Abstract:Objective To systematically evaluate the predictive efficacy of four nutritional assessment tools for hypoglycemia risk in hospitalized patients with type 2 diabetes mellitus（T2DM），explore the independent association between malnutrition and hypoglycemia，and construct a hypoglycemia risk prediction model. Methods A total of 3 864 T2DM patients hospitalized in the Affiliated Hospital of Qingdao University from January 2014 to December 2023 were retrospectively enrolled and grouped according to fasting plasma glucose（FPG）levels at admission：patients with FPG ≤ 3.9 mmol/L were defined as the hypoglycemia group（n=1 273），and those with FPG between 4.0 and 6.0 mmol/L were defined as the euglycemia group（n=2 591）. Four nutritional assessment tools，including Prognostic Nutritional Index（PNI），Nutritional Risk Index（NRI），Controlling Nutritional Status（CONUT）score，and Nutritional Risk Screening 2002（NRS2002），were used to determine nutritional status，and their predictive abilities for hypoglycemia were compared. Least absolute shrinkage and selection operator（LASSO）regression was applied to screen characteristic variables，a multivariate logistic regression model was established，a nomogram was drawn，and the predictive performance of the model was evaluated using receiver operating characteristic（ROC）curve and Bootstrap method. Results ROC curve analysis showed that among the four nutritional tools，NRS2002 had the optimal predictive ability for hypoglycemia［area under the curve（AUC）=0.697，sensitivity=49.10%，specificity=90.04%］，while the AUC values of NRI，PNI，and CONUT were 0.612，0.590，and 0.587，respectively，indicating limited overall predictive efficacy of nutritional tools. Multivariate logistic regression analysis revealed that malnutrition assessed by NRS2002 had the largest odds ratio（OR）for hypoglycemia risk，at 2.68（95%CI：2.39-3.01）. PNI and NRI were not included in the final multivariate logistic regression model after LASSO regression variable screening，suggesting their limited applicability in this population. The final constructed regression equation was as follows：logit（P）=- 1.418 + 0.842 ×NRS2002 + 0.343 × disease duration + 0.233 × oral hypoglycemic agents alone + 1.778 × insulin monotherapy + 1.839 ×combined oral hypoglycemic agents and insulin- 0.662 × history of coronary heart disease- 0.441 × prealbumin- 0.229 ×estimated glomerular filtration rate. The model exhibited good ability to predict hypoglycemia occurrence in T2DM，with an AUC of 0.846 on the ROC curve；a sensitivity of 77.5% and a specificity of 80.1% . Conclusion Malnutrition is independently associated with hypoglycemia in hospitalized T2DM patients. The nomogram model has better predictive efficacy than commonly used nutritional tools and good clinical practicability，providing new ideas for the integrated application of clinical nutritional risk screening tools in blood glucose management.