Developing a Data-Driven Management System Using Machine Learning and Mixed-Methods Research to Predict Job Turnover Among MentalHealth Professionals

项目总结/摘要 本研究的主要目标是建立一个数据驱动的、以证据为基础的组织管理系统， 可以为有效的招聘和保留战略提供信息，以防止过度流动。高更替率 （估计每年25-60%）对精神卫生保健系统是毁灭性的，影响到组织（例如， 成本），雇员（例如，工作福利），最重要的是护理质量。人力资源部门 收集广泛的员工数据，这些数据可以作为预测人员流动的有用指标，但这些数据并不经常 分析以解决心理健康组织中的人员流动问题。计算方法大大 现在，它可以访问和分析大型和复杂的数据。这项试点研究将实现三个具体目标 目标1：通过开发和应用机器学习来构建和测试离职预测模型 目标2：生成关键问题，以加强人员流动预测 目标3：测试增强模型在预测12个月更替率方面的作用。在Aim中 1，使用过去的人力资源数据和服务接触[两个精神卫生组织（农村和 市区）]，我们将开发机器学习算法来预测营业额。算法将解决 周转问题同时（例如，谁最有可能离开？哪些因素预测营业额 在就业的不同时间点？）。在目标2中，我们将采访关键的线人：“离职者”（ 在研究期间自愿终止雇用）;“留下来的人”（在研究期间寿命极长的雇员）; 组织）;和“预测者”（根据我们的算法确定为可能离开）。其成果将 在两个焦点小组中进行讨论，以生成、完善和验证5-10个关键问题， 预测营业额。在目标3中，我们将对所有现有员工进行在线调查，以评估5-10 关键问题，并将调查数据与人力资源和服务部门的数据联系起来， 基于理论的模型（调查中的预测因素）和数据驱动的模型（机器）之间的精度 学习算法）预测12个月后的实际营业额。机器学习可以模拟复杂的， 动态变量关系（例如，处理大量变量，考虑异质性）， 克服了传统离职率研究的局限性，这种研究往往依赖于小规模的、横截面的研究， 方便的样品。本研究的成功完成将促进数据驱动、循证 组织管理实践，以解决营业额，这是符合NIMH的优先事项，资本化 基于现有数据结构并利用技术提高心理健康服务质量。这项研究将是一个 开发高度适应性机器学习算法以预测营业额的关键一步;最终，我们 设想这一系统将与未来的临床干预措施合作，以减少心理健康方面的人员流动。

Project Summary/Abstract The main goal of this study is to build a data-driven, evidence-based organizational management system that can inform effective recruitment and retention strategies to prevent excessive turnover. High turnover rates (estimated 25-60% annually) are devastating for mental health care systems, affecting organizations (e.g., cost), employees (e.g., work well-being), and most critically, the quality of care. Human resource departments collect extensive employee data that can be useful predictors for turnover, but these data are not often analyzed to address turnover issues in mental health organizations. Computational methods have greatly evolved and can now access and analyze large and complex data. This pilot study will achieve three specific aims: Aim 1: build and test turnover prediction models by developing and applying machine learning algorithms to existing human resource data; Aim 2: generate critical questions to enhance turnover prediction through qualitative methods; and Aim 3: test the enhanced model in predicting turnover at 12 months. In Aim 1, using past human resource data and service encounters from [two mental health organizations (rural and urban locations)], we will develop machine learning algorithms to predict turnover. The algorithms will address turnover questions simultaneously (e.g., Who are the most likely to leave? What factors predict turnover at varying time points in employment?). In Aim 2, we will interview key informants: “leavers” (employees who voluntarily terminate employment during the study); “stayers” (employees with extreme longevity in the organization); and “predictees” (identified as likely to leave, based on our algorithms). The findings will be discussed in two focus groups in order to generate, refine, and validate 5-10 critical questions to enhance prediction of turnover. In Aim 3, we will conduct an on-line survey of all current employees to assess the 5-10 critical questions and link survey data with data from human resources and services to examine the improved precision between the theory-based model (predictors in the survey) and the data-driven model (machine learning algorithms) in predicting actual turnover 12 months later. Machine learning can model complex and dynamic variable relationships (e.g., handling a large number of variables, accounting for heterogeneity) and overcome limitations in traditional turnover research that often relies on small, cross-sectional, and convenience samples. Successful completion of this study will promote data-driven, evidence-based organizational management practices to address turnover, which is aligned with NIMH priorities of capitalizing on existing data structures and using technologies to improve mental health service quality. This study will be a critical step in developing highly adaptable machine learning algorithms to predict turnover; ultimately, we envision that this system will be partnered with future clinical interventions to reduce turnover in mental health.