Trial Title:
The AIDPRO-CRC Trial
NCT ID:
NCT06645015
Condition:
Colo-rectal Cancer
Conditions: Official terms:
Rectal Neoplasms
Conditions: Keywords:
Prehabilitation
AI
Artificial Intelligence
Risk-estimation
perioperative
Post-operative complications
Risk-stratification
Algorithm
Study type:
Interventional
Study phase:
N/A
Overall status:
Not yet recruiting
Study design:
Allocation:
Randomized
Intervention model:
Parallel Assignment
Primary purpose:
Supportive Care
Masking:
Single (Participant)
Intervention:
Intervention type:
Device
Intervention name:
AI augmented risk-stratification
Description:
A state-of-the-art artificial intelligence (AI) model called AIDPRO manual CRC is used as
a decision support tool to estimate the 1-year mortality risk for each patient
Arm group label:
AI-augmented risk-stratification
Summary:
The AIDPRO-CRC trial aims to improve outcomes for patients undergoing surgery for
colorectal cancer by using artificial intelligence (AI) to assist surgeons in risk
assessment. The trial will evaluate whether AI can help surgeons better predict the risk
of complications and death, leading to improved care, fewer complications, and better use
of healthcare resources.
In this nationwide, randomized clinical trial, participants will be divided into two
groups. One group will have their risk assessed by a surgeon using standard clinical
methods, while the other group will have their risk assessed by a surgeon using AI
assistance. Based on the risk level, patients will receive varying levels of
perioperative care. The AI-assisted risk assessment aims to tailor the treatment more
precisely to each patient's individual needs, thereby reducing complications,
readmissions, and mortality.
The trial will be conducted at eight hospitals across Denmark, involving patients
diagnosed with colorectal cancer who are scheduled for curative surgery. All participants
will receive standard treatment according to national guidelines, with the only
difference being the method of risk assessment.
The primary hypothesis is that AI-assisted risk assessment will lead to better patient
outcomes compared to standard surgeon-led assessments. The trial also aims to explore
whether this approach can lead to more efficient use of healthcare resources.
The study builds on a successful pilot project (AID-SURG) that showed promising results
in reducing complications, hospital stays, and readmissions.
This study is a researcher-initiated, nationwide, randomized clinical trial involving
patients diagnosed with colorectal cancer across eight hospitals in Denmark. Participants
will be randomly assigned to one of two groups: AI-assisted risk assessment or standard
surgeon-led assessment. The intervention focuses on optimizing perioperative care based
on individual risk levels determined by either AI or the surgeon's clinical judgment.
Detailed description:
Detailed description:
The AIDPRO-CRC trial is an investigator-initiated nationwide multicenter randomized
controlled trial. The trial aims to investigate the clinical effects of an AI-augmented
solution "AIDPRO manual CRC" for optimization of perioperative treatment by personalized
risk stratification of patients undergoing CRC surgery. The protocol adheres to the
SPIRIT Statement recommendations.
The AIDPRO-CRC trial is a pre-market, pivotal stage, confirmatory clinical investigation
designed to evaluate the safety and effectiveness of the AIDPRO manual CRC algorithm. As
a pivotal clinical investigation, this study is critical for generating the robust
evidence required to support regulatory submissions for CE marking. The investigation
involves an interventional approach, meaning that participants will undergo specific
procedures or treatments as part of the study. This design has been specifically chosen
to rigorously assess the performance of the AIDPRO manual CRC device in a real-world
clinical setting, providing the necessary data to demonstrate its safety and
effectiveness. The results from this trial will be used to seek CE marking, enabling the
AIDPRO manual CRC to be brought to market in the future.
All patients referred to a colorectal MDT with clinical suspected or histopathological
verified CRC will undergo eligibility assessment for inclusion in the study. Only
patients with confirmed clinical and/or pathological evidence of CRC and scheduled for
curative CRC surgery will be considered eligible for participation in the study. All
patients will be treated according to local, national and international guidelines and
recommendations by qualified colorectal surgeons. Eligible patients with newly diagnosed
CRC cancer planned for curative intended surgery will be randomly exposed to either an
AI-augmentation risk stratification model (intervention arm) or to surgical
assessment-based risk stratification model (control arm) for perioperative treatment
aiming to optimize their performance leading up to the planned surgery, and thus striving
to reduce post-surgery complications. The randomization allocated 1:1 will happen on the
individual patient basis, and on a per site level.
Only the individual colorectal surgeon, who is doing the initial input of data into the
randomization software and executes the randomization process gets to know the risk
prediction modality of the individual patient.
Throughout the perioperative optimization phase, it is crucial to highlight that the
patient is always blinded to the modality of risk prediction. The surgeon performing the
randomization, who is also involved in the assessment and treatment of the patient, may
therefore be potentially biased in their evaluation. All other assessors involved in the
perioperative treatment are sought to be blinded whenever possible to ensure fair and
unbiased treatment for all participating patients in the trial. An independent data
manager will provide the data in a blinded format. Unblinding to the investigators will
occur only after all analyses have been completed and the primary investigator has
prepared two versions of the abstract for the primary publication-one in which the
results favor the intervention (AI-augmented risk-prediction) group and one in which they
favor the control group (expert-based risk-prediction), if a difference between the
interventions is identified.
Study flow and interventions:
The AIDPRO-CRC trial investigates the impact of using an AI-augmented risk estimation
software device called the "AIDPRO manual CRC" as a decision support tool versus
traditional surgeon-based risk stratification for determining the type of perioperative
treatment trajectories patients undergoing elective curative-intent surgery for newly
diagnosed colorectal cancer (CRC) should be offered. This description will cover the
nature of the intervention, treatment protocols, and follow-up procedures involved in the
trial.
Study flow:
All new patients referred with clinical suspected or histopathological verified CRC to
one of the participating centers will follow local standards of care including diagnostic
evaluation and assessments as defined by the relevant local and national guidelines. This
all takes place at the "screening phase". No new assessments or evaluations to the
patients are introduced because of the AIDPRO-CRC trial at this phase, and all relevant
information to be used at the time of screening will be collected as a part of the
standard workflow at each center. The time a patient spends in the screening phase is not
influenced by the AIDPRO-CRC trial, and it is the responsibility of each participation
center to adhere to the current wait-times in this phase as defined in the relevant
Danish legislation.
At the MDT-conference in the screening phase the decision is made whether the patient is
planned for immediate curative intended surgery, or is qualified for neoadjuvant
treatment with the potential for later curative intended surgery as per relevant clinical
guidelines and assessment. If the patient is scheduled for neoadjuvant treatment, this
treatment must take place before any preoperative optimization is performed.
The AIDPRO manual CRC device and the randomization module for patient allocation to
intervention- and control-arm are hosted within a dedicated application called "MDTplus".
This application is designed in collaboration with Koncern Digitalisering, Region
Sjaelland and is to comply with relevant legislations and rules regarding patient related
health data in a cross-regional trial setup. The MDTplus application is designed as a
platform to display patient related health information of interest in relation to
treatment for CRC. The use of the MDTplus application in the AIDPRO-CRC trial is limited
to providing a platform for execution of the randomization-sequence and AIDPRO manual CRC
device to be used in the trial.
Patients who have been discussed at the multidisciplinary team (MDT) conference with
confirmed CRC and hereafter are planned for curative intended surgery, or have the
potential for curative intended surgery following relevant neo-adjuvant treatment, will
undergo eligibility assessment. When a patient has given informed consent and they are
eligible for enrollment in the study, an investigator can execute the randomization
process within the MDTplus application. To standardize the procedure for the clinical
experts, across the intervention and control arms of the study, the same data needed for
the AIDPRO manual CRC device to make a risk-estimation is entered into the MDTplus
application for all patients, before the randomization procedure takes place. This data
can be populated in the application before the patient is enrolled in the study anytime
during the screening phase, but the randomization sequence can only be executed when the
patient has met all inclusion criteria and has given informed consent for study
participation. If the patient is not enrolled in the study the data in the MDTplus
application will be automatically deleted within 90 days, (except data of age and gender
to be used for the screening log). This approach, with uniform input of data before the
randomization process takes place, ensures that the randomization is then controlled by
the application and conducted at the patient level.
To ensure that the AIDPRO manual CRC device only makes risk estimations on patients which
are enrolled in the study and have provided informed consent, a unique patient consent ID
code should be entered into the software platform under the section of "Patient Samtykke
ID". This unique code is provided enclosed to each informed consent form, and no two
codes throughout the study are alike. When entering this code, the enrolling doctors
hereby ensures that the patient have provided informed consent and are meeting the
inclusion criteria. The patient ID code is also used for randomizations sequence,
ensuring that each patient is randomized following the previous described randomization
sequence, which is built into the backend of the MDTplus application. However, no
references to which trail intervention arm can be made from a patient ID code.
For one group of patients, the application will provide the suggested 1-year risk of
mortality and corresponding risk group A, low risk (<1% risk of 1-year mortality), B,
moderate risk (1-5% risk of 1-year mortality), C, high risk (5-15% risk of 1-year
mortality) or D, very high risk (>15% risk of 1-year mortality) group (based on the
AI-augmentation model) - constituting the intervention arm (augmented decision process).
For the complementary group of patients, the software will only prompt the surgical
experts to provide the risk group (A, B C or D) - and this will constitute the control
arm. The assigned risk groups will explicitly determine the individualized treatment
trajectories each patient is offered, regardless of whether they are in the intervention
(AI-augmentation model-based risk stratification) or control arm (expert-based risk
stratification). This study design implies that all patients enrolled in the study,
regardless of whether the patient is in the intervention- or control arm, gets offered
the same perioperative treatment, the only difference are whether the decision of what
treatment strata to allocate for the individual patient is made by an AI-augmented risk
prediction model or a surcical expert. Patients not enrolled in the trial, but still
eligible for treatments offered by the respective surgical department will still have the
possibility to benefit from the perioperative treatment interventions, as these
perioperative treatment interventions are part of standard clinical practice at each
participating center, and enrollment in the AIDPRO-CRC trial are not a criteria for
receiving the treatment at each participating site.
The randomization design will ensure that the allocation of patients to the different
trial arms is unbiased and patient-blinded at the time of inclusion, balancing potential
inter- and intra- center variation. This approach will further standardize and balance
potential variation due to the effect of learning (all colorectal surgeons will learn
from the same supervised learning paradigm).
After the randomization process and risk stratification procedure the patient will adhere
to one of the four defined trajectories in the preoperative phase, named A, B, C, and D;
which are all part of standard offered treatment to patients with newly diagnosed CRC at
each participating site.
Study flow for patients scheduled for neo-adjuvant treatment If a patient is scheduled
for neo-adjuvant treatment, the relevant treatment as per national guidelines and by the
MDT-conference decision, must be completed before the perioperative optimization included
in the relevant treatment trajectory , to which either an AI-augmented decision
(interventions-arm) or surgical expert (control-arm) have allocated can begin. The
relevant perioperative treatment should be scheduled to begin either 2 or 4 weeks before
the possible date of surgery, depending on which treatment trajectory the patient is
allocated to. If a patient which have received neoadjuvant treatment no longer have
indication for curative intended surgery at the response evaluation they will be excluded
from the study as pr. exclusion criteria's.
The study design implies that patients are to be stratified into relevant treatment
trajectories before any neo-adjuvant treatment has taken place immediately after the
first initial MDT-conference and must complete the neo-adjuvant treatment before any
perioperative optimization can take place.
Intervention: AI-Augmented vs. Surgeon-Based Risk Stratification:
In the AIDPRO-CRC trial, the sole intervention is the method of risk stratification
(AI-augmented vs. surgeon-based) used to determine perioperative treatment plans. All
other aspects of treatment, including the specific perioperative care bundles, follow
established national guidelines and are consistent across all trial sites and
interventions-arms, and are described in appendix 1. The trial's objective is to assess
whether AI-augmented risk stratification can reduce postoperative complications and
improve outcomes compared to traditional clinical assessments by surgical experts. There
are two arms in the trial:
- AI-Augmented Risk Stratification (Intervention Arm)
- Description: A state-of-the-art artificial intelligence (AI) model called
AIDPRO manual CRC is used as a decision support tool to estimate the 1-year
mortality risk for each patient.
- Purpose: The AI model uses various patient data points to predict the risk and
tailor the perioperative treatment bundle accordingly.
- Expected Outcome: By leveraging AI, the trial aims to enhance the precision of
risk stratification, potentially reducing postoperative complications and
improving patient outcomes.
- Surgeon-Based Risk Stratification (Control Arm)
- Description: Experienced surgical experts perform the risk stratification based
on their clinical judgment and existing guidelines.
- Purpose: Traditional clinical assessments are used to categorize patients into
appropriate risk groups and assign perioperative treatment bundles.
- Expected Outcome: This arm serves as the control to evaluate the effectiveness
of the AI model against standard clinical practices.
AI-Augmented risk stratification algorithm:
The AIDPRO manual CRC is an in-house medical device (class IIa) designed for clinical
decision support for colorectal surgeons managing the perioperative care planning of
patients undergoing curative intended elective surgery for colorectal cancer. The
description of the AIDPRO manual CRC adheres to the SPIRIT-AI extension guidelines.
The AIDPRO manual CRC device is developed to be used by specialized colorectal surgeons
or colorectal surgeons in training during multidisciplinary team conferences and the
subsequent surgical consultation at the participating colorectal centers in the
AIDPRO-CRC trial as the specific needs of target patient groups cannot be met, or cannot
be met at the appropriate level of performance, by an equivalent CE-marked device
available on the market. The intention is to obtain a CE-mark for the AIDPRO manual CRC
device.
The AIDPRO manual CRC is version 1.0.0 "v1" of a prediction model designed for clinical
decision support developed and manufactured by Center for Surgical Sciences (CSS) and
builds upon learnings from the "AIS-1" model used in the AID-SURG pilot study.
The tool's core is a prediction model trained on a national cohort in Denmark. The device
takes risk factors as input and outputs an estimation of the risk of dying within one
year following surgery, and a suggestion for risk strata (A, B, C or D) based on the
predicted risk. Colorectal surgeons in multidisciplinary teams or following preoperative
clinical follow-up may use the tool's suggested risk strata for decision support to
choose between evidence based perioperative treatment trajectories.
The device is only intended to provide clinical decision support to colorectal surgeons
at multidisciplinary team conferences and the subsequent surgical consultation and
physical examination in relation to adult patients (age > 18 years) undergoing elective
curative intended surgery for colorectal cancer at the participating colorectal centers
in the AIDPRO-CRC trial. The device is developed to be used by specialized colorectal
surgeons (Doctors of Medicine/"M.D.'s") or colorectal surgeons in-training at the
participating centers. The surgeons should be proficient/native in Danish and English.
The device is NOT intended to be used for deciding if surgery should be performed, for
non-curative intended surgery, for non-elective surgical procedures, or in the surgical
treatment of all children or adults with other oncological diagnoses than colorectal
cancer. The device is NOT intended to be used in acute or emergency care settings. The
device is NOT intended to be used outside the participating colorectal centers in the
AIDPRO-CRC trial at this time.
Algorithm development and covariate selection The dataset upon which the model is build,
and trained are consisting of data sources from the Danish Colorectal Cancer Group (DCCG)
database (DCCG.dk)covering all patients with newly diagnosed CRC in Denmark, the Danish
National Patient Registry (DNPR), the Danish Register of Laboratory Results for Research
database (RLRR)and the Medicinal products Statistics database (MED). These data sources
were all transformed into the Observational Medicine Outcomes Network Common Data Model
(OMOP-CDM) and merged, utilizing open-source tools from the Observational Health Data
Science and Informatics (OHDSI) community. Data was collected with the initial index date
set to the date of surgery and a one-year time-at-risk.
For data quality, OMOP-CDM ensured mandatory fields like sex and age were complete. In
the Danish Colorectal Cancer Group (DCCG) data, mutually exclusive covariates were
separated from missing values. In other sources, distinguishing between absent and
missing records was challenging due to the nature of electronic health record (EHR) data.
Categorical variables were one-hot encoded, interpreting missing and negative values as
zero, while missing continuous values were treated as NA and excluded.
Approximately 15.000 different variables in the harmonized OMOP-database were addressed
using both a data-driven and clinical expert-based approach. Using a supervised feature
selection process utilizing six different modalities of problem-based feature selection
the initial 15.000 variables were narrowed down to approximately 100 clinical variables
were selected for the basis of the AIDPRO-algorithm. Based on a criterion of availability
for manual input by the surgeon this selection of variables was further narrowed down to
40 variables. A selection of 12(excluding patient ID-number, CPR) variables were then
identified for use in the AIDPRO-algorithm for risk assessment using both a data-driven
and clinical expert-based approach.
Model performance was assessed using the area under the receiver operating characteristic
curve (AUROC) and the area under the precision-recall curve (AUPRC). Calibration was
evaluated using calibration intercept, slope, and calibration-in-the-large, with visual
representations of receiver operating characteristic curves, precision-recall curves, and
calibration plots. This approach ensured a systematic procedure for data acquisition and
selection, robust handling of poor-quality data, and effective regression analysis with
sophisticated covariate selection tools.
The surgeons are to access the AIDPRO manual CRC device via a secure connection provided
by the MDTplus application. The manufacturer, in collaboration with the sponsor and
Principal Investigator (PI), will maintain an access log throughout the study, ensuring
that only authorized and qualified personnel can access the algorithm. If the study is
terminated for any reason, access to the software will be remotely disabled for all
investigators.
The surgeon will manually input the required input-data into the AIDPRO manual CRC
device. The covariates needed for AIDPRO manual CRC device to run should be carefully
retrieved from the patient's medical records or clinically assed. It is crucial that the
information for each variable that is used by the model is retrieved and inputted
according to the following description, to guarantee the closest correspondence between
the real-life patient data and the data that were used to train and test the model.
Output of AIDPRO manual CRC device:
The output of the prediction should be interpreted as an estimation of a single patient's
risk of mortality within 1 year following elective curative intended surgery, given the
entered covariates. Based on the estimated 1-year risk of mortality, a corresponding risk
stratum is suggested (A (<1%), B (1-5%), C (5-15%) or D (>15%)). The output should only
be considered as decision support to choose the type of perioperative care offered to the
patient, however, ultimately the decision will rely on the medical doctor treating the
patient. It should be noted, that the individual risk estimated is based on a model
describing a historical national cohort, that several risk factors and covariate
relationships might not be included in the model, and that the covariates contributions'
to the risk or the risk estimate can NOT be interpreted as causal or used in causal
inference. If the individual colorectal surgeon finds that the AIDPRO manual CRC device
is not in accordance with the surgeon's assessment of the patient's risk prediction, the
surgeon always has the option to "overrule" the AI-augmented suggested decision and
thereby place the patient in a different risk strata. This must be recorded in the
MDTplus application, the patients EHR and subsequently noted in the patient's eCRF.
Each of the four possible clinical trajectories (strata A-D) in which the AIDPRO manual
CRC device allocates a patient to in the intervention-arm, includes a number of clinical
interventions pre-, intra-, and postoperatively.
Only qualified colorectal surgeons with extensive experience in assessing patients risk
factors will use the AIDPRO manual CRC algorithm in the trial. The surgeons who will use
the AIDPRO manual CRC device in the AIDPRO-CRC trial will receive both written training
material as well as structured supervised training set in a simulation environment in
order to familiarize with the tool, and use it accordingly to the manufactures
instructions. A detailed description describing the instructions for use can be found in
in the Technical Documentation regarding the Instruction for Use of the AIDPRO manual CRC
device:
Expert-Based Risk Stratification:
In the expert-based risk stratification arm (control-arm), experienced surgical experts
utilize their clinical judgment and adhere to established guidelines to assess and
categorize the risk levels of patients undergoing colorectal cancer surgery.
After the surgeon have provided the needed variables in the MDTplus platform and have
successfully randomized the patient, the surgeon is prompted to categorize the patient
into one of the four risk stratification strata A-D and subsequent treatment
trajectories, based upon their expected 1-year risk of mortality (A (<1%), B (1-5%), C
(5-15%) or D (>15%)). The surgeons are only to categorize the patient in a category and
should not make a precise guess of expected risk of 1-year mortality in percent. In the
MDTplus application, the patients EHR and in the eCRF it will be registered which surgeon
who has provided the risk-stratification.
The surgeon is to perform the risk-stratification based on clinical presentation and
expert-based analysis and evaluation of various patient factors, including medical
history, physical examination findings, diagnostic test results, any relevant
comorbidities of the clinical presentation of the patient, taking into account previous
health record and inputs made at the MDT-conference. The surgeon-based risk-estimation
should not be limited to the covariates the AIDPRO-algorithm has available for
risk-stratification in the intervention-arm, and should thus reflect an average
colorectal expert based assessment of the patients 1-year estimated risk of mortality.
After the surgeon has provided the risk-stratification for patients in the control-arm,
the patients will receive the same perioperative treatment as patients placed in similar
risk-stratification-categories by the intervention-arm.
Perioperative Treatment Bundles Perioperative treatment bundles to which either patients
in the interventions arm of the study (AI-augmentation model-based risk stratification)
or patients in the control arm of the study (expert-based risk stratification) are
assigned to, are not unique to the study. The clinical interventions are evidence-based
and adhere to Danish and international guidelines in colorectal surgery. Interventions,
individually and collectively aim to counteract the perioperative risk to improve
postoperative outcome. The exact distribution of the interventions between the clinical
trajectories is based on the risk profile of patients in each trajectory. When the
patient is allocated, the perioperative course is planned by the coordinating nurse at
each participating center.
Follow-Up Period:
- Duration: Each patient will be monitored for a minimum of one year post-surgery.
- Activities: Follow-up includes regular clinical assessments during the
post-operative stay at the colorectal ward, monitoring for postoperative
complications, and collection of data on outcomes such as mortality, morbidity, and
quality of life.
- Objective: To evaluate the long-term effects of the perioperative treatment
strategies and compare the outcomes between the AI-augmented and surgeon-based risk
stratification groups.
Criteria for eligibility:
Criteria:
Inclusion Criteria:
To be eligible for inclusion, the patient must:
- Have histologically confirmed clinical and/or pathological evidence of suspected
first occurrence of clinical stage I to IV colorectal carcinoma.
- Be ≥ 18 years of age on the date of signing the informed consent.
- Provide written informed consent prior to registration according to the local
regulatory requirements.
- Have indication for elective curative intended surgery decided by the MDT-conference
- Available information about the covariates needed for the AIDPRO manual CRC device
to execute not assessed/directly evaluated by the surgeon (e.g. ASA-score and
WHO-performance score) o Blood test (must not be older than 30 days from the time of
inclusion): Carcinoembryonic Ag(CEA) [Mass/volume] in Serum or Plasma (microgram per
liter) C-reactive protein (CRP) [Mass/volume] in Serum or Plasma (milligram per
liter) Hemoglobin (Hgb) [Moles/volume] in Blood (millimole per liter) Bilirubin
[Moles/volume] in Serum or Plasma (micromole per liter) Albumin [Mass/volume] in
Serum or Plasma (gram per liter) Sodium (Na) [Moles/volume] in Serum or Plasma
(millimole per liter) eGFR [milliliters/minute] mL/min/1,73 m2 calculated from
Creatinin and height
Exclusion Criteria:
The patient may be excluded from participation in the trial if the patient:
- Have indication for elective curative intended surgery revised and canceled in the
preoperative phase.
- Has any serious or uncontrolled medical disorder that, in the opinion of the
investigator or treating physician, may increase the risk associated with study
participation, impair the ability of the subject to receive protocol intervention,
or interfere with the interpretation of study results.
- Has any language and/or mental diasability imparing the ability to actively
participate in perioperative interventions and the ability to answer relevant
questionaries.
Gender:
All
Minimum age:
18 Years
Maximum age:
N/A
Healthy volunteers:
No
Start date:
January 1, 2025
Completion date:
February 1, 2028
Lead sponsor:
Agency:
Zealand University Hospital
Agency class:
Other
Collaborator:
Agency:
Aalborg University Hospital
Agency class:
Other
Collaborator:
Agency:
Randers Regional Hospital
Agency class:
Other
Collaborator:
Agency:
Gødstrup Hospital
Agency class:
Other
Collaborator:
Agency:
Aabenraa Hospital
Agency class:
Other
Collaborator:
Agency:
Odense University Hospital
Agency class:
Other
Collaborator:
Agency:
Hvidovre University Hospital
Agency class:
Other
Collaborator:
Agency:
University Hospital Bispebjerg and Frederiksberg
Agency class:
Other
Collaborator:
Agency:
Viborg Regional Hospital
Agency class:
Other
Source:
Zealand University Hospital
Record processing date:
ClinicalTrials.gov processed this data on November 12, 2024
Source: ClinicalTrials.gov page:
https://clinicaltrials.gov/ct2/show/NCT06645015
http://www.aidpro-crc.dk
