Hormonal patterns in men with prediabetes and diabetes in NHANES III: possible links with prostate cancer

Abstract

Purpose:

Pathways involving sex hormones and insulin-like growth factors (IGFs) have been proposed to explain, in part, the lower risk of prostate cancer among men with diabetes. To gain insights into potential biological mechanisms, we explored differences in serum concentrations of sex hormones and IGFs across the trajectory from normoglycemia to prediabetes to poorly controlled diabetes.

Methods:

Using cross-sectional data from the National Health and Nutrition Examination Survey III, we examined differences in levels of circulating sex hormones, sex hormone binding globulin (SHBG), IGF-1 and IFG-binding protein-3 (IGFBP-3), according to diabetes status: no diabetes [n=648], pre-diabetes [n=578], undiagnosed diabetes [n=106], well-controlled diabetes [n=42], and poorly-controlled diabetes [n=56]. Adjusted geometric mean concentrations were derived using multivariable linear regression, adjusted for age, race and other lifestyle factors.

Results:

Total testosterone concentrations were lower among pre-diabetics (4.89 ng/mL, 95% confidence interval (CI) 4.95–5.21) than men without diabetes (5.29ng/mL, 95%CI 5.06–5.53) but did not reduce further across diabetes groups. Concentrations of estradiol, estimated free testosterone, SHGB, IGF-1 and IGFBP-3 did not differ. While the ratio of IGF-1 to IGF-BP3 was lower among men with prediabetics and undiagnosed diabetes than men without prediabetes/diabetes, there was no trend across groups. A positive trend for the ratio of estradiol to testosterone levels was observed across groups (p-trend=0.045).

Conclusion:

Our findings do not provide clear support for either an androgen-driven or IGF-driven pathway for the inverse association between diabetes and prostate cancer risk.

Introduction

Numerous studies suggest that men with diabetes have reduced prostate cancer incidence [1–5], in contrast to other types of cancer (e.g. breast, colon, kidney, bladder, pancreas) where incidence appears to be an increased among adults with diabetes [6]. Whether the inverse association with prostate cancer applies for both low- and high-grade disease is less clear, with some studies suggesting an association only with less aggressive prostate cancer [7–9].

Potential mechanisms linking diabetes and prostate cancer include: direct or indirect biological effects of diabetes itself that lower risk; protective effects of glucose-lowering medications; lower detection rates resulting from lower PSA levels in men with diabetes, or differences in health care access or quality which influence cancer screening practices [10–12]. Although several studies document lower rates of PSA testing and biopsies among men with diabetes [13, 14], such differences are likely to only partly explain difference in prostate cancer risk [5]. Likewise, evidence for a protective effect from antidiabetic medications such as metformin is inconsistent, with a recent metanalysis indicating no association with prostate cancer risk [15]. Support for a direct causal link comes from observations of a stronger inverse association with increasing duration of diabetes [3, 16]. The two most widely proposed biological pathways to explain the lower prostate cancer risk are: 1) effects of diabetes on sex steroid levels (particularly androgens), reducing the potential for the growth of androgen dependent tumours [10, 11], and 2) that dysregulation of insulin production as diabetes progresses eventually leads to hypo-insulinaemia and reduced levels of insulin-like growth factors (IGF), which limit prostatic epithelial cell growth [17]. Convincing evidence for either mechanism is lacking.

To gain insights into potential biological mechanisms for reduced prostate cancer risk among men with diabetes, we explored differences in serum concentrations of sex hormones and IGF-1, alongside serum markers of glycemia, across the trajectory from normoglycemia to prediabetes to poorly controlled diabetes. We examined serum concentrations of fasting glucose, glycated haemoglobin (HbA1c), insulin C-peptide, sex hormones (testosterone, estradiol, androstanediol glucuronide (AAG)), sex hormone binding globulin (SHBG), IGF-1 and IGF binding protein 3 (IGFBP-3), among a representative sample of US men from NHANES III, according to categories of no diabetes, prediabetes and diabetes.

Materials and Methods

Study population

The Third National Health and Nutrition Examination Survey (NHANES III), conducted by the National Center for Health Statistics (USA) was designed as a cross-sectional survey to assess the health and nutritional status of the US population. Multi-stage, stratified, clustered, probability sampling of the non-institutionalised civilian population was used to select the survey population. Hispanic and African Americans were oversampled to improve precision of analyses within sub-populations. NHANES III was conducted in two waves, the first in 1988–1991 and the second in 1991–1994. Participants were chosen at random to participate in either a morning or afternoon examination. Those attending the morning examination had fasted overnight. Participants were interviewed within their own homes and underwent physical assessment and had blood samples taken at mobile examination centers. Detailed information on NHANES III methodology has been published previously [18].

In the subsequent Hormone Demonstration Program, sex steroid hormone concentrations were assayed in excess serum samples from participants of the morning sample NHANES III phase 1 [19]. Of the 2205 males aged 12 years or older who took part in the morning session of Phase 1, 1637 had serum samples remaining in the NHANES III repository available for sex steroid assays. Our study population comprised 1430 males aged 20 years or older without any prior diagnosis of cancer, after excluding those with missing data for serum sex steroid measures (n=179), those aged under 20 years at assessment (n=69) and men who reported a previous diagnosis of cancer (n=28).

Serum measures

As part of NHANES III, morning serum samples drawn following overnight fasting were centrifuged, aliquoted and stored at −70° C until assayed. Plasma glucose concentrations were assessed by a hexokinase enzymatic reference method (COBAS MIRA; Roche Diagnostics Corporation Laboratories, Montclair New Jersey), serum insulin levels by radioimmunoassay (Pharmacia Diagnostics, Uppsala, Sweden), and C-peptide levels by radioimmunoassay (Bio-Rad Laboratories). HbA1c percentage was measured using ion-exchange high performance liquid chromatography with a glycolated haemoglobin analyser (DIAMAR, Bio-Rad Laboratories, Hercules, California).

For the subsequent Hormone Demonstration Project, sex steroid hormone (serum testosterone, estradiol, SHBG) concentrations were measured using competitive electrochemiluminescence immunoassays on the 2010 Elecsys autoanalyser (Roche Diagnostics, Indianapolis, Indiana) in surplus stored serum samples. AAG concentrations were measured by an enzyme immunoassay (Diagnostic Systems Laboratory (DSL), Webster, Texas). IGF-1 and IGFBP-3 concentrations were measured by an enzyme-linked immunosorbent assay (DSL 10–5600). Free testosterone was estimated from concentrations of total testosterone, SHBG and albumin according to Vermeulen et al. 1999 [20], and free estradiol from concentrations of total estradiol, SHGB and albumin, according to Rinaldi et al 2002 [21]. The estradiol to testosterone ratio was calculated as: total serum Estradiol (pg/mL)/total serum Testosterone (ng/mL) ×1000, and the IGF-1 to IGFBP-3 as: serum IGF-1 (ng/mL)/serum IGFBP-3 (ng/mL). While duplicate and repeated assays of biomarkers were not undertaken, strict quality controls were followed to ensure validity and consistency for all serum measures [22–24].

Exposures and Covariates

The main exposure of interest was diabetes status, categorised according to worsening disease. Men were defined as having diabetes based on a positive response to the question “Have you ever been told by a doctor you have diabetes or sugar diabetes?”. Men reporting a diagnosis of diabetes were further classified into ‘well controlled’ (HbA1c<7%) or ‘poorly controlled’ (HbA1c>= 7%) in line with clinically significant cut-offs [25]. Men were defined as having ‘pre-diabetes’ based on the American Diabetes Association criteria [26]: (i) fasting glucose plasma concentration from 100mg/dL (5.6 mmol/L) to 125 mg/dL (6.9mmol/L); (ii) glucose tolerance test with 2-h plasma glucose concentration in a 75 g oral glucose test from 140mg/dL (7.8mmol/L) to 199mg/dL (11.0mmol/L); or (iii) HbA1c from 5.7 to 6.4%. Men with levels above the maximum cut points for impaired fasting glucose, impaired glucose tolerance or HbA1c but did not report being previously diagnosed with diabetes were classified as having ‘undiagnosed diabetes’. According to the above definitions we grouped the cohort into 5 categories: a) neither pre-diabetic nor diabetic; b) prediabetes; c) undiagnosed diabetes, d) well-controlled diabetes; and e) poorly-controlled diabetes. These categories may have included men with either type 1 or type 2 diabetes mellites, though the proportion with type 1 diabetes would be expected to be quite low (<5%).

Data on additional covariates, considered to be potential confounders for the association between diabetes status and serum hormone levels, were obtained from NHANES interviews and examinations. Age, race, smoking status, alcohol consumption and physical activity were self-reported measures assessed though interview-based questionnaires. Race/ethnicity was grouped as non-Hispanic White American, non-Hispanic Black Americans, Hispanic Americans or other/mixed race; smoking status as never smoked (<100 cigarettes over their lifetime), former smoker or current smoker; alcohol consumption as non-drinker (<1 alcoholic beverage per month) non-drinker and physically active (any moderate or vigorous activity in the past week) or non-active. Waist circumference was measured at examination to the nearest 0.1cm at the iliac crest and classified into 3 groups (<88cm, 88–102cm, >102cm). Self-reported high blood pressure and hypercholesteremia (from response to “Have you ever been told by a doctor you have high blood pressure? / high cholesterol?” Only self-reported data on use of either insulin or “pills” for diabetes management was available in NHANES III. It should also be noted that metformin was not in use during the NHANES III study period. Use of diabetes medication was not included as a covariate due to collinearity (i.e., use of diabetes medication only applied to the latter two “diagnosed” groups).

Statistical analyses

All statistical analyses were performed in Stata Version 15.1 (Stata corporation, College Station, Texas USA) by taking account for survey sample weighting with adjustment for the stratified, cluster sampling design, to provide nationally representative estimates.

Descriptive analyses were undertaken comparing proportions and mean values of the cohort according to diabetes groups, with appropriate weighting for sampling. Correlations between serum concentrations of hyperglycemic biomarkers, sex steroids and insulin-like growth factor components were assessed via pair-wise Pearson’s correlation, without adjustment or weighting.

The main analysis assessed sex steroid hormones concentrations (total and free testosterone, total and free estradiol, estradiol to testosterone), SHBG and insulin-like growth factors (IGF-1 and IGFBP-3 according to diabetes status (no disease, prediabetes, undiagnosed diabetes, well controlled diabetes and poorly-controlled diabetes, as defined above) using multivariable linear regression models. we also examine differences in markers of glycemic control across groups (i.e., fasting glucose, HbA1c, insulin and C-peptide concentrations). Since serum concentrations of the biomarkers and hormones were not normally distributed, log transformations were applied and geometric means and 95% confidence intervals (CI) were calculated. Two sets of regression models were performed to adjust for confounders, one with minimal adjustment (model#1) for age, entered as a continuous variable plus quadratic term, and race/ethnicity, and the second model (model#2) with additional adjustment for poverty level, smoking status, alcohol consumption, physical activity, waist circumference (categorical), self-reported hypertension and self-reported hypercholesterolaemia. Comparison of Akaike and Bayesian Information Criterion estimators indicated that adding body mass index (BMI) to the latter model did not improve model fit, hence BMI was not included. Overall differences and linear trends across categories were assessed using ANOVA methods and Wald Tests for diabetes groups as and ordinal variable, respectively, for both models 1 and 2. Predicted means for each of the diabetes categories, fully adjusted for covariates (model#2) are also graphically presented.

Additional sensitivity analyses (which included models1 and 2 for all biomarkers/hormones) were undertaken to examine differences and trends by diabetes status without potential contamination by exogenous insulin use. In these analyses, men on insulin (n=30) were excluded and the remaining ‘diagnosed’ diabetic men were grouped as one category.

Results

Of the 1430 eligible study participants, 648 (45%) had no evidence of diabetes or pre -diabetes, 578 (40%) were classified as having pre-diabetes based on one or more indicator, 106 (7%) had undiagnosed diabetes, 42 (3%) had well-controlled diabetes and 56 (4%) had poorly control disease. Mean age, BMI and waist circumference generally increased across diabetes categories, as did the proportions reporting hypertension and hypercholesterolaemia (Table 1). Conversely, the proportion of current smokers and drinkers decreased across diabetes categories. Characteristics of men with poor diabetes control differed from those with well controlled diabetes in that they were slightly younger and leaner, and less likely to report high blood pressure or high cholesterol.

Table 1.

Characteristics study population reflecting the US males 20+ years 1988–91, by diabetes category¹

	No pre-DM or DM	Pre-DM	Undiagnosed DM	DM good-control	DM poor-control
N (unweighted)	648	578	106	42	56
Weighted to US population:
Age - mean years (SE)	36 (0.7)	47 (0.9)	56 (2.4)	59 (2.6)	56 (5.1)
Race:
Non-Hispanic White (%)	79	75	75	67	79
Non-Hispanic Black (%)	8	11	10	28	15
Hispanic (%)	5	5	6	5	7
Other/Mixed (%)	6	9	8	0	0
In poverty (%)	8	9	17	13	12
Current Smoker (%)	36	34	29	15	15
Current Drinker (%)	74	66	66	38	41
Physically inactive (%)	7	10	22	18	9
BMI - mean (SE) kg/m2	25.3 (0.2)	26.9 (0.5)	29.7 (1.0)	32.6 (3.3)	27.8 (0.9)
Waist Circ. - mean (SE) cm	91 (0.5)	97 (1.6)	102 (2.1)	113 (7.3)	100 (2.0)
High cholesterol (%)	13	19	17	25	16
Hypertension (%)	12	24	58	65	34

¹Diabetes Categories: No pre-DM or DM - neither prediabetic or diabetic according to criteria or self-report; Pre-DM – fasting glucose 5.6– 6.9 or HbA1c 5.7–6.4 & no self-reported diabetes; undiagnosed DM: fasting glucose>6.4 or HbA1c> 6.9 & no self-reported diabetes; DM well-controlled: self-reported DM & HbA1c<7.0; DM poorly-controlled: self-reported DM & HbA1c≥7.0.

Among men with no diabetes or pre-diabetes moderate inverse correlations were observed between insulin and both total and free testosterone levels (and similarly for C-peptide and total/free testosterone) Table 2. SHBG concentrations were weakly, inversely correlated with both insulin and C-peptide levels. No similar correlations were evident for total or free estradiol and glycemia biomarkers. A weak inverse correlation between HbA1c and IGF-1 serum concentration, and a weak positive correlation between and insulin IGF-BP3, were also observed.

Table 2:

Pearson’s correlation coefficient for association between serum biomarkers and hormone levels among men free of prediabetes or diabetes.

No Disease	Glucose	HbA1c	Insulin	C-pep.	Tot T.	Tot E2	E2/T	Free T	Free E2	SHBG	AAG	IGF-1	IGF-bp3
HbA1c	0.14
	0.001
Insulin	0.24	0.12
	0.00	0.00
C-peptide	0.26	0.14	0.78
	0.000	0.000	0.000
Tot. T	−0.16	−0.09	−0.32	−0.39
	0.000	0.027	0.000	0.000
Tot. E2	−0.13	0.08	−0.05	−0.07	0.38
	0.001	0.050	0.178	0.065	0.000
E2/T	−0.02	0.07	0.03	0.05	−0.23	0.43
	0.571	0.054	0.396	0.184	0.000	0.000
Free T	−0.08	−0.11	−0.14	−0.21	0.76	0.40	−0.19
	0.052	0.006	0.000	0.000	0.000	0.000	0.000
Free E2	−0.04	0.07	0.07	0.05	0.26	0.88	0.17	0.54
	0.262	0.083	0.081	0.220	0.000	0.000	0.000	0.000
SHBG	−0.08	0.07	−0.22	−0.19	0.23	0.11	0.24	−0.35	−0.29
	0.033	0.098	0.000	0.000	0.000	0.005	0.000	0.000	0.000
AAG	−0.01	−0.06	0.04	0.04	0.15	0.07	−0.07	0.18	0.10	−0.10
	0.875	0.153	0.307	0.293	0.000	0.067	0.098	0.000	0.013	0.009
IGF-1	−0.08	−0.13	0.04	−0.01	0.14	0.11	0.057	0.32	0.16	−0.29	0.15
	0.057	0.003	0.349	0.816	0.001	0.013	0.177	0.000	0.000	0.000	0.001
IGFBP-3	−0.03	−0.07	0.11	0.03	−0.03	−0.04	0.073	0.17	0.04	−0.36	0.11	0.64
	0.423	0.109	0.012	0.419	0.414	0.361	0.088	0.000	0.360	0.000	0.009	0.000
IGF-1/IGFBP-3	−0.085	−0.11	−0.04	−0.05	0.22	0.17	0.004	0.28	0.18	−0.08	0.11	0.74	−0.01
	0.045	0.010	0.347	0.244	0.000	0.000	0.928	0.000	0.000	0.050	0.013	0.000	0.894

For each biomarker the 1st row is the Pearsons correlation coefficient (r); 2nd row is p-value - statistically significant correlations (r>0.1) shown in bold. HbA1c: glycolated haemoglobin; T: testosterone; SHBG: sex hormone binding globulin, AAG: androstanediol glucuronide; IGF-1: insulin-like growth factor-1; IGFBP-3: insulin-like growth factor binding protein 3

Geometric mean serum concentrations for serum biomarkers by different diabetes categories, derived from the minimally (model 1) and fully (model 2) adjusted models, are shown Table 3. Accompanying Figures 1 and 2 graphically display the results from fully adjusted model. In both models, mean serum concentrations for fasting glucose, HbA1c, and insulin, were higher in men with pre-diabetes and diabetes compared to men without prediabetes/diabetes, with evidence of linear trends (p<0.001). Concentrations of C-peptide were higher in men with prediabetes and undiagnosed diabetes but were not elevated in men with well controlled diabetes and tended to be lower in men with poorly controlled diabetes (i.e., Model 2: no pre-diabetes or diabetes adj. mean: 0.51 nmol/L, 95% CI 0.46–0.57; pre-diabetes: 0.62, 0.57–0.78; undiagnosed diabetes: 0.75, 0.63–0.91; well-controlled diabetes: 0.62, 0.52–0.74; and poorly-controlled diabetes: 0.36, 0.24–0.55, with no overall trend evident.

Table 3.

Adjusted mean serum concentrations (95% confidence intervals) for glucose metabolism biomarkers, sex steroid hormones, sex hormone binding globulin (SHBG) and insulin-like growth factors (IGF), by diabetes (DM) category¹, (US males, 20+yrs, 1989–91)

	model	No pre-DM or DM	Pre-DM	Undiagnosed DM	Well-controlled DM	Poorly-controlled DM	p-diff	p-trend
Fasting Glucose (mmol/L)	1	5.15 (5.11–5.20)	5.77 (5.72–5.82)	7.39 (6.97–7.84)	6.75 (5.98–7.63)	11.4 (9.98–13.0)	na*	na
	2	5.17 (5.12–5.23)	5.76 (5.70–5.81)	7.29 (6.87–7.74)	6.62 (5.90–7.42)	11.3 (9.83–13.0)
HbA1c (%)	1	5.04 (4.97–5.10)	5.28 (5.21–5.36)	5.96 (5.70–6.24)	5.56 (5.32–5.82)	8.70 (8.18–9.25)	na*	na
	2	5.04 (4.98–5.10)	5.28 (5.20–5.35)	5.97 (5.72–6.23)	5.55 (5.28–5.84)	8.71 (8.19–9.27)
Insulin (pmol/L)	1	46.5 (43.9–49.2)	58.7 (53.1–64.8)	86.0 (70.5–105)	87.5 (61.2–125)	83.8 (62.0–113)	<0.001	<0.001
	2	49.2 (46.1–52.4)	56.3 (51.9–61.3)	70.6 (61.2–81.5)	66.8 (50.7–87.9)	76.6 (57.3–102.4)	<0.001	0.004
C-Peptide (nmol/L)	1	0.48 (0.44–0.52)	0.66 (0.59–0.73)	0.96 (0.81–1.15)	0.83 (0.72–0.97)	0.39 (0.26–0.60)	<0.001	0.757
	2	0.51 (0.46–0.57)	0.62 (0.57–0.78)	0.75 (0.63–0.91)	0.62 (0.52–0.74)	0.36 (0.24–0.55)	<0.001	0.157
IGF-1 (ng/mL)	1	282 (265–301)	272 (258–288)	226 (190–269)	251 (209–301)	259 (186–361)	0.157	0.505
	2	281 (265–301)	273 (258–286)	229 (199–265)	256 (213–306)	266 (196–353)	0.092	0.540
IGF-BP3 (ng/mL)	1	4352 (4214–4495)	4403 (4263–4458)	4408 (4056–4792)	3821 (3243–4573)	4353 (4002–4734)	0.263	0.319
	2	4359 (4217–4494)	4412 (4269–4559)	4345 (4039–4674)	3821 (3200–4562)	4315 (38925–4784)	0.295	0.251
Ratio IGF-1 to IGF-BP3 (%)	1	6.49 (6.20–6.79)	6.19 (5.89–6.50)	5.13 (4.59–5.73)	6.51 (5.82–728)	5.95 (4.58–7.74)	0.011	0.675
	2	6.45 (6.20–6.79)	6.15 (5.89–6.46)	5.29 (4.78–5.85)	6.69 (5.95–7.51)	6.06 (4.81–7.64)	0.015	0.863
Testosterone (T) (ng/mL)	1	5.41 (5.15–5.68)	4.82 (4.52–5.12)	4.62 (3.97–5.38)	4.34 (3.81–4.93)	4.78 (4.21–5.43)	0.010	0.017
	2	5.29 (5.06–5.53)	4.91 (4.61–5.23)	4.89 (4.34–5.51)	4.84 (4.41–5.31	4.91 (4.40–5.48)	0.063	0.174
Estradiol (E2) (pg/mL)	1	36.2 (34.8–37.8)	34.8 (32.9–36.8)	37.9 (35.4–40.6)	37.0 (33.5–40.9)	36.8 (32.5–41.7)	0.301	0.479
	2	36.3 (34.8–37.9)	34.8 (33.1–36.6)	37.2 (34.7–39.8)	37.3 (33.7–41.2)	37.4 (33.2–42.4)	0.117	0.355
Est. Free T (ng/mL)	1	0.105 (0.100–0.110)	0.099 (0.093–0.105)	0.096 (0.085–0.108)	0.085 (0.070–0.103)	0.101 (0.091–0.112)	0.213	0.149
	2	0.104 (0.099–0.109)	0.099 (0.094–0.106)	0.098 (0.088–0.110)	0.091 (0.076–0.108)	0.103 (0.092–0.114)	0.404	0.471
Est. Free E2 (pg/mL)	1	0.907 (0.865–0.951)	0.910 (0.856–0.967)	0.998 (0.916–1.086)	0.952 (0.833–1.088)	0.983 (0.856–1.128)	0.223	0.183
	2	0.917 (0.874–0.962)	0.900 (0.850–0.953)	0.959 (0.889–1.035)	0.933 (0.817–1.066)	0.994 (0.865–1.138)	0.419	0.221
SHBG (nmol/L)	1	36.6 (34.5–38.7)	32.6 (30.5–35.0)	31.4 (26.2–37.6)	34.6 (28.3–42.3)	31.4 (26.3–37.5)	0.123	0.224
	2	35.7 (33.6–37.9)	33.4 (31.2–35.6)	33.8 (29.6–38.5)	38.3 (30.9–47.5)	32.0 (26.3–39.1)	0.324	0.734
AAG (ng/mL)	1	11.6 (10.8–12.4)	11.9 (10.7–13.2)	13.2 (11.0–15.9)	13.8 (11.3–16.7)	10.0 (8.7–11.6)	0.136	0.394
	2	11.6 (10.8–12.4)	11.9 (10.7–13.2)	12.8 (10.4–15.8)	13.2 (10.9–16.0)	9.9 (8.5–11.5)	0.333	0.228
Ratio Total E2/T (×1000)	1	6.69 (6.23–7.18)	7.23 (6.71–7.81)	8.21 (7.17–9.40)	8.53 (7.12–10.2)	7.71 (6.70–8.86)	0.008	0.016
	2	6.86 (6.41–7.34)	7.08 (6.53–7.67)	7.61 (6.82–8.49)	7.71 (6.58–9.03)	7.62 (6.77–8.58)	0.212	0.045

¹Diabetes Categories: No pre-DM or DM - neither prediabetic or diabetic according to criteria or self-report; Pre-DM – fasting glucose 5.6–6.9 or HbA1c 5.7–6.4 & no self-reported diabetes; undiagnosed DM: fasting glucose>6.4 or HbA1c> 6.9 & no self-reported diabetes; DM well-controlled: self-reported DM & HbA1c<7.0; DM poorly-controlled: self-reported DM & HbA1c≥7.0. HbA1c: glycosylated haemoglobin; IGF: insulin-like growth factor; IGFBP: insulin-like growth factor binding protein; AAG: androstanediol glucuronide; T: total testosterone; E₂:.Total estradiol.

Geometric means derived from GLM (linear) models (with survey weighting). adjusting for age and race (model 1), plus smoking status, alcohol consumption, waist circumference, social deprivation, self-reported high blood pressure, self-reported hyper-cholesterol (model 2). Tests for difference/trends not performed for variables used to derive categories.

Figure 1.

Adjusted geometric mean serum concentrations, by diabetes (DM) category, of a) glycosylated haemoglobin (HbA1c) %; b) insulin (pmol/L); c) C-peptide (mg/dL); d) insulin-like growth factor-1 (ng/mL); and e) insulin-like growth factor binding protein-3 (ng/mL); f) ratio of IGF-1 to IGF-BP3, in US men 20+yrs 1988–1991. Derived from fully adjusted model (#2)

[Diabetes Categories: No pre-DM or DM - neither prediabetic or diabetic according to criteria or self-report; Pre-DM – fasting glucose 5.6– 6.9 or HbA1c 5.7–6.4 & no self-reported diabetes; undiagnosed DM: fasting glucose>6.4 or HbA1c> 6.9 & no self-reported diabetes; DM well-controlled: self-reported DM & HbA1c<7.0; DM poorly-controlled: self-reported DM & HbA1c≥7.0.]

Figure 2.

Adjusted geometric mean serum concentrations, by diabetes (DM) category, of a) total testosterone (ng/mL), b) total estradiol (pg/mL), c) free testosterone (ng/mL) d) free estradiol (pg/mL), e) sex hormone binding globulin (SHBG) (nmol/L), f) androstanediol glucuronide (AAG) (ng/mL), g) estradiol to testosterone ratio (x1000), and h) testosterone to SHBG ratio, among US men 20+yrs 1988–1991.

Geometric mean serum concentrations of IGF-1 and IGFBP-3 did not differ across any diabetes groups and no trends were evident. The mean ratio of IGF-1 to IGFBP-3 was significantly lower in men in the prediabetes group, and lower again in men with undiagnosed diabetes, but did not differ for men with either well-controlled or poorly controlled disease compared with those without prediabetes/diabetes (for models 1 and 2). While statistically significant differences were seen among groups (model 2 p=0.015), no statistically significant trend was evident. This U-shaped pattern approximated the inverse of what was observed in relation to C-peptide levels.

In minimally adjusted models (model1), the geometric mean concentration of total testosterone was significantly lower in men with prediabetes (4.82 ng/ml, 95% CI 4.52–5.12) compared with non-diabetic men (5.41, 95% CI 5.15–5.68), with p-value for any difference of 0.010. Geometric mean total testosterone was similarly low in each of the other diabetes groups, however there was no indication of decreasing levels with worsening disease. In the fully adjusted model, differenced in geometric mean total testosterone were attenuated (p-difference=0.063). No trends or differences across diabetes groups were seen for other sex steroid hormone including estradiol, free testosterone, free or total estradiol, and AAG, or for SHBG. The mean ratio of estradiol to testosterone increased across diabetes groups (model 2, p-trend=0.0045), with highest ratios among men with well controlled and poorly controlled diabetes (7.71, 95% CI 6.58–9.03; and 7.62, 6.77–8.58, respectively).

Sensitivity analyses, in which insulin users were excluded, indicated very similar findings to the main analyses. These include: a linear inverse trend in the mean total testosterone concentration in the minimally (p-trend=0.01), but not fully adjusted model and a positive trend in the mean ratio of estradiol to testosterone across diabetes groups in both the minimal (p-trend=0.002) and fully adjusted models (p-trend=0.032). Overall differences in IGF-1 and the ratio of IGF-1 to IGFBP-3 were statistically significant in both models though linear trends across diabetes categories were not observed in either case. Supplementary table 1

Discussion

Findings from our study do not provide clear evidence for either the sex steroid hormone pathway or the IGF pathway as biological mechanisms explaining lower risk of prostate cancer among men with diabetes. However, they do hint at the potential for both mechanisms to be involved. Relative to men without prediabetes/diabetes, we found lower mean total testosterone concentrations in all categories of diabetes including prediabetes. However, the inverse trend for total testosterone seen with progressively worse diabetes when adjusted for age and race, was attenuated and not statistically significant when fully adjusted for other covariates. With the exception of the ratio of total estradiol to testosterone, for which there was an increasing trend across diabetes categories, no differences or trends were evident for free or total estradiol, free testosterone, SHBG or AAG (as a marker of androgen activity). No trends were observed for IGF-1 and IGFBP-3, separately or as a ratio, with worsening categories of diabetes. However, the mean concentration of IGF-1 was lower in men with undiagnosed diabetes, while the ratio of IGF-1 to IGFBP-3 (an indicator of bioavailable IGF-1) was lower among prediabetic and undiagnosed diabetic men compare with men with normoglycemia.

Sex steroid pathway

Studies have consistently shown lower levels of total testosterone in men with type 2 diabetes [27–29]. Whether decreased risk of prostate cancer associated with diabetes is due to reduced levels of circulating total testosterone has been contested. Despite the long-standing acceptance that prostate cancer is an androgen-driven cancer, epidemiological evidence for the association between circulating testosterone and prostate cancer risk is mixed, with meta-analyses and pooled analyses reporting no overall association between total testosterone levels and prostate cancer risk [30, 31]. Whether lower testosterone levels decrease prostate carcinogenesis or reduce detectability (through lower PSA and reduced biopsies) remains an open question.

Other evidence suggests that free testosterone (unbound to SHBG or albumin) rather than total testosterone may be a more important driver of prostate cancer risk. A recent pooled analysis found lower overall risk of prostate cancer among men among men in the lowest vs. highest decile of free testosterone but no association with total serum testosterone levels [32]. Findings from a recent UK Biobank study also point to a positive association between free testosterone and prostate cancer risk, and no association for total testosterone [33].

The current study did not find evidence of a decreasing trend in mean total testosterone levels with worsening disease, when adjusting for multiple covariates. If the level of circulating testosterone were the main driver of the association between diabetes and prostate cancer, we would have expected to see a negative trend, given evidence indicating a stronger inverse association with increasing duration of diabetes. Also, despite lower testosterone levels in men with pre-diabetic compared with non-diabetic men, prostate cancer risk is not reduced among pre-diabetics [34]. One possibility is that sustained periods of low testosterone (or lower free testosterone) are required for prostate cancer risk to be reduced.

While we did not find any differences (or trends) for free testosterone or SHBG levels among pre-diabetic or diabetic men, contrary to previous reports [29, 35], we did see evidence of a positive trend for the ratio of total estradiol to testosterone across worsening diabetes categories. While higher ratios of estrogen to testosterone have been observed in association with insulin resistance [36], evidence for higher ratios of estrogen to testosterone are associated with decreased risk of prostate cancer is lacking. Since the ratio of estrogen to testosterone tends to increase with age, as does prostate cancer, and is higher in men of African descent who are at higher risk of prostate cancer, and lower in Asian men who generally have a lower risk of prostate cancer [10], a positive trend in estrodiol to testosterone ratios seems inconsistent with lower risk of prostate cancer among diabetic men. Alternatively, our findings may reflect some residual confounding rather than a true effect, given that the ratio of estrogen to testosterone increases with greater visceral adipose tissue and age [37], even though our models included a quadratic term for age to limit the potential for residual confounding by age.

Insulin resistance/IGF pathway

Another hypothesis put forward to explain the inverse association between diabetes and prostate cancer relates to changes in the level of (bioavailable) IGF-1 [38]. IGF-1, the most abundant of all IGFs, is thought to play a major role in carcinogenesis for numerous cancer types including prostate cancer, given it functions as a modulator of cell proliferation, differentiation and apoptosis across different tissue types. Bioavailability of IGF-1 is regulated through sequestration by IGF binding proteins, predominantly IGFBP-3. When bound, IGF-1 cannot pass through capillaries to access target tissue, rendering it inactive. Hence, lower levels of IGF-1 (or elevated levels of IGFBPs), which may occur in association with insulin resistance, could explain decreased prostate cancer risk among diabetics. Previous cohort studies and meta-analyses have consistently found a positive association between IGF-1 and risk of prostate cancer [33, 39–42]. However, evidence is weaker for any association between IGFBP-3 and prostate cancer risk, though a pooled analysis by Roddam et al found a positive association for IGFBP-3 as well as IGF-1 [40].

Our findings are somewhat but not entirely consistent with the insulin/IGF-1 hypotheses. The decrease in mean C-peptide concentration from the highest levels in undiagnosed diabetics to lowest levels among poorly-controlled diabetics indicates increasingly impaired insulin production with progressively worse disease. Hence pathways relating to impaired insulin production may be implicated in lowering prostate cancer risk. However, in contrast to a previous NHANES III study which found an inverse association among younger men [43], we found no trend in IGF-1 levels across diabetes groups. Furthermore, while the ratio of IGF-1 to IGFBP-3 (an indicator of bioavailable IGF-1) was lower among men with prediabetes and undiagnosed diabetes, no differences were seen among men with well-controlled or poorly-controlled diabetes compared with non-diabetics, and no trend across diabetes groups was evident. While these findings do not align with evidence indicating no difference in prostate cancer risk among prediabetic men [34], or decreasing risk with increasing duration of diabetes [3], they do hint at a reduction in bioavailable IGF-1 during the early phases of diabetes which may contribute to reduced risk of prostate cancer over the longer term.

Strengths and limitations

We conducted this study using a nationally representative sample from NHANES III, and applied survey weights to ensure generalizable of our findings to adult men in the US. Given the broad nature of data collected in NHANES, we were able to adjust for multiple potential confounders when estimating mean hormone levels. Despite adjusting for age, confounding due to differences in age may still be an issue given the tendency for men without diabetes to be younger. Another major limitation of this study is the potential for measurement error in relation to exposures, outcomes and covariates. While we used only morning samples for measurement of various hormone levels, to reduce extraneous variation due to diurnal production of sex hormones stable measures, the reliance of single serum measurement rather than repeat measures may have increased random error due to natural variability in concentrations. Our classification criteria for different diabetes categories relied, in part, on self-reported diabetes status, which may have led to some inaccuracy in our exposure measure. Likewise, some covariates (e.g., smoking status, alcohol consumption, physical activity) were also self-reported measures and hence prone to recall and reporting bias.

Furthermore, as this was a cross-sectional study, the ability of examine the temporal nature of potential associations is limited. Indeed, there is some evidence that low testosterone and potentially low IGF levels are precursors rather than consequences of diabetes [44, 45]. Longitudinal study designs with multiple serum measures are required to further disentangle potential biological mechanisms that may explain the apparent inverse relationship between diabetes and prostate cancer risk. Finally, the lack of statistically significant findings in relation to differences or trends with worsening disease may be due to lack of power resulting from the small number of cases within self-reported diabetes groups. Despite the low numbers, there are few other nationally representative data sources with measured hormones and biomarkers in which such a study can be conducted.

Conclusions

Biological pathways explaining the inverse relationship between diabetes and prostate cancer are likely to be complex. Our findings do not clearly support either a testosterone driven or IGF-1 driven pathway alone. Initially lower levels of bioavailable IGF-1 in prediabetics and those with undiagnosed diabetes could contribute to lower prostate risk, but since this state did not appear to be sustained as diabetes progressed it is unlikely to be the sole biological pathway. Sustained levels of low testosterone may play some role in lowering risk of prostate cancer among men with diabetes, though or data suggest that the association may be more complex and could involve the balance of androgens with estrogen levels. Further exploration involving longitudinal study designs is warranted to disentangle the relationship between diabetes and prostate cancer risk.

Data Availability Statement:

Data are available at https://wwwn.cdc.gov/nchs/nhanes/nhanes3/default.aspx