Circulating Insulin-Like Growth Factor-1 and Risk of Total and 19 Site-Specific Cancers: Cohort Study Analyses from the UK Biobank

The insulin-like growth factor (IGF) axis plays a crucial role in human growth and development (1). IGF-1 levels signal important early life exposures and are associated with childhood adiposity and height velocity. Perhaps due to its influence on growth, mitosis, and apoptosis, higher circulating IGF-1 has been implicated in multiple malignancies in observational epidemiologic studies, including for breast (2–4), prostate (5, 6), colorectal (7), lung (8), ovarian (9), and central nervous system cancers (glioma; ref. 10). In addition, recent epidemiologic studies suggest that metformin usage may reduce breast cancer incidence and improve overall survival, which may be mediated by metformin's inhibition of the IGF-1 pathway (11, 12). Similarly, a phase II randomized controlled trial (RCT) demonstrated improved survival in patients with lung cancer who received metformin in addition to standard therapy (13). These findings raise interesting questions for the IGF-1 pathway as a biomarker for incident cancer and as a potential target for cancer prevention or therapeutics. However, most existing observational studies were conducted around the 2000s, were relatively limited by sample size or study design (retrospective case–control studies), were inconsistent in their findings, did not comprehensively assess the association of IGF-1 with cancers at uncommon cancer sites, nor examined demographic or clinical characteristics that may modify the associations between IGF-1 and individual cancers.

In turn, we sought to conduct a comprehensive analysis of circulating IGF-1 and incident total and site-specific cancers in the UK Biobank cohort, which is comprised of approximately 500,000 participants. We hypothesized that IGF-1 is positively associated with total and most site-specific cancers.

The UK Biobank cohort recruited 502,656 adult participants, ages 40 to 69, between 2007 and 2010 (14), from approximately 9.2 million people who were invited to participate (response rate: 5.7%). All participants were registered with the UK National Health Service (NHS) and lived within approximately 25 miles of one of the 22 study centers. At the baseline examination, participants completed questionnaires regarding their demographic, lifestyle, disease histories, and medication use and provided biospecimens. Blood samples were collected using standard vacutainers, transported overnight to a central laboratory, where the blood was centrifuged and aliquots of specific fractions were stored at −80°C, prior to biomarker measurements and genotyping (15).

We excluded participants if they reported a history of the cancer being assessed (for the analysis on total cancers, participants with prior diagnosis of any cancer was excluded), prevalent type 2 diabetes or unknown diabetes status at recruitment (due to the potential for hypoglycemic agents to affect IGF-1 levels; N = 25,345; ref. 16), current oral contraceptive or menopausal hormone use at the baseline exam [oral estrogens can affect circulating levels of multiple hormones, including IGF system markers (16), through a first pass effect on the liver; N = 24,448], missing information on genetic ancestry (N = 4,249), discordant self-reported and genetic sex (N = 355), and participants without a baseline IGF-1 measurement (N = 35,464). In addition, 21 participants withdrew consent prior to the submission of this manuscript and were removed from the analysis. Hence, our final analytical sample consisted of 412,645 participants, prior to the exclusion of participants with prevalent cancer.

Circulating IGF-1 were measured in the serum of stored blood samples from 502,527 participants using a chemiluminescent immunoassay with the Liaison XL platform (DiaSorin Ltd), as part of the UK Biobank Biomarker Project (17). Quality control procedures and results were published previously (17). Briefly, average within-laboratory (total) coefficient of variation (CV) for low, medium, and high internal quality control level samples for each biomarker ranged from 1.7% to 15.3% (for IGF-1, the CVs ranged from 5.3% to 6.2%; ref. 17). IGF-1 was measured at both the initial and a follow-up visit (median ∼4 years apart, using blood samples collected in 2012–2013) in 16,357 participants. These participants constitute a subset of the original cohort, who live in the area surrounding the UK Biobank's coordinating center in Stockport, and voluntarily chose to undergo a repeat physical examination, update their answers to the original questionnaires, and provide biospecimens.

Testosterone and sex hormone binding globulin (SHBG) were determined by a chemiluminescent immunoassay using the Beckman Coulter DXI 800. High-sensitivity C-reactive protein (hsCRP) levels were measured with the immunoturbidimetric method (Beckman Coulter DXI 800). Glycated hemoglobin (HbA1c) levels were determined using the HPLC Variant II Turbo 2.0 system (Bio-Rad). Alanine aminotransferase (ALT), aspartate aminotransferase (AST), gamma-glutamyltransferase (GGT) were assayed using enzymatic rate measurements (Beckman Coulter AU5800). Total bilirubin was measured with colorimetric analysis (Beckman Coulter AU5800).

Linkages to NHS medical records with the International Classification of Diseases (ICD-10) diagnosis codes were used to ascertain a cancer diagnosis (18). Since the last data update, cancer diagnoses follow-up is available through March 31, 2016 for participants in England and Wales, and through October 31, 2015 for participants in Scotland. We assessed total cancer (excluding nonmelanoma skin cancer) and the 19 site-specific cancers with the highest incidence rates in the UK Biobank (≥200 incident cases), which included female breast cancer, prostate cancer, nonmelanoma skin cancer, melanoma, colorectal cancer, hematopoietic cancers, lung cancer, uterine cancer, ovarian cancer, urothelial cancer, kidney cancer, pancreatic cancer, esophageal cancer, gastric cancer, liver cancer, central nervous system cancer, mesothelioma, thyroid cancer, and head and neck cancer (HNC). Detailed ICD-10 codes for each cancer site can be found in Supplementary Table S1. The study focused on incidence cases that were developed after the initial exam date (baseline). Prevalent cases, defined as cancer diagnosed before the initial exam date, were excluded.

Reproducibility of IGF-1 over time was evaluated by calculating the intraclass correlation coefficient (ICC) among participants with IGF-1 measured at two time points. Participants were grouped into sex-specific quintiles based on their baseline IGF-1 measurements. Test for linear-trend was calculated by assigning the quintile-specific median IGF-1 value and evaluating it as a continuous variable. IGF-1 was also assessed continuously in per-5 nmol/L increments. To avoid potential biases from outlier measurements, we truncated IGF-1 measurements at 50 nmol/L. Furthermore, to avoid possible underestimation of the true association between IGF-1 and cancers due to within-person variability over time or measurement errors, we additionally corrected for regression dilution by multiplying the baseline IGF-1 by the Pearson correlation coefficients between the baseline and follow-up measurements (19, 20). This correction can provide the true strength of the association between IGF-1 and risk of cancers if there were no random fluctuations.

Cox proportional hazards models were used to calculate HR and corresponding 95% confidence interval (CI). Follow-up time was calculated from the baseline exam date until the first cancer diagnosis, death, loss to follow-up, or end of follow-up, whichever came first. We fit three separate models to examine the associations between IGF-1 and overall cancer risk and each of the 19 cancers. Model 1 included age, sex, assessment center, and the first two principal components (as a proxy for ancestry). Model 2 additionally adjusted for alcohol consumption, smoking status, physical activity, height, body mass index (BMI), waist-to-hip ratio, household income, and family history of breast, lung, colorectal, or prostate cancer (where appropriate). For female-specific cancers, we additionally adjusted in Model 2 for menopausal status, age at menarche, parity, age at first live birth, prior breast cancer screening/mammograms, and history of oral contraceptive or hormone replacement therapy use. We considered Model 2 to be the main model. Model 3 further adjusted for hsCRP, HbA1c, SHBG, and testosterone to examine whether they influenced the association between IGF-1 and individual cancer risk. In post hoc analyses, for kidney cancer, we additionally adjusted for creatinine and cystatin-C in addition to the covariates in Model 3. Similarly, for liver cancer, we additionally adjusted for ALT, AST, GGT, and total bilirubin. The proportional hazards assumption was assessed by testing an interaction term with follow-up time as well as visual examination of Schoenfeld residuals, with no evidence of deviation for any cancers. To account for potential competing risks of deaths from other diseases, in a sensitivity analysis, we performed competing risks analysis using the Fine–Gray model (21).

For cancers with large numbers of incident cases (total cancer, breast, prostate, colorectal, and lung), for which there may be biologically plausible interactions with other demographic or clinical characteristics, we conducted exploratory analysis to assess for potential effect modifiers. These included follow-up duration, sex, menopausal status, BMI, alcohol consumption, smoking status, and SHBG and testosterone levels. Heterogeneity between the subgroups was assessed with an interaction term between these factors and IGF-1.

All statistical analyses were performed using SAS 9.4 (SAS Institute) and Stata 16.1 (StataCorp). Statistical tests were two-sided and a P < 0.05 was considered statistically significant unless otherwise noted. The manuscript was prepared in accordance with STROBE (22).

Ethical approval was obtained for the UK Biobank studies from the North West Multi-centre Research Ethics Committee, the National Information Governance Board for Health and Social Care in England and Wales, and the Community Health Index Advisory Group in Scotland. All participants provided written informed consent.

Statistical codes used for the present analysis can be made available upon request by contacting the corresponding author ([email protected]).

The baseline characteristics of 216,562 female and 196,083 male participants are shown in Table 1. In general, participants with higher circulating IGF-1 tended to be younger, taller, have a lower BMI and waist circumference, and less likely to be a smoker or daily alcohol drinker; and among females, tend to be premenopausal, nulliparous, older at first live birth, and to have ever used oral contraceptives. For circulating biomarkers, participants with higher IGF-1 tended to have lower hsCRP and SHBG. Detailed information on number of participants included in the analysis for each cancer site can be found in Supplementary Table S1.

IGF-1 levels measured at two time points (∼4 years apart) in a subset of participants was highly reproducible over time (ICC: 0.75, Pearson correlation coefficient: 0.77). Regression dilution was corrected by multiplying the measured IGF-1 values by 0.77.

Participants were followed for a median of 7.2 years. Higher IGF-1 was associated with increased total cancer risk in both men [HR (95% CI) comparing Q5 vs. Q1: 1.08 (1.02–1.15)] and women [1.08 (1.02–1.15)]. For site-specific cancers, higher IGF-1 was associated with an increased risk of cancer of the breast (HR Q5 vs. Q1: 1.24; 95% CI, 1.12–1.36), prostate 1.20 (1.10–1.30), colorectum 1.21 (1.08–1.36), kidney 1.29 (1.01–1.64), and thyroid 1.50 (0.97–2.32; Table 2). Conversely, higher circulating IGF-1 was associated with lower risks of lung 0.84 (0.72–0.97), ovarian 0.64 (0.47–0.87), and liver cancers 0.21 (0.13–0.36). The inverse association for lung cancer was only observed among ever-smokers (1,695 cases): 0.79 (0.67, 0.93), whereas among never-smokers (301 cases), the association trended toward increased risk 1.22 (0.84, 1.76). Further adjustment for hsCRP, HbA1c, SHBG, and testosterone did not materially alter these associations (Model 3 in Supplementary Tables S2 and S3). We did not find appreciable associations between IGF-1 and the other cancer sites we evaluated.

When IGF-1 was modeled continuously, we observed similar associations (Fig. 1). For a 5-nmol/L increment in IGF-1 levels that was corrected for regression dilution, we found an increased risk of total cancer [HR (95% CI), men: 1.03 (1.01–1.06); women: 1.03 (1.01–1.06)], breast 1.10 (1.07–1.14), prostate 1.09 (1.05–1.12), colorectal 1.07 (1.02–1.11), kidney 1.10 (1.00–1.20), and thyroid cancers 1.22 (1.05–1.42); and a decreased risk of lung 0.91 (0.86–0.96), ovarian 0.86 (0.77–0.95), and liver cancers 0.32 (0.26–0.38). The positive associations we observed for total cancer in women appeared to be driven by breast cancer whereas the positive association for total cancer in men appeared to be driven by prostate cancer. When we excluded these cancers from the definition of total cancer, the associations were attenuated. For total cancer in women, the HR (95% CI) per 5-nmol/L increment in IGF-1 was 0.98 (0.96–1.01); for total cancer in men, the association was 0.99 (0.97–1.02). We also observed significant associations for HNC: 0.90 (0.82–0.99) and melanoma: 1.08 (1.01–1.15), which were not statistically significant in the quintile analyses. After stringent Bonferroni adjustment (P < 0.05/21), the associations for breast, prostate, colorectal, lung, and liver cancers remained statistically significant. Associations were similar for IGF-1 that was not corrected for regression dilution (Supplementary Table S4). Finally, we performed competing risks analysis using the Fine–Gray model and found nearly identical results to those obtained using Cox models (Supplementary Table S5).

Associations for IGF-1 were largely consistent across subgroups for total, breast, prostate, colorectal, and lung cancers (Table 3; Supplementary Fig. S1). In particular, other circulating hormones such as testosterone or SHBG did not modify the association between IGF-1 and incident breast or prostate cancer (P_int > 0.05 for both cancers). The association between IGF-1 and total cancer in men was stronger among never-smokers [HR (95% CI) per 5-nmol/L increment in IGF-1: 1.08 (1.05–1.12)] compared with ever-smokers [1.00 (0.97–1.03); P_int = 0.0003]. This appeared to be driven by heterogeneity in the association for lung cancer, for which the association was 1.14 (0.99–1.30) among never-smokers and 0.88 (0.83–0.93) among ever-smokers, P_int = 0.0005. Adjusting for smoking intensity (pack-years) did not alter the association among ever-smokers: 0.89 (0.83–0.95). In addition, the association between IGF-1 and lung cancer was only found for men [0.85 (0.79–0.91)] but not for women [1.00 (0.92–1.08); P_int = 0.003]. Both interaction terms remained significant when they were simultaneously included in the same model, P_int for ever-smoking = 0.002 and P_int for sex = 0.009.

To account for potential subclinical liver disease/injury that may confound the relationship between IGF-1 and liver cancer, we additionally adjusted for prediagnostic ALT, AST, GGT, and total bilirubin. The associations were not materially altered, HR (95% CI): 0.45 (0.37, 0.55; Supplementary Table S3). For kidney cancer, we additionally adjusted for baseline creatinine and cystatin-C, to account for baseline kidney function, and observed similar results, 1.08 (0.98, 1.18; Supplementary Table S6).

Our study demonstrated divergent associations between IGF-1 and site-specific cancers, with increased risks for breast, prostate, colorectal, kidney, thyroid cancers, and melanoma, and lower risks for lung, ovarian, HNC, and liver cancers. Our study also shows that IGF-1 has a modest positive association with total cancer risk (excluding nonmelanoma skin cancer) in both men and women. To our knowledge, our study represents the first large-scale prospective investigation of circulating IGF-1 and incident overall cancer.

The positive association between IGF-1 and breast cancer (HR = 1.24) is consistent with a prior pooled analysis (2), which found an OR of 1.28 (95% CI, 1.14–1.44) comparing extreme quintiles of IGF-1. Associations were comparable for premenopausal and postmenopausal women in our study, which is consistent with prior evidence (2). In addition, we demonstrated similar associations for low versus high SHBG or testosterone, suggesting IGF-1 may elevate breast cancer risk independently of other endogenous hormones. During the preparation of our manuscript, an independent analysis of the UK Biobank with a smaller sample of incident breast cancer cases (N = 4,360) obtained nearly identical findings (23). Moreover, the Mendelian randomization analyses conducted in this study showed a positive association between IGF-1 and overall and ER⁺ breast cancer, further supporting IGF-1 as a potential causal factor. Our finding of a positive association between IGF-1 and prostate cancer is consistent with a prior individual participant-level meta-analysis, with OR (95% CI) comparing extreme quintiles: 1.29 (1.16–1.43; ref. 24). Similarly, we confirmed prior findings of a positive association between IGF-1 and colorectal cancer, RR (95% CI) per 1-SD increase in IGF1: 1.07 (1.01–1.14; ref. 7). Finally, our results for thyroid cancer supports a European Prospective Investigation into Cancer report, where higher IGF-1 was positively associated with differentiated thyroid carcinoma, OR (95% CI) per doubling of IGF-1: 1.48 (1.06–2.08; ref. 25).

Prior research has suggested that IGF-1 may play a role in the incidence and progression of renal cell carcinoma (26). We found a significant positive association between IGF-1 and kidney cancer that remained robust to further adjustment for baseline kidney function. Our finding stood in contrast to a nested case–control study of male smokers (100 incident cases), which found an inverse association between IGF-1 and kidney cancer (27). In addition, our finding of a positive association between IGF-1 and melanoma risk stood in contrast with existing studies. One case–control study with IGF-1 measured after melanoma diagnosis found an inverse association (28). Another study utilized a nested case–control design and did not show a significant association between IGF-1 and melanoma (29). Hence, future prospective investigations are warranted on IGF-1 and incident kidney cancer or melanoma.

For lung cancer, a prior meta-analysis of six nested case–control studies (1,043 cases) found no significant associations with IGF-1, OR (95% CI) for high versus low IGF-1: 1.05 (0.80–1.37; ref. 8). Our study showed a modest inverse association between IGF-1 and lung cancer, which was confined to ever-smokers and men. Notably, smoking may impact IGF-1, as individuals who smoke tend to have lower circulating IGF-1 than those who abstain (30). In the UK Biobank, we observed that IGF-1 in never-smokers was 0.61 nmol/L higher on average than in ever-smokers, with an even larger difference between nonsmokers and smokers among men than among women (1.00 vs. 0.43 nmol/L, P < 0.0001). This difference may be partially accounted for by greater smoking intensity among men (5.4 higher pack-years). In turn, residual confounding from smoking is possible as reporting pack-years of smoking may differ between men and women. Alternatively, these interactions may be different depending on lung cancer subtypes, because smoking is a stronger risk factor for small cell and squamous cell lung cancer compared with adenocarcinomas (31). On the other hand, we observed a borderline positive association between IGF-1 and lung cancer risk among never-smokers. Similarly, residual confounding by smoking may have led to our finding of a modest inverse association between IGF-1 and HNC, particularly because studies have suggested IGF-1 receptor as a target for HNC (32). Indeed, in a post hoc analysis, we found that there was a trend toward increased risk for HNC among never-smokers: 1.07 (0.91–1.26), whereas there was a protective association among ever-smokers: 0.83 (0.74–0.93), P_int = 0.01. Nevertheless, we are not aware of other prospective studies that have assessed the association between IGF-1 and HNC. Hence, more research is needed to elucidate whether IGF-1 could play a role in lung or head and neck carcinogenesis, and how smoking may modify this relationship.

Similar to our observations, higher IGF-1 has been found to be associated with a lower risk of ovarian cancer in one prior meta-analysis (9). The included studies tended to observe a stronger inverse association between IGF-1 and ovarian cancer in younger or premenopausal women compared with older women. We similarly observed a nominally stronger inverse association in women younger than 55 years, HR (95% CI): 0.68 (0.55–0.85), than those older than 55 years, 0.92 (0.82–1.05), P_int = 0.02. Potential mechanisms for this inverse association between IGF-1 and ovarian cancer remains unclear, though previous studies have suggested that IGF-1 may downregulate the bioactivity of circulating estrogens due to its affinity for both IGF-1 receptors and estrogen receptors (33, 34). Further studies are warranted to elucidate the role that IGF-1 plays in the pathogenesis of ovarian cancer.

The strong inverse association we observed between higher IGF-1 levels and liver cancer is an outlier of all cancers examined, but this finding is consistent with prior studies (35, 36). Epidemiologic and experimental studies have implicated low IGF-1 as a risk factor for the development and progression of nonalcoholic fatty liver disease (37), which is a common risk factor for liver cancer in high-income countries. Moreover, the majority of circulating IGF-1 is produced by hepatocytes in response to growth hormone (GH), and thus lower circulating IGF-1 may reflect hepatic dysfunction (38). Despite adjustments for hepatic markers, we cannot rule out the possibility that undiagnosed liver disease led to lower IGF-1. Additional studies are needed to examine whether subclinical liver damage may perturb the IGF-pathway and whether low IGF-1 is causally related to hepatocellular carcinoma or simply a marker of hepatic injury/dysfunction.

There are several potential mechanisms for the positive association between IGF-1 and cancers. IGF-1 is positively associated with height, and greater early life height velocity and final attained height are both correlated with higher IGF-1 exposures (39). Height has been found in observational and Mendelian randomization studies to be positively associated with multiple malignancies (40–42). Patients who have acromegaly, a condition of excessive GH secretion and in turn higher IGF-1 levels, have an increased risk of colorectal cancer and possible increased risks of breast and prostate cancer (43). Conversely, Laron syndrome patients, who have a resistance to GH stimulation, experience a lower risk of cancer (44). Genetic variants and mutations in the IGF-1 receptor have been found to be associated with increased risk of breast, colorectal, and non–small cell lung cancers (45–47). Moreover, interactions between insulin and the IGF proteins are likely given the structural similarities of these proteins and their receptors. Previous studies have shown that, irrespective of diabetes status, higher circulating insulin and C-peptides were associated with increased risks of breast and colorectal cancer (48). Dysregulated signaling through these two hormones can activate downstream signaling through the MAPK and PI3K pathways, leading to enhanced protein synthesis, cell growth and division, and evasion of apoptosis (49). Therefore, IGF-1 may be a plausible target for the primary prevention of certain cancers. The potential mechanisms for the inverse associations between IGF-1 and cancers of the lung and ovaries are not clear but probably differ from those discussed above. For lung cancer, the borderline positive association among never-smokers is perhaps in line with the mechanisms described above and is likely to be free from residual confounding by smoking. Nevertheless, mechanistic studies and clinical interventions are needed to assess whether and why IGF-1 may selectively elevate the risks of certain cancers while playing a protective role in others.

Our study has several key strengths. The use of a prospective design with large sample size reduces the likelihood of reverse causation and allows for the comprehensive assessment of the relationship between circulating IGF-1 and multiple common and rare cancers. Our results were robust to extensive statistical adjustments and across a wide variety of participant characteristics. The measurement of IGF-1 using a common assay at a central laboratory can help reduce laboratory variations and random errors that could bias or attenuate associations (17). Several limitations warrant mentioning. Information on cancer stage, grade, subtype, and morphology were not available, so we could not examine whether associations with IGF-1 differed by these characteristics. For cancers with lower incidence rates, we may have lacked statistical power to detect an association if one truly exists. False positive findings due to the examination of multiple cancers is possible, though associations observed for most cancers remained significant even after correcting for multiple testing. We could not assess associations with circulating IGF-binding proteins, which could modify the associations for IGF-1. The median follow-up duration for participants in our study was around 7.2 years, which may be relatively short for several cancer types. While we observed consistent associations according to duration of follow-up, future studies with longer follow-up are warranted. Because of the observational nature of this study, we cannot rule out unmeasured confounding. Mendelian randomization studies and RCTs targeting the suppression of IGF-1 will be necessary to determine whether IGF-1 is causally associated with specific cancers.

In conclusion, our results suggest that serum IGF-1 may be a risk factor for specific invasive cancers. Whether our results indicate a causal role for IGF-1 and whether targeting IGF-1 has therapeutic benefits for the primary prevention of these cancers warrant further research.

No potential conflicts of interest were disclosed.

The funders had no role in the design, implementation, data management/analysis, interpretation of the study, or the decision to submit for publication.

F. Qian: Conceptualization, formal analysis, investigation, visualization, writing–original draft, writing–review and editing. D. Huo: Conceptualization, data curation, software, formal analysis, supervision, funding acquisition, validation, investigation, methodology, writing–original draft, project administration, writing–review and editing.

The current analyses were approved under the application number 49564. The authors thank the participants, investigators, and staff of the UK Biobank for providing them with the resources to pursue this research.

F. Qian was supported by the John D. Arnold Scientific Research Prize from the University of Chicago Pritzker School of Medicine. This study was partially supported by the Breast Cancer Research Foundation and NCI (R01 CA242929; to D. Huo).

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.