• obesity;
  • pancreatic neoplasms;
  • prognosis;
  • diabetes;
  • hyperglycemia


  1. Top of page
  2. Abstract
  6. Acknowledgements


Higher body-mass index (BMI) has been implicated as a risk factor for developing pancreatic cancer, but its effect on survival has not been thoroughly investigated. The authors assessed the association of BMI with survival in a sample of pancreatic cancer patients and used epidemiologic and clinical information to understand the contribution of diabetes and hyperglycemia.


A survival analysis using Cox proportional hazards by usual adult BMI was performed on 1861 unselected patients with pancreatic adenocarcinoma; analyses were adjusted for covariates that included clinical stage, age, and sex. Secondary analyses incorporated self-reported diabetes and fasting blood glucose in the survival model.


BMI as a continuous variable was inversely associated with survival from pancreatic adenocarcinoma (hazard ratio [HR], 1.019 for each increased unit of BMI [kg/m2], P < .001) after adjustment for age, stage, and sex. In analysis by National Institutes of Health BMI category, BMIs of 30 to 34.99 kg/m2 (HR, 1.14; 95% confidence interval [CI], 0.98-1.33), 35 to 39.99 kg/m2 (HR 1.32, 95% CI 1.08-1.62), and ≥40 (HR 1.60, 95% CI 1.26-2.04) were associated with decreased survival compared with normal BMI of 18.5 to 24.99 kg/m2 (overall trend test P < .001). Fasting blood glucose and diabetes did not affect the results.


Higher BMI is associated with decreased survival in pancreatic cancer. Although the mechanism of this association remains undetermined, diabetes and hyperglycemia do not appear to account for the observed association. Cancer 2010. © 2010 American Cancer Society.

High body mass index (BMI) has been consistently reported as a risk factor for pancreatic cancer.1-7 Although the mechanism for this increased risk is not yet established, hyperinsulinemia and insulin resistance have been hypothesized.1, 6 In 1 study using lean and obese mice inoculated with murine pancreatic cancer cells, higher serum insulin and lower adiponectin in obese mice correlated with increased tumor cell proliferation, but there was no change in apoptosis indices.8 The same group has also reported that steatosis in human pancreas at the time of surgical tumor resection correlated with increased likelihood of lymph node metastasis, and hypothesized that steatosis affects the tumor microenvironment.9

Obesity has been reported to be associated with poorer prognosis in multiple cancers,10, 11 perhaps most notably breast cancer.12-15 In breast cancer, it is possible that increased peripheral estrogens related to obesity may contribute to risk of recurrent disease, although this mechanism remains indeterminate. In pancreatic cancer, increased BMI has been reported to be an adverse prognostic factor for survival after surgery in 2 surgical series of 285 and 356 patients, respectively.16, 17 However, another surgical report of 306 patients undergoing resection of pancreatic adenocarcinoma reported increased postoperative complications in obese patients, but the slight decrease in survival did not reach statistical significance.18

An epidemiologic study19 has reported an association of risk of pancreatic cancer with increased BMI at various time points throughout life, and further observed decreased survival from pancreatic cancer, with the strongest effect among resected patients. It should be noted that this sample was drawn from the same institution where 1 of the abovementioned surgical series was reported. Another hospital-based study has also shown a decreased survival among obese patients compared with normal weight among 475 patients with pancreatic cancer, although statistical significance was not reached (hazard ratio [HR], 1.62; 95% confidence interval [CI], 0.76-3.44).20

We more fully examined the effect of BMI on pancreatic cancer survival using the Mayo Clinic Pancreas Biospecimen Resource, a prospective patient series that employs ultra rapid recruitment methods.


  1. Top of page
  2. Abstract
  6. Acknowledgements

The study was reviewed and approved by the Mayo Clinic Institutional Review Board, and written, informed consent was obtained on all subjects.

Patient Recruitment

Pancreatic cancer patients were rapidly and systematically identified and approached, using methodology reported previously21 at Mayo Clinic Rochester, Mayo Clinic Arizona, or Mayo Clinic Florida between October 1, 2000 and January 1, 2009. Of 2746 adenocarcinoma patients identified during this time period, 1898 (69.1%) consented to participate; 1.9% were excluded for missing height, weight, or stage of pancreatic cancer, leaving 1861 patients suitable for analysis. Only histologically (95%) or clinically (5%) confirmed adenocarcinoma cases providing consent were included in survival analyses. Clinically confirmed cases required a pancreatic mass on imaging consistent with adenocarcinoma, and symptoms typical of pancreatic adenocarcinoma (weight loss, abdominal pain, painless jaundice). All cases were reviewed by subspecialist physicians with expertise in pancreatic cancer (oncologist or surgeon) for coding as adenocarcinoma. At the time of enrollment, participants were asked to complete risk factor questionnaires including personal medical history, behaviors, family history, and usual adult height and weight (which were used for calculating BMI), and data from which Karnofsky performance score22 was determined. Fasting blood glucose (FBG) was obtained from the electronic medical record for all subjects at study entry. When risk factor questionnaire information was not completed (n = 629), the medical record was abstracted to ascertain weight at study entry, usual adult weight, recent weight loss, ever/never smoking status, BMI, prior diagnosis of diabetes, and family history of pancreatic cancer. We have previously reported an intermethod reliability study in which usual adult height and weight from 25 cases and 25 healthy controls was obtained from both questionnaires and abstracted from the medical record. Because usual adult weight was not usually available, weight at study entry plus reported recent weight loss was used. A high degree of intermethod reliability was noted (Pearson correlation coefficient >0.93).23 Staging was recorded according to American Joint Committee on Cancer sixth edition staging criteria by a physician with expertise in gastrointestinal cancer. Patients were then grouped into surgically resected, locally advanced, and metastatic disease for survival analysis. Surgical patients were also subcategorized into stages IA, IB, IIA, and IIB. Vital status of subjects was collected by multiple sources as part of routine research follow-up, including periodic mailings, medical records, tumor registry, and death indexes from online services.

Statistical Analysis

Age was used as a continuous variable and defined as age in years at diagnosis of pancreatic cancer. Risk factor questionnaires provided self-report of diabetes (yes/no); cigarette smoking was recorded as ever/never, and also by typical packs-per-day for smoking intensity. BMI was calculated from self-reported usual adult height and weight. Weight loss was recorded as usual adult weight minus weight at study entry. Date of diagnosis was defined as date of tissue diagnosis for those with pathology-proven disease. For those who were clinically diagnosed, date of the first clinical diagnosis was used. Date of death or last known date alive was selected from the most current data among multiple sources as described above. For the primary analysis, BMI was considered as a continuous variable. Secondary analyses also examined BMI categorized according to NIH guidelines of underweight (<18.5 kg/m2), normal weight (referent group, 18.5-24.99 kg/m2), overweight (25.0-29.99 kg/m2), obese class I (30.0-34.99 kg/m2), obese class II (35.0-39.99 kg/m2), and obese class III (≥40.0 kg/m2).24 In addition to age and sex, covariates showing associations (P < .05) in univariate analysis were considered for multivariate analysis.

Cox proportional hazards regression analyses were used to conduct comparisons for survival. Time to event was diagnosis of pancreatic cancer until death or last follow-up with individuals still alive at last follow-up being coded as censored for the event (death) as of that date. All analyses used SAS software, version 9.1.3 (SAS Institute, Cary, NC).


  1. Top of page
  2. Abstract
  6. Acknowledgements

Median follow-up for cases was 306 days. By the time of analysis, 1527 (82.1%) patients were deceased. Table 1 shows the characteristics of patients included in the analysis. Patients who completed questionnaires were more likely to be female, older, earlier stage, and not diabetic, to have longer survival and lower BMI, and to report weight loss at study entry.

Table 1. Characteristics of Pancreatic Adenocarcinoma Patients, Including Completion of Risk Factor Questionnaire
CharacteristicNot Completed, n=629Completed, n=1232Total, n=1861Pa
  • NOS indicates not otherwise specified; SD, standard deviation; BMI, body mass index; FBG, fasting blood glucose.

  • a

    Comparison of subjects who did and did not complete risk factor questionnaires.

Male sex375 (59.6%)669 (54.3%)1044 (56.1%).029
Age at time of pancreatic cancer diagnosis, y.002
 Q1, Q356.0, 73.059.0, 74.058.0, 74.0 
Vital status.164
 Alive102 (16.2%)232 (18.8%)334 (17.9%) 
 Dead527 (83.8%)1000 (81.2%)1527 (82.1%) 
Site of tumor.943
 Head341 (54.6%)700 (57.2%)1041 (56.3%) 
 Body87 (13.9%)165 (13.5%)252 (13.6%) 
 Tail66 (10.6%)116 (9.5%)182 (9.8%) 
 Head and body43 (6.9%)80 (6.5%)123 (6.7%) 
 Body and tail37 (5.9%)79 (6.5%)116 (6.3%) 
 NOS8 (1.3%)16 (1.3%)24 (1.3%) 
 Uncinate process28 (4.5%)45 (3.7%)73 (4%) 
Stage group<.001
  IA12 (1.9%)17 (1.4%)29 (1.6%) 
  IB12 (1.9%)55 (4.5%)67 (3.6%) 
  IIA28 (4.5%)111 (9%)139 (7.5%) 
  IIB93 (14.8%)232 (18.8%)325 (17.5%) 
 Locally advanced246 (39.1%)405 (32.9%)651 (35%) 
 Metastatic238 (37.8%)412 (33.4%)650 (34.9%) 
Days survival from diagnosis<.001
 Mean (SD)334.4 (384.51)511.9 (578.95)451.9 (528.02) 
 Q1, Q3107.0, 426.0198.5, 569.5166.0, 523.0 
Diabetes (self-reported or reported in medical record).036
 No245 (64.8%)812 (70.5%)1057 (69.1%) 
 Yes133 (35.2%)339 (29.5%)472 (30.9%) 
Continuous BMI, usual adult, kg/m2<.001
 Mean (SD)29.4 (6.58)28.1 (5.37)28.5 (5.83) 
 Q1, Q324.6, 32.624.4, 30.724.4, 31.2 
BMI by category, usual adult, kg/m2<.001
 <18.56 (1%)7 (0.6%)13 (0.7%) 
 18.5-24.99 (reference)166 (26.4%)368 (29.9%)534 (28.7%) 
 25.0-29.99228 (36.2%)493 (40%)721 (38.7%) 
 30.0-34.99123 (19.6%)238 (19.3%)361 (19.4%) 
 35.0-39.9958 (9.2%)80 (6.5%)138 (7.4%) 
 ≥40.048 (7.6%)46 (3.7%)94 (5.1%) 
Weight loss as a % of usual adult weight<.001
 None145 (23.1%)211 (17.1%)356 (19.1%) 
 >0, ≤10%255 (40.5%)382 (31%)637 (34.2%) 
 >10%229 (36.4%)639 (51.9%)868 (46.6%) 
Categorical FBG (mg/dL).110
 Missing159 (%)259 (%)418 
 <10082 (17.4%)133 (13.7%)215 (14.9%) 
 100-125177 (37.7%)349 (35.9%)526 (36.5%) 
 126-15085 (18.1%)195 (20%)280 (19.4%) 
 151-20075 (16%)199 (20.5%)274 (19%) 
 >20051 (10.9%)97 (10%)148 (10.3%) 
Performance score (Karnofsky)
 90-100252 (26.8%)252 (26.8%) 
 80315 (33.7%)315 (33.7%) 
 70257 (27.5%)257 (27.5%) 
 6083 (8.9%)83 (8.9%) 
 <5028 (3.0%)28 (3.0%) 

Kaplan-Meier survival comparisons (Fig. 1) showed an association of BMI with survival (chi-square P = .011). Median survival for each BMI category (kg/m2) was as follows: (0-18.49) 276 days, (18.5-24.99) 349 days, (25.0-29.99) 352 days, (30.0-34.99) 310 days, (35.0-39.99) 273 days, and (≥40.0) 246 days.

thumbnail image

Figure 1. Unadjusted Kaplan-Meier analysis of survival among 1861 pancreatic cancer patients is shown by usual adult body mass index (BMI) category.

Download figure to PowerPoint

In multivariate survival analyses, increasing age (HR, 1.017 per year; 95% CI, 1.012-1.022) and locally advanced (HR, 2.99; 95% CI, 2.60-3.43) or metastatic disease (HR, 4.76; 95% CI, 4.14-5.46) versus resectable disease were associated with decreased survival; ever/never smoking, tumor location (head vs body/tail), and family history of pancreatic cancer showed no association with survival in stepwise analysis and were not included in the final multivariate model. BMI was inversely associated with survival as a continuous variable in adjusted analysis (HR, 1.019; P < .001).

Analysis by usual adult BMI National Institutes of Health (NIH) category was also significantly associated with survival (P trend <.001), with a clear dose-dependent association with survival (Table 2, Fig. 2A, first panel). Adjusted hazard ratios for NIH BMI categories compared with the referent group of normal weight were underweight (HR, 1.30; 95% CI, 0.69-2.45), overweight (HR, 1.02; 95% CI, 0.89-1.16), obese class I (HR, 1.14; 95% CI, 0.98-1.33), obese class II (HR, 1.32; 95% CI, 1.08-1.62), and obese class III (HR, 1.60; 95% CI, 1.26-2.04). Although the hazard ratio was increased for underweight persons, the sample size was small (n = 13). For BMI above the normal range, there is a clear, steady decrease in survival with increasing BMI. A sensitivity analysis was performed including only biopsy-proven adenocarcinoma cases, and this did not alter the significance of the associations, with only minimal changes in HRs.

Table 2. Association of Pancreatic Cancer Survival by Usual Adult BMI Category
Subject GroupVariableSubcategoryNo.HR (95% CI)P
  1. BMI indicates body mass index; HR, hazard ratio; CI, confidence interval; ref, reference; RFQ, risk factor questionnaire; FBG, fasting blood glucose; DM, self-reported diabetes mellitus.

OverallBMI (kg/m2)<18.5131.30 (0.69-2.44)<.001
  18.5-24.99 (ref)5341.00 
  25.0-29.997211.02 (0.89-1.16) 
  30.0-34.993611.14 (0.98-1.33) 
  35.0-39.991381.32 (1.08-1.62) 
  ≥40.0941.60 (1.26- 2.04) 
 SexMale10440.99 (0.89-1.10).806
 AgeAge at diagnosis18611.02 (1.01-1.02)<.001
 Surgical stageResectable (ref)5601.00<.001
  Locally advanced6512.99 (2.60-3.43) 
  Metastatic6504.76 (4.14-5.47) 
Completed RFQBMI (kg/m2)<18.571.80 (0.80-4.07).009
  18.5-24.99 (ref)368 
  25.0-29.994931.01 (0.86-1.18) 
  30.0-34.992381.12 (0.93-1.35) 
  35.0-39.99801.32 (1.02-1.72) 
  ≥40.0461.67 (1.19-2.33) 
Did not complete RFQBMI (kg/m2)<18.560.75 (0.27-2.07).222
  18.5-24.99 (ref)1661.00 
  25.0-29.992281.05 (0.84-1.33) 
  30.0-34.991231.23 (0.95-1.59) 
  35.0-39.99581.30 (0.93-1.80) 
  ≥40.0481.43 (1.0-2.05) 
ResectableBMI (kg/m2)<18.544.78 (1.59-14.37).037
  18.5-24.99 (ref)1711.00 
  25.0-29.992201.24 (0.95-1.62) 
  30.0-34.991031.13 (0.82-1.56) 
  35.0-39.99391.29 (0.84-1.99) 
  ≥40.0231.67 (1.00-2.79) 
 SexMale2971.05 (0.84-1.30).687
 AgeAge at diagnosis5601.01 (1.00-1.02).072
Locally advancedBMI (kg/m2)<18.551.37 (0.43-4.32).067
  18.5-24.99 (ref)1791.00 
  25.0-29.992541.03 (0.83-1.28) 
  30.0-34.991231.07 (0.83-1.38) 
  35.0-39.99501.56 (1.12-2.17) 
  ≥40.0401.46 (0.99-2.16) 
 SexMale3650.94 (0.79-1.12).509
 AgeAge at diagnosis6511.02 (1.01-1.03)<.001
MetastaticBMI (kg/m2)<18.540.82 (0.26-2.58).018
  18.5-24.99 (ref)1841.00 
  25.0-29.992470.89 (0.72-1.09) 
  30.0-34.991351.1743 (0.93-1.48) 
  35.0-39.99491.105 (0.80-1.53) 
  ≥40.0311.63 (1.10-2.40) 
 SexMale3820.98 (0.83-1.16).824
 AgeAge at diagnosis6501.02 (1.01-1.03)<.001
Overall model (including FBG)BMI (kg/m2)<18.5101.23 (0.58-2.62).007
  18.5-24.99 (ref)4021.00 
  25.0-29.995630.93 (0.80-1.08) 
  30.0-34.992841.05 (0.89-1.25) 
  35.0-39.991111.22 (0.97-1.53) 
  ≥40.0731.49 (1.13-1.97) 
 SexMale8340.93 (0.83-1.05).251
 AgeAge at diagnosis14431.02 (1.01-1.02)<.001
 FBGFasting blood glucose/1014431.01 (1.00-1.02).082
Overall model (including DM)BMI (kg/m2)<18.5121.42 (0.76-2.68).009
  18.5-24.99 (ref)4381.00 
  25.0-29.996041.03 (0.89-1.18) 
  30.0-34.992941.16 (0.98-1.37) 
  35.0-39.991071.35 (1.07-1.69) 
  ≥40.0741.50 (1.13-1.98) 
 SexMale8640.98 (0.87-1.10).721
 AgeAge at diagnosis15291.02 (1.01-1.02)<.001
 DiabetesYes4721.08 (0.95-1.22).229
thumbnail image

Figure 2. (A) Body mass index (BMI) and survival of pancreatic cancer patients are shown by completion of risk factor questionnaire. Forest plots show multivariate Cox regression survival associations by National Institutes of Health BMI category for pancreatic cancer patients overall and by risk factor questionnaire completion. *Referent group for BMI is 18.5 to 24.99 kg/m2. **Referent group for stage is resectable. (B) BMI and survival of pancreatic cancer patients are shown by stage. Forest plots within stage groupings for resectable, locally advanced, and metastatic patients at initial diagnosis of pancreatic cancer are shown. CI indicates confidence interval.

Download figure to PowerPoint

The dose-dependent association of usual adult BMI and survival was seen in patient groups who either completed or did not complete the questionnaire (Fig. 2A, second and third panels). For those who completed questionnaires, adjusted analysis was also performed using self-reported Karnofsky performance score as a covariate. Performance score was significantly associated with survival (P < 10−5), and the overall association for BMI remained (HR, 1.014; P = .023). Among stage subgroups, BMI was significantly associated with survival in patients presenting with locally advanced (HR, 1.021; P = .005) and metastatic (HR, 1.018; P = .011) disease, but not in resected patients (HR, 1.012; P = .227), although the dose-dependent pattern appeared consistent for all 3 stage groupings (Fig. 2B).

Because weight loss is a prominent feature in clinical presentation of pancreatic cancer, we evaluated its effects on survival. In multivariate analysis, weight loss as a percentage of body weight was also associated with decreased survival (HR, 2.03; P = .015). In weight loss categories (none, >0 and ≤10%, or >10%), there was a significant association (P = .009) with decreased survival seen with higher weight loss, adjusted for age, stage, and sex.

To further investigate whether a hyperglycemic state accounted for our findings, FBG at the time of study recruitment was examined on 1443 cases for whom it was available. When added to the overall model, a modest association of FBG and survival was seen (HR, 1.01), although this did not reach statistical significance (P = .096). Because many diabetic patients were likely already on medication to lower FBG, we also examined the subset of patients who self-reported no history of diabetes (n = 833). There was no association of FBG and survival in this group (HR, 1.00; P = .419). Self-reported diabetes at time of study entry was also not associated with survival (HR, 1.07; P = .273). None of these additions to the model changed the association of BMI and survival (Fig. 3).

thumbnail image

Figure 3. Fasting blood glucose (FBG) and self-reported diabetes are incorporated into models of pancreatic cancer survival with body mass index (BMI), adjusted for stage. Forest plots show multivariate Cox regression survival associations with FBG or self-reported diabetes included as covariates. *Referent group for BMI is 18.5 to 24.99 kg/m2. CI indicates confidence interval.

Download figure to PowerPoint


  1. Top of page
  2. Abstract
  6. Acknowledgements

In a clinic-based survival study, we observed that increasing BMI was strongly and consistently associated with a decreased survival from pancreatic cancer. The mechanism for this finding is not clearly evident, although circulating hyperinsulinemia has been hypothesized as 1 potential etiology.19 For this reason, we examined both self-reported diabetes and FBG levels of patients at presentation to our center. However, diabetes and hyperinsulinemia did not appear to explain the effect of BMI on survival, leaving the explanation for this finding currently unknown. This is in distinction to a recent letter stating diabetes was associated with worsened survival in the Li et al paper.25

Higher BMI is well known to be associated with altered circulating levels of estrogens26, 27 and insulin.28 However, because the survival of women does not substantially differ from that of men in pancreatic cancer, it is difficult to conclude that estrogens contribute to poor survival. Hyperinsulinemia and associated factors are a potential contributor, as has been reported in 1 recent study of colorectal cancer.29 A recent large meta-analysis did not show any effect of diabetes on survival in pancreatic cancer patients.30 However, 1 study of 400 male Finnish smokers suggested higher insulin levels were associated with increased risk for pancreatic cancer.6 Obesity has been associated with low-grade chronic inflammation, with possibly a shift toward a Th2 (tolerant) immune state, which has been associated with carcinogenesis31-33 and cancer progression.34-37 Therefore, an alternate possibility is decreased immune function, which could lead to more rapid tumor progression and thereby poorer survival. A third possibility would be bias in treatment determinations, such as aggressiveness of therapy (multiple agents or modalities) or decreased dosing of chemotherapy in obese patients. Resources did not permit us to fully pursue this possibility.

The strengths of our study include its large size, the inclusion of only specialist-verified adenocarcinoma, the ready availability of electronic medical records when questionnaires were not completed, and the completeness of our follow-up. The dose-dependent nature of our findings, the strength of the association, and the significance of the finding after adjusting for performance score add to the internal validity of this association.

There are several potential limitations to our study. We relied on self-reported usual adult BMI, which can be problematic because weight can vary throughout life, although we felt this was a superior measure than weight at study entry, because pancreatic cancer often causes dramatic weight loss, often just before diagnosis. It could be argued that BMI may affect stage at presentation, or likelihood of a patient undergoing surgery, thus influencing survival. However, we observed no differences in BMI between stage groupings; median usual adult BMIs for resectable, locally advanced, and metastatic tumors were 27.9, 28.2, and 28.0 kg/m2, respectively (P = .647). With regard to limitations of measures, BMI is but 1 measure of obesity, although it is perhaps the most widely accepted measure in clinical studies. FBG and self-report of diabetes are imprecise measures of hyperinsulinemia; more direct measures are required to further elucidate in this setting the relationship of BMI to hormonal changes in insulin and similar species, such as insulinlike growth factor (IGF) and IGF-binding proteins. These parameters should be investigated in future studies.

Numerous studies comparing self-reported to measured BMI have found that self-reported BMI is generally valid to use,38-41 although subjects may underestimate weight, which in our study would bias toward the null. We believe that usual adult BMI is a reasonable surrogate for aggregate exposure to whatever risk high BMI imposes for a given individual over their lifetime, and has similar issues for accuracy and recall as seen for usual alcohol intake or cigarette use, commonly used in epidemiologic studies. Our study did not include therapies administered. However, given the minimal impact on survival of current therapies, and the nearly uniform use of gemcitabine-based regimens for locally advanced and metastatic disease, it could be argued that therapy differences would not notably affect our findings.

Future studies will be necessary to investigate this association of elevated BMI and survival. If the mechanism is ever clearly elucidated, this may lead to potential therapeutic targeting. For instance, because elevated levels of circulating insulin can activate IGF receptors,42 blockade of these receptors through targeted therapy may represent a logical approach to improving outcomes for patients. We believe this association of BMI with survival provides an important clue to the understanding of why this cancer is so deadly and resistant to therapy. A thorough understanding of the mechanisms could lead to improving our care for these patients.


Increasing BMI is associated with decreased survival in those diagnosed with pancreatic cancer, after adjusting for known confounders. BMI should be considered as a covariate in prospective studies of pancreatic cancer. The mechanism of this finding requires further study.


  1. Top of page
  2. Abstract
  6. Acknowledgements

We thank Gloria Petersen, PhD, for allowing us to use the Mayo Clinic Pancreas SPORE Registry for this work, and for her guidance on this project.


  1. Top of page
  2. Abstract
  6. Acknowledgements

Supported by National Institutes of Health grants SPORE P50CA102701 and K07CA116303.


  1. Top of page
  2. Abstract
  6. Acknowledgements