex.
LUS analysis will be performed with methodologies similar to those used for CXR, adapted for
LUS-based outcomes. LUS score will be adapted from the NIH consensus paper for CXR and will
consist of a combination of the quantitative number of findings checked as positive (out of a
total n=5: consolidation, small subpleural consolidation, cavity, pleural effusion, and
miliary pattern) and the number and location of abnormal lung regions, incorporated into a
logistic regression model. LUS will also be interpreted as "consistent with TB" or "not
consistent with TB", based on the same criteria used for CXR in the NIH consensus paper. The
GEE-based approach described above for CXR will also be applied to analysis of LUS data.
