Longitudinal Electronic Health Record Research Articles

Prevalence, incidence, mortality, and healthcare resource use for generalised pustular psoriasis, palmoplantar pustulosis, and plaque psoriasis in England: a population-based cohort study.

Generalised pustular psoriasis (GPP) and palmoplantar pustulosis (PPP) are chronic, inflammatory skin conditions. Accumulating evidence shows that GPP and PPP have different characteristics compared with plaque psoriasis and are distinct clinical entities. To assess the epidemiology, comorbidities, mortality and healthcare use for patients with GPP and PPP compared with those with plaque psoriasis in England. A cohort study involving analyses of longitudinal electronic health record data in the Clinical Practice Research Datalink Aurum database and linked hospital and mortality data between 2008 and 2019. The primary study outcome was the incidence and prevalence rates for GPP, PPP and plaque psoriasis in England. Secondary outcomes included survival rates and healthcare resource use (HCRU) by disease type. We identified 373 patients with GPP, 1,828 with PPP and 224,223 with plaque psoriasis. The mean age was 55.9 years (standard deviation [SD]: 18.6) for patients with GPP, 51.5 years (SD: 16.4) for those with PPP, and 48.5 years (SD: 19.1) for those with plaque psoriasis; 62.5% and 65.9% of patients with GPP and PPP, respectively, were women, compared with 49.4% of those with plaque psoriasis. About half of the patients were overweight or obese at baseline (GPP, 48.6%; PPP, 56.0%; and plaque psoriasis, 45.9%). The incidence rates for GPP, PPP and plaque psoriasis were 0.25 (95% CI: 0.21-0.28), 2.01 (95% CI: 1.92-2.11) and 103.2 (95% CI: 102.5-103.9) per 100,000 person-years, respectively. From 2008 to 2019, the prevalence rates per 100,000 persons ranged from 1.61 to 3.0 for GPP, 1.1 to 18.7 for PPP and 1771.0 to 1903.8 for plaque psoriasis. Survival rates were lower for patients with GPP, particularly those who were over 55 years old and those with a history of ≥1 comorbidity in each cohort. HCRU was lower in the plaque psoriasis cohort and highest in the GPP cohort, particularly among those who had ≥1 GPP flare. Our results provide further evidence that GPP is a distinct disease with different epidemiology, lower survival and higher HCRU than plaque psoriasis in England.

ACM BCB 2023: Predicting Postoperative Pain and Opioid Use with Machine Learning Applied to Longitudinal Electronic Health Record and Wearable Data.

Managing acute postoperative pain and minimizing chronic opioid use is crucial for patient recovery and long-term well-being. This study explored using preoperative electronic health records (EHR) and wearable device data for machine-learning models that predict postoperative acute pain and chronic opioid use. The study cohort consisted of ~347 All of Us Research Program participants who underwent one of eight surgical procedures and shared EHR and wearable device data. We developed four machine learning models and used the Shapley additive explanations (SHAP) technique to identify the most relevant predictors of acute pain and chronic opioid use. The stacking ensemble model achieved the highest accuracy in predicting acute pain (0.68) and chronic opioid use (0.89). The area under the curve (AUC) score for severe pain vs. other pain was highest (0.88) when predicting acute post-operative pain. Values of logistic regression, random forest, extreme gradient boosting, and stacking ensemble ranged from 0.74 to 0.90 when predicting postoperative chronic opioid use. Variables from wearable devices played a prominent role in predicting both outcomes. SHAP detection of individual risk factors for severe pain can help healthcare providers tailor pain management plans. Accurate prediction of postoperative chronic opioid use before surgery can help mitigate the risk for the outcomes we studied. Prediction can also reduce the chances of opioid overuse and dependence. Such mitigation can promote safer and more effective pain control for patients during their recovery.

Enhancing healthcare decision support through explainable AI models for risk prediction

Electronic health records (EHRs) are a valuable source of information that can aid in understanding a patient’s health condition and making informed healthcare decisions. However, modelling longitudinal EHRs with heterogeneous information is a challenging task. Although recurrent neural networks (RNNs) are frequently utilized in artificial intelligence (AI) models for capturing longitudinal data, their explanatory capabilities are limited. Predictive clustering stands as the most recent advancement within this domain, offering interpretable indications at the cluster level for predicting disease risk. Nonetheless, the challenge of determining the optimal number of clusters has put a brake on the widespread application of predictive clustering for disease risk prediction. In this paper, we introduce a novel non-parametric predictive clustering-based risk prediction model that integrates the Dirichlet Process Mixture Model (DPMM) with predictive clustering via neural networks. To enhance the model’s interpretability, we integrate attention mechanisms that enable the capture of local-level evidence in addition to the cluster-level evidence provided by predictive clustering. The outcome of this research is the development of a multi-level explainable artificial intelligence (AI) model. We evaluated the proposed model on two real-world datasets and demonstrated its effectiveness in capturing longitudinal EHR information for disease risk prediction. Moreover, the model successfully produced interpretable evidence to bolster its predictions.

Discovering clinical drug-drug interactions with known pharmacokinetics mechanisms using spontaneous reporting systems and electronic health records

ObjectiveAlthough the mechanisms behind pharmacokinetic (PK) drug-drug interactions (DDIs) are well-documented, bridging the gap between this knowledge and clinical evidence of DDIs, especially for serious adverse drug reactions (SADRs), remains challenging. While leveraging the FDA Adverse Event Reporting System (FAERS) database along with disproportionality analysis tends to detect a vast number of DDI signals, this abundance complicates further investigation, such as validation through clinical trials. Our study proposed a framework to efficiently prioritize these signals and assessed their reliability using multi-source Electronic Health Records (EHR) to identify top candidates for further investigation. MethodsWe analyzed FAERS data spanning from January 2004 to March 2023, employing four established disproportionality methods: Proportional Reporting Ratio (PRR), Reporting Odds Ratio (ROR), Multi-item Gamma Poisson Shrinker (MGPS), and Bayesian Confidence Propagating Neural Network (BCPNN). Building upon these models, we developed four ranking models to prioritize DDI-SADR signals and cross-referenced signals with DrugBank. To validate the top-ranked signals, we employed longitudinal EHRs from Vanderbilt University Medical Center and the All of Us research program. The performance of each model was assessed by counting how many of the top-ranked signals were confirmed by EHRs and calculating the average ranking of these confirmed signals. ResultsOut of 189 DDI-SADR signals identified by all four disproportionality methods, only two were documented in the DrugBank database. By prioritizing the top 20 signals as determined by each of the four disproportionality methods and our four ranking models, 58 unique DDI-SADR signals were selected for EHR validations. Of these, five signals were confirmed. The ranking model, which integrated the MGPS and BCPNN, demonstrated superior performance by assigning the highest priority to those five EHR-confirmed signals. ConclusionThe fusion of disproportionality analysis with ranking models, validated through multi-source EHRs, presents a groundbreaking approach to pharmacovigilance. Our study's confirmation of five significant DDI-SADRs, previously unrecorded in the DrugBank database, highlights the essential role of advanced data analysis techniques in identifying ADRs.

Targeted-BEHRT: Deep Learning for Observational Causal Inference on Longitudinal Electronic Health Records.

Observational causal inference is useful for decision-making in medicine when randomized clinical trials (RCTs) are infeasible or nongeneralizable. However, traditional approaches do not always deliver unconfounded causal conclusions in practice. The rise of "doubly robust" nonparametric tools coupled with the growth of deep learning for capturing rich representations of multimodal data offers a unique opportunity to develop and test such models for causal inference on comprehensive electronic health records (EHRs). In this article, we investigate causal modeling of an RCT-established causal association: the effect of classes of antihypertensive on incident cancer risk. We develop a transformer-based model, targeted bidirectional EHR transformer (T-BEHRT) coupled with doubly robust estimation to estimate average risk ratio (RR). We compare our model to benchmark statistical and deep learning models for causal inference in multiple experiments on semi-synthetic derivations of our dataset with various types and intensities of confounding. In order to further test the reliability of our approach, we test our model on situations of limited data. We find that our model provides more accurate estimates of relative risk [least sum absolute error (SAE) from ground truth] compared with benchmark estimations. Finally, our model provides an estimate of class-wise antihypertensive effect on cancer risk that is consistent with results derived from RCTs.

MedAlign: A Clinician-Generated Dataset for Instruction Following with Electronic Medical Records

The ability of large language models (LLMs) to follow natural language instructions with human-level fluency suggests many opportunities in healthcare to reduce administrative burden and improve quality of care. However, evaluating LLMs on realistic text generation tasks for healthcare remains challenging. Existing question answering datasets for electronic health record (EHR) data fail to capture the complexity of information needs and documentation burdens experienced by clinicians. To address these challenges, we introduce MedAlign, a benchmark dataset of 983 natural language instructions for EHR data. MedAlign is curated by 15 clinicians (7 specialities), includes clinician-written reference responses for 303 instructions, and provides 276 longitudinal EHRs for grounding instruction-response pairs. We used MedAlign to evaluate 6 general domain LLMs, having clinicians rank the accuracy and quality of each LLM response. We found high error rates, ranging from 35% (GPT-4) to 68% (MPT-7B-Instruct), and 8.3% drop in accuracy moving from 32k to 2k context lengths for GPT-4. Finally, we report correlations between clinician rankings and automated natural language generation metrics as a way to rank LLMs without human review. We make MedAlign available under a research data use agreement to enable LLM evaluations on tasks aligned with clinician needs and preferences.

Abstract 3821: Real world outcomes among patients with newly diagnosed acute myeloid leukemia

Abstract Introduction: Treatment options for patients with acute myeloid leukemia (AML) have expanded in recent years. Although new treatments have demonstrated efficacy in clinical trials, less is known about their effect on outcomes in routine clinical practice. We examined outcomes for patients with newly-diagnosed AML (ND-AML) using real world data (RWD). Methods: This retrospective observational study was conducted using the COTA real-world (rw) database, a de-identified source of longitudinal electronic health record data pertaining to the diagnosis, management, and outcomes of patients with cancer. Adult patients with ND-AML, diagnosed January 1, 2013 to March 20, 2023 were included. First-line treatment regimens were grouped into therapeutic categories and outcomes were summarized. The index date was defined as date of initiation of first-line therapy. Results: Among 2516 patients with ND-AML included (Table), the most common regimens used were intensive chemotherapy with cytarabine (IC w/ara-C), hypomethylating agents with venetoclax (HMA+ven), and HMA alone. Complete response rate was highest in patients treated with IC w/ara-C or investigational regimen(s). Median overall survival (OS) was 16 mo overall, with highest median OS seen in patients treated with IC w/ara-C and investigational regimen(s). Median OS for patients treated with HMA+ven was 13.5 mo. Thirty-day mortality was lowest in patients treated with IC w/ara-C, CPX-351 (liposomal daunorubicin + cytarabine), and HMA+ven. Conclusions: Results suggest that IC w/ara-C remains the standard of care for most patients with ND-AML and this regimen has a higher CR rate and median OS. Variation of patient demographics and clinical characteristics across treatment groups did not permit cross-treatment comparisons, though future analyses evaluating these subgroups may provide further insight into effectiveness and safety of AML regimens in routine care for populations underrepresented in clinical trials. rwCR rate Median rwOS 30-day mortality N % (95% CI) Months (95% CI) % (95% CI) Overall 2516 45.3 (43.2, 47.3) 16.0 (14.9, 17.3) 5.9 (5.0, 6.8) IC w/ara-C 1171 60.6 (57.8, 63.4) 36.4 (31.0, 46.0) 1.6 (0.9, 2.3) HMA+ ven 300 33.7 (28.3, 39.3) 13.5 (11.4, 15.5) 3 (1.1, 4.9) Decitabine 242 12.8 (8.9, 17.7) 8.4 (6.7, 9.9) 10.4 (6.5, 14.2) Azacitidine 206 7.8 (4.5, 12.3) 7.2 (5.3, 9.6) 10.7 (6.4, 14.8) Investigational 206 61.6 (54.6, 68.3) 21.3 (15.2, 27.8) 3.4 (0.9, 5.9) Lower dose cytarabine 62 12.9 (5.7, 23.9) 8.7 (5.4, 16.1) 14.9 (5.4, 23.4) CPX-351 based 61 55.7 (42.4, 68.5) 9.3 (7.3, 15.1) 0.0 No treatment 188 NA 1.9 (1.5, 2.8) 30 (22.7, 36.6) Citation Format: Elizabeth D. Pulte, Laura L. Fernandes, Joseph Wynne, Eric Hansen, Andrew J. Belli, Anna Barcellos, Christina M. Zettler, Rebecca Bystrom, Jonathon Vallejo, Wenjuan Gu, Catherine Lerro, Kelly Norsworthy, Angelo DeClaro, Marc R. Theoret, Donna Rivera, Ching-Kun Wang. Real world outcomes among patients with newly diagnosed acute myeloid leukemia [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 3821.

Prediction of Cancer Symptom Trajectory Using Longitudinal Electronic Health Record Data and Long Short-Term Memory Neural Network.

Ability to predict symptom severity and progression across treatment trajectories would allow clinicians to provide timely intervention and treatment planning. However, such predictions are difficult because of sparse and inconsistent assessment, and simplistic measures such as the last observed symptom severity are often used. The purpose of this study is to develop a model for predicting future cancer symptom experiences on the basis of past symptom experiences. We performed a retrospective, longitudinal analysis using records of patients with cancer (n = 208) hospitalized between 2008 and 2014. A long short-term memory (LSTM)-based recurrent neural network, a linear regression, and random forest models were trained on previous symptoms experienced and used to predict future symptom trajectories. We found that at least one of three tested models (LSTM, linear regression, and random forest) outperform predictions based solely on the previous clinical observation. LSTM models significantly outperformed linear regression and random forest models in predicting nausea (P < .1) and psychosocial status (P < .01). Linear regression outperformed all models when predicting oral health (P < .01), while random forest outperformed all models when predicting mobility (P < .01) and nutrition (P < .01). We can successfully predict patients' symptom trajectories with a prediction model, built with sparse assessment data, using routinely collected nursing documentation. The results of this project can be applied to better individualize symptom management to support cancer patients' quality of life.

Genomic data in the All of Us Research Program

Comprehensively mapping the genetic basis of human disease across diverse individuals is a long-standing goal for the field of human genetics1–4. The All of Us Research Program is a longitudinal cohort study aiming to enrol a diverse group of at least one million individuals across the USA to accelerate biomedical research and improve human health5,6. Here we describe the programme’s genomics data release of 245,388 clinical-grade genome sequences. This resource is unique in its diversity as 77% of participants are from communities that are historically under-represented in biomedical research and 46% are individuals from under-represented racial and ethnic minorities. All of Us identified more than 1 billion genetic variants, including more than 275 million previously unreported genetic variants, more than 3.9 million of which had coding consequences. Leveraging linkage between genomic data and the longitudinal electronic health record, we evaluated 3,724 genetic variants associated with 117 diseases and found high replication rates across both participants of European ancestry and participants of African ancestry. Summary-level data are publicly available, and individual-level data can be accessed by researchers through the All of Us Researcher Workbench using a unique data passport model with a median time from initial researcher registration to data access of 29 hours. We anticipate that this diverse dataset will advance the promise of genomic medicine for all.

Sexual orientation and gender identity measures and viral suppression for people living with HIV: a protocol for a population-based cohort study

IntroductionThe measure of sexual orientation and gender identity (SOGI) data in electronic health records (EHR) has been critical for addressing health disparities and inequalities, especially for HIV care. Given that...

A deep clustering-based state-space model for improved disease risk prediction in personalized healthcare

AbstractDecision support systems are being developed to assist clinicians in complex decision-making processes by leveraging information from clinical knowledge and electronic health records (EHRs). One typical application is disease risk prediction, which can be challenging due to the complexity of modelling longitudinal EHR data, including unstructured medical notes. To address this challenge, we propose a deep state-space model (DSSM) that simulates the patient’s state transition process and formally integrates latent states with risk observations. A typical DSSM consists of three parts: a prior module that generates the distribution of the current latent state based on previous states; a posterior module that approximates the latent states using up-to-date medical notes; and a likelihood module that predicts disease risks using latent states. To efficiently and effectively encode raw medical notes, our posterior module uses an attentive encoder to better extract information from unstructured high-dimensional medical notes. Additionally, we couple a predictive clustering algorithm into our DSSM to learn clinically useful representations of patients’ latent states. The latent states are clustered into multiple groups, and the weighted average of the cluster centres is used for prediction. We demonstrate the effectiveness of our deep clustering-based state-space model using two real-world EHR datasets, showing that it not only generates better risk prediction results than other baseline methods but also clusters similar patient health states into groups.

Temporal Phenotyping for End-Stage Renal Disease Using Longitudinal Electronic Health Records.

End Stage Renal Disease (ESRD) is a highly heterogeneous disease with significant differences in prevalence, mortality, complications, and treatment modalities across age, sex, race, and ethnicity. An improved knowledge of disease characteristics results from the use of a data-driven phenotypic classification strategy to identify patients of different subtypes and expose the clinical traits of different subtypes. This study used topic models and process mining techniques to perform subtyping of ESRD patients on hemodialysis based on real-world longitudinal electronic health record data. The mined subtypes are interpretable and clinically significant, and they can reflect differences in the progression of the disease state and clinical outcomes.

DOP33 Distinct perturbances in metabolic pathways associate with disease progression in patients with Inflammatory Bowel Disease

Abstract Background Patients with inflammatory bowel diseases (IBD) exhibit distinct shifts in circulating metabolite levels that have been associated with clinical disease activity and disease phenotype. However, there are limited data on the impact of metabolomic changes on disease progression in IBD. We aimed to understand the association of circulating metabolite levels and related pathways with the risk of disease progression in patients with IBD. Methods We conducted an observational cohort study using the Mount Sinai Crohn’s and Colitis Registry (MSCCR). Patients were prospectively enrolled with baseline questionnaires and biospecimens that were then linked to longitudinal electronic health record data. A total of 1,456 serum metabolites were measured in 277 patients with UC and 375 patients with CD using untargeted metabolomics. Disease progression was defined as any new steroid or biological prescriptions, IBD-related hospitalizations, or surgery. Multivariable Cox proportional hazards regression models, adjusted for age, sex, smoking, biologic use, and surgery were used to explore metabolite associations with risk of disease progression. Metabolite set enrichment analysis (MSEA) was performed to identify underlying metabolic pathways associated with disease progression. Results During a median 2-year follow-up time (IQR 0.3-5.1 years), 375 patients (57.5%) had a disease progression event. Among CD, 151 metabolites significantly correlated with the composite outcome (FDR&amp;lt;0.1). Of these, 81 (53.6%) were linked to higher risk, mainly enriched in amino acids, purine, and pyrimidine metabolism, fibrinogen cleavage peptides, and bile acid pathways. Additionally, 70 (46.4%) metabolites associated with lower risk with enrichment in fatty acid metabolism, steroid biosynthesis, histidine, tryptophan, and antioxidant pathways. In CD, aspartate metabolism was most significantly associated with increased risk and linoleic metabolism with decreased risk. In UC, 84 metabolites were significantly associated with disease progression (FDR&amp;lt;0.1), with 29 (34.5%) associated with increased and 55 (65.5%) with decreased risk. Metabolic pathways linked to higher risk were sphingolipid, benzoate, hydrogen sulfide, polyamine, and tyrosine metabolism, while protective metabolites were enriched in branched-chain amino acid (BCAA), steroid biosynthesis, histidine, and phenylalanine pathways. In UC, sphingolipid metabolism was prominently associated with increased risk and BCAA metabolism with decreased risk. Conclusion Specific metabolites and metabolic pathways are associated with risk of disease progression in patients with IBD, highlighting potential prognostic metabolite biomarkers and pathways that may be important in modulating risk of disease progression.

Determining Distinct Suicide Attempts From Recurrent Electronic Health Record Codes: Classification Study.

Prior suicide attempts are a relatively strong risk factor for future suicide attempts. There is growing interest in using longitudinal electronic health record (EHR) data to derive statistical risk prediction models for future suicide attempts and other suicidal behavior outcomes. However, model performance may be inflated by a largely unrecognized form of "data leakage" during model training: diagnostic codes for suicide attempt outcomes may refer to prior attempts that are also included in the model as predictors. We aimed to develop an automated rule for determining when documented suicide attempt diagnostic codes identify distinct suicide attempt events. From a large health care system's EHR, we randomly sampled suicide attempt codes for 300 patients with at least one pair of suicide attempt codes documented at least one but no more than 90 days apart. Supervised chart reviewers assigned the clinical settings (ie, emergency department [ED] versus non-ED), methods of suicide attempt, and intercode interval (number of days). The probability (or positive predictive value) that the second suicide attempt code in a given pair of codes referred to a distinct suicide attempt event from its preceding suicide attempt code was calculated by clinical setting, method, and intercode interval. Of 1015 code pairs reviewed, 835 (82.3%) were nonindependent (ie, the 2 codes referred to the same suicide attempt event). When the second code in a pair was documented in a clinical setting other than the ED, it represented a distinct suicide attempt 3.3% of the time. The more time elapsed between codes, the more likely the second code in a pair referred to a distinct suicide attempt event from its preceding code. Code pairs in which the second suicide attempt code was assigned in an ED at least 5 days after its preceding suicide attempt code had a positive predictive value of 0.90. EHR-based suicide risk prediction models that include International Classification of Diseases codes for prior suicide attempts as a predictor may be highly susceptible to bias due to data leakage in model training. We derived a simple rule to distinguish codes that reflect new, independent suicide attempts: suicide attempt codes documented in an ED setting at least 5 days after a preceding suicide attempt code can be confidently treated as new events in EHR-based suicide risk prediction models. This rule has the potential to minimize upward bias in model performance when prior suicide attempts are included as predictors in EHR-based suicide risk prediction models.

Multi-Branching Temporal Convolutional Network With Tensor Data Completion for Diabetic Retinopathy Prediction.

Diabetic retinopathy (DR), a microvascular complication of diabetes, is the leading cause of vision loss among working-aged adults. However, due to the low compliance rate of DR screening and expensive medical devices for ophthalmic exams, many DR patients did not seek proper medical attention until DR develops to irreversible stages (i.e., vision loss). Fortunately, the widely available electronic health record (EHR) databases provide an unprecedented opportunity to develop cost-effective machine-learning tools for DR detection. This paper proposes a Multi-branching Temporal Convolutional Network with Tensor Data Completion (MB-TCN-TC) model to analyze the longitudinal EHRs collected from diabetic patients for DR prediction. Experimental results demonstrate that the proposed MB-TCN-TC model not only effectively copes with the imbalanced data and missing value issues commonly seen in EHR datasets but also captures the temporal correlation and complicated interactions among medical variables in the longitudinal clinical records, yielding superior prediction performance compared to existing methods. Specifically, our MB-TCN-TC model provides AUROC and AUPRC scores of 0.949 and 0.793 respectively, achieving an improvement of 6.27% on AUROC, 11.85% on AUPRC, and 19.3% on F1 score compared with the traditional TCN model.

Household Health-Related Social Needs in Newborns and Infant Behavioral Functioning at 6 Months

Dysfunctional patterns of behavior during infancy can predict the emergence of mental health disorders later in childhood. The Baby Pediatric Symptom Checklist (BPSC) can identify indicators of behavioral disorders among children aged 0 to 18 months. Understanding the association of early health-related social needs (HRSNs) with poor infant behavioral functioning can inform interventions to promote early childhood mental well-being. To examine the association between household HRSNs in the first 4 months of life and BPSC results at 6 months. This was a retrospective cohort analysis of longitudinal electronic health record data. Covariates were selected based on the biopsychosocial ecological model. Logistic regression analyses examined the association of HRSN domains and the number of HRSNs with the 6-month BPSC outcomes. Participants were recruited from 6 primary care clinics within 1 health system. Children aged 5 to 8 months who were evaluated for 6-month well-child visits between March 30, 2021, and June 30, 2022, were included in the study. Responses to the first HRSN screening tool that a caregiver completed for infants between 0 and 4 months of age. HRSN domains were examined individually and as the number of positive HRSNs. BPSC screen identified for clinical review due to 1 or more elevated subscales (inflexibility, irritability, and difficulty with routines) at 6 months. A total of 1541 children (mean [SD] age, 6.1 [0.5] months; 775 female [50.3%]) were included in the study. A total of 405 children (26.3%) had a BPSC screen identified for clinical review, and 328 caregivers (21.3%) reported at least 1 HRSN. Food insecurity (174 [11.3%]) and benefits issues (169 [11.0%]) were the most frequently reported HRSN. Children in households with food insecurity had statistically significant higher odds of inflexibility (adjusted odds ratio [aOR], 1.73; 95% CI, 1.14-2.63), difficulty with routines (aOR, 1.64; 95% CI, 1.05-2.57), and irritability (aOR, 1.86; 95% CI, 1.13-3.08) than children in households without food insecurity. Children in households with benefits issues had statistically significant higher odds of difficulty with routines (aOR, 1.70; 95% CI, 1.10-2.65) and irritability (aOR, 1.70; 95% CI, 1.03-2.82). Children in households with 2 or more HRSNs had consistently higher odds of having a BPSC screen identified for clinical review (aOR, 2.16; 95% CI, 1.38-3.39) compared with children with no HRSNs. Results of this cohort study suggest that household food insecurity, benefits issues, and the number of HRSNs were significantly associated with a BPSC screen identified for clinical review at 6 months of age. These findings highlight the urgency of intervening on HRSNs in the newborn period to prevent adverse infant behavioral outcomes.

Insurance coverage and discontinuity during pregnancy: Frequency and associations documented in the PROMISE cohort.

To describe insurance patterns and discontinuity during pregnancy, which may affect the experiences of the pregnant person: their timely access to care, continuity of care, and health outcomes. Data are from the PROMISE study, which utilizes data from community-based health care organizations (CHCOs) (e.g., federally qualified health centers that serve patients regardless of insurance status or ability to pay) in the United States from 2005 to 2021. This descriptive study was a cohort utilizing longitudinal electronic health record data. Insurance type at each encounter was recorded in the clinical database and coded as Private, Public, and Uninsured. Pregnant people were categorized into one of several insurance patterns. We analyzed the frequency and timing of insurance changes and care utilization within each group. Continuous public insurance was the most common insurance pattern (69.2%), followed by uninsured/public discontinuity (11.8%), with 6.4% experiencing uninsurance throughout the entirety of pregnancy. Insurance discontinuity was experienced by 16.6% of pregnant people; a majority of these reflect people transitioning to public insurance. Those with continuous public insurance had the highest frequency of inadequate prenatal care (19.5%), while those with all three types of insurance during pregnancy had the highest percentage of intensive prenatal care (16.5%). The majority (71.7%-81.2%) of those with a discontinuous pattern experienced a single insurance change. Insurance discontinuity and uninsurance are common within our population of pregnant people seeking care at CHCOs. Our findings suggest that insurance status should be regarded as a dynamic rather than a static characteristic during pregnancy and should be measured accordingly. Future research is needed to assess the drivers of perinatal insurance discontinuity and if and how these discontinuities may affect health care access, utilization, and birth outcomes.

Harnessing Artificial Intelligence for Risk Stratification in Acute Myeloid Leukemia (AML): Evaluating the Utility of Longitudinal Electronic Health Record (EHR) Data Via Graph Neural Networks

Background: AML is a life-threatening disease, and to determine which patients need allogeneic stem cell transplantation, hematologists risk-stratify each case. However, standard risk stratification using the European LeukemiaNet (ELN) criteria is focused on baseline mutations and chromosomal aberrations, and the risk estimate is not updated during a patient's course. In other blood cancers, recalculating the risk with treatment response data can help guide the need for more intensive therapy (Kurtz, et al, Cell, 2019). Furthermore, deep learning graph neural networks (GNN) applied to EHR data have strong predictive power in a hematology context (Fouladvand, et al, J Biomed Inform, 2023). Thus, we evaluated the power of a GNN to predict survival in AML using longitudinal EHR data, specifically with labs and histological features that are not included in the ELN but may capture the treatment response. Methods: Patients who were seen at the Stanford Cancer Institute, had EHR data available within six months of diagnosis, and were diagnosed with AML between June 1998 and January 2021 were included in this retrospective analysis. The GNN was trained to predict survival at two years from diagnosis using the first six months of clinical data. Patients were excluded if they were lost to follow-up before two years or died before six months. Data were collected from structured databases associated with Stanford's EHR, except that diagnosis dates were from Stanford's Cancer Registry, and survival data was supplemented with other databases including the Social Security Death Index. Dysplasia, bone marrow cellularity, and bone marrow blast percentages from pathology reports (“pathology report data”) were extracted using text processing algorithms and weakly supervised machine learning (Ratner, et al, ArXiv, 2017). To represent time series information, we framed each patient's timeline as a network (or “graph”) of events. The primary GNN model was a heterogenous graph transformer classifier with two node types: complete blood count (CBC) data and pathology report data (Hu, et al, ArXiv, 2020). Data from the same week were assumed to be from the same timeframe and connected with bidirectional edges. Data separated by longer time periods were connected with unidirectional edges of a separate edge type. The independent test dataset consisted of patients whose ELN 2022 classification was available, and to train the model, the remaining data were divided into train/validation splits of 0.9/0.1. Results: Of the 2,535 patients with survival data, 1,029 met inclusion criteria. Table 1 summarizes the data available in the EHR for each variable, and nearly all patients had CBC and pathology report data. The area under the receiver operating characteristic (AUROC) using the ELN 2022 criteria for predicting survival in the test dataset was 0.79. The AUROC curve for the GNN model was comparable at 0.76, despite not using any variables from the ELN criteria, and the model effectively stratified patients' disease into high- and low-risk in the independent test dataset (hazard ratio [HR] 3.0, log-rank p = 0.0009). Interestingly, despite not having access to mutation or cytogenetic data, the high-risk cases were enriched in known high-risk mutations, like TP53 and RUNX1, and in high-risk chromosomal aberrations, like 5q deletion (Table 1). Although the model predictions correlated with the ELN criteria in some ways, they also stratified the ELN intermediate-risk AML cases into high and low risk (HR 6.1 for model-predicted high risk among ELN intermediate cases, p = 0.07). Conclusions: Risk stratification using artificial intelligence and longitudinal data from the EHR performed comparably to the ELN 2022 criteria and has the potential to further stratify the ELN categories. The model performed well despite only using histological features and lab values, which are more readily available and more frequently updated than next-generation sequencing results. In the future, this approach may further improve with a larger sample size and additional variables, such as measurable residual disease and treatment information. Given the heterogeneity and increasing complexity of AML classification, leveraging artificial intelligence to assist with classification will be crucial, and these results are a step towards a future where data are automatically extracted from the EHR and used for continuously updated risk stratification.

Real-World Treatment Patterns Among Patients with Newly Diagnosed (ND) Acute Myeloid Leukemia (AML) with TP53 Mutation in the United States (US)

Introduction: A unique subgroup of AML is defined by the presence of the TP53 mutation, with worse overall survival (OS) and limited therapeutic options (Döhner et al, Blood 2022; Daver et al, J Hematol Oncol 2023). Clinical trials are essential for establishing the efficacy and safety of new treatment options in AML; yet, they may not be generalizable to patients with TP53 mutation, who are often older adults with comorbidities and prior treatment exposure. Thus, the optimal treatment for TP53 mutated AML is unclear. Real-world data (RWD) can provide insights into demographic and clinical characteristics, as well as the utilization, effectiveness, and safety of treatment regimens used in routine clinical practice. This RWD study aimed to describe the characteristics, first line treatment patterns, and outcomes in patients with ND-AML with TP53 mutation in a real-world cohort of adult patients in the US. Methods: A retrospective observational study was conducted using the COTA real-world database, a de-identified source of longitudinal electronic health record (EHR) data pertaining to the diagnosis, clinical management, and outcomes of patients with cancer. The study population included adult patients (≥ 18 years) with ND-AML and confirmed TP53 mutation status (positive or negative) per Next-Generation Sequencing (NGS), who were diagnosed between January 1, 2013, and March 20, 2023. Patients were excluded if key study data were missing (i.e., if TP53 mutation status was unknown, or if the testing methodology was unspecified), or if an AML-defining recurrent genetic abnormality superseded TP53 in the hierarchical classification of the International Consensus Classification. First-line treatment regimens were mapped to relevant therapeutic categories (e.g., intensive chemo [IC] with cytarabine, hypomethylating agent with venetoclax [HMA+ven]) and summarized across clinical and demographic patient characteristics using descriptive analyses. Results: A total of 884 patients ND-AML with analysis of TP53 status were identified, including 173 TP53 positive (TP53+) and 711 negative (TP53-). Patients who were TP53+ were older, and otherwise demographically similar to those who were TP53- (see table). Testing for TP53 increased over time, with about 75% of tested patients being diagnosed in 2018-2022 versus 25% in 2013-17. All patients who were TP53+ were by definition ELN adverse risk group, whereas 36.8% of patients with TP53- AML were adverse risk. The most commonly used treatment regimens included hypomethylating agent (HMA) alone (17.9% of TP53+ and 11.8% of TP53-), HMA with venetoclax (HMA+ven) (28.9% TP53+, 20.3% TP53-), and IC w/cytarabine (26% TP53+, 46.8% TP53-). Patients with TP53+ AML were more likely to not receive any anti-leukemic treatment. Patients with TP53+ AML were less likely to achieve complete remission (CR) compared to those with TP53- AML (31.2% versus 46.1%, excluding patients in each subgroup who did not receive anti-leukemic treatment). In this cohort, OS was shorter for patients with TP53+ AML compared with those with TP53- AML. This difference was seen consistently in patients treated with the two most common regimens (IC w/cytarabine and HMA+ven) (Figure). OS appeared longer for patients treated with IC w/cytarabine in patients with or without TP53 mutation, but it should be noted that these patients were younger and generally more fit than patients receiving HMA+ven (61 years versus 75 years among patients with TP53+ AML). Conclusions: In patients with TP53+ AML, CR and OS are lower overall and when specific treatment options are compared to patients with TP53- AML. This difference was seen in patients receiving either of the two most commonly used treatment regimens. This study demonstrates the capability of using RWD to characterize the outcomes of a specific genetic subtype of ND-AML despite widespread testing for this marker only occurring in the latter half of the study timeframe. These findings are subject to the inherent limitations associated with observational data, including the challenges in collecting and analyzing real-world data (e.g., missing treatment data or other characteristics) which may result in bias or limited generalizability of study findings. Finally, this study further demonstrates that research into novel treatment of TP53+ AML is needed to improve patient outcomes.

Predicting multiple sclerosis severity with multimodal deep neural networks

Multiple Sclerosis (MS) is a chronic disease developed in the human brain and spinal cord, which can cause permanent damage or deterioration of the nerves. The severity of MS disease is monitored by the Expanded Disability Status Scale, composed of several functional sub-scores. Early and accurate classification of MS disease severity is critical for slowing down or preventing disease progression via applying early therapeutic intervention strategies. Recent advances in deep learning and the wide use of Electronic Health Records (EHR) create opportunities to apply data-driven and predictive modeling tools for this goal. Previous studies focusing on using single-modal machine learning and deep learning algorithms were limited in terms of prediction accuracy due to data insufficiency or model simplicity. In this paper, we proposed the idea of using patients’ multimodal longitudinal and longitudinal EHR data to predict multiple sclerosis disease severity in the future. Our contribution has two main facets. First, we describe a pioneering effort to integrate structured EHR data, neuroimaging data and clinical notes to build a multi-modal deep learning framework to predict patient’s MS severity. The proposed pipeline demonstrates up to 19% increase in terms of the area under the Area Under the Receiver Operating Characteristic curve (AUROC) compared to models using single-modal data. Second, the study also provides valuable insights regarding the amount useful signal embedded in each data modality with respect to MS disease prediction, which may improve data collection processes.