RESUMO
BACKGROUND: COVID-19 is caused by the SARS-CoV-2 virus and has strikingly heterogeneous clinical manifestations, with most individuals contracting mild disease but a substantial minority experiencing fulminant cardiopulmonary symptoms or death. The clinical covariates and the laboratory tests performed on a patient provide robust statistics to guide clinical treatment. Deep learning approaches on a data set of this nature enable patient stratification and provide methods to guide clinical treatment. OBJECTIVE: Here, we report on the development and prospective validation of a state-of-the-art machine learning model to provide mortality prediction shortly after confirmation of SARS-CoV-2 infection in the Mayo Clinic patient population. METHODS: We retrospectively constructed one of the largest reported and most geographically diverse laboratory information system and electronic health record of COVID-19 data sets in the published literature, which included 11,807 patients residing in 41 states of the United States of America and treated at medical sites across 5 states in 3 time zones. Traditional machine learning models were evaluated independently as well as in a stacked learner approach by using AutoGluon, and various recurrent neural network architectures were considered. The traditional machine learning models were implemented using the AutoGluon-Tabular framework, whereas the recurrent neural networks utilized the TensorFlow Keras framework. We trained these models to operate solely using routine laboratory measurements and clinical covariates available within 72 hours of a patient's first positive COVID-19 nucleic acid test result. RESULTS: The GRU-D recurrent neural network achieved peak cross-validation performance with 0.938 (SE 0.004) as the area under the receiver operating characteristic (AUROC) curve. This model retained strong performance by reducing the follow-up time to 12 hours (0.916 [SE 0.005] AUROC), and the leave-one-out feature importance analysis indicated that the most independently valuable features were age, Charlson comorbidity index, minimum oxygen saturation, fibrinogen level, and serum iron level. In the prospective testing cohort, this model provided an AUROC of 0.901 and a statistically significant difference in survival (P<.001, hazard ratio for those predicted to survive, 95% CI 0.043-0.106). CONCLUSIONS: Our deep learning approach using GRU-D provides an alert system to flag mortality for COVID-19-positive patients by using clinical covariates and laboratory values within a 72-hour window after the first positive nucleic acid test result.
Assuntos
COVID-19 , Sistemas de Informação em Laboratório Clínico , Aprendizado Profundo , Algoritmos , Registros Eletrônicos de Saúde , Humanos , Estudos Retrospectivos , SARS-CoV-2RESUMO
To discover lipidomic alterations during pregnancy in mothers who subsequently delivered small for gestational age (SGA) neonates and identify predictive lipid markers that can help recognize and manage these mothers, we carried out untargeted lipidomics on maternal serum samples collected between 24-28 weeks of gestation. We used a nested case-control study design and serum from mothers who delivered SGA and appropriate for gestational age babies. We applied untargeted lipidomics using mass spectrometry to characterize lipids and discover changes associated with SGA births during pregnancy. Multivariate pattern recognition software Collaborative Laboratory Integrated Reports (CLIR) was used for the post-analytical recognition of range differences in lipid ratios that could differentiate between SGA and control mothers and their integration for complete separation between the two groups. Here, we report changes in lipids from serum collected during pregnancy in mothers who delivered SGA neonates. In contrast to normal pregnancies where lysophosphatidic acid increased over the course of the pregnancy owing to increased activity of lysophospholipase D, we observed a decrease (32%; P = 0.05) of 20:4-lysophosphatidic acid in SGA mothers, which could potentially compromise fetal growth and development. Integration of lipid ratios in an interpretive tool (CLIR) could completely separate SGA mothers from controls demonstrating the power of untargeted lipidomic analyses for identifying novel predictive biomarkers. Additional studies are required for further assessment of the lipid biomarkers identified in this report.
Assuntos
Recém-Nascido Pequeno para a Idade Gestacional , Lipidômica , Estudos de Casos e Controles , Feminino , Idade Gestacional , Humanos , Lactente , Recém-Nascido , Lisofosfolipídeos , GravidezRESUMO
Newborn screening for congenital hypothyroidism remains challenging decades after broad implementation worldwide. Testing protocols are not uniform in terms of targets (TSH and/or T4) and protocols (parallel vs. sequential testing; one or two specimen collection times), and specificity (with or without collection of a second specimen) is overall poor. The purpose of this retrospective study is to investigate the potential impact of multivariate pattern recognition software (CLIR) to improve the post-analytical interpretation of screening results. Seven programs contributed reference data (N = 1,970,536) and two sets of true (TP, N = 1369 combined) and false (FP, N = 15,201) positive cases for validation and verification purposes, respectively. Data were adjusted for age at collection, birth weight, and location using polynomial regression models of the fifth degree to create three-dimensional regression surfaces. Customized Single Condition Tools and Dual Scatter Plots were created using CLIR to optimize the differential diagnosis between TP and FP cases in the validation set. Verification testing correctly identified 446/454 (98%) of the TP cases, and could have prevented 1931/5447 (35%) of the FP cases, with variable impact among locations (range 4% to 50%). CLIR tools either as made here or preferably standardized to the recommended uniform screening panel could improve performance of newborn screening for congenital hypothyroidism.
RESUMO
The expansion of the recommend uniform screening panel to include more than 50 primary and secondary target conditions has resulted in a substantial increase of false positive results. As an alternative to subjective manipulation of cutoff values and overutilization of molecular testing, here we describe the performance outcome of an algorithm for disorders of methionine, cobalamin, and propionate metabolism that includes: (1) first tier screening inclusive of the broadest available spectrum of markers measured by tandem mass spectrometry; (2) integration of all results into a score of likelihood of disease for each target condition calculated by post-analytical interpretive tools created byCollaborative Laboratory Integrated Reports (CLIR), a multivariate pattern recognition software; and (3) further evaluation of abnormal scores by a second tier test measuring homocysteine, methylmalonic acid, and methylcitric acid. This approach can consistently reduce false positive rates to a <0.01% level, which is the threshold of precision newborn screening. We postulate that broader adoption of this algorithm could lead to substantial savings in health care expenditures. More importantly, it could prevent the stress and anxiety experienced by many families when faced with an abnormal newborn screening result that is later resolved as a false positive outcome.