This meta-analysis was performed following guidelines for the systematic review and meta-analysis of diagnostic studies[8]. Ethical committee approval and patients’ consents were not required for this work due to its statistic nature. MEDLINE (January 1980 to August 2012) were used for searching English-language articles with the following keywords: (“positron emission tomography OR fluorodeoxyglucose”) AND (“prediction OR response OR response monitoring OR neoadjuvant OR chemotherapy”) AND (“carcinoma OR cancer OR tumor OR tumour”) AND (“sensitivity OR specificity OR accuracy”) AND (“sarcoma”). Two reviewers independently checked the abstracts of retrieved publications and obtained the fulltext of each eligible article. References of the retrieved studies were screened to expand the scope of the computerized literature search. Disagreements were resolved by discussion between the two reviewers.

Studies were eligible for inclusion based on the following criteria: (1) using F-18 FDG PET to assess response to neoadjuvant therapies in patients with primary bone or soft tissue sarcoma; (2) employing pathological outcome as the “gold standard”; and (3) clinical raw data could be drawn from data in the original published study. A minimal sample size for inclusion was not required given the relative rareness of sarcomas in clinical scenarios. Conference abstracts, case reports, comments, and letters to the editors were excluded.

To identify the methodological quality of included article studies, two reviewers independently assessed each article using the QUADAS tool, which is specifically designed and widely used in systematic reviews of diagnostic accuracy studies[9]. It provides a checklist of items regarding the representativeness and methodological quality of investigated studies and presents an accumulated score as a result. Description of the execution of PET scans and the pathological interpretation were assessed. The optimal time interval between F-18 FDG PET and histopathological reading was defined as less than 30 d. Whether the pathological interpretation was blind to the PET scan results was also recorded. However, the PET readers were naturally blind to pathological information, since these scans were performed before the pathological results from following surgeries.

Two reviewers independently extracted the following data by using a standardized data extraction form: (1) characteristics of the study, including investigated tumor pathology (OS or STS), sample size, patient age, timing for PET scans, specific criteria of PET and pathology, etc.; (2) methodological details which needed to be examined with the QUADAS tool; and (3) clinical outcome. A 2 × 2 contingency table was constructed consisting of true-positive, false-positive, true-negative and false-negative for each study. We considered response to therapies as “positive”, while therapy failures as “negative”.

Based on the 2 × 2 contingency tables, sensitivity was calculated as the ratio of the number of true-positive findings showed in the index test to the number of all positive findings supported by the golden standard. Specificity was calculated as the ratio of the number of true-negative findings reported by the index test to the number of all negative findings showed in the reference standard. The diagnostic odds ratio (DOR) was calculated as [True positive (TP) × True negative (TN)]/[False positive (FP)× False negative (FN)], where TP, TN, FP and FN stand for true-positive, true-negative, false-positive and false-negative, respectively. If the DOR could not be calculated because one of the cells in the 2 × 2 table was empty, 0.5 was added to all cells[10]. We used the χ² test to assess the heterogeneity among studies. The I² index was employed, with an I² value of over 50% indicating statistical heterogeneity (a random-effects model was accordingly utilized), and an I² value of less than 50% suggesting that homogeneity was reached (a fixed-effects model was then applied)[11,12].

If statistical heterogeneity was found, subgroup analyses were then performed to explore its possible sources (e.g., different tumor pathology, patient age group and chemotherapy regimen). We employed the Z test to examine the statistical differences between subgroups. Two-sided P < 0.05 was considered statistically significant.

Our summary receiver operating characteristic curves (SROC) was plotted in accordance with Mose’s linear model, reflecting the summary trade-off between sensitivity and specificity across the included studies. The areas under the curves (AUC) and Q* value were also calculated. Q* value is defined by the point where sensitivity and specificity are equal, which is the point closest to the ideal top-left corner of the SROC space, reflecting the diagnostic value[13].

To analyze publication bias, we created inverted funnel plots of individual study log DORs plotted against the effective sample size. Both Begg’s rank correlation test and Egger’s regression test were utilized to assess the asymmetry of the funnel plot[14]. An asymmetrical funnel plot would suggest that additional small-sample studies may have been conducted but not published due to their unfavorable results.

After independent review, our research initially yielded a total of 50 articles in English-language from MEDLINE. Twenty-three articles were considered candidates after careful review of abstracts. After fulltext reading of these articles, eight were excluded because of employing reference standard other than pathological outcome (n = 2); lack of raw data that was essential to built the 2 × 2 contingency tables (n = 5); or focusing on metastasized STS only (n = 1). Finally, 15 articles fulfilled all eligibility criteria and were included for data extraction and following analysis[3,6,7,15-26]. Flow diagram of publication selection is showed in Figure 1.

Open in New Tab   Full Size Figure   Download Figure

Figure 1 Flow diagram of publication selection for this meta-analysis. STS: Soft-tissue sarcoma.

Among the 15 studies, four were retrospectively designed[6,17,21,25], one had unclear design of whether prospective or retrospective[24], and the rest ten were prospective. All were in English. A total of 420 patients were included. Median sample size was 21 (range, 9 to 50), and mean age ranged from 10 to 54 years old. Three studies focused on pediatric patients only[3,6,24].

Detailed characteristics of the 15 included trials are outlined in Table 1. All articles assessed treatment of neoadjuvant chemotherapy, and in four studies some patients received neoadjuvant radiotherapy as well[19,20,23,25]. All trials had a sequential F-18 FDG PET both before and after the neoadjuvant therapies were performed.

All studies included in the meta-analysis fulfilled 9 or more of the 14 standards in the guidelines of the QUADAS tool. Drawbacks were commonly seen in description of how patients were selected and whether histopathological interpretation was blind to PET results. Moreover, the availability of clinical data in the diagnostic test reading was mostly unclear. The median methodological quality score (represented as the number of “Yes” to fourteen questions) was 12.

Estimation of overall accuracy: Table 2 exhibited the absolute numbers from 2 × 2 contingency tables from each article. When more than one standard of PET was present in one trial, we took the one that generated higher accuracy into statistic calculation. The sensitivity of F-18 FDG PET to assess therapy response in sarcoma ranged from 40%-100%, and specificity from 44%-100%. The forest plot displayed the sensitivity, specificity and DOR with 95%CI for each individual study (Figure 2).

Open in New Tab   Full Size Figure   Download Figure

Figure 2 Pooled sensitivity, specificity, and diagnostic odds ratio in forest plots for neoadjuvant therapy response prediction in bone and soft tissue sarcomas by F-18-fluorodeoxy-D-glucose-positron emission tomography.

Based on the random-effects model, the pooled sensitivity and specificity were 87% (95%CI: 81%-91%) and 83% (95%CI: 77%-87%), respectively, and the pooled DOR was 40.5 (95%CI: -14.6-112.0). The SROC indicated an AUC of 0.930 and Q* value of 0.865 (Figure 3). Heterogeneity of sensitivity and specificity was detected, but no significant heterogeneity of DOR was found, as their I² exhibited in Figure 2.

Open in New Tab   Full Size Figure   Download Figure

Figure 3 Summary receiver operating characteristic curves for neoadjuvant therapy response prediction in bone and soft tissue sarcomas by F-18-fluorodeoxy-D-glucose-positron emission tomography.

Several subgroup analyses were then performed to examine the potential sources of heterogeneity displayed among the 15 studies. It was noted that these studies investigated different tumor pathology, patient age group and chemotherapy regimen. Subgroup analysis results are showed in Table 3.

Tumor pathology: The I² values of sensitivity and specificity were close to 50% within both the group “OS or ES” and “STS”, while within the group “STS” the heterogeneity of DOR was eliminated (I² = 0%). Statistic suggested that PET had better specificity and DOR when assessing the “OS or ES” group than the “STS” group.

Patient age group: Three studies focused on pediatric patients, and four studies investigated adult patients only. The heterogeneity of sensitivity, specificity and DOR was all resolved within the pediatric patient group (I²: 0%-37%), and the I² of DOR within the adult patient group was 50%. Results showed that PET had significantly better specificity and DOR when used among pediatric patients than adults.

Chemotherapy regimen: Ifosfamide was reported to cause inflammatory reactions and consequently increase the chance of mismatch between PET scans and pathology[21]. To assess the effect of ifosfamide, we conducted this subgroup analysis. We found that within the group using chemotherapy agents without ifosfamide, the heterogeneity of specificity and DOR was resolved (I²: 0% and 15.7%), and significantly better specificity and DOR than the group using ifosfamide were gained.

PET response criteria: Ten studies employed a lower standardized uptake value (SUV) after chemotherapies (mostly between 2 and 3) and/or a higher SUV reduction rate (27%-60%, mostly around 50%) as PET criteria of good response. Three studies used tumor to background ratio (TBR) related parameters to define good response. However, subgroup analysis did not resolve the heterogeneity of any parameters (sensitivity, specificity and DOR), hindering a reliable further comparison.

The inverted funnel plot suggested some asymmetry of the distribution of the points (Figure 4). Begg’s test indicated that publication bias was unlikely with a P value of 0.06. However, Egger’s test showed that such bias might exist with a P value of 0.003.

Open in New Tab   Full Size Figure   Download Figure

Figure 4 Funnel plot to test publication bias shows some asymmetry.

The difference of pathology and age predilection between bone and soft tissue sarcomas rationalize identification of optimal patient groups of using F-18 FDG PET to assess response to NCT. It is noted that all three studies that focused on pediatric patients investigated OS or ES. Further discrimination of OS and ES was hardly possible, since many studies did not report such data separately. Almost all studies investigating STS were comprised of merely adult patients (one study did not clearly report the patient age range[7]). One study included both bone and soft tissue sarcomas[21]. We conducted subgroup analysis between studies that included merely pediatric patients or adult patient to identify the effect of patient age, and so as to the effect of two pathological groups. Statistical results suggested that PET was especially promising among pediatric patients with OS and ES, with relatively high sensitivity and specificity, both around 90%.

The pooled sensitivity of PET in response assessment of sarcoma was relatively high and stable among all subgroups in this meta-analysis, suggesting a relatively low chance of false negatives in clinical practice. By contrast, the pooled specificity was lower, and almost all the significant difference gained in subgroup analysis was attributed to specificity, calling for clinical caution of false positives. Certain chemotherapy agents have been reported to potentially cause misjudgments in PET scans after treatment of sarcoma, and ifosfamide was among the most widely known ones[21]. It is reported in bone tumors to theoretically increase SUVmax of adjacent inflammatory tissues, causing false negatives consequently. Differing from such possibility, our subgroup analysis found more chance of false positives. The most distinct and largest numbers of false positives were generated in studies by Benz et al[18] and Evilevitch et al[20], both investigating STS, suggesting that ifosfamide might have different mechanisms in STS from bone tumors. Since it was reported in clinical practice that addition of ifosfamide could improve patients’ survival in both OS and ES, and ifosfamide was widely used in STS as well, it is noteworthy that this chemotherapeutic agent has the potential to cause misjudgment on PET scan results of sarcoma.

Different parameters have been employed to explore a better accuracy of F-18 FDG PET, among which maximum SUVs were considered quantitative and reproducible, therefore widely used in clinical trials[17]. Six studies included in this meta-analysis relied on post-treatment SUV (SUV2, mostly maximum SUV) and eight trials relied on SUV reduction rate to identify responders, and overlaps existed. It was pointed out that SUV2 may remain elevated because of inflammation or reactive changes. Moreover, tumors with an initial high pre-treatment SUV might still have a high SUV2, despite a significant decrease in the SUV value. SUV reduction rate is therefore proposed as a more reliable indicator[6]. In contrast to such knowledge, three included studies that compared the accuracy of SUV2 and SUV reduction rate all presented that SUV2 had a better correlation with histopathology than the latter[17,23,24]. TBR was employed in three studies, but was reported as nonreproducible and less practical due to inter-observer variability[3]. Since the subgroup analysis of different criteria still suffered unexplained heterogeneity, the optimal criteria remain to be defined in larger-sample, prospectively designed studies that include and compare directly different parameters.

The insufficient accuracy of current imaging modalities raised reasonable concerns of clinicians regarding the roles of pre-operative PET. Potential clinical applications may include: (1) providing prognostic indicator of disease-free survival after multimodal treatment[5]; (2) guiding the selection of resection technique, because surgical oncologists may be more willing to perform limb-sparing surgical resections in the setting of a favorable response to therapy[27]; and (3) potentially identifying non-responders early (pooled NPV = 90%), since several studies explored the use of interim PET information and all generated positive results[3,7,15,18]. Benz et al[18] compared the performance of PET after the first cycle and after the completion of NCT, reporting that the earlier scans had an even better accuracy. Because the currently available treatments are associated with significant toxicities[28], early identification of non-responders would allow clinicians to minimize the toxicity of ineffective chemotherapies and switch the treatment strategy to more aggressive regimen or addition of radiotherapy. However, further large-scale prospective studies are required to adequately position PET in sarcoma treatment management.

Our meta-analysis had several drawbacks. First, the included studies had relatively small sample sizes in general, giving that sarcoma lacks abundant cases in clinical scenarios. Second, some extent of heterogeneity among studies was detected, while subgroup analysis could not eliminate all of the heterogeneity (although some was resolved). Third, some biases might be present such as publication bias, since P value yielded by Egger’s test was 0.003. Spectrum biases were possible due to the failure of finding out comprehensive relevant articles or unpublished data, caused by the restriction of searching conditions.

In summary, this meta-analysis reveals a beneficial value of F-18 FDG PET to assess histopathological response after NCT in bone and soft tissue sarcomas. It is especially promising to apply to pediatric patients with OS or ES. Post-NCT SUV and SUV reduction rate on PET scans are potential candidates for quantitative criteria of response. It is noteworthy that certain chemotherapeutic agents, specifically ifosfamide, could contribute to false positives of PET results. Given the potential of PET to guide clinical treatment strategies and establish personalized medical plans, large-scale prospective studies are required to adequately position this modality in sarcoma management.

Sarcoma is a rare but highly malignant tumor that frequently requires neoadjuvant chemotherapy, but only a subcollective of patients benefit from such treatment. Therefore, there is an urgent need for non-invasive methods to accurately identify response of sarcoma to chemotherapy, and accordingly decide whether to switch to a more intensified schedule.

Radiologic imaging has some limitations for this utility due to the biological characteristics of sarcoma and the disadvantages of conventional anatomic imaging. Positron emission tomography (PET) using [F-18]-fluorodeoxy-D-glucose (F-18 FDG) is a newly applied imaging modality that measures tissue metabolism and has the unique ability to detect response at the molecular level, therefore is being increasingly used in chemotherapy response monitoring among patients with sarcoma. However, the actual accuracy of PET and some specific issues in clinical practice such as optimal patient groups, standards for response and possible causes for misjudgment remain to be explored.

There have been multiple clinical trials assessing the accuracy of PET to monitor response to chemotherapy in sarcoma. Nonetheless, the sample sizes in conducted trials were relatively small and the results among them were remarkably discrepant. This meta-analysis provided a systematic and quantitative evaluation of the accuracy of PET for this utility. Moreover, it generated further information that clinical physicians concerned with, such as recommended subgroup of patients, potential candidates for quantitative criteria of response and possible causes of false positives of PET.

The study results suggested that F-18 FDG PET is promising to predict neoadjuvant therapy response in sarcoma, especially in pediatric patients with osteosarcoma or Ewing sarcoma. A lower standardized uptake value (SUV) after chemotherapies and/or a higher SUV reduction rate are potential candidates for quantitative criteria of response. Certain chemotherapeutic agents such as ifosfamide could potentially cause false positives of PET.

Positron emission tomography is a functional imaging technique that produces a three-dimensional image of the body. When fludeoxyglucose (FDG), an analogue of glucose, was chosen as the tracer for PET, its concentrations will indicate tissue metabolic activity and therefore explore the possibility of viable tumor; Standardized uptake value is a frequently used quantitative parameter generated on PET scans, indicating concentrations of the tracer (e.g., FDG) in targeted tissues and therefore reflecting their metabolic activity.

Interesting meta-analysis on the use of PET CT in the evaluation of patients with sarcomas of both soft tissue and bone after chemoradiation with a curative intent. The authors have used all the appropriate scientific methodologies used for the preparation of meta-analyses and their results are extremely interesting and supported by the literature. The authors conclude that PET scan is a promising diagnostic tool in the evaluation of neo-adjuvant treatment response in sarcomas.