Clinically feasible pharmacokinetic model of dynamic contrast-enhanced MRI is prognostic of recurrence free survival in breast cancer patients treated with neoadjuvant therapy

Background: Dynamic contrast-enhanced (DCE) MRI has long been used in the diagnosis and treatment planning of cancers. Chief among its use is the identification of tumors in soft tissues, where the tumor progression-induced angiogenic changes result in leaky vasculature which is readily highlighted by contrast enhancement. While allowing tumor morphology assessment, DCE MRI also contains rich information on perfusion and permeability which can be modeled as a pharmacokinetic (PK) process, a subject of extensive investigation in the clinical research setting. Several breast DCE MRI studies have suggested that PK parameters might be indicative of response to therapy and recurrence. However, these studies rely on high temporal resolution DCE MRI for accurate PK modeling, which presents a hurdle to adoption in clinical settings where DCE MRI is acquired with high spatial but low temporal resolutions.

Methods: We present a novel two-parameter modeling framework which circumvents this temporal limitation, allowing kinetic parameters to be extracted from clinical DCE MRI containing 3-6 timepoints spaced 60-90 seconds apart. We sought to examine whether these kinetic parameters (hereafter KP1 and KP2) binned into “high” and “low” values had prognostic value in determining disease recurrence. KP1 reflects the rate of leakiness from vessels into extravascular space, while KP2 reflects the rate of the reverse process. A study was conducted with a cohort of 458 breast cancer patients treated with neoadjuvant therapy (NAT) in which 117 recurrences were observed. Model parameters were extracted from pre-treatment clinical DCE MRI with low temporal resolution. Univariate and multivariate (Cox Proportional Hazard; CPH) analyses were
performed to test the prognostic value covariate in predicting recurrence free survival (RFS).

Results: A CPH model was constructed that included age, receptor status, cancer stage, kinetic parameters, and pathologic complete response (pCR, ypT0/TisN0) as covariates (Table 1). KP2 and pCR were found to be statistically significantly prognostic markers of RFS (p≤0.05). KP1 was found to be a nearly significantly prognostic of RFS (p=0.055). Interestingly, KP2 had a hazard ratio (HR) similar to pCR (HR: -1.04 [-1.75, -0.33]; p<0.0039 vs. HR: -0.97 [-1.5, -0.44], p<0.00034). Using univariate analysis, we investigated whether KP2 was prognostic of recurrence. Patients were stratified by therapy type, whether they achieved pCR or had residual disease, and by disease subtype (as determined by receptor status). KP2 was found to be predictive of RFS for patients with residual disease after treatment with NAT chemotherapy (p≤0.0004, n=241, log- rank test), but not for patients treated with anti-HER2 NAT (p=0.84, n=124) or patients that achieved pCR (p=0.68, n=139). KP2 was also found to be prognostic in Hormone Receptor+/HER2- patients with residual disease (p≤0.012, n=151) and Hormone Receptor-/HER2+ patients treated with NAT chemotherapy (p≤0.001, n=64). The parameter was not found to be prognostic in patients with triple-negative disease.

Conclusions: We developed a new prognostic imaging marker in breast cancer patients treated with NAT that relies solely on pre-treatment, standard-of-care, low temporal resolution DCE MRI. This marker can identify a subset of patients who will have a good outcome regardless of whether or not they achieve a pCR with NAT, and can therefore be used to determine whether or not there is a need for adjuvant chemotherapy in this subgroup of patients