Real-world Electronic Health Record Data Research Articles

Using real-world electronic health record data to predict the development of 12 cancer-related symptoms in the context of multimorbidity.

This study uses electronic health record (EHR) data to predict 12 common cancer symptoms, assessing the efficacy of machine learning (ML) models in identifying symptom influencers. We analyzed EHR data of 8156 adults diagnosed with cancer who underwent cancer treatment from 2017 to 2020. Structured and unstructured EHR data were sourced from the Enterprise Data Warehouse for Research at the University of Iowa Hospital and Clinics. Several predictive models, including logistic regression, random forest (RF), and XGBoost, were employed to forecast symptom development. The performances of the models were evaluated by F1-score and area under the curve (AUC) on the testing set. The SHapley Additive exPlanations framework was used to interpret these models and identify the predictive risk factors associated with fatigue as an exemplar. The RF model exhibited superior performance with a macro average AUC of 0.755 and an F1-score of 0.729 in predicting a range of cancer-related symptoms. For instance, the RF model achieved an AUC of 0.954 and an F1-score of 0.914 for pain prediction. Key predictive factors identified included clinical history, cancer characteristics, treatment modalities, and patient demographics depending on the symptom. For example, the odds ratio (OR) for fatigue was significantly influenced by allergy (OR = 2.3, 95% CI: 1.8-2.9) and colitis (OR = 1.9, 95% CI: 1.5-2.4). Our research emphasizes the critical integration of multimorbidity and patient characteristics in modeling cancer symptoms, revealing the considerable influence of chronic conditions beyond cancer itself. We highlight the potential of ML for predicting cancer symptoms, suggesting a pathway for integrating such models into clinical systems to enhance personalized care and symptom management.

Risk Assessment and Diagnosis Code Prediction from Electronic Health Records (EHR) using deep learning

Knowledge in the medical field is often expressed by distinct and subjective norms. Much research and development efforts have focused on using deep learning algorithms to predict the likelihood of illnesses from EHR in recent years. When it comes to risk prediction, deep learning-based techniques outperform more conventional machine learning models. But nothing in the literature fully accounts for what doctors already know, such as the connections between illnesses and their risk factors. This research examines the use of Multi-layer Perceptron models for the categorisation of diagnoses in electronic health records. The raw data and a modified version of the EHR dataset are used to train two MLPs with distinct topologies. For comparative purposes, a Random Forest is used as a baseline. To phenotype patients using their electronic health records, we provide a deep learning method. Predictive modelling of chronic illnesses is the particular scenario used to verify the suggested model on a real-world EHR data warehouse. Many deep learning applications on EHRs have been effective, and there is still a lot of potential to be realised. It has been discovered that deep learning models can learn from the limited EHR dataset, but not to a level where they outperform the baseline model.

Evaluation of score-based tertiary triage policies during the COVID-19 pandemic: simulation study with real-world intensive care data.

The explicit prohibition of discontinuing intensive care unit (ICU) treatment that has already begun by the newly established German Triage Act in favor of new patients with better prognoses (tertiary triage) under crisis conditions may prevent saving as many patients as possible and therefore may violate the international well-accepted premise of undertaking the "best for the most" patients. During the COVID-19 pandemic, authorities set up lockdown measures and infection-prevention strategies to avoid an overburdened health-care system. In cases of situational overload of ICU resources, when transporting options are exhausted, the question of atertiary triage of patients arises. We provide data-driven analyses of score- and non-score-based tertiary triage policies using simulation and real-world electronic health record data in aCOVID-19 setting. Ten different triage policies, for example, based on the Simplified Acute Physiology Score (SAPSII), are compared based on the resulting mortality in the ICU and inferential statistics. Our study shows that score-based tertiary triage policies outperform non-score-based tertiary triage policies including compliance with the German Triage Act. Based on our simulation model, aSAPSII score-based tertiary triage policy reduces mortality in the ICU by up to 18percentage points. The longer the queue of critical care patients waiting for ICU treatment and the larger the maximum number of patients subject to tertiary triage, the greater the effect on the reduction of mortality in the ICU. ASAPSII score-based tertiary triage policy was superior in our simulation model. Random allocation or "first come, first served" policies yield the lowest survival rates, as will adherence to the new German Triage Act. An interdisciplinary discussion including an ethical and legal perspective is important for the social interpretation of our data-driven results.

Abstract 6467: Causal personalized treatment estimation framework using real-world electronic health record (EHR) data to inform cancer care decisions

Abstract Background: The average treatment effects (ATEs) from randomized controlled trials often do not generalize across populations, or they are not sufficiently personalized to capture the benefits and risks to individual patients. Real-world data aids in validating and extending trial results, but can introduce biases that invalidate the causal interpretation. Here, we propose a robust statistical framework, which uses real-world EHR data to evaluate the efficacy of cancer treatments at both population and individual levels. We demonstrate the ability of our framework to deliver personalized insights in metastatic prostate cancer treatments. Methods: We assembled a retrospective cohort of metastatic prostate cancer patients from a large academic medical center. We gathered demographic, comorbidity, cancer treatment, and outcome data from the EHR or natural language processing of clinician narratives, and social determinants of health from Census data. We used doubly robust machine learning with bootstrap standard errors to compute the ATE and personalized treatment estimates (PTEs), comparing the impact of ADT (androgen depravation therapy) + chemotherapy (ACT) and ADT + combination anti-androgen therapy (ACAT) on percentage reduction in PSA levels in metastatic prostate cancer. Results: The baseline characteristics of our cohort are summarized in Table 1a. Our approach found that ACAT reduces PSA levels significantly more than ACT, and captured the heterogeneity in treatment effects across individuals (Table 1b). Through simulation studies, we demonstrated that any reasonable bias from unobserved confounding or missing information was small compared to the estimated treatment effects and their standard errors. Conclusion: We presented a doubly robust framework and demonstrated its utility in analyzing real-world data, enabling us to uncover the diversity in treatment responses among individuals with metastatic prostate cancer. Results summary a. Baseline demographic characteristics. ADT + chemotherapy N = 302 ADT + combination therapy N = 114 Treatments Any ADT + Docetaxel or Cabazitaxel Any ADT + Abiraterone or apalutamide or Enzalutamide Age Mean (SD) 69.3 (8.76) 66.7 (8.17) Median [Min, Max] 69.0 [47.0, 94.0] 66.5 [46.0, 86.0] Race Asian 8 (2.6%) 2 (1.8%) Black or African American 48 (15.9%) 7 (6.1%) White or Caucasian 224 (74.2%) 99 (86.8%) Other 22 (7.3%) 6 (5.3%) Max PSA treatment start (ng/mL) Mean (SD) 34.9 (98.5) 83.3 (448) b. Treatment effects estimation ATE (95% CI) 32.0% [24.6%, 40.8%] PTEs: 25th and 75th percentiles 29.0%, 34.3% Citation Format: Sumana Srivatsa, David Walsh, Mayada Aljehani, Jason Weinreb, Nathan Becker, Benjamin H. Ellis, Renee George, Xingyao Chen, Naim Matasci, Reva Basho, Mitchell E. Gross, Christine L. Swisher. Causal personalized treatment estimation framework using real-world electronic health record (EHR) data to inform cancer care decisions [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2024; Part 1 (Regular Abstracts); 2024 Apr 5-10; San Diego, CA. Philadelphia (PA): AACR; Cancer Res 2024;84(6_Suppl):Abstract nr 6467.

Prediction of Effectiveness and Toxicities of Immune Checkpoint Inhibitors Using Real-World Patient Data.

Although immune checkpoint inhibitors (ICIs) have improved outcomes in certain patients with cancer, they can also cause life-threatening immunotoxicities. Predicting immunotoxicity risks alongside response could provide a personalized risk-benefit profile, inform therapeutic decision making, and improve clinical trial cohort selection. We aimed to build a machine learning (ML) framework using routine electronic health record (EHR) data to predict hepatitis, colitis, pneumonitis, and 1-year overall survival. Real-world EHR data of more than 2,200 patients treated with ICI through December 31, 2018, were used to develop predictive models. Using a prediction time point of ICI initiation, a 1-year prediction time window was applied to create binary labels for the four outcomes for each patient. Feature engineering involved aggregating laboratory measurements over appropriate time windows (60-365 days). Patients were randomly partitioned into training (80%) and test (20%) sets. Random forest classifiers were developed using a rigorous model development framework. The patient cohort had a median age of 63 years and was 61.8% male. Patients predominantly had melanoma (37.8%), lung cancer (27.3%), or genitourinary cancer (16.4%). They were treated with PD-1 (60.4%), PD-L1 (9.0%), and CTLA-4 (19.7%) ICIs. Our models demonstrate reasonably strong performance, with AUCs of 0.739, 0.729, 0.755, and 0.752 for the pneumonitis, hepatitis, colitis, and 1-year overall survival models, respectively. Each model relies on an outcome-specific feature set, though some features are shared among models. To our knowledge, this is the first ML solution that assesses individual ICI risk-benefit profiles based predominantly on routine structured EHR data. As such, use of our ML solution will not require additional data collection or documentation in the clinic.

A Post-Marketing Drug Evaluation Framework Based on Real-World Electronic Health Records Data.

Real-world data (RWD) could be a new way to evaluate the safety and efficacy of post-marketing drugs, while there is no common method for how to use RWD for drug evaluation. In this paper, we present a framework for real-world drug evaluation based on electronic medical record (EHR) data. We designed a data model customized for post-marketing drug evaluation and a unified post-marketing drug evaluation pipeline. The proposed framework can be applied to drug evaluations with different study paradigms for different purposes by flexible use of the proposed data model and pipeline. A prototype system has been developed according to the framework. Real-world EHRs in a tertiary hospital in China between 2010 and 2020 were converted to the proposed data model, and as a test case, we conducted a research on the risk of allergic reactions to cefodizime and ceftriaxone using the prototype system.

EHR-Safe: generating high-fidelity and privacy-preserving synthetic electronic health records

Privacy concerns often arise as the key bottleneck for the sharing of data between consumers and data holders, particularly for sensitive data such as Electronic Health Records (EHR). This impedes the application of data analytics and ML-based innovations with tremendous potential. One promising approach for such privacy concerns is to instead use synthetic data. We propose a generative modeling framework, EHR-Safe, for generating highly realistic and privacy-preserving synthetic EHR data. EHR-Safe is based on a two-stage model that consists of sequential encoder-decoder networks and generative adversarial networks. Our innovations focus on the key challenging aspects of real-world EHR data: heterogeneity, sparsity, coexistence of numerical and categorical features with distinct characteristics, and time-varying features with highly-varying sequence lengths. Under numerous evaluations, we demonstrate that the fidelity of EHR-Safe is almost-identical with real data (<3% accuracy difference for the models trained on them) while yielding almost-ideal performance in practical privacy metrics.

Parkinson's disease diagnosis codes are insufficiently accurate for electronic health record research and differ by race

BackgroundThere are no evidence-based guidelines for data cleaning of electronic health record (EHR) databases in Parkinson's disease (PD). Previous filtering criteria have primarily used the 9th International Statistical Classification of Diseases and Related Health Problems (ICD) with variable accuracy for true PD cases. Prior studies have not excluded atypical or drug-induced parkinsonism, and little is known about differences in accuracy by race. ObjectiveTo determine if excluding parkinsonism diagnoses improves accuracy of ICD-9 and -10 PD diagnosis codes. MethodsWe included ≥2 instances of an ICD-9 and/or −10 code for PD. We removed any records with at least one code indicating atypical or drug-induced parkinsonism first in all races, and then in Non-Hispanic White and Black patients. We manually reviewed 100 randomly selected charts per group before and after filtering, and performed a test of proportion (null hypothesis 0.5) for confirmed PD. Results5633 records had ≥2 instances of a PD code. 2833 remained after filtering. The rate of true PD cases was low before and after filtering to remove parkinsonism codes (0.55 vs. 0.51, p = 0.84). Accuracy was lowest in Black patients before filtering (0.48, p = 0.69), but filtering had a greater (though modest) impact on accuracy (0.68, p < 0.001). ConclusionsThere was inadequate accuracy of PD diagnosis codes in the largest study of ICD-9 and -10 codes. Accuracy was lowest in Black patients but improved the most with removing other parkinsonism codes. This highlights the limitations of using current real-world EHR data in PD research and need for further study.

AI-assisted prediction of differential response to antidepressant classes using electronic health records

Antidepressant selection is largely a trial-and-error process. We used electronic health record (EHR) data and artificial intelligence (AI) to predict response to four antidepressants classes (SSRI, SNRI, bupropion, and mirtazapine) 4 to 12 weeks after antidepressant initiation. The final data set comprised 17,556 patients. Predictors were derived from both structured and unstructured EHR data and models accounted for features predictive of treatment selection to minimize confounding by indication. Outcome labels were derived through expert chart review and AI-automated imputation. Regularized generalized linear model (GLM), random forest, gradient boosting machine (GBM), and deep neural network (DNN) models were trained and their performance compared. Predictor importance scores were derived using SHapley Additive exPlanations (SHAP). All models demonstrated similarly good prediction performance (AUROCs ≥ 0.70, AUPRCs ≥ 0.68). The models can estimate differential treatment response probabilities both between patients and between antidepressant classes for the same patient. In addition, patient-specific factors driving response probabilities for each antidepressant class can be generated. We show that antidepressant response can be accurately predicted from real-world EHR data with AI modeling, and our approach could inform further development of clinical decision support systems for more effective treatment selection.

Using Real-World Electronic Health Record Data to Assess Chronic Disease Screening in Children: A Case Study of Non-Alcoholic Fatty Liver Disease.

Background: Data sources for assessing pediatric chronic diseases and associated screening practices are rare. One example is non-alcoholic fatty liver disease (NAFLD), a common chronic liver disease prevalent among children with overweight and obesity. If undetected, NAFLD can cause liver damage. Guidelines recommend screening for NAFLD using alanine aminotransferase (ALT) tests in children ≥9 years with obesity or those with overweight and cardiometabolic risk factors. This study explores how real-world data from electronic health records (EHRs) can be used to study NAFLD screening and ALT elevation. Research Design: Using IQVIA's Ambulatory Electronic Medical Record database, we studied patients 2-19 years of age with body mass index ≥85th percentile. Using a 3-year observation period (January 1, 2019 to December 31, 2021), ALT results were extracted and assessed for elevation (≥1 ALT result ≥22.1 U/L for females and ≥25.8 U/L for males). Patients with liver disease (including NAFLD) or receiving hepatotoxic medications during 2017-2018 were excluded. Results: Among 919,203 patients 9-19 years of age, only 13% had ≥1 ALT result, including 14% of patients with obesity and 17% of patients with severe obesity. ALT results were identified for 5% of patients 2-8 years of age. Of patients with ALT results, 34% of patients 2-8 years of age and 38% of patients 9-19 years of age had ALT elevation. Males 9-19 years of age had a higher prevalence of ALT elevation than females (49% vs. 29%). Conclusions: EHR data offered novel insights into NAFLD screening: despite screening recommendations, ALT results among children with excess weight were infrequent. Among those with ALT results, ALT elevation was common, underscoring the importance of screening for early disease detection.

KIT-LSTM: Knowledge-guided Time-aware LSTM for Continuous Clinical Risk Prediction.

Rapid accumulation of temporal Electronic Health Record (EHR) data and recent advances in deep learning have shown high potential in precisely and timely predicting patients' risks using AI. However, most existing risk prediction approaches ignore the complex asynchronous and irregular problems in real-world EHR data. This paper proposes a novel approach called Knowledge-guIded Time-aware LSTM (KIT-LSTM) for continuous mortality predictions using EHR. KIT-LSTM extends LSTM with two time-aware gates and a knowledge-aware gate to better model EHR and interprets results. Experiments on real-world data for patients with acute kidney injury with dialysis (AKI-D) demonstrate that KIT-LSTM performs better than the state-of-the-art methods for predicting patients' risk trajectories and model interpretation. KIT-LSTM can better support timely decision-making for clinicians.

AI-Enabled Screening for Drug Repurposing in Monoclonal Gammopathy of Undetermined Significance

Background: Monoclonal gammopathy of undetermined significance (MGUS) is a benign hematological condition with the potential to progress to malignant conditions including multiple myeloma and Waldenstrom macroglobulinemia. To date, multiple clinical and laboratory findings have been identified as risk factors for MGUS progression, but any medications that modify progression risk have yet to be identified. To investigate this issue further, we propose a low-cost, easily replicated method for screening drug repurposing candidates in MGUS using real-world EHR data. Methods: We extracted MGUS clinical and laboratory data from a manually curated MGUS database, containing 16,752 MGUS patients diagnosed from January 1, 2000, through December 31, 2021, prospectively maintained at Mayo Clinic. We then separately extracted medication data and comorbidities data, corresponding to the time from MGUS diagnosis to last follow-up or disease progression, from our electronic health record (EHR) database. Medications were mapped to 21 drug classes of interest and coded as the presence or absence of a drug from a given class, as there were challenges with ascertaining medication durations from our EHR data. The XGBoost gradient-boosted tree module was then used to fit a Cox survival model to the data, using a standard 70/15/15 train/validation/test split of the data, with 5-fold cross validation. The impact of explanatory features was quantified as hazard ratios using Shapley values averaged over 1000 repetitions of bootstrapping. For continuous features, hazard ratios were calculated by comparing the mean Shapley values for those above and below the median value of the continuous feature; for binary features, hazard ratios were calculated by comparing mean Shapley values for those with 0 versus 1 for a given feature. Results: Of the 16,752 patients in the MGUS database, medications data were available for 12,253; those without medications data were excluded. Our Cox survival model achieved a satisfactory fit of the data with inverse probability of censoring weights concordance index of 0.764. Within the medication groups examined, the presence of multivitamins, metformin, proton-pump inhibitors, non-coronary NSAIDs, vitamin D supplements or statins in patients’ medication lists were associated with significantly lower hazard ratio for progression. The full list of drug classes and associated hazard ratios are described in Table 1. Discussion: We describe a low-cost method for using real-world EHR data and machine-learning to screen for drug repurposing candidates in MGUS. Our analysis identifies statins, metformin, multivitamins and opioids as having potential protective effects for MGUS progression. This work could inform subsequent prospective studies. Similar approaches can be replicated in other disorders allowing development of hypothesis driven research and repurposing of common drugs. Figure 1View largeDownload PPTFigure 1View largeDownload PPT Close modal

Detecting anomalous sequences in electronic health records using higher-order tensor networks

Detecting anomalous sequences is an integral part of building and protecting modern large-scale health information technology (HIT) systems. These HIT systems generate a large volume of records of patients’ state and significant events, which provide a valuable resource to help improve clinical decisions, patient care processes, and other issues. However, detecting anomalous sequences in electronic health records (EHR) remains a challenge in healthcare applications for several reasons, including imbalances in the data, complexity of relationships between events in the sequence, and the curse of dimensionality. Conventional anomaly detection methods use the finite sequence of events to discriminate sequences. They fail to incorporate salient event details under variable higher-order dependencies (e.g., duration between events) that can provide better discrimination of sequences in their models. To address this problem, we propose event sequence and subsequence anomaly detection algorithms that (1) use network-based representations of interactions in the data, (2) account for variable higher-order dependencies in the data, and (3) incorporate events duration for adequate discrimination of the data. The proposed approach identifies anomalies by monitoring the change in the graph after the test sequence is removed from the network. The change is quantified using graph distance metrics so that dramatic changes in the network can be attributed to the removed sequence. Furthermore, the proposed subsequence algorithm recommends plausible paths and salient information for the detected anomalous subsequences. Our results show that the proposed event sequence anomaly detection algorithm outperforms the baseline methods for both synthetic data and real-world EHR data.

Computable phenotypes for cohort identification: core content for a new class of FAIR Digital Objects

Computable phenotypes for cohort identification: core content for a new class of FAIR Digital Objects

Electronic Health Record-Triggered Research Infrastructure Combining Real-world Electronic Health Record Data and Patient-Reported Outcomes to Detect Benefits, Risks, and Impact of Medication: Development Study.

BackgroundReal-world data from electronic health records (EHRs) represent a wealth of information for studying the benefits and risks of medical treatment. However, they are limited in scope and should be complemented by information from the patient perspective.ObjectiveThe aim of this study is to develop an innovative research infrastructure that combines information from EHRs with patient experiences reported in questionnaires to monitor the risks and benefits of medical treatment.MethodsWe focused on the treatment of overactive bladder (OAB) in general practice as a use case. To develop the Benefit, Risk, and Impact of Medication Monitor (BRIMM) infrastructure, we first performed a requirement analysis. BRIMM’s starting point is routinely recorded general practice EHR data that are sent to the Dutch Nivel Primary Care Database weekly. Patients with OAB were flagged weekly on the basis of diagnoses and prescriptions. They were invited subsequently for participation by their general practitioner (GP), via a trusted third party. Patients received a series of questionnaires on disease status, pharmacological and nonpharmacological treatments, adverse drug reactions, drug adherence, and quality of life. The questionnaires and a dedicated feedback portal were developed in collaboration with a patient association for pelvic-related diseases, Bekkenbodem4All. Participating patients and GPs received feedback. An expert meeting was organized to assess the strengths, weaknesses, opportunities, and threats of the new research infrastructure.ResultsThe BRIMM infrastructure was developed and implemented. In the Nivel Primary Care Database, 2933 patients with OAB from 27 general practices were flagged. GPs selected 1636 (55.78%) patients who were eligible for the study, of whom 295 (18.0% of eligible patients) completed the first questionnaire. A total of 288 (97.6%) patients consented to the linkage of their questionnaire data with their EHR data. According to experts, the strengths of the infrastructure were the linkage of patient-reported outcomes with EHR data, comparison of pharmacological and nonpharmacological treatments, flexibility of the infrastructure, and low registration burden for GPs. Methodological weaknesses, such as susceptibility to bias, patient selection, and low participation rates among GPs and patients, were seen as weaknesses and threats. Opportunities represent usefulness for policy makers and health professionals, conditional approval of medication, data linkage to other data sources, and feedback to patients.ConclusionsThe BRIMM research infrastructure has the potential to assess the benefits and safety of (medical) treatment in real-life situations using a unique combination of EHRs and patient-reported outcomes. As patient involvement is an important aspect of the treatment process, generating knowledge from clinical and patient perspectives is valuable for health care providers, patients, and policy makers. The developed methodology can easily be applied to other treatments and health problems.

A hybrid model to identify fall occurrence from electronic health records

IntroductionFalls are a leading cause of unintentional injury in the elderly. Electronic health records (EHRs) offer the unique opportunity to develop models that can identify fall events. However, identifying fall events in clinical notes requires advanced natural language processing (NLP) to simultaneously address multiple issues because the word “fall” is a typical homonym. MethodsWe implemented a context-aware language model, Bidirectional Encoder Representations from Transformers (BERT) to identify falls from the EHR text and further fused the BERT model into a hybrid architecture coupled with post-hoc heuristic rules to enhance the performance. The models were evaluated on real world EHR data and were compared to conventional rule-based and deep learning models (CNN and Bi-LSTM). To better understand the ability of each approach to identify falls, we further categorize fall-related concepts (i.e., risk of fall, prevention of fall, homonym) and performed a detailed error analysis. ResultsThe hybrid model achieved the highest f1-score on sentence (0.971), document (0.985), and patient (0.954) level. At the sentence level (basic data unit in the model), the hybrid model had 0.954, 1.000, 0.988, and 0.999 in sensitivity, specificity, positive predictive value, and negative predictive value, respectively. The error analysis showed that that machine learning-based approaches demonstrated higher performance than a rule-based approach in challenging cases that required contextual understanding. The context-aware language model (BERT) slightly outperformed the word embedding approach trained on Bi-LSTM. No single model yielded the best performance for all fall-related semantic categories. ConclusionA context-aware language model (BERT) was able to identify challenging fall events that requires context understanding in EHR free text. The hybrid model combined with post-hoc rules allowed a custom fix on the BERT outcomes and further improved the performance of fall detection.

Population pharmacokinetic analysis of dexmedetomidine in children using real-world data from electronic health records and remnant specimens.

Our objectives were to perform a population pharmacokinetic analysis of dexmedetomidine in children using remnant specimens and electronic health records (EHRs) and explore the impact of patient's characteristics and pharmacogenetics on dexmedetomidine clearance. Dexmedetomidine dosing and patient data were gathered from EHRs and combined with opportunistically sampled remnant specimens. Population pharmacokinetic models were developed using nonlinear mixed-effects modelling. Stage 1 developed a model without genotype variables; Stage 2 added pharmacogenetic effects. Our final study population included 354 post-cardiac surgery patients aged 0-22 years (median 16 mo). The data were best described with a 2-compartment model with allometric scaling for weight and Hill maturation function for age. Population parameter estimates and 95% confidence intervals were 27.3L/h (24.0-31.1L/h) for total clearance, 161 L (139-187 L) for central compartment volume of distribution, 26.0L/h (22.5-30.0L/h) for intercompartmental clearance and 7903 L (5617-11 119 L) for peripheral compartment volume of distribution. The estimate for postmenstrual age when 50% of adult clearance is achieved was 42.0weeks (41.5-42.5weeks) and the Hill coefficient estimate was 7.04 (6.99-7.08). Genotype was not statistically or clinically significant. Our study demonstrates the use of real-world EHR data and remnant specimens to perform a population pharmacokinetic analysis and investigate covariate effects in a large paediatric population. Weight and age were important predictors of clearance. We did not find evidence for pharmacogenetic effects of UGT1A4 or UGT2B10 genotype or CYP2A6 risk score.

SCEHR: Supervised Contrastive Learning for Clinical Risk Prediction using Electronic Health Records.

Contrastive learning has demonstrated promising performance in image and text domains either in a self-supervised or a supervised manner. In this work, we extend the supervised contrastive learning framework to clinical risk prediction problems based on longitudinal electronic health records (EHR). We propose a general supervised contrastive loss for learning both binary classification (e.g. in-hospital mortality prediction) and multi-label classification (e.g. phenotyping) in a unified framework. Our supervised contrastive loss practices the key idea of contrastive learning, namely, pulling similar samples closer and pushing dissimilar ones apart from each other, simultaneously by its two components: tries to contrast samples with learned anchors which represent positive and negative clusters, and tries to contrast samples with each other according to their supervised labels. We propose two versions of the above supervised contrastive loss and our experiments on real-world EHR data demonstrate that our proposed loss functions show benefits in improving the performance of strong baselines and even state-of-the-art models on benchmarking tasks for clinical risk predictions. Our loss functions work well with extremely imbalanced data which are common for clinical risk prediction problems. Our loss functions can be easily used to replace (binary or multi-label) cross-entropy loss adopted in existing clinical predictive models. The Pytorch code is released at https://github.com/calvin-zcx/SCEHR.

An interpretable outcome prediction model based on electronic health records and hierarchical attention

Outcome prediction aims to predict the future health condition of patients from Electronic Health Record (EHR) data. Because of the sequential characteristic of EHR data, recurrent neural network (RNN)-based outcome prediction methods have achieved state-of-the-art results. However, the major drawback of RNN-based outcome prediction methods is lack of interpretability, which would lead to trust issues. Aiming at this problem, this paper proposes interpretable outcome prediction model with hierarchical attention (IoHAN), an interpretable outcome prediction model by leveraging attention mechanism. The main novelty of IoHAN is that it can pinpoint the fine-grained influence on the final prediction result of each medical component by decomposing the attention weights hierarchically into hospital visits, medical variables, and interactions between medical variables. We evaluated IoHAN on MIMIC-III, a large real-world EHR data set. The experiment results demonstrate that IoHAN can achieve higher prediction accuracy than state-of-the-art outcome prediction models. In addition, the hierarchical decomposed attention weights can interpret the prediction results in a more natural and understandable way.

Applying interpretable deep learning models to identify chronic cough patients using EHR data

Background and ObjectiveChronic cough (CC) affects approximately 10% of adults. Many disease states are associated with chronic cough, such as asthma, upper airway cough syndrome, bronchitis, and gastroesophageal reflux disease. The lack of an ICD code specific for chronic cough makes it challenging to identify such patients from electronic health records (EHRs). For clinical and research purposes, computational methods using EHR data are urgently needed to identify chronic cough cases. This research aims to investigate the data representations and deep learning algorithms for chronic cough prediction. MethodsUtilizing real-world EHR data from a large academic healthcare system from October 2005 to September 2015, we investigated Natural Language Representation of the EHR data and systematically evaluated deep learning and traditional machine learning models to predict chronic cough patients. We built these machine learning models using structured data (medication and diagnosis) and unstructured data (clinical notes). ResultsThe sensitivity and specificity of a transformer-based deep learning algorithm, specifically BERT with attention model, was 0.856 and 0.866, respectively, using structured data (medication and diagnosis). Sensitivity and specificity improved to 0.952 and 0.930 when we combined structured data with symptoms extracted from clinical notes. We further found that the attention mechanism of deep learning models can be used to extract important features that drive the prediction decisions. Compared with our previously published rule-based algorithm, the deep learning algorithm can identify more chronic cough patients with structured data. ConclusionsBy applying deep learning models, chronic cough patients can be reliably identified for prospective or retrospective research through medication and diagnosis data, widely available in EHR and electronic claims data, thus improving the generalizability of the patient identification algorithm. Deep learning models can identify chronic cough patients with even higher sensitivity and specificity when structured and unstructured EHR data are utilized. We anticipate language-based data representation and deep learning models developed in this research could also be productively used for other disease prediction and case identification.