Improving covariance-regularized discriminant analysis for EHR-based predictive analytics of diseases

Electronic health records (EHR) provide a rich source of temporal data that present a unique opportunity to characterize disease patterns and risk of imminent disease. While many data-mining tools have been adopted for EHR-based disease early detection, linear discriminant analysis (LDA) is one of the most commonly used statistical methods. However, it is difficult to train an accurate LDA model for early disease diagnosis when too few patients are known to have the target disease. Furthermore, EHR data are heterogeneous with significant noise. In such cases, the covariance matrices used in LDA are usually singular and estimated with a large variance. This article presents Daehr , an extension of the LDA framework using electronic health record data to address these issues. Beyond existing LDA analyzers, we propose Daehr to (1) eliminate the data noise caused by the manual encoding of EHR data and (2) lower the variance of parameter (covariance matrices) estimation for LDA models when only a few patients’ EHR are available for training. To achieve these two goals, we designed an iterative algorithm to improve the covariance matrix estimation with embedded data-noise/parameter-variance reduction for LDA. We evaluated Daehr extensively using the College Health Surveillance Network, a large, real-world EHR dataset. Specifically, our experiments compared the performance of LDA to three baselines (i.e., LDA and its derivatives) in identifying college students at high risk for mental health disorders from 23 U.S. universities. Experimental results demonstrate Daehr significantly outperforms the three baselines by achieving 1.4%--19.4% higher accuracy and a 7.5%--43.5% higher F1-score.

AB1206 SELF-REPORT OF FRACTURE HISTORY COMPARED TO FRACTURE CODES FROM AN ELECTRONIC HEALTH RECORD DATASET

The increase in hospitalizations of patients with noncommunicable diseases remains a key challenge for healthcare systems. A significant portion of these hospitalizations is potentially preventable with timely outpatient monitoring and effective primary care. Predictive models based on data from health information systems (HIS) and electronic health records (EHRs) make it possible to identify patients at increased risk of hospitalization and improve their management strategies. The article systematizes and summarizes international experience in using hospitalization risk prognostic models, developed on the basis of HIS and EHR data. We made a systematic review of publications presented in Russian and foreign databases (Scopus, PubMed, BMC Health Services Research, BMJ Research, Google Scholar, Elibrary, Oxford Academic, Wiley Online Library) for the period 1993-2023. The analysis included studies that used EHR and HIS data to develop and/or validate prognostic models of hospitalization risk. The information is systematized by following predictor categories: diseases and comorbidities, prescriptions, interaction with healthcare organizations, sociodemographic and laboratory parameters. The final analysis included 14 studies describing 28 models for predicting the hospitalization risk. Most studies used logistic regression. Machine learning methods (gradient boosting, random forest, and Bayesian ensemble models) were used in some studies. The categories with the highest prognostic value were "diseases and comorbidities" (42,3%), "medication prescriptions" (23,6%), and "interaction with health organizations" (19,8%). Including temporal characteristics (frequency and recency of hospitalizations, visits to primary care facilities, and laboratory tests) increased the model accuracy. Predictive models based on EHR and HIS data demonstrate high discriminatory power and enable the assessment of hospitalization risk at the primary care level. Machine learning methods in some studies demonstrated the possibility of a more in-depth analysis of the relationships between predictors and improved prediction accuracy. A promising area of research is the further development, external validation, and adaptation of models using expanded EHR and HIS data sets for its use in outpatient monitoring and preventing noncommunicable diseases.

Performance of the FIND-FH machine learning algorithm for the identification of individuals with suspected familial hypercholesterolemia.

Analyzing “large p small n” data is becoming increasingly paramount in a wide range of application fields. As a projection pursuit index, the Penalized Discriminant Analysis ($\mathrm{PDA}$) index, built upon the Linear Discriminant Analysis ($\mathrm{LDA}$) index, is devised in Lee and Cook (2010) to classify high-dimensional data with promising results. Yet, there is little information available about its performance compared with the popular Support Vector Machine ($\mathrm{SVM}$). This paper conducts extensive numerical studies to compare the performance of the $\mathrm{PDA}$ index with the $\mathrm{LDA}$ index and $\mathrm{SVM}$, demonstrating that the $\mathrm{PDA}$ index is robust to outliers and able to handle high-dimensional datasets with extremely small sample sizes, few important variables, and multiple classes. Analyses of several motivating real-world datasets reveal the practical advantages and limitations of individual methods, suggesting that the $\mathrm{PDA}$ index provides a useful alternative tool for classifying complex high-dimensional data. These new insights, along with the hands-on implementation of the $\mathrm{PDA}$ index functions in the R package classPP, facilitate statisticians and data scientists to make effective use of both sets of classification tools.

Do recent changes in European Society of Human Reproduction and Embryology (ESHRE) clinical guidelines result in more comprehensive diagnosis of women with endometriosis? The latest shift in clinical guidelines results in diagnosis of more women with endometriosis but current ESHRE diagnostic criteria do not capture a sizable percentage of women with the disease. Historically, laparoscopy was the gold standard for diagnosing endometriosis, a complex gynecological condition marked by a heterogeneous set of symptoms that vary widely among women. More recently, changes in clinical guidelines have shifted to incorporate imaging-based approaches such as transvaginal sonography and magnetic resonance imaging. Retrospective, observational cohort study of women aged 15-49 years diagnosed with endometriosis in the United States (US) between January 1, 2013, and December 31, 2023. Data sources include US insurance claims data from the Merative™ MarketScan® Commercial Database (CCAE), Merative™ MarketScan® Multi-State Medicaid Database (MDCD), Optum® de-identified Electronic Health Record dataset (Optum® EHR), and electronic health record (EHR) data from a large academic medical center in New York City (CUIMC EHR). To examine the potential impact of expanding clinical criteria for the diagnosis of endometriosis, we validated and compared five cohort definitions based on different sets of diagnostic guidelines involving combinations of surgical confirmation, diagnostic imaging, guideline-recognized symptoms, and other symptoms commonly reported among women with endometriosis. We performed pairwise comparisons between cohorts and applied Bonferroni corrections to account for multiple comparisons. We identified 491,048 women with a diagnosis of endometriosis across the CCAE, MDCD, Optum EHR, and CUIMC EHR datasets. Each cohort definition demonstrated strong positive predictive value (0.84-0.96), yet only 15-20% of cases were identified by all 5 cohort definitions. Women diagnosed with endometriosis based on imaging and symptoms were three years younger, on average, than women with a diagnosis based on surgical confirmation (mean age = 35 years [SD = 9] vs 38 years [SD = 8]; p<0.001). Women in cohorts based only on symptoms were two years younger than those based on surgery (36 years [SD = 8] vs 38 years [SD = 8]; p<0.001). More than one-fourth of cases presented with endometriosis-related symptoms but lacked surgical or imaging-related documentation required by ESHRE guideline criteria. Pain was reported among nearly all women with endometriosis. Abdominal pain and pain in the pelvis were the most prevalent (ranging from 69% to 90% of women in each cohort). Among approximately 2-5% of all endometriosis cases (14,795 total), women presented with pelvic and/or abdominal pain but none of the other symptoms noted in clinical guidelines. Our study has potential biases associated with documentation practices and secondary data use of insurance claims and EHR data. Further, the phenotyping algorithms used rely on clinical codes that do not necessarily capture all ESHRE diagnostic criteria for endometriosis and may not be generalizable to women with atypical presentation of endometriosis. High positive predictive value among all five cohort definitions despite poor overlap among cases identified illustrates both the heterogeneous presentation of the disease and importance of expanding diagnostic criteria. For example, cohorts derived from updated guidelines identified younger patients at time of diagnosis. Women diagnosed based on imaging had higher rates of emergency room visits while patients diagnosed via laparoscopy had a larger number of hospitalizations. The substantial number of cases with pelvic and/or abdominal pain but none of the other symptoms noted in clinical guidelines underscores the continued need for improved access to timely and appropriate care, particularly among those with non-classical symptoms, different care-seeking patterns, or lack of available surgical intervention.

Objective: To evaluate the continuity and completeness of electronic health record (EHR) data, and the concordance of select clinical outcomes and baseline comorbidities between EHR and linked claims data, from three healthcare delivery systems in Taiwan. Methods: We identified oral hypoglycemic agent (OHA) users from the Integrated Medical Database of National Taiwan University Hospital (NTUH-iMD), which was linked to the National Health Insurance Research Database (NHIRD), from June 2011 to December 2016. A secondary evaluation involved two additional EHR databases. We created consecutive 90-day periods before and after the first recorded OHA prescription and defined patients as having continuous EHR data if there was at least one encounter or prescription in a 90-day interval. EHR data completeness was measured by dividing the number of encounters in the NTUH-iMD by the number of encounters in the NHIRD. We assessed the concordance between EHR and claims data on three clinical outcomes (cardiovascular events, nephropathy-related events, and heart failure admission). We used individual comorbidities that comprised the Charlson comorbidity index to examine the concordance of select baseline comorbidities between EHRs and claims. Results: We identified 39,268 OHA users in the NTUH-iMD. Thirty-one percent (n = 12,296) of these users contributed to the analysis that examined data continuity during the 6-month baseline and 24-month follow-up period; 31% (n = 3,845) of the 12,296 users had continuous data during this 30-month period and EHR data completeness was 52%. The concordance of major cardiovascular events, nephropathy-related events, and heart failure admission was moderate, with the NTU-iMD capturing 49–55% of the outcome events recorded in the NHIRD. The concordance of comorbidities was considerably different between the NTUH-iMD and NHIRD, with an absolute standardized difference >0.1 for most comorbidities examined. Across the three EHR databases studied, 29–55% of the OHA users had continuous records during the 6-month baseline and 24-month follow-up period. Conclusion: EHR data continuity and data completeness may be suboptimal. A thorough evaluation of data continuity and completeness is recommended before conducting clinical and translational research using EHR data in Taiwan.

Causal inference aims to estimate the causal relationships and effect sizes among treatments and outcomes. Electronic health record (EHR) data is a valuable healthcare data source that can support causal inference. However, a large percentage of the data is missing in EHRs and they are missing not at random (MNAR). Ignoring MNAR can lead to severe biases, to the extent where the causal structure underlying the data gets distorted. We proposed a new causal inference methodology that addresses the MNAR problem and thus helps preserve the causal structure. We compared the performance of our proposed method with the traditional causal inference method, structural equation modeling (SEM). We evaluated these methods for their accuracy in estimating the causal effect sizes and their ability to converge at all. We employed both simulation studies and real-world EHR data sets. We demonstrated that imputation under the improper missingness mechanism distorted the causal structure to a degree where SEM found it incompatible with the data and failed to. converge. Even when 20 to 30 % of the values were missing, SEM failed to converge in as many as 50% of the runs. The proposed causal inference method achieved a higher convergence rate and more accurate estimation of latent treatment effects both on the synthetic data and on a real EHR data set. We proposed a new methodology that incorporates the knowledge of missing data mechanisms. It significantly mitigated the biases associated with MNAR in the EHR dataset and substantially outperformed SEM that uses the improper missing data mechanism.

To examine which parental health care and health factors are most strongly associated with a child's receipt of recommended care we must be able to link children to their parents in electronic health record data. Yet, there is not an easy way to link these data.To identify a national cohort of children that link to at least one parent in the same electronic health record dataset and describe their demographics.Methodology to link parents and children in electronic health records and descriptive sociodemographic data.Children with at least one encounter with a primary care clinician between Januray 1, 2007 and December 12, 2018 to a community health center in the OCHIN national network. We identified parents of these children who also had at least one encounter to a community health center in the network using emergency contact and guarantor record fields.A total of 227,552 children had parents with a linkable patient record. After exclusions, our final cohort included 213,513 distinct children with either one or two parent-links. 82% of children linked to a mother only, 14% linked to a father only, and 4% linked to both a mother and a father. Most families consisted of only one linked child (61%).We were able to link 33% of children to a parent in electronic health record data from a large network of community health centers across the United States. Further analyses utilizing these linkages will allow examination of the multi-level factors that impact a child's receipt of recommended health care.

Linear Discriminant Analysis (LDA) is an easy and efficient method for pattern classification, while it is also broadly utilized for the initial discovery of diseases using Electronic Health Records (EHR) data. Nonetheless, the performance of LDA for EHR data classification is recurrently influenced by two major factors: poor evaluation of LDA parameters (e.g., covariance matrix), and “linear inseparability” of the EHR data for classification. In this paper, we propose a novel classifier SDA -Sparse Discriminant Analysis method for heart disease detection. The time complexity will be reduced in this algorithm by optimal scoring analysis of LDA and will be comprehensive to execute sparse discrimination through the combination of Gaussians if limits between classes are nonlinear or if subgroups are available inside every class. On the whole, compared to previous techniques, our proposed technique is more appropriate for the diagnosis of heart disease patients with higher accuracy.

We integrate literature- and data-driven task analysis methods to derive an initial task taxonomy for electronic health record (EHR) and electronic medical record (EMR) data analysis. An EHR (EMR) is a digital and longitudinal version of a patients health(medical) information and may include all key clinical events relevant to that persons health (medical) history, such as provider, demographics, progress notes, medicine, diagnosis, etc. Our goal is to arrive a task taxonomy for analyzing EHR (EMR) datasets because tasks play an important role in the design and evaluation of visualization techniques. Our method has three stages: data collection, task modelling, and task taxonomy summary. In data collection, we first survey related literature from the past two decades and extract typical tasks and corresponding data by extracting goals and scenarios of the particular work. We introduce multiple continuous relations to describe specific binary or multiple continuous relation-seeking tasks. Finally, we arrive an initial set of task types for EHR/EMR analysis that guide the design and evaluation of visualization techniques.

A computer vision system for automatic recognition and classification of five varieties of plant leaves under controlled laboratory imaging conditions, comprising: 1–Cydonia oblonga (quince), 2–Eucalyptus camaldulensis dehn (river red gum), 3–Malus pumila (apple), 4–Pistacia atlantica (mt. Atlas mastic tree) and 5–Prunus armeniaca (apricot), is proposed. 516 tree leaves images were taken and 285 features computed from each object including shape features, color features, texture features based on the gray level co-occurrence matrix, texture descriptors based on histogram and moment invariants. Seven discriminant features were selected and input for classification purposes using three classifiers: hybrid artificial neural network–ant bee colony (ANN–ABC), hybrid artificial neural network–biogeography based optimization (ANN–BBO) and Fisher linear discriminant analysis (LDA). Mean correct classification rates (CCR), resulted in 94.04%, 89.23%, and 93.99%, for hybrid ANN–ABC; hybrid ANN–BBO; and LDA classifiers, respectively. Best classifier mean area under curve (AUC), mean sensitivity, and mean specificity, were computed for the five tree varieties under study, resulting in: 1–Cydonia oblonga (quince) 0.991 (ANN–ABC), 95.89% (ANN–ABC), 95.91% (ANN–ABC); 2–Eucalyptus camaldulensis dehn (river red gum) 1.00 (LDA), 100% (LDA), 100% (LDA); 3–Malus pumila (apple) 0.996 (LDA), 96.63% (LDA), 94.99% (LDA); 4–Pistacia atlantica (mt. Atlas mastic tree) 0.979 (LDA), 91.71% (LDA), 82.57% (LDA); and 5–Prunus armeniaca (apricot) 0.994 (LDA), 88.67% (LDA), 94.65% (LDA), respectively.

Electronic health record (EHR) data are widely used in public health research, including in HIV-related studies, but are limited by potential bias due to incomplete and inaccurate information, lack of generalizability, and lack of representativeness. This study explores how workflow processes among HIV health care providers (HCPs), data scientists, and state health department professionals may potentially introduce or minimize bias within EHR data. One focus group with 3 health department professionals working in HIV surveillance and 16 in-depth interviews (ie, 5 people with HIV, 5 HCPs, 5 data scientists, and 1 health department professional providing retention-in-care services) were conducted with participants purposively sampled in South Carolina from August 2023 to April 2024. All interviews were transcribed verbatim and analyzed using a constructivist grounded theory approach, where transcripts were first coded and then focused, axial, and theoretically coded. The EHR data lifecycle originates with people with HIV and HCPs in the clinical setting. Data scientists then curate EHR data and health department professionals manage and use the data for surveillance and policy decision-making. Throughout this lifecycle, the three primary stakeholders (ie, HCPs, data scientists, and health department professionals) identified challenges with EHR processes and provided their recommendations and accommodations in addressing the related challenges. HCPs reported the influence of socio-structural biases on their inquiry, interpretation, and documentation of social determinants of health (SDOH) information of people living with HIV, the influence of which is proposed to be mitigated through people living with HIV access to their EHRs. Data scientists identified limited data availability and representativeness as biasing the data they manage. Health department professionals face challenges with delayed and incomplete data, which may be addressed statistically but require consideration of the data's limitations. Overall, bias within the EHR data lifecycle persists because workflows are not intentionally structured to minimize bias and there is a diffusion of responsibility for data quality between the various stakeholders. From the perspective of various stakeholders, this study describes the EHR data lifecycle and its associated challenges as well as stakeholders' accommodations and recommendations for mitigating and eliminating bias in EHR data. Based upon these findings, studies reliant on EHR data should adequately consider its challenges and limitations. Throughout the EHR data lifecycle, bias could be reduced through an inclusive, supportive health care environment, people living with HIV verification of SDOH information, the customization of data collection systems, and EHR data inspection for completeness, accuracy, and timeliness. Future research is needed to further identify instances where bias is introduced and how it can best be mitigated and eliminated across the EHR data lifecycle. Systematic changes are necessary to reduce instances of bias between data workflows and stakeholders.

BACKGROUNDElectronic data capture (EDC) has the potential to improve data accuracy and efficiency in oncology. The PBTC and PEP-CTN cooperative groups conducted a pilot project to 1) assess the feasibility of automated transfer of laboratory data from participating institutions’ electronic health records (EHR) to the Medidata Rave EDC housed within each consortium’s Uniform Resource Locator (URL); 2) evaluate discrepancies between retrospective automated laboratory data extraction from the EHR and manual data ascertainment during the original conduct of the study (MANUAL). METHODSData from patients treated on 7 completed PBTC or PEP-CTN trials at 7 hospitals were included. Each consortium created a Rave instance to receive data. Sites elected to use local methods or the ExtractEHR R package to map and extract laboratory data for designated patients. Post-extraction data were uploaded to the Rave Batch Uploader (BU)-associated file transfer protocol (FTP) and BU uploaded data into Rave. After successful upload, discrepancies between EHR and MANUAL data were categorized. Two hospitals reviewed discrepancies at their sites to identify reasons for discrepancies. RESULTSAutomated extraction and upload required significant PBTC/PEP-CTN operational effort and close work with hospital technical teams to map and extract data. All sites successfully extracted specified laboratory data and had ≥1 complete upload via BU to Rave; 2 sites had 2 uploads. Discrepancies were most commonly due to more data in EHR datasets (>1 result on a given day, data on additional study window days) or mismatched values or units. Differences in percentages of discrepant results between hospitals were likely due to mapping differences for EHR extraction. CONCLUSIONSThis study demonstrated that in the pediatric cooperative oncology group setting automated laboratory data ascertainment can be performed. Accomplishing this required upfront central operational and hospital-based effort to prepare for data upload and transfer. This effort varied by site.

BackgroundMicroarray techniques provide promising tools for cancer diagnosis using gene expression profiles. However, molecular diagnosis based on high-throughput platforms presents great challenges due to the overwhelming number of variables versus the small sample size and the complex nature of multi-type tumors. Support vector machines (SVMs) have shown superior performance in cancer classification due to their ability to handle high dimensional low sample size data. The multi-class SVM algorithm of Crammer and Singer provides a natural framework for multi-class learning. Despite its effective performance, the procedure utilizes all variables without selection. In this paper, we propose to improve the procedure by imposing shrinkage penalties in learning to enforce solution sparsity.ResultsThe original multi-class SVM of Crammer and Singer is effective for multi-class classification but does not conduct variable selection. We improved the method by introducing soft-thresholding type penalties to incorporate variable selection into multi-class classification for high dimensional data. The new methods were applied to simulated data and two cancer gene expression data sets. The results demonstrate that the new methods can select a small number of genes for building accurate multi-class classification rules. Furthermore, the important genes selected by the methods overlap significantly, suggesting general agreement among different variable selection schemes.ConclusionsHigh accuracy and sparsity make the new methods attractive for cancer diagnostics with gene expression data and defining targets of therapeutic intervention.Availability: The source MATLAB code are available from http://math.arizona.edu/~hzhang/software.html.