Implicit bias of encoded variables: frameworks for addressing structured bias in EHR-GWAS data

Inherent Bias in Electronic Health Records: A Scoping Review of Sources of Bias

Objectives

1.1Biases inherent in electronic health records (EHRs), and therefore in medical artificial intelligence (AI) models may significantly exacerbate health inequities and challenge the adoption of ethical and responsible AI in healthcare. Biases arise from multiple sources, some of which are not as documented in the literature. Biases are encoded in how the data has been collected and labeled, by implicit and unconscious biases of clinicians, or by the tools used for data processing. These biases and their encoding in healthcare records undermine the reliability of such data and bias clinical judgments and medical outcomes. Moreover, when healthcare records are used to build data-driven solutions, the biases are further exacerbated, resulting in systems that perpetuate biases and induce healthcare disparities. This literature scoping review aims to categorize the main sources of biases inherent in EHRs.

Methods

1.2We queried PubMed and Web of Science on January 19th, 2023, for peer-reviewed sources in English, published between 2016 and 2023, using the PRISMA approach to stepwise scoping of the literature. To select the papers that empirically analyze bias in EHR, from the initial yield of 430 papers, 27 duplicates were removed, and 403 studies were screened for eligibility. 196 articles were removed after the title and abstract screening, and 96 articles were excluded after the full-text review resulting in a final selection of 116 articles.

Results

1.3Systematic categorizations of diverse sources of bias are scarce in the literature, while the effects of separate studies are often convoluted and methodologically contestable. Our categorization of published empirical evidence identified the six main sources of bias: a) bias arising from past clinical trials; b) data-related biases arising from missing, incomplete information or poor labeling of data; human-related bias induced by c) implicit clinician bias, d) referral and admission bias; e) diagnosis or risk disparities bias and finally, (f) biases in machinery and algorithms.

Conclusions

1.4Machine learning and data-driven solutions can potentially transform healthcare delivery, but not without limitations. The core inputs in the systems (data and human factors) currently contain several sources of bias that are poorly documented and analyzed for remedies. The current evidence heavily focuses on data-related biases, while other sources are less often analyzed or anecdotal. However, these different sources of biases add to one another exponentially. Therefore, to understand the issues holistically we need to explore these diverse sources of bias. While racial biases in EHR have been often documented, other sources of biases have been less frequently investigated and documented (e.g. gender-related biases, sexual orientation discrimination, socially induced biases, and implicit, often unconscious, human-related cognitive biases). Moreover, some existing studies lack causal evidence, illustrating the different prevalences of disease across groups, which does not per se prove the causality. Our review shows that data-, human- and machine biases are prevalent in healthcare and they significantly impact healthcare outcomes and judgments and exacerbate disparities and differential treatment. Understanding how diverse biases affect AI systems and recommendations is critical. We suggest that researchers and medical personnel should develop safeguards and adopt data-driven solutions with a "bias-in-mind" approach. More empirical evidence is needed to tease out the effects of different sources of bias on health outcomes.

Open Science Practices in Psychiatric Genetics: A Primer

Open science ensures that research is transparently reported and freely accessible for all to assess and collaboratively build on. Psychiatric genetics has led among the health sciences in implementing some open science practices in common study designs, such as replication as part of genome-wide association studies. However, thorough open science implementation guidelines are limited and largely not specific to data, privacy, and research conduct challenges in psychiatric genetics. Here, we present a primer of open science practices, including selection of a research topic with patients/nonacademic collaborators, equitable authorship and citation practices, design of replicable, reproducible studies, preregistrations, open data, and privacy issues. We provide tips for informative figures and inclusive, precise reporting. We discuss considerations in working with nonacademic collaborators and distributing research through preprints, blogs, social media, and accessible lecture materials. Finally, we provide extra resources to support every step of the research process.

Digital Determinants of Health: Health data poverty amplifies existing health disparities-A scoping review

Artificial intelligence (AI) and machine learning (ML) have an immense potential to transform healthcare as already demonstrated in various medical specialties. This scoping review focuses on the factors that influence health data poverty, by conducting a literature review, analysis, and appraisal of results. Health data poverty is often an unseen factor which leads to perpetuating or exacerbating health disparities. Improvements or failures in addressing health data poverty will directly impact the effectiveness of AI/ML systems. The potential causes are complex and may enter anywhere along the development process. The initial results highlighted studies with common themes of health disparities (72%), AL/ML bias (28%) and biases in input data (18%). To properly evaluate disparities that exist we recommend a strengthened effort to generate unbiased equitable data, improved understanding of the limitations of AI/ML tools, and rigorous regulation with continuous monitoring of the clinical outcomes of deployed tools.

Comorbidity Profiles of Posttraumatic Stress Disorder Across the Medical Phenome

Background

Prior epidemiological research has linked PTSD with specific physical health problems, but the comprehensive landscape of medical conditions associated with PTSD remains uncharacterized. Electronic health records (EHR) provide an opportunity to overcome prior clinical knowledge gaps and uncover associations with biological relevance that potentially vary by sex.

Methods

PTSD was defined among biobank participants (total N=123,365) in a major healthcare system using two ICD code-based definitions: broad (1+ PTSD or acute stress codes versus 0; NCase=14,899) and narrow (2+ PTSD codes versus 0; NCase=3,026). Using a phenome-wide association (PheWAS) design, we tested associations between each PTSD definition and all prevalent disease umbrella categories, i.e., phecodes. We also conducted sex-stratified PheWAS analyses including a sex-by-diagnosis interaction term in each logistic regression.

Results

A substantial number of phecodes were significantly associated with PTSDNarrow (61%) and PTSDBroad (83%). While top associations were shared between the two definitions, PTSDBroad captured 334 additional phecodes not significantly associated with PTSDNarrow and exhibited a wider range of significantly associated phecodes across various categories, including respiratory, genitourinary, and circulatory conditions. Sex differences were observed, in that PTSDBroad was more strongly associated with osteoporosis, respiratory failure, hemorrhage, and pulmonary heart disease among male patients, and with urinary tract infection, acute pharyngitis, respiratory infections, and overweight among female patients.

Conclusions

This study provides valuable insights into a diverse range of comorbidities associated with PTSD, including both known and novel associations, while highlighting the influence of sex differences and the impact of defining PTSD using EHR.

Detecting and Adjusting for Hidden Biases due to Phenotype Misclassification in Genome-Wide Association Studies

With the advent of healthcare-based genotyped biobanks, genome-wide association studies (GWAS) leverage larger sample sizes, incorporate patients with diverse ancestries and introduce noisier phenotypic definitions. Yet the extent and impact of phenotypic misclassification on large-scale datasets is not currently well understood due to a lack of statistical methods to estimate relevant parameters from empirical data. Here, we develop a statistical method and scalable software, PheMED, Phenotypic Measurement of Effective Dilution, to quantify phenotypic misclassification across GWAS using only summary statistics. We illustrate how the parameters estimated by PheMED relate to the negative and positive predictive value of the labeled phenotype, compared to ground truth, and how misclassification of the phenotype yields diluted effect-sizes of variant-phenotype associations. Furthermore, we apply our methodology to detect multiple instances of statistically significant dilution in real-world data. We demonstrate how effective dilution biases downstream GWAS replication and heritability analyses despite utilizing current best practices, and provide a dilution-aware meta-analysis approach that outperforms existing methods. Consequently, we anticipate that PheMED will be a valuable tool for researchers to address phenotypic data quality issues both within and across cohorts.

Leveraging electronic health records for data science: common pitfalls and how to avoid them

Analysis of electronic health records (EHRs) is an increasingly common approach for studying real-world patient data. Use of routinely collected data offers several advantages compared with other study designs, including reduced administrative costs, the ability to update analysis as practice patterns evolve, and larger sample sizes. Methodologically, EHR analysis is subject to distinct challenges because data are not collected for research purposes. In this Viewpoint, we elaborate on the importance of in-depth knowledge of clinical workflows and describe six potential pitfalls to be avoided when working with EHR data, drawing on examples from the literature and our experience. We propose solutions for prevention or mitigation of factors associated with each of these six pitfalls-sample selection bias, imprecise variable definitions, limitations to deployment, variable measurement frequency, subjective treatment allocation, and model overfitting. Ultimately, we hope that this Viewpoint will guide researchers to further improve the methodological robustness of EHR analysis.

What next for eating disorder genetics? Replacing myths with facts to sharpen our understanding

Substantial progress has been made in the understanding of anorexia nervosa (AN) and eating disorder (ED) genetics through the efforts of large-scale collaborative consortia, yielding the first genome-wide significant loci, AN-associated genes, and insights into metabo-psychiatric underpinnings of the disorders. However, the translatability, generalizability, and reach of these insights are hampered by an overly narrow focus in our research. In particular, stereotypes, myths, assumptions and misconceptions have resulted in incomplete or incorrect understandings of ED presentations and trajectories, and exclusion of certain patient groups from our studies. In this review, we aim to counteract these historical imbalances. Taking as our starting point the Academy for Eating Disorders (AED) Truth #5 "Eating disorders affect people of all genders, ages, races, ethnicities, body shapes and weights, sexual orientations, and socioeconomic statuses", we discuss what we do and do not know about the genetic underpinnings of EDs among people in each of these groups, and suggest strategies to design more inclusive studies. In the second half of our review, we outline broad strategic goals whereby ED researchers can expand the diversity, insights, and clinical translatability of their studies.

Penetrance and Pleiotropy of Polygenic Risk Scores for Schizophrenia, Bipolar Disorder, and Depression Among Adults in the US Veterans Affairs Health Care System

Importance

Serious mental illnesses, including schizophrenia, bipolar disorder, and depression, are heritable, highly multifactorial disorders and major causes of disability worldwide.

Objective

To benchmark the penetrance of current neuropsychiatric polygenic risk scores (PRSs) in the Veterans Health Administration health care system and to explore associations between PRS and broad categories of human disease via phenome-wide association studies.

Design, Setting, and Participants

Extensive Veterans Health Administration's electronic health records were assessed from October 1999 to January 2021, and an embedded cohort of 9378 individuals with confirmed diagnoses of schizophrenia or bipolar 1 disorder were found. The performance of schizophrenia, bipolar disorder, and major depression PRSs were compared in participants of African or European ancestry in the Million Veteran Program (approximately 400 000 individuals), and associations between PRSs and 1650 disease categories based on ICD-9/10 billing codes were explored. Last, genomic structural equation modeling was applied to derive novel PRSs indexing common and disorder-specific genetic factors. Analysis took place from January 2021 to January 2022.

Main Outcomes and Measures

Diagnoses based on in-person structured clinical interviews were compared with ICD-9/10 billing codes. PRSs were constructed using summary statistics from genome-wide association studies of schizophrenia, bipolar disorder, and major depression.

Results

Of 707 299 enrolled study participants, 459 667 were genotyped at the time of writing; 84 806 were of broadly African ancestry (mean [SD] age, 58 [12.1] years) and 314 909 were of broadly European ancestry (mean [SD] age, 66.4 [13.5] years). Among 9378 individuals with confirmed diagnoses of schizophrenia or bipolar 1 disorder, 8962 (95.6%) were correctly identified using ICD-9/10 codes (2 or more). Among those of European ancestry, PRSs were robustly associated with having received a diagnosis of schizophrenia (odds ratio [OR], 1.81 [95% CI, 1.76-1.87]; P < 10-257) or bipolar disorder (OR, 1.42 [95% CI, 1.39-1.44]; P < 10-295). Corresponding effect sizes in participants of African ancestry were considerably smaller for schizophrenia (OR, 1.35 [95% CI, 1.29-1.42]; P < 10-38) and bipolar disorder (OR, 1.16 [95% CI, 1.11-1.12]; P < 10-10). Neuropsychiatric PRSs were associated with increased risk for a range of psychiatric and physical health problems.

Conclusions and Relevance

Using diagnoses confirmed by in-person structured clinical interviews and current neuropsychiatric PRSs, the validity of an electronic health records-based phenotyping approach in US veterans was demonstrated, highlighting the potential of PRSs for disentangling biological and mediated pleiotropy.

Cohort design and natural language processing to reduce bias in electronic health records research

Electronic health record (EHR) datasets are statistically powerful but are subject to ascertainment bias and missingness. Using the Mass General Brigham multi-institutional EHR, we approximated a community-based cohort by sampling patients receiving longitudinal primary care between 2001-2018 (Community Care Cohort Project [C3PO], n = 520,868). We utilized natural language processing (NLP) to recover vital signs from unstructured notes. We assessed the validity of C3PO by deploying established risk models for myocardial infarction/stroke and atrial fibrillation. We then compared C3PO to Convenience Samples including all individuals from the same EHR with complete data, but without a longitudinal primary care requirement. NLP reduced the missingness of vital signs by 31%. NLP-recovered vital signs were highly correlated with values derived from structured fields (Pearson r range 0.95-0.99). Atrial fibrillation and myocardial infarction/stroke incidence were lower and risk models were better calibrated in C3PO as opposed to the Convenience Samples (calibration error range for myocardial infarction/stroke: 0.012-0.030 in C3PO vs. 0.028-0.046 in Convenience Samples; calibration error for atrial fibrillation 0.028 in C3PO vs. 0.036 in Convenience Samples). Sampling patients receiving regular primary care and using NLP to recover missing data may reduce bias and maximize generalizability of EHR research.

Psychiatric genomics, mental health equity, and intersectionality: A framework for research and practice

The causal mechanisms and manifestations of psychiatric illness cannot be neatly narrowed down or quantified for diagnosis and treatment. Large-scale genome-wide association studies (GWAS) might renew hope for locating genetic predictors and producing precision medicines, however such hopes can also distract from appreciating social factors and structural injustices that demand more socially inclusive and equitable approaches to mental healthcare. A more comprehensive approach begins with recognizing that there is no one type of contributor to mental illness and its duration that should be prioritized over another. We argue that, if the search for biological specificity is to complement the need to alleviate the social distress that produces mental health inequities, psychiatric genomics must incorporate an intersectional dimension to models of mental illness across research priorities, scientific frameworks, and clinical applications. We outline an intersectional framework that will guide all professionals working in the expanding field of psychiatric genomics to better incorporate issues of social context, racial and cultural diversity, and downstream ethical considerations into their work.

Pharmacogenomics: A road ahead for precision medicine in psychiatry

Psychiatric genomics is providing insights into the nature of psychiatric conditions that in time should identify new drug targets and improve patient care. Less attention has been paid to psychiatric pharmacogenomics research, despite its potential to deliver more rapid change in clinical practice and patient outcomes. The pharmacogenomics of treatment response encapsulates both pharmacokinetic ("what the body does to a drug") and pharmacodynamic ("what the drug does to the body") effects. Despite early optimism and substantial research in both these areas, they have to date made little impact on clinical management in psychiatry. A number of bottlenecks have hampered progress, including a lack of large-scale replication studies, inconsistencies in defining valid treatment outcomes across experiments, a failure to routinely incorporate adverse drug reactions and serum metabolite monitoring in study designs, and inadequate investment in the longitudinal data collections required to demonstrate clinical utility. Nonetheless, advances in genomics and health informatics present distinct opportunities for psychiatric pharmacogenomics to enter a new and productive phase of research discovery and translation.

Genetic Susceptibility to Pneumonia: A GWAS Meta-Analysis Between the UK Biobank and FinnGen

Pneumonia is a respiratory condition with complex etiology. Host genetic variation is thought to contribute to individual differences in susceptibility and symptom manifestation. Here, we analyze pneumonia data from the UK Biobank (14,780 cases and 439,096 controls) and FinnGen (9980 cases and 86,519 controls) and perform a genomewide association study meta-analysis. We use gene-based tests, colocalization, genetic correlation, latent causal variable (LCV) and polygenic prediction in an independent Australian sample (N = 5595) to draw insights into the etiology of pneumonia risk. We identify two independent loci on chromosome 15 (lead single-nucleotide polymorphisms rs2009746 and rs76474922) to be associated with pneumonia (p < 5e-8). Gene-based tests revealed 18 genes in chromosomes 15, 16 and 9, including IL127, PBX3, ApoB receptor (APOBR) and smoking related genes CHRNA3/5, statistically associated with pneumonia. We observed genetic correlations between pneumonia and cardiorespiratory, psychiatric and inflammatory related traits. LCV analysis suggests a strong genetic causal relationship with cardiovascular health phenotypes. Polygenic risk scores for pneumonia significantly predicted self-reported pneumonia in an independent sample, albeit with a small effect size (OR = 1.11 95% CI [1.04, 1.19], p < .05). Sensitivity analyses suggested the associations in chromosome 15 are mediated by smoking history, but the associations in chromosomes 16 and 9, and polygenic prediction were robust to adjustment for smoking. Altogether, our results highlight common genetic variants, genes and potential pathways that contribute to individual differences in susceptibility to pneumonia, and advance our understanding of the genetic factors underlying heterogeneity in respiratory medical outcomes.