Diagnostic Role of Four-Dimensional Computed Tomography for Preoperative Parathyroid Localization in Patients with Primary Hyperparathyroidism: A Systematic Review and Meta-Analysis

We sought to systematically evaluate diagnostic performance of four-dimensional computed tomography (4D-CT) in the localization of hyperfunctioning parathyroid glands (HPGs) in patients with primary hyperparathyroidism (pHPT). We calculated the pooled sensitivity, specificity, positive likelihood ratio (PLR), negative likelihood ratio (NLR), and diagnostic odds ratios (DOR) of 4D-CT on a per-lesion level, as well as pooled sensitivity and positive predictive value (PPV) on a per-patient level with 95% confidence intervals (CIs). Additionally, we plotted summary receiver operating characteristic (SROC) curves and evaluated the areas under the curves (AUC). A total of 16 studies were included in the analysis. Their pooled sensitivity, specificity, PLR, NLR, and DOR of 4D-CT on per-lesion level were 75% (95%CI: 66–82%), 85% (95%CI: 50–97%), 4.9 (95%CI: 1.1–21.3), 0.30 (95%CI: 0.19–0.45), and 17 (95%CI: 3–100), respectively, with an AUC of 81% (95%CI: 77–84%). We also observed heterogeneity in sensitivity (I2 = 79%) and specificity (I2 = 94.7%), and obtained a pooled sensitivity of 81% (95%CI: 70–90%) with heterogeneity of 81.9% (p < 0.001) and PPV of 91% (95%CI: 82–98%) with heterogeneity of 80.8% (p < 0.001), based on a per-patient level. Overall, 4D-CT showed moderate sensitivity and specificity for preoperative localization of HPG(s) in patients with pHPT. The diagnostic performance may improve with 4D-CT’s promotion to first-line use on a lesion-based level, further research is needed to confirm the results.


Introduction
Primary hyperparathyroidism (pHPT) is a common endocrine disorder, defined as hypercalcemia secondary to excessive secretion of parathyroid hormone by 1 or more hyperfunctioning parathyroid glands (HPGs) [1,2]. It more commonly affects elderly individuals, and women four times as often as men [3,4]. A single parathyroid adenoma (85-90%) is the most common cause of pHPT [5]. Parathyroidectomy (PTx) represents the best line of action for management of parathyroid adenoma(s). It is advised for patients with systematic pHPT and those who meet the surgery guideline criteria. On the other hand, it is always an option for some patients with asystematic pHPT, even if they do not meet any criteria for surgery [6,7]. In addition, focused parathyroidectomy and bilateral neck exploration (BNE) have been shown to result in analogous biochemical complications and cure rates. Nevertheless, focused parathyroidectomy has some advantages, including decreased operating time, length of hospital stay as well as reduced medical cost and

Data Extraction and Quality Assessment
Basic information, namely first author name, publication date, country, study design, and patients' characteristics, as well as technical aspects, such as machine model, detector, product, injection rate, dose, scan phase, and imaging procedure, were grouped together. Each included study was analyzed to obtain the numbers of true positives (TP), false positives (FP), true negatives (TN), and false negatives (FN) of per-lesion or per-patient for localization of HPG(s). We applied the Quality Assessment of Diagnostic Accuracy Studies 2 (QUADAS-2) tool to assess studies quality. Each article was reviewed by two reviewers (L.S. and P.H.) and any discrepancies were resolved by discussion. The result was judged as true positive if 4D-CT localized the correct position of the HPG(s) (upper/lower pole of the thyroid left/right lobe, upper/lower part of the mediastinum, or elsewhere) at surgery on a per-lesion level, and if patients with one or several HPG(s) detected on imaging and confirmed by surgery findings on a per-patient level.

Statistical Analysis
The primary purpose of this analysis was to calculate summary sensitivity, specificity, positive likelihood ratio (PLR), negative likelihood ratio (NLR) and diagnostic odds ratios (DOR) with 95% confidence interval (CI) on a per-lesion level for the localization of HPG(s). A DOR can be calculated as the ratio of the odds of positivity in a disease state, relative to the odds of positivity in the non-disease state, with higher values indicating better discriminatory test performance [17]. We applied a bivariate random effects model to calculate the pooled sensitivity and specificity on a per-lesion level, then used the same model to plot summary receiver operating characteristic (SROC) curves and evaluate the areas under the curves (AUC). We only calculated pooled sensitivity and positive predictive value (PPV) on per-patient level, since some articles did not report FP and TN findings.
We assessed between-study heterogeneity of the data using the I-square index (I 2 ) statistic and the Cochrane Q test, based on random-effects analysis [18]. To assess studybetween heterogeneity, we performed the following subgroup analyses: (1) imaging procedure (4D-CT as first line vs. second line examination) on both analyses; and (2) parathyroid glands analysis (all glands vs. resected glands) on lesion-based level. Publication bias was evaluated by Deek's test as previously described [19]. Data analyses were performed using the "Midas" and "Metaprop" modules in Stata software version 15.0 (StataCorp, College Station, TX, USA). Values were considered statistically significant if the two-sided p value was <0.05.

Characteristics of the Included Studies
Basic characteristics of the 4D-CT studies are summarized in Table 1. Detailed technical aspects, including CT, as well as contrast and injection information, scan phase and imaging procedure are listed in Table 2. Among the 16 included studies, five described lesion-based analysis, seven reported patient-based analysis, whereas four were on both of the aforementioned analyses. Overall, these studies comprised a total of 1032 patients. Moreover, nine studies, including 710 patients and 2644 lesions on lesion-based analysis, and 11 studies comprising 503 patients on patient-based analysis allowed evaluation of 4D-CT's role in localizing HPG(s), the diagnostic accuracy of 4D-CT are shown in Table 3. Of the 16 included studies, 13 and two had a retrospective and prospective design, respectively, whereas one study did not report the study design. Two studies focused on the group of patients operated on the second time or experiencing recurrence of the tumor, while the remaining 14 enrolled participants either without a history of prior neck surgery or with mixed type of the first and re-operation. Six and nine studies reported 4D-CT as a first-and second-line of examination, respectively, whereas one reported both examinations.

Characteristics of the Included Studies
Basic characteristics of the 4D-CT studies are summarized in Table 1. Detailed technical aspects, including CT, as well as contrast and injection information, scan phase and imaging procedure are listed in Table 2. Among the 16 included studies, five described lesion-based analysis, seven reported patient-based analysis, whereas four were on both of the aforementioned analyses. Overall, these studies comprised a total of 1032 patients. Moreover, nine studies, including 710 patients and 2644 lesions on lesion-based analysis, and 11 studies comprising 503 patients on patient-based analysis allowed evaluation of 4D-CT's role in localizing HPG(s), the diagnostic accuracy of 4D-CT are shown in Table 3. Of the 16 included studies, 13 and two had a retrospective and prospective design, respectively, whereas one study did not report the study design. Two studies focused on the group of patients operated on the second time or experiencing recurrence of the tumor, while the remaining 14 enrolled participants either without a history of prior neck surgery or with mixed type of the first and re-operation. Six and nine studies reported 4D-CT as a first-and second-line of examination, respectively, whereas one reported both examinations.

Quality Assessment
Results from methodological quality analysis, including patient selection, index test, reference standard, and flow and timing, are summarized in Figure 2. With regards to patient selection, nine studies showed an unclear risk of bias due to insufficient information of consecutive patient enrollment and time limitation, while one study revealed high risk of bias due to exclusion of multiple parathyroid glands. Based on index test, seven studies exhibited an unclear risk of bias owing to the fact that they did not mention whether or not the operators interpreted the images without reference standard, which might result in interpretation bias. We considered one study as high risk of bias, since not all interpretation of images were blinded. All studies revealed an unclear risk of bias in reference standard, owing to the fact that they did not mention whether or not the pathologist were blinded to prior clinical and imaging data, when they interpreted the histological results, which might result in verification bias. However, most of the studies were considered to have low applicability in the patient selection, index test and reference standard domains, therefore, we took all the studies into final analysis.

Summary of 4D-CT's Diagnostic Performance
We regarded histological findings, as well as a combination of these and follow-up of biochemical resolution after surgery, as the reference standard. On a per-lesion analysis, sensitivity and specificity were used to assess the diagnostic performance. Conversely, on a per-patient analysis, the terms specificity and negative predictive value were not meaningful, since diagnosis was biologically confirmed for patients who were all assumed to have at least 1 HPG. Consequently, we only calculated pooled sensitivity and PPV as the metrics of diagnostic accuracy on a per-patient basis.

Summary of 4D-CT's Diagnostic Performance
We regarded histological findings, as well as a combination of these and follow-up of biochemical resolution after surgery, as the reference standard. On a per-lesion analysis, sensitivity and specificity were used to assess the diagnostic performance. Conversely, on a per-patient analysis, the terms specificity and negative predictive value were not meaningful, since diagnosis was biologically confirmed for patients who were all assumed to have at least 1 HPG. Consequently, we only calculated pooled sensitivity and PPV as the metrics of diagnostic accuracy on a per-patient basis.
Further, Deek's test showed no evidence of publication bias (p = 0.19 and p = 0.83) on a lesion-and patient-based level, the figures of funnel plot were supplemented in (Supplementary Figures S1 and S2).

Discussion
The current meta-analysis revealed that 4D-CT has moderate diagnostic accuracy in patients with pHPT to a correct quadrant, based on both per-lesion and per-patient level, as evidenced by moderate pooled sensitivity, specificity and AUC. Moreover, subgroup analyses revealed that 4D-CT may improve diagnostic performance, as a first-line modality for in localizing patients with pHPT on a lesion-based level.

Discussion
The current meta-analysis revealed that 4D-CT has moderate diagnostic accuracy in patients with pHPT to a correct quadrant, based on both per-lesion and per-patient level, as evidenced by moderate pooled sensitivity, specificity and AUC. Moreover, subgroup analyses revealed that 4D-CT may improve diagnostic performance, as a first-line modality for in localizing patients with pHPT on a lesion-based level.
Wan et al. have reported the comparison of diagnostic value between 4D-CT and SPECT/CT head-to-head, their sensitivity and specificity of 4D-CT were slightly higher than the results of this study in a lesion-based analysis [15]. However, there are several methodological differences between the two meta-analyses. First, they only selected studies investigating 4D-CT with SPECT/CT head-to-head. Second, they performed meta-analysis on a lesion-based level without calculations based on a per-patient level. In an attempt to obtain more information evaluating the diagnostic value of 4D-CT, we included more studies. Moreover, we performed subgroup analysis to observe the effectiveness of 4D-CT for localization of HPG(s) in pHPT patients in different clinical settings.
With regards to lesion-based analysis, we found four studies that showed lower sensitivities (from 50 to 75%). Among them, one adopted a prospective and consecutive design, with a resulting sensitivity of 58%, which was in contrast to many other findings [20,32,39]. This might be, in part, due to the study design. The other three articles also demonstrated a relatively lower diagnostic accuracy, in which 4D-CT played a role as a second-line imaging modality for localizing HPG(s). Based on these findings, we performed subgroup analyses on study design and imaging procedure.
Subgroup analysis, by imaging procedure performed targeting lesions and patients, showed that second-line modality had lower diagnostic performance than first-line, with regards to both analyses for localization of HPG(s). This might be attributed to imaging results and the inexperience technicians operating inconvenient shooting techniques. Other factors may also change the effectiveness of 4D-CT, including positioning, large body habitus and position of HPG(s) [28,40,41]. Patients subjected to 4D-CT as a second-line examination always had negative or inconclusive first-line imaging results. Therefore, it was easy to increase the number of false-negatives. Of note, lesions in subgroup of 4DCT used as first-line imaging are significantly larger than those of second-line, the results of this group might be more convincing.
We did not perform subgroup analysis on prior neck surgery (yes vs. no) and scan phase (three phases vs. four phases), due to the limited number of history reports. In ad-dition, among 16 included studies, there were only two studies reporting the diagnostic performance of 4D-CT for localization of pHPT patient with low/mild line hypercalcemia or parathyroid hormone (PTH), two studies focusing on persistent or recurrent pHPT, the published data was insufficient to evaluate the special group of patients. Notably, it is better to choose a protocol that allows minimal radiation dose without sacrificing diagnostic accuracy. For example, Raghavan et al. [42] demonstrated that arterial phase images (with a sensitivity of 91% and specificity of 82.9%) alone was adequate to obtain the diagnostic accuracy for parathyroid adenomas localization. In addition, Campbell et al. [43] suggested the two-phase computed tomography is as effective as 4D-CT for identifying enlarged parathyroid glands. Further research is needed to ascertain an ideal CT protocol.
This meta-analysis had some limitations. Firstly, as shown by the QUADAS 2 tool, an unclear and a high risk of bias, arising from patient selection, index text and reference standard, should be took into consideration. It was difficult to avoid patient selection bias owing to the fact that only two studies had a prospective design, while a majority of the rest were retrospective. Furthermore, whether reviewers/pathologists were blinded to histological/imaging results or not can change the diagnostic accuracy to some extent. The results of the current meta-analysis should therefore be interpreted with caution. Secondly, some studies only exhibited diagnostic accuracy based on a single lesion, and excluded patients with multiple parathyroid glands. This could have lowered sensitivity. Thirdly, not all reference standards combined histological findings with follow up. Fourthly, a previous study reported that radio-labelled choline PET showed excellent diagnostic performance in the detection of HPG(s) in HPT patients. However, in our case, only two studies reported diagnostic accuracy of 4D-CT and F18 choline-PET/CT on lesionbased analysis, which was not enough to evaluate the diagnostic performance of the two imaging modalities.

Conclusions
Overall, 4D-CT exhibited moderate sensitivity and specificity in both patient-and lesion-based analysis of preoperative localization of HPG(s). Its diagnostic performance may improve when it was used as a first-line modality, nevertheless, necessitating further investigation to compare the diagnostic accuracy of 4D-CT with other imaging modalities to confirm our observation.

Supplementary Materials:
The following are available online at https://www.mdpi.com/article/ 10.3390/diagnostics11040664/s1, File S1: details of search algorithm for each electronic literature databases; Figure S1: Deeks funnel plot asymmetry test for publication bias of 4D-CT for detecting HPG(s) on a per-lesion level. Figure S2

Informed Consent Statement: Not applicable.
Data Availability Statement: All analyses were based on previously published studies. Data sharing is not applicable to this article.

Conflicts of Interest:
The authors declare no conflict of interest.