¹⁸F-FDG PET for the Diagnosis and Grading of Soft-Tissue Sarcoma: A Meta-Analysis

Scope and Eligibility Criteria

We addressed the performance of ¹⁸F-FDG PET as a diagnostic test for differentiating STS from benign soft-tissue lesions and for grading STS. We considered all studies with ¹⁸F-FDG PET evaluations on at least 3 subjects with soft-tissue lesions, at least 1 of which was STS. Reports including only STS cases were eligible, provided that they had collected information on tumor grade. Reports including both soft-tissue lesions and lesions at other sites (e.g., bone) were included, provided that ¹⁸F-FDG PET information on soft-tissue lesions could be separated.

We conducted MEDLINE and EMBASE searches (last update, February 2002) using various search terms for PET (PET, positron emission tomography, ¹⁸F-FDG, and fluorodeoxyglucose) and sarcoma and limited the search to “human subjects.” Searches were also conducted using names of specific histologic types of STS (liposarcoma, malignant fibrous histiocytoma, leiomyosarcoma, fibrosarcoma, malignant schwannoma, synovial sarcoma, and peripheral nerve sheath tumor). Furthermore, we also perused the references of retrieved articles to find additional studies and communicated with expert investigators for additional data and clarifications. We set no language restrictions.

Data Extraction

We extracted information on authors, year of publication, age, number of subjects, benign and malignant lesions evaluated, inclusion and exclusion criteria, number of primary lesions and of lesions assessed for recurrence, study design (prospective, retrospective, or unclear), histologic types, technical characteristics of PET, diagnostic parameters considered, definition and interpretation of the reference test (biopsy and type thereof, operative histologic diagnosis, other imaging, clinical, other, and unspecified), and potential for verification bias. Verification bias refers to incomplete confirmation of the results of the test under investigation with the reference test (e.g., no biopsy performed when PET suggests benign lesion).

For each report, we recorded the number of true positives, false positives, true negatives, and false negatives for ¹⁸F-FDG PET in diagnosing malignant versus benign lesions using the following prespecified parameters: (a) qualitative visualization (simple visualization, qualitative interpretation by experts, or assessment based on a tumor-to-background ratio [TBR] ≥ 3.0 without correction for dose of ¹⁸F-FDG, weight, and plasma glucose); (b) standard uptake value (SUV) ≥ 2.0; (c) SUV ≥ 3.0; and (d) metabolic rate of glucose (MRG) ≥ 6 μmol/100 g/min. We also separated evaluations for primary lesions from evaluations for potential recurrences. Furthermore, whenever information was provided on tumor grade, we recorded the number of lesions that were positive by PET (based on each of the above definitions) for intermediate/high-grade (G II/III) and low-grade (G I) tumors. Benign lesions were separated into noninflammatory and inflammatory ones.

For reports that had also used CT scans or MRI, we evaluated the performance of each test in diagnosing primary disease, local recurrence, and metastases. Data were compared on the same patients when 2 imaging procedures had been performed in parallel. For reports describing longitudinal evaluation of therapeutic response with serial ¹⁸F-FDG PET, the baseline (pretherapy) data were used in the quantitative synthesis. In addition, we extracted descriptive information regarding the impact of ¹⁸F-FDG PET on patient management.

Statistical Analysis

Data were combined quantitatively to provide summary information across studies for each of the 4 prespecified diagnostic definitions. We estimated the overall number of true positives, false negatives, true negatives, and false positives, and we estimated the overall sensitivities and specificities using a random- effects model that incorporated between-study variability. We also performed data synthesis using the summary receiver operating characteristic (SROC) approach that takes into consideration the interdependence between sensitivity and specificity (5,6). Combining sensitivity and specificity data independently across studies may underestimate both parameters and provides no information about the effect of diagnostic threshold. However, the overall random-effects estimates fall close to the SROC curve and can provide a useful indicator of where most investigations operated. The SROC curve shows how the true-positive rate (sensitivity) changes as a function of false-positive rate (1 − specificity) across all studies, when the same diagnostic criterion is used to classify cases as benign or malignant. It is estimated by the equation D = a + bS, where D is the difference of the logit of the true-positive rate and the logit of the false-positive rate and S is the respective sum. Both unweighted and weighted regressions were evaluated. When sufficient data are available, the area under the curve (AUC) can also be estimated. For all analyses, data on primary lesions were treated separately from data on evaluation of recurrences whenever feasible. The main analyses combined all data. Subgroup analyses were also performed for each subgroup (primary and recurrent). We also report typical pairs of sensitivity and specificity from these curves, taking as reference value for specificity the one estimated by random effects weighting of specificities across the included studies.

For the analysis of grading, we calculated the percentage of subjects with each type/grade of lesion when ¹⁸F-FDG PET was positive by each of the prespecified diagnostic criteria. Exact 95% confidence intervals (CIs) are also provided. Differences in pooled proportions were tested by Fisher exact test. Analyses were conducted in SPSS (SPSS Inc.), Meta-Test (Joseph Lau), and StatXact 3.0 (Cytel Inc.).

Eligible Studies

Thirty-three potentially relevant articles were identified and examined in full text. Of those, 18 articles were rejected: 10 articles were rejected because they had <3 soft- tissue lesions; 3 other reports (with a total of 37 patients) (7–9) were rejected because they included only STS cases without grading or only benign lesions; and another 2 overlapping reports were rejected (10,11) because they included data on both bone and soft-tissue lesions that could not be separated. We attempted to contact authors for more data but this was not successful. The remaining articles (12–29) were scrutinized for potential overlap. Among 3 overlapping reports by the Heidelberg team (12–14), only the most comprehensive was retained (14). There were 4 publications by the Groningen team (15–18). The most recent (16) was excluded because it overlapped, to a considerable extent, with previous reports and provided only MRG data, whereas the earlier reports also had additional parameters. Two other reports (17,18) had partial overlap: one examined only primary lesions, whereas the other examined some of the same primary lesions and recurrent lesions. To avoid duplication, we retained the first report (17) and the recurrent-lesions group from the other report (18). Finally, the St. Thomas team had published 5 reports (19–23), but they all provided largely nonoverlapping, complementary information. Thus, 15 reports were analyzed (Table 1) (14,15,17–29).

In all, a total of 421 subjects with soft-tissue lesions had been assessed with ¹⁸F-FDG PET, and 416 could be evaluated (5 patients had either a PET technical failure or no diagnostic documentation). There were 441 evaluable lesions (227 malignant and 214 benign). Three small studies included only malignant lesions documented by biopsy. Otherwise, both malignant and benign lesions were included in 12 reports. There was a total of 264 primary lesions and 177 lesions evaluated for recurrence. Seven studies addressed only primary lesions, 4 addressed only lesions suspected for recurrence, and 4 addressed both.

Diagnoses

We retrieved specific histologic diagnoses on malignant tumors from all studies with one exception (22). Among 208 malignant tumors with recorded histology, the most common were liposarcoma and variants (n = 56), malignant fibrous histiocytoma (n = 43), leiomyosarcoma (n = 21), fibrosarcoma (including myxofibrosarcoma) (n = 13), malignant schwannoma (n = 13), synovial sarcoma (n = 12), and peripheral nerve sheath tumor (n = 11). These types accounted for over 80% of STS. Of note, soft-tissue malignancies also included 1 case of non-Hodgkin lymphoma and 1 case of metastatic nasopharyngeal carcinoma.

Specific diagnoses were available for all 214 benign lesions. The most common diagnoses were postsurgical or posttraumatic lesions (typically scars, n = 113), lipoma (n = 23), neurofibroma (n = 21), hemangioma (n = 13), and schwannoma (n = 12). Overall, there were 199 benign noninflammatory lesions and 15 benign lesions with acute or chronic inflammation (infectious or noninfectious).

¹⁸F-FDG PET and Reference Test Characteristics

A variable amount of radiopharmaceutical was used across studies (148–407 MBq). Imaging was typically performed after fasting from a few hours to overnight. With one exception (21), all studies attempted some qualitative interpretation of the PET images, based mostly on various nonquantitative criteria with consensus between radiologists, and in 3 studies, a crude TBR was also used. SUV was estimated in 9 studies, all of which had primary lesions, and 3 also included an evaluation of recurrences (total of 12 case series). MRG was estimated in 5 studies (3 of primary lesions, 2 evaluations of recurrences). Histology was the typical reference standard for all lesions, but there were some notable exceptions. In 3 studies (19,20,22), most or all of the patients with benign-appearing lesions did not have histologic confirmation, and diagnosis depended only on clinical and radiologic criteria. This represents considerable verification bias. A small number of patients with benign lesions did not have histologic confirmation in a fourth study (14) (Table 2). Typically there was no clear mention about whether ¹⁸F-FDG PET evaluations were performed without knowledge of histologic results.

Data Synthesis: Diagnosis of Malignant Versus Benign Lesions

Summary estimates are presented in Table 3 for all the examined diagnostic parameters. Qualitative visualization interpretations were available on 398 subjects from 13 case series comparing malignant and benign lesions. There were 184 true-positives (patients with malignancies detected as such by ¹⁸F-FDG PET), 16 false-negatives (patients with malignancies considered benign by ¹⁸F-FDG PET), 145 true-negatives (patients with benign lesions correctly identified by ¹⁸F-FDG PET), and 53 false-positives (patients with benign lesions incorrectly classified as malignant by ¹⁸F-FDG PET). Respective numbers from 10 case series with SUV data with a cutoff of 2.0 (or 3.0) were 79 (or 60) true-positives, 21 (or 40) false-negatives, 89 (or 97) true-negatives, and 19 (or 11) false-positives. Data on MRG were limited (n = 66 from 4 case series) and included 30 true-positives, 12 false-negatives, 18 true-negatives, and 6 false-positives. The overall accuracy rate was 82.7%, 80.8% (or 75.5%), and 72.7%, respectively, with these diagnostic parameters. The studies seemed to operate at different parts of the SROC curve, and independent estimates of sensitivity and specificity (Table 3) or simple pooled accuracy rates might be underestimating the true performance of the test. The diagnostic performance was better when formal SROC analyses were undertaken (Table 3 and Fig. 1). Both SUV and simple qualitative visualization had somewhat better diagnostic performance than MRG. An AUC would be meaningful only for qualitative visualization, when there were both sufficient data and a wide range (0%–100%) of specificity estimates in individual studies. AUC was estimated at 0.91 (0.92 with weighted regression). There was no evidence of a different overall diagnostic performance (based on AUC estimates) when primary lesions were examined separately from evaluations for recurrence. However, diagnostic performance trended toward larger sensitivity and smaller specificity in primary lesions than in evaluations of recurrent lesions (Table 3). ¹⁸F-FDG PET was highly specific in ruling out recurrence but had limited sensitivity for diagnosing recurrence.

• Download figure
• Open in new tab
• Download powerpoint

FIGURE 1.

SROC curves for (A) qualitative visualization, (B) standard uptake value (cutoff, 2.0), (C) standard uptake value (cutoff, 3.0), and (D) metabolic rate of glucose (cutoff, 6.0 μmol/100 g/min). Curves show trade-off between true-positive rate (sensitivity) and false-positive rate (100 − specificity) across all pertinent studies. Each study is shown by eclipse with diameters approximately proportional to number of subjects evaluated for sensitivity (vertical dimension) and specificity (horizontal dimension) of study. Two SROC curves are shown based on weighted (bold line) and unweighted (thin line) calculations. SROC curves should be used for inferences of diagnostic accuracy preferably within range of sensitivity and specificity values of studies that are included in their calculations. X = random effects estimates of sensitivity and specificity; horizontal and vertical dimensions of rectangle = corresponding 95% confidence intervals.

Data Synthesis: Grading

In the 11 case series in which tumor grading was assessed and qualitative visualization was used, all 133 (100% [95% CI, 97.3%–100%]) intermediate/high-grade malignant lesions and 32/43 (74.4% [95% CI, 58.6%–85.9%]) low-grade malignant lesions were detected. The respective rate for benign lesions was 35/89 (39.3% [95% CI, 29.1%–50.3%]) in the same studies or 74/196 (37.8% [95% CI, 31.0%–44.9%]) when studies without tumor grading were included. Rates differed significantly between the 3 groups. Of 12 inflammatory benign lesions, 11 were positive on ¹⁸F-FDG PET (91.7% [95% CI, 61.5%–99.8%]).

In the 8 case series in which tumor grading was assessed and SUV was estimated, values above 2.0 were seen in 59/66 (89.4% [95% CI, 79.4%–95.6%]) of intermediate/high-grade malignant lesions, 8/24 (33.3% [95% CI, 15.6%–55.3%]) of low-grade malignant lesions, and 13/68 (19.1% [95% CI, 10.6%–30.5%]) of benign lesions. SUV values above 3.0 were seen in 45/66 (68.2% [95% CI, 55.6%–79.1%]), 3/24 (12.5% [95% CI, 26.6%–32.4%]), and 8/68 (11.8% [95% CI, 5.2%–21.9%]), respectively. With either cutoff, there was no significant difference between the low-grade malignant lesions and the benign lesions, whereas intermediate/high-grade malignant lesions differed significantly from both other groups. Inferences were similar when data on benign lesions were included from studies without tumor grading (data not shown).

MRG data were limited, but the inferences were similar. Values ≥ 6.0 μmol/100 g/min were seen in 32/35 (91.4% [95% CI, 76.9%–98.2%]) G II/III tumors, 1/13 (7.7% [95% CI, 0.2%–36.0%]) G I tumors, and 6/24 (25.0% [95% CI, 9.8%–46.7%]) benign lesions (all noninflammatory).

Comparison with MRI and CT

Several studies on primary lesions clarify that part of the inclusion criteria was the prior performance of CT, MRI, or ultrasound, suggestive of malignancy. Because MRI/CT and ¹⁸F-FDG PET are performed in series, rather than in parallel, one cannot address their comparative accuracy for primary soft-tissue lesions. The same applies to studies on evaluation of recurrence, with 2 exceptions: In one study (22), ¹⁸F-FDG PET sensitivity was 13/17 (76%) and specificity was 47/50 (94%) versus 15/17 (88%) and 48/50 (96%), respectively, for paired MRI. ¹⁸F-FDG PET was wrong in 3 cases in which MRI was correct. The study may have substantial verification bias. In the other study (15), PET had 92% sensitivity (12/13) and 100% specificity (2/2). Paired MRI failed to diagnose 3 malignant lesions (including the one also misdiagnosed as benign by PET) and misdiagnosed both benign lesions (scar and Ascaris mass). PET was correct in 4 cases in which MRI was wrong.

Data on the comparative performance of ¹⁸F-FDG PET for the diagnosis of distant metastasis were also sparse. In 1 study (22), PET was positive by qualitative interpretation in 13/15 lung metastases (sensitivity, 87%) and was negative in all 55 cases without metastasis (specificity, 100%). Paired CT scans had 100% sensitivity and 96% specificity. PET failed to detect 2 lung metastases seen on CT but was correct in 2 cases in which CT was falsely positive. The study suffers from substantial verification bias for patients with negative imaging. Inconclusive retrospective paired CT and PET data were also given by 1 of the excluded studies (7) on 8 patients and were also subject to verification bias.

Longitudinal Evaluation of Treated STS

¹⁸F-FDG PET was used for the evaluation of response to therapy in 3 case series with at least 4 subjects involving radiotherapy hyperthermia (n = 4); hyperthermic isolated limb perfusion with tumor necrosis α, interferon γ,and melphalan (n = 20); and various modalities (n = 8), for a total of only 32 patients (24,18,7). PET showed clear changes in the treated tumors but did not discriminate partial from complete response in the largest study (18).