Screen-detected prostate cancer and the insulin-like growth factor axis: Results of a population-based case-control study
Article first published online: 20 NOV 2003
Copyright © 2003 Wiley-Liss, Inc.
International Journal of Cancer
Volume 108, Issue 6, pages 887–892, 1 March 2004
How to Cite
Oliver, S. E., Gunnell, D., Donovan, J., Peters, T. J., Persad, R., Gillatt, D., Pearce, A., Neal, D. E., Hamdy, F. C. and Holly, J. (2004), Screen-detected prostate cancer and the insulin-like growth factor axis: Results of a population-based case-control study. Int. J. Cancer, 108: 887–892. doi: 10.1002/ijc.11631
- Issue published online: 2 JAN 2004
- Article first published online: 20 NOV 2003
- Manuscript Accepted: 25 AUG 2003
- Manuscript Revised: 15 AUG 2003
- Manuscript Received: 7 MAY 2003
- NHS South and West R&D
- prostatic neoplasms;
- insulin-like growth factor axis;
- case-control study
Higher circulating levels of IGF-I have been associated with increased risk of prostate and some other cancers. Most research on prostate cancer has been based on men with symptoms or identified following treatment of benign disease. However, increasing numbers of cancer cases are now detected in asymptomatic men following prostate-specific antigen (PSA) tests. We therefore used a population-based case-finding exercise using the PSA test to examine whether associations between the IGF axis and cancer risk were apparent in this population. A matched case-control study was conducted among 7,383 men (50–70 years) receiving a PSA test as part of a case-finding exercise. Assays of IGF-I, IGF-II, IGFBP-2 and IGFBP-3 were performed on cases and 2 controls matched on age, recruitment center and calendar time. Analyses were based on 176 cases and 324 matched controls. The risk of prostate cancer increased across quartiles of IGF-I (highest vs. lowest quartile, OR = 2.34; 95% CI = 1.26–4.34; ptrend = 0.02) and IGF-II (OR = 1.78; 95% CI = 0.94–3.15; ptrend = 0.09). Controlling for smoking history and IGFBP-3 strengthened associations with cancer for both IGF-I (OR = 3.00; 95% CI = 1.50–6.01; ptrend 0.005) and IGF-II (OR = 2.02; 95% CI = 1.07–3.84; ptrend = 0.04) Associations between the IGFs and cancer risk were stronger for advanced cases. Our findings suggest that both IGF-I and IGF-II are associated with an increased risk of screen-detected prostate cancer. © 2003 Wiley-Liss, Inc.
Prostate cancer is the second most common life-threatening cancer in men in the majority of Western countries1 and accounted for over 31,000 deaths in the United States in 2000.2 Despite its major impact on public health, the etiology of prostate cancer is not well understood, with the only recognized risk factors being age, family history and ethnicity. Several studies now suggest that elements of the IGF axis may have an important role in modifying the risk of prostate,3 breast4, 5 and colorectal cancers.6, 7
The insulin-like growth factors (IGF-I and -II) are peptide growth factors that share structural homology with insulin. Unlike insulin, the IGFs circulate at much higher concentrations due to their association with specific high-affinity binding proteins (IGFBP-1 to IGFBP-6). The circulating concentration of IGF-I is regulated by growth hormone (GH) and IGF-I mediates some of the effects of GH on childhood somatic growth.8 The IGFs promote cell proliferation and inhibit apoptosis in many tissues, including stroma and epithelial cells in the prostate.9 The majority of circulating IGFs are bound to IGFBP-3, which has an important role in modulating their bioavailability and has an independent proapoptotic capacity.9 Men with higher circulating levels of IGF-I have been observed to have a greater risk of prostate cancer in a number of prospective3, 10, 11, 12 and case-control studies.13, 14, 15
Much of the observational epidemiologic research exploring relationships between the IGF axis and prostate cancer has been carried out on symptomatic cases or men identified following surgical treatment for benign prostatic hyperplasia. However, the introduction and widespread uptake of the prostate-specific antigen (PSA) test in some countries have led to a dramatic increase in the proportion of asymptomatic men found to have prostate cancer.16 To date, no studies have investigated the association of screen-detected prostate cancer with the IGF axis. Such studies are crucial as it is possible that previously observed associations with IGF-I reflect its influence on the likelihood of cancer detection rather than it being of etiologic importance. We therefore used a population-based case-finding process using PSA tests to examine whether associations between the IGF axis were present among screen-detected cases of prostate cancer.
MATERIAL AND METHODS
This is a case-control study nested within recruitment to a randomized controlled trial of treatment for screen-detected localized prostate cancer. Between July 1999 and May 2002, 7,383 men aged 50–70 years registered with any of 18 general practices in 3 centers in England were invited to attend a clinic to consider undergoing a PSA test. Men with a known history of prostate cancer, other malignancy or any condition resulting in a life expectancy of less than 10 years were excluded from the study. The PSA test was offered as part of the case-finding process for a multicenter randomized controlled trial of treatments for localized prostate cancer, the Prostate Testing for Cancer and Treatment (ProtecT) study.17 In addition to providing blood specimens for PSA, consent was sought from study participants for further blood specimens to be taken for research. Participants were also given a health and lifestyle questionnaire to complete after the clinic and before their PSA results became available.
For the first 12 months of the PROTECT study, age-specific thresholds were used to interpret results of the PSA test. Men aged 50–59 years with a PSA ≥ 3 ng/ml and men aged over 60 years with a PSA ≥ 4 ng/ml were asked to attend a clinic for further investigation. This was subsequently changed to a single threshold of PSA ≥ 3 ng/ml, regardless of age. Men with a PSA level above the threshold (n = 861; 12%) were invited to a clinic where digital rectal examination, transrectal ultrasound, a repeat PSA and prostatic biopsy were performed. This clinic visit occurred 6 weeks on average after the initial PSA test. Men with a normal biopsy were offered repeat biopsy if the PSA was persistently raised or if there was a high clinical index of suspicion (chiefly evidence of high-grade prostatic interepithelial neoplasia or other suspicious features on initial biopsy). Histologic material obtained at biopsy was reviewed by local pathologists, and all cases of prostate cancer were given a histologic grade using the Gleason score.18 Clinical staging was recorded for all men with cancer using the TNM staging system.19 Men with a clinical stage of ≤ T3 were considered to have localized cancer. The histologic grade of cancer observed at biopsy was classified by pathologists in each of the 3 study centers using the Gleason score; tumors with a Gleason score ≥ 7 were considered high grade.
By April 2002, 224 cases of prostate cancer had been identified (3% of population screened), of whom 176 (79%) had provided both a serum sample and full data on covariates to be included in the analyses. All participants in the ProtecT study who had no evidence of prostate cancer and had provided serum samples were eligible for selection as controls. This included all men with a PSA below the investigation threshold and any men with a PSA above the threshold and one or more sets of negative prostatic biopsies. Cases were individually matched to 2 controls by age (2-year bands), the general practice from which they were recruited and the calendar date of recruitment (closest date). We could only identify a single control for 28 (16%) cases, and so the final control group was made up of 324 men.
All subjects gave voluntary written informed consent for use of their questionnaire data and blood samples for research purposes. The study was reviewed and approved by the multicenter and relevant local research ethics committee.
Measurement of PSA and IGF axis
Blood was taken into plain bottles and kept at 5°C until spun to extract serum. Serum specimens were divided into aliquots and stored at −80°C, generally within 8 hr of venepuncture. The total PSA test used for case-finding in the ProtecT study was performed in 1 of 3 laboratories in the collaborating research centers. Serum specimens were analyzed using either the DPC Immulite 2000 third-generation assay (2 centers) or the Bayer/ACS centaur assay (1 center). All centers participated in a national quality control scheme (United Kingdom National External Quality Assessment Schemes for Autoimmune Serology and Special Immunochemistry, or UKNEQAS).
The concentrations of IGF-I, IGF-II, IGFBP-3 and IGFBP-2 in serum samples were measured by J.H.'s laboratory. Double-antibody enzyme-linked immunosorbant (ELISA) assays were used to measure IGF-I (DSL-10-2800 Active; Diagnostic Systems Laboratories, Webster, TX) and IGF-II (DSL-10-2600; Diagnostic Systems Laboratories). Total levels of IGFBP-2 were measured by radioimmunoassay (DSL-7100; Diagnostic Systems Laboratories). Assays for serum IGFBP-3 were carried out using a previously validated in-house double antibody radioimmunoassay.20 The molar ratio of IGF-I:IGFBP-3, which has been proposed to reflect tissue bioavailability, was calculated as (0.13 × IGF-I concentration in ng/ml)/(0.025 × IGFBP-3 concentration ng/ml).
Results for IGF-I concentration were based on the arithmetic mean of 2 measures and for IGFBP-3 on 3 measures; single measures were made of IGF-II and IGFBP-2. All assays were performed blind to status as either case or control. Serum samples from paired cases and controls were assayed together using the same assay kits wherever possible. The average coefficients of variation for intraassay variability for IGF-I, IGF-II, IGFBP-3 and IGFBP-2 were 3%, 5%, 4% and 5% and for interassay variation were 15%, 26%, 14% and 14%.
Conditional logistic regression was used to compare baseline characteristics between matched case and control subjects. Measures of the IGF axis, PSA and BMI were positively skewed, measures were log-transformed before analysis and geometric means were reported. Conditional logistic regression was also used to estimate the odds ratios (ORs) and 95% confidence intervals (CIs) for the association of IGFs, IGFBPs and the IGF-I:IGFBP-3 ratio with the risk of prostate cancer. For the whole case group, odds ratios were compared across quartiles of IGFs and IGFBPs using the lowest quartile as the reference group (categories were derived from the distribution in control subjects). Tests for linear trend across quartiles were also performed. Models presented in the article were adjusted for cigarette smoking (categorized as nonsmoker, ex-smoker, current smoker). Inclusion of measures of BMI, social class, physical exercise and alcohol intake did not alter the strength of associations seen between IGFs or IGFBPs and cancer risk in these participants and these models are therefore not reported. In addition, models including IGF-I and IGF-II were adjusted for IGFBP-3. It has been suggested that IGFBP-2 may act as an additional IGF carrier when IGFBP-3 levels are inadequate,21 and therefore models including the IGF-I:IGFBP-3 ratio were further adjusted for IGFBP-2.
Associations with cancer risk of IGFs and IGFBPs were assessed separately by stage (localized/advanced) and histologic grade (Gleason score < 7/Gleason score ≥ 7); in these models, IGFs and IGFBPs were categorized in tertiles to ensure sufficient cases in all groups. We also tested for any evidence of an interaction between age (< 60 years, ≥ 60 years) and the associations between both IGFs and IGFBPs and cancer risk, as the relationship between IGF-I and risk of prostate cancer had previously been observed to be stronger in older men.3 All hypothesis tests were 2-tailed and exact p-values are reported. Test probabilities were obtained from likelihood ratio tests; all analyses were performed using the Stata (version 7) computer package (StataCorp, College Station, TX).
Of the 176 cases (mean age, 62 years; interquartile range (IQR), 59–66) with complete data on serum measures and other covariates, 130 (74%) were localized to the prostate gland (≤ T2) and 124 (70%) were low grade (Gleason score ≤ 6). Cases included in the analysis did not differ significantly from those for whom serum was unavailable (n = 48) in terms of age (mean age, 62.2 vs. 62.9 years; p = 0.41), clinical stage (26% vs. 27% advanced; p = 0.90), histologic grade (30% vs. 41.7% high grade; p = 0.10), or PSA level (geometric mean, 7.1 vs. 9.0 ng/ml; p = 0.10).
Table I shows the comparison of measures of the IGF axis in matched cases and controls. Geometric mean levels of IGF-I were higher among cases (130.7 ng/ml; CI = 125.8–135.9) than controls (121.2 ng/ml; CI = 117.4–125.2; p = 0.003); average circulating levels of IGFBP-3, IGF-I:IGFBP-3 molar ratio and IGF-II were also higher among cases. Levels of IGFBP-2 did not differ between cases and controls. As a consequence of the process of case detection, the serum PSA was significantly higher in cases than controls (7.1 vs. 1.2 ng/ml; p < 0.0001). No difference was seen between cases with regard to BMI, social class or levels of physical exercise. Almost all cases and controls reported their ethnicity as “white” (99%). Cases were more likely to report a history of lifetime nonsmoking (49% vs. 37%; p = 0.006) and were less likely to be heavy alcohol consumers (≥ 22 units per week; 19% vs. 26%; p = 0.03).
|Control subjects (n = 324)||Case subjects (n = 176)||p for difference|
|Mean (95% CI)|
|Age||62.2 (61.7–62.6)||62.2 (61.3–62.8)||0.80|
|Geometric mean (95% CI)|
|IGF-I (ng/ml)||121.2 (117.4–125.2)||130.7 (125.8–135.9)||0.003|
|IGFBP-3 (ng/ml)||3220.1 (3102.6–3342.0)||3311.0 (3161.3–3467.7)||0.30|
|IGF-I/IGFBP-3 molar ratio (%)||19.6 (18.9–20.3)||20.5 (19.6–21.6)||0.06|
|IGF-II (ng/ml)||424.1 (411.5–437.1)||437.1 (420.4–454.4)||0.07|
|IGFBP-2 (ng/ml)||506.5 (472.5–543.0)||513.5 (472.3–558.3)||0.74|
|PSA (ng/ml)||1.2 (1.1–1.3)||7.1 (6.3–8.1)||<0.0001|
|BMI (kg/m3)1||26.8 (26.4–27.3)||27.2 (26.7–27.7)||0.26|
|Proportion (number with missing data)|
|Never smoker (%)||37 (0)||49 (0)||0.006|
|Self-reported ethnicity “white” (%)||99 (4)||99 (13)||0.67|
|Nonmanual social class (%)||64 (6)||63 (15)||0.97|
|Alcohol consumption high2 (%)||26 (7)||19 (4)||0.03|
|Regular strenuous exercise3 (%)||48 (3)||52 (6)||0.24|
When all cases (localized and advanced) were included in conditional logistic regression analyses (Table II), a strong association was seen between higher levels of IGF-I and greater cancer risk (highest vs. lowest quartile, OR = 2.34; CI = 1.26–4.34; ptrend = 0.02). This association was strengthened following adjustment for smoking history and IGFBP-3 (highest vs. lowest quartile, OR = 3.00; CI = 1.50–6.01; ptrend = 0.005). No consistent linear trend was observed between IGFBP-3 and cancer risk in the unadjusted analysis; after controlling for smoking history and IGF-I, a weak (p = 0.30) inverse association was seen. Cancer risk was greater in men with a higher IGF-I:IGFBP-3 molar ratio, but the strength of association was less than that seen for IGF-I alone in both unadjusted models (highest vs. lowest quartile, OR = 1.75; CI = 0.94–3.15; ptrend = 0.06) and following control for smoking history. Further adjustment for IGFBP-2 did not alter the strength of this relationship. There was no evidence of a statistically significant interaction between age and associations between cancer risk and any measures of the IGF axis.
|Quartile range||Cases/controls||Unadjusted OR (95% CI)||Fully adjusted1 OR (95% CI)|
|Quartile 1||< 99.5||25/81||1.00 (ref)||1.00 (ref)|
|Quartile 2||99.5–123.9||50/81||2.27 (1.22–4.25)||2.62 (1.36–5.06)|
|Quartile 3||124.0–149.5||49/81||2.18 (1.20–3.97)||2.50 (1.32–4.71)|
|Quartile 4||> 149.5||52/81||2.34 (1.26–4.34)||3.00 (1.50–6.01)|
|Quartile 1||< 2,772.2||47/82||1.00 (ref)||1.00 (ref)|
|Quartile 2||2,772.2–3,328.7||37/81||0.80 (0.46–1.37)||0.62 (0.35–1.12)|
|Quartile 3||3,328.8–4,104.9||55/81||1.16 (0.69–1.93)||0.93 (0.52–1.67)|
|Quartile 4||> 4,104.9||37/80||0.75 (0.41–1.37)||0.58 (0.29–1.15)|
|Quartile 1||< 16.0||41/81||1.00 (ref)||1.00 (ref)|
|Quartile 2||16.0–19.4||37/81||0.99 (0.58–1.72)||0.95 (0.55–1.66)|
|Quartile 3||19.5–23.7||40/81||1.13 (0.63–2.04)||1.11 (0.61–2.02)|
|Quartile 4||> 23.7||58/81||1.75 (0.94–3.15)||1.62 (0.87–2.99)|
|Quartile 1||< 358.6||35/81||1.00 (ref)||1.00 (ref)|
|Quartile 2||358.6–424.5||46/81||1.45 (0.82–2.55)||1.63 (0.91–2.95)|
|Quartile 3||424.6–506.3||45/81||1.49 (0.82–2.73)||1.73 (0.93–3.22)|
|Quartile 4||> 506.3||50/81||1.78 (0.95–3.33)||2.02 (1.07–3.84)|
|Quartile 1||< 355.8||43/82||1.00 (ref)||1.00 (ref)|
|Quartile 2||355.8–533.2||50/81||1.23 (0.73–2.06)||1.19 (0.70–2.01)|
|Quartile 3||533.3–743.2||38/80||0.97 (0.57–1.66)||0.93 (0.54–1.61)|
|Quartile 4||> 743.2||45/81||1.04 (0.61–1.79)||0.95 (0.55–1.67)|
Risk of prostate cancer increased with higher serum levels of IGF-II (highest vs. lowest quartile, OR = 1.78; CI = 0.94–3.15; ptrend = 0.09), and adjustment for smoking and IGFBP-3 accentuated this relationship (highest vs. lowest quartile, OR = 2.02; CI = 1.07–3.84; ptrend = 0.04). No association was seen between serum levels of IGFBP-2 and risk of prostate cancer.
The results of analyses conducted separately among localized (n = 130) and advanced (n = 46) cases are shown in Table III. The pattern of associations seen between cancer risk and measures of the IGF axis were broadly similar in both groups. The strength of association seen between both IGF-I and IGF-II and the risk of prostate cancer was greater in advanced as compared to localized cases. Of the localized cases, 109 (84%) were classified as having a low histologic grade, compared to 15 (33%) of men with advanced disease. Table IV shows the outcome of separate analysis of high- (n = 124) and low-grade (n = 52) cases. The positive relationships between IGF-I and IGF-II and cancer risk were again seen in both groups, stronger for high-grade tumors.
|Tertile range||Localized cases||Advanced cases|
|Cases/controls||Unadjusted OR (95% CI)||Fully adjusted OR (95% CI)||Cases/controls||Unadjusted OR (95% CI)||Fully adjusted1 OR (95% CI)|
|Tertile 1||< 110.2||33/77||1.00 (ref)||1.00 (ref)||9/31||1.00 (ref)||1.00 (ref)|
|Tertile 2||110.2–140.1||45/80||1.47 (0.81–2.68)||1.54 (0.82–2.92)||22/28||2.75 (1.05–7.24)||3.25 (1.15–9.18)|
|Tertile 3||> 140.1||52/81||1.65 (0.93–2.93)||1.77 (0.93–3.34)||15/27||1.88 (0.69–5.14)||2.09 (0.65–6.76)|
|Tertile 1||< 2,952.7||42/82||1.00 (ref)||1.00 (ref)||16/27||1.00 (ref)||1.00 (ref)|
|Tertile 2||2,952.7–3,800.8||49/81||1.15 (0.68–1.92)||1.14 (0.66–1.98)||16/27||0.95 (0.37–2.44)||0.93 (0.30–2.89)|
|Tertile 3||> 3,800.8||39/75||1.00 (0.55–1.83)||0.96 (0.49–1.90)||14/32||0.66 (0.24–1.78)||0.48 (0.15–1.55)|
|Tertile 1||< 382.6||37/79||1.00 (ref)||1.00 (ref)||1.00 (ref)||1.00 (ref)|
|Tertile 2||382.6–472.1||42/81||1.21 (0.67–2.18)||1.36 (0.72–2.55)||11/29||1.93 (0.68–5.47)||2.35 (0.76–7.21)|
|Tertile 3||> 472.1||51/78||1.73 (0.91–3.26)||2.16 (1.08–4.33)||16/27||2.26 (0.74–6.91)||2.77 (0.77–9.87)|
|Tertile range||Low grade cases||High grade cases|
|Cases/Controls||Unadjusted OR (95% CI)||Fully adjusted* OR (95% CI)||Cases/Controls||Unadjusted OR (95% CI)||Fully adjusted1 OR (95% CI)|
|Tertile 1||< 110.2||33/77||1.00 (ref)||1.00 (ref)||9/31||1.00 (ref)||1.00 (ref)|
|Tertile 2||110.2–140.1||42/77||1.40 (0.78–2.55)||1.43 (0.76–2.70)||25/31||3.23 (1.16–9.04)||3.80 (1.28–11.31)|
|Tertile 3||> 140.1||49/73||1.66 (0.94–2.93)||1.88 (1.00–3.54)||18/35||1.93 (0.69–5.39)||2.38 (0.75–7.57)|
|Tertile 1||< 2952.7||35/70||1.00 (ref)||1.00 (ref)||23/39||1.00 (ref)||1.00 (ref)|
|Tertile 2||2952.7–3800.8||50/79||1.22 (0.72–2.09)||1.22 (0.68–2.16)||15/29||0.81 (0.34–1.93)||0.74 (0.28–1.94)|
|Tertile 3||> 3800.8||39/78||0.97 (0.53–1.78)||0.88 (0.45–1.75)||14/29||0.74 (0.28–1.96)||0.50 (0.16–1.58)|
|Tertile 1||< 382.6||30/64||1.00 (ref)||1.00 (ref)||18/44||1.00 (ref)||1.00 (ref)|
|Tertile 2||382.6–472.1||41/82||1.21 (0.65–2.27)||1.46 (0.75–2.83)||17/26||1.72 (0.70–4.24)||2.19 (0.83–5.75)|
|Tertile 3||> 472.1||53/81||1.69 (0.88–3.24)||2.16 (1.06–4.38)||17/27||2.10 (0.71–6.19)||2.96 (0.90–9.78)|
This is the first epidemiologic study to examine specifically associations between the IGF axis and the risk of prostate cancer detected through PSA-based screening. We have confirmed that the strong positive association between circulating IGF-I levels and cancer risk seen for clinically detected cancer is also apparent for screen-detected cases. In addition, we observed that higher serum levels of IGF-II were also associated with an elevated risk of cancer. Adjustment for circulating IGFBP-3 strengthened the association between the IGFs and prostate cancer risk.
Our findings in relation to IGF-I are in keeping with a number of other case-control13, 14, 15 and prospective studies3, 10, 11, 12 that have observed an increased risk of cancer in men with higher levels of IGF-I. The magnitude of the effect we observed before adjustment for IGFBP-3 (highest quartile vs. lowest quartile, OR = 2.34; 95% CI = 1.26–4.34) is also similar to that seen in the initial analysis of the U.S. Physicians Health Study (highest vs. lowest quartile, OR = 2.41; 95% CI = 1.23–4.74), and in prospective studies in Sweden (highest vs. lowest quartile, OR = 1.57; 95% CI = 0.88–2.81) and Baltimore, Maryland (highest vs. lowest tertile, OR = 1.65; 95% CI = 0.71–3.86). Positive associations have also been observed between higher levels of IGF-I and a number of other epithelial cancers, in particular breast4, 5 and colorectal cancer.6, 7
The possibility that associations between IGF-I and prostate cancer risk might differ by disease stage has been examined in 5 studies3, 10, 12, 14, 15 with mixed results. In both the case-control study by Wolk et al.,14 and the prospective study by Stattin et al.,10 there was no evidence of any difference in risk by stage. In a case-control study conducted in a Chinese population in Shanghai, Chokkalingam et al.15 reported a stronger relationship between IGF-I and the risk of localized rather than advanced cancer. In the first published report on prostate cancer and the IGF axis within the Physicians' Health Study, Chan et al.3 categorized cases as high grade/stage or low grade/stage and found that associations between IGF-I and cancer risk were essentially the same in both groups. In an updated analysis, including cases diagnosed up to 1995, Chan et al.12 observed a positive association between IGF-I only for advanced-stage cancer (Whitmore-Jewett classification C or D, equivalent to TNM stage ≥ T3). Associations between the IGF axis and the histologic grade of prostate cancer at diagnosis have only been examined within the Physicians' Health Study.12 In these data, associations of IGF-I and IGFBP-3 and cancer risk were broadly similar for high- (Gleason score ≥ 7) and low-grade (Gleason score < 7) tumors.
The observation that adjustment for IGFBP-3 strengthened the relationship between IGF-I and cancer risk has been seen in most previous studies,3, 11, 14, 15 but not all.10, 22 Only 3 epidemiologic studies have previously reported on the relationship between IGF-II and prostate cancer,3, 11, 15 and results have been inconsistent. In the Physicians' Health Study, there was no evidence of any association between IGF-II and cancer risk,3 while Harman et al.11 observed a reduction in risk of cancer at higher levels of IGF-II in a prospective study nested in the Baltimore Longitudinal Study of Aging (highest tertile vs. lowest tertile, OR = 0.24; 95% CI = 0.10–0.59). In the one cross-sectional case-control study by Chokkalingam et al.15 that measured IGF-II, the risk of cancer was greater in men with higher levels. This association was strongest in men with advanced cancer and following adjustment for IGFBP-3, sex hormone binding globulin (SHBG) and androstanediol glucaronide (3α-diol G; highest quartile vs. lowest quartile, OR = 2.56; 95% CI = 0.78–8.42).
There is some evidence that higher circulating levels of IGF-II may be associated with increased risk of colorectal neoplasia. A small case-control study from Greece reported that higher levels of IGF-II were associated with increased risk of colorectal cancer,23 and one prospective study from China reported a more than 2-fold increase in risk of colorectal cancer in men with the highest levels of IGF-II.24 Interestingly, in the latter study, the association between IGF-II and colorectal cancer was strongest in men with the shortest follow-up, suggesting that elevation of circulating IGF-II might be a disease effect. This hypothesis is consistent with the observation that, in a case-control study nested within a sigmoidoscopy screening trial, while higher IGF-II (but not IGF-I) levels were associated with increased risk of colorectal adenomas, these levels fell significantly after removal of the adenoma.25
While higher levels of IGFBP-2 were seen among prostate cancer cases in some early comparative case series,26, 27 the lack of any association with cancer risk seen in our study is consistent with the one prospective population-based study to include measures of this binding protein.10
There was some evidence in our data that the association between both IGF-I and IGF-II and cancer risk was greater for more advanced disease. While this is consistent in part with recent results from the Physicians' Health Study,12 there are also some differences between the findings. In particular, the positive association between higher levels of IGF-I and the presence of screen-detected localized cancer persisted in our study, while in the analyses by Chan et al.,12 no association was seen for localized cancer when cases were limited to those thought more likely to have been PSA-detected (those diagnosed after 1990). One possible explanation for this apparent divergence in results may relate to the different levels of information available on staging within the 2 studies. In our study, all staging was clinical and it is recognized that a proportion of clinically localized cancers will subsequently be revealed at the time of surgery to have been advanced. Information on staging in the study by Chan et al.12 was abstracted from medical records and may be based on pathologic data as well as clinical examination; it is possible, therefore, that more cases of advanced cancer have been misclassified as localized in our study. Chan et al.12 also reported a weaker association between IGF-I and the risk of high-grade tumors, while in our study the opposite was true.
The strengthening of effect estimates between IGF-I and prostate cancer that occurred after adjustment for smoking history indicates a positive association between smoking history and IGF-I level. Such a relationship has some support within the published literature,28, 29 although other studies have reported either null30 or inverse associations.31 Other studies have included adjustment for smoking in multivariable models, but it is unclear exactly what effect smoking has in isolation from other covariates.3, 10
Some limitations of the study should be considered. Although the nested study design limits the impact of selection bias within the case-control study, it should be recognized that all analyses were cross-sectional. Disease status and measures of the IGF axis were ascertained at the same time, leaving open the possibility that serum measures might be influenced by the presence of cancer. It seems unlikely that the findings for IGF-I and cancer risk can be explained in this way. Serum levels of IGF-I are dominated by hepatic production; the contribution made by production in other tissues is negligible by comparison. In addition, similar associations between IGF-I and cancer risk have been seen in prospective studies with specimens taken over 5 years before diagnosis.3 A factor of more importance than tumor-derived IGF-I might be the effect of the host-response to cancer on systemic levels of the IGFs. As cancer develops, systemic illness usually results and frequently progresses to cancer cachexia. This decline in health is accompanied by a progressive decrease in circulating IGF-I levels.32 In the presence of symptomatic cancer, this cancer-related decrease in IGF-I could weaken any positive association between IGF-I and cancer risk, but such a bias is unlikely among asymptomatic screen-detected cases of prostate cancer. With regard to IGF-II, it is of interest that increased risk of prostate cancer has only been observed in cross-sectional15 rather than prospective studies,3 and that there is some evidence from studies of colorectal neoplasia that IGF-II levels may be raised by tumors.25
Although controls in the study had either a low PSA (≤ 3 ng/ml) or had undergone at least one negative prostatic biopsy, the possibility remains that this group may include some men with occult malignancy. Such a misclassification would tend to bias associations toward the null. Grading of prostate tumors can vary between pathologists, and these analyses were conducted in advance of the final pathologic review within the ProtecT study. As with the clinical staging, it is possible that the histologic grade is an underestimate.
Our study adds further evidence of the role of the IGF axis as an important mediator of prostate cancer risk, with higher levels of both IGF-I and IGF-II strongly linked to cancer risk. While these associations were slightly enhanced for advanced cancer, our results suggest that the IGF axis may also be associated with localized tumors. Given the accumulating strength of research evidence pointing to a key role for IGF-I in prostate cancer, 2 important questions are now raised. First, whether the IGF axis influences cancer progression, and second, whether interventions to modify levels of IGF-I as a risk reduction strategy are merited. Overdiagnosis and overtreatment are major impediments to prostate cancer screening based on the PSA test, with levels of overdiagnosis of 50–80% being reported.33 If measures of the IGF axis could identify prostate tumors that will progress rapidly, aggressive treatments could be restricted to those most likely to benefit. Pharmaceutical and dietary interventions targeted at the IGF axis are currently under investigation.34 Their overall risks and benefits will need to be carefully evaluated as the IGF axis has a wide range of physiologic effects, including a role in maintaining bone density.
The authors thank Sara Bright, Zoe Wilkins, Tracey Calthorpe and Andrea Wilson for providing clerical support; Mark Sidaway and Daniel Dedman for database management; as well as research nurses Peter Holding, Teresa Mewes, Sally Burton, Liz Salter, Louise Goodwin, Ingrid Emmerson, Miranda Benney, Sue Kilner, Lyn Wilkinson, Clare Kennedy, Christine Hardy and Andrew Robinson.
- 1Cancer in five continents. vol. 7. Lyon: IARC, 1997., , , , .
- 2Deaths: final data for 2000—national vital statistics reports. Hyattsville, MD: National Center for Health Statistics, 2002., , , , .
- 20The effects of recombinant human IGF-I administration on concentrations of acid labile subunit, IGF binding protein-3, IGF-I, IGF-II and proteolysis of IGF binding protein-3 in adolescents with insulin-dependent diabetes mellitus. J Endocrinol 1998; 157: 81–7., , , et al.
- 31Serum insulin-like growth factor I in a random population sample of men and women: relation to age, sex, smoking habits, coffee consumption and physical activity, blood pressure and concentrations of plasma lipids, fibrinogen, parathyroid hormone and osteocalcin. Clin Endocrinol 1994; 41: 351–7., , , et al.