| ObjectivesTo investigate the association between fasting blood glucose(FBG)levels and the risk of overall cancer and cancer by sites among Chinese population.Furthermore,based on the evidence we found,develop the risk prediction model for liver cancer.Besides providing high-quality evidence for the association between FBG and cancer,the prediction model that we developed could be potentially served as a tool to stratify the risk of liver cancer among general population.Methods1.According to the kailuan cohort that would be used throughout the whole project,the comprehensive description of the socio-demographic characteristics,lifestyle factors,personal health and blood test information of the baseline population was carried out.The incidence of overall cancer and cancer by sites were calculated.2.A baseline study cohort was derived from Kailuan cohort,Cox proportional hazards regression model,restricted cubic spline model(RCS),and the piecewise linear regression model were applied to evaluate the association and non-linear dose-response relationship between FBG and the risk of overall cancer and cancer by sites,respectively.3.The two-point FBG cohort of population without diabetes was derived from Kailuan cohort.Cox proportional risk regression model was used to explore the association of the percentage change of FBG in short term with the risk of overall cancer and cancer by sites.4.The three-point FBG cohort of population without diabetes was derived from Kailuan cohort.The group-based trajectory modeling(GBTM)was used to form the FBG trajectory within 4 years,the Cox proportional hazards regression model and competing risk model were applied to investigate the association between the FBG trajectory and the risk of overall and gastrointestinal cancer.5.The Kailuan cohort was 1:1 randomly divided into a training cohort and a validation cohort to construct an 8-year prediction model for the risk of liver cancer.In the training cohort,we screened for potential predictors and developed the prediction model.Afterwards,the performance of the model was evaluated in the validation cohort.Hosmer-Lemeshow's statistic and Harrell's C-index were used to evaluate calibration and discrimination,respectively.Results1.After four rounds of questionnaire survey and health examination,a total of 138,150 participants were included in the kailuan cohort,including 110,612 men and 27,538 women.The mean age was 51 years,population aged between 40-69 years was accounted for 71.96%of the total population.The sociodemographic characteristics were in line with the national level,and the prevalence of smoking and alcohol was lower than the national level.By December 31,2015,with a median follow-up of 8.8 years,a total of 4143 cancer cases(men:3378,women:765)were newly diagnosed in the cohort.The crude incidence of cancer was 381.16/100,000,and the standardized incidence was 160.91/100,000(men:155.30/100,000,women:170.14/100,000).The top 10 cancer sites in the cohort were lung cancer,colorectal cancer,liver cancer,stomach cancer,breast cancer,kidney cancer,esophageal cancer,bladder cancer,lymphoma,pancreatic cancer,and tumors of the brain and nervous system.2.The U-shaped association was observed between FBG level and the risk of overall cancer.In addition,the threshold effect indicated that,when the FBG?4.5 mmol/L,each 0.56 mmol/L increase in FBG was linearly associated with a 11%decrease of overall cancer,and when the FBG>4.5 mmol/L,each 0.56 mmol/L increase in FBG was linearly associated with a 1%increase of overall cancer.The dose-response relationship between FBG and the risk of cancer was differed by cancer sites.When FBG?4.5mmol/L,the risk of oral and lung cancer decreased with the increase of FBG concentration.When FBG>4.5mmol/L,the risk of colorectal cancer and leukemia increased with the increase of FBG concentration.The risk of liver cancer increased with the increase of FBG level,but the dose-response relationship was significantly different(18%vs.3%)at the threshold of 5.6 mmol/L.3.Compared with population with stable FBG(FBG change<5%),those with increases in FBG of>15%were observed to have a 15%increased risk of overall cancer.The similar results could be observed for colorectal cancer,pancreatic cancer,brain and nervous system tumors,and female breast cancer.4.Relative to the moderate-stable group,we found a higher hazard ratio for overall cancer in the low-increasing group and elevated-stable group,and BMI may modify the associations.5.Based on the information of lifestyle factors,routine blood test indicators and hepatitis virus factor,three prediction models of 8-year risk of liver cancer were derived,namely HBsAg model,general model and full model.The full model was composed of age,sex,HBsAg,smoking status,alcohol drinking status,tea consumption,FBG with diabetes,TC,Hs-CRP,and ALT,and showed good discrimination(Harrell's C-index=0.84,95%CI:0.80-0.87)in the validation data set.In subgroup analysis,the full model also performed well in the non-cirrhotic population(Harrell's c-index=0.83,95%CI:0.80-0.87),and the general model showed moderate performance in the HBsAg negative population(Harrell's c-index=0.72,95%CI:0.67-0.77).Conclusions1.According to the kailuan cohort,the participants were suitable for analysis,the information of exposures were comprehensive,the follow-up system was stable,and the collection of cancer cases were reliable,in addition,there was conducted four rounds of questionnaires survey and blood tests.Therefore,this cohort enables us to systematically investigate the association of baseline and changes of FBG with the risk of cancer.2.There was a U-shaped association between FBG level and the risk of overall cancer,and there was a threshold effect for the dose-response relationship between FBG level and risk of overall cancer.3.During the follow-up,population with increases in FBG or those with elevated-stable pattern have a higher risk of cancer,BMI could modify the association.4.The 8-year prediction model of liver cancer risk which composed of age,sex,HBsAg,smoking status,alcohol drinking status,tea consumption,FBG with diabetes,TC,Hs-CRP and ALT could effectively estimate the risk of liver cancer for the general population. |
PDF Full Text Request