A flexible symbolic regression method for constructing interpretable clinical prediction models.

Machine learning (ML) models trained for triggering clinical decision support (CDS) are typically either accurate or interpretable but not both. Scaling CDS to the panoply of clinical use cases while mitigating risks to patients will require many ML models be intuitively interpretable for clinicians. To this end, we adapted a symbolic regression method, coined the feature engineering automation tool (FEAT), to train concise and accurate models from high-dimensional electronic health record (EHR) data. We first present an in-depth application of FEAT to classify hypertension, hypertension with unexplained hypokalemia, and apparent treatment-resistant hypertension (aTRH) using EHR data for 1200 subjects receiving longitudinal care in a large healthcare system. FEAT models trained to predict phenotypes adjudicated by chart review had equivalent or higher discriminative performance (p < 0.001) and were at least three times smaller (p < 1 × 10-6) than other potentially interpretable models. For aTRH, FEAT generated a six-feature, highly discriminative (positive predictive value = 0.70, sensitivity = 0.62), and clinically intuitive model. To assess the generalizability of the approach, we tested FEAT on 25 benchmark clinical phenotyping tasks using the MIMIC-III critical care database. Under comparable dimensionality constraints, FEAT's models exhibited higher area under the receiver-operating curve scores than penalized linear models across tasks (p < 6 × 10-6). In summary, FEAT can train EHR prediction models that are both intuitively interpretable and accurate, which should facilitate safe and effective scaling of ML-triggered CDS to the panoply of potential clinical use cases and healthcare practices.

Artificial Intelligence and Mapping a New Direction in Laboratory Medicine: A Review.

BACKGROUND: Modern artificial intelligence (AI) and machine learning (ML) methods are now capable of completing tasks with performance characteristics that are comparable to those of expert human operators. As a result, many areas throughout healthcare are incorporating these technologies, including in vitro diagnostics and, more broadly, laboratory medicine. However, there are limited literature reviews of the landscape, likely future, and challenges of the application of AI/ML in laboratory medicine. CONTENT: In this review, we begin with a brief introduction to AI and its subfield of ML. The ensuing sections describe ML systems that are currently in clinical laboratory practice or are being proposed for such use in recent literature, ML systems that use laboratory data outside the clinical laboratory, challenges to the adoption of ML, and future opportunities for ML in laboratory medicine. SUMMARY: AI and ML have and will continue to influence the practice and scope of laboratory medicine dramatically. This has been made possible by advancements in modern computing and the widespread digitization of health information. These technologies are being rapidly developed and described, but in comparison, their implementation thus far has been modest. To spur the implementation of reliable and sophisticated ML-based technologies, we need to establish best practices further and improve our information system and communication infrastructure. The participation of the clinical laboratory community is essential to ensure that laboratory data are sufficiently available and incorporated conscientiously into robust, safe, and clinically effective ML-supported clinical diagnostics.

Why Is the Electronic Health Record So Challenging for Research and Clinical Care?

BACKGROUND: The electronic health record (EHR) has become increasingly ubiquitous. At the same time, health professionals have been turning to this resource for access to data that is needed for the delivery of health care and for clinical research. There is little doubt that the EHR has made both of these functions easier than earlier days when we relied on paper-based clinical records. Coupled with modern database and data warehouse systems, high-speed networks, and the ability to share clinical data with others are large number of challenges that arguably limit the optimal use of the EHR OBJECTIVES: Our goal was to provide an exhaustive reference for those who use the EHR in clinical and research contexts, but also for health information systems professionals as they design, implement, and maintain EHR systems. METHODS: This study includes a panel of 24 biomedical informatics researchers, information technology professionals, and clinicians, all of whom have extensive experience in design, implementation, and maintenance of EHR systems, or in using the EHR as clinicians or researchers. All members of the panel are affiliated with Penn Medicine at the University of Pennsylvania and have experience with a variety of different EHR platforms and systems and how they have evolved over time. RESULTS: Each of the authors has shared their knowledge and experience in using the EHR in a suite of 20 short essays, each representing a specific challenge and classified according to a functional hierarchy of interlocking facets such as usability and usefulness, data quality, standards, governance, data integration, clinical care, and clinical research. CONCLUSION: We provide here a set of perspectives on the challenges posed by the EHR to clinical and research users.

The effect of the Covid-19 shutdown on glycemic testing and control.

BACKGROUND: The coronavirus disease 2019 (COVID-19) pandemic caused a halt to in-person ambulatory care. We evaluated how the reduction in access to care affected HbA1c testing and patient HbA1c levels. METHODS: HbA1c data from 11 institutions were extracted to compare testing volume and the percentage of abnormal results between a pre-pandemic period (January-June 2019, period 1) and a portion of the COVID-19 pandemic period (Jan-June 2020, period 2). HbA1c results greater than 6.4% were categorized as abnormal. RESULTS: HbA1C testing volumes decreased in March, April and May by 23, 61 and 40% relative to the corresponding months in 2019. The percentage of abnormal results increased in April, May and June (25, 23, 9%). On average, we found that the frequency of abnormal results increased by 0.31% for every 1% decrease in testing volume (p < 0.0005). CONCLUSION: HbA1c testing volume for outpatients decreased by up to 70% during the early months of the pandemic. The decrease in testing was associated with an increase in abnormal HbA1c results.

Side-Effects of COVID-19 on Patient Care: An INR Story.

BACKGROUND: Numerous studies have documented reduced access to patient care due to the COVID-19 pandemic, including access to diagnostic or screening tests, prescription medications, and treatment for an ongoing condition. In the context of clinical management for venous thromboembolism, this could result in suboptimal therapy with warfarin. We aimed to determine the impact of the pandemic on utilization of International Normalized Ratio (INR) testing and the percentage of high and low results. METHODS: INR data from 11 institutions were extracted to compare testing volume and the percentage of INR results ≥3.5 and ≤1.5 between a pre-pandemic period (January-June 2019, period 1) and a portion of the COVID-19 pandemic period (January-June 2020, period 2). The analysis was performed for inpatient and outpatient cohorts. RESULTS: Testing volumes showed relatively little change in January and February, followed by a significant decrease in March, April, and May, and then returned to baseline in June. Outpatient testing showed a larger percentage decrease in testing volume compared to inpatient testing. At 10 of the 11 study sites, we observed an increase in the percentage of abnormal high INR results as test volumes decreased, primarily among outpatients. CONCLUSION: The COVID-19 pandemic impacted INR testing among outpatients which may be attributable to several factors. Increased supratherapeutic INR results during the pandemic period when there was reduced laboratory utilization and access to care is concerning because of the risk of adverse bleeding events in this group of patients. This could be mitigated in the future by offering drive-through testing and/or widespread implementation of home INR monitoring.

Inference for the Case Probability in High-dimensional Logistic Regression.

Labeling patients in electronic health records with respect to their statuses of having a disease or condition, i.e. case or control statuses, has increasingly relied on prediction models using high-dimensional variables derived from structured and unstructured electronic health record data. A major hurdle currently is a lack of valid statistical inference methods for the case probability. In this paper, considering high-dimensional sparse logistic regression models for prediction, we propose a novel bias-corrected estimator for the case probability through the development of linearization and variance enhancement techniques. We establish asymptotic normality of the proposed estimator for any loading vector in high dimensions. We construct a confidence interval for the case probability and propose a hypothesis testing procedure for patient case-control labelling. We demonstrate the proposed method via extensive simulation studies and application to real-world electronic health record data.

Testing for Primary Aldosteronism and Mineralocorticoid Receptor Antagonist Use Among U.S. Veterans : A Retrospective Cohort Study.

BACKGROUND: Primary aldosteronism is a common cause of treatment-resistant hypertension. However, evidence from local health systems suggests low rates of testing for primary aldosteronism. OBJECTIVE: To evaluate testing rates for primary aldosteronism and evidence-based hypertension management in patients with treatment-resistant hypertension. DESIGN: Retrospective cohort study. SETTING: U.S. Veterans Health Administration. PARTICIPANTS: Veterans with apparent treatment-resistant hypertension (n = 269 010) from 2000 to 2017, defined as either 2 blood pressures (BPs) of at least 140 mm Hg (systolic) or 90 mm Hg (diastolic) at least 1 month apart during use of 3 antihypertensive agents (including a diuretic), or hypertension requiring 4 antihypertensive classes. MEASUREMENTS: Rates of primary aldosteronism testing (plasma aldosterone-renin) and the association of testing with evidence-based treatment using a mineralocorticoid receptor antagonist (MRA) and with longitudinal systolic BP. RESULTS: 4277 (1.6%) patients who were tested for primary aldosteronism were identified. An index visit with a nephrologist (hazard ratio [HR], 2.05 [95% CI, 1.66 to 2.52]) or an endocrinologist (HR, 2.48 [CI, 1.69 to 3.63]) was associated with a higher likelihood of testing compared with primary care. Testing was associated with a 4-fold higher likelihood of initiating MRA therapy (HR, 4.10 [CI, 3.68 to 4.55]) and with better BP control over time. LIMITATIONS: Predominantly male cohort, retrospective design, susceptibility of office BPs to misclassification, and lack of confirmatory testing for primary aldosteronism. CONCLUSION: In a nationally distributed cohort of veterans with apparent treatment-resistant hypertension, testing for primary aldosteronism was rare and was associated with higher rates of evidence-based treatment with MRAs and better longitudinal BP control. The findings reinforce prior observations of low adherence to guideline-recommended practices in smaller health systems and underscore the urgent need for improved management of patients with treatment-resistant hypertension. PRIMARY FUNDING SOURCE: National Institutes of Health.

Ideas for how informaticians can get involved with COVID-19 research.

The coronavirus disease 2019 (COVID-19) pandemic has had a significant impact on population health and wellbeing. Biomedical informatics is central to COVID-19 research efforts and for the delivery of healthcare for COVID-19 patients. Critical to this effort is the participation of informaticians who typically work on other basic science or clinical problems. The goal of this editorial is to highlight some examples of COVID-19 research areas that could benefit from informatics expertise. Each research idea summarizes the COVID-19 application area, followed by an informatics methodology, approach, or technology that could make a contribution. It is our hope that this piece will motivate and make it easy for some informaticians to adopt COVID-19 research projects.

Marked Up-Regulation of ACE2 in Hearts of Patients With Obstructive Hypertrophic Cardiomyopathy: Implications for SARS-CoV-2-Mediated COVID-19.

OBJECTIVE: To explore the transcriptomic differences between patients with hypertrophic cardiomyopathy (HCM) and controls. PATIENTS AND METHODS: RNA was extracted from cardiac tissue flash frozen at therapeutic surgical septal myectomy for 106 patients with HCM and 39 healthy donor hearts. Expression profiling of 37,846 genes was performed using the Illumina Human HT-12v3 Expression BeadChip. All patients with HCM were genotyped for pathogenic variants causing HCM. Technical validation was performed using quantitative real-time polymerase chain reaction (qRT-PCR) and Western blot. This study was started on January 1, 1999, and final analysis was completed on April 20, 2020. RESULTS: Overall, 22% of the transcriptome (8443 of 37,846 genes) was expressed differentially between HCM and control tissues. Analysis by genotype revealed that gene expression changes were similar among genotypic subgroups of HCM, with only 4% (1502 of 37,846) to 6% (2336 of 37,846) of the transcriptome exhibiting differential expression between genotypic subgroups. The qRT-PCR confirmed differential expression in 92% (11 of 12 genes) of tested transcripts. Notably, in the context of coronavirus disease 2019 (COVID-19), the transcript for angiotensin I converting enzyme 2 (ACE2), a negative regulator of the angiotensin system, was the single most up-regulated gene in HCM (fold-change, 3.53; q-value =1.30×10-23), which was confirmed by qRT-PCR in triplicate (fold change, 3.78; P=5.22×10-4), and Western blot confirmed greater than 5-fold overexpression of ACE2 protein (fold change, 5.34; P=1.66×10-6). CONCLUSION: More than 20% of the transcriptome is expressed differentially between HCM and control tissues. Importantly, ACE2 was the most up-regulated gene in HCM, indicating perhaps the heart's compensatory effort to mount an antihypertrophic, antifibrotic response. However, given that the severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) uses ACE2 for viral entry, this 5-fold increase in ACE2 protein may confer increased risk for COVID-19 manifestations and outcomes in patients with increased ACE2 transcript expression and protein levels in the heart.

Challenges in obesity and primary aldosteronism: Diagnosis and treatment.

BACKGROUND: Obese patients may have unrecognized primary aldosteronism due to high rates of concomitant hypertension. We hypothesized that obesity impacts the diagnosis and management of patients with primary aldosteronism. METHODS: We conducted a retrospective analysis of all primary aldosteronism patients (n = 418) who underwent adrenal vein sampling (1997-2017). Patients were classified by body mass index as obese (body mass index ≥35) or nonobese (body mass index <35) and diagnostic evaluation was compared between groups. Within the operative cohort (n = 285), primary outcomes were changes in both blood pressure and antihypertensive medications after adrenalectomy. Secondary outcome was clinical resolution by Primary Aldosteronism Surgery Outcomes criteria. RESULTS: Thirty-five percent of patients were obese. Obese patients were more likely to be male (67.8% vs 56.1%, P = .025), somewhat younger (51.5 vs 54.4 years old, P < .012), and require more preoperative antihypertensive medications (6.7 vs 5.7, P = .04) than nonobese patients. Obese patients had lesser rates of radiologic evidence of adrenal tumors (68.4 vs 77.9%, P = .038) despite similar rates of lateralization on adrenal vein sampling. In the operative subset, obese patients had somewhat smaller tumors on final pathology (1.1 vs 1.5 cm, P = .014) but similar rates of complete and partial clinical resolution (P = 1.000). CONCLUSION: Obese primary aldosteronism patients have lesser rates of localization by imaging, likely due to smaller tumor size, however, experience similar benefit from adrenalectomy.

A maximum likelihood approach to electronic health record phenotyping using positive and unlabeled patients.

OBJECTIVE: Phenotyping patients using electronic health record (EHR) data conventionally requires labeled cases and controls. Assigning labels requires manual medical chart review and therefore is labor intensive. For some phenotypes, identifying gold-standard controls is prohibitive. We developed an accurate EHR phenotyping approach that does not require labeled controls. MATERIALS AND METHODS: Our framework relies on a random subset of cases, which can be specified using an anchor variable that has excellent positive predictive value and sensitivity independent of predictors. We proposed a maximum likelihood approach that efficiently leverages data from the specified cases and unlabeled patients to develop logistic regression phenotyping models, and compare model performance with existing algorithms. RESULTS: Our method outperformed the existing algorithms on predictive accuracy in Monte Carlo simulation studies, application to identify hypertension patients with hypokalemia requiring oral supplementation using a simulated anchor, and application to identify primary aldosteronism patients using real-world cases and anchor variables. Our method additionally generated consistent estimates of 2 important parameters, phenotype prevalence and the proportion of true cases that are labeled. DISCUSSION: Upon identification of an anchor variable that is scalable and transferable to different practices, our approach should facilitate development of scalable, transferable, and practice-specific phenotyping models. CONCLUSIONS: Our proposed approach enables accurate semiautomated EHR phenotyping with minimal manual labeling and therefore should greatly facilitate EHR clinical decision support and research.

Falsely Increased Plasma Lactate Dehydrogenase without Hemolysis Following Transport through Pneumatic Tube System.

BACKGROUND: Lactate dehydrogenase (LDH) is a nonspecific biomarker for diseases including lymphoma. Serum and plasma are generally considered interchangeable for LDH testing. Investigation into falsely increased plasma LDH concentration results led to the hypothesis that a workflow change that included pneumatic tube system (PTS) transportation caused the errors. The following study was conducted to test the hypothesis. METHODS: Plasma and serum separator tube samples were each drawn in duplicate, centrifuged, transported either through the PTS or by hand courier, and evaluated by means of clinical chemistry and hematology assays. Smear slides were made out of the plasma and examined. Aggregate patient results before and after the PTS workflow change were compared. RESULTS: In post-PTS plasma samples, LDH activity was 26%-149% higher. Similarly, white blood cells (WBCs) were 14- to 156-fold higher and platelets were 1- to 13-fold higher. Smear examination revealed dramatically more cells and cell fragments. No significant hemolysis was observed in plasma by chemistry hemolysis indices or hemoglobin testing. These effects were not observed in similarly transported serum samples in gel separator tubes. Aggregate LDH patient results, including moving medians, demonstrated dramatic changes following PTS workflow implementation. CONCLUSIONS: PTS transportation led to falsely increased LDH concentration in plasma. These LDH concentration elevations are not heralded by standard indicators of hemolysis. These errors can be prevented by restricting LDH concentration testing to serum collected in gel separator tubes. Moving patient statistics can effectively detect important testing process changes not revealed by external QC or indices.

A gene-centric strategy for identifying disease-causing rare variants in dilated cardiomyopathy.

PURPOSE: We evaluated strategies for identifying disease-causing variants in genetic testing for dilated cardiomyopathy (DCM). METHODS: Cardiomyopathy gene panel testing was performed in 532 DCM patients and 527 healthy control subjects. Rare variants in 41 genes were stratified using variant-level and gene-level characteristics. RESULTS: A majority of DCM cases and controls carried rare protein-altering cardiomyopathy gene variants. Variant-level characteristics alone had limited discriminative value. Differentiation between groups was substantially improved by addition of gene-level information that incorporated ranking of genes based on literature evidence for disease association. The odds of DCM were increased to nearly 9-fold for truncating variants or high-impact missense variants in the subset of 14 genes that had the strongest biological links to DCM (P <0.0001). For some of these genes, DCM-associated variants appeared to be clustered in key protein functional domains. Multiple rare variants were present in many family probands, however, there was generally only one "driver" pathogenic variant that cosegregated with disease. CONCLUSION: Rare variants in cardiomyopathy genes can be effectively stratified by combining variant-level and gene-level information. Prioritization of genes based on their a priori likelihood of disease causation is a key factor in identifying clinically actionable variants in cardiac genetic testing.

Quality Control Practices for Chemistry and Immunochemistry in a Cohort of 21 Large Academic Medical Centers.

OBJECTIVES: In the United States, minimum standards for quality control (QC) are specified in federal law under the Clinical Laboratory Improvement Amendment and its revisions. Beyond meeting this required standard, laboratories have flexibility to determine their overall QC program. METHODS: We surveyed chemistry and immunochemistry QC procedures at 21 clinical laboratories within leading academic medical centers to assess if standardized QC practices exist for chemistry and immunochemistry testing. RESULTS: We observed significant variation and unexpected similarities in practice across laboratories, including QC frequency, cutoffs, number of levels analyzed, and other features. CONCLUSIONS: This variation in practice indicates an opportunity exists to establish an evidence-based approach to QC that can be generalized across institutions.

Efficient Detection of Copy Number Mutations in PMS2 Exons with a Close Homolog.

Detection of 3' PMS2 copy-number mutations that cause Lynch syndrome is difficult because of highly homologous pseudogenes. To improve the accuracy and efficiency of clinical screening for these mutations, we developed a new method to analyze standard capture-based, next-generation sequencing data to identify deletions and duplications in PMS2 exons 9 to 15. The approach captures sequences using PMS2 targets, maps sequences randomly among regions with equal mapping quality, counts reads aligned to homologous exons and introns, and flags read count ratios outside of empirically derived reference ranges. The method was trained on 1352 samples, including 8 known positives, and tested on 719 samples, including 17 known positives. Clinical implementation of the first version of this method detected new mutations in the training (N = 7) and test (N = 2) sets that had not been identified by our initial clinical testing pipeline. The described final method showed complete sensitivity in both sample sets and false-positive rates of 5% (training) and 7% (test), dramatically decreasing the number of cases needing additional mutation evaluation. This approach leveraged the differences between gene and pseudogene to distinguish between PMS2 and PMS2CL copy-number mutations. These methods enable efficient and sensitive Lynch syndrome screening for 3' PMS2 copy-number mutations and may be applied similarly to other genomic regions with highly homologous pseudogenes.

Variability and Error in Cardiac Troponin Testing: An ACLPS Critical Review.

OBJECTIVES: To provide a comprehensive overview of the complexities associated with cardiac troponin (cTn) testing. An emphasis is placed on the sources of error, organized into the preanalytical, analytical, and postanalytical phases of the testing pathway. Controversial areas are also explored. METHODS: A case scenario and review of the relevant literature describing laboratory considerations involving cTn testing are described. RESULTS: Advanced comprehension of the specific assay used in a given laboratory is necessary for optimal reporting, utilization, and quality monitoring of cTn. CONCLUSIONS: cTn assays are reliable diagnostic tests for acute myocardial infarction, but understanding their limitations is required for appropriate result interpretation.

Endogenous alkaline phosphatase interference in cardiac troponin I and other sensitive chemiluminescence immunoassays that use alkaline phosphatase activity for signal amplification.

BACKGROUND: False positive cardiac troponin results can lead to inappropriate diagnosis. Our laboratory workflow includes systematic quality practices to identify false positive cardiac troponin I (cTnI) results reported by the DxI AccuTnI+3 assay, which uses alkaline phosphatase (ALP) for signal amplification. Recently, a sample with elevated cTnI failed our quality standards and was found to have extremely elevated endogenous ALP activity. The objective of this study was to determine the true cTnI concentration and evaluate whether ALP was the source of interference. METHODS: The suspicious cTnI result was evaluated by repeat analyses, dilution, heterophile blocking treatment, alternative methodology (Vista), and heat treatment. Purified ALP was added to reference serum and we quantified DxI cTnI and human chorionic gonadotropin (hCG). Next, cTnI and/or hCG was measured in specimens with normal (N=20) or elevated (N=26) ALP using DxI and Vista assays. Finally, cTnI was quantified using a prototype, ALP-dependent high-sensitivity assay. RESULTS: The sentinel sample's DxI-cTnI results were imprecise on repeat, linear on dilution, unaffected by heterophile blocking antibodies, and correlated with ALP lability following heat treatment. The Vista-cTnI concentrations were ~7-fold lower. Addition of purified ALP to reference serum linearly increased the DxI-cTnI results. DxI-hCG results also appeared affected by ALP. Several independent patients' specimens with elevated ALP appeared to have falsely elevated DxI-cTnI and DxI-hCG. CONCLUSIONS: Elevated ALP can interfere with contemporary, ALP-dependent immunoassays, including DxI-cTnI and DxI-hCG. The validation of such methods should include evaluations for endogenous ALP interference. Specimens with ALP >1000U/L and elevated DxI-cTnI should be evaluated for ALP interference.

Actionable exomic incidental findings in 6503 participants: challenges of variant classification.

Recommendations for laboratories to report incidental findings from genomic tests have stimulated interest in such results. In order to investigate the criteria and processes for assigning the pathogenicity of specific variants and to estimate the frequency of such incidental findings in patients of European and African ancestry, we classified potentially actionable pathogenic single-nucleotide variants (SNVs) in all 4300 European- and 2203 African-ancestry participants sequenced by the NHLBI Exome Sequencing Project (ESP). We considered 112 gene-disease pairs selected by an expert panel as associated with medically actionable genetic disorders that may be undiagnosed in adults. The resulting classifications were compared to classifications from other clinical and research genetic testing laboratories, as well as with in silico pathogenicity scores. Among European-ancestry participants, 30 of 4300 (0.7%) had a pathogenic SNV and six (0.1%) had a disruptive variant that was expected to be pathogenic, whereas 52 (1.2%) had likely pathogenic SNVs. For African-ancestry participants, six of 2203 (0.3%) had a pathogenic SNV and six (0.3%) had an expected pathogenic disruptive variant, whereas 13 (0.6%) had likely pathogenic SNVs. Genomic Evolutionary Rate Profiling mammalian conservation score and the Combined Annotation Dependent Depletion summary score of conservation, substitution, regulation, and other evidence were compared across pathogenicity assignments and appear to have utility in variant classification. This work provides a refined estimate of the burden of adult onset, medically actionable incidental findings expected from exome sequencing, highlights challenges in variant classification, and demonstrates the need for a better curated variant interpretation knowledge base.