Percutaneous Preoperative Biliary Drainage for Resectable Perihilar Cholangiocarcinoma: No Association with Survival and No Increase in Seeding Metastases

Consecutive patients who underwent a resection with curative intent for PHC were identified from prospectively maintained databases at the Academic Medical Center (AMC) in Amsterdam, the Netherlands, and the Memorial Sloan Kettering Cancer Center (MSKCC), New York. PHC was defined according to the 7th edition of the American Joint Committee on Cancer staging manual.15 Patients were included from 1991 to 2012 if they had undergone preoperative biliary drainage before resection of PHC using extrahepatic bile duct resection and reconstruction with or without concomitant liver resection. Exclusion criteria were R2 resection, repeat resection after initial resection at another hospital, and 90 day postoperative mortality. Additional data were collected through retrospective chart review. The institutional review board at both institutions approved this study.

Patient selection for resection and preoperative management was similar between the 2 centers, as described previously.16 Biliary drainage was initiated in either a regional center before referral, or after referral to AMC or MSKCC. Patients were treated with initial EBD or initial PTBD according to the treating physician’s preference. Additional preoperative PTBD was performed when biliary decompression was inadequate after EBD or if EBD was associated with complications, such as cholangitis. The PTBD group in this study included patients treated with initial PTBD and patients treated with additional PTBD after inadequate EBD. The control group consisted of patients treated with preoperative EBD without previous or subsequent PTBD.

All patients in the AMC in Amsterdam were routinely treated with a preoperative low-dose irradiation protocol (3 × 3.5 Gy in the 3 days before the resection) with the aim of preventing seeding metastases.17 Entry sites of percutaneous drain tracts in the abdomen were not routinely excised after resection.

All patients were followed with CT imaging at 3 and 6 months after resection to detect early recurrence. Thereafter, patients at AMC were followed with clinical assessments until 5 years after resection; imaging was performed when indicated by rising tumor markers, symptoms, or findings at physical examination. At MSKCC, follow-up included CT or MRI imaging every 4 to 6 months. Pathologic confirmation of recurrences was often obtained, but it was not required if imaging unambiguously demonstrated recurrent disease in patients who were unfit to undergo further treatment. Suspect lesions at the laparotomy scar or in the prior PTBD drainage tract were always confirmed with a biopsy.

The primary end point in this study was OS, measured from the date of surgery to the date of death. Patients alive at follow-up were censored at the date of last contact before April 1, 2014. We used propensity score method rather than traditional multivariable analysis because this method is considered superior in reducing confounding and bias, especially when analyzing relatively small observational data sets.18,19

Secondary end points were directed toward the incidence of seeding metastases after resection. Analysis of recurrences focused on the pattern of initial recurrences based on the assumption that prognosis was unlikely to be affected by seeding metastases that arose after recurrence at another site. Seeding metastases were defined as recurrences either in the percutaneous catheter tract (i.e., any recurrence along the catheter tract from skin to the intrahepatic bile duct) or in the laparotomy scar (i.e., any recurrence in the abdominal wall at the laparotomy scar).6,8 In addition, the incidence of peritoneal recurrences (i.e., intra-abdominal recurrence in the peritoneum or ascites with malignant cells) was assessed, although these recurrences were not necessarily regarded as seeding metastases.

Our method of propensity score adjustment was straightforward. First, we estimated propensity scores for the probability of PTBD assignment on the basis of all observed baseline characteristics. Second, we analyzed OS with a Cox proportional hazards model including 2 variables: drainage method (PTBD vs. EBD only) and propensity score (continuous variable). This model adjusts the survival analysis conditional on the propensity score. Thus, the model calculates the effect of PTBD compared to EBD only given that the propensity scores (i.e., the observed baseline characteristics) are hold equal.

In more detail, we calculated propensity scores using multivariable logistic regression with preoperative PTBD as the outcome of interest and with adjustment for observed baseline characteristics, including demographics, comorbidities, total bilirubin level at referral, the level of bile duct involvement (Bismuth class), preoperative imaging variables (Blumgart T stage), cholangitis, extended hepatectomy, and treating center. Three baseline characteristics had missing data, including bilirubin level at presentation (27.8 % missing), Blumgart T stage (6.9 % missing), and preoperative cholangitis (5.3 % missing). To avoid bias, multiple imputation with 10 imputed data sets was performed for these missing data before estimation of the propensity scores, using a regression model that included all baseline characteristics. To evaluate residual bias after adjustment for propensity score, logistic regressions with drainage method as outcome were performed for each of the baseline characteristics with and without adjustment for propensity scores. We then estimated OS using the Kaplan–Meier method, and compared the groups with the log-rank test in univariable analysis. Finally, a multivariable Cox proportional hazards model was used to compare OS between the PTBD and EBD-only groups after adjustment for the propensity score as a continuous variable. To assess the proportional hazards assumption, we inspected the hazard ratio plots and found no violation.

Analysis of secondary end points was performed by χ ² tests, and 95 % confidence intervals (CIs) were determined using the standard deviation of the mean. The type of liver resection and pathologic characteristics were also compared by χ ² tests. All analyses were performed in SPSS v22 (IBM, Armonk, NY).

Baseline characteristics of the study groups are presented in Table 1. Nearly all variables were different between groups, indicating severe bias at baseline. The mean ± standard deviation propensity scores for patients in the PTBD and EBD-only groups were 0.53 ± 0.24 and 0.27 ± 0.19, respectively, with an area under the curve of 0.79. Only minimal differences in baseline characteristics remained after adjustment for propensity score, as indicated by the adjusted P values in Table 1. Type of resection and pathologic characteristics of both study groups are shown in Table 2.

Among the 245 included patients, 173 patients (71 %) died during follow-up. The median OS was 38 months (95 % CI 32–44), and 5-year survival was 32 %. Median follow-up among survivors was 52 months (range 6–251 months). The unadjusted OS was comparable between the PTBD group (36 months) and the EBD-only group (41 months; P = 0.25; Fig. 1a). Stratifying patients in the PTBD group between those who underwent EBD plus PTBD and those who underwent PTBD only did not reveal a difference when these 2 groups were compared to patients who underwent EBD only (P = 0.44; Fig. 1b). After using propensity score adjustment to account for potential confounders, OS between the PTBD group and EBD-only group was similar (adjusted hazard ratio, 1.05; 95 % CI 0.74–1.49; P = 0.80; Fig. 2).

A total of 87 patients in the PTBD group and 147 patients in the EBD-only group were available for recurrence analysis (1 patient missing in the PTBD group and 10 patients missing in the EBD-only group; total 4 %). Seeding metastases occurred as initial recurrence in 3 (3.4 %) of 87 patients in the PTBD group (95 % CI 0–7.3), and in 4 (2.7 %) of 147 patients in the EBD-only group (95 % CI 0–5.3; P = 0.71). Among the total 7 patients who developed a seeding metastasis, 3 had developed a concurrent local recurrence. Time to diagnosis of the seeding metastases was 8, 13, 14, 17, 20, 21, and 66 months (median, 17 months), and OS in these 7 patients was 13, 30, 20, 21, 21, 27, and 142 months (median, 21 months), respectively. All 7 seeding metastases developing as initial recurrence were abdominal wall recurrences at the site of the laparotomy scar. No initial recurrences were observed in a percutaneous catheter tract. The incidence of seeding metastases was not significantly different between both centers: 5 (3.9 %) of 128 patients at MSKCC (95 % CI 0.5–7.3) and 2 (1.9 %) of 106 patients at AMC (95 % CI 0–4.5; P = 0.46).