A Framework for Using Real-World Data and Health Outcomes Modeling to Evaluate Machine Learning–Based Risk Prediction Models

ObjectivesWe propose a framework of health outcomes modeling with dynamic decision making and real-world data (RWD) to evaluate the potential utility of novel risk prediction models in clinical practice. Lung transplant (LTx) referral decisions in cystic fibrosis offer a complex case study.MethodsWe used longitudinal RWD for a cohort of adults (n = 4247) from the Cystic Fibrosis Foundation Patient Registry to compare outcomes of an LTx referral policy based on machine learning (ML) mortality risk predictions to referral based on (1) forced expiratory volume in 1 second (FEV₁) alone and (2) heterogenous usual care (UC). We then developed a patient-level simulation model to project number of patients referred for LTx and 5-year survival, accounting for transplant availability, organ allocation policy, and heterogenous treatment effects.ResultsOnly 12% of patients (95% confidence interval 11%-13%) were referred for LTx over 5 years under UC, compared with 19% (18%-20%) under FEV₁ and 20% (19%-22%) under ML. Of 309 patients who died before LTx referral under UC, 31% (27%-36%) would have been referred under FEV₁ and 40% (35%-45%) would have been referred under ML. Given a fixed supply of organs, differences in referral time did not lead to significant differences in transplants, pretransplant or post-transplant deaths, or overall survival in 5 years.ConclusionsHealth outcomes modeling with RWD may help to identify novel ML risk prediction models with high potential real-world clinical utility and rule out further investment in models that are unlikely to offer meaningful real-world benefits.     

Validating Restricted Mean Survival Time Estimates From Reconstructed Kaplan-Meier Data Against Original Trial Individual Patient Data From Trials Conducted by the Canadian Cancer Trials Group

ObjectivesThe development of novel cancer therapies, including immuno-oncology agents, has increased interest in reconstructed individual patient data (IPD) based restricted mean survival time (RMST) analyses. Additionally, reconstructed IPD–based RMST is recommended in cost-effectiveness analyses when original trial IPD are not available. Nevertheless, recently concerns regarding potential bias of reconstructed-IPD RMST have been presented, because reconstructed-IPD RMSTs have not been validated and previous validation endpoints may not capture the entire Kaplan-Meier (KM) curve, especially the “tail.” Our study aims to validate the recommended method of IPD reconstruction by comparing reconstructed IPD– and original trial IPD–based RMST.MethodsCanadian Cancer Trials Group trials from 1990 to 2017 were included. Overall survival and progression-free survival IPD were reconstructed based on published KM curves using the Guyot method. Analysts were blinded to original trial IPD. RMST was calculated at 1 year and over the entire KM curve. Reconstructed-IPD and original trial–IPD (gold-standard) RMSTs were compared for accuracy and predictive error via mean deviation, mean absolute error (MAE), mean percentage bias, and Bland-Altman plots and across KM curve quality (vector traced or bitmapped).ResultsWe identified 39 trials. The mean deviation, MAE, and mean percentage bias of RMST between the reconstructed IPD and original trial IPD were small. In particular, the mean deviation was ?0.01 months and ?0.04 months, MAE was 0.19 months and 0.24 months, and mean percentage bias was 0.82% and 0.84% in overall survival KM curves in control and experimental arms, respectively. Accuracy was generally not associated with KM curve quality.ConclusionsRMST derived from reconstructed IPD displayed excellent accuracy and predictive error compared with the gold standard. Reconstructed IPD could be used to calculate RMST in lieu of original trial IPD, to facilitate decision making for clinicians, researchers, and policy makers.     

Using aggregated mobile phone location data to compare the realised foodscapes of different socio-economic groups.

The foodscape (the built food environment) is considered one of the driving factors of the higher burden of obesity and chronic disease observed in low socio-economic status (SES) groups. Traditional data collection methods struggle to accurately capture actual access and exposure to the foodscape (realised foodscape). We assess the use of anonymised mobile phone location data (location data) in foodscape studies by applying them to a case study in Perth, Western Australia to test the hypothesis that lower SES groups have poorer realised foodscapes than high SES groups. Kernel density estimation was used to calculate realised foodscapes of different SES groups and home foodscape typologies, which were compared to home foodscapes of the different groups. The location data enabled us to measure realised foodscapes of multiple groups over an extended period and at the city scale. Low SES groups had poor availability of food outlets, including unhealthy outlets, in their home and realised foodscapes and may be more susceptible to a poor home foodscape because of low mobility.     

中医药随机对照试验中采用真实世界数据作为对照组的研究设计及挑战

在设计随机对照试验（RCT）时,如果对照组存在患者招募和入组困难的情况,就会影响试验整体实施。近年来,真实世界数据（RWD）作为除RCT之外的数据来源,在医疗领域中发挥着越来越重要的作用。中医药RCT中可以尝试采用将RWD作为对照组的研究设计,不仅可以有效解决中医药RCT西医对照组患者入组困难的问题,同时能提供有力证据来评价中医药的疗效。倾向评分法目前已广泛应用于真实世界研究中混杂因素的处理,该文对RCT采用RWD作为对照组的这类设计中,基于倾向评分法常见的4种研究设计形式以实例分别进行了介绍,包括不对称随机分配、基于倾向评分分层法的两阶段设计、倾向评分联合复合似然法及倾向评分多种方法的联合。同时,这种设计类型也存在着方法学的挑战,包括RWD数据源必须是高质量且关键信息需要规范收集、RCT和RWD患者基线特征应该具有可比性、协变量选择时需要把所有已知与干预措施和结局相关的协变量都纳入进行分析等。在中医药领域采用这种设计时,还存在着有些RWD中医证型信息缺失、中医结局指标缺失等问题,在使用RWD时,需要根据数据实际情况决定如何分析。该文对以RWD作为RCT对照组的设计类型及面临的方法学挑战进行了介绍,期望能为研究者今后使用这类设计提供方法学借鉴。     

基于数据挖掘探讨全国名中医王行宽诊疗胸痹心痛遣方用药规律及学术经验

目的：运用中医传承计算平台V3.0软件,研究王行宽教授治疗胸痹心痛遣方用药规律,传承王行宽教授诊疗胸痹心痛的学术经验。方法：收集整理2017—2020年王行宽教授于湖南中医药大学门诊诊疗冠心病心绞痛患者的原始病历资料,录入中医传承计算平台V3.0,运用软件进行方药规律研究。结果：共收集王行宽教授治疗胸痹心痛处方1 044则,所用药物多为甘、苦药物,归经以肺经为主,其次为心、脾、肝、胃、肾经;所用方剂中使用最多的经方是生脉散,最高的经验方是心痛灵Ⅲ号方;高频数药物主要有麦冬、半夏、丹参、瓜蒌皮、黄连、五味子、柴胡等药物;药物的常用剂量多为3、5、10、15 g;组方规律分析得到常用药组合129个,置信度>0.99的组合有58个,并得到常见证型核心药物;药物聚类得到6个核心药物组合。结论：王行宽教授论治胸痹心痛以益气养营、豁痰化瘀、疏肝利胆为治疗思路,并根据胸痹心痛不同证型予以辨证施治,体现其“多脏调燮、综合治理”的学术思想,其核心处方可供临床从业者参考,但仍需要进一步的临床及实验研究验证其疗效。     

基于数据挖掘的早发性卵巢功能不全动物模型分析

目的：研究早发性卵巢功能不全动物模型的造模特点,为早发性卵巢功能不全动物模型制备的规范化提供参考,为该病发病机制及诊疗研究提供更好的研究基础。方法：通过检索中国知网、万方、维普、中国生物医学文献及PubMed数据库获取近十年早发性卵巢功能不全动物实验文献,归纳实验动物种类、造模方式、给药剂量、给药方案、成模标准及检测指标,应用Excel软件进行频数分析,应用SPSS Modeler 18.0进行关联规则分析并运用Cytoscape 3.6.1对结果进行可视化升级。结果：共纳入文献281篇,建立早发性卵巢功能不全动物模型多选用SD大鼠或BALB/c小鼠,造模方式多选用医源性诱导,造模药物以环磷酰胺为主,成模标准多选用动情周期紊乱,检测指标多选用卵巢组织形态学变化、雌二醇、卵泡刺激素及动情周期以从多方面综合评价模型。结论：早发性卵巢功能不全动物实验多选用SD大鼠,以环磷酰胺首次腹腔注射给药50 mg·kg-1,第2天起以8 mg·kg-1·d-1维持14 d诱导造模,具有简便易行、成模率高且符合疾病发生过程等优点,可为该病基础动物实验研究提供参考。     

Association between Vitamin D and Heart Failure Mortality in 10,974 Hospitalized Individuals

A broad range of chronic conditions, including heart failure (HF), have been associated with vitamin D deficiency. Existing clinical trials involving vitamin D supplementation in chronic HF patients have been inconclusive. We sought to evaluate the outcomes of patients with vitamin D supplementation, compared with a matched cohort using real-world big data of HF hospitalization. This study was based on the Diagnosis Procedure Combination database in the Japanese Registry of All Cardiac and Vascular Datasets (JROAD-DPC). After exclusion criteria, we identified 93,692 patients who were first hospitalized with HF between April 2012 and March 2017 (mean age was 79 ± 12 years, and 52.2% were male). Propensity score (PS) was estimated with logistic regression model, with vitamin D supplementation as the dependent variable and clinically relevant covariates. On PS-matched analysis with 10,974 patients, patients with vitamin D supplementation had lower total in-hospital mortality (6.5 vs. 9.4%, odds ratio: 0.67, p < 0.001) and in-hospital mortality within 7 days and 30 days (0.9 vs. 2.5%, OR, 0.34, and 3.8 vs. 6.5%, OR: 0.56, both p < 0.001). In the sub-group analysis, mortalities in patients with age < 75, diabetes, dyslipidemia, atrial arrhythmia, cancer, renin-angiotensin system blocker, and β-blocker were not affected by vitamin D supplementation. Patients with vitamin D supplementation had a lower in-hospital mortality for HF than patients without vitamin D supplementation in the propensity matched cohort. The identification of specific clinical characteristics in patients benefitting from vitamin D may be useful for determining targets of future randomized control trials.     

Composition of Dietary Fatty Acids and Health Risks in Japanese Youths

In the overall composition of dietary fatty acids (FAs), the quantity of each FA is interrelated with that of others. We examined the associations between dietary FA composition and cardiometabolic risk in Japanese youths. Risk factors (anthropometric characteristics, serum lipid and liver enzyme levels, and blood pressure) were measured in 5485 junior-high-school students. Dietary intake was assessed using a food frequency questionnaire. The mean saturated FA (SFA), monounsaturated FA (MUFA), omega-6 polyunsaturated FAs (PUFAs), and omega-3 PUFAs intake were 9.6%E, 10.3%E, 6.3%E, and 1.1%E, respectively. In compositional regression analysis controlled for confounders, a high intake of omega-6 PUFAs relative to others was associated with low low-density-lipoprotein cholesterol levels (LDL-C; p = 0.003), and relative SFA intake was associated with high levels of gamma-glutamyl transpeptidase (p = 0.019). Relative omega-3 PUFAs intake was associated with low blood pressure (p = 0.005–0.034) but had unfavorable effects on adiposity and alanine transaminase. Substitutional models showed similar results for omega-6 PUFAs on LDL-C, but MUFA had inconsistent effects on risk factors. The results from the compositional data analysis were consistent with previous studies and clinical practice/knowledge. Focusing on increasing omega-6 PUFAs in Japanese youths could have favorable consequences in the long term.     

A Log Ratio-Based Analysis of Individual Changes in the Composition of the Oral Microbiota in Different Dietary Phases

Background: Investigating the influence of nutrition on oral health has a long scientific history. Due to recent technical advances like sequencing techniques for the oral microbiota, this topic has gained scientific interest again. A basic challenge is to understand the influence of nutrition on the oral microbiota and on the interaction between the oral bacteria, which is also statistically challenging. Methods: Log-transformed ratios of two bacteria concentrations are introduced as the basic analytic tool. The framework is illustrated by application in an experimental study exposing eleven participants to different nutrition schemes in five consecutive phases. Results: The method could be sufficiently used to analyse the interrelation between the bacteria and to identify some bacterial groups with the same as well as different reactions to additional dietary components. It was found that the strongest changes in bacterial concentrations were achieved by the additional consumption of dairy products. Conclusion: A log ratio-based analysis offers insights into the relation of different bacteria while taking specific features of compositional data into account. The presented methods allow becoming independent of the behaviour of other bacteria, which is a disadvantage of common analysis methods of compositions. The results indicate that modulations of the oral biofilm microbiota due to nutrition change can be attained.     

Dietary Patterns Derived from UK Supermarket Transaction Data with Nutrient and Socioeconomic Profiles

Poor diet is a leading cause of death in the United Kingdom (UK) and around the world. Methods to collect quality dietary information at scale for population research are time consuming, expensive and biased. Novel data sources offer potential to overcome these challenges and better understand population dietary patterns. In this research we will use 12 months of supermarket sales transaction data, from 2016, for primary shoppers residing in the Yorkshire and Humber region of the UK (n = 299,260), to identify dietary patterns and profile these according to their nutrient composition and the sociodemographic characteristics of the consumer purchasing with these patterns. Results identified seven dietary purchase patterns that we named: Fruity; Meat alternatives; Carnivores; Hydrators; Afternoon tea; Beer and wine lovers; and Sweet tooth. On average the daily energy intake of loyalty card holders -who may buy as an individual or for a household- is less than the adult reference intake, but this varies according to dietary purchase pattern. In general loyalty card holders meet the recommended salt intake, do not purchase enough carbohydrates, and purchase too much fat and protein, but not enough fibre. The dietary purchase pattern containing the highest amount of fibre (as an indicator of healthiness) is bought by the least deprived customers and the pattern with lowest fibre by the most deprived. In conclusion, supermarket sales data offer significant potential for understanding population dietary patterns.