Ethical and Human Factors Impacting Successful Translation of Perceptual Computing to Improve Clinical Care

项目总结 感知计算(PC)与人工智能和机器学习(AI/ML)相结合，有望 革命性的临床诊断、个性化治疗(精准医学)、症状和出院方法 涉及各种疾病的监测、远程医疗/移动医疗和初级预防。PC工具 迅速扩大但尚未解决的道德和实践挑战阻碍了负责任的翻译 变成了临床护理。这些挑战源于PC指标的特定、新颖特征，这些特征不同于传统的 情绪和社会行为的传统测量，因为它们1)代表客观观察到的，而不是亚 客观上引出的状态，可能涉及收集患者可能不知道或希望的数字数据 与他们的临床医生共享；2)使用数字设备被动收集数据，这些设备可以观察和记录瞬间到 瞬间的情绪和行为信息；3)产生难以处理的海量材料(即“大数据” 在个人层面上扩展为可操作的信息；以及4)依赖于易于收集和做出推断的数据 来自，邀请商业和其他实体的参与，他们的目标可能是利润驱动的，而不是固定的- 托儿所，就像医疗保健一样。为了帮助实现这项具有广泛临床影响的技术的潜力， 本研究的目标是确定和预测收益和关注事项(目标1)；确定这些关注事项的优先顺序 并评估风险/收益权衡(目标2)；以及评估将PC整合到临床护理中的影响(目标3)。在……里面 目标1，我们将与不同的利益相关者(PC工具的研究人员/开发人员)进行深入访谈 改善医疗保健；跨医学专科的临床医生；患者；和护理人员)以确定高优先级 解释PC结果并将其整合到临床护理中的关注事项和信息需求。访谈结果 将形成由目标2中的专家利益相关者使用3阶段修改的Delphi进行评估的内容。在 第一轮，我们将进行一项调查，需要进行多标准决策分析，以确定重点并确定优先顺序 专家利益相关者的好处和潜在危害。代表性参与者的子集(统计 )将被邀请在随后的两轮中召开会议，每一轮都涉及一次决定会议以进行审查 MCDA结果，并产生可行的解决方案和政策指导方针。在AIM 3中，我们将与Re- 开发多模式PC工具(NIH R01MH125958)的搜索者向心理健康临床医生呈现视频 以及患者摄入量的录音，除了标准摄入量测量外，还包括PC观察。CLI- 多个地点的医生(与NIH R01MH125958无关)将被要求提供他们最好的临床 使用从音频/视频数据导出的PC度量呈现之前和之后的估计，并评估 他们在认知访谈中的可解释性、相关性、适当性和可接受性。Trian- 这些目标的汇总结果将有助于对不同利益相关者需要了解的内容提供具体的见解 以便理解PC指标并将其转化为可操作的临床知识，并将有助于NCAT的 旨在确保将PC指标转化为临床护理的成功和可预测的影响。

PROJECT SUMMARY Perceptual computing (PC), in combination with artificial intelligence and machine learning (AI/ML), is poised to revolutionize clinical approaches to diagnosis, personalized treatment (precision medicine), symptom and out- come monitoring, telemedicine/mobile health, and primary prevention across a wide range of disorders. PC tools are rapidly expanding but unresolved ethical and practical challenges stand in the way of responsible translation into clinical care. These challenges stem from the specific, novel features of PC metrics, which differ from tradi- tional measures of emotional and social behavior in that they 1) represent objectively observed rather than sub- jectively elicited states and may involve collection of digital data that patients may not be aware of or wish to share with their clinicians; 2) collect data passively using digital devices that observe and register moment-to- moment emotional and behavioral information; 3) yield voluminous material (i.e. “big data”) that is difficult to scale into actionable information at the individual level; and 4) rest on data easy to collect and make inferences from, inviting engagement from commercial and other entities whose goals may be profit-driven rather than fidu- ciary, as in healthcare. To help realize the potential of this technology with widespread clinical impacts, the objective of this research is to identify and anticipate benefits and concerns (Aim 1); prioritize these concerns and assess risk/benefit tradeoffs (Aim 2); and evaluate impacts of integrating PC into clinical care (Aim 3). In Aim 1, we will conduct in-depth interviews with diverse stakeholders (researcher/developers of PC tools intended to improve healthcare; clinicians across medical specialties; patients; and caregivers) to identify high priority concerns and information needs for interpreting and integrating PC findings into clinical care. Interview findings will form the content to be evaluated by expert stakeholders in Aim 2 using a 3-phase modified Delphi. In the first round, we will conduct a survey entailing Multi-criteria Decision Analysis to confirm and prioritize salient benefits and potential harms among expert stakeholders. A subset of representative participants (statistically defined) will be invited to convene in two subsequent rounds, each involving a Decision Conference to review MCDA results and generate actionable solutions and policy guidelines. In Aim 3, we will collaborate with re- searchers developing a multimodal PC tool (NIH R01MH125958) to present mental health clinicians with video and audio recordings of patient intakes involving PC observations in addition to standard intake measures. Cli- nicians across multiple sites (unaffiliated with NIH R01MH125958) will be asked to provide their best clinical estimates before and after being presented with PC metrics derived from the audio/video data, and to evaluate their interpretability, relevance, appropriateness, and acceptability in the context of a cognitive interview. Trian- gulated results from these aims will contribute concrete insights into what diverse stakeholders need to know in order to understand and translate PC metrics into actionable clinical knowledge and will contribute to NCAT’s aims by ensuring the success and predictable impacts of translating PC metrics into clinical care.