The current study demonstrates that pre-operative MRI and MRSI measurements are associated with time-to-BCR after radical prostatectomy. The MRI risk score was positively correlated with the probability of BCR and was an independent predictor of BCR in a model of recurrence which included standard pre-operative clinical data. This agrees with the results of an MRI-only study of a larger population conducted at our institution of which the current population was a subset 13
(see Methods). Very few studies to date have attempted to correlate endorectal MRI data with BCR after radical prostatectomy. D’Amico et. al. found that after controlling for clinical T stage, PSA, biopsy Gleason score and percentage of biopsy cores positive, MRI stage was prognostic in patients who were considered “intermediate risk” but not in low risk or high risk patients 16
The number of voxels in the MRSI index lesion as a continuous variable and the presence of high grade voxels were significantly associated with BCR in a multivariate model including pre-operative clinical parameters. This indicates that not only the size, but the degree of metabolic abnormality in the MRSI index lesion is related to the risk of BCR. The predictive utility of the MRSI parameters studied here may be related to their ability to stand as surrogates for pathological parameters determined at surgery such as Gleason score and tumor volume. Multiple studies have linked one or both of these parameters to the risk of biochemical failure 4, 17–20
. We previously showed in the same population a positive correlation between the metabolic abnormality as represented by [Cho+Cr]/Cit and the pathological lesion Gleason score 9
. In the current study, the number and percentage of high grade voxels in the MRSI index lesion were correlated with BCR. Our results strikingly reflect the results of a pathological study by Stamey and colleagues which found that the volume of the largest tumor and the percentage of high grade (Gleason 4/5) cancer in the tumor were independently associated with BCR in 379 patients with peripheral zone cancer 18
. Others have also reported that the amount or percentage of high grade cancer in either the biopsy specimen 21–22
or the radical prostatectomy specimen 21–23
correlated with BCR. In addition, Stamey, et. al. also noted that patients with relatively high percentages of high grade disease who did not fail had much smaller tumors than those who failed. This is similar to our result that patients with smaller index lesions (≤ 0.5 cm3
) with at least one high grade voxel fared better than patients with larger lesions and at least one high grade voxel.
One might expect the presence of high grade MRSI voxels to correlate with increased biopsy Gleason grade. However, in this study, the presence of one or more high grade voxels remained significant in the model adjusted for PSA, clinical stage and biopsy Gleason score. This suggests that the MRSI grade information is independent of the biopsy Gleason score. One factor which may contribute to this is that due to sampling limitations, the biopsy Gleason score may not fully reflect the aggressiveness of the tumor. Others have shown that if the length or percentage of high grade cancer in the biopsy specimen is included in a multivariate pre-operative prognostic model, the biopsy Gleason score becomes insignificant 21
In our multivariate model which included both the number of voxels in the index lesion and presence/absence of HG voxels, clinical stage became insignificant (p = 0.15)as a predictor of BCR. It is possible that the combined information provided by MRSI about tumor volume and degree of metabolic abnormality supplants the extent of disease given by the clinical stage. In addition, there were no clinical T3 patients which could have given clinical stage more impact. The MRI index, which also indicates local extent of disease, was also in significant when either MRSI variable was included. However, these results should be interpreted with caution due to the limited patient numbers in this study.
A limited number of studies have assessed endorectal MRSI as a prognostic indicator in prostate cancer. Yu, et. al. found that within a given lobe of the prostate, the number of suspicious MRSI voxels per cross-section correlated with the chance of pathologic ECE 24
. Since ECE is a predictor of poor outcome in prostate cancer 25–26
, this study tends to support the relationship between the number of index lesion voxels and BCR in the current study. The same group found the strongest predictor of BCR after radiation therapy to be MRSI tumor volume 27
; however, the degree of MRSI metabolic abnormality was not reported. The finding with respect to MRSI tumor volume is similar to ours, suggesting MRSI may be prognostic for multiple treatment modalities. Finally, in a study of 92 men who selected active surveillance, the finding of in apparent tumor on MRI/MRSI did not predict which patients had eventually recurred 28
The results of our study suggest that we can identify a population with a high likelihood of recurrence after radical prostatectomy based on high MRSI tumor volume and grade. This finding should be verified in a larger population. With such data, we could also determine whether baseline MRI/MRSI data gives incremental value to the pre-operative nomogram. Potentially, these high risk patients could be considered for inclusion in clinical trials investigating men with high risk but clinically localized prostate cancer.
The current study had some limitations. Information on the percentage of positive cores in the biopsy was not available for all patients and thus was not included in the model. MRI tumor volume was not assessed because the accuracy of MRI for measuring prostate cancer volume is limited 29
. Patients with tumors only in the transition zone, as identified at surgical pathology, were not included (see Methods). It is well-known that biochemical failure does not necessarily reflect clinical failure and further followup on these patients would be valuable. Finally, due to the limited number of subjects, this study should be considered a preliminary analysis which requires validation in larger populations.
We did not include post-surgery pathological data in the multivariate model of recurrence. The goal of the current study was to test the hypothesis that baseline MRI/MRSI data correlated with biochemical recurrence and thus could have value in the pre-operative setting. We believe this value comes from acting as a non-invasive surrogate for tumor volume and aggressiveness.