Patients and follow-up
Patients were selected from a validated quality assurance database at the Department of Gynecological Cancer at the Oslo University Hospital. The database covers all patients treated at Oslo University Hospital (The Norwegian Radium Hospital, Ullevål University Hospital and Rikshospitalet) for endometrial cancer. The database is linked to Statistics Norway and individual survival data are available through this linkage. The database provides detailed information on the primary diagnosis, the preoperative work-up, comorbidity, surgical treatment, adjuvant treatment, incident relapse, localization of relapse, and date of death. We included all patients with endometrioid endometrial cancer stage I, treated at the Oslo University Hospital between 2005 and 2012. Histopathological diagnosis was confirmed by review of the hematoxylin-eosin slides by three surgical pathologists (BD, BR, BK) specialized in gynecologic pathology at the Norwegian Radium Hospital. Patients with synchronous ovarian cancer were excluded (n = 33). For this study the optimal slide from each individual case was selected, based on the presence of sufficient amount of tumor, good fixation and the presence of normal myometrium as a control.
Patients were staged according to the 2009 FIGO classification and categorized as low (stage IA, grade 1 and 2), intermediate (stage IA grade 3, or stage IB grade 1 and 2), or high risk (stage IB grade 3). They were treated with surgery and adjuvant chemotherapy according to the national treatment policy. In general, low risk patients were treated with extrafascial hysterectomy and bilateral salpingo-oophorectomy alone, while intermediate and high-risk patients underwent lymph node staging. After surgery, high-risk patient were offered systemic treatment with platinum-based chemotherapy. Patients were followed every 3 months for the first 2 years and every 6 months for the next 3 years. Visits included thorough clinical examination and vaginal ultrasound, supplemented by CT or MR scan on clinical indication.
Immunohistochemical staining and evaluation of expression
The formalin fixed paraffin-embedded (FFPE) tissue blocks were collected and cut into 3-4μm sections and mounted on Superfrost slides. The sections were analyzed for L1CAM protein expression using the Dako FLEX+ protocol (Dako, Glostrup, Denmark). The L1CAM antibody was a mouse monoclonal antibody (clone 14.10, cat. # SIG-3911) from Covance (Princeton NJ), applied at 1:300 dilutions. Antigen retrieval was performed in low pH buffer (Dako). Visualization was achieved using 3′3-diaminobenzidine tetrahydrochloride substrate (DAB) and hematoxylin counterstaining. Positive control consisted of a high-grade serous carcinoma shown to be positive in antibody testing, and was satisfactory in all reactions. Negative controls consisted of slides stained with IgG1k murine melanoma immunoglobulin (Sigma-Aldrich, St. Louis MO; cat. # M9035) at the same concentration.
Staining extent was scored as positive (>10 % of cells) (Fig. 1) vs. negative (≤10 % of cells). This cut-off has previously been reported to result in the strongest model and was confirmed by Bosse et al. [2, 3]. Scoring was performed by one gynecologic pathologist (BD), which was blinded for clinical outcome.
Statistical analysis
This study is a ”prospective-retrospective” design that used archived tumor specimen as suggested by Simon et al. [4]. Our power calculation was based on results reported in the literature. Relapse rates of 3 % in L1CAM negative tumors and 50 % in L1CAM positive tumors have been reported [3]. Given a statistical power of 80 % and alpha of 5 % we would have needed 17 patients in each group and 10 relapses in total. Further, an absolute difference of 37 % in the number of deaths has been reported previously [3], with 40 % death rate in the L1CAM positive group and 3 % death rate in the negative group. With the same statistical assumptions as above (HR = 3), we would have needed 24 patients in each group and 10 deaths to replicate the results. Therefore, given the sample size available and the above stated assumptions, our study was sufficiently powered.
Continuous variables were descried as median and range. Categorical variables were presented with counts and proportions. Crude associations between L1CAM negative and L1CAM positive patients and categorical variables were assessed with χ2 test.
When studying the risk of relapse, follow-up time was calculated from the date of EC diagnosis until date of relapse, date of death from any cause or end of follow-up, August 31, 2014, whichever occurred first.
For risk of death, follow-up time was calculated from the date of EC diagnosis until date of death from any cause or end of follow-up, whichever occurred first. Survival curves were plotted with the Kaplan-Meier method. The log rank test was used to compare survival between the groups. Crude hazard ratios (HRs) of relapse and death with 95 % confidence intervals (CI) associated with L1CAM overexpression were calculated using Cox proportional hazard models. The proportionality assumption was tested using Schoenfeld’s residuals. All variables revealing prognostic significance in the univariate analysis or previously have been reported to be associated with risk of relapse or death, were included in the multivariable model. The final multivariate model was adjusted for FIGO stage and age < vs ≥ 60 years. An alternative model was adjusted for age as attained age at the diagnosis of endometrial cancer and contained L1CAM status, FIGO stage and grade. However the estimates remained unchanged. In exploratory subgroup analysis by treatment with adjuvant chemotherapy or not, we studied the association of L1CAM expression with the risk of relapse. P-values <0.05 were considered statistically significant and all tests were two-sided.
The analyses were performed using IBM SPSS version 22 (SPSS, Chicago, IL) and the STATA statistical package, version 11.0, (Stata Corp LP, Texas, USA).