Impact of Diabetes and Insulin Use on Prognosis in Patients With Resected Pancreatic Cancer: An Ancillary Analysis of NRG Oncology RTOG 9704

Abstract

Purpose:

Diabetes mellitus (DM) has been proposed to be tumorigenic; however, prior studies of the association between DM and survival are conflicting. The goal of this ancillary analysis of RTOG 9704, a randomized controlled trial of adjuvant chemotherapy in pancreatic cancer, was to determine the prognostic effects of DM and insulin use on survival.

Methods and Materials:

Eligible patients from RTOG 9704 with available data on DM and insulin use were included. Overall survival (OS) and disease-free survival (DFS) were estimated using the Kaplan-Meier method, and variable levels were compared using log-rank test. Cox proportional hazards models were created to assess the associations among DM, insulin use, and body mass index phenotypes on outcomes.

Results:

Of 538 patients enrolled from 1998 to 2002, 238 patients were eligible with analyzable DM and insulin use data. Overall 34% of patients had DM and 66% did not. Of patients with DM, 64% had insulin-dependent DM, and 36% had non-insulin-dependent DM. On univariable analysis, neither DM nor insulin dependence were associated with OS or DFS (P > .05 for all). On multivariable analysis, neither DM, insulin use, nor body mass index were independently associated with OS or DFS. Nonwhite race (hazard ratio [HR], 2.18; 95% confidence interval [CI], 1.35–3.50; P = .0014), nodal involvement (HR, 1.74; 95% CI, 1.24–2.45; P = .0015), and carbohydrate antigen 19–9 (CA19-9) ≥ 90 U/mL (HR, 3.61; 95% CI, 2.32–5.63; P < .001) were associated with decreased OS. Nonwhite race (HR, 1.67; 95% CI, 1.05–2.63; P = .029) and CA19-9 ≥ 90 U/mL (HR, 2.86; 95% CI, 1.85–4.40; P < .001) were associated with decreased DFS.

Conclusions:

DM and insulin use were not associated with OS or DFS in patients with pancreatic cancer in this study. Race, nodal involvement, and increased CA19-9 were significant predictors of outcomes. These data might apply to the more modern use of neoadjuvant therapies for potentially resectable pancreatic cancer.

Introduction

Pancreatic cancer and diabetes mellitus (DM) are strongly associated diseases, sharing several risk factors such as age, sex, obesity, and family history of DM.¹ DM patients are approximately twice as likely to develop pancreatic cancer as their nondiabetic counterparts, and the prevalence of DM is higher in pancreatic cancer than in other common cancers.^2–5 In addition, the prevalence of DM rises to approximately 70% in the months before a pancreatic cancer diagnosis, and up to 88% of patients with diabetic pancreatic cancer receive a diagnosis of DM within 1 year of their cancer diagnosis, suggesting that pancreatic cancer itself is diabetogenic.^2,6–8

The underlying mechanisms of the relationship between DM and pancreatic cancer are unclear and likely multifactorial. In longstanding DM, hyperinsulinemia and insulin-like growth factor can promote acinar cell transformation and carcinogenesis, raising the concern that insulin use could affect pancreatic cancer survival.⁹ In addition, a high level of advanced glycation end products in the setting of uncontrolled blood glucose has been implicated in the release of proinflammatory cytokines and S100 proteins,^10,11 which, together with obesity,¹² can contribute to an inflammatory state that increases cancer risk. Alternatively, potential mechanisms by which pancreatic cancer can induce diabetes include mass-effect within the pancreas leading to β-cell loss and paraneoplastic phenomena.^13,14 The findings that recent-onset DM often occurs before the tumor is radiologically detectable¹⁵ and that DM can resolve after resection of the pancreatic cancer¹⁶ supports the paraneoplastic hypothesis.

Studies assessing the relationship between DM and survival in pancreatic cancer have reported conflicting results, with most epidemiologic studies in resectable disease showing that DM is associated with worse survival.^17–20 However, cancer-directed therapy was heterogeneous in these studies, and analyses are complicated by many potential confounding clinical variables. In addition, few studies have assessed the independent association between insulin dependence and pancreatic cancer outcome. Understanding the true association of DM and insulin use with survival in pancreatic cancer might help to guide clinical decision making and to prioritize potential therapeutic targets in this deadly disease. An ancillary analysis, therefore, was performed of RTOG 9704, a phase 3 trial of patients with resected pancreatic cancer randomized to 5fluoruracil (5-FU) or gemcitabine given before and after 5-FU–based chemoradiotherapy,^21,22 to understand the prognostic significance of DM and insulin use in resected pancreatic cancer.

Methods and Materials

Patient population and treatment

This is an ancillary analysis of RTOG 9704, a prospective randomized controlled trial, the details of which have previously been reported.^21,22 After surgery for resectable pancreatic adenocarcinoma, patients were randomized to either adjuvant 5-FU versus gemcitabine, both delivered before and after 5-FU–based chemoradiation. Institutional review board approval was obtained from each participating site. Each patient provided informed consent before enrollment. DM and insulin dependence were collected on a case report form along with other baseline information provided by participating sites. All eligible patients who had data on DM status, insulin use, and carbohydrate antigen 19–9 (CA19-9) were included in this analysis. Patients with unanalyzable CA19-9 levels and missing DM or insulin use data were excluded.

Statistical methods

The following baseline characteristics were dichotomized into meaningful groupings, both for ease in interpretation of results and at times owing to small numbers in some individual categories: age (<65 vs ≥65 years), race (white vs nonwhite), body mass index (BMI; under-/normal weight vs overweight/ obese),primary tumor location (head vs non-head), Karnofsky performance status score (90–100 vs 60–80), pathologic T stage (T1-T2 vs T3-T4), American Joint Commission on Cancer fifth edition stage (I-II vs III-IV), and largest tumor diameter (<3 cm vs ≥3 cm). Statistical comparisons to assess potential associations between baseline characteristics and (1) availability of DM and insulin use data, (2) DM status, and (3) DM/insulin use group (no DM vs insulin-dependent DM [IDDM] vs non-insulin-dependent DM [NIDDM]) were completed using the χ² test.

Overall survival (OS) and disease-free survival (DFS) were estimated univariately with the Kaplan-Meier method,²³ and the levels of DM and insulin use were compared using the log-rank test. Univariate and multivariable Cox proportional hazards models²⁴ were used to assess the association of DM and insulin use with OS and DFS. For the multivariable analyses, (1) DM status, (2) DM/insulin use group, (3) DM/BMI group (no DM and under-/normal weight vs DM and under-/normal weight vs no DM and overweight/obese vs DM and overweight/ obese), or (4) DM/insulin use/BMI group (under-/normal weight vs no DM and overweight/obese vs NIDDM and overweight/obese vs IDDM and overweight/obese) were included in the models along with the following variables to control for potential confounders: treatment arm, age, sex, BMI (for DM and DM/insulin use group models), race, Karnofsky performance status score, tumor location, N stage, tumor diameter, CA19-9, and surgical margin status. To adjust for the multiple comparisons in this analysis, levels were set for associations such that P < .01 was considered to be statistically significant and P ≥ .01 but <.05 was considered to be a trend. Statistical analyses were performed using SAS statistical software (version 9.4; SAS Institute, Cary, NC).

Results

From 1998 to 2002, 538 patients at 164 institutions in the United States and Canada were entered onto RTOG 9704 and 451 were eligible. Of the 87 patients excluded, 85 patients were ineligible and 2 patients withdrew consent. Of the 451 eligible patients, 66 patients had no analyzable CA19-9. These 66 patients were Lewis antigen–positive cases for which tissue was sent to the RTOG tissue bank, per the protocol, but CA19-9 levels could not be determined. An additional 140 patients had no DM data, and an additional 7 had no insulin use data. Thus, 238 patients were included in this analysis. Follow-up data collection was terminated for this trial on June 30, 2010.

Comparison of baseline characteristics between patients with and without DM and insulin use data showed that those with DM and insulin use data were more likely to be ≥65 years old (52% vs 30%; P < .0001) and have a BMI 25 kg/m² compared with those without these data (54% vs. 42%; P = .015; Table E1). There was no difference in OS (hazard ratio [HR], 0.98; 95% confidence interval [CI], 0.80–1.20; P = .84) or DFS (HR, 1.05; 95% CI, 0.86–1.28; P = .64) between patients with and without DM and insulin use data.

Baseline clinical characteristics

Baseline patient characteristics by DM status are shown in Table 1. Eighty patients (34%) had DM, and 158 patients (66%) did not. Of patients with DM, 37 of 115 (32%) were in the 5-FU arm, and 43 of 123 (35%) were in the gemcitabine arm. Patients with DM showed a trend toward having bigger tumors (≥3 cm) than those without DM (68% vs 53%; P = .034).

Table 1.

Baseline demographic and clinical characteristics

	No DM		DM
	(n = 158)		(n = 80)
Characteristics	n	%	n	%	P value*
Age (y)					.24
<65	80	51	34	43
≥65	78	49	46	58
Sex					.10
Male	83	53	51	64
Female	75	47	29	36
Race					.44
White	139	88	73	91
Nonwhite	19	12	7	9
BMI					.27
Under-/normal weight (<25 kg/m2)	77	49	33	41
Overweight/obesity (>25 kg/m2)	81	51	47	59
Primary location					.49
Head	133	84	70	88
Not head	25	16	10	13
KPS score					.056
90–100	105	66	43	54
60–80	53	34	37	46
Pathologic T stage					.41
T1-T2	37	23	15	19
T3-T4	121	77	65	81
Pathologic N stage					.80
N0	54	34	26	33
N1	104	66	54	68
AJCC stage					.97
I-II	49	31	25	31
III-IV	109	69	55	69
Largest tumor dimension of primary					.034
<3 cm	74	47	26	33
≥3 cm	84	53	54	68
Primary tumor status					.073
Complete resection with negative margins	69	44	23	29
Complete resection with positive margins	51	32	35	44
Complete resection with unknown margins	38	24	22	28
CA19-9					.14
Lewis antigen-negative	56	35	29	36
<90 U/mL	86	54	36	45
≥90 U/mL	16	10	15	19
Treatment arm					.65
RT + 5-FU	78	49	37	46
RT + Gemcitabine	80	51	43	54

Abbreviations: 5-FU = 5-fluoruracil; AJCC = American Joint Commission on Cancer; BMI = body mass index; CA19-9 = carbohydrate antigen 19-9; DM = diabetes mellitus; KPS = Karnofsky performance status; RT = radiation therapy.

^*P value from χ² test.

Of patients who had DM, 51 patients (64%) had IDDM, and 29 (36%) had NIDDM. Of those with DM, 28 of 37 (76%) and 23 of 43 (53%) in the 5-FU and gemcitabine arms, respectively, had IDDM (Table E2). There was no statistically significant difference in any of the variables by DM/insulin use group.

Survival analysis

Median follow-up was 1.5 years (range, 0.2–9.0 years) for all patients and 6.8 years (range, 0.3–9.0 years) for alive patients. There were 191 OS events and 211 DFS events in the 238 patients. Two-year OS for patients with and without DM was 33% (95% CI, 23%−43%) and 38% (95% CI, 30%−46%), respectively (HR, 1.27; 95% CI, 0.95–1.71; log-rank P Z .11; Fig. 1A). Two-year DFS for patients with and without DM was 18% (95% CI, 10%−27%) and 23% (95% CI, 17%−30%), respectively (HR, 1.16; 95% CI, 0.87–1.54; log-rank P = .31; Fig. 1B). There was no association between OS and DM/insulin use (IDDM vs no DM: HR, 1.24; 95% CI, 0.87–1.76; and NIDDM vs no DM: HR, 1.33; 95% CI, 0.87–2.04; log-rank P = .27) or DFS and DM/insulin (IDDM vs no DM: HR, 1.15; 95% CI, 0.82–1.60; and NIDDM vs no DM: HR, 1.18; 95% CI, 0.78–1.79; log-rank P = .59; Fig. E1).

Fig. 1.

Kaplan-Meier estimated (A) overall survival and (B) disease-free survival, stratified by diabetes mellitus (DM) (solid blue line) versus no DM (dashed red line). Abbreviations: CI = confidence interval; HR = hazard ratio; RL = reference level. (A color version of this figure is available at https://doi.org/10.1016/j.ijrobp.2020.08.042.)

Table 2 shows the multivariable models for OS and DFS, with DM status. There was no association between DM status and OS or DFS after adjusting for known prognostic factors. In addition, BMI was not associated with OS or DFS. Nonwhite race (vs white; HR, 2.18; 95% CI, 1.35–3.50; P = .0014), N1 disease (vs N0; HR, 1.74; 95% CI, 1.24–2.45; P = .0015), and CA19-9 90 U/mL (vs <90 U/mL; HR, 3.61; 95% CI, 2.32–5.63; P < .0001) were associated with OS. Nonwhite race (vs white; HR, 1.67; 95% CI, 1.05–2.63; P = .029) and CA19-9 ≥90 U/mL (vs <90 U/mL; HR, 2.86; 95% CI, 1.85–4.40; P < .0001) were associated with DFS.

Table 2.

Multivariable Cox regression models for OS and DFS with DM status

Endpoint	Adjustment variables	Comparison	Adjusted HR*	95% CI LL	95% CI UL	P value†
OS	DM	Yes vs no	1.13	0.81	1.57	.49
	Treatment arm	RT + 5-FU vs RT + gemcitabine	1.23	0.91	1.66	.17
	Age (y)	≥65 vs <65	0.93	0.69	1.27	.66
	Sex	Male vs female	0.85	0.63	1.15	.29
	Race	Nonwhite vs white	2.18	1.35	3.50	.0014
	BMI	Overweight/obese vs under-/normal weight	0.88	0.65	1.19	.41
	KPS score	60–80 vs 90–100	1.18	0.85	1.62	.32
	Tumor diameter	≥3 vs <3 cm	1.22	0.88	1.69	.23
	Tumor location	Nonhead vs head	1.11	0.73	1.68	.63
	Surgical margin status	Negative	1.00	—	—	—
		Positive	0.82	0.57	1.18	.28
		Unknown	0.93	0.63	1.37	.71
	Pathologic N stage	N1 vs N0	1.74	1.24	2.45	.0015
	CA19-9	Lewis antigen-negative	1.37	0.99	1.91	.062
		<90 U/mL	1.00	—	—	—
		≥90 U/mL	3.61	2.32	5.63	<.0001
DFS	DM	Yes vs no	1.01	0.74	1.38	.94
	Treatment arm	RT + 5-FU vs RT + gemcitabine	1.12	0.85	1.48	.43
	Age (y)	≥65 vs <65	1.02	0.76	1.36	.92
	Sex	Male vs female	1.01	0.75	1.34	.97
	Race	Nonwhite vs white	1.67	1.05	2.63	.029
	BMI	Overweight/obese vs under-/normal weight	0.89	0.67	1.18	.42
	KPS score	60–80 vs 90–100	1.03	0.76	1.40	.85
	Tumor diameter	≥3 vs <3 cm	1.26	0.92	1.72	.16
	Tumor location	Non-head vs head	1.07	0.72	1.58	.74
	Surgical margin status	Negative	1.00	—	—	—
		Positive	0.96	0.68	1.35	.81
		Unknown	0.74	0.51	1.07	.10
	Pathologic N stage	N1 vs N0	1.23	0.88	1.72	.22
	CA19-9	Lewis antigen-negative	1.06	0.77	1.45	.71
		<90 U/mL	1.00	—	—	—
		≥90 U/mL	2.86	1.85	4.40	<.0001

Abbreviations: 5-FU = 5-fluoruracil; BMI = body mass index; CA19-9 = carbohydrate antigen 19-9; CI = cons\fidence interval; DFS = disease-free survival; DM = diabetes mellitus; HR = hazard ratio; KPS = Karnofsky performance status; LL = lower limit; OS = overall survival; RT = radiation therapy; UL = upper limit.

^*HRs with a CI containing 1 indicate no difference between the levels of the listed variable.

^†P value from χ² test using the Cox proportional hazards model.

Multivariable models for OS and DFS comparing DM/insulin use groups, DM/BMI groups, and DM/insulin use/BMI groups similarly showed decreased OS to be associated with nonwhite race, N1 disease, and higher CA19-9. The models also showed decreased DFS to be associated with nonwhite race and higher CA19-9 (Tables 3–5). OS or DFS did not vary with DM status, insulin use, or BMI in any of these models.

Table 3.

Multivariable Cox regression models for OS and DFS with DM/insulin use groups (n = 238)

Endpoint	Adjustment variables	Comparison	Adjusted HR*	95% CI LL	95% CI UL	P value†
OS	DM/insulin use	No DM	1.00	—	—	—
		IDDM	1.21	0.77	1.90	.42
		NIDDM	1.08	0.73	1.59	.70
	Treatment arm	RT + 5-FU vs RT + gemcitabine	1.24	0.91	1.67	.17
	Age (y)	≥65 vs <65	0.93	0.68	1.27	.64
	Sex	Male vs female	0.85	0.63	1.16	.31
	Race	Nonwhite vs white	2.18	1.35	3.51	.0013
	BMI, kg/m2	Overweight/obese vs under-/normal weight	0.89	0.65	1.21	.46
	KPS score	60–80 vs 90–100	1.18	0.86	1.63	.31
	Tumor diameter	≥3 vs <3 cm	1.22	0.89	1.69	.22
	Tumor location	Non-head vs head	1.10	0.73	1.68	.64
	Surgical margin status	Negative	1.00	—	—	—
		Positive	0.82	0.57	1.17	.27
		Unknown	0.93	0.63	1.37	.71
	Pathologic N stage	N1 vs N0	1.74	1.24	2.45	.0015
	CA19-9	Lewis antigen-negative	1.36	0.97	1.89	.073
		<90 U/mL	1.00	—	—	—
		≥90 U/mL	3.60	2.31	5.62	<.0001
DFS	DM/insulin use	No diabetes	1.00	—	—	—
		NIDDM	0.98	0.63	1.52	.92
		IDDM	1.03	0.72	1.48	.86
	Treatment arm	RT + 5-FU vs RT + gemcitabine	1.12	0.84	1.48	.44
	Age (y)	≥65 vs <65	1.02	0.76	1.36	.91
	Sex	Male vs female	1.00	0.75	1.34	.98
	Race	Nonwhite vs white	1.66	1.05	2.63	.03
	BMI, kg/m2	Overweight/obese vs under-/normal weight	0.88	0.66	1.18	.40
	KPS score	60–80 vs 90–100	1.03	0.76	1.40	.86
	Tumor diameter	≥3 vs <3 cm	1.26	0.91	1.73	.16
	Tumor location	Non-head vs head	1.07	0.72	1.59	.74
	Surgical margin status	Negative	1.00	—	—	—
		Positive	0.96	0.68	1.35	.81
		Unknown	0.73	0.51	1.06	.10
	Pathologic N stage	N1 vs N0	1.23	0.88	1.72	.22
	CA19-9	Lewis antigen-negative	1.06	0.78	1.46	.70
		<90 U/mL	1.00	—	—	—
		≥90 U/mL	2.87	1.86	4.43	<.0001

Abbreviations: 5-FU = 5-fluoruracil; BMI = body mass index; CA19-9 = carbohydrate antigen 19–9; CI = confidence interval; DFS = disease-free survival; DM = diabetes mellitus; HR = hazard ratio; IDDM = insulin-dependent DM; LL = lower limit; KPS = Karnofsky performance status; NIDDM = non—insulin dependent DM; OS = overall survival; RT = radiation therapy; UL = upper limit.

^*HRs with a CI containing 1 indicate no difference between the levels of the listed variable.

^†P value from χ² test using the Cox proportional hazards model.

Table 5.

Multivariable Cox regression models for OS and DFS with DM/insulin use/BMI groups (n = 238)

Endpoint	Adjustment variables	Comparison	Adjusted HR*	95% CI LL	95% CI UL	P value†
OS	DM/Insulin	Under-/normal weight	1.16	0.82	1.64	.40
	Use/BMI	No DM and overweight/obese	1.00	—	—	—
		NIDDM and overweight/obese	1.51	0.82	2.77	.18
		IDDM and overweight/obese	0.94	0.58	1.51	.78
	Treatment arm	RT + 5-FU vs RT + gemcitabine	1.25	0.93	1.69	.15
	Age (y)	≥65 vs <65	0.93	0.68	1.27	.64
	Sex	Male vs female	0.85	0.63	1.16	.31
	Race	Nonwhite vs white	2.20	1.37	3.55	.0012
	KPS score	60–80 vs 90–100	1.21	0.88	1.67	.23
	Tumor diameter, cm	≥3 vs <3	1.24	0.89	1.71	.20
	Tumor location	Non-head vs head	1.11	0.73	1.70	.61
	Surgical margin status	Negative	1.00	—	—	—
		Positive	0.85	0.59	1.21	.35
		Unknown	0.95	0.65	1.40	.80
	Pathologic N stage	N1 vs N0	1.79	1.27	2.53	.0009
	CA19-9	Lewis antigen-negative	1.33	0.95	1.86	.10
		<90 U/mL	1.00	—	—	—
		≥90 U/mL	3.81	2.45	5.90	<.0001
DFS	DM/Insulin	Under-/normal weight	1.13	0.82	1.56	.46
	Use/BMI	No DM and overweight/obese	1.00	—	—	—
		NIDDM and overweight/obese	1.31	0.73	2.38	.37
		IDDM and overweight/obese	0.91	0.58	1.42	.66
	Treatment arm	RT + 5-FU vs RT + gemcitabine	1.14	0.86	1.51	.38
	Age (y)	≥65 vs <65	1.01	0.75	1.35	.96
	Sex	Male vs female	1.00	0.75	1.33	.99
	Race	Nonwhite vs White	1.69	1.07	2.68	.025
	KPS score	60–80 vs 90–100	1.05	0.77	1.42	.78
	Tumor diameter, cm	≥3 vs <3	1.27	0.92	1.74	.14
	Tumor location	Non-head vs head	1.08	0.73	1.60	.71
	Surgical margin status	Negative	1.00	—	—	—
		Positive	0.97	0.69	1.36	.85
		Unknown	0.75	0.52	1.08	.12
	Pathologic N stage	N1 vs N0	1.23	0.88	1.72	.22
	CA19-9	Lewis antigen-negative	1.05	0.77	1.44	.75
		<90 U/mL	1.00	—	—	—
		≥90 U/mL	2.92	1.90	4.48	<.0001

Abbreviations: 5-FU = 5-fluoruracil; BMI = body mass index; CA19-9 = carbohydrate antigen 19–9; CI = confidence interval; DFS = disease-free survival; DM = diabetes mellitus; HR = hazard ratio; IDDM = insulin-dependent DM; KPS = Karnofsky performance status; LL = lower limit; NIDDM = non—insulin dependent DM; OS = overall survival; RT = radiation therapy; UL = upper limit.

^*HRs with a CI containing 1 indicate no difference between the levels of the listed variable.

^†P value from χ² test using the Cox proportional hazards model.

Another model was evaluated with the following more granular DM/insulin use/BMI grouping: DM and under-/normal weight versus NIDDM and under-/normal weight versus IDDM and under-/normal weight versus no DM and overweight/obese versus NIDDM and overweight/obese versus IDDM and overweight/obese (Table E3). There was no association between any of these categories and OS or DFS after adjusting for known prognostic factors.

Discussion

In this ancillary analysis of RTOG 9704, no association was found between DM status and OS or DFS on multivariable analysis after adjusting for potential confounders. Insulin use was not associated with oncologic outcomes, and a DM/insulin use/BMI phenotype was not identified as being associated with survival. As previously reported, nonwhite race, higher CA19-9, and nodal involvement were independently associated with survival.^25–27

DM and pancreatic cancer have overlapping risk factors, and each disease is a risk factor for the other.^2–8 This relationship has motivated numerous investigations into the prognostic value of DM in pancreatic cancer. Given the challenge of accounting for the many variables that can confound such studies, the major strength of this current analysis is that all patients were treated in a standardized fashion with stratified randomization in a prospective, multi-institutional cooperative group trial setting, allowing for many clinical variables to be tracked in the attempt to understand the true prognostic value of DM with adjuvant treatment.

Although most studies have shown DM to be associated with worse outcomes in resected pancreatic cancer,^{17–20,28–32} some have not.^33–35 Most studies in the advanced stage setting have not shown a prognostic value of DM, although it bears noting that a potential signal is likely more challenging to identify in this population given the uniformly poor prognosis.^35–40 In a large meta-analysis of 17 observational studies including 5407 patients with resectable pancreatic cancer, DM was found to be associated with worse OS (risk ratio, 1.24; 95% CI, 1.05–1.45; P = .01).¹⁸ On stratification, this association was significant only among patients with recent-onset DM (defined as <27 months’ duration before pancreatic cancer diagnosis).However, in this meta-analysis, survival was not adjusted in a standard manner for clinical variables, nor were neoadjuvant or adjuvant treatments reported, complicating interpretation of results. Conversely, and in agreement with the current study, Hart et al³³ undertook a detailed retrospective cohort study of 488 patients who underwent resection for pancreatic cancer at a single institution. Patients who received neoadjuvant treatment were excluded, 86% of patients received adjuvant chemotherapy, and 66% of patients received adjuvant radiation therapy. DM was not associated with OS after adjusting for potential confounders, including age, sex, BMI, weight loss, smoking status, family history of DM, DM treatment, stage, tumor size, tumor grade, number of positive lymph nodes, margin status, and adjuvant chemotherapy (HR, 1.06; 95% CI, 0.811.38; P = .68). Although details of adjuvant treatment were not included, most patients received adjuvant chemotherapy, and median survival in this study was similar to the long-term results of RTOG 9704.²² Taken together with the present study, these results provide evidence that DM is not a prognostic factor in patients with pancreatic cancer who receive adjuvant therapy.

There are several plausible explanations for the inconsistent findings of studies assessing the relationship between DM and pancreatic cancer. First, many prior studies are small, retrospective series or heterogeneous meta-analyses that do not adjust for prognostic factors in pancreatic cancer, including performance status, tumor stage, nodal status, margin status, CA19-9, and neoadjuvant and adjuvant treatment. Furthermore, many of the studies include patients from an earlier era, and they frequently do not consider chemotherapy or radiation therapy details, raising the question of whether results are relevant with modern treatment for pancreatic cancer, as illustrated by Hart et al³³ and the findings detailed here. A potential DM-induced tumor growth–promoting effect that was observed in studies of patients treated before adjuvant therapy was standardized can now be neutralized with more effective adjuvant treatment. In addition, and perhaps most critically, patients with cancer and poorly controlled DM or DM with vascular complications might receive less aggressive cancer-directed therapy than those without DM.⁴¹ This could have introduced significant selection bias in retrospective studies, wherein DM patients might do worse because of differences in cancer-directed therapy and comorbidities rather than the DM itself. All patients in the current study received relatively homogenous treatment on trial, minimizing this important source of bias.

It has been proposed that patients with pancreatogenic DM, or type 3c DM, are clinically distinct from those with long-standing type 2 diabetes.^1,13,42 In contrast to type 2 DM, pancreatogenic DM is characterized by recent onset of DM, normal or increased peripheral insulin sensitivity, and weight loss at the time of DM diagnosis.^42,43 However, the clinical significance of these differences in pancreatic cancer is unclear. Although some studies have identified an association between survival and DM only among patients with presumably pancreatogenic recent-onset DM,^17,18,28,29 others have found an association only among patients with long-standing DM, presumably type 2.^19,31 Interestingly, a prospective observational study of patients with pancreatic cancer who underwent resection attempted to identify distinct DM laboratory-based features based on timing of DM onset. The study authors found that recent-onset DM overlapped with long-standing type 2 DM.⁴⁴ These results argue against a distinct DM tumor phenotype associated with survival in pancreatic cancer. The duration of DM before pancreatic cancer diagnosis was not collected in the current study; however, several multivariable models with subgroups of DM patients by insulin dependence and BMI were assessed to identify a subpopulation of DM patients who had worse survival. A separate DM phenotype that was independently associated with oncologic outcomes was not identified.

There is preclinical evidence that insulin and insulin-like growth factor promote cancer growth.⁹ However, the current analysis found no association between insulin dependence and survival in pancreatic cancer, and prior retrospective studies have been inconsistent. Similar to these findings, a multi-institutional analysis of 2793 patients with pancreatic cancer showed that long-term DM was associated with survival in resectable patients, but insulin dependence was not.¹⁹ A single-institution retrospective study of 1071 resected pancreatic cancer patients found that IDDM was associated with worse OS, whereas NIDDM was not.³² These studies include heterogeneous and non-overlapping patient populations, which might explain the disparate findings. Overall, there is currently no strong clinical evidence that exogenous insulin itself is independently associated with worse outcomes in pancreatic cancer.

The finding that nonwhite race is associated with worse survival has been reported previously in analyses of RTOG 9704,²⁷ and similar race and ethnicity disparities have been identified in other studies of patients with early stage pancreatic cancer.^45,46 Nonwhite patients face significant biases and barriers to care, they are less likely to be referred to specialist care,⁴⁷ and they receive less aggressive treatment.^45,46 However, the disparities observed on this clinical trial suggest an additional unaccounted for barrier or an underlying biologic difference. Investigating and addressing the biologic and socioeconomic drivers of worse outcomes in nonwhite populations is a particularly critical target to improve pancreatic cancer outcomes. Increasing access to clinical trials so that study populations are reflective of society’s racial and ethnic diversity will be essential to develop high-quality, equitable treatment for all patients in the future.

This study has limitations that should be considered. First, the applicability of these findings to the more modern use of neoadjuvant therapies in the potential resectable setting is unknown. In addition, 47% of analyzable patients on RTOG 9704 were excluded from this analysis due to missing DM and insulin use data; however, the baseline clinical characteristics and outcomes were comparable between the excluded and included groups, and in the patients included, OS (n = 191) and DFS (n = 211) events were more than sufficient for statistical power to detect potential associations with DM and insulin use. DM was provider reported in this study, and patients with undiagnosed DM might have been included in the no-DM group. This limitation could contribute to a type II error, where a true association between DM and survival was not observed. As discussed previously, duration of DM before pancreatic cancer diagnosis was not available for analysis. In addition, the association between DM and pancreatic cancer could be related to the degree of DM control.⁴⁸ Although an adjustment was made for insulin dependence, a likely proxy for DM severity, blood glucose level data were unavailable; therefore, a detailed assessment of this relationship was not performed. Oral antihyperglycemic use was not assessed, which might be an unaccounted confounder. Although metformin has been implicated as an anticancer agent⁴⁰ and metformin use was associated with improved OS in patients with pancreatic cancer and DM,⁴⁹ the result of a randomized phase 2 trial of metformin in advanced pancreatic cancer was negative.⁵⁰ Prior studies have reported a wide range of DM prevalence in patients with pancreatic cancer (40%−70%),^6,7 but the DM prevalence was slightly lower (34%) in this analysis. This finding suggests that the population included in this study was overall healthier, likely owing to referral and selection bias, thus limiting generalizability. Finally, the results of this analysis might not be reflective of non-Western populations, where the difference in DM phenotype may be due to differences in DM risk factors.

Conclusion

The results of this study suggest that there is no association between DM or insulin use and survival endpoints among patients with resected pancreatic cancer enrolled in a randomized controlled trial of adjuvant 5-flurouracil or gemcitabine, given both before and after chemoradiation therapy. Nodal involvement, increased CA19-9, and nonwhite race were significant predictors of OS in this study, suggesting that efforts should continue to focus on the development of novel therapies that address the disease biology.

Table 4.

Multivariable Cox regression models for OS and DFS with DM/BMI groups (n = 238)

Endpoint	Adjustment variables	Comparison	Adjusted HR*	95% CI LL	95% CI UL	P value†
OS	DM/BMI	No DM and under-/normal weight	1.00	—	—	—
		DM and under-/normal weight	1.17	0.71	1.93	.54
		No DM and overweight/obese	0.90	0.62	1.31	.58
		DM and overweight/obese	0.99	0.64	1.53	.95
	Treatment arm	RT + 5-FU vs RT + gemcitabine	1.23	0.91	1.67	.17
	Age (y)	≥65 vs <65	0.93	0.68	1.27	.67
	Sex	Male vs female	0.85	0.63	1.16	.30
	Race	Nonwhite vs white	2.18	1.35	3.50	.0014
	KPS score	60–80 vs 90–100	1.17	0.85	1.62	.34
	Tumor diameter, cm	≥3 vs <3	1.22	0.88	1.69	.23
	Tumor location	Non-head vs head	1.10	0.73	1.68	.64
	Surgical margin status	Negative	1.00	—	—	—
		Positive	0.82	0.57	1.18	.28
		Unknown	0.92	0.62	1.37	.69
	Pathologic N stage	N1 vs N0	1.73	1.23	2.45	.0017
	CA19-9	Lewis antigen-negative	1.37	0.98	1.90	.063
		<90 U/mL	1.00	—	—	—
		≥90 U/mL	3.59	2.30	5.61	<.0001
DFS	DM/BMI	No DM and under-/normal weight	1.00	—	—	—
		DM and under-/normal weight	1.01	0.64	1.62	.95
		No DM and overweight/obese	0.89	0.62	1.27	.51
		DM and overweight/obese	0.90	0.60	1.35	.61
	Treatment arm	RT + 5-FU vs RT + gemcitabine	1.12	0.85	1.48	.43
	Age (y)	≥65 vs <65	1.02	0.76	1.36	.92
	Sex	Male vs female	1.01	0.75	1.34	.97
	Race	Nonwhite vs white	1.66	1.05	2.63	.029
	KPS score	60–80 vs 90–100	1.03	0.76	1.40	.85
	Tumor diameter, cm	≥3 vs <3	1.26	0.92	1.73	.16
	Tumor location	Non-head vs head	1.07	0.72	1.59	.74
	Surgical margin status	Negative	1.00	—	—	—
		Positive	0.96	0.68	1.35	.81
		Unknown	0.74	0.51	1.07	.11
	Pathologic N stage	N1 vs N0	1.23	0.88	1.72	.22
	CA19-9	Lewis antigen-negative	1.06	0.77	1.45	.71
		<90 U/mL	1.00	—	—	—
		≥90 U/mL	2.86	1.85	4.41	<.0001

Abbreviations: 5-FU = 5-fluoruracil; BMI = body mass index; CA19-9 = carbohydrate antigen 19–9; CI = confidence interval; DFS = disease-free survival; DM = diabetes mellitus; HR = hazard ratio; KPS = Karnofsky performance status; LL = lower limit; OS = overall survival; RT = radiation therapy; UL = upper limit.

^*HRs with a CI containing 1 indicate no difference between the levels of the listed variable.

^†P value from χ² test using the Cox proportional hazards model.