Introduction
For early stage pancreatic cancer, surgery followed by adjuvant therapy is widely used in the setting of recent advances in chemotherapy regimens. Reference Neoptolemos, Palmer and Ghaneh1,Reference Conroy, Hammel and Hebbar2 Post-operative complications, however, may prevent more than 30% of patients from receiving systemic adjuvant therapy. Reference Merkow, Bilimoria and Tomlinson3 The benefit of NT has been thus investigated as a potential alternative to the upfront surgery. Neoadjuvant chemoradiation or chemotherapy has potential advantages, including early systemic therapy that may allow tumour downstaging with a greater likelihood of clear-margin resection and better selection of surgical candidates. Reference Abbott, Baker and Talamonti4
With NT and surgery, prior studies added AC with mixed results. Reference Roland, Katz and Tzeng5–Reference Lutfi, Talamonti and Kantor8 National Comprehensive Cancer Network (NCCN) guideline recommends the consideration of AC for those who underwent NT followed by surgery, 9 and the survival benefit of AC remains unclear. This NCDB study examined the long-term outcome of those who received AC compared to that of those who received no further therapy, following NT and surgery.
Method
Patient population
The NCDB is a de-identified, national database with approximately 70% of new cancer cases in the United States of America Reference Bilimoria, Stewart, Winchester and Ko10 It is queried for patients with pancreatic cancer diagnosed between 2004 and 2015. This dataset was exempt from institutional review board review.
Patient selection criteria are shown in Figure 1. Final cohort comprised patients with resected stages I–II, clinical T1-3N0-1M0 pancreatic adenocarcinoma who had been treated with neoadjuvant chemotherapy or chemoradiation followed by curative-intent resection and either AC or observation. Tumour staging in 2004–2015 was based on American Joint Committee on Cancer (AJCC) 7th edition. Reference Edge, Byrd, Carducci and Compton11
Patient selection diagram.
Baseline characteristics were extracted for analysis. Pancreatic surgery has been categorised by Whipple and Whipple-variant surgeries as previously described. Reference Zhong, Patel and Switchenko12 Based on their median values, age, tumour size and post-operative inpatient admission, duration were stratified by <64 years or ≥ 64 years, <3·1 cm or ≥3·1 cm and <1 week or ≥1 week, respectively. Baseline carbohydrate antigen (CA) 19-9 level was predetermined with a cut-off value at 98 U/mL within the NCDB. Surgical margins were stratified by negative (R0) and positive (R1, R2, positive margin not otherwise specified).
All missing values were re-coded as unknown for our analysis. Other pertinent variables are unavailable in the NCDB, such as performance status, type and duration of chemotherapy, local or distant recurrence and toxicity profile. Patients were excluded if they have incomplete follow-up data, distant metastases, palliative-intent treatments, adjuvant radiation or missing surgery data. To address immortal time bias, patients with a survival of less than 6 months were excluded as a conditional landmark. Reference Levesque, Hanley, Kezouh and Suissa13 OS, defined as the duration from the diagnosis to either the last follow-up or death, was used as the primary end point.
Statistical analysis
Fisher’s exact and Mann–Whitney U tests were used to compare variables between AC and observation cohorts. Kaplan–Meier method and log-rank tests were performed for survival. Logistic regression univariate (UVA) and multivariate analyses (MVA) identified predictors for the receipt of AC. Cox proportional hazard UVA and MVA identified variables associated with OS.
MVA models were initially built based on all statistically significant variables from UVA. They were finalised after a backward stepwise elimination. By adding interaction terms on the Cox MVA model, treatment interactions with select variables were examined. Reference Barraclough and Govindan14 If statistically significant, the final Cox MVA model and forest plot were analysed for each subgroup of such variables. Reference Barraclough and Govindan14
Propensity score matching was performed to minimise the selection bias. Matched variables included facility, age, Charlson–Deyo comorbidity score (CDS), year of diagnosis, tumour size, tumour grade, pathologic stages, baseline CA 19-9 level, treatment regimens, surgical margin, duration of post-operative inpatient admission and unplanned readmission within 30 days post-operatively. Using the nearest neighbour method with a caliper distance of 0·2 of the standard deviation of the logit of the propensity score, cohorts were matched in a 1:1 ratio without any replacement. Reference Austin15 Cox UVA using matched sample evaluated the association between AC and OS.
R software (version 3.5.1, R Foundation for Statistical Computing, Vienna, Austria) was used for all analyses. All P-values were two-sided. P-values were considered statistically significant if they were less than 0·05.
Results
Baseline characteristics
A total of 1,737 patients met inclusion criteria. Of those, 1,247 patients had no adjuvant therapy while 490 patients had AC (Figure 1). The majority of patients had pathologic T3N0-1 adenocarcinoma of the pancreatic head (Table 1).
Baseline characteristics prior to matching.
Abbreviations: NA, not available; well diff, well differentiated; mod diff, moderately differentiated; poor diff, poorly differentiated; IQR, interquartile range.
On logistic MVA, patients with male gender [OR (Odds Ratio) 1·24, p = 0·046] and positive lymph node diseases (OR 1·54, p < 0·001) are more likely to receive AC. Those with older age (OR 0·78, p = 0·026), African-American race (OR 0·57, p = 0·0097), pancreatic body tumours (OR 0·60, p = 0·022) and higher radiation dose (OR 0·9995, p = 0·010) were less likely to undergo AC.
On Cox MVA (Table 2), treatments at academic facilities (HR 0·84, p = 0·0055), receiving multi-agent chemotherapy (HR 0·77, p < 0·001) and AC (HR 0·78, p < 0·001) are associated with improved survival. Patients with older age (HR 1·24, p < 0·001), poorly differentiated tumours (HR 1·31, p = 0·047), pathologic N1 stages (HR 1·60, p < 0·001), elevated CA 19-9 level (HR 1·18, p = 0·039) and positive surgical margin (HR 1·76, p < 0·001) had worse mortality. After Cox MVA, treatment interactions were seen favouring AC for patients with smaller tumours (<3·1 cm: HR 0·67, p < 0·001; ≥3·1 cm: HR 0·93, p = 0·48; Fig. 3). No other treatment interactions were observed in other variables, including age (p = 0·20), CDS (p = 0·15), duration of post-operative inpatient admission (p = 0·055), readmission within 30 days post-operatively (unplanned: p = 0·68; others: p = 0·71), grade (moderately differentiated: p = 0·83; poorly differentiated: p = 0·95; others: p = 0·98), pathologic T (p = 0·60) and N stage (p = 0·58), surgical margin (p = 0·27), single- versus multi-agent chemotherapy (p = 0·17) and radiation (p = 0·71).
Cox multivariate analysis.
Abbreviations: CI, confidence interval; ref, reference; well diff, well differentiated; mod diff, moderately differentiated; poor diff, poorly differentiated.
A total of 490 pairs were matched. All variables were well balanced (Table 3). After matching, there were no statistically significant differences in characteristics between the cohorts including age, Charlson–Deyo score, tumour grade, tumour size and tumour stage. The median follow-up was 31·1 months [IQR (Interquartile Range) 23·2–49·7] for post-operative observation cohort and 36·0 months (IQR 25·7–53·2) for the AC cohort. The median OS was 22·3 months (IQR 14·5–33·6) and 26·3 months (17·8–38·9) for the post-operative observation and AC cohorts, respectively (p < 0·001). OS at 2 years was 52·9% for the post-operative observation cohort and 63·9% for the AC cohort (Figure 2).
Baseline characteristics for matched cohorts.
Abbreviations: NA, not available; well diff, well differentiated; mod diff, moderately differentiated; poor diff, poorly differentiated; IQR, interquartile range.
Forest plot for subgroup analysis.
Abbreviations: Obs, observation; adj, adjuvant chemotherapy.
Overall survival for adjuvant therapy versus observation after matching.
Note: P < 0.001.
Abbreviations: Obs, observation; adj, adjuvant chemotherapy.
Discussion
Recent trials showed the survival benefit of AC among patients with upfront resection in pancreatic cancer patients. Reference Neoptolemos, Palmer and Ghaneh1,Reference Conroy, Hammel and Hebbar2 Combination of gemcitabine and capecitabine was compared with gemcitabine alone in ESPAC-4 trial. Reference Neoptolemos, Palmer and Ghaneh1 Median OS was 28 months for the combined regimen and 25·5 months for the gemcitabine alone group, with comparable toxicity profile. Reference Neoptolemos, Palmer and Ghaneh1 In the study by Neoptolemos et al., they found elevated CA 19-9 level, worse tumour grade, positive lymph nodes and single-agent chemotherapy were associated with worse survival outcomes, which are consistent with our study. Reference Neoptolemos, Palmer and Ghaneh1 Similarly, a modified FOLFIRINOX regimen was compared with gemcitabine alone in PRODIGE 24/CCTG PA.6 trial, with improved median OS of 54·4 months for the modified FOLFIRINOX group compared to 35·0 months for the gemcitabine alone group. Reference Conroy, Hammel and Hebbar2 Toxicity was shown to be worse for the modified FOLFIRINOX group. Reference Conroy, Hammel and Hebbar2
However, the benefits of AC following NT and surgery remain unclear as only a handful of retrospective studies have addressed this topic. Reference Roland, Katz and Tzeng5–Reference Lutfi, Talamonti and Kantor8 Several recent prospective studies suggested the feasibility of AC for those treated with neoadjuvant chemotherapy or chemoradiation, although only 30–60% of such patients completed the full course of adjuvant gemcitabine. Reference Versteijne, Suker and Groothuis16,Reference Ahmad, Duong and Sohal17 Our study suggests a benefit to survival with AC compared to observation following NT and surgery for pancreatic adenocarcinoma. Its results are in line with the majority of prior studies, which have found survival benefits in subsets of patients treated with AC. Reference Roland, Katz and Tzeng5,Reference Barnes, Krepline and Aldakkak6,Reference Lutfi, Talamonti and Kantor8
Unlike prior studies, the effect of AC in our study did not depend upon the lymph node stage. Reference Roland, Katz and Tzeng5,Reference Barnes, Krepline and Aldakkak6 A single-institution study of 263 patients showed survival benefits in patients treated with AC following surgery and NT compared with patients that received NT and surgery alone. Reference Roland, Katz and Tzeng5 This association was only found in patients with a lymph node ratio <0·15. However, the authors did not account for immortal time bias, which may have led to an overestimation of the treatment effect size. A similarly designed study involving 234 patients suggested improved survival in lymph node-positive patients who received adjuvant therapy compared to those who did not. Reference Barnes, Krepline and Aldakkak6 No survival benefit of adjuvant therapy was noted among lymph node-negative patients. The authors state that prior to the increased adoption of adjuvant therapy in 2011, patients may have been selected for adjuvant treatment in a biased manner based upon their risk factors.
Two prior NCDB studies have examined the role of adjuvant therapy with mixed results. Reference de Geus, Kasumova and Eskander7,Reference Lutfi, Talamonti and Kantor8 Lutfi et al. suggested perioperative chemotherapy, defined as preoperative and post-operative chemotherapy, improved survival compared to neoadjuvant chemotherapy alone. Reference Lutfi, Talamonti and Kantor8 de Geus et al. performed a similar study and reported no difference in survival between patients who did and did not receive adjuvant therapy. Reference de Geus, Kasumova and Eskander7 The advent of improved chemotherapy regimens could explain why our more recent study was able to find improved survival rates with AC while de Geus et al. did not.
In our study, multi-agent chemo was associated with improved survival, consistent with recent trials. Reference Neoptolemos, Palmer and Ghaneh1,Reference Conroy, Hammel and Hebbar2 In our analysis, AC was more effective in smaller tumours. A prior study reported that very small primary tumours were associated with decreased OS compared to some larger tumours. Reference Muralidhar, Nipp and Ryan18 The authors posit the possibility that these small tumours represent biologically aggressive cancers. Rapidly proliferating cancers are typically more affected by chemotherapeutic agents. More traditionally, as the probability of metastatic spread is related to the size of the primary tumour, adjuvant therapies will be more effective early in the disease course when tumours are small. Reference Tubiana19
Due to the retrospective nature of our study and the use of the NCDB, there exist several limitations in interpreting the data. First, the clinical reasoning as to why an individual patient received AC or observation is not fully captured by the data fields of the NCDB and creates the possibility of a selection bias. Patients with less post-operative complications may be more likely to receive AC. Though NCDB does not capture performance status, 30-day unplanned post-operative admission and post-operative inpatient admission duration were included for matching variables and MVA as proxy measures for performance status. Reference Stitzenberg, Chang, Smith and Nielsen20–Reference Zafar, Shah and Nembhard22 Secondly, important details about treatment regimens including which chemotherapy agents were used, radiographic response to NT, the duration of chemotherapy, time to AC, tumour recurrence and subsequent salvage therapy are not available. Third, given our small sample size and since our study did not find the survival benefit of pathologic downstaging and adding neoadjuvant radiation to chemotherapy on Cox MVA, our patients were not stratified for analysis based on such factors. Despite these limitations, our NCDB study captured larger sample sizes that would have been difficult in single-institution studies. In our practice, increasingly, we will consider adding AC after NT and surgery in pancreatic cancer. Further prospective trials to more carefully examine the benefits of AC after NT and surgery, in addition to honing down on optimal multi-agent regimens and dosing are needed.
Conclusion
Using propensity score-matched analysis, our findings suggest a survival benefit for AC compared to observation following NT and surgery for resectable pancreatic adenocarcinoma, especially in patients with smaller tumours and independent of lymph node involvement. Prospective studies are needed to identify a subset of patients that would benefit from AC.
Acknowledgements
The National Cancer Database is a joint project of the Commission on Cancer of the American College of Surgeons and the American Cancer Society. The data used in the study are derived from a de-identified NCDB file. The American College of Surgeons and the Commission on Cancer have not verified and are not responsible for the analytic or statistical methodology employed, or the conclusions drawn from these data by the investigator.
Financial Support
Dr Singh is currently supported by the National Cancer Institute (grant number 5P30CA016056-42) Cancer Center Support Grant.
Conflicts of Interest
The authors declare none.
Ethical Standards
Our study received Roswell Park Institutional Review Board approval to use the NCDB (STUDY00000621/BDR 099918). No consent to participate was needed.
Availability of Data and Materials
The data that support the findings of this study are available from the Commission on Cancer of the American College of Surgeons and the American Cancer Society. The data is publicly accessible by any investigator affiliated with the Fellow of the American College of Surgeons applying to gain access. More information can be found at: http://ncdbpuf.facs.org/.
