Between November 1, 2000, and November 1, 2006, 70 patients with primary brain tumors were enrolled onto a prospective study of early tumor response. Sixty-seven of these patients had WHO grade 3 or IV astrocytoma, and 60 had assessable results form the population for this study. Seven patients were excluded for the following reasons: three had repeat surgical procedures within one month, one had claustrophobia, and three declined treatment. Pretreatment scans were performed 6 (± 3.9) days from start of treatment, 54 patients had a scan at 1 week (6 ± 2.8 days), all 60 at 3 weeks (21 ± 5.6 days), and 55 at 10 weeks (71 ± 14.2 days). Median survival is 13.9 months, and at last contact, 30% of patients (18 of 60) were still alive with a median follow-up of 23.1 months.
Evaluation of Response Measures
A total of five metrics were measured at the three time points, including the percentage of change in tumor volume, the percentage of change in mean tumor ADC, and three submetrics of fDM (increasing ADC [fDM-VI], decreasing ADC [fDM-VD], or any change in ADC [fDM-VT]. Comparisons with standard RR were limited to the 55 patients with scans at 10 weeks. ROC curve analysis was performed to predict patient survival 1 year from diagnosis (34 of 55 patients were alive 1 year from diagnosis, and 21 if 55 died; ). All but two deaths were secondary to tumor progression.
Pretreatment and Treatment-Related Patient Characteristics
Change in Tumor Volume
There were modest changes in tumor volume; median at 1, 3, and 10 weeks, respectively, was +0.2% (interquartile range [IQR], –19.4 to +19.4), +2.0% (IQR, –30.0 to 3.5), and +0.3% (IQR, –32.2 to +56.6). With smaller increases in volume at 10 weeks for those who were alive at 1 year compared with those who died (median, −0.1% [IQR, –38.2 to +41.7] v +46.9% [IQR, –17.5 to 122.2]; P < .09). By ROC curve analysis () the change in tumor volume at each time point exhibited a trend toward predicting patient survival at 1 year but did not reach statistical significance (P < .09 at each time). When tumor response at 10 weeks was stratified by Macdonald criteria, this increased the predictive value (). No patient had CR, three had PR, 27 had SD, and 25 had PD. The presence of SD or PR at 10 weeks was the best volume-based correlate with survival at 1 year (P < .04).
Changes in Mean ADC
The changes in mean tumor ADC at 1, 3, and 10 weeks were, respectively, +0.4% (IQR, –4.5 to +5.6), +2.9% (IQR, –3.1 to +6.9), and +10.3% (IQR, –1.6 to +21.6). Three-week mean ADC was associated with 1-year survival, with those alive exhibiting increased ADC (median, +3.4% [IQR, –2.0 to + 12.0]) compared with a decreased ADC (median, −1.5% [IQR, –6.9 to +0.9]) in those who died (P < .03). By ROC curve analysis (), the change in mean tumor ADC at 3 weeks was associated with 1-year survival (P < .02) but the change at 1 and 10 weeks was not (P > .1). After correcting for multiple comparisons, even the 3-week metric was of only of borderline significance.
Changes in fDM
When regional tumor diffusion data were analyzed by fDM (), the percentage of tumor with increasing diffusion (fDM-VI) over time was associated with survival 1 year from diagnosis. VI increased linearly over time, with median increases of 1.6% (IQR, 0.4 to 4.3), 4.0% (IQR, 1.0 to 7.5), and 12.2% (IQR, 3.8 to 27.5) at 1, 3, and 10 weeks, respectively, with greater increases in VI for those alive at 1 year compared with those who had died (Cochran-Armitage P < .001; Appendix Fig A1, online only). To assess fDM as an early biomarker we focused on VI at 3 weeks. By ROC curve analysis the strongest relationship between any of the imaging metrics, and survival at 1 year was observed for VI at 3 weeks (P < .0002; ; Appendix Fig A2, online only).
both increasing and decreasing fDM at 3 weeks was correlated with RR at 10 weeks. In the present analysis, however, no correlation was found between patient survival at 1 year and decreasing diffusion by fDM (P
> .1 at 1, 3, and 10 weeks; Fig A1). Adding VD
(to yield VT
) was, therefore, associated with a lower predictive value for survival, and all analysis focused on fDM-VI
at 3 weeks.
Optimization of fDM-VI
When assessed as a continuous variable, increasing VI at 3 weeks was correlated with increasing OS (P < .02). Given the continuous nature of VI, ROC curve analysis suggested a threshold of 4.7%, where VI 4.7% or greater at 3 weeks was stratified as response and VI less than 4.7% as nonresponse. After leave-one-out cross-validation, VI remained a significant predictor of patient survival at 1 year (P < .001; AUC = 0.723; sensitivity = 69.7% [95% CI, 51.3 to 84.4]; specificity = 75.0% [95% CI, 50.9 to 91.2]; positive predictive value [PPV] = 82.1%; negative predictive value [NPV] = 60.0%).
Overall Survival As a Function of fDM Stratification and RR
Using the VI threshold of 4.7%, those with higher VI had median survival 52.6 months whereas those with lower VI had median survival of only 10.9 months (P < .003; HR = 2.7; 95% CI, 1.5 to 5.9; ). Conventional RR at 10 weeks was similarly prognostic (). Those with SD/PR had median survival of 31.6 months, whereas those with PD had median survival of 10.9 months (P < .0007; HR = 2.9; 95% CI, 1.7 to 7.2). For comparison with RR, fDM was limited to the 55 patients who had RR at 10 weeks, but if this analysis is extended to include all 60 patients, fDM was similarly prognostic (P < .005; HR = 2.4; 95% CI, 1.4 to 4.8).
Fig 2. Overall survival as a function of functional diffusion map (fDM), radiologic response (RR), and their composite. (A) Overall survival by log-rank test based on fDM stratification at 3 weeks from the start of treatment where the yellow curve (n = 27) represents (more ...)
There was an association between RR and fDM stratification (P < .001; ) with concurrence in 75% of cases (41 of 55). presents two patients in whom fDM and RR differed in their stratification of response. The patient on the left was classified as PD by RR, but in contrast fDM documented a VI of 26.4% at 3 weeks (middle panel), and the patient was classified as responding by fDM. Despite PD, this patient clinically stabilized and is alive without progression at 33 months. In contrast, the patient on the right had SD by RR, but had minimal change in tumor ADC at 3 weeks by fDM (middle panel, 1.6%), clinically progressed within 5 months, and died at 7 months.
Fig 3. Correlation of functional diffusion map (fDM) mediated evaluation of response with radiologic response (RR). There was a strong correlation between increasing fDM-VI at 3 weeks and subsequent RR at 10 weeks going from worst to best (progressive disease (more ...)
Given the differences between conventional RR and fDM in 25% of patients, a composite index of response was developed based on fDM and RR, and was the most robust response-based model for OS (P < .0002; ). The composite identified three groups of patients. Those with the best prognosis were without radiographic progression (SD/PR) and responsive by fDM, and had a median survival of 52.6 months. Those with the worst prognosis had low VI by fDM and PD by RR, and their median survival was 8.1 months. The intermediate group, comprising patients in whom fDM and RR differed, had a median survival of 14.4 months. Both the intermediate group (P < .02; HR = 2.4; 95% CI, 1.2 to 4.6) and the best-prognosis group (P < .0001; HR = 4.2; 95% CI, 2.4 to 12.9) were distinct from the worst-prognosis composite group.
Evaluation of Other Prognostic Variables
To further assess the utility of fDM, we evaluated common variables previously found to correlate with survival in high-grade glioma (). Those disproportionately represented in the fDM responding group were younger age (P < .03), higher baseline tumor ADC (P < .005), and increased frequency of surgical resection (P < .002). On univariate analysis, only age (< 50 v ≥ 50 years; P < .006) and pathologic grade (WHO grade 3 v 4; P < .05) correlated with OS. Performance status, surgical resection, pretreatment tumor ADC, use of chemotherapy, and radiation dose did not (). If limited to the patients with grade 4 tumors treated with definitive radiation therapy (≥ 60 Gy; n = 41) there was prolonged survival associated with concurrent temozolomide and radiation compared with radiation alone (P < .05). When these individual variables were included in a multivariate model, only age and fDM were retained (, model 1).
Variables Associated With Patient Survival
The best predictor of OS was the Radiation Therapy Oncology Group (RTOG) recursive partition analysis44
(RPA; ; P
< .0004). When fDM was added to the RPA, both retained prognostic value (, model 2). Interestingly, across the five categories (there were no patients in class 2) there was an inverse relationship between class and the likelihood of response by fDM: 75%, 73%, 50%, 33%, and 25%, for classes 1, 3, 4, 5, and 6, respectively (P
< .01; Table A1, online only). For each class, median survival was longer in the group responding by fDM than in those not responding. Thus, although the numbers were small, it does appear that fDM retained prognostic value across the whole spectrum of disease. Patients predicted to have a worse outcome by RPA were also less likely to be responsive to therapy even as early as 3 weeks into treatment.