Large Language Model Capabilities in Perioperative Risk Prediction and Prognostication.

Importance: General-domain large language models may be able to perform risk stratification and predict postoperative outcome measures using a description of the procedure and a patient's electronic health record notes. Objective: To examine predictive performance on 8 different tasks: prediction of American Society of Anesthesiologists Physical Status (ASA-PS), hospital admission, intensive care unit (ICU) admission, unplanned admission, hospital mortality, postanesthesia care unit (PACU) phase 1 duration, hospital duration, and ICU duration. Design, Setting, and Participants: This prognostic study included task-specific datasets constructed from 2 years of retrospective electronic health records data collected during routine clinical care. Case and note data were formatted into prompts and given to the large language model GPT-4 Turbo (OpenAI) to generate a prediction and explanation. The setting included a quaternary care center comprising 3 academic hospitals and affiliated clinics in a single metropolitan area. Patients who had a surgery or procedure with anesthesia and at least 1 clinician-written note filed in the electronic health record before surgery were included in the study. Data were analyzed from November to December 2023. Exposures: Compared original notes, note summaries, few-shot prompting, and chain-of-thought prompting strategies. Main Outcomes and Measures: F1 score for binary and categorical outcomes. Mean absolute error for numerical duration outcomes. Results: Study results were measured on task-specific datasets, each with 1000 cases with the exception of unplanned admission, which had 949 cases, and hospital mortality, which had 576 cases. The best results for each task included an F1 score of 0.50 (95% CI, 0.47-0.53) for ASA-PS, 0.64 (95% CI, 0.61-0.67) for hospital admission, 0.81 (95% CI, 0.78-0.83) for ICU admission, 0.61 (95% CI, 0.58-0.64) for unplanned admission, and 0.86 (95% CI, 0.83-0.89) for hospital mortality prediction. Performance on duration prediction tasks was universally poor across all prompt strategies for which the large language model achieved a mean absolute error of 49 minutes (95% CI, 46-51 minutes) for PACU phase 1 duration, 4.5 days (95% CI, 4.2-5.0 days) for hospital duration, and 1.1 days (95% CI, 0.9-1.3 days) for ICU duration prediction. Conclusions and Relevance: Current general-domain large language models may assist clinicians in perioperative risk stratification on classification tasks but are inadequate for numerical duration predictions. Their ability to produce high-quality natural language explanations for the predictions may make them useful tools in clinical workflows and may be complementary to traditional risk prediction models.

Intra- and post-pandemic impact of the COVID-19 outbreak on Stanford Health Care.

Stanford Health Care, which provides about 7% of overall healthcare to approximately 9 million people in the San Francisco Bay Area, has undergone significant changes due to the opening of a second hospital in late 2019 and, more importantly, the COVID-19 pandemic. We examine the impact of these events on anatomic pathology (AP) cases, aiming to enhance operational efficiency in response to evolving healthcare demands. We extracted historical census, admission, lab tests, operation, and AP data since 2015. An approximately 45% increase in the volume of laboratory tests (P < 0.0001) and a 17% increase in AP cases (P < 0.0001) occurred post-pandemic. These increases were associated with progressively increasing (P < 0.0001) hospital census. Census increase stemmed from higher admission through the emergency department (ED), and longer lengths of stay mostly for transfer patients, likely due to the greater capability of the new ED and changes in regional and local practice patterns post-pandemic. Higher census led to overcapacity, which has an inverted U relationship that peaked at 103% capacity for AP cases and 114% capacity for laboratory tests. Overcapacity led to a lower capability to perform clinical activities, particularly those related to surgical procedures. We conclude by suggesting parameters for optimal operations in the post-pandemic era.

An immune signature of postoperative cognitive decline in elderly patients.

Postoperative cognitive decline (POCD) is the predominant complication affecting elderly patients following major surgery, yet its prediction and prevention remain challenging. Understanding biological processes underlying the pathogenesis of POCD is essential for identifying mechanistic biomarkers to advance diagnostics and therapeutics. This longitudinal study involving 26 elderly patients undergoing orthopedic surgery aimed to characterize the impact of peripheral immune cell responses to surgical trauma on POCD. Trajectory analyses of single-cell mass cytometry data highlighted early JAK/STAT signaling exacerbation and diminished MyD88 signaling post-surgery in patients who developed POCD. Further analyses integrating single-cell and plasma proteomic data collected before surgery with clinical variables yielded a sparse predictive model that accurately identified patients who would develop POCD (AUC = 0.80). The resulting POCD immune signature included one plasma protein and ten immune cell features, offering a concise list of biomarker candidates for developing point-of-care prognostic tests to personalize perioperative management of at-risk patients. The code and the data are documented and available at https://github.com/gregbellan/POCD . Teaser: Modeling immune cell responses and plasma proteomic data predicts postoperative cognitive decline.

Liquid-biopsy proteomics combined with AI identifies cellular drivers of eye aging and disease in vivo.

Single-cell analysis in living humans is essential for understanding disease mechanisms, but it is impractical in non-regenerative organs, such as the eye and brain, because tissue biopsies would cause serious damage. We resolve this problem by integrating proteomics of liquid biopsies with single-cell transcriptomics from all known ocular cell types to trace the cellular origin of 5,953 proteins detected in the aqueous humor. We identified hundreds of cell-specific protein markers, including for individual retinal cell types. Surprisingly, our results reveal that retinal degeneration occurs in Parkinson's disease, and the cells driving diabetic retinopathy switch with disease stage. Finally, we developed artificial intelligence (AI) models to assess individual cellular aging and found that many eye diseases not associated with chronological age undergo accelerated molecular aging of disease-specific cell types. Our approach, which can be applied to other organ systems, has the potential to transform molecular diagnostics and prognostics while uncovering new cellular disease and aging mechanisms.

Variability and relative contribution of surgeon- and anesthesia-specific time components to total procedural time in cardiac surgery.

BACKGROUND: Decreasing variability in time-intensive tasks during cardiac surgery may reduce total procedural time, lower costs, reduce clinician burnout, and improve patient access. The relative contribution and variability of surgeon control time (SCT) and anesthesia control time (ACT) to total procedural time is unknown. METHODS: A total of 669 patients undergoing coronary artery bypass graft (CABG) surgery were enrolled. Using linear regression, we estimated adjusted SCTs and ACTs, controlling for patient and procedural covariates. The primary endpoint compared overall SCTs and ACTs. The secondary endpoint compared the variability in adjusted SCTs and ACTs. Sensitivity analyses quantified the relative importance of the specific surgeon and anesthesiologist in the adjusted linear models. RESULTS: The median SCT was 4.1 hours (interquartile range [IQR], 3.4-4.9 hours) compared to a median ACT of 1.0 hours (IQR, 0.8-1.2 hours; P < .001). Using linear regression, the variability in adjusted SCT among surgeons (range, 1.8 hours) was 3.5-fold greater than the variability in adjusted ACT among anesthesiologists (range, 0.5 hour; P < .001). The specific surgeon and anesthesiologist accounted for 50% of the explanatory power of the predictive model (P < .001). CONCLUSIONS: SCT variability is significantly greater than ACT variability and is strongly associated with the surgeon performing the procedure. Although these results suggest that SCT variability is an attractive operational target, further studies are needed to determine practitioner specific and modifiable attributes to reduce variability and improve efficiency.

Early response evaluation by single cell signaling profiling in acute myeloid leukemia.

Aberrant pro-survival signaling is a hallmark of cancer cells, but the response to chemotherapy is poorly understood. In this study, we investigate the initial signaling response to standard induction chemotherapy in a cohort of 32 acute myeloid leukemia (AML) patients, using 36-dimensional mass cytometry. Through supervised and unsupervised machine learning approaches, we find that reduction of extracellular-signal-regulated kinase (ERK) 1/2 and p38 mitogen-activated protein kinase (MAPK) phosphorylation in the myeloid cell compartment 24 h post-chemotherapy is a significant predictor of patient 5-year overall survival in this cohort. Validation by RNA sequencing shows induction of MAPK target gene expression in patients with high phospho-ERK1/2 24 h post-chemotherapy, while proteomics confirm an increase of the p38 prime target MAPK activated protein kinase 2 (MAPKAPK2). In this study, we demonstrate that mass cytometry can be a valuable tool for early response evaluation in AML and elucidate the potential of functional signaling analyses in precision oncology diagnostics.

In-silico generation of high-dimensional immune response data in patients using a deep neural network.

Technologies for single-cell profiling of the immune system have enabled researchers to extract rich interconnected networks of cellular abundance, phenotypical and functional cellular parameters. These studies can power machine learning approaches to understand the role of the immune system in various diseases. However, the performance of these approaches and the generalizability of the findings have been hindered by limited cohort sizes in translational studies, partially due to logistical demands and costs associated with longitudinal data collection in sufficiently large patient cohorts. An evolving challenge is the requirement for ever-increasing cohort sizes as the dimensionality of datasets grows. We propose a deep learning model derived from a novel pipeline of optimal temporal cell matching and overcomplete autoencoders that uses data from a small subset of patients to learn to forecast an entire patient's immune response in a high dimensional space from one timepoint to another. In our analysis of 1.08 million cells from patients pre- and post-surgical intervention, we demonstrate that the generated patient-specific data are qualitatively and quantitatively similar to real patient data by demonstrating fidelity, diversity, and usefulness.

Objective Activity Parameters Track Patient-specific Physical Recovery Trajectories After Surgery and Link With Individual Preoperative Immune States.

OBJECTIVE: The longitudinal assessment of physical function with high temporal resolution at a scalable and objective level in patients recovering from surgery is highly desirable to understand the biological and clinical factors that drive the clinical outcome. However, physical recovery from surgery itself remains poorly defined and the utility of wearable technologies to study recovery after surgery has not been established. BACKGROUND: Prolonged postoperative recovery is often associated with long-lasting impairment of physical, mental, and social functions. Although phenotypical and clinical patient characteristics account for some variation of individual recovery trajectories, biological differences likely play a major role. Specifically, patient-specific immune states have been linked to prolonged physical impairment after surgery. However, current methods of quantifying physical recovery lack patient specificity and objectivity. METHODS: Here, a combined high-fidelity accelerometry and state-of-the-art deep immune profiling approach was studied in patients undergoing major joint replacement surgery. The aim was to determine whether objective physical parameters derived from accelerometry data can accurately track patient-specific physical recovery profiles (suggestive of a 'clock of postoperative recovery'), compare the performance of derived parameters with benchmark metrics including step count, and link individual recovery profiles with patients' preoperative immune state. RESULTS: The results of our models indicate that patient-specific temporal patterns of physical function can be derived with a precision superior to benchmark metrics. Notably, 6 distinct domains of physical function and sleep are identified to represent the objective temporal patterns: ''activity capacity'' and ''moderate and overall activity (declined immediately after surgery); ''sleep disruption and sedentary activity (increased after surgery); ''overall sleep'', ''sleep onset'', and ''light activity'' (no clear changes were observed after surgery). These patterns can be linked to individual patients preopera-tive immune state using cross-validated canonical-correlation analysis. Importantly, the pSTAT3 signal activity in monocytic myeloid-derived suppressor cells predicted a slower recovery. CONCLUSIONS: Accelerometry-based recovery trajectories are scalable and objective outcomes to study patient-specific factors that drive physical recovery.

Integrated Single-cell and Plasma Proteomic Modeling to Predict Surgical Site Complications: A Prospective Cohort Study.

OBJECTIVE: The aim of this study was to determine whether single-cell and plasma proteomic elements of the host's immune response to surgery accurately identify patients who develop a surgical site complication (SSC) after major abdominal surgery. SUMMARY BACKGROUND DATA: SSCs may occur in up to 25% of patients undergoing bowel resection, resulting in significant morbidity and economic burden. However, the accurate prediction of SSCs remains clinically challenging. Leveraging high-content proteomic technologies to comprehensively profile patients' immune response to surgery is a promising approach to identify predictive biological factors of SSCs. METHODS: Forty-one patients undergoing non-cancer bowel resection were prospectively enrolled. Blood samples collected before surgery and on postoperative day one (POD1) were analyzed using a combination of single-cell mass cytometry and plasma proteomics. The primary outcome was the occurrence of an SSC, including surgical site infection, anastomotic leak, or wound dehiscence within 30 days of surgery. RESULTS: A multiomic model integrating the single-cell and plasma proteomic data collected on POD1 accurately differentiated patients with (n = 11) and without (n = 30) an SSC [area under the curve (AUC) = 0.86]. Model features included coregulated proinflammatory (eg, IL-6- and MyD88- signaling responses in myeloid cells) and immunosuppressive (eg, JAK/STAT signaling responses in M-MDSCs and Tregs) events preceding an SSC. Importantly, analysis of the immunological data obtained before surgery also yielded a model accurately predicting SSCs (AUC = 0.82). CONCLUSIONS: The multiomic analysis of patients' immune response after surgery and immune state before surgery revealed systemic immune signatures preceding the development of SSCs. Our results suggest that integrating immunological data in perioperative risk assessment paradigms is a plausible strategy to guide individualized clinical care.

Human influenza virus challenge identifies cellular correlates of protection for oral vaccination.

Developing new influenza vaccines with improved performance and easier administration routes hinges on defining correlates of protection. Vaccine-elicited cellular correlates of protection for influenza in humans have not yet been demonstrated. A phase-2 double-blind randomized placebo and active (inactivated influenza vaccine) controlled study provides evidence that a human-adenovirus-5-based oral influenza vaccine tablet (VXA-A1.1) can protect from H1N1 virus challenge in humans. Mass cytometry characterization of vaccine-elicited cellular immune responses identified shared and vaccine-type-specific responses across B and T cells. For VXA-A1.1, the abundance of hemagglutinin-specific plasmablasts and plasmablasts positive for integrin α4ß7, phosphorylated STAT5, or lacking expression of CD62L at day 8 were significantly correlated with protection from developing viral shedding following virus challenge at day 90 and contributed to an effective machine learning model of protection. These findings reveal the characteristics of vaccine-elicited cellular correlates of protection for an oral influenza vaccine.

Measuring the human immune response to surgery: multiomics for the prediction of postoperative outcomes.

PURPOSE OF REVIEW: Postoperative complications including infections, cognitive impairment, and protracted recovery occur in one-third of the 300 million surgeries performed annually worldwide. Complications cause personal suffering along with a significant economic burden on our healthcare system. However, the accurate prediction of postoperative complications and patient-targeted interventions for their prevention remain as major clinical challenges. RECENT FINDINGS: Although multifactorial in origin, the dysregulation of immunological mechanisms that occur in response to surgical trauma is a key determinant of postoperative complications. Prior research, primarily focusing on inflammatory plasma markers, has provided important clues regarding their pathogenesis. However, the recent advent of high-content, single-cell transcriptomic, and proteomic technologies has considerably improved our ability to characterize the immune response to surgery, thereby providing new means to understand the immunological basis of postoperative complications and to identify prognostic biological signatures. SUMMARY: The comprehensive and single-cell characterization of the human immune response to surgery has significantly advanced our ability to predict the risk of postoperative complications. Multiomic modeling of patients' immune states holds promise for the discovery of preoperative predictive biomarkers, ultimately providing patients and surgeons with actionable information to improve surgical outcomes. Although recent studies have generated a wealth of knowledge, laying the foundation for a single-cell atlas of the human immune response to surgery, larger-scale multiomic studies are required to derive robust, scalable, and sufficiently powerful models to accurately predict the risk of postoperative complications in individual patients.

High-dimensional profiling clusters asthma severity by lymphoid and non-lymphoid status.

Clinical definitions of asthma fail to capture the heterogeneity of immune dysfunction in severe, treatment-refractory disease. Applying mass cytometry and machine learning to bronchoalveolar lavage (BAL) cells, we find that corticosteroid-resistant asthma patients cluster largely into two groups: one enriched in interleukin (IL)-4+ innate immune cells and another dominated by interferon (IFN)-γ+ T cells, including tissue-resident memory cells. In contrast, BAL cells of a healthier population are enriched in IL-10+ macrophages. To better understand cellular mediators of severe asthma, we developed the Immune Cell Linkage through Exploratory Matrices (ICLite) algorithm to perform deconvolution of bulk RNA sequencing of mixed-cell populations. Signatures of mitosis and IL-7 signaling in CD206-FcεRI+CD127+IL-4+ innate cells in one patient group, contrasting with adaptive immune response in T cells in the other, are preserved across technologies. Transcriptional signatures uncovered by ICLite identify T-cell-high and T-cell-poor severe asthma patients in an independent cohort, suggesting broad applicability of our findings.

Predicting Acute Pain After Surgery: A Multivariate Analysis.

OBJECTIVES: To identify perioperative practice patterns that predictably impact postoperative pain. BACKGROUND: Despite significant advances in perioperative medicine, a significant portion of patients still experience severe pain after major surgery. Postoperative pain is associated with serious adverse outcomes that are costly to patients and society. METHODS: The presented analysis took advantage of a unique observational data set providing unprecedented detailed pharmacological information. The data were collected by PAIN OUT, a multinational registry project established by the European Commission to improve postoperative pain outcomes. A multivariate approach was used to derive and validate a model predictive of pain on postoperative day 1 (POD1) in 1008 patients undergoing back surgery. RESULTS: The predictive and validated model was highly significant (P = 8.9E-15) and identified modifiable practice patterns. Importantly, the number of nonopioid analgesic drug classes administered during surgery predicted decreased pain on POD1. At least 2 different nonopioid analgesic drug classes (cyclooxygenase inhibitors, acetaminophen, nefopam, or metamizol) were required to provide meaningful pain relief (>30%). However, only a quarter of patients received at least 2 nonanalgesic drug classes during surgery. In addition, the use of very short-acting opioids predicted increased pain on POD1, suggesting room for improvement in the perioperative management of these patients. Although the model was highly significant, it only accounted for a relatively small fraction of the observed variance. CONCLUSION: The presented analysis offers detailed insight into current practice patterns and reveals modifications that can be implemented in today's clinical practice. Our results also suggest that parameters other than those currently studied are relevant for postoperative pain including biological and psychological variables.

Mid-gestation serum lipidomic profile associations with spontaneous preterm birth are influenced by body mass index.

Spontaneous preterm birth (sPTB) is a major cause of infant morbidity and mortality. While metabolic changes leading to preterm birth are unknown, several factors including dyslipidemia and inflammation have been implicated and paradoxically both low (<18.5 kg/m2) and high (>30 kg/m2) body mass indices (BMIs) are risk factors for this condition. The objective of the study was to identify BMI-associated metabolic perturbations and potential mid-gestation serum biomarkers of preterm birth in a cohort of underweight, normal weight and obese women experiencing either sPTB or full-term deliveries (n = 102; n = 17/group). For this purpose, we combined untargeted metabolomics and lipidomics with targeted metabolic profiling of major regulators of inflammation and metabolism, including oxylipins, endocannabinoids, bile acids and ceramides. Women who were obese and had sPTB showed elevated oxidative stress and dyslipidemia characterized by elevated serum free fatty acids. Women who were underweight-associated sPTB also showed evidence of dyslipidemia characterized by elevated phospholipids, unsaturated triglycerides, sphingomyelins, cholesteryl esters and long-chain acylcarnitines. In normal weight women experiencing sPTB, the relative abundance of 14(15)-epoxyeicosatrienoic acid and 14,15-dihydroxyeicosatrienoic acids to other regioisomers were altered at mid-pregnancy. This phenomenon is not yet associated with any biological process, but may be linked to estrogen metabolism. These changes were differentially modulated across BMI groups. In conclusion, using metabolomics we observed distinct BMI-dependent metabolic manifestations among women who had sPTB. These observations suggest the potential to predict sPTB mid-gestation using a new set of metabolomic markers and BMI stratification. This study opens the door to further investigate the role of cytochrome P450/epoxide hydrolase metabolism in sPTB.

Preferential inhibition of adaptive immune system dynamics by glucocorticoids in patients after acute surgical trauma.

Glucocorticoids (GC) are a controversial yet commonly used intervention in the clinical management of acute inflammatory conditions, including sepsis or traumatic injury. In the context of major trauma such as surgery, concerns have been raised regarding adverse effects from GC, thereby necessitating a better understanding of how GCs modulate the immune response. Here we report the results of a randomized controlled trial (NCT02542592) in which we employ a high-dimensional mass cytometry approach to characterize innate and adaptive cell signaling dynamics after a major surgery (primary outcome) in patients treated with placebo or methylprednisolone (MP). A robust, unsupervised bootstrap clustering of immune cell subsets coupled with random forest analysis shows profound (AUC = 0.92, p-value = 3.16E-8) MP-induced alterations of immune cell signaling trajectories, particularly in the adaptive compartments. By contrast, key innate signaling responses previously associated with pain and functional recovery after surgery, including STAT3 and CREB phosphorylation, are not affected by MP. These results imply cell-specific and pathway-specific effects of GCs, and also prompt future studies to examine GCs' effects on clinical outcomes likely dependent on functional adaptive immune responses.

iRhom2 inhibits bile duct obstruction-induced liver fibrosis.

Chronic liver disease can induce prolonged activation of hepatic stellate cells, which may result in liver fibrosis. Inactive rhomboid protein 2 (iRhom2) is required for the maturation of A disintegrin and metalloprotease 17 (ADAM17, also called TACE), which is responsible for the cleavage of membrane-bound tumor necrosis factor-α (TNF-α) and its receptors (TNFRs). Here, using the murine bile duct ligation (BDL) model, we showed that the abundance of iRhom2 and activation of ADAM17 increased during liver fibrosis. Consistent with this, concentrations of ADAM17 substrates were increased in plasma samples from mice after BDL and in patients suffering from liver cirrhosis. We observed increased liver fibrosis, accelerated disease progression, and an increase in activated stellate cells after BDL in mice lacking iRhom2 (Rhbdf2-/- ) compared to that in controls. In vitro primary mouse hepatic stellate cells exhibited iRhom2-dependent shedding of the ADAM17 substrates TNFR1 and TNFR2. In vivo TNFR shedding after BDL also depended on iRhom2. Treatment of Rhbdf2-/- mice with the TNF-α inhibitor etanercept reduced the presence of activated stellate cells and alleviated liver fibrosis after BDL. Together, these data suggest that iRhom2-mediated inhibition of TNFR signaling protects against liver fibrosis.

A topological view of human CD34+ cell state trajectories from integrated single-cell output and proteomic data.

Recent advances in single-cell molecular analytical methods and clonal growth assays are enabling more refined models of human hematopoietic lineage restriction processes to be conceptualized. Here, we report the results of integrating single-cell proteome measurements with clonally determined lymphoid, neutrophilic/monocytic, and/or erythroid progeny outputs from >1000 index-sorted CD34+ human cord blood cells in short-term cultures with and without stromal cells. Surface phenotypes of functionally examined cells were individually mapped onto a molecular landscape of the entire CD34+ compartment constructed from single-cell mass cytometric measurements of 14 cell surface markers, 20 signaling/cell cycle proteins, and 6 transcription factors in ∼300 000 cells. This analysis showed that conventionally defined subsets of CD34+ cord blood cells are heterogeneous in their functional properties, transcription factor content, and signaling activities. Importantly, this molecular heterogeneity was reduced but not eliminated in phenotypes that were found to display highly restricted lineage outputs. Integration of the complete proteomic and functional data sets obtained revealed a continuous probabilistic topology of change that includes a multiplicity of lineage restriction trajectories. Each of these reflects progressive but variable changes in the levels of specific signaling intermediates and transcription factors but shared features of decreasing quiescence. Taken together, our results suggest a model in which increasingly narrowed hematopoietic output capabilities in neonatal CD34+ cord blood cells are determined by a history of external stimulation in combination with innately programmed cell state changes.

Single-cell analysis identifies a CD33+ subset of human cord blood cells with high regenerative potential.

Elucidation of the identity and diversity of mechanisms that sustain long-term human blood cell production remains an important challenge. Previous studies indicate that, in adult mice, this property is vested in cells identified uniquely by their ability to clonally regenerate detectable, albeit highly variable levels and types, of mature blood cells in serially transplanted recipients. From a multi-parameter analysis of the molecular features of very primitive human cord blood cells that display long-term cell outputs in vitro and in immunodeficient mice, we identified a prospectively separable CD33+CD34+CD38-CD45RA-CD90+CD49f+ phenotype with serially transplantable, but diverse, cell output profiles. Single-cell measurements of the mitogenic response, and the transcriptional, DNA methylation and 40-protein content of this and closely related phenotypes revealed subtle but consistent differences both within and between each subset. These results suggest that multiple regulatory mechanisms combine to maintain different cell output activities of human blood cell precursors with high regenerative potential.

Single-cell developmental classification of B cell precursor acute lymphoblastic leukemia at diagnosis reveals predictors of relapse.

Insight into the cancer cell populations that are responsible for relapsed disease is needed to improve outcomes. Here we report a single-cell-based study of B cell precursor acute lymphoblastic leukemia at diagnosis that reveals hidden developmentally dependent cell signaling states that are uniquely associated with relapse. By using mass cytometry we simultaneously quantified 35 proteins involved in B cell development in 60 primary diagnostic samples. Each leukemia cell was then matched to its nearest healthy B cell population by a developmental classifier that operated at the single-cell level. Machine learning identified six features of expanded leukemic populations that were sufficient to predict patient relapse at diagnosis. These features implicated the pro-BII subpopulation of B cells with activated mTOR signaling, and the pre-BI subpopulation of B cells with activated and unresponsive pre-B cell receptor signaling, to be associated with relapse. This model, termed 'developmentally dependent predictor of relapse' (DDPR), significantly improves currently established risk stratification methods. DDPR features exist at diagnosis and persist at relapse. By leveraging a data-driven approach, we demonstrate the predictive value of single-cell 'omics' for patient stratification in a translational setting and provide a framework for its application to human cancer.