Diabetes Among Finnish Pancreatic Cancer Patients: Prevalence, Time Interval Between Diabetes and Cancer Diagnosis, and Overall Survival

ABSTRACT

Introduction

New‐onset diabetes (NOD) potentially associated with pancreatic cancer (PC) requires further investigation to assess its potential as a tool for PC screening.

Material and Methods

In this longitudinal nationwide register study, which includes 14,712 PC patients and 18 years of follow‐up time (2000–2018), we aimed to assess the prevalence of diabetes by using a large national registry. Survival was analyzed according to diabetes status and duration (long‐term diabetes, i.e., DM diagnosed > 3 years prior to PC, NOD, i.e., DM diagnosed 0–3 years prior to PC, or DM diagnosed after PC diagnosis) and treatment status (curative intent vs. palliative) using multivariable Cox regression analysis.

Results

Overall, 18% of PC patients had diabetes. The median time interval between DM and PC diagnoses was 4.8 months. In the multivariable Cox regression analysis, only DM diagnosed after PC was associated with worse survival, both in the palliative intent treatment group (HR 1.12, 95% CI 1.03–1.23, p = 0.011) and in the curative intent treatment group (HR 1.44, 95% CI 1.14–1.81, p = 0.002). Among patients diagnosed with diabetes prior to PC, neither long‐term nor NOD significantly affected survival.

Conclusions

Most of the diabetic PC‐patients have been diagnosed with DM within a close proximity of cancer diagnosis. The progression from diabetes to clinically detectable cancer seems to be rapid—less than 6 months. Given the aggressive nature of PC and the fact that NOD can be an early sign, screening may need to begin at the time of diabetes diagnosis. The fact that a DM diagnosis made after a PC diagnosis is associated with worse survival may suggest that pancreatic tumors that cause hyperglycemia are particularly aggressive.

Abbreviations

Cur

curative intent

DM

diabetes mellitus

FCR

Finnish Cancer Registry

HILMO

the Care Register for Health Care in Finland

ICD

the International Classification of Diseases

NCSP

Nordic Classification of Surgical Procedures

NOD

new‐onset diabetes

PC

pancreatic cancer

TNM

tumor, node, and metastasis

1. Introduction

Pancreatic cancer (PC) has a dismal prognosis since most patients have a reported median overall survival of less than 6 months, regardless of disease stage [1, 2]. Aggressive tumor biology, poor response to systemic oncologic therapies and early metastatic spread, combined with PC's asymptomatic course of disease, lead to late diagnosis [3, 4]. Moreover, even though PC is not the most common gastrointestinal cancer, its incidence is rising by an annual increase of approximately 0.5% [5]. The only potential curative treatment for PC is radical surgical resection combined with systemic therapy, but only 10%–15% of PC patients are eligible for surgery with curative intent [6]. One of the fundamental approaches to improve survival is to identify and screen high‐risk populations [7].

Various risk factors for PC have been identified. Cigarette smoking and excessive alcohol consumption are among potential lifestyle hazards [8, 9]. A positive family history (i.e., more than one first‐degree relative with PC) is a strong risk factor for PC, in addition to various pathological gene variants, such as those in CDKN2A [10, 11]. Diabetes mellitus (DM), especially new‐onset diabetes (NOD), is a well‐known risk factor for PC [12, 13]. However, it is important to note that in some newly diagnosed diabetics, NOD is often an early manifestation of the underlying pancreatic pathology, rather than a risk factor per se [14].

Current guidelines recommend screening high‐risk patients [15, 16], since PC screening may potentially lead to diagnosis at an earlier stage, and thus to improved survival [17, 18]. Currently, there are no surveillance recommendations for DM [16], and the potential role of NOD in PC screening and the clinical features of PC‐related NOD need further study.

To assess the current prevalence of DM and the time interval between DM and PC diagnosis, we designed a longitudinal register study utilizing high‐quality population registry data. Firstly, we studied the overall prevalence of diabetes among PC patients in Finland in groups based on curative vs. palliative treatment intent. Furthermore, we studied the time interval between the DM and PC diagnoses. Finally, we explored whether survival of diabetic PC patients differs from non‐diabetic PC patients, both overall and according to treatment intent and diabetes duration.

2. Material and Methods

2.1. Diabetes Status and Choice of Treatment

In this nationwide register study, individuals with a PC diagnosis between the years 2000 and 2016 were identified from the finnish cancer registry (FCR). Follow‐up was extended to 2018. Diabetes and treatment status were defined by using information from the care register for health care in finland (HILMO). HILMO is a data collection system maintained by Statistics Finland and the Finnish Institute for Health and Welfare (THL), and it includes data on both in‐ and outpatient care. HILMO provided nationwide data on diabetes‐ and PC‐related healthcare visits and treatment periods. Diabetes status was defined according to the corresponding ICD‐codes for diabetes in the HILMO data. Surgical procedures were identified based on the nordic classification of surgical procedures (NCSP). Oncological treatments were identified from the HILMO data for PC treatment periods by using oncological treatment procedure codes (NCSP). Though in 71% of patients, data on stage was available in the FCR registry, it was inconsistently classified and only partly reliable; therefore, we did not include cancer stage in the analysis [19].

Patients were further divided into subgroups to investigate the prevalence of DM in PC patients overall and according to treatment intent. The options used were treatment with curative or palliative (Cur vs. Pal) intent, and whether a DM diagnosis was present or not (DM+ vs. DM‐). Pancreatoduodenectomy, distal pancreatic resection, total pancreatectomy and unspecified pancreatic resections were classified as treatments with curative intent. Open surgery and minimally invasive operations were grouped together. Patients without these procedure codes were classified as being treated with palliative intent. Patients who underwent ambulatory explorative or biliodigestive bypass procedures, but not pancreatic surgery with curative intent, were also included in the palliative group. The final four groups were: DM+/Cur, DM+/Pal, DM‐/Cur and DM‐/Pal. Figure 1 illustrates the selection criteria for different study groups in more detail.

FIGURE 1.

The selection criteria for study groups according to diabetes status and treatment intent.

2.2. Time Interval Between DM and PC Diagnosis

To analyze the time interval between DM and PC diagnosis, all patients with a PC diagnosis between 2004 and 2016 were identified. Patients were further divided depending on whether they had received their DM diagnosis before or after the PC diagnosis. The earliest healthcare visit linked to a DM record was defined as the DM diagnosis date.

2.3. Statistical Analysis on Survival Between Subgroups

Survival was analyzed by comparing the subgroups based on DM status, duration, and treatment intent. The index date for survival analysis was the cancer diagnosis date. Survival was estimated using Kaplan–Meier curves in non‐diabetic PC patients and diabetic PC patients who had received their DM diagnosis prior to their cancer diagnosis. The comparison between these two survival functions was assessed using the log‐rank test. In cases where diabetes was diagnosed after the cancer diagnosis, the Simon–Makuch method was used to estimate the survival functions of diabetic and non‐diabetic PC patients. The comparison between these two survival functions was assessed with the Mantel–Bryar test.

Multivariable Cox proportional hazards regression was used to analyze survival in PC patients treated with either curative or palliative intent. DM duration was also taken into consideration. Duration was grouped into long‐term diabetes, i.e., DM diagnosed > 3 years prior to PC; NOD, i.e., DM diagnosed 0–3 years prior to PC; or DM diagnosed after PC diagnosis. DM diagnosis after PC and oncological treatment were included as a time‐dependent covariate. The Cox regression assumption of constant hazard ratios over time (proportional hazards) was assessed with the Schoenfeld residuals plotted over time and testing for a trend. Split time axis was used if necessary to account for deviances from the proportional hazard assumption. Patients with more than 1 year between PC diagnosis and surgery with curative intent or oncological treatment were removed. Interactions in multivariable models were considered. The Cox model accounts for the confounders (age, sex, oncological treatments and Charlson comorbidity index).

Participants with missing data on any of the variables included in the analysis were excluded from the respective analysis. All statistical analyses were made using IBM's SPSS Statistics for Macbook, version 27.0.1 (Armonk, NY: IBM Corp) and R (version 4.2.0 with survival package (Therneau T, 2022)).

3. Results

3.1. The Prevalence of Diabetes in Finnish Pancreatic Cancer Patients and Choice of Treatment

A total of 14,712 individuals were included in the analysis. The overall sex distribution was 6903 (47%) males and 7809 (53%) females. The median age was 73 years (range 19–103).

Diabetic patients in the two different treatment groups (Cur vs. Pal) are displayed in Figure 1. Out of the PC patients, 2681 (18%) had a diagnosis of DM. Two hundred and eighty out of 2681 (10%) diabetic PC patients and 902 out of 12,031 (7.5%) non‐diabetic PC patients were treated with curative intent, respectively. Based on the ICD‐10 codes, 118 (4.4%) were reported to have Type 1 diabetes, including both complicated and uncomplicated disease stages. Type 2 diabetes without complications was found in 1686 patients (62.9%), whereas 633 patients (23.6%) had Type 2 diabetes with complications. The remaining 244 patients had other or unspecified types of diabetes diagnoses.

3.2. Time Interval Between DM and PC Diagnosis

Between 2004 and 2016, a total of 1606 (60%) patients had received their DM diagnosis prior to their PC diagnosis, and 718 were diagnosed with diabetes posterior to their PC diagnosis (40%). The median time between their DM diagnosis and their PC diagnosis was −0.4 years/4.8 months (25%–75% quartiles −3.08–0.03). There was an increase prior to the PC diagnosis and a rapid decline posterior to the PC diagnosis (Figure 2). Five hundred and forty‐eight patients were diagnosed with DM 6 months prior to their PC diagnosis.

FIGURE 2.

Diabetes diagnoses done before and after pancreatic cancer diagnosis (PC). This histogram illustrates the number of diabetes diagnoses made every six months on the y axis, with time measured in years from a baseline marked at the point of PC diagnosis (0 on the x axis).

3.3. Survival

The overall median for survival time was 4.24 months (4.1–4.4, 95% CI, p = 0.007).

For non‐diabetic PC patients, the median survival time was 4.4 months (95% CI 4.2–4.6), while it was 3.4 months (95% CI 3.0–3.7, p < 0.001) for diabetic PC patients.

In the survival analysis including patients treated with palliative intent (i.e., groups DM+/Pal vs. DM‐/Pal), diabetic PC patients had a worse prognosis with a median survival of 2.9 months (95% CI 2.6–3.2) vs. 3.7 months (95% CI 3.6–3.9, p < 0.001) for non‐diabetic PC patients. When comparing the patients treated with curative intent (i.e., groups DM+/Cur and DM‐/Cur), there was no statistically significant difference in survival, with median survival of 2.2 (95% CI 1.98–2.37) years in diabetic PC patients vs. 1.8 years (95% CI 1.32–2.35, p = 0.267) in non‐diabetic PC patients, respectively. Figure 3 presents the Kaplan–Meier survival functions and Simon–Makuch plots for survival.

FIGURE 3.

Survival functions according to diabetes status and treatment options.

In the multivariable Cox regression analysis, only the DM diagnosed after PC was associated with worse survival both in the palliative intent treatment group (HR 1.12, 95% CI 1.03–1.23, p = 0.011) and the curative intent treatment group (HR 1.44, 95% CI 1.14–1.81, p = 0.002). Among patients diagnosed with DM prior to PC, neither long‐term nor NOD significantly affected survival. Estimates from the Cox regression model are presented in Table 1.

TABLE 1.

Cox multivariable regression analysis on the DM‐diagnosis made both prior and after PC‐diagnosis and its effect on the overall survival.

	Curative intent treatment			p	Palliative intent treatment			p
	HR	lower 95% CI	upper 95% CI		HR	lower 95% CI	upper 95% CI
Age (median)	1.24	1.05	1.48	0.014	1.61	1.51	1.71	< 0.001
Sex	0.99	0.85	1.15	0.850	0.85	0.80	0.90	< 0.001
CCI score 1 vs. 0	0.90	0.73	1.11	0.317	1.04	0.99	1.10	0.142
CCI score 2 vs. 0	0.96	0.77	1.20	0.738	1.06	1.00	1.12	0.059
CCI score 3 vs. 0	1.08	0.82	1.41	0.594	1.22	1.15	1.30	0.000
Diabetes diagnosis 0–3 years before PC	1.25	0.95	1.65	0.107	1.07	0.99	1.15	0.085
Diabetes over 3 years before PC	1.16	0.74	1.82	0.515	1.08	0.99	1.19	0.079
Diabetes diagnosis after PC diagnosis	1.44	1.14	1.81	0.002	1.12	1.03	1.23	0.011
Oncological treatment within 1 year after PC diagnosis	0.50	0.37	0.69	< 0.001	0.83	0.77	0.9	< 0.001
Oncological treatment > 1 year after PC diagnosis	0.91	0.76	1.08	0.282	1.69	1.51	1.88	< 0.001

Note: Age is a categorical variable divided at the median age (71.4 years). For sex male is the reference level. Significant interactions were observed between oncological treatment and age (Age*Oncological 0–1 years 0.64, 0.56–0.72 95% CI, p = < 0.001 and Age*Oncological > 1 years 0.61, 0.51–0.73 95% CI, p < 0.001) and between age and sex (1.18, 1.09–1.28 95% CI, p < 0.001) among patients treated with palliative intent. These interactions were included in the palliative model. The time axis was split by oncological treatment to address violations of the proportional hazard's assumption in the Cox regression model.

Abbreviations: CCI, Charlson Comorbidity Index (CCI = 0 is the reference level); CI, Confidence interval.

4. Discussion

Interestingly, a less than 20% prevalence of DM in PC patients is observed in our study, which is lower than that reported in comparable studies. Amri et al. presented that 38% of their 197 PC patients had diabetes [20], and the corresponding numbers in the study by Hsieh et al. was 36% out of 3200 PC‐patients [21]. However, this discrepancy may reflect differences in the number of PC patients and the definitions used to identify diabetes. The increased frequency of DM diagnoses around the time of PC, and the fact that a substantial proportion of these diagnoses were made either shortly before or after the cancer diagnosis, suggest that a significant number of DM cases could represent pancreatogenic diabetes that is actually caused by the tumor itself [22].

New‐onset diabetics seem to be an effective group for screening, considering the large proportion of DM/PC patients who are newly diagnosed diabetics [23]. Given the high prevalence of DM in the general population [24], compared with the relatively low incidence of PC [25], surveying all newly diagnosed diabetics for PC is unlikely to be justified. Older age at DM onset, weight loss and a worsening glucose status have been proposed as possible additional clinical clues for PC‐related DM [26, 27, 28]. Over the last few years of NOD/PC research, interesting tools for screening have emerged, such as the END‐PAC model, that takes into account weight loss and changes in blood sugar levels in NOD patients. This model was first introduced by Sharma et al. [29] and later reviewed by Hajibandeh et al. [30]. However, the time interval between DM and PC was relatively short in our study, with a median of 4.8 months. Given the biological aggressiveness of PC, monitoring newly diagnosed DM patients for weight loss and blood sugar levels could delay the diagnosis of an underlying pancreatic pathology.

Several previous studies have reported how DM impairs survival among PC patients. The systematic review and meta‐analysis by Mao et al. [31] and the study by Hank et al. [32] found that diabetes is associated with higher mortality in all PC patients. The role of DM duration in survival has also been studied, and at least one meta‐analysis has been conducted on this topic: Lv et al. [33] showed that only NOD was associated with worse survival.

In our study, although the Kaplan–Meier survival estimates suggested poorer survival among DM patients, this association was not so apparent in Cox regression analysis after adjustment for other relevant covariates, for we found a negative effect of DM on survival only among those diabetics who were diagnosed with DM after their PC diagnosis. In DM diagnoses done prior to PC, diabetes duration (i.e., long‐term DM vs NOD) did not seem to affect survival. It is challenging to explain why only DM diagnosed after the cancer diagnosis was associated with worse survival, rather than diabetes diagnosed relatively shortly before the PC. However, if these patients who developed diabetes after the cancer diagnosis could be considered to have DM secondary to the PC, one might interpret this to mean that PC causing hyperglycemia is a more aggressive tumor. Nevertheless, caution should be exercised in this interpretation.

The strengths of this study include its very large sample size with data extracted from high‐quality nationwide registers. The quality of FCR is considered very good in solid tumors with a completeness of over 90%, and approximately 97% of these are histologically verified [34]. The HILMO data are maintained by the Finnish Institute for Health and Welfare (THL), and they comprise a significant portion of health information, since THL has a legislative right to utilize information extracted from both public and private health care providers.

A clear limitation of this study is that the oncological stage could not be utilized in the multivariable analysis. There was furthermore no reliable data available on simultaneous vascular procedures during pancreatic surgeries, and thus this could not be taken into consideration in survival, nor could we distinguish between local and locally advanced cancers.

Information regarding smoking status, BMI, and use of antidiabetic drugs was also not available for this study. The lack of these important prognostic factors may cause some distortion in survival rates. Additionally, information on the pancreatitis history of patients as well as their family history of PC was not available. It should also be noted that as laboratory values and other diagnostic methods for diabetes were not available, some individuals may have been misclassified as non‐diabetic, potentially leading to misclassification bias.

Pancreatic neuroendocrine tumors were excluded, but other than that, the FCR register does not offer information on histological subtype, and thus a small portion of rare cancers of the exocrine pancreas may be included (such as adenosquamous and acinar cell carcinoma). Furthermore, HILMO is based on electronic patient information systems, the implementation of which was gradually established in Finland in the early 2000s. Thus, for some patients, the actual duration of diabetes is unreliable. The exclusion of participants with missing data may have introduced selection bias.

5. Conclusions

In this large nationwide cohort, approximately one‐fifth of PC patients have a diagnosis of DM, and most of them are diagnosed within a few years of their PC diagnosis, or nearly immediately after their PC diagnosis. This close temporal relationship is consistent with previous findings that demonstrate how NOD can present itself as an early symptom of PC. The progression from diabetes to clinically detectable cancer seems to be rapid. The fact that a DM diagnosis made after a PC diagnosis is associated with worse survival may suggest that pancreatic tumors that cause hyperglycemia are particularly aggressive. However, in patients with pre‐existing diabetes, neither long‐term nor NOD influenced overall survival.

6. Future Directions

We recommend future research to look for clinical differences compared to conventional diabetic patients as early as at the time of diabetes diagnosis, since we observed a relatively short time interval between DM and PC.

Novelty and Impact: Diabetes mellitus (DM), especially new‐onset diabetes (NOD), is a known risk factor for pancreatic cancer (PC). New‐onset diabetics are an attractive group for PC screening since we have observed that in some cases, NOD is an early manifestation of PC. However, screening diabetics for PC presents a clear problem due to the vast differences in the prevalence of these two diseases: DM is much more common than PC, and the prevalence of DM within the pancreatic population is relatively low. In addition, in our study, we found that the time interval between DM and PC is relatively short, and given the biological aggressiveness of PC, we encourage future researchers to search for differences between PC‐associated DM and conventional, adult‐type DM, ideally simultaneously with the DM diagnosis. This study offers a comprehensive analysis of the prevalence of diabetes among PC patients and how diabetes status and diabetes duration affect survival. The data are extracted from reliable and comprehensive national registries, and the follow‐up time is considerably long.

Author Contributions

Suitiala Roosa: conceptualization, methodology, writing – original draft, visualization, formal analysis. Aaltonen Panu: writing – review and editing. Mustonen Harri: formal analysis, writing – review and editing. Sund Malin: conceptualization, methodology, writing – review and editing, supervision. Seppänen Hanna: conceptualization, methodology, writing – review and editing, supervision.

Funding

This work was supported by Sigrid Juséliuksen Säätiö, TYH2022329. The study was funded by Helsinki University Library.

Ethics Statement

The National Institute for Health and Welfare, Statistics Finland, and the Helsinki University Hospital approved the study protocol.

Consent

This registry‐based study utilized data from the Finnish Cancer Registry (FCR) and the Finnish Care Register for Health Care (HILMO). The study was approved by The Finnish Institute for Health and Welfare (THL/4492/14.06.00/2022) and the University of Helsinki (HUS/12199/2022). Due to the secondary use of registry data, individual informed consent was not required.

Conflicts of Interest

The authors declare no conflicts of interest.

Data Availability Statement

The data that support the findings of this study are available on request from the corresponding author. The data are not publicly available due to privacy or ethical restrictions.