Table_1_Development and validation of a novel nomogram for prediction of ketosis-prone type 2 diabetes.xlsx

BackgroundKetosis-prone type 2 diabetes (KPD), as a unique emerging clinical entity, often has no clear inducement or obvious clinical symptoms at the onset of the disease. Failure to determine ketosis in time may lead to more serious consequences and even death. Therefore, our study aimed to develop and validate a novel nomogram to predict KPD.
MethodsIn this retrospective study, clinical data of a total of 398 newly diagnosed type 2 diabetes in our hospital who met our research standards with an average age of 48.75 ± 13.86 years years old from January 2019 to December 2022 were collected. According to the occurrence of ketosis, there were divided into T2DM groups(228 cases)with an average age of 52.19 ± 12.97 years, of whom 69.74% were male and KPD groups (170cases)with an average age of 44.13 ± 13.72 years, of whom males account for 80.59%. Univariate and multivariate logistic regression analysis was performed to identify the independent influencing factors of KPD and then a novel prediction nomogram model was established based on these independent predictors visually by using R4.3. Verification and evaluation of predictive model performance comprised receiver-operating characteristic (ROC) curve, corrected calibration curve, and clinical decision curve (DCA).
Results4 primary independent predict factors of KPD were identified by univariate and multivariate logistic regression analysis and entered into the nomogram including age, family history, HbA1c and FFA. The model incorporating these 4 predict factors displayed good discrimination to predict KPD with the area under the ROC curve (AUC) of 0.945. The corrected calibration curve of the nomogram showed good fitting ability with an average absolute error =0.006 < 0.05, indicating a good accuracy. The decision analysis curve (DCA) demonstrated that when the risk threshold was between 5% and 99%, the nomogram model was more practical and accurate.
ConclusionIn our novel prediction nomogram model, we found that age, family history, HbA1c and FFA were the independent predict factors of KPD. The proposed nomogram built by these 4 predictors was well developed and exhibited powerful predictive performance for KPD with high discrimination, good accuracy, and potential clinical applicability, which may be a useful tool for early screening and identification of high-risk population of KPD and therefore help clinicians in making customized treatment strategy.

背景：易酮症型2型糖尿病（Ketosis-prone type 2 diabetes, KPD）是一种新兴的独特临床实体，患者发病时常无明确诱因及明显临床症状；若未能及时识别酮症，可能引发严重后果甚至死亡。因此本研究旨在构建并验证一款新型列线图，用于预测KPD的发生。
方法：本项回顾性研究收集了2019年1月至2022年12月期间，我院符合研究纳入标准的398例新诊断2型糖尿病患者的临床资料，患者平均年龄为48.75±13.86岁。根据酮症发生情况将受试者分为两组：2型糖尿病（Type 2 Diabetes Mellitus, T2DM）组（228例，平均年龄52.19±12.97岁，男性占比69.74%）与KPD组（170例，平均年龄44.13±13.72岁，男性占比80.59%）。采用单因素及多因素logistic回归分析筛选KPD的独立影响因素，基于筛选出的独立预测因子，借助R4.3软件可视化构建新型预测列线图模型。通过受试者工作特征（Receiver-operating characteristic, ROC）曲线、校正校准曲线以及临床决策曲线（Decision Curve Analysis, DCA）对预测模型的性能进行验证与评估。
结果：经单因素及多因素logistic回归分析，共筛选出4项KPD的独立预测因子，分别为年龄、糖尿病家族史、糖化血红蛋白（Hemoglobin A1c, HbA1c）与游离脂肪酸（Free Fatty Acid, FFA），并将其纳入列线图模型。纳入该4项预测因子的模型对KPD具有良好的区分能力，受试者工作特征曲线下面积（Area Under the Curve, AUC）为0.945。列线图的校正校准曲线显示出良好的拟合效能，平均绝对误差为0.006，小于0.05，提示模型具有较好的准确性。临床决策曲线分析（DCA）结果显示，当风险阈值处于5%~99%区间时，该列线图模型具有更高的临床实用性与准确性。
结论：本研究构建的新型预测列线图模型显示，年龄、糖尿病家族史、HbA1c及FFA为KPD的独立预测因子。基于该4项预测因子构建的列线图开发完善，对KPD具有优异的预测性能：区分度高、准确性好且具备潜在临床应用价值，可作为KPD高危人群早期筛查与识别的有效工具，助力临床医师制定个体化治疗方案。