Insulin-like growth factors (IGFs) are potent mitogens and antiapoptotic agents that are involved in the regulation of cellular proliferation, differentiation, and apoptosis.1–3 The action and availability of IGFs to bind to IGF receptors to exert their action is modulated by at least 6 high-affinity IGF-binding proteins (IGFBPs)4, 5; > 95% of the circulating IGF-I is complexed with IGFBP-3 and an acid-labile subunit.6 Although most circulating IGF-I and most IGFBPs are produced by the liver, they also are produced locally by a wide variety of tissues.2
There are several reasons to suspect that the circulating levels of IGF-I and IGFBP-3 may influence the risk of prostate carcinoma. In in vitro studies, IGF-I modulates the growth of the prostate carcinoma cell lines LNCaP, PC-3, and DU-145.7, 8 In the human LNCaP model of prostate carcinoma progression, there is a marked increase in autocrine production of IGF-I in cells that metastasize to the bone.9 The transgenic adenocarcinoma prostate model in mice involves elevated prostatic IGF-I mRNA expression that correlates with tumor progression, and serum IGF-I levels in these mice are elevated compared with the levels in nontransgenic animals.10 It has been shown that prostate cells and tissues contain IGF-I and IGF-II proteins7, 11–13 as well as IGF mRNA,12, 14–19 IGF-I receptor,14, 15 and IGFBP-2–IGFBP-6. In addition to the modulation of IGF availability and action, it has been shown that IGFBPs inhibit cellular growth18, 20–22 and IGF-I receptor activation.23
Many epidemiologic studies have investigated the association between the risk of prostate carcinoma and circulating levels of IGFs and IGFBPs. To overcome the potential bias resulting from the presence of prostate carcinoma and its influence on levels of IGF-I and IGFBP-3, six of those studies24–29 assayed blood specimens that were drawn from men prior to their prostate carcinoma diagnosis. However, no consistent association has emerged either from those six studies or from the other studies in which blood samples were obtained from men after they were diagnosed. Similarly, although the results of some studies suggest that high levels of IGFBP-3 are associated with a reduced risk of prostate carcinoma,24, 27, 30–32 the results from other studies do not.25, 29, 33–36
Insulin regulates the bioavailability of IGF-I acutely through its regulation of IGFBP synthesis in the liver.37 In a case–control study in China,38 but not in another case–control study in Sweden,25 relatively high plasma levels of insulin were observed in men with prostate carcinoma. In the current study, we attempted to reexamine the possible influence of circulating levels of IGF-I, IGFBP-3, and insulin on the occurrence of prostate carcinoma using data from a large cohort study of elderly Americans.
MATERIALS AND METHODS
This case–control study was conducted among participants in the Cardiovascular Health Study (CHS), a population-based cohort study that was designed primarily to investigate coronary heart disease and stroke in the elderly.39 The study recruited individuals age ≥ 65 years using Medicare eligibility lists from 4 communities in the U.S.: Sacramento County, CA; Forsyth County, NC; Pittsburgh, PA; and Washington County, MD.40 The Institutional Review Board at each participating center approved the study, and each participant provided informed consent. In total, 5888 men were enrolled, including 5201 men who were enrolled in 1989–1990 and an additional 687 African-American men who were enrolled in 1992–1993. Participants completed up to 10 annual clinic examinations from baseline until 1998–1999. A fasting blood specimen was collected during these examinations as well as demographic data and medical history, which were collected using a standardized questionnaire. Surveillance for all hospitalizations with extensive medical record collection was initiated and continues to be performed.41
Case and Control Ascertainment
The group of men with incident prostate carcinoma (cases) was identified by linking CHS data to available records from 1989–1999 of population-based cancer registries serving the 4 CHS communities. The registries were the California Cancer Registry for the years 1989–1999; North Carolina Central Cancer Registry for the years 1990–1999; Pennsylvania Department of Health, Bureau of Health Statistics and Research for the years 1989–1999; Maryland Cancer Registry for the years 1992–1999; and Johns Hopkins Training Center for Public Health Research for the years 1989–1999. Data from both Johns Hopkins University and the state of Maryland were used to allow us cover all years of interest for the study. It was estimated that cancer registry records were 93–100% complete, as measured by the standards of the North American Association of Central Cancer Registries. We included cases either if they had a registry-confirmed diagnosis of prostate carcinoma or if they had documentation of prostate carcinoma by both self-report (at the time of an annual CHS examination) and a hospital discharge diagnosis code in CHS records.
For 80% of both cases and controls, a plasma specimen from the initial blood draw was made available to us; for most of the remaining men, the specimens had been obtained 3 years later. The mean time (± standard deviation) between blood collection and diagnosis was 3.4 years ± 1.6 years. A minimum of 1 year after enrollment into the CHS was required prior to the date of the prostate carcinoma diagnosis. Information regarding body mass index (BMI) and other potential confounding variables was taken from the data collected nearest in time to the blood draw.
Men were eligible to serve as controls only if they had no documentation of prostate carcinoma either in the cancer registries, in self-report at the examination, or in the abstracted medical records for all hospitalizations. Potential controls, as well as 23 cases, were excluded if they had incurred an incident myocardial infarction, angina pectoris, or stroke and therefore their blood specimens were reserved for studies of cardiovascular disease. Cases were frequently matched to potential controls by age at enrollment (within 3 years), race, year of entry, and year of blood draw. Controls also were required to have survived at least to the age at diagnosis of the matched case to ensure that cases and controls experienced a comparable risk period for the development of disease.
Determination of Plasma Concentrations of IGF-I, IGFBP-3, and Prostate-Specific Antigen
Plasma samples containing ethylenediamine tetraacetic acid as an anticoagulant were shipped from the CHS Central Blood Laboratory at the University of Vermont to the Fred Hutchinson Cancer Research Center Chen Laboratory for analyses. Plasma IGF-I and IGFBP-3 concentrations were determined using 2-site immunoradiometric assay (IRMA) kits from Diagnostic Systems Laboratories, Inc. (Webster, TX). This is a noncompetitive assay in which the IGF-I or IGFBP-3 in the sample is sandwiched between 2 antibodies specific to these analytes. The first antibody is immobilized to the inside wall of the tubes, and the second antibody is radiolabeled with 125I for detection. Unbound material is removed by decanting and washing the tubes. Prior to the IRMA for IGF-I, the sample was treated with acid-ethanol to release IGF-I from its binding proteins. There is no cross-reactivity with IGF-II or insulin, and the results correlate well with those obtained using the reference method of acid-column chromatography. Assay performance for each run was monitored by analysis of two levels of kit controls and one in-house pooled control. For IGF-I, the intraassay percentage coefficients of variation were 3.0%, 3.5%, and 3.8% at 59 ng/mL, 214 ng/mL, and 275 ng/mL, respectively, and the interassay percentage coefficients of variation were 6.6%, 10.7%, and 12.3% at 65 ng/mL, 173 ng/mL, and 194 ng/mL, respectively. For IGFBP-3, the intraassay percentage coefficients of variation were 2.1%, 2.9%, 5.8% at 7.8 ng/mL, 16.4 ng/mL, and 4347 ng/mL, respectively; and the interassay percentage coefficients of variation were 6.4%, 4.1%, and 7.1% at 7.3 ng/mL, 15.5 ng/mL, and 3943 ng/mL, respectively. Prostate-specific antigen (PSA) concentrations were measured using an IRMA kit from Diagnostic Systems Laboratories. The intraassay percentage coefficients of variation of three DSL controls and BioRad Lyphochek Immunoassay Plus Control Level 2 were 4.0%, 1.9%, 1.6%, and 2.5% at 0.24 ng/mL, 3.91 ng/mL, 20.71 ng/mL, and 4.64 ng/mL, respectively; and the interassay percentage coefficients of variation were 7.2%, 5.7%, 8.2%, and 4.0% at 0.25 ng/mL, 4.07 ng/mL, 20.78 ng/mL, and 4.70 ng/mL, respectively.
Determination of Fasting Insulin
Fasting serum insulin levels were analyzed by the CHS Central Blood Laboratory at the University of Vermont using a competitive radioimmunoassay from Diagnostic Products Corporation (Los Angeles, CA).42
One hundred seventy-five cases and 175 controls had serum or plasma samples. One case with serum and one control with plasma were excluded because neither serum nor plasma was available for the matched control/case. This left 173 matched case–control pairs with plasma and one case with serum (the matched control had plasma) for analysis.
The serum level for IGF-I was divided by 1.08 according to the Diagnostic Systems Laboratories package insert, so that the IGF-I value would be equivalent to plasma values of IGF-I. Quartile cut-off points for IGF-I, IGFBP-3, the IGF-I/IGFBP-3 molar ratio, and insulin were determined from the distribution of these values in controls. The natural logs of serum levels of IGF-I, IGFBP-3, and insulin and the BMI were used to compute the partial Pearson correlation coefficient using STATA software (version 8; Stata Corporation, College Station, TX). “Aggressive tumors” were defined as those that were Stage C or D (i.e., spread beyond the prostate itself) at the time of diagnosis or those that were poorly differentiated (i.e., Grade 3) irrespective of stage. Stage and grade were based on SEER Registries criteria.
Conditional logistic regression was used to calculate odds ratios as an estimate of the relative risk and to compute 95% confidence intervals (95% CIs) using version 8 of the STATA software. Each case–control set was matched based on race (white, African American, or other), age at entry (± 4 years), year of entry into cohort (1989–1990 or 1992–1993), and year of blood draw. The following variables were evaluated as potential confounders: marital status, education, aspirin use, any nonsteroidal antiinflammatory drug use, and ratio of waist-to-hip circumference. None of these variables altered the odds ratios by > 10% or improved model fit.
Cases and controls were similar with respect to their distributions of the matching variables (age and race), anthropometric measurements, and educational level (Table 1). Plasma levels of IGF-I were found to be correlated strongly with those of IGFBP-3 (partial Pearson correlation coefficient = 0.75; P < 0.001) but were not correlated at all with levels of insulin. Plasma levels of insulin were correlated positively with BMI (partial Pearson correlation coefficient = 0.46; P < 0.001).
Table 1. Selected Characteristics of Cases and Controls at the Time of Blood Draw
|Age (yrs)|| || |
| 65–69||56 (32.2)||57 (32.8)|
| 70–74||75 (43.1)||74 (42.5)|
| 75–79||36 (20.7)||36 (20.7)|
| ≥ 80||7 (4.0)||7 (4.0)|
|Height (inches)|| || |
| ≤ 66.0||34 (19.5)||34 (19.5)|
| From 66.1 to< 70.0||94 (54.0)||89 (51.2)|
| ≥ 70.0||46 (26.4)||51 (29.3)|
|BMI (kg/m2)|| || |
| < 25.0||58 (35.6)||59 (35.5)|
| 25.0–28.99||75 (46.0)||74 (44.6)|
| ≥29.0||30 (18.4)||33 (19.9)|
|Waist-to-hip circumference ratio|| || |
| ≤ 0.9||19 (10.9)||15 (8.6)|
| 0.91–0.99||109 (62.6)||115 (66.1)|
| ≥ 1.0||46 (26.4)||44 (25.3)|
|Race|| || |
| Caucasian||131 (75.3)||131 (75.3)|
| African American||41 (23.6)||41 (23.6)|
| Other||2 (1.2)||2 (1.2)|
|Smoking history|| || |
| Never smoked||59 (33.9)||47 (27.0)|
| Former smoker||100 (57.5)||103 (59.2)|
| Current smoker||15 (8.6)||24 (13.8)|
|Education|| || |
| ≤ High school or GED||86 (49.7)||84 (48.6)|
| Some college or vocation school||40 (23.1)||40 (23.1)|
| ≥ College graduate||27 (27.2)||49 (28.3)|
|Diabetes|| || |
| No||122 (70.5)||117 (67.2)|
| Impaired fasting glucosea||28 (16.2)||29 (16.7)|
| Diabetesb||23 (13.3)||28 (16.1)|
| Using insulin||1 (0.6)||2 (1.2)|
| Using hypoglycemics||12 (6.9)||12 (6.9)|
There were only modest overall differences noted between cases and controls with regard to the mean plasma concentrations of IGF-I, IGFBP-3, and insulin (Table 2). We did not observe a positive association between plasma IGF-I concentrations with the incidence of prostate carcinoma among the CHS study participants (Table 3). Relative to the men who had IGF-I levels in the first (lowest) quartile of the distribution, the risk of prostate carcinoma among men who had IGF-I levels in the second, third, and fourth (upper) quartiles were 0.77 (95% CI, 0.43–1.38), 0.73 (95% CI, 0.41–1.31), and 0.67 (95% CI, 0.37–1.25), respectively. There was a suggestion of decreased risk of prostate carcinoma with an increased plasma level of IGFBP-3: The corresponding relative risks were 0.91 (95% CI, 0.49–1.68), 0.47 (95% CI, 0.24–0.94), and 0.65 (95% CI, 0.34–1.20). The relative risks associated with plasma levels of both IGF-I and IGFBP-3 were found to be influenced little by adjustment for the other analyte (Table 3). For the molar IGF-I/IGFBP-3 ratio, the relative risks in men who had ratios in the second, third, and fourth quartiles of distribution were 0.71 (95% CI, 0.40–1.27), 0.83 (95% CI, 0.48–1.44), and 0.76 (95% CI, 0.43–1.36), respectively.
Table 2. Mean ± Standard Deviation of Plasma Analyte Concentrations in Men with Prostate Carcinoma (Cases) and in Controls
|IGF-1 ng/mL||157.7 ± 94.5||163.2 ± 77.7|
|IGFBP-3 ng/mL||3101 ± 924||3210 ± 843|
|Insulin IU/mLa||14.5 ± 8.6||16.3 ± 17.9|
Table 3. The Risk of Prostate Carcinoma in Relation to Plasma Levels of Insulin-Like Growth Factor I, Insulin-Like Growth Factor Binding Protein 3, and Insulin
|IGF-I (ng/mL)|| || || || || |
| Median||87||131||163||259|| |
| Range||38–112||113–148||148–192||193–917|| |
| No. of cases/controls||54/43||43/44||40/44||37/43|
| OR (95% CI)|| || || || || |
| Matched||1.0||0.77 (0.43–1.38)||0.73 (0.41–1.31)||0.67 (0.37–1.25)||0.20|
| Adjusteda||1.0||0.85 (0.44–1.68)||0.72 (0.37–1.43)||0.77 (0.33–1.84)||0.45|
|IGFBP-3 (ng/mL)|| || || || || |
| Median||2169||2942||3489||4118|| |
| Range||1098–2601||2608–3263||3273–3779||3796–6081|| |
| No. of cases/controls||54/43||54/44||28/43||38/44|| |
| OR (95% CI)|| || || || || |
| Matched||1.0||0.91 (0.49–1.68)||0.47 (0.24–0.94)||0.65 (0.34–1.20)||0.06|
| Adjusted ORb||1.0||0.85 (0.43–1.69)||0.45 (0.21–0.99)||0.63 (0.29–1.37)||0.11|
|IGF-I/IGFBP-3 molar ratio|| || || || || |
| No. of cases/controls||54/44||38/44||41/42||41/44|
| OR (95% CI)|| || || || || |
| Matched||1.0||0.71 (0.40–1.27)||0.83 (0.48–1.44)||0.76 (0.43–1.36)||0.44|
| Adjustedc||1.0||0.73 (0.40–1.31)||0.81 (0.46–1.43)||0.75 (0.41–1.36)||0.40|
|Insulin (IU/mL)|| || || || || |
| Median||8||11||14||24|| |
| No. of cases/controls4||43/47||49/42||40/40||40/43|
| OR (95% CI)|| || || || || |
| Matched||1.0||1.30 (0.72–2.32)||1.08 (0.57–2.04)||1.00 (0.54–1.85)||0.88|
| Adjustede||1.0||1.20 (0.66–2.18)||0.98 (0.50–1.91)||0.89 (0.45–1.76)||0.65|
Of the 174 men with prostate carcinoma, information regarding disease stage and/or tumor grade was available for all but 24 men. Of these, 98 men had nonaggressive disease at the time of diagnosis. A comparison of this subgroup with their controls suggested either no association or, at most, a modest positive association with IGF-I levels in plasma: The risks of prostate carcinoma among men in the second, third, and fourth quartiles of distribution, relative to men in the first (lowest) quartile, were 1.29 (95% CI, 0.56–2.99), 0.80 (95% CI, 0.33–1.94), and 1.69 (95% CI, 0.72–4.0), respectively (Table 4). The corresponding relative risks among men in the second, third, and fourth quartiles of the IGFBP-3 distribution were 1.44 (95% CI, 0.63–3.28), 0.8 (95% CI, 0.32–2.03), and 1.26 (95% CI, 0.56–2.82). The levels of IGF-I and IGFBP-3 tended to be lower among men who had aggressive tumors that spread beyond their prostate compared with their controls. The risk of prostate carcinoma among men in the fourth (upper) quartile of the IGF-I distribution, compared with men in the first (lowest) quartile, was only 0.17 (95% CI, 0.05–0.62) (Table 4). The corresponding relative risk among men in the fourth (upper) quartile of the IGFBP-3 distribution was 0.24 (95% CI, 0.07–0.84).
Table 4. Odds Ratios and 95% Confidence Intervals for Prostate Carcinoma According to Plasma Levels of Insulin- Like Growth Factor 1 and Insulin-Like Growth Factor Binding Protein 3, the Molar Ratio of Insulin-Like Growth Factor 1/Insulin-Like Growth Factor Binding Protein 3, and Insulin Levels by Aggressiveness of Diseasea
|IGF-1|| || || || |
| No. of nonaggressive cases||21||28||20||29|
| No. of matched controls||24||26||27||21|
| OR (95% CI) for nonaggressive disease||1.0||1.29 (0.56–2.99)||0.80 (0.33–1.94)||1.69 (0.72–4.00)|
| No. of aggressive cases||24||10||12||6|
| No. of matched controls||10||14||13||15|
| OR (95% CI) for aggressive disease||1.0||0.34 (0.11–1.02)||0.46 (0.16–1.29)||0.17 (0.05–0.62)|
|IGFBP-3|| || || || |
| No. of nonaggressive cases||24||30||17||27|
| No. of matched controls||27||23||24||24|
| OR (95% CI) for nonaggressive disease||1.0||1.44 (0.63–3.28)||0.80 (0.32–2.03)||1.26 (0.56–2.82)|
| No. of aggressive cases||24||14||5||9|
| No. of matched controls||12||14||11||15|
| OR (95% CI) for aggressive disease||1.0||0.46 (0.14–1.52)||0.18 (0.04–0.75)||0.24 (0.07–0.84)|
|IGF-1/IGFBP-3c|| || || || |
| No. of nonaggressive cases||27||22||18||31|
| No. of matched controls||24||24||25||25|
| OR (95% CI) for nonaggressive disease||1.0||0.85 (0.37–1.98)||0.68 (0.32–1.46)||1.10 (0.52–2.31)|
| No. of aggressive cases||19||10||15||8|
| No. of matched controls||13||13||11||15|
| OR (95% CI) for aggressive disease||1.0||0.55 (0.21–1.47)||1.07 (0.41–2.83)||0.34 (0.10–1.10)|
|Insulin|| || || || |
| No. of nonaggressive casesd||20||28||27||22|
| No. of matched controlsd||26||20||23||28|
| OR (95% CI) for nonaggressive disease||1.0||1.88 (0.82–4.31)||1.68 (0.70–4.06)||1.08 (0.48–2.44)|
| No. of aggressive cases||14||16||8||13|
| No. of matched controls||14||13||12||12|
| OR (95% CI) for aggressive disease||1.0||1.40 (0.47–4.17)||0.57 (0.17–1.92)||1.11 (0.34–3.57)|
|Insulin levels adjusted for IGF-1 and IGFBP-3|| || || || |
| OR (95% CI) for nonaggressive diseased||1.0||1.90 (0.82–4.41)||1.77 (0.72–4.36)||1.14 (0.49–2.62)|
| OR (95% CI) for aggressive diseased||1.0||1.23 (0.36–4.14)||0.70 (0.19–2.59)||1.06 (0.29–3.91)|
The majority of men who were diagnosed later with prostate carcinoma had a PSA concentration > 4 ng/mL in their prediagnostic blood specimen. In these men, it is likely that the carcinoma already was present at that time. In an analysis based on the subset of cases (and their matched controls) in whom the baseline plasma PSA level was not elevated (and, thus, in whom prostate carcinoma may not yet have been present), increasing levels of both IGF-I and IGFBP-3 were associated with a decreasing risk of prostate carcinoma (Table 5). Serum insulin levels bore no relation to the incidence of prostate carcinoma whether in all participants (Table 3) or in subsets of participants defined by stage of disease (Table 4) or baseline PSA value (Table 5).
Table 5. Risk of Prostate Carcinoma in Relation to Plasma Levels of Insulin-Like Growth Factor 1, Insulin-Like Growth Factor Binding Protein 3, and Insulin by Prostate-Specific Antigen Values among Cases and Matched Controls
|IGF-I|| || || || || |
| Cases with PSA < 4 ng/mL|| || || || || |
| No. of cases/controls||17/9||9/9||8/11||5/10|| |
| Matched OR (95% CI)a||1.0||0.53 (0.14–2.05)||0.38 (0.10–1.43)||0.30 (0.08–1.14)||0.04|
| Cases with PSA ≥ 4 ng/mL|| || || || || |
| No. of cases/controls||37/34||34/35||32/33||32/33|| |
| Matched OR (95% CI)a||1.0||0.89 (0.46–1.72)||0.89 (0.46–1.73)||0.88 (0.43–1.79)||0.74|
|IGFBP-3|| || || || || |
| Cases with PSA < 4 ng/mL|| || || || || |
| No. of cases/controls||17/7||11/14||5/8||6/10|| |
| Matched OR (95% CI)a||1.0||0.17 (0.03–0.98)||0.14 (0.02–0.99)||0.18 (0.03–1.03)||0.02|
| Cases with PSA ≥ 4 ng/mL|| || || || || |
| No. of cases/controls||37/36||43/30||23/35||32/34|| |
| Matched OR (95% CI)a||1.0||1.34 (0.67–2.68)||0.63 (0.29–1.34)||0.89 (0.45–1.78)||0.35|
|Insulin|| || || || || |
| Cases with PSA < 4 ng/mL|| || || || || |
| No. of cases/controls||11/12||11/11||9/7||8/8|| |
| Matched OR (95% CI)a||1.0||1.32 (0.24–7.18)||1.49 (0.25–8.73)||1.25 (0.23–6.70)||0.81|
| Cases with PSA ≥ 4 ng/mL|| || || || || |
| No. of cases/controls||32/35||38/31||31/33||32/35|| |
| Matched OR (95% CI)a||1.0||1.32 (0.71–2.47)||1.01 (0.51–2.01)||0.94 (0.48–1.85)||0.78|
The current results do not support the hypothesis that circulating IGF-I levels are associated positively with the risk of prostate carcinoma. Men who went on to be diagnosed with prostate carcinoma did not have higher plasma levels of IGF-I compared with their controls, regardless of their extent of disease at the time of diagnosis and whether IGF-I was considered on its own or was adjusted for plasma levels of IGFBP-3. There was a suggestion of a decline in the risk of prostate carcinoma with increasing levels of IGFBP-3, particularly for men with more advanced disease, but no association was evident with serum insulin concentrations.
There were a number of limitations to this study. The results were based on a single measurement of IGF-I, IGFBP-3, and insulin concentrations. However, there have been reports of relative stability of IGF-I and IGFBP-3,43 lack of appreciable diurnal or seasonal variation, and good correlation between repeated measurements over time,24, 44 supporting the usefulness of a single measure. Because of the limited size of the study, small-to-moderate associations between growth factor levels and the risk of prostate carcinoma may have been missed, especially in subgroups defined by extent of disease. Furthermore, because the relation between serum and tissue IGF-I level is not understood well,2, 45 results based on circulating levels may or may not reflect what impact these peptide hormones may have locally to influence prostate carcinoma risk. Finally, a substantial fraction of the men who were diagnosed later with prostate carcinoma already may have had this condition in an occult form when they provided a blood specimen, based on their elevated plasma PSA levels at that time. However, in analyses restricted to cases with PSA levels < 4 ng/mL and their matched controls, negative associations with both IGF-I and IGFBP-3 became more pronounced.
There have been a number of comparisons between the IGF-I levels in serum or plasma specimens from controls and the levels in specimens obtained from men after a diagnosis of prostate carcinoma. In some (but not in all) of those comparisons, it was observed that average IGF-I levels in cases were higher.24, 25, 27, 32 However, because circulating IGF-I levels appear to rise during a period of several years prior to a man's diagnosis of prostate carcinoma,29 the etiologic relevance of any association based on postdiagnosis blood samples is unclear. There have now been seven cohort studies of circulating levels of IGF-I and prostate carcinoma that were conducted among participants from whom blood samples were obtained prior to a diagnosis of prostate carcinoma. Unfortunately, the findings from those studies are inconsistent and cannot be interpreted in any straightforward way.
In the first and largest of those nested case–control studies, Chan et al.24, 46 observed a strong, positive association among participants in the Physicians Health Study. In their first report, which was based on plasma samples obtained from 152 cases and 152 controls, men in the upper quadrant of the distribution had 2.4 times the risk of men in the lower quadrant (P value for trend = 0.006). Adjusting for plasma levels of IGFBP-3, with which levels of IGF-I are highly correlated, the corresponding relative risk was 4.3. In a follow-up report, which was based on 530 cases and a similar number of controls, the association with elevated IGF-I levels (adjusted for levels of IGFBP-3) was restricted to patients with relatively advanced prostate carcinoma. Somewhat similar results were obtained in the 149 cases and 298 controls who participated in the Northern Sweden Health and Disease Cohort Study.25 In that study, there was a gradual rise in risk (P value for trend = 0.02) with an increasing plasma IGF-I level: The risk associated with being in the upper quartile of the distribution, relative to the lower quartile, was 1.57. However, adjustment for IGFBP-3 somewhat attenuated the association (adjusted relative risk for men in the upper quartile = 1.32). Furthermore, if anything, the association was weaker for patients who had more advanced disease compared with patients who had less advanced disease. Another suggestion of an association emerged from the Baltimore Longitudinal Study on Aging27: Based on 72 cases and 127 matched controls, the risk of prostate carcinoma among men in the upper one-third of the serum IGF-I distribution was 1.65 times that among men in the lower one-third of the serum IGF-I distribution.
Authors of the reports from the Physicians and Swedish studies examined the possibility that the positive association with circulating IGF-I levels was attributable to the presence of as-yet-undiagnosed prostate carcinoma giving rise to the elevated levels, rather than the reverse. Those investigators restricted their analyses to men who were diagnosed at least 9 years and 2 years, respectively, after blood samples were drawn, and they obtained results similar to those based on all cases. In addition, in the Physicians Study, a positive association was observed in an analysis that was restricted to men whose IGF-I status was assessed at a time when PSA levels were not elevated.
In opposition to these results are those from nested case–control studies that were conducted in cohorts in Washington County (MD),28 Northern California,26 and Finland.29 The first 2 of those were relatively small studies (n = 30 cases and 45 cases, respectively), and, in the second study, no adjustment for levels of IGFBP-3 could be made. Nonetheless, no hint of a positive association was seen: The risk among men in the upper quartile of the IGF-I distribution, relative to the risk among men in the lowest quartile, was 0.7 and 0.8, respectively. The Finnish study was larger (n = 100 cases), although the analysis was restricted to men in whom at least 5 years elapsed between blood draw and the diagnosis of prostate carcinoma. In that study, the risk in men in the upper and lower quartiles of the distribution of serum IGF-I levels was identical (P value for trend = 0.74). After adjustment for serum levels of IGFBP-3, the risk associated with an IGF-I serum level in the upper quartile of the distribution was 0.52 (P value for trend = 0.16). With our results (particularly those in the subset of cases with aggressive tumors) added to the tally of the other negative results, it is difficult to argue that circulating IGF-I levels have an appreciable impact on the risk of prostate carcinoma.
The interpretation of the data pertaining to the incidence of prostate carcinoma in relation to the serum or plasma levels of IGFBP-3 is equally problematic. In the nested case–control study within the Physicians cohort,46 levels of IGFBP-3 were found to be related inversely to risk, but only to the risk of advanced-stage disease and only when the analysis was adjusted for levels of IGF-I (for advanced disease, the adjusted relative risk associated with being in the fourth [upper] quartile of the distribution = 0.2; P value for trend = 0.002). However, in both the Swedish study and the Finnish study, a modest, positive association was observed between prostate carcinoma and circulating IGFBP-3 levels. Thus, our observation of a negative association between prostate carcinoma risk and levels of IGFBP-3 should not be accorded too much weight.
In a Chinese case–control study, men with prostate carcinoma had relatively higher levels of insulin38 and, as a closely correlated, derived measure, greater insulin resistance47 compared with a control group. However, in neither the current study nor the nested case–control study in Sweden25 was any association observed between prostate carcinoma and circulating insulin levels. The explanation for the divergent results—chance, timing of blood draw relative to time of diagnosis, or effect modification by race–remains unidentified.