Understanding frailty, multimorbidity and renal failure in clinical trials: Attrition, retention and heterogeneity of treatment effects in trials for diabetes, cancer and a heterogenous set of index conditions

Clinical trials are the key method for determining treatment effects, but are known to be unrepresentative with respect to a range of important clinical characteristics such as age, kidney function, race/ethnicity, multimorbidity and frailty. Declining kidney function is associated both with increasing age and multimorbidity and has important additional implications for drug selection and dosing which are not adequately considered in the clinical trial setting.

We have previously made a number of important observations regarding trials and representativeness. First, multimorbidity and frailty are present (albeit underrepresented) in clinical trials and both predict serious adverse event rates (Hanlon 2019, Hanlon 2021 and Hanlon 2022). Secondly, morbidity count (a metric of multimorbidity) is not associated with heterogeneity in treatment effects (File attachments), but is associated with increased rates of trial attrition (failure for any reason to complete the final trial visit, including intentional and non-intentional withdrawals) (Lees 2022). Finally, clinical trials can be calibrated to real-world populations in order to improve the representativeness of trial findings (Butterly 2022).

However, while these findings are novel, they are difficult to operationalise. We intend to take these findings in three directions. First, we wish to better understand the relationship between frailty and kidney function and trial outcomes and characteristics. Secondly, we wish to determine whether it is practical to increase trial representativeness by recruiting more individuals with frailty, multimorbidity and impaired kidney function, or whether this would simply lead to greater trial attrition. Thirdly, we wish to determine whether calibrating trials to target populations modifies treatment effects, and whether the magnitude of this differs depending on the level of frailty, multimorbidity and kidney function within each trial.

We propose addressing these questions using two specific exemplars: 1. Novel antidiabetics for type 2 diabetes (see this protocol document for a description of the trials); 2. Cancer drugs for 4 common cancer sites: kidney/renal tract, colorectal, lung and melanoma. Where feasible (e.g. for the attrition risk prediction scores (the probability that an individual will experience trial attrition)) we will also validate the findings against the heterogeneous set of trials (21 index conditions across general medicine) which we have already extensively studied and for which we have considerable R code within the Vivli repository.

临床试验是确定治疗效果的核心方法，但众所周知，其在诸多重要临床特征方面缺乏代表性，包括年龄、肾功能、种族/民族、共病（multimorbidity）与衰弱（frailty）。肾功能减退与年龄增长和共病均存在关联，且对药物选择与给药剂量具有重要的额外影响，但这些影响在临床试验场景中并未得到充分考量。

此前我们针对临床试验及其代表性开展了多项重要研究观察。其一，共病与衰弱在临床试验中确实存在（尽管代表性不足），且二者均可预测严重不良事件发生率（Hanlon 2019、Hanlon 2021及Hanlon 2022）。其二，共病计数（共病的量化指标）与治疗效应异质性无相关性（详见附件文件），但与试验脱落（trial attrition）率升高相关——试验脱落指因任何原因无法完成最终临床试验访视的情况，涵盖主动与非主动退出（Lees 2022）。其三，可针对真实世界人群校准临床试验，以提升试验结果的代表性（Butterly 2022）。

然而，尽管上述研究发现具有创新性，但难以付诸实践。我们计划从三个方向推进这些研究成果的落地应用：其一，进一步明确衰弱、肾功能与临床试验结局及特征之间的关联；其二，旨在明确通过招募更多衰弱、共病及肾功能受损的受试者以提升试验代表性是否具有可行性，抑或此举仅会加剧试验脱落；其三，希望明确针对目标人群校准临床试验是否会改变治疗效应，且该效应的强弱是否会因各试验内部的衰弱、共病及肾功能水平而异。

我们拟通过两个具体的范例来解答上述问题：1. 用于2型糖尿病的新型抗糖尿病药物（详见本方案文件了解相关临床试验的描述）；2. 针对4种常见癌种的抗肿瘤药物：肾/泌尿道癌、结直肠癌、肺癌及黑色素瘤。在可行的情况下（例如针对脱落风险预测评分——即受试者发生试验脱落的概率），我们还将利用此前已开展广泛研究的异质性临床试验集（涵盖全科医学领域的21种指征）对研究发现进行验证，且我们已在Vivli库中存储了针对该数据集的大量R代码。