Methods
The study was approved by the institutional Bioethics Committee at the University of Rzeszow and by the all appropriate administrative bodies. The study was carried out in accordance with ethical standards laid down in Polish regulations and in an appropriate version of the Declaration of Helsinki (as revised in Brazil 2013).
After Bioethics Committee approval, we performed retrospective analysis of existing individual patients’ records in the databases of 3 diabetic and 1 primary care clinics. Inclusion criteria for the ‘case’ group included: cancer diagnosed after diagnosis of type 2 diabetes, at least one HbA1c measurement before or at the time of cancer diagnosis, date of diabetes diagnosis, diabetes treatment, BMI and history of comorbidities available. We identified 203 patients (115 women) with T2DM eligible for analysis. Data analysis covered the period from January 1998 (the first eligible patient with diagnosed cancer) to 30 April 2015. The mean age of diabetic patients at the time of cancer diagnosis was 67.1 ± 9.7 years, and 141 persons were aged ≥65 years. The control group consisted of 203 strictly age- and gender matched subjects with T2DM without cancer. Patients were selected from the same databases in the case-control manner, with the 1:1 ratio. For each ‘case’ patient, an eligible ‘control’ subject with the same gender, and with the nearest possible date of birth was chosen, and any given pair was recruited always at one center to avoid impact of different treatment algorithms used in different clinics. Individuals who died within the analyzed time period but before the moment of data collection were also included into the analysis if their data were available. Data for patients with malignancy were taken from the period preceding cancer diagnosis (index time). Data for the ‘control’ subjects were assessed from the same index time, i.e., if the ‘case’ patient had cancer diagnosed in April 2009, the data for his/her comparator were taken from the same period.
In both groups metabolic control of diabetes (mean HbA
1c from the preceding up to 3 years before index time, if available), diabetes duration, antidiabetic medications (also from the preceding up to 3 years, and each drug was classified as “used” if it was taken for at least 6 months), mean insulin dose from the preceding 6 months, and duration of insulin treatment up to the moment of cancer diagnosis were analyzed. Also place of residence (rural, small cities or urban), smoking habits (current, former or never smoker), presence of comorbidities (hypertension, hyperlipidemia and cardiovascular disease), BMI and use of aspirin were also included into the analysis. All included patients were of Caucasian ethnicity. Detailed characteristics of both groups is presented in the Table
1.
Table 1
Characteristics of the case and control groups
Age at index time (years) (mean ± SD) | 67.1 ± 9.7 | 67.1 ± 9.7 | - |
< 65 years (n) | 62 (30.5 %) | 62 (30.5 %) | |
≥ 65 years (n) | 141 (69.5 %) | 141 (69.5 %) | |
Gender | | | - |
male (n) | 88 (43.3 %) | 88 (43.3 %) | |
female (n) | 115 (56.7 %) | 115 (56.7 %) | |
Place of residence | | | 0.264 |
rural (n) | 38 (18.7 %) | 44 (21.7 %) | |
cities <50,000 inhabitants (n) | 27 (13.3 %) | 36 (17.7 %) | |
cities >50,000 inhabitants (n) | 138 (68.0 %) | 123 (60.6 %) | |
Smoking habits | | | 0.838 |
never smokers (n) | 115 (56.7 %) | 117 (57.6 %) | |
former smokers (n) | 54 (26.6 %) | 57 (28.1 %) | |
current smokers (n) | 33 (16.3 %) | 29 (14.3 %) | |
unknown status (n) | 1 (0.5 %) | - | |
BMI (kg/m2) (mean ± SD) | 30.8 ± 5.3 | 30.1 ± 4.7 | 0.103 |
Comorbidities |
cardiovascular disease (n) | 50 (24.6 %) | 58 (28.6 %) | 0.369 |
hypertension (n) | 177 (87.2 %) | 179 (88.2 %) | 0.763 |
hyperlipidemia (n) | 153 (75.4 %) | 159 (78.3 %) | 0.480 |
Diabetes duration (years) (mean ± SD) | 10.7 ± 7.4 | 10.3 ± 8.1 | 0.262 |
HbA1c (%) (mean ± SD) | 7.39 ± 1.21 % | 7.30 ± 1.06 % | 0.755 |
Antidiabetic medications |
metformin (n) | 126 (62.1 %) | 167 (82.3 %) | <0.001 |
sulfonylurea (n) | 83 (40.9 %) | 101 (49.8 %) | 0.090 |
acarbose (n) | 18 (8.9 %) | 14 (6.9 %) | 0.581 |
DPP-4 inhibitor (n) | 11 (5.4 %) | 6 (3.0 %) | 0.322 |
insulin (n) | 110 (54.2 %) | 81 (39.9 %) | 0.005 |
insulin dose (IU/kg/24 h) (mean ± SD) | 0.59 ± 0.31 | 0.53 ± 0.24 | 0.407 |
insulin duration (years) (mean ± SD) | 6.2 ± 5.6 | 6.5 ± 4.9 | 0.529 |
Aspirin use (n) | 102 (52.6 %) | 102 (52.8 %) | 0.957 |
unknown status | 9 | 10 | |
Current place of residence was taken into analysis with the exception of patients, who moved in the last year. In such cases a previous place of residence was taken into account. Patients were considered as a current, former or never smokers according to definition stated by Centers for Disease Control and Prevention [
20]. Hypertension was considered if blood pressure values were ≥140 mmHg for systolic, and/or ≥90 mmHg for diastolic blood pressure, or if antihypertensive medications were used. Hyperlipidemia was recognized if LDL-cholesterol level was ≥2.6 mmol/L (100 mg/dl) and/or triglycerides concentration was ≥1.7 mmol/L (150 mg/dl), or hypolipemic drugs were used. Cardiovascular disease was confirmed if the patient had a history of non-fatal myocardial infarction, hospitalization for acute coronary syndrome, non-fatal stroke, revascularization or amputation.
Statistical analysis of the data was performed using SigmaPlot for Windows version 12.5 (Systat Software Inc., San Jose, CA, USA). The analysis was performed in 2 stages. In the first stage comparison of the two groups was made. The continuous data were analyzed using an unpaired two-tailed Student’s t-test or by a Mann-Whitney rank sum test where appropriate. The categorical data were compared using χ
2 test. In the second stage patients were divided into subgroups according to BMI (<25.0, 25.0–29.9, 30.0–34.9 and ≥35.0 kg/m2), diabetes duration (<5.0, 5.0–9.9, 10.0–14.9 and ≥15.0 years), insulin dose (no insulin, <0.50 and ≥0.50 IU/kg) and duration of insulin treatment (no insulin, <5.0, 5.0–9.9 and ≥10.0 years). For the assessment of the effect of treatment or analyzed risk factors on cancer occurrence OR (odds ratios) and 95 % CI (confidence intervals) were calculated in univariate and in multiple logistic regression models. A P value <0.05 was considered statistically significant.
Discussion
Although the link between diabetes and malignant neoplasms is well known and many site-specific cancers are more prevalent in diabetic patients [
3‐
11], the exact risk factors of cancer in diabetic population have not been fully determined. Also in our study prevalence of breast, colorectal, uterine, kidney, pancreatic and gastric cancers among diabetic patients was higher, while proportion of patients with prostatic and lung cancers was lower than observed in the general Polish population [
21], which is in line with majority of other observations [
4‐
7,
22].
Our retrospective, multicenter, case-control study revealed that the following diabetes-related factors may be associated with cancer occurrence: poor metabolic control, obesity and antidiabetic medications use. Importantly, age and gender were not included into risk analysis, because the patients were strictly matched according to them.
Data regarding association between HbA
1c level and risk of malignancy are divergent. Some studies showed continuous relationship between increasing HbA
1c level and cancer risk [
23‐
25], others found non-linear association [
26,
27] (the second one only among women), or elevated risk of malignancy above the specified HbA
1c threshold of 7.5 % (colorectal cancer) [
28]. In the study by Miao Jonasson et al. no relationship between HbA
1c level and cancer was found [
29]. In the recent meta-analysis significant association between chronic hyperglycemia and elevated risk of several types of malignancies with the exception of prostate cancer was demonstrated [
30]. We took to the analysis the mean HbA
1c level from a longer period of time preceding cancer diagnosis considering it as better reflecting the overall exposure to glucose than single measurement. In our study risk of malignancy was rising rapidly at the HbA
1c level equal or above 8.5 %. de Beer and Liebenberg found similar HbA
1c threshold for the risk of colorectal and breast cancers [
30]. In our study these cancers were 37 % of all malignancies, which partly explains our findings.
Some of the studies cited above were performed in diabetic populations [
23‐
25,
29], while other were conducted in both non-diabetic and diabetic subjects [
26‐
28]. Although deleterious effect of glucose and elevated cancer risk as a function of HbA
1c may be seen also in a higher versus lower values within a normal range, it can be more pronounced at high glucose concentrations. The higher threshold found in our study can be explained by the fact that prolonged hyperglycemia leads to formation of ROS (reactive oxygen species) and to accumulation of AGEs (advanced glycation end products). AGEs stimulate their specific receptor RAGE (receptor for AGE) which leads to increased inflammation through the activation of the nuclear transcription factor NF-κB and formation of ROS in the cells, which have mutagenic effect and cause DNA damage. This pathway is considered to play an important role in both inflammation and carcinogenesis [
31,
32]. Chronic hyperglycemia activates these biological processes and thus a higher HbA
1c value may reflect higher cancer risk in poorly controlled diabetes. In addition, glucose serves is a primary energy source for cancer cells, and higher glucose concentrations may accelerate cancer growth [
33,
34].
Data regarding effect of diabetes duration on cancer risk are scarce. Johnson et al. and the Danish registry study documented highest cancer risk occurring immediately after diabetes diagnosis [
35,
36]. However, Li et al. demonstrated opposite results, with the lowest risk of malignancy in the first 5 years from the onset of diabetes, and the highest cancer risk among patients with diabetes lasting over 15 years [
37]. In our study there was a clear tendency towards lowest cancer risk in the first years, and the highest risk between 10 and 15 years after diagnosis of diabetes. The increasing risk of cancer incidence with duration of diabetes can be explained by cumulative effect of hyperglycemia, use of insulin, and weight gain developing in the course of the disease. In addition, increasing age itself is strongly associated with increasing cancer risk both in diabetic and non-diabetic population [
21,
38].
Impact of antidiabetic medications on cancer risk has been widely discussed in recent years. As a progressive disease type 2 diabetes requires intensification of treatment over time, from lifestyle modification through oral therapy in different regimens, to insulin treatment. Thus, a clear impact of antidiabetic medications on cancer risk is difficult to determine.
Our study demonstrated highly significant reduction of cancer risk among metformin users, which is in line with many [
17,
19,
39,
40] but not all [
41] studies. The mechanisms of the anti-cancer effect of metformin include inhibition of cancer cells growth through stimulation of AMPK (AMP-activated protein kinase) and its regulator LKB1 (liver kinase B1), which is known to act as a tumor suppressor protein. In addition, metformin also directly inhibits mTOR (mammalian target of rapamycin) pathway [
42] and may have a role of immunomodulator [
43].
Sulfonylurea (SU) use in our study did not show relationship with cancer risk. Data from other observations are divergent. Soranna et al. demonstrated neutral effect of SU derivates on the risk of malignancy [
39]. This was confirmed by Monami et al. with the exception of gliclazide, which appeared to be protective [
44]. On the other hand, recently Thakkar et al. revealed increased cancer risk among SU users [
40].
Current evidence from observational studies indicate harmful effect of insulin on the cancer risk [
18,
37]. In our study insulin use was associated with a dose-dependent elevated risk of malignancy. Similar relationship was also observed by Holden et al. [
45]. Regarding duration of insulin treatment, risk of malignancy in our observation was highest in the first 10 years of treatment, and become insignificant with a longer insulin use. This phenomenon can be explained by increased cancer and also coronary heart disease mortality observed among diabetic patients treated with insulin [
46]. Other studies showed increased risk of malignancy associated with insulin use after 4 years of insulin treatment [
18]. In general, insulin is a potent growth stimulating hormone acting through insulin and IGF-1 (insulin-like growth factor-1) receptors [
34,
47,
48]. On the other hand, landmark prospective studies in type 2 diabetes did not confirm elevated risk of malignancy associated with more intensified treatment [
49]. Also Outcome Reduction with an Initial Glargine Intervention (ORIGIN) trial did not demonstrated raised cancer risk among insulin users [
50]. However, these studies were of limited duration, also relatively low doses of insulin were used in the ORIGIN trial.
The number of patients treated with other antidiabetic medications in our study was small, therefore we were unable to determine their relationship with cancer risk. Acarbose (the α-glucosidase inhibitor) is not popular due to its well-known side effects. DPP-4 inhibitors are not reimbursed in Poland and their utilization is low. In our study, after adjustment to all analyzed variables they demonstrated significant association with cancer risk. However, it is worth to notice that only 17 patients were treated with these agents and in this case random effect cannot be excluded. Pioglitazone and SGLT-2 inhibitors are also not reimbursed and, in addition, pioglitazone was not available on market in Poland up to 2014, thus number of patients using these medications is extremely low.
Obesity is a well-recognized risk factor of several types of cancer [
51,
52]. Our study confirmed association between obesity and risk of malignancy in diabetic population. Obese patients, especially with severe obesity (BMI ≥35 kg/m
2), had significantly higher risk of cancer occurrence compared with non-obese subjects. It should be remembered that with cancer development body weight frequently decreases, the fact which may mar analysis of the results. Insulin resistance, hyperinsulinemia, elevated levels of IGF-1, inflammation, increased sex hormones bioavailability and hyperglycemia are considered to be responsible for increased cancer risk in obese individuals [
53].
Smoking is known to be associated with elevated risk of several site-specific cancers, especially lung cancer [
54]. Interestingly, in our study number of never, former and current smokers was not significantly different in case and control groups, and smoking was not associated with elevated overall cancer risk. However, not surprisingly, number of current and former smokers was significantly higher in patients with lung cancer related to their comparators.
For other analyzed variables, including place of residence, presence of comorbidities and aspirin use relationship with risk of malignancy was not revealed.
The limitations of our study include its retrospective and observational design, and relatively small sample size which has influenced the statistical power of our findings, and has not allowed to demonstrate other possible relationships for which positive trends were observed. Also immortal time, time-window and time-lag biases despite our best efforts cannot be excluded [
55]. Another limitation is low number of users of oral drugs other than sulfonylurea derivatives and metformin. In addition, all main cancer risk factors confirmed in the study e.g., insulin use, HbA
1c level and obesity heavily influence one another which may also confound the results. And finally, due to the characteristics of Polish society, only patients of Caucasian ethnicity were included and our findings may have not be applicable for persons from other ethnic groups.
This study has also several strengths. One of them is its case–control design with strictly matched pairs of case subjects and their comparators, with each pair taken from the very same center. The study was based on a high-quality data sources using samples with a long follow-up time (mean time from diabetes diagnosis to index time has exceeded 10 years) and extensive covariate information, including the date of the onset of diabetes, date of cancer diagnosis, treatment details, metabolic control and other common risk factors, which allowed to explore the relationship between T2DM and cancer risk.
Acknowledgements
Not applicable.