Risk of cancer in a large cohort of U.S. veterans with diabetes

Abstract

Prior studies of cancer risk among diabetic men have reported inconsistent findings. The aim of the current investigation was to assess the risk of cancer among a large cohort (n=4,501,578) of black and white U.S. veterans admitted to Veterans Affairs hospitals. The cancer risk among men with diabetes (n=594,815) was compared to the risk among men without diabetes (n=3,906,763). Poisson regression was used to estimate adjusted relative risks (RRs) and 95% confidence intervals (CIs). Overall, men with diabetes had a significantly lower risk of cancer (RR=0.93, 95%CI=0.93-0.94). Men with diabetes, however, had increased risks of cancers of the liver (RR=1.95, 95%CI=1.82-20.9), pancreas (RR=1.50, 95%CI=1.42-1.59), biliary tract (RR=1.41, 95%CI=1.22-1.62), colon (RR=1.20, 95%CI=1.16-1.25), rectum (RR=1.12, 95%CI=1.07-1.18), and kidney (RR=1.09, 95%CI=1.03-1.16), as well as leukemia (RR=1.14, 95%CI=1.08-1.21) and melanoma (RR=1.13, 95%CI=1.03-1.24). In contrast, men with diabetes had decreased risks of cancers of the prostate (RR=0.89, 95%CI=0.87-0.91), brain (RR=0.91, 95%CI=0.81-0.99), buccal cavity (RR=0.85, 95%CI=0.82-0.89), lung (RR=0.79, 95%CI=0.77-0.80), esophagus (RR=0.77, 95%CI=.072-0.82), and larynx (RR=0.76, 95%CI=0.71-0.80). These findings indicate that black and white men with diabetes are at significantly lower risk of total cancer and of two of the most common cancers among U.S. males; lung and prostate cancers. These decreased risks were offset, however, by increased risks of cancer at several sites. Hyperinsulinemia may explain the increased risks of the digestive cancers, while lower testosterone levels, in the case of prostate cancer, and higher BMI, in the case of lung cancer, may explain the decreased risks of those tumors.

Introduction

In the United States, it is estimated that 23.5 million adults have diabetes and another 57 million adults are at increased risk of developing diabetes due to impaired fasting glucose levels ¹. Non-Hispanic black adults have a notably higher prevalence of diabetes (14.7%), than do non-Hispanic white adults (9.8%). The numbers of affected adults are projected to increase in the future as the prevalence of diabetes has been estimated to rise by 165% by 2050 ². Given the large number of individuals with diabetes and the expectation that the number will increase in the future, concerns about morbidities subsequent to diabetes have increased. Persons with diabetes are known to be at increased risk of blindness, kidney damage, cardiovascular disease and of requiring lower-limb amputations ¹. Whether diabetics are also at increased risk of cancer has been debated for a number of years ³.

Several cohort studies from the U.S. ^4-8 as well as other countries ^9-13, have previously examined cancer risk or cancer mortality among men with diabetes. Increased risk or mortality of total cancer has been reported by some studies ^{6, 8, 11-13} but others have reported no association or decreased risk or mortality ^{4, 7, 9-10}. As interpretation of prior U.S. studies may have been hampered by inadequate size or inability to account for race, a very large cohort study of more than 4.5 million men was conducted among black and white U.S. military veterans.

Methods

The study cohort was identified using hospital discharge records dated between July 1, 1969 and September 30, 1996 from 142 nationwide United States Veterans Affairs (VA) hospitals. Based on US census data, there were an estimated 30 million US veterans entitled to admission to a VA hospital during the study period ¹⁴. The study population included all black (n=832,334) and white (n=3,669,244) male veterans between the ages of 18 and 100 years, hospitalized at least once during the study period. Other ethnic/racial groups and females were not included in the study due to small numbers. A diagnosis of diabetes was identified by code 250 of the International Classification of Diseases, Eighth or Ninth Revision (ICDA-8 or ICD-9-CM).

To estimate the risk of cancer, the cohort was followed from 1 year after first hospital discharge until the diagnosis of a first cancer, death, or the end of the observation period (September 30, 1996), whichever came first. Dates of death were ascertained from record linkage to Social Security Administration mortality files. The study period (referred to as latency) was defined by subtracting the date of study entry (date of discharge for the first hospitalization) from the date of exit (date of admission for the first hospitalization that listed a discharge diagnosis of cancer, or death or end of study). The ICD codes (ICDA- 8 and ICD 9-CM) for cancers examined were: total cancer (140-207/8), buccal cavity and pharynx (hereafter referred to as buccal cavity) (140-149), esophagus (150), stomach (151), small intestine (152), colon (153), rectum (154), liver and intrahepatic bile ducts (hereafter, referred to as liver) (155), biliary tract (156), pancreas (157), larynx (161), trachea, bronchus and lung (hereafter referred to as lung) (162), melanoma of skin (hereafter referred to as melanoma) (172), prostate (185), bladder (188), kidney (189), brain (191), thyroid (193), non-Hodgkin lymphoma (200,202), Hodgkin lymphoma (201), multiple myeloma (203), leukemia (204-207/8).

Cancer risks among men with diabetes were compared to risks among men without diabetes. Estimates of relative risk (RR) and 95% confidence intervals (95%CI) were calculated using Poisson regression methods for cohort data ¹⁵. All models were adjusted for attained age (<40, 40-49, 50-59, 60-69, 70-79, and >79 yrs), attained calendar year (1969-1974, 1975-1979, 1980-1984, 1985-1989, 1990-1996), latency in years (2-3, 4-5 yrs, 6-10, 11-15, >15 yrs), race (black, white), number of hospital visits (<3, 3-4, 5+). Additional analyses were stratified on race, latency (2-5, 6-10, >10 yrs), calendar year of study entry (1969-1979,1980-1996), age at study entry (<50 yrs, 50+ yrs), and whether diabetes was the only, primary or a secondary diagnosis at the time of the first diabetes visit.

As smoking, alcohol consumption and obesity (ICD-8 code: 277; ICD-9 code: 278.0) are associated with a large number of cancers, the analyses attempted to adjust for these variables. Neither questionnaire data on lifestyle factors (smoking, alcohol consumption) nor measurements of body size were available, however. As a consequence, diagnostic codes were used as surrogate variables. A diagnosis of chronic obstructive pulmonary disease (COPD) was used as a surrogate for smoking (ICD-8 codes: 490, 491, 492, 493; ICD-9 codes: 490, 491, 492, 493, 494, 495, 496). Alcohol-related diagnoses served as surrogate variables for alcohol consumption. Alcohol-related diagnoses included alcoholism, alcoholic gastritis, alcoholic cirrhosis and toxic effect of alcohol (ICD-8 codes: 291, 303, 571.0, 980.0; ICD-9 codes: 291, 303, 535.3, 571.0, 571.1, 571.2, 571.3, 980.0). Obesity diagnoses served as the proxy variable for obesity. A secondary analysis for risk of liver cancer also included diagnoses of chronic liver disease (ICD-8 codes: 570-573, 070, 571.9; ICD-9 codes: 570-573, 070, 571.4). Although all proxy variables almost certainly underestimated the actual exposures, they were included because they were the only measures available and because they were likely to include the most extreme levels of exposures (e.g., heaviest drinker and smokers and obesity at the highest levels).

The dates for all diagnoses except those of cancer were assigned to the date of hospital discharge. The date of cancer diagnosis was assigned to the date of hospital admission because cancer was the outcome of interest and thus, was treated conservatively. As a result, all variables except race and number of hospital visits were time dependent. All p-values were two-sided and statistical significance was set at p<0.05. Calculations of relative risk were performed using the Poisson regression module (AMFIT) in the Epicure software package (Epicure, version 2.0; HiroSoft International Corp, Seattle, Wash). The study was approved by the NIH Office of Human Subjects Research.

Results

Of the 4,501,578 men in the study cohort, 3,669,244 (81.5%) were white and 832,334 (18.5%) were black (Table 1). A hospital diagnosis of diabetes was recorded for 594,815 (13.2%) of the total cohort. Diabetes was more common among black (14.8 %) than among white men (12.9 %). The median follow-up time was 10.5 years for men with diabetes, resulting in a total of 6,264,889.9 person-years of risk. For men without diabetes, the median follow-up time was 11.9 years, resulting in a total of 46,389,292.40 person-years of risk. The median age at first VA hospital discharge diagnosis of diabetes was 59.1 years. The mean age at first VA cancer diagnosis was 67.2 years for men with diabetes and 65.5 years for men without diabetes. Compared with white men, black men had longer median follow-up times and lower median ages at study entry and hospital diagnoses of diabetes and cancer.

Table 1.

Characteristics of white and black male veterans with at least one hospital admission, 1969-1996, who were followed for at least one year. U. S. Veterans Affairs Inpatient Database.

	All Men				White Men				Black Men
	Men with diabetes		Men without diabetes		Men with diabetes		Men without diabetes		Men with diabetes		Men without diabetes
Number of men	594,815	13.2%	3,906,763	86.8%	471,789	12.9%	3,197,455	87.1%	123,026	14.8%	709,308	85.2%
Median age (yrs)
Study entry	57.5		51.5		58.2		52.5		54.4		46.1
Diabetes diagnosis	59.1				60.3				56.6
Cancer diagnosis	67.2		65.5		67.4		65.8		66.3		64.5
Mean years of follow up	10.5		11.9		10.3		11.9		11.3		12.0
Median number of hospital visits	5		2		5		2		5		2
Concomitant medical conditions
Alcohol-related conditions	116707	19.6%	1020135	26.1%	87260	18.5%	797274	24.9%	29447	23.9%	222861	31.4%
COPD1	134778	22.7%	630324	16.1%	115712	24.5%	555591	17.4%	19066	15.5%	74733	10.5%
Obesity	98916	16.6%	159236	4.1%	82651	17.5%	139103	4.4%	16265	13.2%	20133	2.8%
Age (yrs)
<50	157245	26.4%	1816045	46.5%	112,611	23.9%	1,402,711	43.9%	44,634	36.3%	413,334	58.3%
50+	437570	73.6%	2090718	53.5%	359,178	76.1%	1,794,744	56.1%	78,392	63.7%	295,974	41.7%

¹Chronic obsructive pulmonary disease

Risks of cancer among all men, white men and black men are shown in Table 2 and Figure 1. There was very little difference between the results of the multivariate analyses and the results of the analyses that additionally adjusted for diagnoses of alcohol-related conditions, obesity and COPD. Overall, men with diabetes had a significantly lower risk of developing total cancer than did men without diabetes (RR_adj=0.93, 95%CI=0.93-0.94). Cancer risks varied by tumor site, however. Cancers that were significantly less likely to occur among men with diabetes were cancers of the buccal cavity (RR_adj=0.85, 95%CI=0.82-0.89), esophagus (RR_adj=0.77; 95%CI=0.72-0.82), larynx (RR_adj=0.76, 95%CI=0.71-0.80), lung (RR_adj=0.79; 95%CI=0.77-0.80), prostate (RR_adj=0.89; 95%CI=0.87-0.91) and brain (RR_adj=0.91; 95%CI=0.82-0.99). Cancers that were significantly more likely to occur among men with diabetes were cancers of the colon (RR_adj=1.20; 95%CI=1.16-1.25), rectum (RR_adj=1.12, 95%CI=1.07-1.18), liver (RR_adj=1.95, 95%CI=1.95-2.09), biliary tract (RR_adj=1.41, 95%CI=1.22-1.62), pancreas (RR_adj=1.50, 95%CI=1.42-1.59), and kidney (RR_adj=1.09, 95%CI=1.03-1.16). Risks of leukemia (RR_adj=1.14; 95%CI=1.08-1.21) and melanoma (RR_adj=1.13; 95%CI=1.03-1.24) were also significantly elevated among men with diabetes.

Table 2.

Risks of cancer among men with diabetes and men without diabetes in the US Veterans Affairs Inpatient Hospitalization Database, 1969-1996.

	All Men						White Men						Black Men
	Men with diabetes	Men without diabetes	RR1	95% CI1	RR2	95% CI2	Men with diabetes	Men without diabetes	RR1	95% CI1	RR2	95% CI2	Men with diabetes	Men without diabetes	RR1	95% CI1	RR2	95% CI2
All Cancers	51,271	320,539	0.90	0.90-0.91	0.93	0.93-0.94	39,908	259,016	0.93	0.92-0.94	0.95	0.94-0.96	11,363	61,523	0.84	0.82-0.86	0.88	0.86-0.89
Buccal cavity	3,036	26,210	0.75	0.72-0.78	0.85	0.82-0.89	2,457	21,421	0.78	0.75-0.81	0.89	0.85-0.93	579	4,789	0.63	0.58-0.69	0.75	0.69-0.82
Esophagus	1,103	9,240	1.04	1.04-1.04	0.77	0.72-0.82	761	5,576	1.04	1.03-1.04	0.95	0.88-1.03	342	3,664	1.06	1.05-1.06	0.54	0.48-0.60
Stomach	1,063	6,452	0.94	0.88-1.01	0.95	0.89-1.02	740	4,581	1.00	0.92-1.08	1.01	0.93-1.10	323	1,871	0.83	0.74-0.94	0.85	0.75-0.96
Small intestine	144	621	1.24	1.02-1.50	1.21	0.99-1.46	94	482	1.11	0.88-1.40	1.07	0.84-1.36	50	139	1.60	1.14-2.25	1.60	1.13-2.26
Colon	3,976	18,341	1.24	1.20-1.28	1.20	1.16-1.25	3,138	15,081	1.25	1.20-1.30	1.22	1.17-1.27	838	3,260	1.20	1.11-1.30	1.15	1.06-1.25
Rectum	1,990	10,507	1.13	1.08-1.19	1.12	1.07-1.18	1,650	3,138	1.15	1.09-1.22	1.15	1.08-1.21	340	838	1.03	0.91-1.17	1.02	0.90-1.15
Liver	1,299	3,861	1.78	1.67-1.91	1.95	1.82-2.09	1,011	2,938	1.94	1.80-2.08	2.11	1.96-2.28	288	923	1.37	1.19-1.58	1.51	1.31-1.74
Biliary Tract	235	1,013	1.41	1.22-1.62	1.41	1.22-1.62	193	866	1.41	1.20-1.65	1.39	1.18-1.63	42	147	1.39	0.97-1.98	1.35	0.94-1.94
Pancreas	1,656	5,983	1.47	1.39-1.56	1.50	1.42-1.59	1,235	4,673	1.48	1.38-1.58	1.51	1.41-1.62	421	1,310	1.44	1.28-1.62	1.48	1.32-1.67
Larynx	1,413	12,384	0.69	0.65-0.73	0.76	0.71-0.80	1,098	10,015	0.71	0.66-0.75	0.78	0.73-0.83	315	2,369	0.62	0.55-0.70	0.69	0.61-0.78
Lung	11,862	90,565	0.74	0.72-0.75	0.79	0.77-0.80	9,330	30,348	0.75	0.74-0.77	0.80	0.78-0.82	2,532	22,478	0.68	0.65-0.71	0.73	0.70-0.77
Melanoma	599	3,528	1.16	1.07-1.27	1.13	1.03-1.24	581	3,450	1.16	1.06-1.27	1.13	1.03-1.24	18	78	1.25	0.74-2.13	1.15	0.66-1.98
Prostate	9,967	55,496	0.90	0.88-0.92	0.89	0.87-0.91	6,987	42,439	0.89	0.97-0.91	0.88	0.86-0.90	2,980	13,147	0.93	0.89-0.97	0.91	0.88-0.95
Bladder	2,844	16,456	0.97	0.93-1.01	0.96	0.92-1.01	2,523	14,900	0.97	0.93-1.02	0.97	0.93-1.02	321	1,556	0.89	0.79-1.01	0.89	0.79-1.01
Kidney	1,393	6,812	1.15	1.08-1.22	1.09	1.03-1.16	1,131	5,624	1.20	1.12-1.28	1.13	1.06-1.22	262	1,188	0.96	0.84-1.11	0.93	0.81-1.07
Brain	527	3,856	0.91	0.82-1.00	0.91	0.82-0.99	453	3,416	0.91	0.82-1.01	0.91	0.82-1.01	74	440	0.87	0.67-1.12	0.87	0.67-1.12
Thyroid	147	906	1.19	1.00-1.43	1.12	0.93-1.34	118	764	1.19	0.97-1.45	1.13	0.92-1.38	29	142	1.22	0.81-1.85	1.11	0.72-1.69
NHL3	1,277	8,219	0.96	0.91-1.03	0.95	0.90-1.02	1,043	6,956	0.96	0.90-1.03	0.95	0.89-1.02	234	1,263	0.99	0.85-1.15	0.97	0.84-1.13
Hodgkin lymphoma	178	1,402	1.01	0.86-1.20	1.00	0.84-1.18	140	1,178	1.00	0.83-1.20	0.99	0.82-1.20	38	224	1.08	0.73-1.58	1.04	0.71-1.54
Multiple myeloma	734	3,907	0.98	0.91-1.07	0.96	0.88-1.04	458	2,582	1.02	0.92-1.13	1.00	0.90-1.11	276	1,325	0.92	0.81-1.06	0.90	0.79-1.03
Leukemia	1,662	8,173	1.18	1.11-1.24	1.14	1.08-1.21	1,334	7,018	1.14	1.08-1.21	1.11	1.04-1.18	328	1,155	1.35	1.19-1.54	1.29	1.13-1.47

¹adjusted for age, time, latency, race and number of visits

²adjusted for factors above and diagnoses of alcohol-related conditions, obesity and chronic obstructive pulmonary disease

³NHL=non-Hodgkin lymphona

Figure 1.

Relative risks and 95% CI of cancer among black and white veterans with diabetes, 1969-1996, U.S. Veterans Affairs Inpatient Database.

An examination of cancers risks among each racial group separately found that both white (RR_adj=0.95, 95%CI=0.94-96) and black men (RR_adj=0.88, 95%CI=0.86-0.89) with diabetes had significantly decreased risks of total cancer (Table 2). Both groups also had significantly decreased risks of cancers of the buccal cavity, larynx, lung and prostate and significantly increased risks of cancers of the colon, liver, pancreas and of leukemia. In contrast to these similarities, only black men had significantly decreased risks of esophageal cancer (RR_adj=0.54, 95%CI=0.48-0.60) and stomach cancer (RR_adj=0.85, 95%CI=0.75-0.96) and a significantly increased risk of cancer of the small intestine (RR_adj=1.60, 95%CI=1.13-2.26). Only white men had significantly increased risks of cancers of the rectum (RR_adj=1.15, 95%CI=1.08-1.21), biliary tract (RR_adj=1.39, 95%CI=1.18-1.63), kidney (RR_adj=1.13, 95%CI=1.06-1.22) and of melanoma (RR_adj=1.13, 95%CI=1.03-1.24). In general, however, the point estimates of relative risk for black men and white were in the same direction, even when only one of the two was statistically significant.

An examination of cancer risks by interval between first diabetes visit and diagnosis of cancer (2-5 yrs, 6-10 yrs, >10 yrs) found that the decreased risk of total cancer persisted through all latency intervals (Table 3). The decreased risks of buccal cavity, esophageal, laryngeal, and lung cancer also persisted through all three time intervals. Although the risk of prostate cancer was decreased throughout all time intervals, the decreased risk did not become statistically significant until 6 years after the first diabetes visit. The significantly increased risks of cancers of the colon, liver, biliary tract, and pancreas also persisted through all intervals. In contrast, the risks of cancers of the rectum and kidney declined over time and were no longer significantly increased after 10 years, in the case of rectal cancer, and no longer significantly increased after 5 years in the case of kidney cancer. The risks of melanoma and leukemia were inconsistent across latency intervals.

Table 3.

Risks of cancer, stratified by latency between diagnoses of diabetes and cancer, among men with diabetes and men without diabetes in the US Veterans Affairs Inpatient Hospitalization Database, 1969-1996.

	2-5 years		6-10 years		>10 years
	RR1	(95% CI)1	RR1	(95% CI)1	RR1	(95% CI)1
All Cancers	0.98	0.97-1.00	0.94	0.92-0.96	0.90	0.89-0.92
Buccal cavity	0.86	0.80-0.92	0.86	0.80-0.92	0.86	0.81-0.92
Esophagus	0.83	0.74-0.93	0.73	0.64-0.81	0.76	0.69-0.85
Stomach	0.93	0.83-1.05	0.96	0.85-1.09	0.97	0.87-1.09
Small intestine	1.34	0.98-1.84	0.97	0.65-1.44	1.29	0.94-1.77
Colon	1.25	1.17-1.33	1.24	1.16-1.33	1.14	1.07-1.21
Rectum	1.22	1.12-1.33	1.17	1.07-1.29	1.02	0.94-1.11
Liver	2.01	1.78-2.29	2.03	1.79-2.31	1.89	1.71-2.09
Biliary Tract	1.36	1.05-1.77	1.51	1.17-1.96	1.31	1.03-1.67
Pancreas	1.83	1.66-2.02	1.38	1.23-1.54	1.33	1.20-1.47
Larynx	0.78	0.70-0.86	0.78	0.70-0.87	0.73	0.67-0.80
Lung	0.80	0.77-0.83	0.78	0.75-0.81	0.79	0.77-0.82
Melanoma	1.15	0.99-1.34	1.07	0.90-1.27	1.17	1.00-1.37
Prostate	0.97	0.93-1.01	0.91	0.87-0.95	0.86	0.83-0.89
Bladder	1.02	0.95-1.09	0.99	0.92-1.08	0.90	0.84-0.97
Kidney	1.24	1.11-1.38	1.09	0.97-1.22	1.00	0.91-1.10
Brain	0.85	0.72-1.01	1.04	0.87-1.23	0.86	0.73-1.02
Thyroid	1.23	1.17-1.67	1.23	0.89-1.68	0.93	0.67-1.30
Non-Hodgkin lymphoma	1.03	0.93-1.15	0.98	0.87-1.10	0.88	0.76-0.97
Hodgkin lymphoma	1.13	0.87-1.46	1.17	0.87-1.57	0.71	0.51-1.00
Multiple myeloma	0.94	0.81-1.10	0.98	0.84-1.15	0.96	0.84-1.09
Leukemia	1.26	1.14-1.39	1.08	0.97-1.20	1.09	1.00-1.19

¹adjusted for age, time, race and number of visits and diagnoses of alcohol-related conditions, obesity and chronic obstructive pulmonary disease

As liver cancer was the most significantly increased cancer among men with diabetes, an analysis was undertaken which adjusted for diagnoses of liver disease, as well as diagnoses of obesity, alcohol-related conditions and COPD. Although the point estimates were somewhat attenuated, the risks among all men (RR_adj=1.44, 95%CI=1.63-1.86), white men (RR_adj=1.86, 95%CI=1.72-2.00), and black men (RR_adj=1.41, 95%CI=1.22-1.62) all remained statistically significant in the models that included liver disease.

Among the men without a diabetes diagnosis, the most common primary diagnoses at first hospitalization were mental disorders (ICD codes 240-279, 25.0%), diseases of the circulatory system (ICD codes 390-459, 14.8%), diseases of the digestive system (ICD codes 520-579) and injuries and poisoning (ICD codes 800-999, 7%). Among the men with diabetes, the most comment primary diagnoses at first hospitalization were diseases of the circulatory system (ICD codes 390-459, 22.4%), diabetes (ICD code 250), mental disorders (ICD codes 290-319, 13.3%) and diseases of the digestive system (ICD codes 520-570, 8.9%) (data not shown). Among the men with diabetes, diabetes was the only diagnosis for 4%, the primary diagnosis for 22% and a secondary diagnosis for the remaining 74%. As a secondary diagnosis, diabetes was most commonly listed as secondary to heart disease, cerebrovascular disease, alcoholism, schizophrenia, hypertension, COPD, benign prostatic hypertrophy and cataracts. An examination of risk by whether diabetes was the only, primary or a secondary diagnosis found no differences (data not shown). The risks of cancer did not differ when the analysis was stratified by age at study entry (<50yrs, 50+yrs) or calendar year of study entry (data not shown).

Discussion

Analysis of a very large cohort of black and white men yielded evidence that men with diabetes have different risks of developing cancer than do men without diabetes. Overall, the men with diabetes were 7% less likely to develop cancer. By tumor site, men with diabetes had significantly reduced risks of cancers of the buccal cavity, esophagus, larynx, lung, prostate and brain. These reduced risks, however, were offset by significantly increased risks of cancers of the colon, rectum, liver, biliary tract, pancreas and kidney as well as increased risks of melanoma and of leukemia.

Prior cohort studies of total cancer risk among men with diabetes have reported inconsistent findings. Among six cohort studies of cancer incidence, three found a significantly increased risk of total cancer ^{8, 11, 13} while three found no increased risk ^{7, 9, 16}. Among five cohort studies of cancer mortality, one found a significantly increased risk ¹², one found a significantly decreased risk ⁵ and three found no significant difference in risk ^{6, 17-18}. The geographic variability in location of the studies does not explain the differing results as there are inconsistencies even among U.S. studies. Among prior U.S. studies of cancer incidence, one study reported a significantly increased risk ⁸, while the other did not ⁷. Among the U.S. studies of cancer mortality, one reported a significantly decreased risk ⁵, while two found no difference in risk ^{6, 18}. The single greatest difference between prior studies and the current study is population size. Prior studies ranged between 1134 men ⁷ and 144,427 men ¹². The current study includes more than four and a half million men. Given the rather modest decrease in risk in the current study, it is likely that smaller studies could miss such an effect. Why some studies would find an increased risk is unclear, but seems unlikely to be related to the source of the patient population as two of the studies that found an increased risk ^12-13 were based on hospital diagnoses of diabetes, as is the current study. Differences in patient populations among studies, however, can not be ruled out.

The risk of developing a specific type of cancer was most significantly elevated for liver cancer. This finding is consistent with at least 10 case-control and 5 cohort studies that previously found statistically significant positive associations between diabetes and liver cancer with risks ranging up to 240% ¹⁹. The relative risks in the current study were very stable over all latency intervals (2-5, 6-10, >10 years), making it unlikely that diabetes was caused by incipient liver cancer. Diabetes, however, is known to be associated with chronic liver disease, which precedes liver cancer in the vast majority of cases. To account for pre-existing liver disease, a secondary analysis in the current study adjusted for liver disease and saw very little change in result. If diabetes itself increases risk of liver cancer, it has been hypothesized to do so by increasing insulin resistance and altering levels of insulin-like growth factors ²⁰. For example, it has also been shown that hyperinsulinemia causes a decrease in IGF-1 and IGFBP-3 levels ²¹. Such reduced levels have been shown to increase the proliferation of cancer cells ²⁰. In addition, several studies have reported that insulin replacement therapy is itself related to risk ^22-24.

A link between diabetes and pancreatic cancer has long been suggested ²⁵ as it is estimated that up to 80% of persons with pancreatic cancer also have diabetes or impaired glucose tolerance ²⁶. It has been difficult to determine, however, whether diabetes is a risk factor for pancreatic cancer or a consequence of pancreatic cancer ²⁷. With an even poorer prognosis than liver cancer, however, it is unlikely that diabetes diagnosed many years prior to pancreatic cancer is a consequence of the carcinogenic process ⁹. An early meta-analysis ²⁸ of diabetes and pancreatic cancer studies reported a summary risk of 2.1 (95%CI=1.6-2.8), which was similar both to the results of a more recent meta-analysis ²⁹ (RR=1.8, 95%CI=1.7-1.9) and to the results of the current study (RR=1.5, 95%CI=1.4-1.6). Cohort studies prior to the current one, also found increased risks of pancreatic cancer and found that the risk persisted more than 10 years after the diagnosis of diabetes ³⁰. As with liver cancer, the most frequently suggested hypothesis explaining the link between diabetes and pancreatic cancer is the insulin hypothesis as in vitro experiments have shown that insulin can stimulate pancreatic carcinogenesis ³¹. Insulin has also been shown to activate insulin-like-growth-factor I receptors which increase pancreatic cell proliferation ³². Bolstering the hyperinsulinemia hypothesis is the finding of no increased risk of pancreatic cancer risk among persons with type I diabetes ³³. As type I diabetes is not characterized by hyperinsulinemia, there should be no relationship between type I diabetes and pancreatic cancer, if the insulin hypothesis is accurate.

An elevated risk of biliary tract cancer has been suggested by several prior studies ^{7, 11, 13, 34-35} with risk estimates very similar to the estimate found in the current study (RR_adj=1.41). As suggested by Adami et al. ⁹, possible explanations for an increased risk include the propensity of diabetic men to develop gallstones, and the association of obesity with both diabetes and biliary tract tumors. Increased risks of kidney cancer have also been reported by some ^{11, 13, 36}, but not all ^{9, 16, 35, 37-39} studies. As with liver cancer, a suggested mechanism for a diabetes-kidney cancer association is the increased levels of insulin-like growth factor 1 seen in diabetic men ³⁶. It should be noted, however, that the increased risk of kidney cancer was only seen among white men and was only significant during the first 2-5 years after the diagnosis of diabetes.

Elevated risks of both colon and rectal cancers have been previously reported in a number of studies ^{9, 16, 18, 35, 40-42}, although several have reported that the risk is confined to women ^43-45. A meta-analysis of 15 studies of colorectal cancer ⁴⁶, however, found statistically significant relative risks of both colon cancer (RR=1.43, 95%CI=1.28-1.60) and rectal cancer (RR=1.33, 95%CI=1.14-1.54), which are very similar to the relative risks of colon cancer (RR=1.20) and rectal cancer (RR=1.12) found in the current study. As with liver and pancreatic cancers, hyperinsulinemia, insulin-like growth factor-I and insulin-like growth factor binding protein-3 have all been associated with colorectal carcinogenesis ^47-50.

Increased risks of melanoma ^{7, 9, 13, 35, 38, 51} and leukemia ^{5, 7, 11, 13, 17, 35, 38} have not been generally reported in association with diabetes. In the current study, increased risk of melanoma was quite modest (RR=1.13, 95%CI=1.03-1.24), was confined to white men, was confined to the study entry period 1980-1996 and was not statistically significant during each latency period; suggesting that the association between diabetes and melanoma was tenuous at best. The risk of leukemia was seen in both white and black men and was evident in both the earlier and later study entry periods, but was most significantly elevated during the first latency period, suggesting that undiagnosed leukemia may have resulted in diabetes rather than diabetes predisposing to leukemia.

In the current study, diabetes was associated with reduced risks of cancers of the buccal cavity, esophagus, larynx, lung, brain and prostate. With the exception of prostate and brain cancers, all other sites are strongly linked to cigarette smoking and alcohol consumption. The reduced risk of lung cancer has previously been found in some studies ^{5, 16-17, 38, 51-52}, but not in others ^{8-9, 11, 18, 35}. One hypothesis suggested to explain the inverse association is that diabetics might be less likely to smoke than non-diabetics, or might be more likely to quit when given a diabetes diagnosis ⁵. Most research, however, does not support this postulate as there are now consistent epidemiologic results showing that smoking increases risk of diabetes in a dose-response manner ^53-54. An alternative explanation may be related to the higher body mass index (BMI) of diabetic men, as low BMI has been reported to be risk factor for lung cancer ⁵⁵, buccal cavity cancer ⁵⁶, esophageal squamous cell carcinoma ⁵⁷ and laryngeal cancer ⁵⁸. Whether diabetic men who smoke have a lower risk of smoking-related cancers because they tend to be heavier than non-diabetic men who smoke remains to be studied.

An inverse relationship between diabetes and development of prostate cancer has been reported by a number of prior studies ^{11, 13, 59-63} and examined in two meta-analyses ^64-65. Based on a meta-analysis of 19 studies, Kasper et al. reported a relative risk of 0.84 (95%CI=0.76-0.93), which is very similar to the relative risk found in the current study (RR=0.89, 95%CI=0.87-0.91). The mechanism for this apparent protective effect may be that diabetics tend to have altered hormone profiles ⁵⁹. In support of this hypothesis are data examining plasma levels of C-peptide, testosterone, sex hormone binding globulin, insulin-like growth factor-1 and insulin-like growth factor binding protein-3 over time in diabetic and non-diabetic men ⁶⁶. IGF-1 levels were found to be lower in diabetic men and bioavailable testosterone levels decreased over time after the diagnosis of diabetes ⁶⁶. These results support the findings of the current study as the decreased risk of prostate cancer, which was of borderline significance in the earliest interval (RR=0.97, 95%CI=0.93-1.01) became more significant in the subsequent time intervals (RR=0.91, 95%CI=0.87-0.95; RR=0.86, 95%CI=0.83-0.89). An alternative explanation is a lower prostate cancer detection rate among diabetic men than among non-diabetic men. Diabetes is associated with both prostatic enlargement ⁶⁷ and lower serum levels of prostate-specific antigen ⁶⁸, which may result in screening bias by rendering the detection of prostate cancer more difficult in diabetic men.

The current study had several notable strengths. The study included a very large population (>4.5 million men) with relatively stable and standardized access to medical care and lengthy follow-up (up to 27 years). In addition, the study was able to examine risks separately among black and white men and was based on medical records rather than self-reported questionnaire data. The study was limited, however, by lack of information on relevant covariates, such as obesity, smoking habits and alcohol consumption. Although medical conditions were used as surrogate variables for these covariates, they do not correlate perfectly and they almost certainly underestimate the prevalence of the exposures. It is likely, however, that the surrogate variables were conservative estimates of the true variables, and so statistical adjustment for the variables is unlikely to have overstated associations in the current study. Another limitation of the current study is that it was based on inpatient data, thereby including, in all likelihood, the most severe forms of the medical conditions. The study was also unable to differentiate between type I and type II diabetes. However, the age of the population (mean age 59 years at first hospitalization) and the proportion of all diabetics that have type II disease (90-95%) suggests that the great majority of the men were type II diabetics. Finally, Veterans Affairs’ patients, particularly during the time period of the current study, were likely to be of lower socio-economic status and to be in poorer health than were men in the general population. For example, a study of veterans who used V.A. health system found that the veterans were 14.7 times more likely to have poor health status than were men in the general population ⁶⁹. Veterans might also be different from men in the general population in regard to exposures and lifestyles. As a result, caution should be exercised when extrapolating the results of the current study to men in the general population.

In summary, the current study suggests that differences in cancer risks exist among diabetic and non-diabetic men. In particular, diabetic men may be at significantly increased risk of cancers of the liver, pancreas, biliary tract and colorectum. The overall decreased risk of cancer may be related to higher body-mass index among diabetic versus non-diabetic men who smoke, and to decreased risk of total prostate cancer. More study is warranted to examine the possible effects of diabetic treatment and to control more extensively for lifestyle habits and other potentially confounding variables.