Non-invasive seizure forecasting system using e-diaries, internal and external factors

这个职业发展奖将为我提供一个机会，让我发展成为一名 独立调查者利用先进的机器学习和大规模的临床队列来应用于癫痫。 研究项目的中心是使用电子日记(e-diary)非侵入性地预测癫痫发作的风险。 和生物传感器数据。目前尚不清楚非侵入性癫痫发作的预测是否足够准确，足以在临床上 实用程序。我之前的回顾工作表明，使用先进的机器学习算法来评估电子日记 数据(即内部因素)，对癫痫风险的预测比机会预测更准确。目前还不清楚是否 使用来自睡眠生物传感器、服药依从性、压力、天气的额外数据来增强这些预测 模式、压力和锻炼(即外部因素)会进一步提高准确性。初步数据 暗示这些额外的数据可能是有价值的。对于目标1，该项目将前瞻性地验证机器 仅使用电子日记预测癫痫患者队列中癫痫发作风险的学习算法(内部 因素)。这些预测将与利率匹配的随机预测进行比较，作为基准。对于目标2， 预测将使用可穿戴生物传感器、自动服药依从性以及 关于压力、荷尔蒙周期和天气(外部因素)的信息。这项研究的预期结果是 用非侵入性技术预测癫痫发作风险的有效方法，具有较高的准确性。此外, 该项目包括通过指导、在线课程和当地课程工作实现的教育目标 旨在为研究独立做准备。主要的教育目标是(1)培养学生的技能 先进的数据科学技术，(2)管理大型临床队列，(3)建立强大的 生物统计学/信息学基础，以及(4)专业发展。布兰登·韦斯特弗博士，其中一位 癫痫领域最顶尖的数据科学专家将担任该项目的主要导师。其他内容 导师将来自机器学习和计算机科学方面的专家孙继蒙博士，以及孙继蒙博士。 托马斯·特拉维森，一位生物统计学家和临床试验专家。我的目标是建立一个最先进的， 独立实验室专注于应用数据科学来降低癫痫的发病率和死亡率。

This career development award will provide me with an opportunity to develop the needed skills to become an independent investigator using advanced machine learning and large-scale clinical cohorts applied to epilepsy. The research project centers on forecasting the risk of seizures non-invasively using electronic diaries (e-diaries) and biosensor data. It is unknown if non-invasive seizure forecasting can be sufficiently accurate to have clinical utility. My prior retrospective work suggests that using advanced machine learning algorithms to evaluate e-diary data (i.e., internal factors), forecasts of seizure risk are more accurate than chance forecasts. It is unknown if enhancing these forecasts using additional data from sleep biosensors, medication adherence, stress, weather patterns, stress, and exercise (i.e., external factors) would improve the accuracy further. Preliminary data suggest that this additional data may be valuable. For Aim 1, this project will prospectively validate the machine learning algorithm to forecast seizure risk in a cohort of people with epilepsy using e-diaries alone (internal factors). The forecasts will be compared with a rate-matched random forecast as a baseline. For Aim 2, the forecasts will be enriched using data from a wearable biosensor, automated medication adherence, as well as information about stress, hormonal cycles and weather (external factors). The expected outcome of this study is a validated method with higher accuracy for forecasting seizure risk using non-invasive techniques. In addition, this project includes educational objectives through mentorship and online courses and local coursework designed to prepare for research independence. The main educational objectives are (1) developing skills in advanced data science techniques, (2) managing a large clinical cohort, (3) build a strong biostatistics/informatics foundation, and (4) professional development. Dr. Brandon Westover, one of the foremost data science experts in the field of epilepsy, will serve as the primary mentor for this project. Additional mentorship will come from Dr. Jimeng Sun, an expert in machine learning and computer science, as well as Dr. Thomas Travison, an expert biostatistician and clinical trialist. My goal is to establish a state-of-the-art, independent laboratory focused on data science applied to decrease morbidity and mortality from epilepsy.