Estimating BrAC/BAC from Transdermal Alcohol: Combining First-Principles Physiological Models with Machine-Learning to Create Software to Optimally Process and Quantitatively Interpret Biosensor Data

摘要 经皮酒精生物传感器为不引人注目地连续收集酒精水平提供了一种有前途的方法 在自然环境中长期存在。现在有设备可以可靠地测量经皮 酒精浓度（TAC），通过皮肤扩散的酒精量，但经常被忽视， 使这些生物传感器有价值的关键问题是，TAC并不总是与生物传感器的易变性相关。 个体之间呼吸和血液酒精浓度（BrAC/BAC）的可解释测量， 环境条件和设备。本研究的目标是制作软件，将TAC数据转换为 BrAC/BAC（eBrAC/eBAC）。我们将通过以下方式实现这一目标：1）开发数学模型， 从TAC数据中产生定量的eBrAC，2）研究校准这些模型的替代方案，3） 使用不同类型和数量的非常精确的数据来测试模型拟合，以及4）将模型打包成 综合资料同化软件程序。具体而言，我们将通过以下方式提高模型的逼真度： 将先进的物理/生理模型与统计方法和数据驱动的机器相结合， 学习技巧为了减轻目前为每个人校准模型的负担，我们将 测试一些校准程序，包括更换实验室酒精管理 会议与更多不同的饮用协议，以及与人口为基础的参数估计。我们将测试 我们的模型和协议使用详细的消费数据收集1）在两个调查人员，2）在40 每个参与者将参加四个受控的实验室饮酒会议，以及3）40名参与者 他们将分别参加田间试验和实验室会议。我们将研究不同饮酒量的模型拟合 当使用不同量的个体化酒精数据时的模式（例如，呼吸分析仪、饮料日记）， 校准模型，并且在具有不同特征的个体内和个体之间（例如，性别、体重）和 在可变条件下（例如，湿度、心率），这可能会影响模型拟合。我们将创建一个数据同化 软件系统，BrAC Estimate软件，它结合了所有可用的数据， 准确的EBRAC测量。软件输出将包括饮酒事件的识别、连续的 eBrAC信号和eBrAC汇总评分（例如，eBrAC峰值、eBrAC峰值时间、饮水下面积 曲线）与置信带。该软件将是平台可移植的，可以单独运行或集成到其他 移动的医疗技术或精确医疗协议。这项提案在技术上是创新的， 先进，可行，并将导致第一个工具来完成TAC-eBrAC转换，最后 使得有可能获得自然酒精消费的可解释的定量测量， 领域这项研究的预期结果是扩大了TAC生物传感器对研究人员，临床医生， 和个人监测自然酒精消费和容易理解的结果。

Abstract Transdermal alcohol biosensors offer a promising method for unobtrusively collecting continuous alcohol levels in naturalistic settings over long periods of time. Devices are now available to reliably measure transdermal alcohol concentration (TAC), the amount of alcohol diffusing through the skin, but an often overlooked yet critical issue for making these biosensors valuable is that TAC does not consistently correlate with the easily interpretable measures of breath and blood alcohol concentrations (BrAC/BAC) across individuals, environmental conditions, and devices. The goal of this study is to produce software to convert TAC data into estimates of BrAC/BAC (eBrAC/eBAC). We will meet this goal by 1) developing mathematical models to produce quantitative eBrAC from TAC data, 2) examining alternative options for calibrating these models, 3) testing the model fits using varied types and amounts of very precise data, and 4) packaging the models into a comprehensive data assimilation software program. Specifically, we will enhance the fidelity of the models by integrating advanced physics/physiological-based models with statistical methods and data-driven machine- learning techniques. To reduce the burden currently required to calibrate the models for each individual, we will test a number of calibration procedures, including the replacement of the laboratory alcohol administration session with more varied drinking protocols as well as with population-based parameter estimates. We will test our models and protocols using detailed consumption data collected 1) on two of the investigators, 2) on 40 participants who will each participate in four controlled laboratory drinking sessions, and 3) on 40 participants who will each participate in a field trial and laboratory sessions. We will examine model fits across drinking patterns when using varying amounts of individualized alcohol data (e.g., breath analyzer, drink diary) to calibrate the models, and within and across individuals with differing characteristics (e.g., gender, weight) and under variable conditions (e.g., humidity, heart rate) that may affect model fit. We will create a data assimilation software system, the BrAC Estimator software, that incorporates all available data to produce the most accurate eBrAC measures. The software output will include the identification of drinking episodes, continuous eBrAC signal, and eBrAC summary scores (e.g., peak eBrAC, time of peak eBrAC, area under the drinking curve) with confidence bands. The software will be platform-portable to run alone or to be integrated into other mobile health technologies or precision medicine protocols. This proposal is innovative, technologically sophisticated, and feasible, and would result in the first tool to accomplish the TAC-eBrAC conversion, finally making it possible to obtain interpretable quantitative measurement of naturalistic alcohol consumption in the field. The anticipated result of this study is the expanded utility of TAC biosensors for researchers, clinicians, and individuals to monitor naturalistic alcohol consumption and easily understand the results.