Results
We yielded a total of 1261 publications by database searches, of which 131 studies were identified after exclusion of reduplicative and irrelevant studies. After abstracts screening, 73 unique studies remained, and all full texts were obtained. After reviewing of the full texts, we excluded 44 studies. The reasons for each exclusion of these studies were listed in supplementary (p8–9). Besides, we obtained five reports of unpublished data and two of them were eligible (supplementary p15–16). Finally, 31 eligible studies (29 publications) were included. Figure 1 displays the PRISMA (2009) flow diagram for study selection.
The publication years of the 31 identified studies were between 2002 and 2019. Most of these studies were done in Europe, Asia, and the USA. A total of 1887 eligible patients were identified from the 31 studies. Each of them underwent SLNB for cervical cancer, with at least one SLN detected and examined by intraoperative FSE.
The characteristics of the 31 finally included studies were listed in supplementary (p17). Patients with early-stage disease accounted for 66.1–100% of whole populations, with 19 out of 31 (61.3%) studies having 100% patients at early stage. The most common histological types were squamous cell carcinoma, followed by adenocarcinoma. The rate of nodal metastasis varied between 5.4–36.4% among the included studies.
The surgical approaches were affected by the FSE results in 18 studies, of which 13 performed additional para-aortic lymphadenectomy in positive cases. Only four studies omitted PLND in cases of negative FSE. Four studies applied negative FSE as a prerequisite of fertility preservation. Eight studies transferred radical surgeries to concurrent chemoradiotherapy in positive cases. Besides, one study performed simple hysterectomy in cases of negative FSE. Twelve studies reported follow-up results and the oncological outcomes were generally good (supplementary p22).
The sectioning protocols of FSE consisted of three different approaches as following:
  1. SLN was bisected, one section was taken from the maximum surface of one half SLN (9 studies);
  2. SLN was bisected, adjacent sections were taken from the maximum surface of each half SLN (4 studies);
  3. SLN was cut into pieces at certain intervals (varied from 2 to 5 mm, perpendicularly to their long axis), and one or more sections were taken from each piece (13 studies).
For subgroup analysis, we defined the two “bisected” protocols together as equatorial (E), and the third one as latitudinal (L) (figure 2). The next steps were similar in which the sections were examined after staining with hematoxylin and eosin (H&E) and the rest tissue were embedded in paraffin. For DPE, additional ultrastaging were performed in 24 studies, while routine pathological examination was performed alone in seven. Among the 24 studies using ultrastaging, serial sectioning combined with immunohistochemistry (IHC), serial sectioning alone, and IHC alone were performed in 18, 2 and 4, respectively.
In published pathological studies at least four further step sections combined with IHC examination were recommended for SLN ultrastaging.38 We utilized this criterion to evaluate the stringencies of ultrastaging techniques and their potential influences. DPEs meeting this criterion were judged as adequate, otherwise as inadequate. The descriptions and judgements of FSE and DPE protocol for each study were presented in supplementary (p20–21).
The process of data extraction for meta-analysis were detailed in supplementary p10–16. Overall, there were 363 patients having SLN metastases confirmed by DPE, of whom 115 were misdiagnosed by FSE. The sensitivities of FSE varied over a wide range of 0%–100% among the 31 included studies (table 1). Specificities were 100% in all studies except two reporting false positive. Both were due to misdiagnosis of endosalpingiosis in SLN.6,30Pooled analysis using mixed-effect model yielded an estimated rate of 0.77 (95%CI: 0.66–0.85, figure 2) for sensitivity. Heterogeneity test for sensitivities showed high heterogeneity among the included studies (Q=99.09,I 2=69.73%, p<0.001).
Among 26 studies with definable pathological protocols, 19 reported FN results. The metastatic types of FNs were available in 18 studies, including 24 MAM, 51 MIM, and 29 ITC in 101 patients (table 2). The metastatic types of TPs were available in nine studies, including 95 MAM, 14 MIM, and one ITC in 110 patients (table 3). The pooled sensitivity of FSE were 0.79 (95%CI: 0.70–0.86) if ITC were not considered, and 0.94 (95%CI: 0.85–0.98) if only MAM was considered. Notably, only four of the 24 (16.7%) FN-MAM were missed, whereas 13 of the 15 (86.7%) TP-MIM/ITC were detected under L-protocol.
We conducted meta-regressions to investigate the source of heterogeneity in sensitivities. Studies were categorized into subgroups by the study design (prospective and retrospective), SLNB strategy (whether the surgical approaches were affected by FSE results: yes and no), sample size (<60 and ≥60), overall metastatic rate (<20% and ≥20%), reference standard (DPE protocol: adequate and inadequate), and index test (FSE protocol: E and L). Five studies were excluded from meta-regression due to undefinable methodologies. Finally, FSE protocol was found to be the only source of heterogeneity (p<0.001, table 4).
Subgroup analysis showed decreased heterogeneities in both subgroups stratified by FSE protocol (Q=7.59, I 2=0%, p=0.82 for L-protocol; Q=28.90, I 2=58.47%, p<0.001 for E-protocol). The sensitivity pooled achieved 0.86 (95%CI: 0.79–0.91) in the L-protocol subgroup (13 studies, 650 patients), whereas it was 0.59 (0.46–0.72) in the E-protocol subgroup (13 studies, 1047 patients). The difference reached statistical significance (P<0.001). If ITC was not considered, the pooled sensitivities would be 0.88 (0.81–0.93) and 0.64 (0.52–0.75) for L and E-protocol subgroup, respectively (p<0.001). If only MAM was considered, the pooled sensitivities would be 0.97 (95%CI 0.89–0.99) and 0.86 (0.74–0.93) for L and E-protocol subgroup, respectively (p=0.01). In sensitivity analyses (supplementary p23–27), whatever the alterations made in study setting or statistical model, the observation that sectioning protocol determined the accuracy of FSE, remained unchanged.