Effect of Diabetes on Survival after Resection of Pancreatic Adenocarcinoma. A Prospective, Observational Study

Aim To investigate the effect of diabetes mellitus (DM) on disease-free and overall post-resection survival of patients with pancreatic ductal adenocarcinoma (PDAC) Methods Prospective observational study on patients admitted for pancreatic disease from January 2008 to October 2012. DM was classified as recent-onset (<48 months before PDAC diagnosis), longstanding (≥48 months before PDAC) or new onset (after surgery). Results Of 296 patients, 140 had a diagnosis of DM prior to surgery (26 longstanding, 99 recent-onset, 15 with unknown duration). Median follow-up time was 5.4 ± 0.22 years. Patients with recent onset DM had poorer postoperative survival than patients without DM: disease-free survival and overall survival were 1.14±0.13 years and 1.52±0.12 years in recent onset DM, versus 1.3±0.15 years and 1.87±0.15 years in non-diabetic patients (p = 0.013 and p = 0.025, respectively). Longstanding DM and postoperative new onset DM had no impact on prognosis. Compared to cases without DM, patients with recent onset DM were more likely to have residual disease after surgery and to develop liver metastases during follow-up. Multivariate analysis confirmed recent onset DM was independently associated with PDAC relapse (hazard ratio 1.45 [1.06–1.99]). Conclusion Preoperative recent onset DM has an impact on survival after the resection of PDAC.

first-degree relatives, date of diagnosis of PDAC and presenting sign/symptoms, concomitant diseases (classified as ever/never: non-pancreatic tumour, dyslipidaemia, hypertension, coronary artery disease, hepatic diseases, autoimmune diseases, thyroid dysfunction). In patients with previous diagnosis of DM the medical records were also reviewed to compute disease duration and diabetes medication. All abstracted data were validated with either an interview with the patient and/or a relative. Perioperative and pathologic data were also collected in a prospective electronic pancreatic surgery database.

Blood biochemistry during hospitalization
Fasting blood samples were obtained during the preoperative work-up [upon or just before hospital admission, median 6 days (1-13) before surgery] and prior to hospital discharge [median 9 (7-12) days after surgery]. Baseline hemoglobin A 1c (HbA 1c ; Bio-Rad Variant II HbA 1C analyzer; Bio-Rad Laboratories, Munich, Germany), baseline and discharge fasting plasma glucose (FPG; glucose-oxidase method, Advia 2400; Siemens Diagnostics, Deerfield, IL) and baseline and discharge serum creatinine (kinetic alkaline picrate method, Advia 2400) were measured in all patients. Serum insulin (AIA-PACK IRI; Tosoh, Tokyo, Japan) was measured upon admission and prior to hospital discharge in 291 and 122 out of 364 patients, respectively. C-peptide (AIA-PACK C-Peptide; Tosoh) was measured at baseline and prior to hospital discharge in 135 and 67 out of 364 patients, respectively. Autoantibodies to glutamic acid decarboxylase (GADA), insulinoma-associated protein 2 (IA-2A), insulin (IAA) and zinc transporter 8 antigen (ZnT8A) were measured in 300 out of 364 patients by radio binding and immunoprecipitation assays as previously described [32]. Insulin resistance and β-cell function were estimated using the HOMA2 model (available from www.OCDEM.ox.ac.uk), following the recommendations for its appropriate use [33]. Estimation of glomerular filtration rate (eGFR) was calculated using the simplified modification of diet in renal disease (MDRD) formula [34]. To asses inflammatory biomarkers we calculated peripheral neutrophil-to-lymphocyte ratio (NLR) [35] and platelet-to-lymphocyte ratio (PLR), both reported as poor prognostic indicators in several malignancies, and lymphocyte-to-monocyte ratio (LMR) previously associated with a favorable prognosis for certain hematologic and solid tumors [36]. Preoperative nutritional status was evaluated in all study participants using Onodera's prognostic nutrition index (PNI) calculated as 10 x albumin (g/dl) + 0.005 x total lymphocyte count (per mm 3 ). A PNI value of at least 50 was defined as normal nutritional status, between 50 and 46 was regarded as mild malnutrition, between 45 and 40 as moderate malnutrition and less than 40 as severe malnutrition [37]. To grade nutritional status we computed the Geriatric Nutritional Risk Index (GNRI) [1.519 x albumin (g/L) + 41,7 x (weight / ideal weight)] [38]. A GNRI value of at least 98 was defined as normal, less than 98 was regarded as mild malnutrition, between 92 and 82 as moderate malnutrition and less than 82 as severe malnutrition.

Diabetes definition and its duration
Study participants were defined as having diabetes if at least one fasting plasma glucose (FPG) was !7.0 mmol/l or HbA 1C was ! 6.5% (48 mmol/mol) or they were prescribed diabetes medications. At the time of PDAC diagnosis participants were classified as having: a) long-standing diabetes if they had a documented diagnosis of DM for !48 months; b) short-standing diabetes if they had a documented diagnosis of DM for <48 months [48-months is a period of time during which a PDAC is likely to become manifest (16)]; c) previously undiagnosed diabetes if participants were diagnosed with DM at the time of the diagnosis of PDAC; d) DM of uncertain duration if participants had a documented diagnosis of DM prior to PDAC diagnosis but duration was unknown. We define recent onset DM as the combination of short-standing and previously unknown DM. During follow up participants were classified as having new-onset DM if a documented diagnosis of DM was made after pancreatic surgery.

Follow-up
Outpatient's visits were scheduled one month after hospital discharge and every six months thereafter. Follow-up data were collected for all patients by reviewing electronic medical records and/or telephone interviews. Adjuvant chemo-or radiotherapy was prescribed when indicated and CT scan and blood tumour markers were performed every three to six months, based on the risk of recurrence. Medical history, including diagnosis of diabetes, disease-free survival and overall survival were recorded.

Statistical Analysis
Data are presented as mean ± standard deviation (SD) or median and (interquartile range, IQR), according to their distribution. Variables with a normal distribution were compared with one-way unpaired or paired Student's t test (two groups) or one-way ANOVA test (three or more groups). Variables with a non-normal distribution were compared with Wilcoxon signed-rank test, or Mann-Whitney U test (two groups), or Kruskal-Wallis test (three or more groups). Categorical variables were compared with the chi-square test or Fisher's exact test, as appropriate. Disease-free and overall survival was estimated according to Kaplan-Meier. Association between variables were evaluated by logistic regression and Cox regression analysis as appropriate. The multivariate analysis was performed using variables significant in the univariate analysis at the level of 0.20. All statistical analyses were performed using the SPSS statistical software, version 13.0 (SPSS Inc, Chicago, Illinois, USA).

Study participants
When PDAC was diagnosed, 140 out of 296 patients (47.3%) already had DM. Long-duration DM, short-duration DM and previously undiagnosed DM accounted for 18.6% (n = 26), 40% (n = 56) and 30.7% (n = 43) of the patients with DM, respectively, while 10.7% (n = 15) of the patients had a DM of uncertain duration. Patients with DM of uncertain duration (n = 15) were excluded from subsequent analysis. Of 281 cases in our series, 173 (61.6%) underwent pancreaticoduodenectomy, 72 (25.6%) distal pancreatectomy and 36 (12.8%) total pancreatectomy. As of December 2015, median follow-up time after diagnosis was 5.4 ± 0.22 (standard error) years. Two-hundred and sixteen patients died during follow-up (76.9%). All the 65 patients known to be alive at the end of the observation period were censored at the latest follow-up visit, with a median follow-up time of 5.1 ± 0.29 years.

Effect of pre-operative DM on PDAC outcome
The general characteristics of the study participants are reported in Table 1. Patients with recent onset DM were older than patients without DM (69.1±9 vs 65.3±11 years, p<0.05), had higher FPG and HbA 1C at admission and higher prevalence of family history of diabetes. The median duration of recent-onset diabetes was 0.77±1.1 years prior to PDAC diagnosis (Table 1). Both beta-cell dysfunction and insulin resistance contributed to the pathogenesis of recent onset diabetes: with respect to patients without DM, C-pep HOMA2-%B was 71 (52-99) vs 121 (89-155; p<0.05) and insulin HOMA2-%S was 74 (43-133) vs 102 (75-160; p<0.05), respectively ( Table 1). As for the treatment of diabetes 23.2% of patients were treated with insulin, 25.3% with oral diabetes medications, 3% with both insulin and oral diabetes medications, and 8.1% with lifestyle modifications. Seventeen percent of patients with recentonset DM tested positive for at least one autoantibody and 1% for more than one: IAA was the most frequent specificity (12%) followed by GADA (3%), IA-2A (2%) and ZnT8A (1%). Also long-lasting DM patients were older than patients without DM (69.6±7.6 years, p<0.05), had higher FPG and HbA 1C at admission and higher prevalence of family history of diabetes. The age at onset of long-lasting diabetes was 54±9.9 years, with an overall median DM duration of 15±9.9 years at PDAC diagnosis (Table 1). Similarly to recent-onset diabetes, both beta-cell dysfunction and insulin resistance contributed to the pathogenesis of this diabetes: C-pep HOMA2-%B was 61 (25-131) vs 121 (89-155; p<0.05) and insulin HOMA2-%S was 60 (24-117) vs 102 (75-160; p<0.05) in long-standing diabetes and patients without DM, respectively ( Table 1). As for the treatment of diabetes 61.5% of the patients were treated with insulin, 38.5% with oral diabetes medications, and 11.5% with both insulin and oral diabetes medications. Twenty-nine percent of the patients with long-standing DM tested positive for at least one autoantibody and 4% for more than one: IAA was the most frequent specificity (29%), followed by GADA (8%). Cardiovascular comorbidity, hyperlipidaemia and prostatic hypertrophy were all associated with long-standing DM. Based on preoperative glycaemic status, disease free-survival and overall survival were analysed (Fig 1) and to develop liver metastases during the follow-up, while tumour location, stage, grading, therapy were not different. Concordantly recent-onset diabetes had a significant impact on liver metastasis free-survival and local recurrence-free survival (Fig 2). Taken together these data suggest that preoperative recent-onset diabetes, but not long-lasting DM, has an impact on the prognosis of PDAC.
To better characterized factors associated with disease-free and overall survival, a Cox regression analysis was performed including patients with recent-onset DM and patients without diabetes (n = 255). In the univariate analysis (Fig 3) we found an association between disease relapse and recent-onset diabetes [ .001] were associated with lower survival in the univariate analysis after adjusting for age and sex (Fig 3). The multivariate analysis (S2 Table) confirmed tumour stage, tumour size and development of liver metastasis as independent predictors of PDAC mortality.

Effect of post-operative DM on PDAC outcome
Among the 156 patients without DM at the time of PDAC diagnosis, 60 (20.3% of total) developed DM after pancreatic resection, while 96 (32.4%) were without DM at the time of PDAC  Table 2. The age at onset of patients with new-onset diabetes was 66.1 ±10.2 years, with a diabetes-free median time of 30±48 days after discharge. As for the treatment of diabetes, at last follow-up 38.3% of the patients were treated with insulin, 15% with oral diabetes medications. Factors associated with postoperative new onset DM in the univariate Cox regression analysis were: family history of DM, distal pancreatectomy, pre-operative higher weight, BMI and FPG; all these factors, with the exception of weight and BMI, were confirmed in the multivariate analysis (S1 Fig). To characterize the association between new onset DM and disease-free or overall survival, a Cox regression with time dependent covariates was performed. Tumor size, pT, pN and tumor stage were associated with both PDAC relapse and overall mortality in the univariate analysis (Fig 4). Residual disease and tumor grade were associated with overall mortality. Multivariate analysis (S2 Table) confirmed tumor stage [HR = 1.44 (1.01-2.06); p = 0.042] as a variable independently associated with PDAC relapse. New onset diabetes was not associated with PDAC relapse or mortality.

Discussion
In the last years several cohort studies have showed that cancer patients with preexisting DM are generally at increased risk of adverse long-term outcomes compared to patients without DM [11,39,40]. In the field of PDAC, the prognostic relevance of preoperative DM is still controversial. In retrospective studies DM was associated with reduced survival in patients undergoing resection for PDAC [21,26,29,[41][42][43][44][45]; on the other hand, DM had no impact on overall survival in other studies [46][47][48][49][50] or was unexpectedly associated to a reduced risk of death [51][52][53]. Inconsistent findings were also reported when DM was stratified by duration of diabetes: recent-onset DM, but not long-standing DM was associated with reduced PDAC survival in some studies [10,29], while others did not find this association [23,24,43,54,55]. The inconsistency of those observations is likely due to differences in DM definition, small sample size, study populations with different stages of disease and the retrospective design, with no controlling for potential confounders. Consequently there is a need for further investigation of the relationship between DM and prognosis of PDAC [56]. Taking advantage of a prospective observational study conducted to describe T3cDM [30], we evaluated the impact of DM on disease-free survival, recurrence pattern and overall survival after potentially curative resection of PDAC limiting the effect of the potential confounders which may have biased previous studies. Taken together our data suggest that pre-operative recent-onset DM, but not long-lasting DM or post-operative new-onset DM, has an impact on the prognosis of PDAC. Our results are in agreement with a recent meta-analysis reporting that preexisting DM in PDAC is associated with increased risk of mortality [57]. Of note, the meta-analysis showed that the risk of mortality was higher in patients with resected or resectable cancer than in those with nonresectable cancer and was higher in patients with recent-onset DM than in those with long-lasting DM [57].
It remains unclear the possible cause of the relation between recent onset DM and prognosis of PDAC. Since long-standing DM and postoperative new onset DM did not affect survival, the poorer prognosis of patients with recent onset DM does not seem related to hyperglycemia, nor to insulin resistance and hyperinsulinemia, nor to poor overall health that may render patients more prone to complications following surgical and adjuvant therapies. In fact glucose exposure, both in terms of FPG and HbA 1c , was higher in long-standing than in recent-onset DM, either before or after PDAC removal. Similarly, during follow-up FPG and HbA 1C were as expected higher in new-onset DM than in patients without DM. Hyperinsulinemia and low insulin sensitivity were both present in recent-onset and, even more prevalent, in long-standing DM. Mortality within 90 day from surgery in the different groups were similar, and did not significantly contribute to the observed differences in long-term survival. Moreover, long-standing DM showed the strongest association with cardiovascular comorbidity and dyslipidemia. Taken together these data suggest that the association of DM with survival of PDAC patients is more complex than previously hypothesized. There are several potential explanations for our results. We can speculate that the loss of pancreatic parenchyma leading to the DM recent onset may be the consequence of a more aggressive cancer or of a tumour with longer duration before diagnosis. This hypothesis could be supported by the presence of a R1 resection rate higher in recent onset diabetes than in the other two groups, a difference that could be responsible for the worse behavior. To address this issue we evaluated the disease-free survival and overall survival for each of the 3 groups for patients with R0 resection alone and R1 resections alone (data not shown). The recent onset diabetes showed a clear trend for less median overall survival and disease free survival independently by the R1 status, speaking against this hypothesis. The association between baseline variables and PDAC relapse or mortality was assessed by a Cox regression analysis including patients with recent-onset DM and without DM. All analyzed variables are presented. Dots represent the Hazard Ratio (HR) after natural log transformation, lines the 95% confidence intervals. *p<0.05; **p<0.01; ***p<0.001. FPG = Fasting Plasma Glucose; HOMA = Homeostatic Model Assessment; GNRI = Geriatric Nutritional Risk Index; NLR = neutrophil-to-lymphocyte ratio; LMR = lymphocyte-to-monocyte ratio; PLR = platelet-to-lymphocyte ratio.
doi:10.1371/journal.pone.0166008.g003 Another possibility could be that under recent-onset hyperglycemic conditions, an increased level of oxidative stress and proinflammatory factors cause pancreatic nerve damage and an inflammatory response [58], which simultaneously facilitates cancer cell proliferation, migration, and metastasis [59]. Testing these postulated explanations is beyond the scope of this study, but their exploration should be continued in future research. As previously reported [30] our cohort of patients presents some biases: patients studied were admitted for surgery (potential selection bias) in a tertiary-care academic center (potential referral bias). Moreover, in about 10% of patients with DM, it was not possible to document disease duration. Our use of 48 months as the cutoff for classification of recent-onset versus long-standing DM is questionable. However, the utilization of this threshold is based on studies that demonstrate progressive increase in prevalence of DM within 5 years prior to index date of cancer diagnosis [15]. We also analyzed our data using 24 months as an alternative cutoff for recent-onset DM (data not shown). In this scenario, 16 patients moved from the recentonset category to the long-standing group. Major trends in clinicopathologic profile and survival remained the same.

Conclusion
In conclusion, our study has clearly shown that a recent diagnosis of DM preceding the diagnosis of PDAC is independently associated with survival after PDAC resection. Etiologies underlying this association are unclear, and require additional investigations.