Glycemic Status and Risk of Prostate Cancer

Among men in most western countries, prostate cancer is a leading cause of morbidity and mortality (1, 2), yet its etiology remains largely unknown. At present, increasing age, Black race, and family history of prostate cancer are the only established risk factors (1). Biological evidence supports a role for androgens in the etiology of prostate cancer (3-5). Furthermore, risk of this malignancy has been associated with high circulating levels of insulin as well as insulin resistance (6, 7). Insulin is a mitogen and might affect prostate cancer risk by either the androgen pathway or insulin-like growth factor system, exerting growth factor and antiapoptotic effects on prostatic epithelial cells (8, 9). Type 2 diabetes mellitus is a complex metabolic disease, initially characterized by hyperinsulinemia and insulin resistance. With increasing duration and severity of disease, insulin levels decrease and hyperglycemia tends to worsen (10), which can adversely affect Leydig cell function and, ultimately, testosterone secretion (11). Lower circulating androgen levels have been reported in men with type 2 diabetes mellitus (11-13). This could be the result of hyperglycemia-mediated testicular cell damage (in later-stage type 2 diabetes mellitus; ref. 11). There is also evidence that low testosterone levels may be predictive of metabolic syndrome and type 2 diabetes mellitus (14-16).

Several epidemiologic studies have investigated the association between diabetes mellitus and risk of prostate cancer; however, findings have been inconsistent (17-36). Some have investigated the association between prostate cancer risk and glycemic status, examined as a continuum ranging from normoglycemia to diabetes mellitus (17-19, 26, 37, 38). A recent meta-analysis of 19 published studies found an overall inverse association between diabetes mellitus and prostate cancer risk [relative risk (RR), 0.84; 95% confidence interval (95% CI), 0.76-0.93; ref. 39]. The authors did several subgroup analyses but lacked sufficient data to examine the relationship between diabetes mellitus and prostate cancer by stage of disease. Moreover, they noted that future studies investigating this association should be conducted among more racially diverse cohorts. The purpose of this study was to examine the risk of prostate cancer, both advanced and localized, across several degrees of glucose tolerance in a large, racially diverse cohort of men followed for up to 39 years.

The source population for this study was the men who were members of Kaiser Permanente Medical Care Program and had completed at least one Multiphasic Health Checkup (MHC) between July 1964 and August 1973. The MHC, initiated in 1964, was a voluntary, comprehensive health evaluation used by members and their physicians as part of regular checkups. It included automated multiphasic health testing, consisting of a battery of tests and procedures (including a 75-g oral glucose challenge), a follow-up appointment with a physician, and a group of specialty examinations (40). In addition, participants completed a questionnaire that requested information on a variety of health-related factors. Generally, members who received the MHC were inclined to be healthier, better educated, and less prone to unhealthy lifestyle habits, such as smoking and heavy drinking (41). A 75-g oral glucose challenge was done in individuals without history of diabetes mellitus between 8:00 a.m. and 9:00 p.m. regardless of fasting status. Serum glucose levels were measured 1 h after the glucose challenge by a Technicon AutoAnalyzer (1964-1972), AutoChemist (1968-1972), or SMA-12 AutoAnalyzer (1972).

The study population was restricted to 52,641 men ages 35 to 80 years with complete data on glucose tolerance status, including history of diabetes mellitus, and no history of prostate cancer at MHC examination. We then excluded 1,887 with either self-reported or unknown history of cancer or tumor, 1 with in situ prostate tumor, 15 with missing race, and 3,529 with missing or invalid height or weight. Thus, the final analytic cohort consisted of 47,209 men, of whom 2,833 developed prostate tumors. Median age at diagnosis was 72 years (range, 45-97). Subjects were followed for development of prostate cancer using the Kaiser Permanente tumor registry, which is part of the Surveillance, Epidemiology and End Results program, and through computer-stored hospitalization records. Any records with a diagnosis date before 1973 (that is, institution of Surveillance, Epidemiology and End Results program) were verified as cancer cases after review of full abstract by registry personnel. We used the Surveillance, Epidemiology and End Results summary stage of disease to assign subcategories of prostate cancers (stage 1, localized; stages 2-5, regional; and stage 7, distant).

MHC participants who did not report a history of diabetes mellitus at MHC examination were assigned to one of four categories of glucose tolerance based on the serum glucose measurement obtained 1 h after a 75-g oral glucose challenge, as described previously in the Chicago Heart Association Detection Project in Industry (42): (a) <140 mg/dL, (b) 140 to 159 mg/dL, (c) 160 to 199 mg/dL, and (d) ≥200 mg/dL (asymptomatic hyperglycemia). Men with a history of diabetes mellitus, based on self-report (at MHC examination) of physician diagnosis or diabetes mellitus–related medication usage during the past year or two, were placed in a fifth category (“diabetes”). Follow-up began at “index” MHC examination and ended at diagnosis of prostate cancer (n = 2,833), termination of membership in the health plan for any reason including death (n = 36,013), or December 31, 2003 (n = 8,363), whichever occurred first. We determined that membership termination was attributable to death for 14,528 of the 36,013 men in the cohort who left the health plan before a prostate cancer diagnosis or end of study. The remaining 21,485 (45% of the entire cohort) were considered lost to follow-up. Cox proportional hazards modeling with participant's age as the time structure was used to estimate RR and 95% CI while controlling for confounding variables. The risk sets were further stratified by calendar year of birth, in narrow intervals of 2 to 4 years, and the regression model also adjusted for year of examination as a covariate.

Thus, at the time of each outcome, we formed a risk set in which the person diagnosed with cancer was compared with all others in the cohort who were born during the same range of years and who were still in follow-up at the same age. Other covariates examined as possible confounders or effect modifiers included race/ethnicity, body mass index (BMI; entered as categories based on the WHO classification; ref. 43), educational attainment categories (less than high school, high school graduate, some college, college graduate and higher), cigarette smoking status (never, former, current), high aspirin use (that is, >6 per day versus not), history of prostate surgery (yes, no), prior treatment to genitals (yes, no), age at shaving initiation (dichotomized as <15 versus ≥15 years), and history of venereal disease (yes, no).

A factor was evaluated for confounding by examining whether or not it altered the risk ratio for prostate cancer associated with glucose tolerance level by >10%. Only race/ethnicity changed the risk ratios associated with the principal exposure by >5%. BMI, modeled in various ways (that is, as continuous, WHO classification categories, quintiles, and quartiles), did not appear to be associated with risk of prostate cancer. Nevertheless, it was included in the final model because of the established association of obesity with insulin resistance and diabetes mellitus (9, 44, 45).

In addition to models using the entire analytic cohort and all prostate tumor cases, several subgroup analyses were conducted, including separate analyses by stage of disease (localized versus regional/distant cancer) and after omitting the first 5 years of follow-up (“lagged”) because prostate tumors diagnosed within that period may have been present at the baseline examination. We also examined whether the association between glycemic status and prostate cancer risk varied by race (White versus African American), age at MHC examination (<55 versus ≥55 years), and BMI (<25 versus ≥25 kg/m²). Additionally, we examined the risk of prostate cancer stratified by age in two separate analyses. One ended follow-up before age 65 and the second began follow-up at age ≥65. To examine the potential effect of prostate cancer screening, one analysis ended follow-up at December 31, 1991 or before widespread use of prostate-specific antigen (PSA) testing. Finally, we conducted analyses for two different time intervals. For the first, follow-up began at MHC and continued for up to 10 years. In the second analysis, follow-up began 10 years after MHC and continued until the end of study. We did this to examine whether the observed associations varied by time since exposure status (that is, diabetes mellitus, post-load glucose levels). All analyses were conducted using SAS version 9.13 (SAS Institute).

The median follow-up time for the cohort was 18.4 years (range, <0.5-39.4 years; interquartile range, 22.8 years). Approximately 16% had <5 years of follow-up. Median follow-up time for the 21,485 men in the cohort who were lost to follow-up was 9 years (interquartile range, 10.5 years). Compared with the subset of the cohort not lost to follow-up, these individuals were slightly more likely to be younger at MHC examination (median age, 46 versus 48 years) and White (85% versus 77%) but had similar glycemic status profiles (that is, there was little difference between the two groups for all categories, including asymptomatic hyperglycemia and diabetes mellitus). Among the 2,833 men diagnosed with prostate cancer during follow-up, 1,883 (67%) were diagnosed with localized cancer, 375 with regional, 335 with distant, and 240 had unknown disease stage (Table 1). The proportion of African Americans was higher among men diagnosed with prostate cancer compared with the entire cohort. In comparison with the overall study population, cases less frequently reported a history of diabetes mellitus or had a 1-h post-challenge serum glucose measurement ≥200 mg/dL. In general, cases also participated in more MHC examinations during the 1964 to 1973 phase, were less obese, better educated, and less commonly cigarette smokers. Despite small differences in age group distribution, median age at MHC examination was similar for cases and the full cohort (48 and 47 years, respectively).

Hazard ratios for the association between glucose tolerance and risk of prostate cancer are shown in Table 2. After adjusting for race and BMI, a statistically significant inverse association between increasing levels of reduced glucose tolerance and prostate cancer risk was observed. This trend was somewhat more consistent for localized than for regional/distant disease.

When entry into the study was lagged 5 years, risk ratios (and 95% CIs) changed little relative to when the entire cohort was included in the model (data not shown). When follow-up ended at December 31, 1991, we observed a modest attenuation of the inverse association between diabetes mellitus and localized prostate cancer risk that was no longer statistically significant (data not shown). Risk of regional/distant disease among men with diabetes mellitus neither changed materially nor was statistically significant. When follow-up began at MHC examination and continued for up to 10 years, prostate cancer risk was slightly increased for all levels of glucose tolerance though more substantively for serum glucose ≥200 mg/dL and diabetes mellitus (P_trend = 0.05; Table 2). In contrast, when follow-up began 10 years after MHC and continued until end of study, a statistically significant inverse association between both asymptomatic hyperglycemia and diabetes mellitus and prostate cancer risk was observed (P_trend = 0.005).

When the cohort was stratified by age during follow-up (<65 versus ≥65 years), an inverse association between diabetes mellitus and prostate cancer risk was observed (∼20% diminished risk), although the 95% CI included unity (RR, 0.78; 95% CI, 0.42-1.47). In contrast, when entry into the study was delayed until age ≥65 years, we observed a statistically significant inverse association between increasing levels of reduced glucose tolerance and prostate cancer risk (P_trend = 0.03 for all and localized cancers; Table 3). The results from this subgroup analysis were virtually the same as when we included the entire cohort. In race-stratified models (Table 3), there was a stronger inverse association between diabetes mellitus and risk of prostate cancer in African American men (RR, 0.59; 95% CI, 0.37-0.95) compared with Whites (RR, 0.82; 95% CI, 0.63-1.08). When the cohort was stratified by BMI (Table 3), <25 kg/m² (underweight and normal weight) versus ≥25 kg/m²(overweight and obese), there was a stronger inverse association between diabetes mellitus and localized versus regional/distant prostate cancer risk in overweight and obese men (RR for localized disease, 0.58; 95% CI, 0.37-0.89; P_trend = 0.03 versus RR for regional/distant disease, 0.85; 95% CI, 0.49-1.48). This pattern was reversed in normal weight men (RR for localized disease, 0.93; 95% CI, 0.63-1.38 versus RR for regional/distant disease, 0.51; 95% CI, 0.20-1.29). Among men ages <55 years at examination, we observed an inverse association between diabetes mellitus and risk of prostate cancer for both localized and advanced disease (data not shown). In contrast, the inverse association for localized disease was weaker and not observed at all for advanced disease among men ages ≥55 years at start of follow-up (data not shown).

To our knowledge, this is one of the first studies to examine the association between risk of prostate cancer and both diabetes mellitus and post-load serum glucose levels in nondiabetics. Our findings suggest that prostate cancer risk associated with decreasing glucose tolerance varied significantly over time. The initially positive and later inverse association is consistent with other reports on diabetes mellitus and future prostate cancer risk (22, 25).

The current study has several limitations that should be considered when interpreting results. First, classification of glucose tolerance and diagnosed diabetes mellitus was based on a single oral glucose challenge test or self-reported history of diabetes mellitus, and this likely changed over time for some men. It did, however, allow us to examine whether a single assessment could predict future risk ≥10 years later. Second, self-reported diabetes mellitus did not include type. It has been estimated that >95% of diabetics over age 40 are type 2 (46). Although age at entry into the study was restricted to ≥35 years, the cohort likely included a small percentage of men with type 1 diabetes mellitus who do not have hyperinsulinemia and insulin resistance. The glucose challenge procedure was administered at different times throughout the day and therefore not necessarily following an overnight fast. Nevertheless, 90% of the men for whom these data were captured had fasted for ≥4 h before ingestion of the 75-g load (17% for ≥8 h).

A third limitation is that a sizeable proportion of the cohort (45%) was lost to follow-up. In spite of this, characteristics for these individuals, with respect to age, race, and glucose tolerance status, were similar to those not lost. Moreover, median follow-up time was still 9 years. Therefore, while possible, we think it unlikely that our results are biased because those lost to follow-up were different than those not lost with respect to both glycemic status and risk of prostate cancer. Fourth, we had no data on either duration or onset of diabetes mellitus and asymptomatic hyperglycemia. As a proxy for examining the association between glycemic status and prostate cancer risk by duration of principal exposure, we conducted analyses during two separate periods, within 10 years of MHC and lagged 10 years from index examination.

Findings from epidemiologic studies investigating the association between diabetes mellitus and risk of prostate cancer have been inconsistent. An insignificant overall elevated risk was observed in two cohort studies (20, 23) and two hospital-based case-control studies (27, 28). Our results are largely consistent with seven other cohort studies (17, 21, 22, 24-26, 36), six case-control studies (29-33, 35), and a recent meta-analysis (39) that found an inverse association between diabetes mellitus and/or glycemic status and prostate cancer risk, including several studies that were conducted before widespread PSA testing (21, 24, 29).

Several studies also have examined prostate cancer risk and time since diabetes mellitus diagnosis (22, 23, 25). Some (22, 25) but not all (23) have found an early increased risk followed by a later inverse association. Unlike other (22, 29, 31) but not all (30) studies that reported an association between diabetes mellitus and decreased risk of advanced prostate cancer, we saw a stronger inverse association between diabetes mellitus and/or asymptomatic hyperglycemia and localized prostate cancer risk compared with all and advanced-stage tumors. Some confounding of our results by PSA screening seems plausible given that we initially observed a stronger inverse association between diabetes mellitus and risk of localized, compared with regional/distant, prostate cancer and that this association was slightly attenuated when we truncated follow-up at December 1991 (that is, before the pre-PSA screening era). Men with diabetes mellitus may be less likely to undergo PSA testing because their physicians are more focused on managing the diabetes mellitus and attendant complications. Some studies have reported diabetic men to be screened less often (23, 47), whereas others have noted no difference or increased screening in some age groups (22). We examined two interval cohorts within our setting (48, 49) and found little evidence of differential PSA screening by diabetes mellitus.

To our knowledge, only four studies have examined the potential association between glucose and prostate cancer risk based on post-load glucose measurements (18, 19, 37, 38). In contrast to our findings, none observed an inverse association with glucose levels. However, differences in methods may explain some of this inconsistency. In follow-up studies with the Whitehall cohort, diabetes mellitus was included among the exposures and defined by either self-report or a 2-h post-challenge (50-g) plasma glucose measurement of ≥200 mg/dL (18, 19). Studies using participants in the Chicago Heart Association Detection Project excluded men who self-reported a history of diabetes mellitus and measured glucose 1 h after a 50-g oral load (37, 38). In addition, the outcome was prostate cancer mortality.

More recently, Jee et al. evaluated the association of fasting (versus post-load) serum glucose with cancer risk in a large cohort of Korean men ages 35 to 90 years at the beginning of follow-up (26). They also examined risk associated with a diagnosis of diabetes mellitus. Similar to our findings, an inverse association with prostate cancer risk was seen only among men with impaired glucose tolerance (fasting serum glucose concentrations ≥126 mg/dL) or with diabetes mellitus.

A biologically plausible mechanism to explain our findings is based on the complex temporal interrelationships between metabolic alterations concomitant with development of type 2 diabetes mellitus and hormonal factors associated with prostate cancer risk. Saad et al. have proposed a two-step model for development for development of type 2 diabetes mellitus (10). The first step, transition from normal to impaired glucose tolerance, is driven by insulin resistance and reflected by hyperinsulinemia. Over time, the transition from impaired glucose tolerance to type 2 diabetes mellitus occurs (step 2) and progression of diabetes mellitus is accompanied by diminished insulin secretion, secondary to pancreatic β-cell dysfunction (either from exhaustion or glucose toxicity).

Insulin may increase free testosterone, which itself may increase prostate cancer risk (8). However, insulin-mediated effects on androgen levels likely are offset by the stronger influence of hyperinsulinemia on increasing bioavailability of insulin-like growth factor-I, which has been shown in vitro to have mitogenic and antiapoptotic effects on prostatic epithelial cells (9, 50-52). Several epidemiologic studies have reported a positive association between circulating insulin-like growth factor-I concentration and prostate cancer risk (53-57). Furthermore, in a recent meta-analysis of studies examining the association of circulating insulin-like growth factor-I concentrations with prostate cancer risk, Shi et al. found that insulin-like growth factor-I blood levels were higher in prostate cancer patients than in controls (58). Thus, during the early stage of abnormal glucose metabolism and development of type 2 diabetes mellitus (the insulin-resistant step), prostate cancer risk may be increased due to increasing levels of insulin.

The progression of type 2 diabetes mellitus, during which β-cell exhaustion plays a critical role, is accompanied by diminished insulin response to (increasing) glycemic stimuli. This reduction in insulin secretion may contribute to decreased prostate cancer risk in part due also to a possible decrease in testosterone levels. Lower total and free testosterone levels have been reported in men with diabetes mellitus (12-16, 59, 60). Androgens, particularly testosterone, are essential for normal growth and maintenance of the prostate and may contribute (either directly or indirectly) to the pathogenesis of prostate cancer (8). Results of epidemiologic studies have been mixed, with some finding a positive association between high circulating levels of testosterone and prostate cancer risk (4, 5) and others unable to show any association (61) or even finding a modest decreased risk associated with total testosterone (62). Nonetheless, it is plausible that with progression of type 2 diabetes mellitus, the concomitant diminution in both insulin response and bioavailable testosterone confers protection from prostate cancer (22, 38).

In summary, our findings are consistent with a growing body of evidence indicating an inverse association between diabetes mellitus and/or impaired glucose intolerance and risk of prostate cancer. While biologically plausible, more research is needed to clarify the possible changing pattern of risk over time and whether risk is different by stage of disease. In addition, further examination of potential confounding by PSA testing is warranted.