The complexity of preterm birth (PTB), both spontaneous and medically indicated, and its various etiologies and associated risk factors pose a significant challenge for developing tools to accurately predict risk. This review focuses on the discovery of proteomics signatures that might be useful for predicting spontaneous PTB or preeclampsia, which often results in PTB. We describe methods for proteomics analyses, proteomics biomarker candidates that have so far been identified, obstacles for discovering biomarkers that are sufficiently accurate for clinical use, and the derivation of composite signatures including clinical parameters to increase predictive power.
Biomarkers , Premature Birth , Proteomics , Humans , Female , Pregnancy , Biomarkers/metabolism , Pre-Eclampsia/diagnosis , Pre-Eclampsia/metabolism , Infant, Newborn , Predictive Value of Tests
[This corrects the article DOI: 10.3389/falgy.2023.1149008.].
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.
Adoption of high-content omic technologies in clinical studies, coupled with computational methods, has yielded an abundance of candidate biomarkers. However, translating such findings into bona fide clinical biomarkers remains challenging. To facilitate this process, we introduce Stabl, a general machine learning method that identifies a sparse, reliable set of biomarkers by integrating noise injection and a data-driven signal-to-noise threshold into multivariable predictive modeling. Evaluation of Stabl on synthetic datasets and five independent clinical studies demonstrates improved biomarker sparsity and reliability compared to commonly used sparsity-promoting regularization methods while maintaining predictive performance; it distills datasets containing 1,400-35,000 features down to 4-34 candidate biomarkers. Stabl extends to multi-omic integration tasks, enabling biological interpretation of complex predictive models, as it hones in on a shortlist of proteomic, metabolomic and cytometric events predicting labor onset, microbial biomarkers of pre-term birth and a pre-operative immune signature of post-surgical infections. Stabl is available at https://github.com/gregbellan/Stabl .
Advanced measurement and data storage technologies have enabled high-dimensional profiling of complex biological systems. For this, modern multiomics studies regularly produce datasets with hundreds of thousands of measurements per sample, enabling a new era of precision medicine. Correlation analysis is an important first step to gain deeper insights into the coordination and underlying processes of such complex systems. However, the construction of large correlation networks in modern high-dimensional datasets remains a major computational challenge owing to rapidly growing runtime and memory requirements. Here we address this challenge by introducing CorALS (Correlation Analysis of Large-scale (biological) Systems), an open-source framework for the construction and analysis of large-scale parametric as well as non-parametric correlation networks for high-dimensional biological data. It features off-the-shelf algorithms suitable for both personal and high-performance computers, enabling workflows and downstream analysis approaches. We illustrate the broad scope and potential of CorALS by exploring perspectives on complex biological processes in large-scale multiomics and single-cell studies.
Preterm birth (PTB) is the leading cause of infant mortality globally. Research has focused on developing predictive models for PTB without prioritizing cost-effective interventions. Physical activity and sleep present unique opportunities for interventions in low- and middle-income populations (LMICs). However, objective measurement of physical activity and sleep remains challenging and self-reported metrics suffer from low-resolution and accuracy. In this study, we use physical activity data collected using a wearable device comprising over 181,944 h of data across N = 1083 patients. Using a new state-of-the art deep learning time-series classification architecture, we develop a 'clock' of healthy dynamics during pregnancy by using gestational age (GA) as a surrogate for progression of pregnancy. We also develop novel interpretability algorithms that integrate unsupervised clustering, model error analysis, feature attribution, and automated actigraphy analysis, allowing for model interpretation with respect to sleep, activity, and clinical variables. Our model performs significantly better than 7 other machine learning and AI methods for modeling the progression of pregnancy. We found that deviations from a normal 'clock' of physical activity and sleep changes during pregnancy are strongly associated with pregnancy outcomes. When our model underestimates GA, there are 0.52 fewer preterm births than expected (P = 1.01e - 67, permutation test) and when our model overestimates GA, there are 1.44 times (P = 2.82e - 39, permutation test) more preterm births than expected. Model error is negatively correlated with interdaily stability (P = 0.043, Spearman's), indicating that our model assigns a more advanced GA when an individual's daily rhythms are less precise. Supporting this, our model attributes higher importance to sleep periods in predicting higher-than-actual GA, relative to lower-than-actual GA (P = 1.01e - 21, Mann-Whitney U). Combining prediction and interpretability allows us to signal when activity behaviors alter the likelihood of preterm birth and advocates for the development of clinical decision support through passive monitoring and exercise habit and sleep recommendations, which can be easily implemented in LMICs.
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.
BACKGROUND: Blood proteins are frequently measured in serum or plasma, because they provide a wealth of information. Differences in the ex vivo processing of serum and plasma raise concerns that proteomic health and disease signatures derived from serum or plasma differ in content and quality. However, little is known about their respective power to predict feto-maternal health outcomes. Predictive power is a sentinel characteristic to determine the clinical use of biosignatures. OBJECTIVE: This study aimed to compare the power of serum and plasma proteomic signatures to predict a physiological pregnancy outcome. STUDY DESIGN: Paired serum and plasma samples from 73 women were obtained from biorepositories of a multinational prospective cohort study on pregnancy outcomes. Gestational age at the time of sampling was the predicted outcome, because the proteomic signatures have been validated for such a prediction. Multivariate and cross-validated models were independently derived for serum and plasma proteins. RESULTS: A total of 1116 proteins were measured in 88 paired samples from 73 women with a highly multiplexed platform using proximity extension technology (Olink Proteomics Inc, Watertown, MA). The plasma proteomic signature showed a higher predictive power (R=0.64; confidence interval, 0.42-0.79; P=3.5×10-6) than the serum signature (R=0.45; confidence interval, 0.18-0.66; P=2.2×10-3). The serum signature was validated in plasma with a similar predictive power (R=0.58; confidence interval, 0.34-0.75; P=4.8×10-5), whereas the plasma signature was validated in serum with reduced predictive power (R=0.53; confidence interval, 0.27-0.72; P=2.6×10-4). Signature proteins largely overlapped in the serum and plasma, but the strength of association with gestational age was weaker for serum proteins. CONCLUSION: Findings suggest that serum proteomics are less informative than plasma proteomics. They are compatible with the view that the partial ex-vivo degradation and modification of serum proteins during sample processing are an underlying reason. The rationale for collecting and analyzing serum and plasma samples should be carefully considered when deriving proteomic biosignatures to ascertain that specimens of the highest scientific and clinical yield are processed. Findings suggest that plasma is the preferred matrix.
Preterm birth (PTB) is the leading cause of death in children under five, yet comprehensive studies are hindered by its multiple complex etiologies. Epidemiological associations between PTB and maternal characteristics have been previously described. This work used multiomic profiling and multivariate modeling to investigate the biological signatures of these characteristics. Maternal covariates were collected during pregnancy from 13,841 pregnant women across five sites. Plasma samples from 231 participants were analyzed to generate proteomic, metabolomic, and lipidomic datasets. Machine learning models showed robust performance for the prediction of PTB (AUROC = 0.70), time-to-delivery (r = 0.65), maternal age (r = 0.59), gravidity (r = 0.56), and BMI (r = 0.81). Time-to-delivery biological correlates included fetal-associated proteins (e.g., ALPP, AFP, and PGF) and immune proteins (e.g., PD-L1, CCL28, and LIFR). Maternal age negatively correlated with collagen COL9A1, gravidity with endothelial NOS and inflammatory chemokine CXCL13, and BMI with leptin and structural protein FABP4. These results provide an integrated view of epidemiological factors associated with PTB and identify biological signatures of clinical covariates affecting this disease.
Premature Birth , Infant, Newborn , Pregnancy , Child , Humans , Female , Premature Birth/epidemiology , Developing Countries , Multiomics , Proteomics , Chemokines, CC
The prevalence of food allergy continues to rise globally, carrying with it substantial safety, economic, and emotional burdens. Although preventative strategies do exist, the heterogeneity of allergy trajectories and clinical phenotypes has made it difficult to identify patients who would benefit from these strategies. Therefore, further studies investigating the molecular mechanisms that differentiate these trajectories are needed. Large-scale omics studies have identified key insights into the molecular mechanisms for many different diseases, however the application of these technologies to uncover the drivers of food allergy development is in its infancy. Here we review the use of omics approaches in food allergy and highlight key gaps in knowledge for applying these technologies for the characterization of food allergy development.
High-content omic technologies coupled with sparsity-promoting regularization methods (SRM) have transformed the biomarker discovery process. However, the translation of computational results into a clinical use-case scenario remains challenging. A rate-limiting step is the rigorous selection of reliable biomarker candidates among a host of biological features included in multivariate models. We propose Stabl, a machine learning framework that unifies the biomarker discovery process with multivariate predictive modeling of clinical outcomes by selecting a sparse and reliable set of biomarkers. Evaluation of Stabl on synthetic datasets and four independent clinical studies demonstrates improved biomarker sparsity and reliability compared to commonly used SRMs at similar predictive performance. Stabl readily extends to double- and triple-omics integration tasks and identifies a sparser and more reliable set of biomarkers than those selected by state-of-the-art early- and late-fusion SRMs, thereby facilitating the biological interpretation and clinical translation of complex multi-omic predictive models. The complete package for Stabl is available online at https://github.com/gregbellan/Stabl.
Although prematurity is the single largest cause of death in children under 5 years of age, the current definition of prematurity, based on gestational age, lacks the precision needed for guiding care decisions. Here, we propose a longitudinal risk assessment for adverse neonatal outcomes in newborns based on a deep learning model that uses electronic health records (EHRs) to predict a wide range of outcomes over a period starting shortly before conception and ending months after birth. By linking the EHRs of the Lucile Packard Children's Hospital and the Stanford Healthcare Adult Hospital, we developed a cohort of 22,104 mother-newborn dyads delivered between 2014 and 2018. Maternal and newborn EHRs were extracted and used to train a multi-input multitask deep learning model, featuring a long short-term memory neural network, to predict 24 different neonatal outcomes. An additional cohort of 10,250 mother-newborn dyads delivered at the same Stanford Hospitals from 2019 to September 2020 was used to validate the model. Areas under the receiver operating characteristic curve at delivery exceeded 0.9 for 10 of the 24 neonatal outcomes considered and were between 0.8 and 0.9 for 7 additional outcomes. Moreover, comprehensive association analysis identified multiple known associations between various maternal and neonatal features and specific neonatal outcomes. This study used linked EHRs from more than 30,000 mother-newborn dyads and would serve as a resource for the investigation and prediction of neonatal outcomes. An interactive website is available for independent investigators to leverage this unique dataset: https://maternal-child-health-associations.shinyapps.io/shiny_app/.
Infant Health , Infant, Premature , Adult , Child , Infant, Newborn , Humans , Child, Preschool , Gestational Age , Morbidity , Risk Assessment
OBJECTIVES: The aim of the study was to: (1) Identify (early in pregnancy) psychosocial and stress-related factors that predict risk of spontaneous preterm birth (PTB, gestational age <37 weeks); (2) Investigate whether "protective" factors (e.g., happiness/social support) decrease risk; (3) Use the Dhabhar Quick-Assessment Questionnaire for Stress and Psychosocial Factors (DQAQ-SPF) to rapidly quantify harmful or protective factors that predict increased or decreased risk respectively, of PTB. STUDY DESIGN: This is a prospective cohort study. Relative risk (RR) analyses investigated association between individual factors and PTB. Machine learning-based interdependency analysis (IDPA) identified factor clusters, strength, and direction of association with PTB. A nonlinear model based on support vector machines was built for predicting PTB and identifying factors that most strongly predicted PTB. RESULTS: Higher levels of deleterious factors were associated with increased RR for PTB: General anxiety (RR = 8.9; 95% confidence interval [CI] = 2.0,39.6), pain (RR = 5.7; CI = 1.7,17.0); tiredness/fatigue (RR = 3.7; CI = 1.09,13.5); perceived risk of birth complications (RR = 4; CI = 1.6,10.01); self-rated health current (RR = 2.6; CI = 1.0,6.7) and previous 3 years (RR = 2.9; CI = 1.1,7.7); and divorce (RR = 2.9; CI = 1.1,7.8). Lower levels of protective factors were also associated with increased RR for PTB: low happiness (RR = 9.1; CI = 1.25,71.5); low support from parents/siblings (RR = 3.5; CI = 0.9,12.9), and father-of-baby (RR = 3; CI = 1.1,9.9). These factors were also components of the clusters identified by the IDPA: perceived risk of birth complications (p < 0.05 after FDR correction), and general anxiety, happiness, tiredness/fatigue, self-rated health, social support, pain, and sleep (p < 0.05 without FDR correction). Supervised analysis of all factors, subject to cross-validation, produced a model highly predictive of PTB (AUROC or area under the receiver operating characteristic = 0.73). Model reduction through forward selection revealed that even a small set of factors (including those identified by RR and IDPA) predicted PTB. CONCLUSION: These findings represent an important step toward identifying key factors, which can be assessed rapidly before/after conception, to predict risk of PTB, and perhaps other adverse pregnancy outcomes. Quantifying these factors, before, or early in pregnancy, could identify women at risk of delivering preterm, pinpoint mechanisms/targets for intervention, and facilitate the development of interventions to prevent PTB. KEY POINTS: · Newly designed questionnaire used for rapid quantification of stress and psychosocial factors early during pregnancy.. · Deleterious factors predict increased preterm birth (PTB) risk.. · Protective factors predict decreased PTB risk..
Premature Birth , Pregnancy , Infant, Newborn , Female , Humans , Infant , Premature Birth/prevention & control , Prospective Studies , Pregnancy Outcome , Gestational Age , Pain , Risk Factors
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.
Benchmarking , Exercise , Humans , Monocytes , Physical Examination , Postoperative Period
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.
Machine Learning , Neural Networks, Computer , Humans , Proteomics
Preeclampsia is a complex disease of pregnancy whose physiopathology remains unclear. We developed machine-learning models for early prediction of preeclampsia (first 16 weeks of pregnancy) and over gestation by analyzing six omics datasets from a longitudinal cohort of pregnant women. For early pregnancy, a prediction model using nine urine metabolites had the highest accuracy and was validated on an independent cohort (area under the receiver-operating characteristic curve [AUC] = 0.88, 95% confidence interval [CI] [0.76, 0.99] cross-validated; AUC = 0.83, 95% CI [0.62,1] validated). Univariate analysis demonstrated statistical significance of identified metabolites. An integrated multiomics model further improved accuracy (AUC = 0.94). Several biological pathways were identified including tryptophan, caffeine, and arachidonic acid metabolisms. Integration with immune cytometry data suggested novel associations between immune and proteomic dynamics. While further validation in a larger population is necessary, these encouraging results can serve as a basis for a simple, early diagnostic test for preeclampsia.
Psychosocial and stress-related factors (PSFs), defined as internal or external stimuli that induce biological changes, are potentially modifiable factors and accessible targets for interventions that are associated with adverse pregnancy outcomes (APOs). Although individual APOs have been shown to be connected to PSFs, they are biologically interconnected, relatively infrequent, and therefore challenging to model. In this context, multi-task machine learning (MML) is an ideal tool for exploring the interconnectedness of APOs on the one hand and building on joint combinatorial outcomes to increase predictive power on the other hand. Additionally, by integrating single cell immunological profiling of underlying biological processes, the effects of stress-based therapeutics may be measurable, facilitating the development of precision medicine approaches. Objectives: The primary objectives were to jointly model multiple APOs and their connection to stress early in pregnancy, and to explore the underlying biology to guide development of accessible and measurable interventions. Materials and Methods: In a prospective cohort study, PSFs were assessed during the first trimester with an extensive self-filled questionnaire for 200 women. We used MML to simultaneously model, and predict APOs (severe preeclampsia, superimposed preeclampsia, gestational diabetes and early gestational age) as well as several risk factors (BMI, diabetes, hypertension) for these patients based on PSFs. Strongly interrelated stressors were categorized to identify potential therapeutic targets. Furthermore, for a subset of 14 women, we modeled the connection of PSFs to the maternal immune system to APOs by building corresponding ML models based on an extensive single cell immune dataset generated by mass cytometry time of flight (CyTOF). Results: Jointly modeling APOs in a MML setting significantly increased modeling capabilities and yielded a highly predictive integrated model of APOs underscoring their interconnectedness. Most APOs were associated with mental health, life stress, and perceived health risks. Biologically, stressors were associated with specific immune characteristics revolving around CD4/CD8 T cells. Immune characteristics predicted based on stress were in turn found to be associated with APOs. Conclusions: Elucidating connections among stress, multiple APOs simultaneously, and immune characteristics has the potential to facilitate the implementation of ML-based, individualized, integrative models of pregnancy in clinical decision making. The modifiable nature of stressors may enable the development of accessible interventions, with success tracked through immune characteristics.
The biological determinants underlying the range of coronavirus 2019 (COVID-19) clinical manifestations are not fully understood. Here, over 1,400 plasma proteins and 2,600 single-cell immune features comprising cell phenotype, endogenous signaling activity, and signaling responses to inflammatory ligands are cross-sectionally assessed in peripheral blood from 97 patients with mild, moderate, and severe COVID-19 and 40 uninfected patients. Using an integrated computational approach to analyze the combined plasma and single-cell proteomic data, we identify and independently validate a multi-variate model classifying COVID-19 severity (multi-class area under the curve [AUC]training = 0.799, p = 4.2e-6; multi-class AUCvalidation = 0.773, p = 7.7e-6). Examination of informative model features reveals biological signatures of COVID-19 severity, including the dysregulation of JAK/STAT, MAPK/mTOR, and nuclear factor κB (NF-κB) immune signaling networks in addition to recapitulating known hallmarks of COVID-19. These results provide a set of early determinants of COVID-19 severity that may point to therapeutic targets for prevention and/or treatment of COVID-19 progression.
COVID-19 , Humans , NF-kappa B/metabolism , Proteomics , SARS-CoV-2 , Signal Transduction
INTRODUCTION: Current treatments for chronic musculoskeletal (MSK) pain are suboptimal. Discovery of robust prognostic markers separating patients who recover from patients with persistent pain and disability is critical for developing patient-specific treatment strategies and conceiving novel approaches that benefit all patients. Given that chronic pain is a biopsychosocial process, this study aims to discover and validate a robust prognostic signature that measures across multiple dimensions in the same adolescent patient cohort with a computational analysis pipeline. This will facilitate risk stratification in adolescent patients with chronic MSK pain and more resourceful allocation of patients to costly and potentially burdensome multidisciplinary pain treatment approaches. METHODS AND ANALYSIS: Here we describe a multi-institutional effort to collect, curate and analyse a high dimensional data set including epidemiological, psychometric, quantitative sensory, brain imaging and biological information collected over the course of 12 months. The aim of this effort is to derive a multivariate model with strong prognostic power regarding the clinical course of adolescent MSK pain and function. ETHICS AND DISSEMINATION: The study complies with the National Institutes of Health policy on the use of a single internal review board (sIRB) for multisite research, with Cincinnati Children's Hospital Medical Center Review Board as the reviewing IRB. Stanford's IRB is a relying IRB within the sIRB. As foreign institutions, the University of Toronto and The Hospital for Sick Children (SickKids) are overseen by their respective ethics boards. All participants provide signed informed consent. We are committed to open-access publication, so that patients, clinicians and scientists have access to the study data and the signature(s) derived. After findings are published, we will upload a limited data set for sharing with other investigators on applicable repositories. TRIAL REGISTRATION NUMBER: NCT04285112.
Chronic Pain , Musculoskeletal Pain , Adolescent , Humans , Multicenter Studies as Topic , Musculoskeletal Pain/diagnosis , National Institutes of Health (U.S.) , Pain Management , Prospective Studies , United States
Whereas prematurity is a major cause of neonatal mortality, morbidity, and lifelong impairment, the degree of prematurity is usually defined by the gestational age (GA) at delivery rather than by neonatal morbidity. Here we propose a multi-task deep neural network model that simultaneously predicts twelve neonatal morbidities, as the basis for a new data-driven approach to define prematurity. Maternal demographics, medical history, obstetrical complications, and prenatal fetal findings were obtained from linked birth certificates and maternal/infant hospitalization records for 11,594,786 livebirths in California from 1991 to 2012. Overall, our model outperformed traditional models to assess prematurity which are based on GA and/or birthweight (area under the precision-recall curve was 0.326 for our model, 0.229 for GA, and 0.156 for small for GA). These findings highlight the potential of using machine learning techniques to predict multiple prematurity phenotypes and inform clinical decisions to prevent, diagnose and treat neonatal morbidities.
