Poisson Regression (PR).

Diabetes mellitus stands out as one of the most prevalent chronic conditions affecting pediatric populations. The escalating incidence of childhood type 1 diabetes (T1D) globally is a matter of increasing concern. Developing an effective model that leverages Key Performance Indicators (KPIs) to understand the incidence of T1D in children would significantly assist medical practitioners in devising targeted monitoring strategies. This study models the number of monthly new cases of T1D and its associated KPIs among children aged 0 to 14 in Saudi Arabia. The study involved collecting de-identified data (n=377) from diagnoses made between 2010 and 2020, sourced from pediatric diabetes centers in three cities across Saudi Arabia. Poisson regression (PR), and various machine learning (ML) techniques, including random forest (RF), support vector machine (SVM), and K-nearest neighbor (KNN), were employed to model the monthly number of new T1D cases using the local data. The performance of these models was assessed using both numbers of KPIs and metrics such as the coefficient of determination (), root mean squared error (RMSE), and mean absolute error (MAE). Among various Poisson and ML models, both model considering birth weight over 3.5 kg, maternal age over 25 years at the child’s birth, family history of T1D, and nutrition history, specifically early introduction to cow milk and model taking into account birth weight over 3.5 kg, maternal age over 25 years at the child’s birth, and nutrition history (early introduction to cow milk) emerged as the best-reduced models. They achieved of (0.89,0.88), RMSE (0.82, 0.95) and MAE(0.62,0.67). Additionally, models with fewer KPIs, like model that considers maternal age over 25 years and early introduction to cow milk, achieved consistently high values ranging from 0.80 to 0.83 across all models. Notably, this model demonstrated smaller values of RMSE (0.92) and MAE (0.67) in the KNN model. Simplified models facilitate the efficient creation and monitoring of KPIs profiles. The findings can assist healthcare providers in collecting and monitoring influential KPIs, enabling the development of targeted strategies to potentially reduce, or reverse, the increasing incidence rate of childhood T1D in Saudi Arabia.

糖尿病是影响儿科人群最普遍的慢性疾病之一。全球儿童1型糖尿病（type 1 diabetes, T1D）发病率的持续上升已成为日益令人担忧的公共卫生议题。构建一款借助关键绩效指标（Key Performance Indicators, KPIs）解析儿童T1D发病情况的有效模型，将极大助力医疗从业者制定针对性监测方案。本研究针对沙特阿拉伯0至14岁儿童的T1D月度新增病例数及其相关KPIs开展建模。研究收集了2010年至2020年间，来自沙特阿拉伯三座城市的儿科糖尿病中心的去标识化数据（n=377）。本研究采用泊松回归（Poisson regression, PR）以及多种机器学习（machine learning, ML）技术，包括随机森林（random forest, RF）、支持向量机（support vector machine, SVM）与K近邻（K-nearest neighbor, KNN），基于本地数据对T1D月度新增病例数进行建模。采用KPIs数量以及决定系数（coefficient of determination）、均方根误差（root mean squared error, RMSE）与平均绝对误差（mean absolute error, MAE）等指标对各模型的性能进行评估。在多款泊松回归与ML模型中，以下两款精简模型表现最优：其一纳入了新生儿出生体重超过3.5kg、产妇分娩时年龄超过25岁、T1D家族史以及喂养史（具体为早期引入牛乳）这几项指标；其二纳入了新生儿出生体重超过3.5kg、产妇分娩时年龄超过25岁以及喂养史（早期引入牛乳）这几项指标。二者的决定系数分别为0.89与0.88，均方根误差分别为0.82与0.95，平均绝对误差分别为0.62与0.67。此外，仅纳入较少KPIs的模型（例如仅考虑产妇分娩时年龄超过25岁与早期引入牛乳这两项指标的模型），其性能指标始终维持在0.80至0.83的较高区间。值得注意的是，在KNN模型中，该精简模型的RMSE与MAE分别达到0.92与0.67，表现更优。精简模型有助于高效构建并监测KPIs特征画像。本研究结果可帮助医疗工作者收集并监测具有影响力的KPIs，从而助力制定针对性策略，以潜在减缓甚至逆转沙特阿拉伯儿童T1D发病率持续上升的趋势。