Standardized uptake value of ¹⁸F-fluorodeoxyglucose positron emission tomography for prediction of tumor recurrence in breast cancer beyond tumor burden

Between April 2004 and May 2009, 1,053 women consecutively underwent surgery for breast cancer at our institution. Of these 1,053 patients, 835 underwent preoperative FDG-PET as a part of their routine preoperative staging. Patients were excluded on the basis of the following criteria: known bilateral breast cancer (n = 31), preoperative chemotherapy (because chemotherapy can affect tumor characteristics related to FDG uptake) (n = 94), ductal carcinoma in situ (n = 135) and distant metastases at initial assessment (n = 42). Among these patients, 501 women of interest were identified. Patients missing data for any immunohistochemical marker were excluded (n = 3). Patients with an immunohistochemistry (IHC) scores of 2+ for human epidermal growth factor receptor 2 (HER2), but without fluorescence in situ hybridization (FISH) results for HER2 amplification, were also excluded (n = 2). Data for the remaining 496 patients were entered into the analysis (Figure 1).

For the immunohistochemical study of four markers, formalin-fixed, paraffin-embedded tissue sections obtained from the surgical specimens were stained with appropriate antibodies for estrogen receptor (ER) (Novocastra; Leica Microsystems, Newcastle upon Tyne, UK), progesterone receptor (PR) (Novocastra; Leica Microsystems), HER2 (Ventana Medical Systems, Tucson, AZ, USA) and Ki-67 (MIB-1; Dako, Glostrup, Denmark). For HER2 evaluation, membranous staining was graded with a score of 0, 1+, 2+ or 3+ [13]. HER2 status was considered positive with a score of 3+ and negative with a score of 0 or 1+. Tumors with a score of 2+ were sent for FISH testing performed using the PathVysion HER-2 DNA Probe Kit (Abbott Molecular, Des Plaines, IL, USA).

The staging was performed according to the American Joint Committee on Cancer (AJCC) system [14]. The Elston-Ellis modification of the Scarff-Bloom-Richardson grading system was used for tumor grading. Adjuvant systemic therapy and/or radiotherapy were administered according to the standard guidelines based on patient age, primary tumor characteristics and axillary lymph node status. Endocrine therapy was given to patients whose tumors were positive for hormone receptor expression. The follow-up protocol included planned regular visits every 6 months, and missed appointments were followed by telephone calls to minimize the number of patients lost to follow-up and improve the accuracy of the survival data. The final update to the clinical database was made in December 2013.

The institutional review board (IRB) of Gangnam Severance Hospital, Yonsei University, Seoul, Korea, approved the study in accordance with Good Clinical Practice guidelines and the Declaration of Helsinki. The IRB granted a waiver of written documentation of informed consent from all participants because of the retrospective study design.

Prior to undergoing FDG-PET, each patient was asked to fast for a minimum of 8 hours, and blood glucose levels were controlled to <130 mg/dl. Patients received an intravenous injection of ¹⁸F-FDG (0.14 MBq) in the arm contralateral to the primary tumor. Sixty minutes after injection of ¹⁸F-FDG, whole-body emission scans were obtained using a Philips Allegro PET camera (Philips Medical Systems, Cleveland, OH, USA). Scans were obtained with the patient in the supine position with the arms raised. Attenuation-corrected transaxial images were reconstructed with an iterative transmission algorithm (row-action maximum likelihood three-dimensional protocol) using a three-dimensional image filter in a 128 × 128 matrix. For semiquantitative evaluations, maximum standardized uptake value (SUV_max) was calculated by measuring the ¹⁸F-FDG absorption by tumors in the region of interest (ROI) using the following equation: SUV_max = (maximal radioactivity concentration in the ROI (μCi/g)/injected dose (μCi)/patient’s weight (kg)). All FDG-PET scans were reviewed by two nuclear medicine radiologists who were blinded to survival data. SUV_max was obtained at the time of the imaging procedure.

The cutoff point of SUV_max was obtained by using the time-dependent receiver operating characteristic (ROC) curve. Age is presented in the study as median value with a range and was compared by the Mann-Whitney U test. Discrete variables were compared by performing a χ² test. The primary endpoint was recurrence-free survival (RFS), which was measured from the date of the first curative surgery to the date of the first tumor recurrence, including locoregional recurrence or distant metastasis or death. Breast cancer–specific survival (BCSS) was measured from the date of the first curative surgery to the date of the last follow-up or until death due to breast cancer during the follow-up period. The Kaplan-Meier method was utilized to estimate RFS or BCSS. Using Harrell c-statistics [15], the concordance index (c-index) was calculated to measure the concordance for time-to event data, in which increasing values between 0.5 and 1.0 indicated improved prediction. The significant prognostic factors associated with RFS were selected based on the c-index (Additional file 1). The Cox proportional hazards regression model was used for multivariable survival analysis. To assess the additional prognostic value of SUV_max, we used changes in the likelihood ratio values (LR − Δχ²) to quantitatively measure the relative amount of information for SUV_max compared to the model without SUV_max. The cutoff value of young age was defined as 35 years in accordance with a previous Korean study [16]. SPSS version 18 (SPSS, Chicago, IL, USA) and R [17] were used to perform these analyses. Statistical significance was defined by a P-value <0.05 or a 95% confidence interval (CI).

The cutoff point of SUV_max was obtained using the time-dependent ROC. The time-dependent ROC curve for SUV_max in relation to RFS yielded an area under the curve of 0.673 (95% CI, 0.588 to 0.753) (Additional file 2). Youden’s index was the highest for SUV_max of 4.2. Considering the clinical application, we defined the SUV_max cutoff as 4.

A total of 496 patients with breast cancer were included in the analysis. The median age of the cohort was 48 years (range, 25-80 years). The median and mean SUV_max were 4.3 ± 3.1 and 3.2 (range, 0.3-32.9), respectively. When patients were divided into two groups according to SUV_max, these groups differed significantly in T stage, N stage, AJCC stage, which represent tumor burden. They also differed in characteristics reflecting tumor biology, including histologic grade, ER, PR, HER2, and Ki67. In considering the distribution of tumor subtypes, the group with high SUV_max had a higher rate of luminal B, HER2, and triple-negative subtypes. In contrast, the proportion of patients with the luminal A subtype was relatively low in the group with high SUV_max (Table 1). A higher rate of mastectomy was noted in the group with high SUV_max (Table 1).

At a median follow-up of 6.03 years, tumors had recurred in 40 patients. There were 13 patients with locoregional recurrences, 25 with distant metastases and 2 with combined local recurrence and distant metastases. During the follow-up period, 11 deaths occurred, 8 of which were breast cancer–specific and 3 of which were not breast cancer–specific. The probability of RFS at 6 years was 95.6% for patients with low SUV_max and 86.8% for patients with high SUV_max. High SUV_max was significantly predictive of decreased RFS (P < 0.001 by log-rank test) (Figure 2A). Furthermore, patients with high SUV_max showed a reduced BCSS (P = 0.007 by log-rank test) (Figure 2B). When adjusted for age of diagnosis, T stage, nodal status and ER status using the Cox proportional hazards regression model, high SUV_max was significantly associated with risk of tumor relapse (hazard ratio, 2.39, 95% CI, 1.20 to 4.76) (Table 2). For this model, the Harrell c-index was 0.745. The c-index for the multivariate model without SUV_max was 0.724. The LR-Δχ² showed a significant improvement of the additional prognostic utility of SUV_max.

Four patient groups were classified according to SUV_max and tumor size: (1) tumor size ≤2 cm and SUV_max <4; (2) tumor size >2 cm and SUV_max <4; (3) tumor size ≤2 cm and SUV_max ≥4; and (4) tumor size >2 cm and SUV_max ≥4. The RFS of the four groups differed significantly (P < 0.001) (Figure 3A). Within the groups of large tumor size (>2 cm) and small tumor size (≤2 cm), RFS differed significantly according to the SUV_max (P = 0.049 and P = 0.009, respectively). Conversely, within the groups of high SUV_max and low SUV_max, RFS did not differ according to tumor size (P = 0.350 and P = 0.096, respectively).

Furthermore, SUV_max was significantly predictive of RFS in combination with nodal status (P < 0.001) (Figure 3B). Node-positive patients with high SUV_max had worse outcomes, whereas node-negative patients with low SUV_max had better outcomes. Similarly, SUV_max combined with stage was significantly correlated with RFS (P = 0.001) (Figure 3C).

After the patients were divided into three subtypes (luminal, HER2 or triple-negative), multivariate analysis for RFS was performed in each subtype. In luminal subtypes, which were defined as hormone receptor–positive breast cancer (ER-positive and/or PR-positive), SUV_max was found to be a significant prognostic factor for RFS based on multivariate analysis (Table 3). However, in HER2 or triple-negative subtypes, SUV_max was not an independent prognostic factor (Additional file 3).

The prognostic value of SUV_max combined with tumor burden was also assessed in hormone receptor–positive breast cancer. When the patients were classified into four groups according to both combined factors, RFS differed significantly (P < 0.001) (Figure 4A). There was no difference in RFS when patients were stratified by tumor size within the groups with high SUV_max and low SUV_max (P = 0.950 and P = 0.688, respectively). However, within the groups with small tumor sizes (≤2 cm), a significantly reduced RFS was found in patients with high SUV_max (P = 0.044). In patients with large tumor sizes (>2 cm), RFS did not differ significantly according to SUV_max (P = 0.065), possibly due to the limited number of patients (n = 122).

In luminal breast cancer, SUV_max was still predictive of RFS in combination with nodal status (negative vs. positive) and stage (I vs. II and III) (P < 0.006 and P = 0.029, respectively) (Figure 4B and 4C).