Table of Contents 1.0 TRIAL SUMMARY 2.0 TRIAL DESIGN 2.1 Trial Design 2.1 Trial Diagram
3.0 OBJECTIVE(S) & HYPOTHESIS(ES) 3.1 Primary Objective(s) & Hypothesis(es) 3.2 Secondary
Objective(s) & Hypothesis(es) 4.0 BACKGROUND & RATIONALE 4.1 Background 4.2 Rationale
4.2.1 Rationale for the Trial 4.2.2 Rationale for Endpoints 5.0 METHODOLOGY 5.1
Demographics and Baseline Characteristics Collection 5.2 Imaging Data Collection and
Process 5.2.1 PET/CT Imaging Collection and Anonymization 5.2.2 PET/CT Imaging Quality
Control 5.2.3 PET/CT Imaging Annotation 6.0 TRIAL PROCEDURES 7.0 STATISTICAL ANALYSIS 7.1
Statistical Analysis Plan Summary 7.2 Hypotheses/Estimation 7.3 Analysis Population 7.4
Statistical Methods for Study Endpoints 7.5 Statistical Methods for Baseline
Characteristics and Demographics 7.6 Sample Size and Power Calculations 8.0
ONFIDENTIALITY AND DATA SHARING PLAN 8.1 Confidentiality of Data 8.2 Data Sharing Plan
9.0 SPONSORS AND COLLABORATORS 9.1 Sponsors and Collaborators 9.2 Responsible
Party/Investigator 9.3 Role of Funding 9.4 Ethics Committee.1.0 TRIAL SUMMARY Brief Title: Prognostic Value of Neurometabolic Networks in CRC
(PVNM-CRC) Official Title: An Observational Study on the Prognostic Value of
Neurometabolic Networks in Colorectal Cance Trial Type: Observational Time Perspective:
Prospective Medical Context: Prognostic Study Population: Patients with colorectal cancer
Study Procedure: Eligible patients will be prospectively enrolled, and their images of
PET/CT Imaging and clinical data will be collected and analyzed, respectively. The data
will be applied to the prediction models to evaluate prognosis.
Study Groups: Patients will be recruited into one group, and will receive response
prediction by four distinct predictors respectively based on required images, which have
been previously constructed by investigators.
Sample Size: Approximately 213 patients will be enrolled. Estimated Duration:
Approximately 12 months from the time the first subject enrolled until the last subject.
Outcome Measures: Primary endpoints: area under curve (AUC) Secondary endpoints:
sensitivity, specificity, positive prediction value (PPV), negative prediction value
(NPV)
2.0 TRIAL DESIGN 2.1 Trial Design This is a prospective, observational clinical study for
validation of artificial intelligence (AI)-based prediction models for predicting
prognosis for colorectal cancer (CRC). Specifically, investigators are intended to verify
the prediction accuracy of the Brain-gut Risk Index for Disease in Gastrointestinal
Cancer Evaluation (BRIDGE), and whether it outperforms other conventional prediction
models based on clinical data.
Approximately 213 patients will be prospectively enrolled from The First Affiliated
Hospital of Zhengzhou University into a prospective validation dataset. All patients
perform whole body PET/CT imaging diagnosis and pathological biopsy. Qualified images of
PET/CT sequences will be collected and uploaded to a cloud platform, within which the
regions of tumor (ROIs) will be annotated by a professional team of radiologists. The
images data will be employed to distinct prediction models to generate prediction labels
for individuals, which are blind to both participants and physicians-in-charge.
3.0 OBJECTIVE(S) & HYPOTHESIS(ES) 3.1 Primary Objective(s) & Hypothesis(es)
- (1) Objective 1: Evaluate area under curve (AUC) of BRIDGE in predicting overall survival
(OS) for CRC patients.
Hypothesis (H1): BRIDGE achieves an AUC over 0.80 in predicting OS for CRC patients.
3.2 Secondary Objective(s) & Hypothesis(es)
1. Objective 2: Compare the AUC of BRIDGE with clinical prediction models in predicting
survival status for CRC patients. Hypothesis (H2): BRIDGE is superior to clinical
prediction models in terms of AUC in predicting OS for CRC patients.
2. Objective 5: Evaluate the sensitivity of BRIDGE in predicting OS for CRC patients.
3. Objective 6: Evaluate the specificity of BRIDGE in predicting OS for CRC patients.
4. Objective 7: Evaluate the positive prediction value (PPV) of BRIDGE in predicting OS
for CRC patients.
5. Objective 8: Evaluate the negative prediction value (NPV) of BRIDGE in predicting OS
for CRC patients.
4.0 BACKGROUND & RATIONALE 4.1 Background Colorectal cancer (CRC), with annually
increasing incidence and mortality worldwide, has become the second leading cause of
cancer-related death. The gastrointestinal (GI) tract comprises a complex ecosystem
with extensive interactions between normal or neoplastic epithelial cells with
immune, neuronal, and other cell types, as well as microorganisms and metabolites
within the gut lumen. Specifically, the intricate relationship between the GI tract
and the central nervous system (CNS), collectively known as the brain-gut axis,
plays a pivotal role in the pathogenesis of gastrointestinal disorders and neoplasm.
For instance, chronic stress increased the risk of colon cancer via activating the
COX-2/PEG2 system and promoted tumor cell dissemination by remodeling lymph
vasculature. The bidirectional communications of the brain-gut axis are generally
found to be mediated by neurotransmitters, inflammatory cytokines, metabolites, or
gut microbiota. Nonetheless, the spotlight has shone primarily on the brain-gut
crosstalk mechanisms in experimental cellular or animal models, with less attention
paid to the structural and functional alterations on the brain networks at the
patient level.
The evolution of functional neuroimaging modalities and neuroscience technologies
has enabled accurate delineation of CNS activities. Specifically, nuclear medicine
imaging technology using 2-[18F] fluoro-2-deoxy-D-glucose ([18F] FDG) to adopt
whole-body imaging information, is the optimal in vivo method for the investigation
of regional human brain metabolism and associations with systemic disorders. The
investigators have previously identified the neuronal metabolic-ventricular
dyssynchronization axis which might related to major arrhythmic events using
myocardial perfusion imaging and the brain [18F]FDG positron emission tomography
(PET).15 Given the potential dual interactions of the brain-gut axis, identification
of specific brain regions associated with CRC development and progression might lead
to a better understanding of the disease's neurobiological underpinnings and inform
the development of targeted therapeutic strategies. Hence, this study was structured
to elucidate the role of neuro-metabolism and its potential mediator in regulating
CRC tumorigenesis and metastasis. By delving into the neurometabolic-gut axis in
CRC, the resulting mechanistic insights might be leveraged to identify diagnostic
and prognostic biomarkers and to develop novel therapeutic interventions for CRC
patients.
4.2 Rationale 4.2.1 Rationale for the Trial Previously, investigators have
constructed a BRIDGE based on retrospective datasets. The study is conducted to
further prospectively verify the clinical applicability and generalizability of
BRIDGE in predicting OS for CRC patients. The prediction performance of BRIDGE will
be evaluated in a prospective dataset, and compared to conventional clinical-based
prediction models in the trial, which might potentially provide important evidence
for the feasibility and clinical value of integration of brain images for artificial
intelligence-aided GI cancer medicine.
4.2.2 Rationale for Endpoints The primary accuracy endpoint in the study is the AUC,
a significant indicator of classification performance of a binary classifier, which
has been widely used to evaluate model performance in the field of machine learning.
5.0 METHODOLOGY 5.1 Basic Information Collection and Serial Number Generation Once
enrollment, the basic information including demographics and baseline clinical
characteristics of each subject will be recorded. A unique tracking number will be
generated randomly for each subject that will be used to identify the subject for
all procedures in the trial.
5.2 Imaging Data Collection and Process 5.2.1 PET/CT Imaging Collection and
Anonymization For each subject, initial tumor imaging by PET/CT should have been
performed within 1-2 weeks after enrolled. The process for imaging collection and
transmission is manipulated in a uniform imaging protocol by radiologists and
technicians in participating institutions. The whole series of PET/CT scans should
be exported as DICOM files and completely anonymized with unique tracking number
before uploaded to the designated cloud platform.
5.2.2 PET/CT Imaging Quality Control Images of PET/CT scans acquired in sites will
be downloaded and reviewed by an independent radiologist experienced in PET/CT in
the central laboratory to ensure high image quality for analysis. Case that is lack
of any requisite sequence will be firstly ruled out. Images with insufficient
clarity, low resolution, motion artifacts or other disturbing factors that might
potentially affect imaging analysis will be further excluded.
5.2.3 PET/CT Imaging Annotation The regions of interest (ROIs) of brain within the
adequate sequences will be manually annotated by expert radiologists with at least 5
years' experience in PET/CT imaging. The Statistical Parametric Mapping (SPM) is the
preferred tool for imaging segmentation.
All cerebral images were first standardized into the Montreal Neurological Institute
(MNI) space using a 12-parameters affine transformation and subsequently non-linear
registration and resampled to 2 × 2 × 2 mm3 voxels. All normalized images were then
smoothed using an 8 mm3 full-width-at-half-maximum Gaussian kernel. Subsequently,
the intracranial tissues in the smoothed images were extracted using a brain mask
image in the MNI space. The standardized uptake values (SUVs) of all intracranial
voxels were summed to obtain the whole-brain SUV (SUVwhole-brain), which was used to
reflect the total glucose consumption of the brain. Ninety bilateral regions of
interest (ROIs) were selected in the MNI space using the Automated Anatomical
Labeling (AAL) atlas. The cerebellum was used as the reference region 1. The SUVmean
in a specific cerebral region was divided by the SUVmean of the cerebellum. Thus,
the SUV ratios (SUVRs) of the ROIs were calculated. A third senior professional
radiologist is responsible for dispute settlement and annotation review.
6.0 TRIAL PROCEDURES The trial procedures are summarized in the flow charts as
described below. The asterisk under the time course line indicates the possible
timepoint a subject might be enrolled in the trial. The procedure will be performed
accordingly.
7.0 STATISTICAL ANALYSIS This section outlines the statistical strategy and
procedures for the trial. If, after the trial has begun, changes made to primary
and/or key secondary hypotheses, or the statistical methods related to those
hypotheses, then the protocol will be amended accordingly.
7.1 Statistical Analysis Plan Summary Study Design Overview A Prospective,
Observational, Prognostic Trial to Validate the Prediction Accuracy and Performance
Superiority of a BRIDGE in Predicting OS for Colorectal Cancer (CRC) Prediction
Assignment Approximately 100 subjects newly diagnosed as CRC undergoing nCRT with
pathological tumor response unknown will be consecutively recruited into one
collective group. All subjects will be evaluated as 'predicted survival' or
'predicted non-survival' by four predictors independently based on required image
data 7.2 Hypotheses/Estimation Objectives and hypotheses of the study are stated in
Section 3.0. 7.3 Analysis Population All enrolled subjects that receive prospective
evaluation of predictors and provide prognosis will be included in this population.
7.4 Statistical Methods for Study Endpoints This section describes the statistical
methods that address the primary and secondary objectives.
Primary Accuracy Endpoints Area under curve (AUC) AUC is defined as the probability
that a randomly chosen positive example is ranked higher than a randomly chosen
negative example. A higher AUC indicates a better classification performance of a
definite predictor. The AUC is evaluated by calculating the area under curve of
receiver operating characteristics (ROC) which plots the proportion of true positive
cases (sensitivity) against the proportion of false positive cases (1-specificity)
based on various predictive probability threshold. The 95% confidence intervals
(95%CI) of AUC are generated by bootstrapping strategy in 1000 sampling times.
Secondary Accuracy Endpoints Sensitivity Sensitivity is defined as the proportion of
the predicted positive cases among the total actual positive cases, also known as
the true positive rate.
In the study, sensitivity of BRIDGE is evaluated by calculating the proportion of
the 'predicted survival' subjects among the total 'actual survival' subjects. The
95% confidence intervals (95%CI) of sensitivity are generated by bootstrapping
strategy in 1000 sampling times.
Specificity Specificity is defined as the proportion of the predicted negative cases
among the total actual negative cases, also known as the true negative rate.
In the study, specificity of BRIDGE is evaluated by calculating the proportion of
the 'predicted non-survival' subjects among the total 'actual non-survival'
subjects. The 95% confidence intervals (95%CI) of specificity are generated by
bootstrapping strategy in 1000 sampling times.
Positive prediction value Positive prediction value (PPV) is defined as the
proportion of the actual positive cases among the total predicted positive cases.
In the study, PPV of BRIDGE is evaluated by calculating the proportion of the
'actual survival' subjects among the total 'predicted survival' subjects. The 95%
confidence intervals (95%CI) of PPV are generated by bootstrapping strategy in 1000
sampling times.
Negative prediction value Negative prediction value (NPV) is defined as the
proportion of the actual negative cases among the total predicted negative cases.
In the study, NPV of BRIDGE is evaluated by calculating the proportion of the
'actual non-survival' subjects among the total 'predicted non-survival' subjects.
The 95% confidence intervals (95%CI) of NPV are generated by bootstrapping strategy
in 1000 sampling times.
Secondary Comparative Endpoints Delong's test Delong's test is a nonparametric test
for comparing AUC of two or more correlated ROC curves.
The AUC of BRIDGE will be compared to the AUCs of other two predictors respectively
by Delong's test. A two-sided p-value of less than 0.05 was considered significant.
Student's T test Student's T test is an inferential statistic used to determine if
there is a significant difference between the means of two datasets which follow
normal distributions and homogeneity variance.
The mean AUC of BRIDGE will be compared to the mean AUCs of other three predictors
respectively by Student's T test. A two-sided p-value of less than 0.05 was
considered significant.
7.5 Statistical Methods for Baseline Characteristics and Demographics The
comparability of the two groups with distinct pathological response (survival versus
non-survival) for each relevant baseline characteristics will be assessed by the use
of tables and/or graphs. The number and percentage of subjects in subgroups will be
displayed. Demographic variables (such as age and gender) and baseline
characteristics (such as clinical stage) will be summarized either by descriptive
statistics or categorical tables. Student's t-test or Wilcoxon signed-rank test will
be performed to compare continuous variables, while χ2 test or Fisher's exact test
for categorical variables.
7.6 Sample Size and Power Calculations The study will consecutively enroll
approximately 213 subjects. For AUC (H1), the study has ~85% power to detect an AUC
of 0.80 in BRIDGE at alpha = 0.05 (two-sided) with 213 subjects.
The sample size and power calculations were performed in the software PASS 15. 8.0
CONFIDENTIALITY AND DATA SHARING PLAN 8.1 Confidentiality of Data Data generated in
the trials will be considered confidential by the investigators, except to the
extent that it is included in a publication. During the trial, the subject will be
identified by unique tracking number.
8.2 Data Sharing Plan The subject data (deidentified participant information and
origin images of PET/CT and pathological slides) and the full study protocol will be
made available to the scientific community, immediately on publication, with as few
restrictions as possible. All requests should be submitted to the investigators for
consideration. A data use agreement will be required before the release of subject
data and institutional review board approval as appropriate.
9.0 SPONSORS AND COLLABORATORS 9.1 Sponsors The trial is sponsored by the First
Affiliated Hospital of Zhengzhou University.
9.2 Responsible Party/Investigator. The Study Protocol is completed and reviewed by all the authors. Principal
Investigators are responsible for the study protocol. The central contact
information is as following:
Yujie Bai, PhD Telephone: 0371-18801221165 Email:
[email protected] 9.3 Role of
Funding The funding of study is supported by National Natural Science Foundation of
China under Grant No.81872188, No.81902867, No.82001986, and No.81903152. The
funders have no role in the study design, data collection, data analysis, data
interpretation, writing of the report, or the decision to submit the report for
publication.
9.4 Ethics Committee This study is approved by the ethics committee of the First
Affiliated Hospital of Zhengzhou University.
