基于MLP神经网络特征筛选与服务质量动态监测的家庭医生“医-助”共管模式在分级诊疗中的探索与实践

董修全; 常菲; 王海棠; 崔晨辉

doi:10.16766/j.cnki.issn.1674-4152.004538

基于MLP神经网络特征筛选与服务质量动态监测的家庭医生“医-助”共管模式在分级诊疗中的探索与实践

doi: 10.16766/j.cnki.issn.1674-4152.004538

上海市浦东新区上钢社区卫生服务中心，上海 200126

基金项目:

上海市浦东新区科技发展基金事业单位民生科研专项医疗卫生项目 PKJ2022-Y87

详细信息

通讯作者:
常菲，E-mail：changfei2g8@163.com

中图分类号: R197.61 R192.3
计量
- 文章访问数: 7
- HTML全文浏览量: 2
- PDF下载量: 0
- 被引次数: 0
出版历程
- 收稿日期: 2025-06-10
- 网络出版日期: 2026-11-10

Exploration and practice of a family doctor "doctor-assistant" co-management model based on MLP neural network feature screening and dynamic monitoring of service quality in hierarchical diagnosis and treatment

Shanghai Pudong Shanggang Community Health Service Center, Shanghai 200126, China

摘要

摘要: 目的探索家庭医生“医-助”共管模式对签约服务质量的影响，构建机器学习驱动的服务质量评价体系。方法选取2023年6月—2024年6月上海市浦东新区上钢社区卫生服务中心60名全科医生为研究对象，采用计算机随机数字法按1∶1分为试验组和对照组，各30名。对照组实施常规家庭医生签约服务自我管理模式，试验组在常规服务基础上实施“医-助”共管模式。比较2组签约数量、就诊流向等10项指标差异，并采用MLP神经网络分析特征重要性。结果年度评估中，试验组总得分为75（64, 83）分，显著高于对照组[69（60, 75）分, P < 0.05]，2组签约人数[20(15, 25) vs. 15(10, 20)]、重点人群比例[6(5, 7) vs. 5(4, 7)]等7项核心指标差异均有统计学意义（P < 0.05）。试验组7~12月总得分显著高于对照组（P < 0.05）。机器学习分析显示，签约人数、重点人群比例、组合内就诊率和糖化检测数为关键驱动指标，累计贡献度为69.1%。试验组干预后总得分提升44.1%，签约人数、重点人群比例、组合内就诊率等核心指标显著改善（P < 0.05），其余非核心指标亦明显提高（P < 0.05）。对照组总得分提升27.3%，部分非核心指标改善（P < 0.05）。结论 “医-助”共管模式可有效提升家庭医生签约服务质量，结合MLP特征筛选技术实现动态监测与精准优化，为分级诊疗制度深化提供实践范式。
- 家庭医生 /
- 共管模式 /
- 签约服务 /
- 机器学习
Abstract: Objective To explore the impact of the family doctor "doctor-assistant co-management" model on the quality of contract services, and to construct a machine learning-driven service quality evaluation system. Methods Sixty general practitioners from Shanghai Pudong Shanggang Community Health Service Center between June 2023 and June 2024 were enrolled and randomly assigned at a 1∶1 ratio into an experimental group and a control group using a computer-generated random number method, with 30 practitioners in each group. The control group received the conventional self-management model for family doctor contract services, whereas the experimental group received the "doctor-assistant" co-management model in addition to conventional services. Differences in 10 indicators, including the number of signed contracts and visit flow, were compared between the two groups, and an MLP neural network was used to analyze feature importance. Results In the annual assessment, the total score of the experimental group was 75 (64, 83), which was significantly higher than that of the control group [69 (60, 75), P < 0.05]. Significant differences were observed in seven core indicators, such as the number of signed contracts [20 (15, 25) vs. 15 (10, 20)] and the proportion of key populations [6 (5, 7) vs. 5(4, 7)], P < 0.05. The total score of the experimental group from July to December was significantly higher than that of the control group (P < 0.05). Machine learning analysis showed that the number of signed contracts, the proportion of key populations, within-team visit rate, and the number of glycated hemoglobin tests as the key driving indicators, with a cumulative contribution rate of 69.1%. Following the intervention, the total score of the experimental group increased by 44.1%. Core indicators, such as the number of signed contracts, the proportion of key populations, and within-team visit rate, showed significant improvements (P < 0.05), along with significant enhancement in non-core indicators (P < 0.05). The total score of the control group increased by 27.3%, with improvements observed only in some non-core indicators (P < 0.05). Conclusion The "doctor-assistant co-management" model can effectively improve the quality of family doctor contract services. When integrated with MLP-based feature screening technology, it enables dynamic monitoring and precise optimization, providing a practical paradigm for deepening the hierarchical diagnosis and treatment system.
- Family doctor /
- Co-management model /
- Contract services /
- Machine learning

HTML全文

图 1 家医年度平均总分箱线图

Figure 1. Box plot of the average total score of family doctors for the year

下载: 全尺寸图片幻灯片

图 2 家医年度各指标平均得分小提琴图

Figure 2. Annual average score bar chart of various indicators for family doctors

下载: 全尺寸图片幻灯片

图 3 MLP神经网络分类模型特征变量分析结果图

Figure 3. Analysis result graph of feature variables of the MLP neural network classification model

下载: 全尺寸图片幻灯片

表 1 家庭医生签约服务质量考核评分表

Table 1. Family doctor contracting service quality assessment scoring form

一级指标	二级指标	指标分值	指标解释	考核方案
签约数量	签约人数	25	按要求完成当月任务	新增签约每1人5分，封顶25分。
	重点比例	8	≥60%	≥60%得8分，每降1%扣1分，扣完为止。
就诊流向	组合内就诊率	10	≥70%	≥70%得1分，每升1%加1分；＜70%同比升1%加1分。封顶10分。
	社区内就诊率	12	≥60%	≥60%得1分，每升1%加1分；＜60%同比升1%加1分。封顶12分。
	组合外延方	5	＜20%	＜20%得5分，≥20%得0分。
频次及费用	社区居民就诊次数	10	不得高于月份×2次	≤月份×2次得10分，＞月份×2次得4分，同比少1次加1分，封顶10分。
	社区居民人均累计门诊费用	15	不得高于每月的预估金额	达月指标得10分；每降100元加1分，封顶15分；超20%倒扣5分。
健康管理	签约糖尿病人随访数	6	完成门诊糖尿病随访	完成率≥95%得6分；90%~94%得5分；80%~89%得4分；＜80%得0分。
	糖化血红蛋白检测数	4	糖化检测每月5次	每月5次得4分，每少1次扣1分。
信息化工具使用	费用APP登录	5	每月登录2次	每月≥2次得5分；连续2个月未登录倒扣5分。

下载: 导出CSV

表 2 2组家医全年各考核指标比较[M(P₂₅, P₇₅)，分]

Table 2. Comparison of all-year-round assessment indicators for the two groups of family doctors [M(P₂₅, P₇₅), points]

组别	人数	签约人数得分	重点人群比例得分	组合内就诊率得分	社区内就诊率得分	组合外延方得分	月平均累计就诊次数得分	累计门诊费用得分	糖尿病人随访数得分	糖化检测数得分	费用APP登录得分	总得分
对照组	30	15(10, 20)	5(4, 7)	7(4, 8)	8(5, 10)	5(0, 5)	7(6, 9)	13(11, 14)	5(4, 6)	2(1, 4)	5(0, 5)	69(60, 75)
试验组	30	20(15, 25)	6(5, 7)	7(5, 9)	9(6, 11)	5(0, 5)	8(7, 9)	13(12, 14)	5(4, 6)	3(2, 4)	5(0, 5)	75(64, 83)
Z值		-3.796	-4.363	-3.233	-2.468	-0.323	-4.847	-3.032	-0.937	-3.260	-0.774	-6.258
P值		＜0.001	＜0.001	0.001	0.013	0.708	＜0.001	0.002	0.320	＜0.001	0.366	＜0.001

下载: 导出CSV

表 3 2组家医每月服务考核总得分比较(x±s，分)

Table 3. Comparison of the total scores of monthly service assessment for the two groups of family doctors (x±s, points)

组别	人数	1月	2月	3月	4月	5月	6月	7月	8月	9月	10月	11月	12月
对照组	30	60.333±10.765	57.567±9.810	58.267±10.736	61.133±9.552	65.000±5.586	69.500±7.809	70.000±7.916	71.400±8.589	72.567±8.192	75.400±6.681	74.033±7.722	76.867±7.210
试验组	30	61.833±13.508	60.667±12.349	60.700±11.677	64.467±9.787	66.167±8.034	70.500±7.894	75.467±6.125	77.433±7.584	79.367±7.054	83.733±4.041	85.467±4.773	89.133±4.808
Z值		-0.468	-1.059	-0.826	-1.313	-0.642	-0.485	-2.941	-2.836	-3.387	-5.747	-6.782	-7.623
P值		0.642	0.294	0.412	0.194	0.523	0.630	0.005	0.006	0.001	＜0.001	＜0.001	＜0.001

下载: 导出CSV

表 4 MLP神经网络分类模型特征变量重要度得分排序

Table 4. Ranking of feature importance scores for the MLP neural network classification model

变量	重要度得分
签约人数得分	0.183
重点人群比例得分	0.177
组合内就诊率得分	0.170
糖化检测数得分	0.161
月平均累计就诊次数得分	0.096
累计门诊费用得分	0.071
费用APP登录得分	0.059
糖尿病人随访数得分	0.051
社区内就诊率得分	0.019
组合外延方得分	0.014

下载: 导出CSV

表 5 试验组试验前后家医服务各考核指标评分及总分差异比较(分)

Table 5. Comparison of differences in the scores of each assessment item and the total score of home medical services before and after the trial in the experimental group (points)

时间	签约人数得分[M(P₂₅, P₇₅)]	重点人群比例得分[M(P₂₅, P₇₅)]	组合内就诊率得分[M(P₂₅, P₇₅)]	社区内就诊率得分[M(P₂₅, P₇₅)]	组合外延方得分[M(P₂₅, P₇₅)]	月平均累计就诊次数得分[M(P₂₅, P₇₅)]	累计门诊费用得分[M(P₂₅, P₇₅)]	糖尿病人随访数得分[M(P₂₅, P₇₅)]	糖化检测数得分[M(P₂₅, P₇₅)]	费用APP登录得分[M(P₂₅, P₇₅)]	总得分(x±s)
第1个月	10(5, 20)	6(3, 7)	7(4, 9)	7(5, 11)	5(0, 5)	7(5, 9)	12(11, 14)	5(4, 5)	2(1, 3)	5(0, 5)	61.833±13.508
第12个月	25(20, 25)	8(7, 8)	9(8, 10)	10(10, 11)	5(0, 5)	9(8, 10)	14(13, 15)	6(5, 6)	4(3, 4)	5(5, 5)	89.133±4.808
统计量	-4.650^a	-4.775^a	-3.622^a	-3.134^a	-0.222^a	-3.541^a	-3.023^a	-3.770^a	-4.302^a	-0.665^a	-10.253^b
P值	＜0.001	＜0.001	＜0.001	0.001	0.790	＜0.001	0.002	＜0.001	＜0.001	0.399	＜0.001
注：^a为Z值，^b为t值。

下载: 导出CSV

表 6 对照组试验前后家医服务各考核指标评分及总分差异比较(分)

Table 6. Comparison of differences in the scores of each assessment item and the total score of home medical services before and after the trial in the control group (points)

时间	签约人数得分[M(P₂₅, P₇₅)]	重点人群比例得分[M(P₂₅, P₇₅)]	组合内就诊率得分[M(P₂₅, P₇₅)]	社区内就诊率得分[M(P₂₅, P₇₅)]	组合外延方得分[M(P₂₅, P₇₅)]	月平均累计就诊次数得分[M(P₂₅, P₇₅)]	累计门诊费用得分[M(P₂₅, P₇₅)]	糖尿病人随访数得分[M(P₂₅, P₇₅)]	糖化检测数得分[M(P₂₅, P₇₅)]	费用APP登录得分[M(P₂₅, P₇₅)]	总得分(x±s)
第1个月	15(5, 25)	4(2, 7)	6(4, 8)	7(4, 10)	5(0, 5)	7(5, 8)	11(10, 13)	5(4, 6)	2(0, 3)	5(0, 5)	60.333±10.765
第12个月	15(10, 20)	6(5, 8)	7(6, 9)	9(8, 11)	5(5, 5)	8(7, 9)	13(12, 14)	5(5, 6)	3(2, 4)	5(5, 5)	76.867±7.210
统计量	-0.495^a	-3.267^a	-2.299^a	-3.290^a	-1.331^a	-1.774^a	-3.992^a	-1.478^a	-3.511^a	-0.665^a	-6.872^b
P值	0.618	＜0.001	0.021	＜0.001	0.105	0.073	＜0.001	0.119	＜0.001	0.357	＜0.001
注：^a为Z值，^b为t值。

下载: 导出CSV

参考文献(20)

[1]	中华人民共和国国务院. 国务院办公厅关于推进分级诊疗制度建设的指导意见[EB/OL]. (2015-09-18)[2025-04-30]. https://www.gov.cn/gongbao/content/2015/content_2937321.htm. The State Council of the People's Republic of China. Guidelines of the General Office of the State Council on Promoting the Construction of the Tiered Diagnosis and Treatment System[EB/OL]. (2015-09-18)[2025-04-30]. https://www.gov.cn/gongbao/content/2015/content_2937321.htm.
[2]	韩志鹏, 陈星, 王煜琳, 等. 家庭医生签约对居民基层医疗卫生机构服务利用的影响研究: 基于倾向得分匹配法[J]. 现代预防医学, 2025, 52(6): 1074-1078, 1143. Han Z P, Chen X, Wang Y L, et al. Effect of family doctor contract on the service utilization of residents'primary medical and health institutions: based on propensity score matching method[J]. Modern Preventive Medicine, 2025, 52(6): 1074-1078, 1143.
[3]	高俊. 上海市家庭医生签约服务机制分析: 基于利益相关者理论[J]. 中国初级卫生保健, 2022, 36(6): 53-55. Gao J. Analysis of Shanghai family doctor contract service mechanism: based on the stakeholder theory[J]. Chinese Primary Health Care, 2022, 36(6): 53-55.
[4]	钟捷, 张永春, 郭瑾. 上海市家庭医生助理聘任工作实践探索[J]. 中国初级卫生保健, 2022, 36(1): 20-22, 25. Zhong J, Zhang Y C, Guo J. Practice of employment of general practice assistants in Shanghai from 2011 to 2019[J]. Chinese Primary Health Care, 2022, 36(1): 20-22, 25.
[5]	本刊记者. 中共中央国务院印发《"健康中国2030"规划纲要》大健康产业大有可为健康中国势在必行[J]. 中国品牌与防伪, 2023(S1): 4-6. The reporter of this magazine. The Central Committee of the Communist Party of China and the State Council have issued the "Healthy China 2030" Planning Outline. The big health industry holds great potential and the realization of Healthy China is imperative[J]. China Brand and Anti-counterfeiting, 2023(S1): 4-6.
[6]	王琨, 李明, 张晓波, 等. 人工智能在心血管健康管理中的应用现状[J]. 医疗卫生装备, 2024, 45(2): 92-96. Wang K, Li M, Zhang X B, et al. Application progress of artificial intelligence in cardiovascular health management[J]. Chinese Medical Equipment Journal, 2024, 45(2): 92-96.
[7]	廖华龙, 曾小茜, 李华凤, 等. 机器学习在疾病预测中的应用[J]. 生物医学工程研究, 2021, 40(2): 203-209. Liao H L, Zeng X Q, Li H F, et al. Application of machine learning in disease prediction[J]. Journal of Biomedical Engineering Research, 2021, 40(2): 203-209.
[8]	茅荣智, 陆小花, 裴佳璐, 等. 基于机器学习的支原体肺炎诊断中临床表现与生化指标的关键特征分析[J]. 现代计算机, 2024, 30(14): 9-17. Mao R Z, Lu X H, Pei J L, et al. Key feature analysis of clinical manifestations and biochemical indicators in mycoplasma pneumonia diagnosis based on machine learning[J]. SmartTech Innovations, 2024, 30(14): 9-17.
[9]	尤红俊, 赵倩倩, 苟棋玲, 等. 基于机器学习算法筛选肥厚性心肌病铁死亡的潜在疾病特征基因[J]. 实用心脑肺血管病杂志, 2024, 32(5): 59-65. You H J, Zhao Q Q, Gou Q L, et al. Screening of potential disease characteristic genes of ferroptosis in hypertrophic cardiomyopathy based on machine learning algorithm[J]. Practical Journal of Cardiac Cerebral Pneumal and Vascular Disease, 2024, 32(5): 59-65.
[10]	钟捷, 张永春, 郭瑾. 上海市家庭医生助理主要聘任现况与政策建议[J]. 中国初级卫生保健, 2021, 35(11): 17-19. Zhong J, Zhang Y C, Guo J. The current situation and policy suggestions of the mainstream employment mode of general practice assistant in Shanghai[J]. Chinese Primary Health Care, 2021, 35(11): 17-19.
[11]	窦雄, 郑传芬, 武书兴, 等. 我国家庭医生签约服务现状问题及对策研究进展[J]. 现代医药卫生, 2021, 37(2): 229-232. Dou X. Research progress on the current situation and countermeasures of family doctor contract services in our country[J]. Journal of Modern Medicine & Health, 2021, 37(2): 229-232.
[12]	何武, 平文俊, 张鑫宇, 等. 中国社区全科医生工作压力现状及其影响因素研究[J]. 中国全科医学, 2021, 24(28): 3590-3596. He W, Ping W J, Zhang X Y, et al. Status and influencing factors of work stress among Chinese general practitioners[J]. Chinese General Practice, 2021, 24(28): 3590-3596.
[13]	朱宇霞, 王礼文, 赵亮, 等. 公共卫生视角下家庭医生助理服务现状与发展探讨[J]. 中国卫生产业, 2023, 20(10): 224-227. Zhu Y X, Wang L W, Zhao L, et al. Exploring the current situation and development of family physician assistant services from a public health perspective[J]. China Health Industry, 2023, 20(10): 224-227.
[14]	徐炜, 金璐, 刘杰, 等. 社区家庭医生签约服务模式发展现况及对策分析[J]. 中国社区医师, 2021, 37(35): 178-179. Xu W, Jin L, Liu J, et al. Development status and countermeasures of community family doctor signing service model[J]. Chinese Community Doctors, 2021, 37(35): 178-179.
[15]	朱纯, 李冬华, 张幸娜, 等. 基于全科医生签约决策模型提升全科医生签约数量与质量的策略研究[J]. 中华全科医学, 2022, 20(3): 441-445. doi: 10.16766/j.cnki.issn.1674-4152.002373 Zhu C, Li D H, Zhang X N, et al. Strategy research on improving the quantity and quality of general practitioner contract based on general practitioner contract decision model[J]. Chinese Journal of General Practice, 2022, 20(3): 441-445. doi: 10.16766/j.cnki.issn.1674-4152.002373
[16]	李娅玲, 杜兆辉, 潘志刚. 上海市某社区卫生服务中心签约居民就诊情况的相关因素分析[J]. 中华全科医师杂志, 2023, 22(3): 283-287. Li Y L, Du Z H, Pan Z G. Relevant factors of clinic attendance for contracted residents in a community health service center in Shanghai[J]. Chinese Journal of General Practitioners, 2023, 22(3): 283-287.
[17]	陈晓燕, 陈斌, 栾伟, 等. 基于上海"社区云管理"大数据的家庭医生改革后签约居民就诊情况分析[J]. 现代医院管理, 2021, 19(2): 6-8. Chen X Y, Chen B, Luan W, et al. Analysis of medical treatment of contracting residents after the reform of family doctors based on the big data of "Community Cloud Management" in Shanghai[J]. Modern Hospital Management, 2021, 19(2): 6-8.
[18]	季晖. "1+1+1"医疗机构组合签约影响因素研究[J]. 中华全科医学, 2021, 19(3): 436-440. Ji H. Study on the influencing factors of "1+1+1" type of contracted family doctor services[J]. Chinese Journal of General Practice, 2021, 19(3): 436-440.
[19]	唐雷, 钟世镇. 人工智能及其在医学领域中的应用[J]. 中国实用妇科与产科杂志, 2024, 40(9): 876-878. Tang L, Zhong S Z. Artificial intelligence and its applications in the field of medicine[J]. Chinese Journal of Practical Gynecology and Obstetrics, 2024, 40(9): 876-878.
[20]	张驰, 郭媛, 黎明. 人工神经网络模型发展及应用综述[J]. 计算机工程与应用, 2021, 57(11): 57-69. Zhang C, Guo Y, Li M. Review of development and application of artificial neural network models[J]. Computer Engineering and Applications, 2021, 57(11): 57-69.