AUTHOR COPY ONLY Predicting values of lipids and white blood cell count for all-site cancer in type 2 diabetes

Type 2 diabetic patients have increased cancer risk. We developed and validated an all-site cancer risk score in a prospective cohort of 7374 Chinese type 2 diabetic patients free of known history of cancer at enrolment, using split-half validation. Spline Cox model was used to detect common risk factors of cancer and to guide linear transformation of non-linear risk factors. After a median follow-up period of 5.45 years, 365 patients (4.95%) developed cancer. Body mass index (BMI; !24.0 or R27.6 kg/m 2 ), triglyceride (R0.81 to !1.41 mmol/l), high-density lipoprotein cholesterol (!0.9 or R1.8 mmol/l), total cholesterol (!4.3 mmol/l) and white blood cell (WBC) count (!5.8!10 9 count per litre) were associated with increased cancer risks and exhibited non-linear relationships. We further linear transformed these terms for selection using backward Cox regression (P!0.05 for stay) in the training dataset. In the test dataset, calibration was checked using Hosmer–Lemeshow test and discrimination checked using area under receiver operating characteristic curve. In addition to age and current smoking, only linear-transformed total cholesterol and WBC count were selected. The risk score was 0.0488!age (years)K0.5810!total cholesterol (mmol/l, coded to 4.3 if O4.3)K0.3596!WBC count (10 9 counts/l, 5.8 if O5.8)C0.6390!current smoking status (1 if yes). The 5-year probability of cancer was 1K0.9590 EXP(0.9382!(RISK SCOREC1.5903)) .The predicted cancer probability was not significantly different from the observed cancer probability during the 5-year follow-up. The adjusted area under receiver operating characteristic curve was 0.712. In conclusion, BMI, lipids and WBC count have predicting values for cancer. Endocrine-Related Cancer (2008) 15 597–607