Dataset Information

Joint Fairness Model with Applications to Risk Predictions for Under-represented Populations.

ABSTRACT: Under-representation of certain populations, based on gender, race/ethnicity, and age, in data collection for predictive modeling may yield less-accurate predictions for the under-represented groups. Recently, this issue of fairness in predictions has attracted significant attention, as data-driven models are increasingly utilized to perform crucial decision-making tasks. Methods to achieve fairness in the machine learning literature typically build a single prediction model subject to some fairness criteria in a manner that encourages fair prediction performances for all groups. These approaches have two major limitations: i) fairness is often achieved by compromising accuracy for some groups; ii) the underlying relationship between dependent and independent variables may not be the same across groups. We propose a Joint Fairness Model (JFM) approach for binary outcomes that estimates group-specific classifiers using a joint modeling objective function that incorporates fairness criteria for prediction. We introduce an Accelerated Smoothing Proximal Gradient Algorithm to solve the convex objective function, and demonstrate the properties of the proposed JFM estimates. Next, we presented the key asymptotic properties for the JFM parameter estimates. We examined the efficacy of the JFM approach in achieving prediction performances and parities, in comparison with the Single Fairness Model, group-separate model, and group-ignorant model through extensive simulations. Finally, we demonstrated the utility of the JFM method in the motivating example to obtain fair risk predictions for under-represented older patients diagnosed with coronavirus disease 2019 (COVID-19).

SUBMITTER: Do H

PROVIDER: S-EPMC8132236 | biostudies-literature | 2021 May

REPOSITORIES: biostudies-literature

ACCESS DATA

Publications

A Joint Fairness Model with Applications to Risk Predictions for Under-represented Populations.

Do Hyungrok H Nandi Shinjini S Putzel Preston P Smyth Padhraic P Zhong Judy J

ArXiv 20210510

In data collection for predictive modeling, under-representation of certain groups, based on gender, race/ethnicity, or age, may yield less-accurate predictions for these groups. Recently, this issue of fairness in predictions has attracted significant attention, as data-driven models are increasingly utilized to perform crucial decision-making tasks. Existing methods to achieve fairness in the machine learning literature typically build a single prediction model in a manner that encourages fair ...[more]

PMID: 34012993

Similar Datasets

Project description:Neuromuscular diseases (NMDs) affect ∼15 million people globally. In high income settings DNA-based diagnosis has transformed care pathways and led to gene-specific therapies. However, most affected families are in low-to-middle income countries (LMICs) with limited access to DNA-based diagnosis. Most (86%) published genetic data is derived from European ancestry. This marked genetic data inequality hampers understanding of genetic diversity and hinders accurate genetic diagnosis in all income settings. We developed a cloud-based transcontinental partnership to build diverse, deeply-phenotyped and genetically characterized cohorts to improve genetic architecture knowledge, and potentially advance diagnosis and clinical management. We connected 18 centres in Brazil, India, South Africa, Turkey, Zambia, Netherlands and the UK. We co-developed a cloud-based data solution and trained 17 international neurology fellows in clinical genomic data interpretation. Single gene and whole exome data were analysed via a bespoke bioinformatics pipeline and reviewed alongside clinical and phenotypic data in global webinars to inform genetic outcome decisions. We recruited 6001 participants in the first 43 months. Initial genetic analyses 'solved' or 'possibly solved' ∼56% probands overall. In-depth genetic data review of the four commonest clinical categories (limb girdle muscular dystrophy, inherited peripheral neuropathies, congenital myopathy/muscular dystrophies and Duchenne/Becker muscular dystrophy) delivered a ∼59% 'solved' and ∼13% 'possibly solved' outcome. Almost 29% of disease causing variants were novel, increasing diverse pathogenic variant knowledge. Unsolved participants represent a new discovery cohort. The dataset provides a large resource from under-represented populations for genetic and translational research. In conclusion, we established a remote transcontinental partnership to assess genetic architecture of NMDs across diverse populations. It supported DNA-based diagnosis, potentially enabling genetic counselling, care pathways and eligibility for gene-specific trials. Similar virtual partnerships could be adopted by other areas of global genomic neurological practice to reduce genetic data inequality and benefit patients globally.

Project description:IntroductionSustained viral suppression, an indicator of long-term treatment success and mortality reduction, is one of four strategic areas of the 'Ending the HIV Epidemic' federal campaign launched in 2019. Under-represented populations, like racial or ethnic minority populations, sexual and gender minority groups, and socioeconomically disadvantaged populations, are disproportionately affected by HIV and experience a more striking virological failure. The COVID-19 pandemic might magnify the risk of incomplete viral suppression among under-represented people living with HIV (PLWH) due to interruptions in healthcare access and other worsened socioeconomic and environmental conditions. However, biomedical research rarely includes under-represented populations, resulting in biased algorithms. This proposal targets a broadly defined under-represented HIV population. It aims to develop a personalised viral suppression prediction model using machine learning (ML) techniques by incorporating multilevel factors using All of Us (AoU) data.Methods and analysisThis cohort study will use data from the AoU research programme, which aims to recruit a broad, diverse group of US populations historically under-represented in biomedical research. The programme harmonises data from multiple sources on an ongoing basis. It has recruited ~4800 PLWH with a series of self-reported survey data (eg, Lifestyle, Healthcare Access, COVID-19 Participant Experience) and relevant longitudinal electronic health records data. We will examine the change in viral suppression and develop personalised viral suppression prediction due to the impact of the COVID-19 pandemic using ML techniques, such as tree-based classifiers (classification and regression trees, random forest, decision tree and eXtreme Gradient Boosting), support vector machine, naïve Bayes and long short-term memory.Ethics and disseminationThe institutional review board approved the study at the University of South Carolina (Pro00124806) as a Non-Human Subject study. Findings will be published in peer-reviewed journals and disseminated at national and international conferences and through social media.

Dataset Information

Joint Fairness Model with Applications to Risk Predictions for Under-represented Populations.

Publications

A Joint Fairness Model with Applications to Risk Predictions for Under-represented Populations.

Similar Datasets

OmicsDI is part of the ELIXIR infrastructure

Tweets