Forecasting Migraine Attacks

项目摘要 对于每年经历偏头痛的数百万人来说，治疗通常包括 只有在经历了破坏性疼痛和继发症状后才反应性地治疗发作。因为个别 偏头痛发作对大多数患者来说是不可预测的，流产药物不能早期或有效地使用， 并且不能制定先发制人地停止发展攻击的策略。通过将日常风险形式化， 一旦发生攻击，个人将更好地准备使用现有的流产疗法，并减少痛苦 与任何一次袭击有关。我们的团队之前已经构建并测试了头痛预测-I HAPRED-I和HAPRED-II模型，它们是简单的偏头痛预测模型 这些都是基于日常压力的。尽管这些模型很有前途，但它们存在一些弱点， 使其无法广泛应用于临床。该项目的目标是评估一种新的预测模型， 提高了预测能力。为了实现这一点，已经向 现有模型，新模型的参数将使用贝叶斯估计不断更新。 在新的HAPRED-III模型（目标1）中，预报窗口从24小时缩短至12小时（时间） 已经添加了统计预测器，并且另外的预测器（例如，睡眠，情绪，药物使用，前驱期 症状和自我预测）将被测试以提高性能。为了让模型更容易 部署（目标2），将检查模型参数的预测因子。这些预测因素将更好地告知 先验概率的模型参数，并将减少需要收集数周或数月的数据， 每个人在做出可靠的预测之前。

Project Summary For the millions of individuals who experience migraine each year, treatment typically consists of reactively treating attacks only after experiencing disruptive pain and secondary symptoms. Because individual migraine attacks are unpredictable to most sufferers, abortive medications are not used early or effectively, and strategies to preemptively stop developing attacks cannot be formulated. By formalizing the daily risk for an attack, individuals will be better prepared to use existing abortive therapies and reduce the suffering associated with any single attack. Our team has previously built and tested the Headache Prediction-I (HAPRED-I) and Headache Prediction-II (HAPRED-II) models, which are simple migraine forecasting models that are based on daily stress. Despite their promise, these models exhibit several weaknesses that would prevent them from broad clinical use. The objective of this project is to evaluate a new forecasting model that has improved predictive power. To accomplish this, several important predictors have been added to the existing model, and the parameters of the new model will be continuously updated using Bayesian estimation. In the new HAPRED-III model (Aim 1), the forecasting window is reduced from 24 to 12 hours, temporal statistical predictors have been added, and additional predictors (e.g., sleep, mood, medication use, prodromal symptoms, and self-prediction) will be tested for improved performance. To allow the model to be more easily deployed (Aim 2), predictors of the model parameters will be examined. These predictors will better inform the prior probabilities of the model parameters and will reduce the need to collect weeks or months of data from each individual before generating reliable forecasts.