CA19.9 Response and Tumor Size Predict Recurrence Following Post-neoadjuvant Pancreatectomy in Initially Resectable and Borderline Resectable Pancreatic Ductal Adenocarcinoma

After Institutional Review Board approval (PAD-R, n.1101CESC), patients undergoing post-neoadjuvant pancreatectomy for PDAC at the Unit of General and Pancreatic Surgery, University of Verona Hospital Trust, and at the Pancreatic Surgery Unit, San Raffaele University Hospital, Milan, from 2013 to 2017 were retrieved from prospectively maintained electronic databases. Resectability was classified according to the National Comprehensive Cancer Network (NCCN) guidelines¹¹ and only patients who were resectable or BR at the time of diagnosis were included, in compliance with a rigorous definition of NAT.^11,12 Additional exclusion criteria were distant metastases, macroscopically incomplete (R2) resection, in-hospital mortality, and missing information on recurrence or early censoring (<6 months, Study flowchart in Supplementary Fig. 1). Standard demographic, clinical, and surgical details were captured. Radiologic staging was integrated with the concepts of “biologic” and “conditional” BR disease, as proposed in the MD Anderson Cancer Center (MDACC) classification¹³ (Supplementary Table 1). Radiographic features and CA19.9 levels were assessed both at baseline and posttreatment. Tumor size was measured as the biggest diameter on computed tomography (CT) imaging, and radiographic response was assessed using the Response Evaluation Criteria in Solid Tumors (RECIST) v1 criteria.¹⁴ CA19.9 levels were considered evaluable only when the total bilirubin level was <2 U/mL. For patients experiencing jaundice at diagnosis (around 55% of the cohort), only CA19.9 values captured after endoscopic drainage (and subsequent bilirubin normalization—i.e., total bilirubin level <2 U/mL—were included in the analysis. When post-drainage CA19.9 values were not available, the data were considered as missing. The upper limit of normality used for CA19.9 was 37 U/ml. CA19.9 response was calculated as the percentage variation in response to NAT [ΔCA19.9 = (baseline CA19.9 – posttreatment CA19.9)/baseline CA19.9]. Patients whose baseline levels were <5 U/mL were considered nonsecretors and analyzed as a separate group.¹⁵

Throughout the study period, NAT was indicated for all BR patients and favored in anatomically resectable tumors exhibiting risk features (e.g., “biologic” and “conditional” BR tumors according to the MDACC classification).¹³ Chemotherapy regimens were assigned by the treating medical oncologist, and predominantly entailed FOLFIRINOX and gemcitabine + nab-paclitaxel. While the planned duration of NAT was 6 months in both institutions, the actual amount of chemotherapy depended on patient tolerance and radiological and biochemical response. Multidisciplinary evaluation of each case was performed following restaging. Minimum requirements for surgical eligibility were a stable disease per RECIST criteria and a performance status of 0–1 Eastern Cooperative Oncology Group (ECOG). Determinants of intraoperative resectability were absence of distant metastases, reconstructible superior mesenteric vein/portal vein, and no need for superior mesenteric artery resection. Pancreatectomies were performed in a standard fashion as previously described.¹⁶ Microscopic residual disease (R1) was determined based on the presence of tumor cells within 1 mm from any margin. The 8th Edition of the American Joint Committee of Cancer Staging Manual was applied.¹⁷ Active postoperative surveillance was carried out at 3–6-month intervals through physical examination, cross-sectional imaging, and measurement of CA19.9 serum levels. Disease recurrence was diagnosed radiographically in conjunction with clinical picture and/or CA19.9 levels; tissue diagnosis was occasionally performed. Follow-up was closed on July 2020.

The RFS was computed from the date of surgery to the date of last follow-up or disease recurrence. For patients experiencing recurrence, the median post-recurrence survival (PRS) was evaluated, from the date of recurrence to the last follow-up. The location of first recurrence was classified as local (in the pancreatic remnant, resection bed, or along the peripancreatic vasculature), distant, or combined local and distant. Distant metastases were further classified based on the specific site (liver-only, lung-only, or multiple sites, including peritoneal carcinomatosis). The disease-specific survival (DSS) was calculated from the date of surgery to the date of last follow-up or disease-related death.

Data were analyzed using the R.4.0.0 software (Foundation for Statistical Computing, Vienna, Austria; https://​www.​r-project.​org). Continuous variables were expressed as medians with interquartile range (IQR) and compared using Mann–Whitney U test. Categorical variables were presented as frequencies with percentages and compared using Chi-square or Fisher’s exact tests, as appropriate. All tests were two-tailed. Recurrence estimates were derived through life tables. Survival curves were constructed using the Kaplan–Meier method, and pairwise differences between groups were assessed using the log-rank test. While tumor size and CA19.9 parameters were initially handled as continuous variables, a minimum p-value approach was employed to identify clinically meaningful cut-off points. This entails selecting the threshold maximizing differences in RFS between groups. The association between clinically relevant preoperative variables and RFS was investigated through uni- and multivariable Cox regression models. The possible interplay between the various CA19.9 parameters and radiological features was investigated including interaction terms. A statistically significant interaction term indicates that the association of a given variable with RFS differs depending on the value of the covariate.¹⁸ The effect of the interaction was visualized plotting the conditional effects, which are the predictive values of one interaction term conditioned on certain (reference) levels of the other, using the ggeffects package. To avoid multicollinearity, different CA19.9 parameters were evaluated in distinct multivariable models. Other set of uni- and multivariable Cox regression models were also designed, including postoperative and pathologic data.

The amount of missing information for each variable accounted for less than 10% (Table 1). Preoperative data were considered to be missing at random and handled with multiple imputations with five permutations. Continuous variables were imputed by predictive mean matching, and binary variables by logistic regression. Pathologic, postoperative, and outcome variables (recurrence/survival) were not considered to be missing at random and were not imputed. The p-values are presented with odds ratios (OR) or hazard ratios (HR) and 95% confidence intervals (CI) as appropriate. Statistical significance was determined by a p-value <0.05.

The study population consisted of 315 patients, of whom 152 (48.3%) were anatomically resectable at diagnosis. Their characteristics are presented in Table 1. The median follow-up was 24.9 months (IQR 33.3–13.8 months) from surgery and 33.3 months (IQR 24.1–45.2 months) from diagnosis. At the time of the last contact, 166 patients (52.7%) were still alive, with a median follow-up of 30.8 months from surgery (IQR 20.9–43.2 months) and 39.8 months from diagnosis (IQR 29.8–50.4 months). The median RFS was 15.7 months (95% CI 12.7–18.7 months) (Fig. 1a). Disease recurrence manifested in 215/315 patients (68.3%). The estimated recurrence rate exceeded 40% at 1-year postoperatively and approached 75% at 3-years postoperatively (Fig. 1b). Isolated local recurrence occurred in 16.7% of patients (n = 36), distant metastases in 49.8% (n = 107) and combined recurrence in 33.5% (n = 72) of cases (Fig. 2a). The proportion of recurrence location as a function of time from pancreatectomy is shown in Fig. 2b. The median postoperative DSS was 41.3 months (95% CI 35.0–47.5 months). Survival outcomes varied depending on the specific recurrence pattern, with lung-only and multiple-distant sites exhibiting the most and least favorable prognostic outlook, respectively (Table 2 and Supplementary Fig. 2).

The median RFS was not significantly different based on resectability status, either at baseline (16.3 vs 14.3 months for resectable and borderline resectable patients, p = 0.318) or posttreatment (15.7 vs 14.3 months, p = 0.233). RECIST response was indeed associated with RFS (20.0 months vs 12.7 months for partial response vs stable disease, p = 0.002). Tumor size, analyzed as a continuous variable, was not associated with RFS at baseline (HR 1.004, 95% CI 0.995–1.013, p = 0.364), yet turned to be significant on posttreatment evaluation (HR 1.037, 95% CI 1.023–1.052, p < 0.001). Differences in RFS were maximized by a threshold of 19 mm (p = 7.34 × 10⁻⁷). Rounding this to 20 mm, 170/315 patients had a tumor size above the threshold (54.0%), with RFS being 25.0 versus 10.8 months for ≤20 mm versus >20 mm (Supplementary Fig. 3). Stratified analyses by tumor size are presented in Table 3. Notably, RECIST response did not remain statistically significant, while a posttreatment tumor size >20 mm was significantly associated with RFS in both the partial response and stable disease groups.

Baseline CA19.9 levels were not significantly associated with RFS (16.3, 14.3, and 29.1 months for normal, elevated, and not expressed, respectively, p = 0.120), while there were significant differences in RFS based on posttreatment CA19.9 (17.7, 11.5, and 29.1 months for normal, elevated, and not expressed, respectively, p = 0.009). After excluding CA19.9 nonsecretors (n = 30), ΔCA19.9 was significantly associated with RFS (HR 0.992, 95% CI 0.985–0.999, p = 0.023), with a critical value maximizing RFS differences set at 53.8% (p = 7.26 × 10⁻⁴), which was approximated at 50.0%. Based on this definition, 199 patients (69.8%) experienced a CA19.9 response, while 86 (30.2%) were nonresponders, with RFS being 17.7 months in the former group and 11.5 months in the latter (Supplementary Fig. 4). On stratified analyses, an elevated posttreatment CA19.9 was associated with a shorter RFS only in the cohort of patients with normal baseline values, and was not a significant predictor of RFS when stratifying patients by CA19.9 response (Table 3). Conversely, ΔCA19.9 was significantly associated with RFS both in patients with normal and elevated baseline CA19.9 values. Moreover, ΔCA19.9 remained significantly associated with RFS in patients with elevated levels, but not in those with normal posttreatment CA19.9 values.

The uni- and multivariable analyses of preoperative variables associated with RFS in the study cohort are presented in Table 4. Tumor location (tail), duration of chemotherapy, elevated posttreatment CA19.9, and posttreatment tumor size were associated with shorter RFS. Indeed, ΔCA19.9 ≥ 50% was associated with longer RFS. When including interaction terms in the model, in the cohort of CA19.9 expressors, a significant interaction was confirmed between ΔCA19.9 and posttreatment CA19.9 (HR 0.551, 95% CI 0.364–0.835, p = 0.005), suggesting a significant risk reduction in patients with elevated posttreatment CA19.9 values, when ΔCA19.9 exceeded 50% (Fig. 3a). Moreover, the interaction between baseline and posttreatment CA19.9 was also significant (HR 0.371, 95% CI 0.217–0.453, p = 0.020), indicating a particularly elevated risk in the instance of CA19.9 elevation during NAT (Fig. 3b).

When combining radiologic and CA19.9 parameters, posttreatment tumor size was found to significantly interact with both posttreatment CA19.9 (HR 1.619, 95% CI 1.134–2.310, p = 0.008, Fig. 3c) and ΔCA19.9 (HR 0.566, 95% CI 0.392–0.817, p = 0.002, Fig. 3d). This suggests an increased risk in the instance of an elevated posttreatment CA19.9, and a protective effect associated with CA19.9 response in the cohort of patients with greater tumor size. Conversely, baseline CA19.9 did not significantly interact with tumor size.

The analysis of pathologic and clinical factors associated with RFS is presented in Supplementary Table 2. After multivariable adjustment, R-status (HR 1.350, 95% CI 1.001–1.821, p = 0.049) together with AJCC T-status (HR 1.550, 95% CI 1.093–2.197, p = 0.014 for ypT2 vs ypT1; HR 2.200, 95% CI 1.202–4.028, p = 0.011 for ypT3 vs ypT1) and N-status (HR 1.127, 95% CI 0.791–1.605, p = 0.509 for ypN1 vs ypN0; HR 2.244, 95% CI 1.517–3.317, p < 0.001 for ypN2 vs ypN0) remained independent predictors of RFS.