BACKGROUND
Colorectal cancer (CRC) is the third most commonly diagnosed cancer in Canada and the
second leading cause of death in both men, and women (1). In 2021, 24800 Canadians were
diagnosed with CRC and 9,600 died from the disease (1). Over their lifetime, 1 in 18
Canadians will be diagnosed with CRC and 1 in 37 will die (1). Accurate staging is
essential to improving outcomes, providing appropriate patient management, and improving
the health care costs associated with caring for patients with CRC.
London Health Sciences Centre (LHSC) is a tertiary care referral centre for a catchment
area of 2 million people in Southwestern Ontario. Annually, approximately 200 patients
present to the London Regional Cancer Program with a diagnosis of colorectal cancer. Of
these, about 100 patients will have potentially resectable colorectal liver metastasis
(CRCLM).
Staging algorithms for CRC include contrast enhanced computed tomography (CECT) of the
thorax/abdomen/pelvis, with MRI of the liver in some centres. The objective for
performing imaging tests is to accurately determine the extent of local and distant
disease to direct patient management. Accurate assessment of the hepatic disease burden
is crucial for surgical planning since resection of liver metastases is a core component
of CRCLM treatment (2). At LHSC, all patients are initially imaged with CECT of the
thorax/abdomen/pelvis. MRI of the liver is reserved for patients that require further
characterization of equivocal liver lesions detected on CT. When performed, liver MRI is
often performed with extracellular agents such as gadobutrol (Gadovist), i.e. EC-MRI.
Hepatobiliary MRI contrast agents such as gadoxetic acid (aka gadoxetate, trade name
Primovist in Canada), i.e. EOB-MRI, provide superior accuracy in detection of CRCLM
compared to both CECT (3) and EC-MRI (4). Moreover, the use of EOB-MRI can alter
management decisions and improve patient outcomes (3,5,6). It is also the modality of
choice in CRCLM patients post-systemic therapy as per the 9th International Forum for
Liver MRI Consensus Report (7).
Despite these data, hepatobiliary agents are being used sparingly in most Canadian
hospitals, including at LHSC as a problem-solving tool. This is due to two factors: (a)
the higher unit cost of gadoxetate compared to gadobutrol and iodine-based CT contrast
agents, and (b) the increased MRI scan time required for EOB-MRI compared to EC-MRI or
CECT. The increased scan time is a result of the need to acquire images in the
"hepatobiliary (HPB) phase" for EOB-MRI, typically 20 minutes post-injection, a longer
delay than is required for EC-MRI or CECT. These factors result in increased operational
costs for EOB-MRI and opportunity costs from reduced magnet time for other MRI studies.
To address the increased scan time with EOB-MRI, some studies have retrospectively
examined the potential role of abbreviated MRI protocols (aMRI) compared to a full
protocol (fMRI) (8-11). The premise of EOB-aMRI protocols involves an injection of
gadoxetate at the outset of the study, often outside the scanner room. During the 20 min
waiting period prior to image acquisition in the HPB phase, an "abbreviated" set of
sequences is acquired, usually including DWI/ADC and sometimes T2 weighted images. At the
20 min mark, the HPB phase images are acquired, and the study is complete. The aim of
abbreviated protocols is to increase patient throughput without compromising diagnostic
accuracy.
The initial results in this relatively nascent field are promising, showing high
interobserver agreement and high diagnostic accuracy not significantly different from the
full protocol. For example, Canellas et al reported both κ and area under the ROC curve
(AUC) of greater than 0.9 for both aMRI and fMRI, with an estimated cost savings of 41%
per scan (10). Ghorra et al found similar detection rates of about 86% for both aMRI and
fMRI with slightly lower accuracy of the aMRI protocol of about 87% vs 93% for fMRI, but
no consistent statistical trends were present (11).
However, existing studies in the literature have simulated an aMRI examination by using a
subset of fMRI sequences; some sequences, including the dynamic post contrast sequences
acquired before 20 min are removed retrospectively (8-11). Currently there are no
published studies comparing fMRI with prospectively acquired aMRI. As retrospective
studies may overestimate accuracy and cost savings, there is a need for higher quality,
prospective evidence (7). Additionally, retrospective studies are unable to perform a
formal economic analysis of costs related to the imaging procedure itself, and
importantly downstream costs related to patient management.
RATIONALE
The primary aim of this study is to prospectively compare the diagnostic accuracy of aMRI
compared to fMRI regarding CRCLM, using a composite reference standard. Our hypothesis is
that aMRI is noninferior to fMRI in this regard, as measured by sensitivity, specificity,
and the AUC. If this is the case, it may serve as evidence that EOB-MRI utilization can
be increased even within resource constraints inherent to all Healthcare systems. The
rationale for using a composite reference standard is that due to varying patient
management strategies, the optimal reference standard (surgical pathology) is not always
available, and therefore alternative methods must be considered. The rationale for using
fMRI as the control group is that this protocol is the current standard of care for
EOB-MRI.
A secondary aim is to quantify the economic impact of aMRI vs fMRI both in terms of
imaging costs and downstream patient management costs. Our hypothesis is that aMRI will
not cost more than fMRI on a per patient basis (i.e. noninferiority). If this is the
case, higher patient throughput can be achieved at no increased economic expense.
A second secondary aim is to evaluate patient outcomes (overall survival, cancer-specific
survival, and hepatic recurrence / progression free survival) at 1-year post-baseline
EOB-MRI, using clinical data and the 1-year follow-up CECT. Our hypothesis is that aMRI
will be noninferior to fMRI, indicating that there is no adverse effect on patient
outcomes from the using an abbreviated protocol.
The third secondary aim is to retrospectively compare the diagnostic accuracy of fMRI to
a simulated aMRI created from a subset of fMRI pulse sequences. Our hypothesis is that
the simulated aMRI will be noninferior to fMRI. This constitutes a 3-factor multireader
multicase design, analogous to multiple prior investigations (3,4), enabling direct
comparison of our study and adding to the body of literature on the subject.
The final study aim is to compare the diagnostic accuracy and interobserver agreement on
aMRI and fMRI. Our hypothesis is that there will be no significant difference for
diagnostic accuracy. We expect interobserver agreement to be moderate to high.
The rationale for choosing a study cohort comprised of patients with CRCLM is: 1) this is
a large patient population / common patient presentation, and 2) EOB-MRI has been shown
to provide added value for staging CRCLM but is likely underutilized in Canada, as
detailed above.
The rationale for choosing a 1-year follow-up period is that about 30% to 50% of CRCLM
will recur or progress within this interval (12,13), enabling a compromise between
capturing a significant portion of adverse patient outcomes while minimizing loss to
follow-up and unnecessarily prolonging the study, as this is not the primary objective.
STUDY DESIGN
This is a prospective, block randomized, allocation concealed, unblinded, multireader
study with case-nested-within-test split-plot design.
The baseline abbreviated or full Primovist MRI will be acquired between day 2 and 14 and
a follow-up contrast enhanced CT abdomen pelvis will be performed 1 year from baseline. A
combination of histopathology, biological behavior, and imaging findings applied in a
hierarchical manner will determine the reference standard for each focal hepatic lesion,
i.e. metastasis or not. Sample size is 200 subjects, with equal distribution of 100 per
arm.
Statistical analysis of the primary endpoint will be conducted via the updated
Obuchowski-Rockette (OR) method (14).
