Statistical Methods to Correct for Proxy Bias in Studies of Older Adults

描述（由申请人提供）：本提案有两个目标：1)将生物统计学家Shardell博士培养成一个独立的研究者，在生物统计学和老年学的界面上进行研究；2)在老年髋部骨折患者的流行病学研究中，开发更适当地使用代理受访者（如亲属或照护者）数据所需的统计方法。对于后一个目标，髋部骨折患者流行病学研究的一个重要问题是选择偏倚，因为最虚弱和认知受损患者的大量数据缺失，这可能导致结果偏倚和研究结论不准确。目前，为了最大限度地减少由于缺少数据而导致的选择偏差，采用代理来代替不能或不愿回答访谈问题的患者。然而，代理的反应可能存在系统性偏见。这种偏差是一个重要的问题，因为它意味着将缺失的患者数据与代理的反应相关联的标准统计方法可能导致不准确的研究结论。因此，这些指标的偏倚会妨碍研究者准确识别有希望的干预目标，从而改善骨折后患者的预后。为了解决这个问题，最初设计用于调整缺失数据的选择偏差的统计方法将扩展到包括代理数据。可用的统计方法可以纠正代理偏差，有助于为骨折后恢复设计精确的针对性干预措施。这些方法的计算机程序将在网上提供，供老年学社区使用。新方法将作为未来R01提案的一部分进行验证和比较。该提案中描述的为期三年的指导研究计划包括一个职业发展计划，其中包括生物学、社会学和老龄化的社会心理方面的课程；定期与导师会面；参加研讨会、讲习班和专业会议；通过跟随临床研究人员接触临床老年病学环境。迫切需要具有生物统计学和老年学专业知识的训练有素的研究人员来解决与老龄化研究相关的研究设计，结果测量和统计方面的问题。

DESCRIPTION (provided by applicant): This proposal has two goals: 1) to develop Dr. Shardell, a biostatistician, into an independent investigator performing research at the interface of biostatistics and gerontology; and 2) to develop statistical methods needed to more appropriately use data from proxy respondents (e.g., relatives or care givers) in epidemiological studies of elderly hip-fracture patients. Regarding the latter goal, a significant problem in epidemiologic studies of hip fracture patients is selection bias due to the large amount of data missing from the most frail and cognitively impaired patients, which may lead to biased results and inaccurate study conclusions. Currently, in order to minimize selection bias due to missing data, proxies are recruited to supply responses in place of patients who are unable or unwilling to respond to interview questions. However, responses from proxies may be systematically biased. This bias is a significant problem because it implies that the standard statistical approach of imputing missing patient data with responses from proxies can lead to inaccurate study conclusions. Therefore, bias from these proxies can impede investigators' ability to accurately identify promising targets of intervention that may improve patients' post-fracture prognosis. To solve this problem, statistical methods originally designed to adjust for selection bias from missing data will be extended to include proxy data. Availability of statistical methods that can correct for proxy bias can help in designing accurately-targeted interventions for postfracture recovery. Computer programs for the methods will be made available online for use by the gerontology community. The new approaches will be validated and compared as part of a future R01 proposal. The three-year mentored research program described in this proposal involves a career development plan that includes coursework in the biology, sociology, and psychosocial aspects of aging; regular meetings with mentors; participation at seminars, workshops, and professional conferences; and exposure to clinical geriatrics settings by shadowing clinician-researchers. Trained investigators with expertise in both biostatistics and gerontology are urgently needed to solve problems in study design, outcome measurement, and statistics that are relevant to aging research. PUBLIC HEALTH RELEVANCE: Accurately identifying behaviors and other targets of intervention that can improve physical and emotional functioning and protect against age-related diseases is the primary goal of gerontological research. Effective policy for care of older adults requires the ability to determine factors that can improve prognosis. Current statistical methods used to inform these decisions rely on imputing missing patient data with proxy data, methods which are known to lead to biased results. Improved statistical methods will allow correction for proxy bias to improve the accuracy of results, leading to more accurate conclusions regarding the effectiveness of interventions on health outcomes.