Luke V. Rasmussen

Objective: To describe the development and implementation of an automated platform for genomic risk prediction that integrates multiple data types. Materials and methods: Using the REDCap infrastructure, we constructed the R4 (Recruitment, Results, and Risk Reduction) platform to intake data from clinical sites, partner laboratories, participant surveys, and electronic health record (EHR) data across 13 institutions. Results: The R4 Portal successfully integrated data to generate genome-informed risk assessments (GIRAs) across 11 conditions for a 23 840 person cohort. Testing phases and quality control led to network-wide protocols ensuring consistency and accuracy. Discussion: As the science of estimating disease risk evolves, standardized and high-throughput methods of collecting and manipulating complex data are required. Platforms should be open-source, modular, and reusable, ensuring flexibility, security, and integration across healthcare environments. Conclusion: The electronic MEdical Records and GEnomics (eMERGE) network successfully generated and returned comprehensive risk profiles using logic and data specific to 11 conditions in a secure and semi-automated fashion employing a customized REDCap database.

Pneumonia is the leading cause of death from infectious disease worldwide. The diagnosis and treatment of patients with pneumonia lag behind other major conditions, relying on syndromic definitions that lack molecular resolution and ignore underlying endotypes. We sought to test the hypothesis that dynamic pathogen-specific host responses in the alveolar space represent distinct pneumonia endotypes linked to different clinical features and outcomes. We prospectively enrolled a cohort of 690 patients (including immunocompromised patients) with known or suspected pneumonia receiving mechanical ventilation in whom the etiology of pneumonia was determined by gold-standard analysis of distal lung fluid obtained by bronchoalveolar lavage (BAL) combined with clinical adjudication. From these patients, we analyzed 792 BAL fluid samples, including 310 serial samples, using flow cytometry (482 patients) and single-cell RNA-sequencing (170 patients; 263 samples, complemented by 9 healthy controls and 25 post-COVID-19 patients, yielding ~2.4 million single cells across 28 cell types), and extracted daily clinical data from the electronic health record (>15,000 patient-days). We used machine learning models to identify pathogen-specific host responses in the transcriptome of alveolar immune cells that were associated with changes in alveolar cell abundance and clinical features. Our results suggest that therapeutic strategies for pneumonia should be individualized to specific host-pathogen interactions.

Many analyses of critically ill patients focus on admission features and discharge outcomes, overlooking daily events and intercurrent complications. Pneumonia is a syndromic entity that is difficult to adjudicate outcomes without clinical expertise. Here, we examine clinical features on a day-by-day basis, mirroring the practice of daily multidisciplinary rounds, with annotated pneumonia episodes and outcomes. The Successful Clinical Response in Pneumonia Therapy (SCRIPT) CarpeDiem Dataset includes 21,931 patient-hospital-days from 704 patients enrolled in the SCRIPT study between June 2018 and May 2023. These patients were all receiving mechanical ventilation and underwent bronchoalveolar lavage to diagnose and establish the etiology of their pneumonia or an alternative diagnosis. The dataset captures patient demographics, daily clinical parameters, including vital signs, laboratory results, and mechanical support data, detailed pneumonia episode results adjudicated by critical care physicians, and outcomes. All data are de-identified per HIPAA Safe Harbor guidelines. This dataset offers a unique resource for analyzing the clinical trajectories of patients with severe pneumonia.

Purpose: Ventilator-associated pneumonia (VAP) remains one of the most serious hospital-acquired infections in the intensive care unit (ICU), with high morbidity and mortality. Early identification of patients at risk for developing VAP could enable timely diagnostics and intervention. However, current clinical tools are limited in their ability to detect early physiologic signals preceding VAP onset. We aimed to build supervised machine learning models to predict short term onset of VAP. Methods: We analyzed electronic health record data from a prospective observational cohort of ICU patients, where VAP was adjudicated using a standardized published protocol by a panel of critical care physicians. Clinical features (including vital signs, ventilator settings, laboratory values, and support devices) were extracted for each patient-ICU-day. We explored unsupervised clustering to characterize feature dynamics associated with VAP onset. We built multiple machine learning models across different prediction windows (3, 5, 7 days before VAP). We examined model performance in two external cohorts, MIMIC-IV and secondary analysis of the AMIKINHAL trial. Results were evaluated with discrimination metrics such as AUROC. Results: The internal cohort included 507 patients with BAL-confirmed diagnoses: 261 developed VAP and 246 did not have VAP. Visualization using clustering identified distinct physiologic states enriched for VAP-labeled days. The best-performing model achieved an AUROC of 0.866 in predicting VAP up to seven days before clinical diagnosis. Temporal model probability trajectories showed rising model confidence in the days leading up to VAP. On external validation in MIMIC-IV, the best model achieved an AUROC of 0.817 for forecasting VAP within five days. There was low feature overlap with the AMIKINHAL trial data, leading to poor model performance. Feature analysis revealed that platelet count, positive end-expiratory pressure (PEEP), ventilator duration, and inflammatory markers were key drivers of model predictions. Conclusions: Machine learning models trained on routinely collected ICU data with careful labeling can anticipate VAP onset up to a week in advance with strong predictive performance. Model performance generalized to data from an entirely different hospital system despite differences in practice and labeling patterns, but did not perform well when there was poor feature overlap. Future work should focus on real-time prospective evaluation.

Background: Computational phenotyping from electronic health records (EHRs) is essential for clinical research, decision support, and quality/population health assessment, but the proliferation of algorithms for the same conditions makes it difficult to identify which algorithm is most appropriate for reuse. Objective: To develop a framework for assessing phenotyping algorithm fitness for purpose and reuse. Fitness for Purpose: Phenotyping algorithms are fit for purpose when they identify the intended population with performance characteristics appropriate for the intended application. Fitness for Reuse: Phenotyping algorithms are fit for reuse when the algorithm is implementable and generalizable—that is, it identifies the same intended population with similar performance characteristics when applied to a new setting. Conclusions: The PhenoFit framework provides a structured approach to evaluate and adapt phenotyping algorithms for new contexts increasing efficiency and consistency of identifying patient populations from EHRs.

Objective: Rule-based structured data algorithms and natural language processing (NLP) approaches applied to unstructured clinical notes have limited accuracy and poor generalizability for identifying immunosuppression. Large language models (LLMs) may effectively identify patients with heterogenous types of immunosuppression from unstructured clinical notes. We compared the performance of LLMs applied to unstructured notes for identifying patients with immunosuppressive conditions or immunosuppressive medication use against 2 baselines: (1) structured data algorithms using diagnosis codes and medication orders and (2) NLP approaches applied to unstructured notes. Materials and Methods: We used hospital admission notes from a primary cohort of 827 intensive care unit (ICU) patients at Northwestern Memorial Hospital and a validation cohort of 200 ICU patients at Beth Israel Deaconess Medical Center, along with diagnosis codes and medication orders from the primary cohort. We evaluated the performance of structured data algorithms, NLP approaches, and LLMs in identifying 7 immunosuppressive conditions and 6 immunosuppressive medications. Results: In the primary cohort, structured data algorithms achieved peak F1 scores ranging from 0.30 to 0.97 for identifying immunosuppressive conditions and medications. NLP approaches achieved peak F1 scores ranging from 0 to 1. GPT-4o outperformed or matched structured data algorithms and NLP approaches across all conditions and medications, with F1 scores ranging from 0.51 to 1. GPT-4o also performed impressively in our validation cohort (F1 = 1 for 8/13 variables). Discussion: LLMs, particularly GPT-4o, outperformed structured data algorithms and NLP approaches in identifying immunosuppressive conditions and medications with robust external validation. Conclusion: LLMs can be applied for improved cohort identification for research purposes.

Objective: Covariate-adaptive randomization algorithms (CARAs) can reduce covariate imbalance in randomized controlled trials (RCTs), but a lack of integration into Research Electronic Data Capture (REDCap) has limited their use. We developed a software pipeline to seamlessly integrate CARAs into REDCap as part of the all2GETHER study, a 2-armed RCT concerning HIV prevention. Materials and Methods: Leveraging REDCap’s Data Entry Trigger and a separate server, we implemented software in PHP and R to automate randomizations for all2GETHER. Randomizations were triggered by saving a specific REDCap form and were automatically communicated to unblinded study personnel. Results: Study arms were highly comparable, with differences across covariates characterized by Cohen’s d = 0.003 for continuous variables and risk differences <2.4% for categorical/binary variables. Conclusions: Our pipeline proved effective at reducing covariate imbalance with minimal additional effort for study personnel. Discussion: This pipeline is reproducible and could be used by other RCTs that collect data via REDCap.

Pharmacogenomics implementation efforts have increased over the last decade, but no published prospective pharmacogenomics trials have formally evaluated utility in underrepresented populations. We present final results from the ACCOuNT trial, a multi‐institutional prospective study [NCT03225820] in which African American inpatients were genotyped and pharmacogenomic results made available via an integrated, scalable pharmacogenomics clinical decision support (PGx CDS) system to guide prescribing. We hypothesized that PGx CDS utilization would improve genomically concordant prescribing during inpatient admissions. We recruited 531 patients, of which 187 were readmitted (60.4% female, median age = 55 [range: 19–88], median readmissions/patient = 2 [range: 1–17]), resulting in 518 evaluable admissions during which the effect of available pharmacogenomic results was measured. In 50% of admissions, care teams utilized PGx CDS. 12% of inpatient prescriptions had pharmacogenomic annotations, with only 3.6% pharmacogenomically associated with increased caution (“discordant”); no PGx high‐risk medications were prescribed. PGx CDS use by provider teams did not decrease genomically discordant prescribing (3.9% vs. 3.3% without PGx CDS use, P = 0.53). Nevertheless, inpatient care teams repeatedly consulted PGx CDS ostensibly to affirm genomically concordant prescriptions, as 48% of all admissions and 91% of discharges included pharmacogenomically annotated medications, mostly genomically concordant. Notably, timing of PGx CDS utilization correlated tightly with prescribing timing (correlation coefficient = 0.84, P < 0.05). In the first results of a multi‐institutional, prospective pharmacogenomics trial focused on underrepresented patients, we found the relevance of currently actionable pharmacogenomics information for guiding inpatient prescribing to be modest; but observed frequent use of concordant pharmacogenomics to affirm prescribing.

Severe community-acquired pneumonia (CAP) remains a major cause of critical illness, yet there are no validated early clinical criteria to predict short-term treatment outcomes in these patients. Short-term pneumonia treatment outcomes are less affected by confounding factors introduced by a prolonged hospital course, and early prediction of short-term treatment outcomes can help physicians identify those who are likely to fail the current treatment and implement adjustments to existing diagnostic and therapeutic plans. Traditional clinical stability criteria such as Halm’s criteria are not calibrated for early outcome prediction in critically ill severe pneumonia patients. We applied the XGBoost algorithm to predict pneumonia cure by day 7-8 post- intubation with clinical features from days 1-3 in mechanically ventilated patients with severe CAP from the Successful Clinical Response in Pneumonia Therapy (SCRIPT) study, a prospective cohort study at a tertiary academic center. Pneumonia episodes were adjudicated for day 7-8 cure status by a panel of critical care physicians using a structured review process. Clinical features that inform Halm’s criteria, including vital signs, oxygenation parameters, mental status, and vasopressor use, were extracted from the electronic health record. We also examined model performance by including additional features, such as laboratory data, ventilator settings, and medications. Basic demographic characteristics including age and BMI were also incorporated. Among 85 patients, 42 (49.4%) were cured by day 7–8. The best-performing model, which used Halm’s clinical features and ventilator features from days 1–3, achieved a cross-validated AUROC of 0.757. Inclusion of lab and medication data did not significantly improve performance. Key predictors included GCS, norepinephrine requirement, and BMI. We prove the feasibility of using ML models to predict short-term treatment outcomes of severe CAP among critically ill patients with basic clinical features. Future studies should focus on external validation and clinical integration to inform prognosis and early reevaluation of treatment strategy in patients with predicted poor outcomes.

Commonly used bisulfite-based procedures for DNA methylation sequencing can degrade DNA, worsening signal-to-noise ratios in samples with low DNA input. Enzymatic methylation sequencing (EM-seq) has been proposed as a less biased alternative for methylation profiling with greater genome coverage. Reduced representation approaches enrich samples for CpG-rich genomic regions, thereby enhancing throughput and cost effectiveness. We hypothesized that enzyme-based technology could be adapted for reduced representation methylation sequencing to enable DNA methylation profiling of low-input samples. We leveraged the well-established differences in methylation profile between mouse CD4+ T cell populations to compare the performance of our reduced representation EM-seq (RREM-seq) procedure against an established reduced representation bisulfite sequencing (RRBS) protocol. While the RRBS method failed to generate reliable DNA libraries when using 10-fold higher DNA input. RREM-seq also successfully detected lineage-defining methylation differences between alveolar conventional T and regulatory T cells obtained from patients with severe SARS-CoV-2 pneumonia. Our RREM-seq method enables single-nucleotide resolution methylation profiling using low-input samples, including from clinical sources.