Diagnostic performance of choline PET/CT for the detection of bone metastasis in prostate cancer: A systematic review and meta-analysis

Purpose The aim of this study was to evaluate the diagnostic performance of choline positron emission tomography/computed tomography (PET/CT) for the detection of bone metastasis in patients with prostate cancer. Methods MEDLINE, EMBASE and the Cochrane Library were searched up to 20 February 2018 for studies that used 11C-choline or 18F-choline PET/CT for the detection of bone metastasis in patients with prostate cancer and “histopathology and/or clinical follow-up” as the reference standard. Methodological quality was assessed using the Quality Assessment of Diagnostic Accuracy Studies-2 (QUADAS-2) tool. Pooled diagnostic accuracy with the 95% confidence interval (CI) was calculated using a bivariate random effects model. We also constructed hierarchical summary receiver operating characteristic curves and performed meta-regression analyses. Results Fourteen studies with reasonable methodological quality were included in the analysis. On a per-patient basis, the pooled sensitivity, specificity, positive likelihood ratio (PLR), negative likelihood ratio (NLR), and diagnostic odds ratio (DOR) were 0.89 (95% CI 0.80–0.94), 0.98 (95% CI 0.95–0.99), 40.4 (95% CI 19.7–82.6), 0.12 (95% CI 0.07–0.20), and 344 (95% CI 148–803), respectively. On a per-lesion basis, the pooled sensitivity, specificity, PLR, NLR, and DOR were 0.91 (95% CI 0.85–0.94), 0.97 (95% CI 0.95–0.98), 34.1 (95% CI 20.0–58.1), 0.10 (95% CI 0.06–0.16), and 358 (95% CI 165–778), respectively. In the meta-regression analysis, the clinical setting (staging vs. restaging) was the only source of study heterogeneity on a per-patient basis. Conclusions Choline PET/CT shows excellent diagnostic performance for the detection of bone metastasis. However, a negative choline PET/CT result cannot ensure the lack of bone metastasis.


Study selection
Two investigators independently screened titles and abstracts of all citations. Discrepancies were resolved by mutual agreement. We then reviewed the full text of these studies deemed relevant to determine eligibility. Studies were included based on the following criteria: (1) patients diagnosed with PC regardless of disease stage and treatment status, (2) 11C-choline or 18F-choline PET/CT used as the index test for detecting bone metastasis, (3) sufficient data to construct 2×2 contingency tables regarding sensitivity and specificity, (4) histopathological results and/or clinical follow-up served as the reference standard, and (5) publications written in English. The exclusion criteria were as follows: (1) case series with fewer than 10 patients; (2) insufficient data to construct contingency tables; (3) duplicated studies enrolling the same cohort; and (4) reviews, conference abstracts, case reports, and letters. In the case of an overlapping population, only the largest and most informative study was included.

Data extraction and quality assessment
Two investigators independently performed data extraction, and disagreements were resolved by consensus or consultation with a third reviewer. The following information was extracted using a standardized form: authors, publication year, country, study design, reference standard, blinding to reference standard, patient characteristics, clinical setting, prostate specific antigen (PSA) level, PET/CT characteristics, and absolute number of true positive, false positive, true negative, and false negative results for either patient-based analysis or lesion-based analysis. The authors of the eligible studies with inadequate data were contacted through email for additional information.
The quality of each study was independently appraised by two observers using the QUADAS-2 tool [18]. The QUADAS-2 tool assesses the risk of bias and applicability based on four domains: patient selection, index text, reference standard, and flow and timing. The provided signaling questions of the QUADAS-2 tool were used to reach a judgment as 'low', 'high', or 'unclear' rating.

Statistical analysis
The pooled summary estimates of sensitivity, specificity, positive likelihood ratio (PLR), negative likelihood ratio (NLR), and diagnostic odds ratio (DOR) were calculated. We used a bivariate random effects regression approach to synthesize data. This method estimated pairs of logit transformed sensitivities and specificities from studies following a bivariate normal distribution, incorporating both the between-study and within-study variability. Summary estimates of sensitivity and specificity were plotted in forest plots and hierarchical summary receiver operating characteristic (HSROC) curves with 95% confidence and prediction regions. A Spearman correlation coefficient of greater than 0.6 was considered to indicate a considerable threshold effect. Deeks' funnel plot was conducted to detect publication bias [19].
Heterogeneity was assessed using Cochran's Q test (p < 0.05 was considered significant) and the I 2 index (I 2 > 50% was considered substantial heterogeneity) [20]. We performed metaregression to investigate the potential source of heterogeneity within the included studies. The covariates included in the analysis were as follows: study design (prospective vs. retrospective), tracer (11C-choline vs. 18F-choline), clinical setting (staging vs. restaging), reference standard (histopathology or clinical follow-up vs. only clinical follow-up), diagnostic criteria (qualitative and semi-quantitative vs. qualitative), and blinding to reference standard (yes vs. no). Statistical analyses were performed using STATA version 12.0 (STATA Corporation, College Station, TX, USA). The association was considered statistically significant if the p value was less than 0.05.

Eligible studies and study description
The process of study selection is shown in Fig 1. The systematic search retrieved 760 articles after removing 231 duplicates. Among these articles, 57 articles were selected for reading of the Choline PET/CT for the detection of bone metastasis in prostate cancer full text. Finally, 14 studies [12][13][14][21][22][23][24][25][26][27][28][29][30][31] were included based on the inclusion and exclusion criteria. The general study characteristics are presented in Table 1. Six studies were analyzed on a per-patient basis, four studies were analyzed on a per-lesion basis, and four studies were analyzed on a per-patient basis as well as per-lesion basis. Five studies used "histopathology and/or clinical follow-up" as the reference standard, while the other nine used only clinical follow-up. The PET/CT characteristics are shown in Table 2. 11C-choline as a tracer was used in seven studies, and 18F-choline was used in the other seven studies. There was a wide variation in imaging protocols, particularly regarding the injection dose of tracers and the time from injection to scan.

Quality assessment
The summary of the quality assessment is illustrated in Table 3. With regard to the patient selection domain, four studies [12,13,24,30] were considered to have an unclear risk of bias because they did not explicitly mention whether patient recruitment was consecutive or not. High applicability concerns for patient selection domain were found in two studies because one study [13]  only included patients showing a single lesion on BS, and the other [27] excluded patients with more than four metastatic bone lesions. In terms of the index test domain, there was high risk of bias in two studies [12,24], as the interpretation of PET/CT was not blinded to the reference standard. There was no concern for applicability of the index test in all included studies. With regard to the reference standard domain, there was an unclear risk of bias in all studies except two [14,25] because it was unclear whether the reference standard assessments were blinded to the index test for most studies. The risk of bias for the flow and timing domain was judged as high in five studies [12,13,25,27,29] because different reference standards were applied within these studies. In general, the quality of the currently available studies was considered reasonable, with 12 of the 14 studies satisfying at least four of the seven QUADAS-2 domains. recorded no threshold effect (Spearman correlation coefficient = 0.457; p = 0.184). The forest plot of the sensitivity and specificity also revealed the lack of a threshold effect (Fig 2).  (Table 4). No threshold effect was shown (Spearman correlation coefficient = 0.357; p = 0.385). The coupled forest plot of the sensitivity and specificity also indicated no threshold effect (Fig 3). The heterogeneity was substantial with regard to sensitivity (Q = 56.72; p = 0.00; I 2 = 87.66%), and it was moderate with regard to specificity (Q = 12.95; p = 0.07; I 2 = 45.95%). The HSROC curves are presented in Fig 4, and the area under the

Exploration of heterogeneity
The results of the meta-regression analyses are shown in Table 5. On a per-patient basis, the clinical setting was likely the only source of study heterogeneity. Specifically, studies including only restaging PC patients reported a significantly higher specificity than those including only initial staging PC patients (0.99 vs. 0.91; p = 0); however, the pooled sensitivity estimates were not significantly different (0.87 vs. 0.95; p = 0.16). Upon analysis of the other covariates, study design, tracer, reference standard, diagnostic criteria and blinding to reference standard were not shown to be significant factors affecting the heterogeneity. We also performed a sensitivity analysis excluding a single study [25] that had a high risk of bias for the reference standard and showed a particularly low sensitivity of 0.50. The analysis yielded a lower degree of heterogeneity (I 2 = 0 and 46.12 for sensitivity and specificity, respectively), with a sensitivity of 0.90 (95%

Publication bias
The Deek's funnel plot asymmetry test suggested the presence of publication bias for perpatient basis (p = 0), while no publication bias was found for per-lesion basis (p = 0.95) (Fig 5).

Discussion
Previously, two meta-analyses have investigated the diagnostic accuracy of choline-PET/CT for detecting bone metastases in PC. Fanti et al. [15] conducted a meta-analysis of the literature published until December 2014 assessing 11C-choline PET/CT for its accuracy in the restaging  of patients. He reported that in all eight studies the overall detection rate for bone metastases was 25%. Unfortunately, in only four studies the data of sensitivity and specificity for skeletal metastases were reported, and with very high heterogeneity. Thus, it is not wise to draw hasty conclusion of diagnostic accuracy from this meta-analyses. In a meta-analysis from 2014, Shen et al. [16] summarized the literature on choline PET/CT, MRI, SPECT, and BS in detecting bone metastases for PC. Shen indicated that MRI was better than choline PET/CT and BS on a per-patient basis, and choline PET/CT was better than BS and SPECT on a per-lesion analysis. However, this meta-analysis has substantial shortcomings in its quantitative data analysis. It summarized pairs of sensitivity and specificity into a single measure of diagnostic accuracy. Thus, important information is missing, and furthermore, the researchers did not assess the heterogeneity of patients from different settings or other study-specific covariates. To retain the two-dimensional character, we used the bivariate random effects regression model to synthesize data, which was more likely to be scientific. In this meta-analysis, we demonstrated that choline PET/CT performs well as a diagnostic modality for assessing skeletal metastases in PC, with an area under the HSROC curve of 0.99 both on a per-patient basis and on a perlesion basis. Such adjacency of the area under the HSROC curve to 1 is a strong indicator of high diagnostic accuracy. According to our analysis, the pooled DOR values on a per-patient basis and on a per-lesion basis were respectively 344 and 358, also suggesting a high level of overall accuracy. Currently, given its low cost, easy availability and large clinical experience, 99mTc-phosphonates BS (99mTc-BS) is the most widely used agent for assessing bone metastases in PC despite its well-known limited diagnostic performance [9]. A recent meta-analysis by Treglia discovered that the discordance rate was 10.9% between choline PET/CT and BS in detecting bone metastases in PC [32]. Treglia considered that discordant findings were likely related to the different mechanism of uptake of radioactive tracers. 99mTc-phosphonates accumulate in osteoplastic lesions, which are the response to bone destruction and are not tumor specific [8]. Thus, metastatic bone lesions are identified indirectly by 99mTc-BS. Conversely, choline is a substrate for the synthesis of phospholipids that are necessary for the formation of cell membranes [33,34]. Accordingly, PET radiotracers such as 11C-choline or 18F-choline are supposed to target tumor cells directly [35]. In this study, choline PET/CT showed a pooled sensitivity of 89% and specificity of 98% to diagnose bone metastases in PC, which is superior to the reported performance of BS (i.e., pooled sensitivity and specificity of 71% and 91%, respectively) [36].
The optimum tracer for PET/CT in PC remains a matter of debate. Both 11C-choline and 18F-choline have been investigated. 11C-choline presents a shorter half-life (20 min) that makes its use limited to institutions with a cyclotron [37]. In contrast, 18F-choline is available to institutions without an onsite cyclotron because of its longer half-life (110 min), but the urinary excretion of 18F-choline is higher than 11C-choline, which may interfere with the interpretation of imaging findings in the pelvis [38,39]. A previous review reported that 11Ccholine and 18F-choline had similar diagnostic performance for malignant lesions in different clinical settings [40]. Our meta-analysis also found that both 11C-choline and 18F-choline PET/CT had excellent sensitivity (0.87 vs. 0.90) and specificity (0.98 vs. 0.97) for detecting bone metastases in PC patients. Our findings strengthen the current evidence for the use of PET/CT with 11C-choline or 18F-choline as tracers.
In the restaging setting, it is important to determine whether there is recurrent disease, lymph nodes or bone metastases, for the purpose of seeking an appropriate therapeutic planning [13]. Many researchers have demonstrated that choline PET/CT is useful for restaging PC, especially for detecting distant skeletal metastases. Fuccio et al. [41] detected a total of 30 bone lesions not revealed by BS in 18 of 123 restaging PC patients (14.6%) through 11C-choline PET/CT. Garcia et al. [27] found that choline PET/CT allowed early detection of bone metastases in 19.6% of restaging patients with negative BS results, thereby avoiding unnecessary treatment. Usefulness of choline PET/CT in initial staging setting is limited but encouraging. It was reported that choline PET/CT was not recommended for the initial diagnosis considering its low sensitivity in detecting primary lesions [42]. However, Evangelista's study had emerged that 18F-choline PET/CT could accurately stage PC patients with an intermediate to high risk of systemic disease [22]. A prospective study [43] demonstrated that based on choline PET/CT results, the therapy plan was changed from curative intent to palliative care in 20% of staging patients. The application of choline PET/CT in initial staging of PC patients warrants further investigation.
We further examined the diagnostic accuracy by calculating the PLR and NLR, which were more clinically meaningful than the HSROC and DOR. A PLR greater than 10 or an NLR less than 0.1 provide convincing evidence to rule in or rule out disease [44]. Our study revealed that the pooled PLR values on a per-patient basis and on a per-lesion basis were 40.4 and 34.1, respectively, which were high enough to verify bone metastases. At the same time, the pooled NLR values on a per-patient basis and on a per-lesion basis were 0.12 and 0.10, respectively. Hence, a negative choline PET/CT result may be insufficient to exclude bone metastases in PC. Lower 11C-choline uptake was observed in osteoblastic metastases compared to osteolytic lesions [45,46]. Beheshti et al. [12] confirmed that there was a significant correlation between tracer uptake and the density of malignant lesions with HU (Hounsfield unit) levels < 825 on CT. Besides, no choline uptake was detected in sclerotic bone lesions (with HU > 825). These performances, therefore, proves that the imaging may yield false-negative results. Moreover, it has been demonstrated that the PSA level significantly influenced the sensitivity of choline PET/CT [31,47]. Choline PET/CT may not be routinely indicated in case the serum PSA level rises < 1 ng/ml. Reported data showed a variable detection rate according to PSA level, ranging from 36% if PSA at relapse was lower than 1 ng/ml to 73% if PSA was higher than 3 ng/ml [25,48]. However, few papers included in this meta-analysis reported accurate data stratified by PSA values, which made it impossible to deliver a reasonable comparison. The recent introduction of gallium-68 prostate specific membrane antigen (Ga-68 PSMA) as a PET tracer might further improve results [49]. Ga-68 PSMA PET/CT has a detection rate of 50% and 68%, respectively for PSA levels < 0.5 ng/ml and 0.5-2 ng/ml [50]. However, its use is restricted due to limited availability and high costs.
There was significant heterogeneity among the included studies. The meta-regression analysis indicated that the clinical setting may be the only source of heterogeneity on a perpatient basis. Compared with studies including only initial staging patients, studies including only restaging patients showed a significantly higher specificity (0.99 vs. 0.91; p = 0) and a tendency for lower sensitivity (0.87 vs. 0.95; p = 0.16). We hypothesized that if patients have already received systemic therapies, the imaging features of malignant bone lesions may change. As mentioned above, Beheshti et al. [46] found that no choline uptake was detected in densely sclerotic bone lesions, almost all of which were observed in patients with hormone therapy. Another study [51] reported a significant reduction of choline uptake following androgen deprivation therapy in androgen-sensitive patients with recurrent PC. The presence of systemic therapies may cause false-negative PET/CT findings and affect diagnostic performance. However, Picchio et al. [21] documented that the accuracy of 11C-choline PET/CT for detecting skeletal metastasis in hormone-resistant patients did not significantly differ from patients who did not receive anti-androgenic treatment. A similar phenomenon was described by Kitajima1 et al [26]. In our study, we were unable to perform separate analyses based on the type of treatments, as the separate diagnostic performance values could not be extracted from the included studies. The impact of systemic therapies prior to choline PET/CT scanning remains unclear.
Among the included studies, only a few used "histopathology and clinical follow-up" as the gold standard, while most relied on clinical follow-up and conventional imaging modalities to confirm the existence of bone metastasis. Different reference standards might be an important source of heterogeneity, although our analysis suggested no significant difference among the subgroups. Furthermore, other covariates, including study design, tracer, diagnostic criteria and blinding to the reference standard, were not shown to be significant factors influencing the heterogeneity. However, in one study [25], all negative PET/CT scans were considered false negatives due to a PSA rise, except in two patients, who had a negative 24-month followup. This study showed an unusually inferior sensitivity of 0.50 and was considered to have a high risk of bias for the reference standard. Therefore, a sensitivity analysis was performed excluding this single study as a result, we achieved consistent diagnostic performance with substantially decreased heterogeneity, with a sensitivity of 0.90 (95% CI 0.84-0.94, I 2 = 0) and a specificity of 0.98 (95% CI 0.94-0.99, I 2 = 46.12).
This meta-analysis has several limitations [52]. First, the lack of a well-accepted gold standard may have affected the evaluation of choline PET/CT. The gold reference standard for any diagnostic study is histological confirmation of the findings. Nevertheless, in clinical practice, we could not perform a biopsy on each lesion. In this meta-analysis, we had to use "histopathology and/or clinical follow-up" as the suboptimal reference tests. Another limitation is that we detected publication bias on a per-patient basis. Generally, studies with desirable results may be more likely to be published than those with neutral or unfavorable results [53]. To minimize the possibility of publication bias, we searched the databases and reference lists of included articles again for further potential studies, but we could not obtain additional relevant publications. In addition, limiting the search to the English language and the exclusion of abstracts, case reports, letters, and comments may have also produced potential publication bias. Furthermore, the characteristics of the clinical variables among these selected studies, such as PSA level, technical parameters and measurements, were heterogeneous, making a stratified analysis for different risk groups impossible. Finally, caution is needed when applying our results because most of the included studies were from the United States and Europe, and only one Asian study with a relatively small sample size was included.
In conclusion, this systematic literature review and meta-analysis demonstrated that choline PET/CT had excellent sensitivity and specificity for the detection of bone metastasis in PC patients, both on a per-patient basis and on a per-lesion basis. However, a negative choline PET/CT result could not ensure the lack of bone metastasis.