Introduction

Ovarian cancer is the most common cause of cancer-related deaths in women with approximately 22,000 newly diagnosed cases annually [1]. More than 70% of patients are diagnosed at advanced stages and as a result, five-year survival rates are less than 50% [2]. Debulking surgery remains the standard-of-care and optimal debulking with residual tumors less than 1 cm is the goal in patients with advanced disease [3]. Tumor burden is a prognostic factor itself as well as predictors of optimal cytoreduction and response to chemotherapy [4,5,6].

Due to its unique properties associated with glucose metabolism, 18F-fluorodeoxyglucose (FDG) positron emission tomography/computed tomography (PET/CT) has been successfully used for diagnosis, staging, re-staging, response assessment, and prognosis evaluation in ovarian cancer [2]. 18F-FDG uptake, a marker of tumor glucose metabolism, is usually measured semi-quantitatively using maximum standardized uptake value (SUVmax) defined as the highest single voxel value of radiotracer uptake [7]. Unlike SUVmax which is susceptible to noise and cannot reflect the tumor in its entirety, volume-based parameters such as metabolic tumor volume (MTV) and total lesion glycolysis (TLG) have potential to represent metabolic activity in the whole tumor [8]. Several meta-analyses have shown that MTV and TLG are significant prognostic factors in various malignancies [9, 10]. In the past few years, studies on the prognostic value of volume-based parameters in ovarian cancer have been published; however, due to small study population in individual studies, their prognostic value is not well established. Hence, we performed a systematic review and meta-analysis to evaluate the prognostic value of volume-based metabolic parameters of 18F-FDG PET/CT in patients with ovarian cancer.

Materials and methods

This systematic review and meta-analysis was performed in accordance with Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines [11]. The protocol was registered to the International Prospective Register of Systematic Reviews (PROSPERO) (registration no. CRD42018088589). The research question for this meta-analysis was as follows: “What is the prognostic value of volume-based metabolic parameters of 18F-FDG PET/CT in patients with ovarian cancer?”

Literature search

A systematic search on Pubmed and EMBASE databases was done for articles published up to Feburary 12, 2018. Search queries included synonyms and related terms for “ovarian cancer”, “18F-FDG-PET/CT”, “volume-based metabolic parameters” and “prognosis” as the following: (ovary OR ovarian) AND (PET OR “positron emission tomography” OR fluorodeoxyglucose OR FDG) AND (“Metabolic tumor volume” OR MTV OR “total lesion glycolysis” OR TLG) AND (Prognos* OR predict* OR survival OR outcome). No language restriction was applied. The bibliographies of identified studies were checked to find additional relevant papers.

The inclusion criteria was based on the patient/intervention/comparator/outcome/study design (PICOS) criteria (12): (1) “patients” with ovarian cancer; (2) volume-based metabolic parameters on 18F-FDG-PET/CT as “intervention”; (3) no “comparator” relevant to this study; (4) progression-free survival (PFS) and overall survival (OS) as “outcome”; and (5) “study design” as original articles. The following exclusion criteria were applied: (1) patients less than 10; (2) non-original articles; (3) not in field of interest (i.e., ovarian tumor characterization by means of 18F-FDG-PET/CT or articles which performed Cox proportional hazards regression incorporating MTV and TLG as a continuous variable); (4) insufficient survival information; and (5) overlapping study population. The literature search and selection was performed independently by two reviewers. Discrepancies were resolved via consensus.

Data extraction and quality assessment

The following study, clinicopathological, and 18F-FDG-PET/CT characteristics were extracted with a standardized form: (1) study: first author, publication year, institution, patient enrollment period, number of patients, and design (prospective/multicenter/consecutive enrollment); (2) clinicopathological: age, FIGO stages, histology, serum cancer antigen (CA)-125 level, clinical setting (pretreatment, post-operative, or recurrence), endpoint (PFS or OS), and follow-up; and (3) 18F-FDG-PET/CT: vendor, model, injected dose, injection-to-imaging time, acquisition time, fasting duration, blood glucose level, normalization of SUV, reconstruction methods, matrix size, attenuation correction, assessed volume-based parameters, volume-of-interest (VOI) (location and delineation threshold), cut-off values of MTV/TLG, and method for determining cut-off values.

The methodologic quality of included studies were assessed using a quality scale in previous meta-analyses evaluating prognostic value of volume-based metabolic parameters. Each reviewer rated a score of 0–2 for 19 items in the four domains―scientific design, generalizability, analysis of results, and PET reports [9, 10]. The overall quality scores were presented as a percentage of the maximum achievable score. Two independent reviewers performed data extraction and quality assessment, with any disagreement resolved by discussion.

Statistical analysis

Primary and secondary outcomes were PFS and OS, respectively. Events in PFS included any recurrence, progression, or death. Hazard ratios (HRs) were used as the effect size for prognostic value of MTV and TLG. Univariate HRs and their 95% confidence intervals (CIs) were extracted. If these were not provided, other available information including raw data showing number of events and follow-up duration, or p values from log-rank test were used to calculate HR indirectly [12]. If more than one HR were provided within a single study due to multiple cut-off values, we used the results showing the best prognostic value. When Kaplan–Meier curves were presented, HRs and their standard error were extracted via Engauge Digitizer version 3.0 (http://digitizer.sourceforge.net) and suggested methodology by Tierney et al. [13].

The HRs were meta-analytically pooled using the random-effects model. Heterogeneity was evaluated with Higgins I2 [14]. Funnel plots and Egger’s test were used to assess publication bias [15]. Statistical analyses were done with Review Manager (The Cochrane Collaboration, version 5.3.5) and “metafor” package in R (R Foundation for Statistical Computing, version 3.4.3). Subgroup analyses were planned for studies stratified by clinical setting, determination method of cut-off value, and delineation of VOI. Meta-regression was done to compare the pooled HRs of MTV, TLG, and SUVmax. P value of less than 0.05 was considered statistically significant.

Results

Study characteristics

The study selection process is described in Fig. 1. Initial literature search retrieved 47 articles. After the removal of 16 duplicates and screening of the remaining 31 titles and abstracts, 12 articles were considered potentially eligible. After full-text review, four articles were excluded as following: not in field of interest (tumor characterization [n = 2] [16, 17], and MTV/TLG as continuous variable [n = 1] [18]), or insufficient survival data (n = 1) [19]. Thus, eight original articles with 473 patients were included in this meta-analysis [20,21,22,23,24,25,26,27].

Fig. 1
figure 1

Flow diagram showing study selection process

Full size image

The study and clinicopathological and PET characteristics are summarized in Tables 1 and 2, respectively. All included studies were retrospectively designed. 18F-FDG PET was performed at initial staging in three [20, 23, 27], post-operatively in two [21, 24], and at the time of recurrence in three studies [22, 25, 26]. In all studies, prognostic values of both MTV and TLG were evaluated and VOI was placed on tumors in whole body. For the threshold of VOI, a certain percentage of SUVmax (40%, 42%, or 50%) were used in six studies [20, 21, 23, 25,26,27]. The cut-off values dichotomizing into high or low MTV/TLG were from receiver-operating characteristics (ROCs) analyses in six [20,21,22,23,24, 27], and median values in two studies [25, 26]. The quality of the studies was good in general (mean 73.0%; range 65.8–84.2%). There was no publication bias upon visual assessment of the funnel plot (Fig. 2) and based on the Egger’s test (p = 0.4626 and 0.1421 for MTV and TLG, respectively).

Table 1 Study and clinicopathological characteristics of studies included in meta-analysis
Full size table
Table 2 PET/CT characteristics
Full size table
Fig. 2
figure 2

Funnel plots and Egger’s test shows that no significant publication bias was present in studies assessing MTV (a) (p = 0.4626) and TLG (b) (p = 0.1421) for disease-free survival

Full size image

Progression-free survival

Five studies evaluated HRs of MTV and TLG for PFS [20, 23, 25,26,27]. Forest plots are presented in Fig. 3. The pooled HR of MTV for PFS was 2.50 (95% CI 1.79–3.48; p < 0.00001). No heterogeneity was present (I2 = 0%). Overall pooled HR of TLG for PFS was 2.42 (95% CI 1.61–3.65; p < 0.0001) with no heterogeneity (I2 = 0%). Results of subgroup analyses based on clinical setting and cut-off determination method consistently showed that MTV and TLG were significant prognostic factors with pooled HRs ranging from 1.73 to 3.35 (Table 3).

Fig. 3
figure 3

Forest plots for hazard ratios of disease-free survival using MTV (a) and TLG (b)

Full size image
Table 3 Subgroup analyses for PFS and OS of volume-based parameters using covariates
Full size table

Overall survival

Three studies evaluated OS using MTV and TLG [21,22,23]. Overall pooled HR of MTV for OS was 8.06 (95% CI 4.32–15.05; p < 0.00001) (Fig. 4). No heterogeneity was present (I2 = 0%). The pooled HR of TLG for OS was 7.23 (95% CI 3.38–15.50; p < 0.00001) without significant heterogeneity (I2 = 26%). Subgroup analyses based on clinical setting and VOI delineation method were performed. Each subgroup with the exception of a study [21] using 42% of SUVmax for VOI (HR of 3.20; 95% CI 0.95–10.79; p = 0.06) showed significant pooled HRs ranging from 4.80 to 9.93.

Fig. 4
figure 4

Forest plots for hazard ratio of overall survival using MTV (a) and TLG (b)

Full size image

Combined SUVmax data

Data from survival analysis for SUVmax were extracted from four studies for PFS and three studies for OS. The SUVmax was significant prognostic factor for PFS with pooled HR of 1.45 (95% CI 1.16–1.82; p = 0.001) (Table 4). No heterogeneity was present (I2 = 0%). The pooled HR for OS was 2.93 (95% CI 0.80–10.74; p = 0.11) with significant heterogeneity (I2 = 74%). The pooled HRs of MTV and TLG for PFS and OS were higher than those of SUVmax, but without statistical significance (p = 0.1014 and 0.1418 for MTV and TLG, respectively).

Table 4 Pooled hazard ratios for PFS and PS of 18F-FDG PET parameters
Full size table

Discussion

In the present meta-analysis, we evaluated the prognostic value of volume-based metabolic parameters of 18F-FDG PET/CT in patients with ovarian cancer in terms of the HRs for PFS and OS. When pooling all available studies, we found that patients with a high MTV showed a 2.50-fold increase in HR for PFS and 8.06-fold increase in HR for OS than patient with a low MTV. Likewise, patients with a high TLG showed a 2.42-fold increase in HR for PFS and 7.23-fold increase in HR for OS than patient with a low TLG. Identifying the prognosis in ovarian cancer is crucial for predicting outcomes and allows to plan the most appropriate management for individual patient. As shown in our meta-analysis, MTV and TLG may be potentially useful prognostic markers in patients with ovarian cancer.

It is important to note that each included studies used different methods for assessing MTV and TLG varying from fixed threshold of VOI based on certain percentages of SUVmax (i.e., 40%, 42%, and 50%), mediastinal blood pool SUV, or background activities around the lesion to the dichotomization of high vs low MTV/TLG on the basis of ROCs of median values, although all included studies placed the VOIs on whole body. Such variable methodology may have resulted in widely dispersed cut-off values ranging from 2.27 to 92 cm3 for MTV and from 6.62 to 563 for TLG in the prediction of survival. Use of different PET scanner model also affected these variability in cut-off value. Due to such variability, it is difficult to demonstrate the optimal cut-off values of these volumetric parameters of 18F-FDG-PET/CT. Despite the fact that included studies in our meta-analysis were heterogeneous in terms of methodological and clinicopathological characteristics, pooled estimates did not show significant heterogeneity with I2 ranging from 0 to 26%. Furthermore, subgroup analyses evaluating the effect of clinical setting and methodology showed that HRs for PFS were consistently significant.

Although the incremental value as prognostic factors value of volume-based metabolic parameters could not be identified on this study, MTV and TLG might have several advantages over SUVmax as prognostic markers of 18F-FDG PET/CT in ovarian cancer as the following reasons: first, the pooled HRs of MTV and TLG were higher than those of SUVmax, although there was no statistically significant difference, presumably due to the paucity of included studies. Second, the tumor burden itself has prognostic impact in ovarian cancer [4]. Third, these volume-based metabolic parameters reflect glucose metabolism in the entire tumor, while SUVmax determined by highest metabolic activity within tumor which is susceptible to noise [8]. Fourth, Volumetric approaches can reflect intratumoral heterogeneity which also have prognostic value in ovarian cancer [18]. Fifth, four of included studies performed multivariate analyses and MTV [22] or TLG [23, 24, 27] were found to be significant independent prognostic factors, whereas SUVmax was not. These findings further support that MTV and TLG may provide incremental prognostic value in ovarian cancer.

Our results demonstrated that volume-based parameters of 18F-FDG PET/CT are effective tools to predict PFS and OS. In addition, MTV and TLG also have shown its potential to monitor the therapeutic effect. Vallius et al. [28] showed that MTV decreased after neoadjuvant chemotherapy in the advanced ovarian cancer cohort, and post-treatment MTV and MTV reduction during treatment were significantly associated with therapeutic response and PFS. This may further support PET volumetric parameters as indicators of patients outcome in the management of ovarian cancer.

There were a few limitations in this meta-analysis. First, the number of included studies was small. Only recently has volume-based metabolic parameters become easily available with the dispersion of commercial softwares that provide automatic tumor 3-dimensional segmentation. Nevertheless, but to the best of our knowledge, this is the first meta-analysis providing a general overview on this topic. Second, all included studies were retrospective in nature. Pooled estimates from predominantly retrospective studies might be subject to the overestimation of effect sizes. Prospective studies may be needed to verify the results of our meta-analysis. Third, variability in methodology and accordingly in cut-off value of MTV/TLG may generate concerns about external validity. Further studies with well-established tumor segmentation methods and cross-calibration of the scanners across multi-centers will be needed to validate this issue. Fourth, volume-based parameters in the included studies did not distinguish the locations of metastasis (i.e., retroperitoneal lymph node metastases, peritoneal disseminations nor distant metastases). As metastasis to different sites have different impact on patient prognosis, the prognostic values of MTV/TLG are highly likely to be stratified by their anatomical location. Therefore, future studies adding these anatomical factors on volumetric parameters are required.

In conclusion, 18F-FDG-PET/CT-derived volume-based metabolic parameters were statistically significant prognostic factors in terms of PFS and OS in patients with ovarian cancer. Despite the clinical and methodological differences across the studies, patients with a high MTV or TLG were at higher risk of disease progression or death.