【作者】 陈楠；

【导师】 潘杰；

【作者基本信息】 四川大学 ， 流行病与卫生统计学， 2023， 博士

【摘要】 背景:作为完成联合国“千年发展目标”中妇幼健康幅度提升最快最显著的国家之一,中国的孕产妇死亡率由1990年的10万分之94.7下降到2020年的10万分之16.9。数据显示,中国的妇幼健康水平已达到中高收入国家水平,基本完成了“生存”目标,正在向“繁荣”目标迈进。中国目前处于较低死亡率和较低生育率的阶段。根据其他国家的经验判断,生育率的下降可能导致助产机构数量的减少,区位分布集中在少数区位点造成服务区位集中化。区位集中化很可能损害资源可及性,使得住院分娩服务利用不足的地区利用情况进一步恶化。同时,分娩服务集中在幸存助产机构中造成服务规模的集中化。有研究结果显示,适度的规模集中化有提高医疗服务质量的潜力。随着生育政策的调整,我国近年来高龄产妇、经产妇等占比增高,分娩风险升高,对助产服务质量提出了更高要求。集中化从资源配置角度为提高医疗质量提供了新思路。我国此前的研究多关注分散化配置,对集中化研究少,认识不充分。本研究在分娩服务集中化的背景下,探索应如何配置助产资源,达到尽可能降低集中化对可及性的负面影响,同时提高医疗质量,改善健康结局的目的。人民群众健康水平的提高是资源配置优化的最终目标。本研究首先结合三阶段延迟模型(three delays model)和Andersen卫生服务利用模型建立理论框架。通过理论框架的建立,研究梳理了资源空间配置影响健康结局的两条可能路径,结合助产资源在我国的研究现状和发展趋势,本研究将重点放在了资源空间配置对于第一阶段和第三阶段延迟的影响上,并提出两个研究假设。针对第一阶段的延迟提出假设1:孕产妇选择住院分娩的意愿随着孕产妇到助产机构距离的增加而下降。针对第三阶段的延迟提出假设2:助产机构分娩量的增加有助于改善孕产妇分娩健康结局。在理论框架的基础上进行数据实证分析。本研究使用的数据来自多个方面,共纳入了:包含医疗卫生机构基本信息和服务能力的医疗机构年报数据;包含孕产妇个人基本信息和住院服务利用信息的病案首页数据;住院分娩率、经济发展水平、人口教育水平的区县数据;出生率的市州数据;包含人口点坐标和估计人数的World Pop人口栅格数据;地理坐标数据;以及包含道路交通信息的路网数据。研究首先对数据进行了清理、筛选。从2016年到2020年,共计1,042,878条分娩出院病例、73,208个人口点(人口栅格数据)、5,546个助产机构年(其中不重复的助产机构有1,516家)、以及915个区县年(每年183个区县)被纳入分析。分娩服务利用指标选用了最具代表性的住院分娩率。孕产妇的健康结局指标选用了与孕产妇死亡率相关的孕产妇危重事件发生率(severe maternal morbidity,SMM)。信效度评价结果显示,SMM具有较高的重测信度(五年ICC为0.937)和较好的区分效度(在年龄、住院时长、总费用中差异明显),可以用于孕产妇健康结局的评估和监测。为实现研究目的,研究内容分为以下四部分。研究内容一:助产资源配置现状的精准测度。精准的描述资源配置的现状是配置优化的基础。方法:使用供需比法、最短路径法和clm FCA(Conditional Logit Model with two-step Floating Catchment Area)三种方法对资源的空间可及性进行精准测度。基于行政区划的供需比法是该类研究中最常见的方法。该方法尺度粗略,测量结果依赖于区域边界的划分并忽略了边界跨区域就医的问题。相比而言,最短路径法考虑了边界问题,更加灵活不受边界区划的限制,但该方法仅考虑了空间距离因素并未考虑供需双方的规模和特征。本研究除使用以上两种常用方法外,创新性的使用结合了两种方法优点的空间可及性测度方法,得到调整了孕产妇就医偏好的人口点尺度的供需比。该方法使用离散选择模型(Discrete Choice Model,DCM)中的条件logit模型(Conditional Logit Model)对孕产妇助产机构选择的倾向性进行了刻画,然后结合基于重力模型的两步移动搜寻法(Two-step Floating Catchment Area)在人口点尺度进行计算,得到嵌入了孕产妇偏好的调整后供需比。结果:需方结果显示分娩需求从2017年开始显著下降。供方结果显示,从2017年开始助产机构数量逐年减少,其中三级医院占比从11.7%增加到24.1%、城市地区助产机构占比从41.7%增长到45.4%。助产资源呈现往高级别、城市地区集中化的趋势。区县尺度供需比呈上升趋势,提示资源没有随着需求减少而调整,资源配置存在可优化的空间。最短路径法的车程时间距离结果显示,五年间人口点到助产机构的最短车程时间距离中位数从52.812分钟增长到57.272分钟,空间可及性下降。分布图显示东部可及性优于西部。离散选择模型中条件logit回归显示每增加1分钟车程时间与选择可能性平均降低6.6%左右有关,选择私立非营利性助产机构的可能性比公立非营利助产机构少30%,增加一个产科床位与平均增加1.1%的选择可能性相关。嵌入了孕产妇选择偏好结果显示,人口点尺度调整后供需比数值从2017年的0.015上升到2020年的0.020。调整后供需比与调整前供需比相比增加了人口点尺度的可及性结果;调整后供需比的区县中位数上无明显差异,同样呈上升趋势;调整后供需比数值分布在中间区间段的区县占比较多,最小和最大区间段占比较少;调整后供需比东西部差异更加明显。结论:三种方法结果均提示在资源集中化的趋势下,资源的空间可及性下降。由于供给减少速度小于需求,供需比呈现上升趋势,说明资源配置存在可优化的空间。偏好调整后供需比相比前两种方法可提供更精准测度的空间可及性结果。研究内容二:区位集中化对于分娩服务利用的影响。在分娩服务区位集中化对住院分娩服务利用影响的探索中,研究聚焦于以距离为代表的空间可及性对住院分娩率的影响。该部分研究基于因果推断理论和方法,在区县级别验证研究假设1是否成立,即距离是否会影响住院分娩服务的利用。方法:首先对所有可能影响距离对住院分娩服务的因素进行文献综述,在此基础上使用DAGitty绘制有向无环图(Directed Acyclic Graphs,DAGs)。该步目的在于将所有潜在的混杂因素及其之间的关系可视化,找出混杂因素的最小校正子集。在建立回归模型时,使用了区县内人口点距离助产机构最短距离的最大值、两小时车程内覆盖率两种指标代表空间可及性的距离,选择了区县的住院分娩率来测量服务利用。由于数据存在嵌套层次结构,研究采用多水平回归模型分析。结果:DAG模型结果显示,在探讨距离对住院分娩率的作用时,需要校正孕产妇居住地的特征(包括城乡分类、教育水平、经济因素、民族地区分类)来减少偏倚。多水平回归模型显示,人口点距离助产机构最短车程时间在区县内的最大值越低,助产机构周围两小时车程内分娩量覆盖率越高,该区县住院分娩率越高。两种空间可及性指标相比,分娩量覆盖率对于住院分娩率的影响更大。在校正人均GDP的情况下,覆盖率回归系数估计值为13.052,P<0.001。同时通过在模型中增加覆盖率与人均GDP的交互项可知,覆盖率对于住院分娩率的影响随着人均GDP的增加而减小(交互项回归系数估计值-0.576,P<0.001)。随机斜率模型显示区县间住院分娩率的差异在覆盖率较低的区县更大。固定效应模型和分位数回归的敏感性分析结果提示分析结果稳健。结论:覆盖率的提升有助于住院分娩服务利用的提升,该效应在不同区县具有异质性。研究结果提示在助产机构减少的区位集中化趋势下,优化助产资源配置时需要保证资源的空间可及性来减少集中化对服务利用的不利影响。研究内容三:规模集中化对于健康结局的影响。针对分娩服务规模集中化对孕产妇健康结局作用进行探索。该部分研究重点关注助产机构分娩量和孕产妇高危事件发生率的因果关系,即假设2的验证。方法:采用了多元回归模型和工具变量法,验证是否存在因果效应,然后使用分段回归模型对效应阈值进行估计。选用计量经济学中在观察性研究因果推断中广泛使用的工具变量法来解决助产机构分娩量可能存在的内生性问题。本研究选用了助产机构车程两小时覆盖范围内平均分娩需求量、基于条件logit回归模型的助产机构分娩量的估计值两种指标作为工具变量。结果:工具变量法结果显示助产机构分娩量的增加有助于SMM率的降低(平均边际效应值为-0.013,P<0.001)。在此基础上,分段回归模型结果显示,当助产机构一个季度的分娩量超过118之后,分娩量的增加对健康结局的改善效果有限。结论:研究结果显示服务规模提升对于健康结局有提升效应,可推测在规模集中化的情况下,医疗质量有提升潜力。阈值结果为优化后医疗质量提升效果的评估提供了参考值。研究内容四:助产资源配置的优化。方法:在进行了资源配置现状描述并明确了配置对健康结局作用路径的基础上,本研究使用最大覆盖问题模型和遗传算法对助产机构的区位进行了优化。假设1的研究结果显示,提升助产机构两小时车程内分娩需求覆盖率对服务利用的影响较大。因此在区位配置优化时选择了最符合研究目的的最大覆盖模型。研究构建了两种优化情境,情境1中的助产机构数目与显示数据中2020年助产机构数目一致均为951家。优化情境2中助产机构数目依据往年减少速率推算而得,为884家。区位优化时将2016年到2020年所有提供助产服务的区位点作为候选列表,以最大化覆盖分娩量为目标函数,运用启发式的遗传算法进行求解。在区位优化的基础上,研究运用改进区位配置模型结合遗传算法对产科床位数进行了优化。将“每千分娩量设置17张产科床位数”作为助产资源床位配置参考值,将最小化偏好调整后供需比与配置参考值之间的差异作为优化的目标函数进行了运算。结果:优化后的方案与2020年真实情况相比,不论是在情境1还是情景2中,助产机构两小时车程内的覆盖率均增大。尤其是在助产机构减少的情境2中,覆盖率增加了0.4个百分点,对应住院分娩率的提升为5.2个百分点。助产机构的分娩量模拟结果显示,助产机构分娩量中位数有明显提升,中高分娩量机构占比增多。在助产机构数量更少的情境2中,分娩量中位数提高了42个单位,对应0.05个百分点SMM率的降低。同时,分娩量达到118的机构占比提高了6%左右。床位优化结果显示,在优化情境1和2中,优化后助产机构产科床位中位数显著增加,床位数在助产机构间的分布更加均匀分散,偏好调整后供需比有所提升。结论:区位优化可改善资源的空间可及性,减缓分娩服务集中化对可及性的不利影响。根据助产机构分娩量和孕产妇健康结局的关系推测,优化后整个分娩服务质量可能会有所提升。床位优化提升了资源配置的公平性,优化后床位资源分布更加均匀,更多的人口点或区县更接近配置标准。综上所述,本研究以助产资源为例,通过理论分析确定了研究框架,通过实证研究进行了因果效应探讨,并在此基础上运用区位配置模型提出了区位优化和床位优化的解决方案。研究补充了国内对于妇幼卫生资源空间配置研究的不足,也为其他医疗资源空间配置优化研究提供了可行的参考方案。更多还原

【Abstract】 Backgrounds: As one of the countries with the most significant progress in health among "Countdown countries",China’s maternal mortality has decreased from 94.7 per 100,000 live births in 1990 to 16.9 per 100,000 live births in 2020.The core maternal and child healthcare indicators data show that China has been ranked at the forefront levels among high-and middle-income countries,marking a shift from mere survival to thriving.Presently,China finds itself in the low mortality rate and low fertility rate stage,a testament to its successful health policies and initiatives.However,lessons from other nations suggest that declining fertility rates can precipitate a reduction in obstetric facilities,potentially limiting the availability of obstetric delivery services.This concentration of services might exacerbate underutilization issues,affecting accessibility.Conversely,service concentration in retained facilities may present an opportunity to enhance medical quality,as observed in previous research correlating increased service volume with improved quality.In the context of China’s evolving fertility policy over the past four decades,adjustments have resulted in a higher proportion of high-risk mothers,characterized by advanced age and previous births.This demographic shift underscores the increasing demand for high-quality obstetric services.While prior research has predominantly focused on the decentralization of obstetric resources to enhance accessibility,there exists a notable academic gap regarding the concentration of obstetric resources in China.This study seeks to address this gap by examining the rational optimization of obstetric resource allocation.The primary objective of this study is to identify a pathway for the rational optimization of obstetric resource allocation,ensuring both resource accessibility and the enhancement of service quality.The ultimate goal of resources allocation is to improve health outcomes.Firstly,this study combined the Three Delays Model and Andersen’s behavioral model to form the theoretical framework,which clarified the possible pathways through which resource spatial allocation influences health outcomes.In light of the current research landscape and development trends in China,this study focuses on addressing the first and third delays,which leads to the formulation of two research hypotheses.Hypothesis 1 is formulated to address the first delay,proposing that the willingness of pregnant women to choose in-hospital delivery decreases as the distance to the obstetric facility increases.Hypothesis 2 is proposed to address the third delay,suggesting that an increase in the number of deliveries performed by obstetric facilities contributes to improving maternal health outcome.Based on the theoretical framework,this study conducted data empirical analysis.The data used in this study was from a wide range of sources which included healthcare facility annual report data containing basic information and service capabilities of healthcare facilities,hospital discharge data containing personal information and inpatient service utilization information of pregnant women,inhospital delivery rates,economic development level,and population education level data at county level,city-level birth rate data,World Pop grid data containing population points coordinates and population number,geographical coordinate data and road network data containing road traffic information.The data was cleaned,screened,merged and matched.From 2016 to 2020,a total of 1,042,878 delivery discharge cases,73,208 population points for 3km grid data,5,546 facility-years with1,516 unique obstetric facilities,and 915 counties(183 counties per year)data were analyzed.The utilization indicator for obstetric services was the in-hospital delivery rate,and the maternal health outcome indicator was the severe maternal morbidity rate(SMM).SMM demonstrated high test-retest reliability(five-year ICC of 0.937)and discriminant validity(significant differences among different age groups,hospitalization duration,and total cost),making it suitable for obstetric quality assessment and monitoring.For achieving research goal,this sudy composite four sections.Section 1: Precisely Measuring Obstetric Resource Allocation.Accurate characterization of current resource allocation is fundamental to optimization.Methods: Three measurement methods are applied for precise measurement:the supply-demand ratio method,the shortest path method,and the conditional logit model-based Floating Catchment Area(clm FCA)method.The supply-demand ratio method is commonly used but suffers from a coarse measurement scale,high dependence on broad definitions,and an oversight of cross-boundary service utilization.The shortest path method addresses the boundary problem and offers greater flexibility across broadlines.However,its limitation lies in its sole consideration of distance,overlooking factors related to supply and demand scales.This study innovatively applies the conditional logit model-based Floating Catchment Area(clm FCA)method to characterize the spatial distribution of obstetric resources.This approach combines supply-demand ratios,travel time distance,and maternal facility preferences.It employs the conditional logit regression model from the discrete choice model to describe pregnant women’s preferences for choosing obstetric facilities.The two-step floating catchment area method is integrated to calculate preference-adjusted supply-demand ratio at the population point level.Results: County-level supply-demand ratio results indicated a decrease in both delivery demand and the number of obstetric facilities since 2017.The proportion of tertiary-level hospitals increased from 11.7% to 24.1%,and urban facilities proportion rose from 41.7% to 45.4%,signaling a concentration trend in obstetric services.Travel time distance results showed an increase in the median value of the shortest travel time from 52.8 to 57.3 minutes over five years.Conditional logit regression revealed that every additional minute of travel time was associated with an average 6.6% decrease in the likelihood of choosing an obstetric facility.The likelihood of choosing a private non-profit obstetric facility was 30% lower than that of a public non-profit facility.Additionally,an increase in obstetric beds was associated with a 1.1% higher likelihood of facility selection.Incorporating pregnant women’s preferences,the adjusted supply-demand ratio increased from 0.015 in 2017 to 0.020 in 2020,suggesting an improvement in obstetric resources accessibility.Using the government-recommended bed allocation standard(17 obstetric beds per1,000 deliveries)as a benchmark,the adjusted supply-demand ratios highlighted significant differences between western and eastern areas.However,unadjusted supply-demand ratios at the county level did not show significant differences between the east and west.Accurate population point level supply-demand ratios prove valuable in identifying resource allocation imbalances.Conclusions: The three measurements indicate a decrease in spatial accessibility of resources under the concentration trend.The rate of supply decrease is slower than that of demand,resulting in an increase in supply-demand ratios.This suggests the potential for optimizing obstetric resource allocation.The clm FCA method demonstrates its capability to provide accurate results for spatial accessibility.Section 2: The impact of location concentration on in-hospital delivery services utilization.In exploring the impact of location concentration on the utilization of inhospital delivery services,this section focuses on the influence of travel distance on the in-hospital delivery rate.Based on causal inference theory and methods,this section aims to verify Hypothesis 1 at the county level,which examines whether travel distance affects the utilization of in-hospital delivery services.Methods: First,a literature review was conducted on all potential factors that may influence the use of inpatient delivery services related to distance.Based on this review,Directed Acyclic Graphs(DAGs)were constructed using DAGitty to visualize all potential confounding factors and their relationships,identifying the minimal subset of confounding factors.In building regression models,two indicators representing spatial accessibility by distance were used: the maximum travel time from population points to obstetric facilities within counties and the coverage proportion of delivery demand within a two-hour drive within counties.County-level in-hospital delivery rates were selected to measure service utilization.Due to the nested hierarchical structure of the data,two-level regression models were employed for analysis.Results: DAG model results indicated that,when investigating the effect of distance on in-hospital delivery rates,it was necessary to adjust for characteristics of pregnant women’s residences(including urban-rural classification,education level,and economic development level,ethnical minority type)to reduce bias.The twolevel regression models showed that as the maximum travel time decreased and the coverage rate increased,the inpatient delivery rate in the county increased.Of the two spatial accessibility indicators,coverage rate had a greater impact on inpatient delivery rates.With the adjustment for per capita GDP,the estimated coefficient of coverage rate was 13.052(P<0.001),as shown in the regression analysis.Additionally,adding an interaction term between coverage rate and per capita GDP in the model revealed that the influence of coverage rate on in-hospital delivery rates decreased as per capita GDP increased(interaction term coefficient estimate-0.576,P<0.001).Sensitivity analysis of the fixed-effects models and quantile regression model confirmed the robustness of these results.Conclusions: The promotion of in-hospital delivery service utilization is evident with an increasing coverage rate,yet this effect exhibits heterogeneity.The findings underscore the significance of accessibility in the optimization process amid the trend of location concentration.Section 3: The impact of service scale concentration on health outcomes.Regarding the exploration of the impact of the concentration of obstetric delivery services on maternal health outcomes,this study employed multiple regression models and instrumental variable methods.This section focuses on the causal effect of the delivery volume and the maternal health outcomes,specifically testing Hypothesis 2.Methods: First,the existence of causal effects was verified,and then a segmented regression model was used to estimate effect threshold.The study adopted instrumental variable(IV)methods which was widely used in causal inference in observational studies in econometrics to address potential endogeneity issues.Two indicators were chosen as instrumental variables: the average delivery demand within a two-hour drive coverage area of obstetric facilities,and the estimates of the delivery volume based on the conditional logit regression model.Results: The instrumental variable(IV)results showed that an increase in the delivery volume of obstetric facilities contributed to a reduction of severe maternal morbidity(SMM)rates(average marginal effect is-0.013,P<0.001).Building upon this,the segmented regression model results indicated that the increase in delivery volume had limited effects on health outcomes once the volume exceeded 118 deliveries per quarter.Conclusions: Results shows the increasing of services scale can improve the health outcomes,implies the possibility of obstetric services quality improvement under the scale concentration trend.The result provide a reference for assessing the improvement in health outcomes resulting from the optimization of obstetric resources.Section 4: Optimization of obstetric resources.Methods: Based on the description of resource allocation and the validation of Hypotheses 1 and 2,this study proceeded to optimize the location of obstetric facilities using the Maximum Coverage Problem model and Genetic Algorithm.Hypothesis 1’s research results indicated that improving the delivery demand coverage proportion within a two-hour drive of facilities had a significant impact on service utilization.Therefore,among the three commonly used methods in locationallocation optimization,this chapter had chosen the maximum coverage model,which aligned most closely with the research objectives.Two optimization scenarios were constructed.In Scenario 1,the number of obstetric facilities remained the same as the data displayed in 2020,with a total of 951 facilities.In Scenario 2,the number of obstetric facilities was estimated based on the reduction rate,resulting in 884 facilities.All locations providing obstetric delivery services from 2016 to 2020 were considered as candidates for optimization.The optimization function goal was to maximize the coverage of delivery demand.The resolution was solved by using a heuristic Genetic Algorithm.Based on the location optimization,this study further employed a modified location-allocation model combined with a genetic algorithm to optimize the allocation of obstetric beds.The reference value for obstetric resource bed allocation was set as "17 obstetric beds per 1,000 deliveries".The objective function for optimization was to minimize the difference between the preference adjusted supply-demand ratio and the reference bed allocation value for each population point.Results: Compared to the actual situation in 2020,the optimized results showed an increase in the coverage proportions both in Scenario 1 and Scenario 2.Particularly,in Scenario 2 where obstetric facilities were reduced,the coverage rate had increased by 0.4 percentage points,corresponding to a 5.2 percentage point increase in in-hospital delivery rates.This indicates that location optimization can indeed improve the accessibility of resource and mitigate the adverse impact of concentration of obstetric services on accessibility.The simulation of delivery volumes showed a noticeable increase in the median value of delivery volume,with a higher proportion of facilities with medium to high delivery volume.In Scenario 2,where there are fewer obstetric facilities,the median delivery volume has increased by 42 units,corresponding to a 0.05 percentage point reduction in the SMM rates.Simultaneously,there has been an approximately 6% increase in the proportion of facilities with a delivery volume of 118 or more.The results showed that in both optimization scenarios(Scenario 1 and Scenario 2),the median number of obstetric beds in obstetric facilities significantly increased after optimization.Conclusions: Optimizing location can enhance spatial accessibility and alleviate negative outcomes associated with concentration.Based on the relationship between obstetric facility delivery volume and maternal health outcomes,it is inferred that the overall quality of delivery services may improve after optimization.Obstetric beds optimization findings shows that the distribution of beds among obstetric facilities became more even and dispersed after optimization,leading to an improvement in the adjusted supply-demand ratio.indicate that the optimization of resource allocation has significantly enhanced the allocation of obstetric beds.In summary,this study focused on obstetric resources and established a research framework through theoretical analysis.Causal relationships were explored through empirical research,and based on these findings,location-allocation models were applied to propose solutions for location and bed optimization.The research has addressed the deficiencies in domestic studies on spatial allocation of maternal and child health resources,and also offers feasible reference solutions for optimizing the spatial allocation of other medical resources.更多还原