Incident Type 2 Diabetes Duration and Cancer Risk

A Prospective Study in Two US Cohorts

Yang Hu, ScD; Xuehong Zhang, MD, ScD; Yanan Ma, PhD; Chen Yuan, ScD; Molin Wang, PhD; Kana Wu, MD, PhD; Fred K. Tabung, MSPH, PhD; Deirdre Tobias, ScD; Frank B. Hu, MD, PhD; Edward Giovannucci, MD, ScD; Mingyang Song, MD, ScD


J Natl Cancer Inst. 2021;113(4):381-389. 

In This Article


Study Population

The NHS cohort was initiated in 1976 when 121 700 female registered nurses aged 30–55 years were recruited, and the HPFS cohort was established in 1986 and included 51 529 male health professionals aged 40–75 years. Participants were followed biennially through mailed questionnaires inquiring medical and lifestyle information.[7,8]

We set as baseline for the current study the first follow-up cycle for NHS (1978) and HPFS (1988) to identify patients with incident T2D. Prevalent diabetes in 1976 (n = 1955) and 1986 (n = 1640) were excluded because their diabetes diagnosis dates were not available. We also excluded participants with prevalent cancer at baseline (n = 4350 in NHS; n = 2652 in HPFS) and those who answered only the baseline questionnaire (n = 932 in NHS; n = 771 in HPFS). For incident T2D cases without known diagnosis date, we set the diagnosis date as the return date of the follow-up questionnaire in which participants first reported T2D diagnosis. After exclusion, we had 113 429 participants from NHS and 45 604 participants from HPFS.

This study was approved by the institutional review board at Brigham and Women's Hospital and Harvard T.H. Chan School of Public Health and those of participating registries as required. The return of a completed questionnaire was considered informed consent. We obtained written consent to acquire medical records to document cases of cancer.

Assessment of Type 2 Diabetes

Participants who reported a diagnosis of T2D were mailed a supplementary questionnaire regarding symptoms, diagnostic tests, and hypoglycemic therapy. For cases diagnosed before 1998, T2D was confirmed using the National Diabetes Data Group criteria,[9] and for cases of T2D identified after 1998, the cutoff point for elevated fasting plasma glucose concentrations was lowered to 7.0 mmol/L according to the American Diabetes Association criteria. We further considered HbA1c of 6.5% and higher in the diagnosis criteria for confirming T2D cases identified after January 2010.[10] Validation studies in the NHS and HPFS demonstrated the validity of using our supplementary questionnaire to adjudicate T2D diagnosis (Supplementary Methods, available online).[11,12]

Assessment of Cancer Incidence

Total cancer involved all types of cancer with International Classification of Diseases–9 code between 140 and 239 except nonmelanoma skin cancers and nonfatal prostate cancer. Nonmelanoma skin cancers were excluded from total cancer because of their high incidence and extremely low degree of malignancy, which would strongly influence the results. Nonfatal prostate cancer was not considered either, because of its consistent inverse association with T2D and potential detection bias resulting from prostate-specific antigen screening, all of which would introduce substantial heterogeneity to the risk estimates. Obesity-related cancer and diabetes-related cancer were defined according to previous studies.[13,14] Cancer diagnosis was confirmed by medical record review among participants who reported a diagnosis of cancer on the follow-up questionnaires. For the current analysis, we counted only the first documented cancer for participants diagnosed with multiple cancers.

Assessment of Diet, Physical Activity, Screening, and Medication use

A validated semiquantitative food frequency questionnaire was administered to collect information of diet and alcohol intake approximately every 4 years since 1980 in NHS and 1986 in HPFS. We used the alternative health eating index score to quantify overall diet quality.[15] In both cohorts, information on demographic profiles, lifestyle characteristics, and body weight were collected at baseline and in biennial questionnaires. Physical activity was self-reported via a validated questionnaire.[16] Starting from 1988 and every 2 years thereafter, the follow-up questionnaires asked participants whether they had had a colonoscopy and/or sigmoidoscopy in the previous 2 years in the 2 cohorts. Similarly, the information regarding fasting glucose examination was collected from 1998 in the NHS and 2000 in the HPFS. From 1988, women were asked whether they underwent a mammography screening during the previous 2 years. The information on insulin use and oral hypoglycemic drug use were collected from 1982 in NHS, but such data were not available in HPFS.

Assessment of C-Peptide and HbA1c

Blood samples were collected from subpopulations of the NHS (n = 32 826) in 1989–1990 and HPFS (n = 18 225) from 1993 to 1995. Detailed information on the blood collection has been described elsewhere.[17] In these subpopulations, multiple biomarkers including C-peptide and HbA1c were measured from several nested case-control studies for various outcomes. Data were pooled from 2 cohorts to increase statistical power, and a total of 10 896 (362 prevalent diabetes patients) and 11 105 (1689 prevalent diabetes patients) participants were included in the analysis for C-peptide and HbA1c, respectively (Supplementary Methods, available online).

Statistical Analysis

To capture the dynamic characteristics for participants with and without T2D during the follow-up, the demographic, lifestyle, and dietary information was presented according to the person-time of T2D status. For each participant, follow-up time was counted from the return date of the baseline questionnaire to the date of cancer diagnosis, death date, date of return of last available follow-up questionnaire, or the end of follow-up (June 2014 for NHS and January 2014 for HPFS), whichever happened first. T2D duration was defined in years as the difference between the questionnaire return date in each follow-up cycle and the diabetes diagnosis date. We first analyzed the 2 cohorts separately and then combined the estimates using a fixed-effect meta-analysis.

A multivariable Cox proportional hazards model with age (months) as the time scale was used to calculate hazard ratios (HRs) and 95% confidence intervals (CIs). Given the importance of body mass index (BMI) for both T2D and cancer, we additionally adjusted for the cumulative-averaged BMI in the multivariable model. The proportional hazards assumption was tested by including the interaction terms between T2D duration and follow-up time for all models, and no statistically significant violations were detected.

We calculated the hazard ratios according to categories of T2D duration using prespecified cutoffs: no more than 2 years, 2.1–4.0 years, 4.1–6.0 years, 6.1–8.0 years, 8.1–10.0 years, 10.1–15 years, and more than 15 years. We also employed the restricted cubic spline analysis[18,19] to test for potential nonlinear relationships between T2D duration and cancer risk. Three knots were specified for diabetes duration of 0, 6, and 10 years based on the results of the categorical analysis. To assess potential effect modification, we performed stratified analyses according to age (60 years or younger, older than 60 years), cumulative-averaged BMI (<25 kg/m2, 25–30 kg/m2, ≥30 kg/m2), and smoking status (never smoker, ever smoker). Product terms between T2D status and dichotomized age, smoking status, and continuous BMI were additionally included in the models, and Wald test was used to calculate the P value for interaction.

In a secondary analysis, we conducted a cross-sectional analysis by calculating the least square geometric means of C-peptide and HbA1c by diabetes duration to further understand the nonlinear association between T2D duration and cancer risk in the primary analysis. The diabetes duration was calculated as year of blood draw (1990 for NHS and 1994 for HPFS) minus the year of T2D diagnosis. The multiple linear regression model was adjusted for age at blood draw, aspirin and/or NSAID use, presence of chronic diseases or conditions (hypercholesterinemia, hypertension, cancer, osteoporosis, rheumatoid arthritis, stroke, and myocardial infarction), physical activity, smoking status, case-control status, sex, and BMI.[20,21]

Because participants may change their lifestyle and dietary behaviors following a diagnosis of T2D, the results might have been attenuated by overadjustment for these changes. Therefore, we conducted a sensitivity analysis by using baseline values for covariate adjustment. All statistical tests were 2-sided with a statistical significance level of .05 and were performed using SAS 9.4 (SAS Institute, Cary, NC).