This article is mentioned in:

Introduction

With the exception of stage, no definitive prognostic factors have been established for head and neck cancer. Cancer stage is currently the most important prognostic factor for outcome. Traditional staging of head and neck squamous cell carcinoma (HNSCC) depends on the site of disease origin, size and extent of the primary tumor, cervical lymph node involvement and presence or absence of distant metastasis. This staging system relies on conventional imaging, such as computed tomography (CT) and magnetic resonance imaging (MRI); however, this approach is limited by outcome heterogeneity within stage categories, hampering accurate prognostication for individual patients (1).

Clinical characteristics, including gender, age, weight loss and biological markers, such as p53 (2), cyclin D (3) and epidermal growth factor receptor (4), have been investigated but are not sufficiently accurate for individual patient prognosis. More accurate markers would be helpful to stratify patients for therapy and predict outcomes.

As with other types of cancer, there is great interest in trying to identify factors predictive of outcome other than stage and biological markers. The development of functional imaging studies, particularly positron emission tomography (PET) with the glucose analogue 2-(18F)-fluoro-2-deoxy-D-glucose (18F-FDG) has emerged as a useful tool for several malignancies. PET scanning enables non-invasive study of the physiology of cancers (5). Tumor uptake of18 F-FDG, standardized uptake value (SUV), as measured by PET, has been associated with various cellular characteristics, such as cell viability (6) and proliferative activity (7). Furthermore, SUV as a semi-quantitative simplified measurement of tissue deoxyglucose metabolic rate has suggested that tumor FDG uptake may be of prognostic significance, as patients with high FDG uptake generally have a less favorable outcome (8).

Other retrospective studies have investigated the prognostic significance of SUV, but the majority of those reports only included a small patient sample. Based on these considerations, we performed a systematic review of the literature on18 F-FDG-PET scan and local control and a meta-analysis of the data to determine the prognostic value of primary tumor SUV in patients with head and neck cancer.

Materials and methods

Search strategy

We conducted a search, without language restrictions, through PubMed, Embase, the Cochrane Controlled Trials Register and OVID databases. We employed both medical subject headings and free-language terms for ‘HNSCC’ (i.e., ‘head and neck cancer’, ‘head and neck carcinoma’, ‘squamous cell carcinoma’), combined with each of the following: ‘positron emission tomography or PET or PET imaging tomography’ and ‘FDG-F18 or FDG or18 F-fluorodeoxyglucose or18 F-FDG’ and ‘SUV or standardized uptake value’ as search terms. The references reported in all the identified studies were used to complete this search, which ended in May, 2014. Further searches were conducted by scanning the abstracts of major ENT and nuclear medicine meetings.

Inclusion and exclusion criteria

In order to be eligible for the systematic review, a study was required to fulfill the following criteria: i) Limited to HNSCC, with any stage or any histological grading; ii) assessed the association between pretherapeutic SUV and local control, at least in a univariate analysis; iii) SUV referred to the primary tumor; and iv) reports using all modalities of care were included.

Abstracts were excluded, as they do not provide sufficient details to assess methodology or relative information to perform a meta-analysis. Furthermore, we carefully examined the possibility of patient duplication by reporting the same cohorts in different publications. This led us to suppress one article, although no reference to such duplicates was reported by the authors.

Quality assessment

To improve quality assessment, 7 physicians and 1 biostatistician reviewed all the publications to assess their methodological quality and extract the most important information determining the clinical and PET characteristics. A methodological quality scale was designed for the purpose of this study, using the variables available in the publications (9). This score assessed the clinical and PET reports. The clinical report included the distribution of the expected ‘prognostic factors’ (age, gender, stage, performance status, histological grading and weight loss), tumor stage description, staging characteristics [definition of the size of pathological metastasis, systematic use of the head and neck CT for head and neck staging, systematic metastatic work-up, systematic use of a CT or MRI for distant metastasis, histological confirmation of metastasis and if the analysis of the association between SUV and each expected prognostic factor was performed without knowledge of clinical results and vice versa (double blind)], description of results of local control and analyses (number of patients, number of local control, follow-up duration, number of patients lost to follow-up, univariate and multivariate analyses, description of statistical tests, definition of local control and SUV cut-off definition). The PET report included patient characteristics (weight/height, glycaemia and histological subtype),18 F-FDG-PET acquisition protocol characteristics (injected dose of18 F-FDG, delay between injection and data acquisition and fasting duration) and technical parameters (investigation area, delay between head and neck CT and PET acquisition, SUV formula, type of PET engine, duration of emission time, duration of transmission time, attenuation and reconstruction parameters and type of SUV). The clinical and PET reports were scored on 21 and 14 points, respectively. A value between 0 and 2 was attributed to each item. The scores were expressed in percentage of the maximal theoretical value that can be obtained. When the results of a particular study were reported in more than one publication, only the most recent and complete data were included in the meta-analysis.

Statistical analysis

Data were extracted by Zhang and Nie and discrepancies were discussed and resolved by Dong. Data were entered into the Cochrane Collaboration Review Manager program RevMan 4.2.2. We measured the impact of SUV by relative risk (RR) between the local control distributions of two groups. For each trial, this RR was estimated by a method depending on the results provided in the publication. The most accurate method was through determining the total number of events and the number of patients at risk in each group, allowing calculation of the RR estimation. If the only exploitable data were in the form of graphical representations of local control distribution, they were used to extract the corresponding rates at specified times to reconstruct the RR estimate and its variance, with the hypothesis that the rate of patients censored was constant during the study follow-up. The individual RR point estimates were combined following acceptation of the null hypothesis of the homogeneity of the treatment effect across the various trials, using the RevMan 4.2.2 software to obtain a global RR estimate of the treatment effect. The RR was calculated using the fixed-effects method. In case of significant heterogeneity (P<0.05), the random-effects method was applied. This impact of SUV on local control was considered as statistically significant if the 95% confidence interval (95% CI) for the overall RR did not overlap 1. All reported P-values were two-tailed. Finally, funnel plot asymmetry was used to detect any publication bias in the meta-analysis.

Results

Study characteristics

A total of 7 articles (10–16) on SUV in HNSCC were retrieved. One study was excluded from the analysis due to patient duplication (16) and the remaining 6 studies, published between 2002 and 2009, were considered eligible for this review. The sites of primary tumor in the studies included the oral cavity, oropharynx, nasopharynx, hypopharynx, larynx and maxilla. The follow-up time for local control, disease-free survival and overall survival ranged between 2 and 82.8 months per study. The principal characteristics of the 6 studies eligible for the meta-analysis are described in Table I. For the present study, SUV referred to the primary tumor, except in 5 patients (3 with unknown primary tumors and 2 with T1-2 tumors), in whom only the lymph nodes exhibited increased uptake. Consequently, for these 5 patients, the SUV of the lymph node was used as reference for correlation with local control. The main SUV characteristics reported in the publications are described in Table II.

Table I. Characteristics of studies included in this meta-analysis.

Table I.

Characteristics of studies included in this meta-analysis.

Author (year)	Total patient no.	SUV<threshold patient no.	SUV>threshold patient no.	Histology	(Refs.)
Brun et al (2002)	44	23	21	HNSCC	(10)
Allal et al (2004)a	119	62	57	HNSCC	(11)
Kim et al (2007)	52	27	25	HNSCC	(12)
Roh et al (2007)	79	48	31	HNSCC	(13)
Liao et al (2009)	109	97	12	HNSCC	(14)
Torizuka et al (2009)	50	21	29	HNSCC	(15)

a Include d 3 patients with unknown primary tumors and 2 with T1-2 tumors; in these 5 patients, the SUV of the lymph node was used as reference. SUV, standardized uptake value; HNSCC, head and neck squamous cell carcinoma.

Table II. Main SUV characteristics reported in the six publications assessable for meta-analysis.

Table II.

Main SUV characteristics reported in the six publications assessable for meta-analysis.

Study (year)	Type of SUV	Correction of SUV	SUV threshold definition	SUV threshold	(Refs.)
Brun et al (2002)	SUVmean	Weight	Median	9.0	(10)
Allal et al (2004)a	SUVmax	Weight	Median	4.76	(11)
Kim et al (2007)	SUVmax	Lean body weight	Median	6.0	(12)
Roh et al (2007)	SUVmax	Weight	Best cut-offb	8.0	(13)
Liao et al (2009)	SUVmax	Weight	Best cut-offb	19.3	(14)
Torizuka et al (2009)	SUVmax	Weight	Best cut-offb	7.0	(15)

a Include d 3 patients with unknown primary tumors and 2 with T1-2 tumors; in these 5 patients, the SUV of the lymph node was used as reference.

b The authors maximized the log-rank test statistic to determine the optimal cut-off. SUV, standardized uptake value.

Study quality

The methodological quality of the studies was moderate. Overall, the median quality score was 64%, ranging between 53 and 70%. The respective median values for the clinical and PET reports were 62 % (range, 55–69%) and 66% (range, 50–82%), respectively.

High FDG uptake is a marker of poor outcome in HNSCC

The combined RR from the 6 reports for local control was 0.72 (95% CI: 0.63-0.81). The combined RR was calculated using the fixed-effects method. The combined RR confirmed that high FDG uptake on PET is a marker for poor outcome in primary HNSCC. The results are detailed in Fig. 1. The funnel plot revealed a symmetrical distribution, indicating no evidence of substantial publication bias. The results are detailed in Fig. 2.

Figure 1. Graph of SUV>threshold vs. SUV<threshold on local control of patients with head and neck squamous cell carcinoma. SUV, standardized uptake value; RR, relative risk; 95% CI, 95% confidence interval.

Figure 2. Funnel plot of included studies on local control. SUV, standardized uptake value; SE, standard error; RR, relative risk.

Discussion

Over the last decade, FDG-PET has become an important tool used to stage patients with HNSCC. Furthermore, in vivo imaging of human tumors with FDG-PET is a clinical extension of classical studies on carbohydrate metabolism. It was demonstrated that a high rate of glycolysis contributes to tumor growth, which may be exploited for malignancy grading with PET (17). The specific goal of this study was to evaluate the potential of SUV as a prognostic marker. The present meta-analysis confirmed that increased SUV of the primary tumor is a poor prognostic factor in patients with HNSCC.

Under the balanced circumstances of glucose metabolism,18 F-FDG is phosphorylated. The retention products of FDG metabolism are consistent with the amount of glucose consumed in the cell. Therefore,18F-FDG may reflect the utilization of glucose in vivo. SUV is a semi-quantitative index that shows the characteristics of the18 F-FDG tracer uptake, hence approximating the glucose metabolic rate. Commonly, some factors affecting the SUV include outlining of the region of interest, focus size, system resolution, reconstruction algorithm, patient factors (body weight, lean body weight and body surface area), non-tumor uptake caused by activating and acquisition time following drug injection. The time between injection and PET data acquisition in the eligible articles ranged between 45 and 60 min; that makes SUV closer to the glucose metabolic rate. However, SUV estimates suffer from poor reproducibility between centers, due to the lack of standardization of the acquisition and processing protocols for its assessment. In our research, this poor reproducibility was proven by the broad range of threshold values that have been used in the literature to distinguish between patients with low and high survival. In order for it to be used as a functional prognostic factor in routine practice, a single SUV threshold distinguishing between patients should be agreed on, or an optimization method, used to determine the threshold for each center, should be established. To set a unanimous threshold, most variable factors affecting the SUV estimations should be defused or at least controlled. Reducing the large variability currently affecting SUV estimates is likely to enhance the prognostic value of SUV. In our research, we did not take into consideration the variable conditions under which SUV was obtained, due to the poor quality scores of the PET reports. Despite variability in the abovementioned factors, we were able to demonstrate that SUV was correlated with patient local control. Indeed, our research design calculated a RR for each study center to control the deviation of data, based on the SUV threshold used in the corresponding study, which somehow cancelled the difference in threshold factors used in different centers. By doing so, we were able to ignore the difference in SUV values between different centers and demonstrated that SUV is certainly worth considering as a prognostic factor.

Literature-based meta-analyses have the advantage of including published trials immediately available for analysis, with results that can be readily checked. In our meta-analysis, some biases may have occured. Although the funnel plot did not indicate publication bias, bias cannot definitely be ruled out, due to the small number of the studies and the low power of the test used to detect publication bias. Certain studies were not included, as separate data for head and neck cancer patients could not be obtained. Furthermore, one limitation of the present study is the lack of data from multicenter and large-scale perspective studies. To avoid some of the biases of literature-based meta-analyses, we aim to confirm our results in a meta-analysis of individual patient data, incorporating unpublished trials and updating results.

In conclusion, this meta-analysis provided evidence regarding the potential value of FDG uptake, as measured by SUV, in predicting local control in head and neck carcinomas. We are currently planning a meta-analysis based on individual patient data that will potentially reduce biases associated with literature-based meta-analyses.

Acknowledgements

The authors are indebted to the authors of the primary studies.

References