目的 构建科学、可操作的居民健康评价指标体系及分级评价模型。方法 通过专家函询构建居民健康评价指标体系并采用层次分析法确定指标权重，进而以评价体系的健康指标值为输入、健康分级为输出，采用反向传播神经网络算法建立居民健康分级评价模型并验证。结果 2轮函询专家积极程度均为100.00%，专家权威系数均为0.89，条目重要性评分均值分别为3.90~5.00、4.00~5.00，变异系数分别为0.00~0.34、0.00~0.28，肯德尔协调系数分别为0.202、0.289，最终形成的指标体系由5个一级指标、14个二级指标、25个三级指标构成；居民健康分级评价模型在训练集和验证集的总体准确率为98.54%和91.63%，验证集模型曲线下面积分别为0.995、0.975、0.965、0.982、0.998。结论 本研究构建的指标体系涵盖居民健康的关键影响要素，健康分级评价模型具有良好区分度，可实现对居民健康状况的客观量化评价。

Objective To develop a scientific and operable indicator system and a grading evaluation model for the assessment of resident health.Methods The resident health evaluation indicator system was constructed through expert consultations,with indicator weights determined by the Analytic Hierarchy Process (AHP).Using the health indicator values from the evaluation system as input and health grading as output,a resident health grading evaluation model was established and validated using the backpropagation neural network algorithm.Results The response rate for both rounds of expert consultation was 100.00%,with an expert authority coefficient of 0.89.The mean importance scores of each item in the 2 rounds ranged from 3.90 to 5.00 and 4.00 to 5.00,respectively,with coefficients of variation of 0.00-0.34 and 0.00-0.28.The Kendall’s coordination coefficients were 0.202 and 0.289,respectively.The final indicator system comprised 5 primary indicators,14 secondary indicators,and 25 tertiary indicators.The overall accuracy of the resident health grading evaluation model was 98.54% in the training set and 91.63% in the validation set.The area under the curve (AUC) values for the validation set were 0.995,0.975,0.965,0.982,and 0.998,respectively.Conclusions The constructed indicator system covers key factors influencing resident health,and the health grading evaluation model demonstrates good discriminative ability,enabling objective and quantitative assessment of resident health status.

R47-05

北京市自然科学基金-海淀原始创新联合基金（L222103）