Technologies for integrated point-of-care testing and ai-powered adverse pregnancy outcome prediction

Abstract

Technologies for predicting health risk for a pregnant person include a computing device configured to collect a training data set that includes predetermined categories of demographic information and health data associated with a training population of pregnant persons. The computing device trains a predictive model to predict a risk of adverse pregnancy outcome with the training data set. A computing device is configured to receive feature data associated with a patient that includes predetermined categories of demographic information and health information associated with the patient, generate a predicted risk of adverse pregnancy outcome by inputting the feature data to the trained predictive model, and determine a predicted risk score. The computing device may generate a recommendation as a function of the predicted risk score and provide the recommendation to the patient. Other embodiments are described and claimed.

Description

Docket No. 85007-428789TECHNOLOGIES FOR INTEGRATED POINT-OF-CARE TESTING AND AI- POWERED ADVERSE PREGNANCY OUTCOME PREDICTIONCROSS-REFERENCE TO RELATED APPLICATION

[0001] This application claims the benefit of and priority to U.S. Patent Application No. 63 / 681,671, entitled “INTEGRATED POINT-OF-CARE TESTING AND AI-POWERED PREECLAMPSIA PREDICTION,” which was filed on August 9, 2024, and which is incorporated herein by reference in its entirety.BACKGROUND

[0002] Preeclampsia and other adverse pregnancy outcomes such as preterm delivery, intrauterine growth restriction (IUGR), and fetal demise remain significant contributors to maternal and perinatal morbidity and mortality worldwide. These conditions are often interrelated and can present with overlapping clinical features, making early and accurate predictions a persistent challenge in obstetric care. Especially, preeclampsia is a serious and complex condition affecting pregnant women, characterized by its unpredictability and significant health risks to both the mother and the baby. Currently, the only definitive cure for preeclampsia is the delivery of the placenta. Given these risks, early prediction of preeclampsia and other complications is crucial in mitigating health threats. However, existing prediction approaches primarily rely on periodic clinical assessments, maternal history, and advanced biomarker assays, which often require specialized laboratory infrastructure and expertise.

[0003] For example, existing prediction approaches include the sFIt-l / PIGF ratio test. This is a biochemical marker test that measures the ratio of soluble fms-like tyrosine kinase-1 (sFlt-1) to placental growth factor (P1GF). High levels of sFlt-1 and low levels of P1GF are indicative of preeclampsia. This test is used clinically to help predict the short-term absence of preeclampsia in women with suspected cases. Recently, the U.S. FDA endorsed this ratio as a tool to evaluate the risk of severe preeclampsia in pregnant women experiencing hypertensive disorders. This test demonstrates an excellent negative predictive value (NPV) exceeding 95%, but its positive predictive value (PPV) is relatively modest, ranging from 30% to 60%. Consequently, it is primarily used to rule out the condition. Additionally, there are specific clinical guidelines that must be adhered to when using this biomarker to predict preeclampsia.Docket No. 85007-428789

[0004] Another approach includes the FullPIERS model. Developed for use in a hospital setting, this predictive model utilizes variables such as maternal symptoms, signs, and laboratory tests to estimate the risk of adverse maternal outcomes in women with preeclampsia. This model requires clinical input and is used to guide treatment decisions.

[0005] Automated preeclampsia risk calculators integrate electronic health records with algorithm-based tools to assess risk levels based on individual patient data. The effectiveness of those tools varies depending on the integration with hospital IT systems and the specific algorithms used. Additionally, some calculators incorporate additional biomarkers such as sFlt-1, P1GF, PAPP-A, soluble Endoglin, and measurements from ultrasound Doppler to enhance their predictive accuracy.

[0006] Certain home monitoring kits are designed to enable patients to monitor their symptoms from home. These kits, often including blood pressure cuffs and urine protein tests, allow for convenient tracking of preeclampsia-related symptoms. However, while these tools facilitate at-home monitoring and can help in diagnosing preeclampsia, they generally require professional interpretation and follow-up. These home monitoring tools do not possess predictive capabilities; that is, they cannot forecast the onset of preeclampsia before symptoms actually appear.SUMMARY

[0007] According to one aspect of the disclosure, a computing device for predicting health risk for a pregnant person includes a trained predictive model, a feature data manager, and a risk monitor. The feature data manager is configured to receive feature data associated with a patient, wherein the feature data comprises predetermined categories of demographic information and predetermined categories of health data associated with the patient. The risk monitor is configured to generate a predicted risk of adverse pregnancy outcome by an input of the feature data to the trained predictive model, and to determine a predicted risk score based on the predicted risk of adverse pregnancy outcome. In an embodiment, the adverse pregnancy outcome comprises preeclampsia, other hypertensive disorders of pregnancy, preterm delivery, intrauterine growth restriction, or stillbirth.

[0008] In an embodiment, the predetermined categories of demographic information comprises maternal characteristics. In an embodiment, the predetermined categories ofDocket No. 85007-428789 demographic information comprises age, race, pregravid body mass index (BMI), smoking status, alcohol use, nulliparity, gestational age at delivery, gravidity, number of fetuses, baby weight, delivery method, medication use, and medical history. In an embodiment, the predetermined categories of health data comprises hematology test results. In an embodiment, the predetermined categories of health data comprises chemistry test results. In an embodiment, the predetermined categories of health data comprises urinalysis results. In an embodiment, the predetermined categories of health data comprises routine blood test results. In an embodiment, the predetermined categories of health data comprises urine protein analysis test results.

[0009] In an embodiment, the feature data associated with the patient comprises age, race, pregravid body mass index (BMI), smoking status, alcohol use, nulliparity, gestational age at delivery, gravidity, number of fetuses, baby weight, delivery method, medication (aspirin, antihypertensive medications, insulin, heparin) use, history of chronic hypertension, chronic kidney diseases, diabetes mellitus, auto-immune diseases or stroke, blood pressure and laboratory tests (Hematology tests: white blood cell count (WBC), red blood cell count (RBC), platelet count, hemoglobin, hematocrit, mean cell volume (MCV), mean cell hemoglobin (MCH), mean cell hemoglobin concentration (MCHC), red cell distribution width (RDW), mean platelet volume (MPV), nucleated RBC (NRBC), neutrophil count, lymphocyte count, monocyte count, eosinophil count, basophil count; chemistry test results: aspartate aminotransferase (AST), alanine aminotransferase (ALT), creatinine level, blood urea nitrogen (BUN), potassium level, alkaline phosphatase (ALP), sodium level, total protein level, albumin level, carbon dioxide, anion gap, globulin, calcium level, bilimbin total, bilimbin direct, estimated glomerular filtration rate (eGFR), glucose, and chloride level; Urinalysis: urine color, urine pH, urine nitrite, urine bilimbin, urine ketones, urine appearance, urine glucose, urine blood, urine protein, urine creatinine, urine gravidity, urine leukocyte esterase).

[0010] In an embodiment, the predetermined categories of health data comprises lab test results. In an embodiment, the predetermined categories of health data comprises sFlt-1 test results, Pl GF test results, sFIt-l / PIGF ratio test results, or other biomarkers associated with the patient. In an embodiment, the predetermined categories of health data comprises blood pressure data.

[0011] In an embodiment, the computing device further includes an input device configured to receive the predetermined categories of demographic information.Docket No. 85007-428789

[0012] In an embodiment, the computing device further includes a sample analyzer configured to analyze a blood sample collected from the patient to generate blood sample test results, wherein the predetermined categories of health data comprises the blood sample test results. In an embodiment, the blood sample is collected by venipuncture or capillary blood collection. In an embodiment, the computing device further includes a sample analyzer to analyze a urine sample collected from the patient to generate urine sample test results, wherein the predetermined categories of health data comprises the urine sample test results.

[0013] In an embodiment, the trained predictive model comprises a random forest model, a deep learning model, a decision tree model, a logistic regression model, or a gradient-boosted tree model. In an embodiment, the trained predictive model comprises a plurality of trained predictive models, and wherein to generate a predicted risk of adverse pregnancy outcome comprises to perform multi-stage cascading inference with the plurality of trained predictive models. In an embodiment, the risk monitor is further configured to preprocess the feature data; wherein to input the feature data comprises to input the feature data in response to preprocessing of the feature data. In an embodiment, to preprocess the feature data comprises to encode categorical variables, to perform missing value imputation, to remove outliers, or to perform feature value normalization.

[0014] In an embodiment, the computing device further includes a recommendation engine configured to generate a recommendation as a function of the predicted risk score based on a one or more predetermined recommendation rules; and provide the recommendation to the patient. In an embodiment, the recommendation is indicative of frequency or location of maternal and fetal monitoring from routine monitoring to emergency monitoring. In an embodiment, the recommendation is indicative of a recommended medication or a recommendation to start labor.

[0015] In an embodiment, the computing device comprises a point of care testing (POCT) device. In an embodiment, the computing device comprises a healthcare facility information technology system. In an embodiment, the computing device comprises a cloud-based system.

[0016] According to another aspect, a computing device for pregnancy health risk model training includes a predictive model, a data manager, and a model trainer. The data manager is configured to collect a first training data set. The first training data set includes predetermined categories of demographic information and predetermined categories of health data associated with a training population of pregnant persons. The model trainer is configured to train the predictiveDocket No. 85007-428789 model to predict a risk of adverse pregnancy outcome with the first training data set. In an embodiment, the adverse pregnancy outcome comprises preeclampsia, other hypertensive disorders of pregnancy, preterm delivery, intrauterine growth restriction, or stillbirth.

[0017] In an embodiment, the predetermined categories of demographic information comprises maternal characteristics. In an embodiment, the predetermined categories of demographic information comprises age, race, first-trimester body mass index (BMI), smoking status, nulliparity, gestational age, and medical history. In an embodiment, the predetermined categories of health data comprises hematology test results. In an embodiment, the predetermined categories of health data comprises chemistry test results. In an embodiment, the predetermined categories of health data comprises urinalysis chemistry test results. In an embodiment, the predetermined categories of health data comprises routine blood test results. In an embodiment, the predetermined categories of health data comprises urine protein analysis test results.

[0018] In an embodiment, the first training data set comprises age, race, pregravid body mass index (BMI), smoking status, alcohol use, nulliparity, gestational age at delivery, gravidity, number of fetuses, baby weight, delivery method, medication (aspirin, anti-hypertensive medications, insulin, heparin) use, history of chronic hypertension, chronic kidney diseases, diabetes mellitus, auto-immune diseases or stroke, blood pressure and laboratory tests (Hematology tests: white blood cell count (WBC), red blood cell count (RBC), platelet count, hemoglobin, hematocrit, mean cell volume (MCV), mean cell hemoglobin (MCH), mean cell hemoglobin concentration (MCHC), red cell distribution width (RDW), mean platelet volume (MPV), nucleated RBC (NRBC), neutrophil count, lymphocyte count, monocyte count, eosinophil count, basophil count; Chemistry test results: aspartate aminotransferase (AST), alanine aminotransferase (ALT), creatinine level, blood urea nitrogen (BUN), potassium level, alkaline phosphatase (ALP), sodium level, total protein level, albumin level, carbon dioxide, anion gap, globulin, calcium level, bilirubin total, bilirubin direct, estimated glomerular filtration rate (eGFR), glucose, and chloride level; Urinalysis: urine color, urine pH, urine nitrite, urine bilirubin, urine ketones, urine appearance, urine glucose, urine blood, urine protein, urine creatinine, urine gravidity, urine leukocyte esterase).

[0019] In an embodiment, the predetermined categories of health data comprises lab test results. In an embodiment, the predetermined categories of health data comprises sFlt-1 test results, Pl GF test results, sFIt-l / PIGF ratio test results, or other biomarkers associated with theDocket No. 85007-428789 patient. In an embodiment, the predetermined categories of health data comprises blood pressure data.

[0020] In an embodiment, the trained predictive model comprises a random forest model, a deep learning model, a decision tree model, a logistic regression model, or a gradient-boosted tree model. In an embodiment, the data manager is further configured to preprocess the first training data; and to train the predictive model comprises to train the predictive model in response to preprocessing of the first training data. In an embodiment, to preprocess the first training data comprises to encode categorical variables, to perform missing value imputation, to remove outliers, or to perform feature value normalization. In an embodiment, to preprocess the first training data comprises to enrich the first training data with a statistical feature or a temporal feature based on the demographic information or the health data. In an embodiment, to train the predictive model comprises to perform nested cross-validation with a plurality of patient-level folds of the first training data. In an embodiment, to train the predictive model comprises to train a plurality of predictive models for multi-stage cascading inference.

[0021] In an embodiment, the computing device further includes a model analyzer configured to identify a most predictive feature of the first training data in response to training of the predictive model. In an embodiment, the computing device further includes a model analyzer configured to evaluate performance of the predictive model in response to training of the predictive model.

[0022] According to another aspect, a method for predicting health risk for a pregnant person includes receiving, by a computing device, feature data associated with a patient, wherein the feature data comprises predetermined categories of demographic information and predetermined categories of health data associated with the patient; inputting, by the computing device, the feature data to a trained predictive model to generate a predicted risk of adverse pregnancy outcome; and determining, by the computing device, a predicted risk score based on the predicted risk of adverse pregnancy outcome. In an embodiment, the adverse pregnancy outcome comprises preeclampsia, other hypertensive disorders of pregnancy, preterm delivery, intrauterine growth restriction, or stillbirth.

[0023] In an embodiment, the predetermined categories of demographic information comprises maternal characteristics. In an embodiment, the predetermined categories of demographic information comprises age, race, pregravid body mass index (BMI), smoking status,Docket No. 85007-428789 alcohol use, nulliparity, gestational age at delivery, gravidity, number of fetuses, baby weight, delivery method, medication use, and medical history. In an embodiment, the predetermined categories of health data comprises hematology test results. In an embodiment, the predetermined categories of health data comprises chemistry test results. In an embodiment, the predetermined categories of health data comprises urinalysis results. In an embodiment, the predetermined categories of health data comprises routine blood test results. In an embodiment, the predetermined categories of health data comprises urine protein analysis test results.

[0024] In an embodiment, the feature data associated with the patient comprises age, race, pregravid body mass index (BMI), smoking status, alcohol use, nulliparity, gestational age at delivery, gravidity, number of fetuses, baby weight, delivery method, medication (aspirin, antihypertensive medications, insulin, heparin) use, history of chronic hypertension, chronic kidney diseases, diabetes mellitus, auto-immune diseases or stroke, blood pressure and laboratory tests (Hematology tests: white blood cell count (WBC), red blood cell count (RBC), platelet count, hemoglobin, hematocrit, mean cell volume (MCV), mean cell hemoglobin (MCH), mean cell hemoglobin concentration (MCHC), red cell distribution width (RDW), mean platelet volume (MPV), nucleated RBC (NRBC), neutrophil count, lymphocyte count, monocyte count, eosinophil count, basophil count; Chemistry test results: aspartate aminotransferase (AST), alanine aminotransferase (ALT), creatinine level, blood urea nitrogen (BUN), potassium level, alkaline phosphatase (ALP), sodium level, total protein level, albumin level, carbon dioxide, anion gap, globulin, calcium level, bilirubin total, bilirubin direct, estimated glomerular filtration rate (eGFR), glucose, and chloride level; Urinalysis: urine color, urine pH, urine nitrite, urine bilirubin, urine ketones, urine appearance, urine glucose, urine blood, urine protein, urine creatinine, urine gravidity, urine leukocyte esterase).

[0025] In an embodiment, the predetermined categories of health data comprises lab test results. In an embodiment, the predetermined categories of health data comprises sFlt-1 test results, Pl GF test results, sFIt-l / PIGF ratio test results, or other biomarkers associated with the patient. In an embodiment, the predetermined categories of health data comprises blood pressure data.

[0026] In an embodiment, the method further includes receiving, by the computing device, the predetermined categories of demographic information with an input device of the computing device.Docket No. 85007-428789

[0027] In an embodiment, the method further includes collecting a blood sample from the patient; and analyzing the blood sample to generate blood sample test results, wherein the predetermined categories of health data comprises the blood sample test results. In an embodiment, wherein collecting the blood sample comprises collecting the blood sample by venipuncture or capillary blood collection. In an embodiment, the method further includes collecting a urine sample from the patient; and analyzing the urine sample to generate urine sample test results, wherein the predetermined categories of health data comprises the urine sample test results.

[0028] In an embodiment, the trained predictive model comprises a random forest model, a deep learning model, a decision tree model, a logistic regression model, or a gradient-boosted tree model. In an embodiment, generating the predicted risk of adverse pregnancy outcome comprises performing multi-stage cascading inference with a plurality of trained predictive models, wherein the trained predictive model comprises the plurality of trained predictive models. In an embodiment, the method further includes preprocessing, by the computing device, the feature data; wherein inputting the feature data comprises inputting the feature data in response to preprocessing the feature data. In an embodiment, preprocessing the feature data comprises encoding categorical variables, performing missing value imputation, removing outliers, or performing feature value normalization.

[0029] In an embodiment, further the method further includes generating, by the computing device, a recommendation as a function of the predicted risk score based on a one or more predetermined recommendation rules; and providing, by the computing device, the recommendation to the patient. In an embodiment, the recommendation is indicative of frequency or location of maternal and fetal monitoring from routine monitoring to emergency monitoring. In an embodiment, the recommendation is indicative of a recommended medication or a recommendation to start labor.

[0030] In an embodiment, the computing device comprises a point of care testing (POCT) device. In an embodiment, the computing device comprises a healthcare facility information technology system. In an embodiment, the computing device comprises a cloud-based system.

[0031] According to another aspect, a method for pregnancy health risk model training includes collecting, by a computing device, a first training data set, wherein the first training data set comprises predetermined categories of demographic information and predetermined categoriesDocket No. 85007-428789 of health data associated with a training population of pregnant persons; and training, by the computing device, a predictive model to predict a risk of adverse pregnancy outcome using the first training data set. In an embodiment, the adverse pregnancy outcome comprises preeclampsia, other hypertensive disorders of pregnancy, preterm delivery, intrauterine growth restriction, or stillbirth.

[0032] In an embodiment, the predetermined categories of demographic information comprises maternal characteristics. In an embodiment, the predetermined categories of demographic information comprises age, race, pregravid body mass index (BMI), smoking status, alcohol use, nulliparity, gestational age at delivery, gravidity, number of fetuses, baby weight, delivery method, medication use, and medical history. In an embodiment, the predetermined categories of health data comprises hematology test results. In an embodiment, the predetermined categories of health data comprises chemistry test results. In an embodiment, the predetermined categories of health data comprises urinalysis results. In an embodiment, the predetermined categories of health data comprises routine blood test results. In an embodiment, the predetermined categories of health data comprises urine protein analysis test results.

[0033] In an embodiment, the first training data set comprises age, race, pregravid body mass index (BMI), smoking status, alcohol use, nulliparity, gestational age at delivery, gravidity, number of fetuses, baby weight, delivery method, medication (aspirin, anti-hypertensive medications, insulin, heparin) use, history of chronic hypertension, chronic kidney diseases, diabetes mellitus, auto-immune diseases or stroke, blood pressure and laboratory tests (Hematology tests: white blood cell count (WBC), red blood cell count (RBC), platelet count, hemoglobin, hematocrit, mean cell volume (MCV), mean cell hemoglobin (MCH), mean cell hemoglobin concentration (MCHC), red cell distribution width (RDW), mean platelet volume (MPV), nucleated RBC (NRBC), neutrophil count, lymphocyte count, monocyte count, eosinophil count, basophil count; Chemistry test results: aspartate aminotransferase (AST), alanine aminotransferase (ALT), creatinine level, blood urea nitrogen (BUN), potassium level, alkaline phosphatase (ALP), sodium level, total protein level, albumin level, carbon dioxide, anion gap, globulin, calcium level, bilirubin total, bilirubin direct, estimated glomerular filtration rate (eGFR), glucose, and chloride level; Urinalysis: urine color, urine pH, urine nitrite, urine bilirubin, urine ketones, urine appearance, urine glucose, urine blood, urine protein, urine creatinine, urine gravidity, urine leukocyte esterase).Docket No. 85007-428789

[0034] In an embodiment, the predetermined categories of health data comprises lab test results. In an embodiment, the predetermined categories of health data comprises sFlt-1 test results, Pl GF test results, sFIt-l / PIGF ratio test results, or other biomarkers associated with the patient. In an embodiment, the predetermined categories of health data comprises blood pressure data.

[0035] In an embodiment, the trained predictive model comprises a random forest model, a deep learning model, a decision tree model, a logistic regression model, or a gradient-boosted tree model. In an embodiment, the method further includes preprocessing, by the computing device, the first training data; wherein training the predictive model comprises training the predictive model in response to preprocessing the first training data. In an embodiment, preprocessing the first training data comprises encoding categorical variables, performing missing value imputation, removing outliers, or performing feature value normalization. In an embodiment, preprocessing the first training data comprises enriching the first training data with a statistical feature or a temporal feature based on the demographic information or the health data. In an embodiment, training the predictive model comprises performing nested cross-validation with a plurality of patient-level folds of the first training data. In an embodiment, training the predictive model comprises training a plurality of predictive models for multi-stage cascading inference.

[0036] In an embodiment, the method further includes identifying, by the computing device, a most predictive feature of the first training data in response to training the predictive model. In an embodiment, the method further includes evaluating, by the computing device, performance of the predictive model in response to training the predictive model.BRIEF DESCRIPTION OF THE DRAWINGS

[0037] The concepts described herein are illustrated by way of example and not by way of limitation in the accompanying figures. For simplicity and clarity of illustration, elements illustrated in the figures are not necessarily drawn to scale. Where considered appropriate, reference labels have been repeated among the figures to indicate corresponding or analogous elements.

[0038] FIG. 1 is a simplified block diagram of at least one embodiment of a system for integrated point-of-care testing and Al-powered adverse pregnancy outcome prediction;Docket No. 85007-428789

[0039] FIG. 2 is a simplified block diagram of at least one embodiment of an environment that may be established by a computing device of the system of FIG. 1 ;

[0040] FIG. 3 is a simplified flow diagram of at least one embodiment of a method for training a machine learning model for adverse pregnancy outcome prediction that may be executed by the system of FIGS. 1 and 2;

[0041] FIG. 4 is a simplified flow diagram of at least one embodiment of a method for adverse pregnancy outcome including integrated point-of-care testing in some embodiments that may be executed by the system of FIGS. 1 and 2; and

[0042] FIG. 5 is a schematic diagram illustrating at least one potential embodiment of an integrated point-of-care instrument that may be used with the method of FIG. 4.DETAILED DESCRIPTION OF THE DRAWINGS

[0043] While the concepts of the present disclosure are susceptible to various modifications and alternative forms, specific embodiments thereof have been shown by way of example in the drawings and will be described herein in detail. It should be understood, however, that there is no intent to limit the concepts of the present disclosure to the particular forms disclosed, but on the contrary, the intention is to cover all modifications, equivalents, and alternatives consistent with the present disclosure and the appended claims.

[0044] References in the specification to “one embodiment,” “an embodiment,” “an illustrative embodiment,” etc., indicate that the embodiment described may include a particular feature, structure, or characteristic, but every embodiment may or may not necessarily include that particular feature, structure, or characteristic. Moreover, such phrases are not necessarily referring to the same embodiment. Further, when a particular feature, structure, or characteristic is described in connection with an embodiment, it is submitted that it is within the knowledge of one skilled in the art to effect such feature, structure, or characteristic in connection with other embodiments whether or not explicitly described. Additionally, it should be appreciated that items included in a list in the form of “at least one A, B, and C” can mean (A); (B); (C); (A and B); (A and C); (B and C); or (A, B, and C). Similarly, items listed in the form of “at least one of A, B, or C” can mean (A); (B); (C); (A and B); (A and C); (B and C); or (A, B, and C).

[0045] The disclosed embodiments may be implemented, in some cases, in hardware, firmware, software, or any combination thereof. The disclosed embodiments may also beDocket No. 85007-428789 implemented as instructions carried by or stored on a transitory or non-transitory machine-readable (e g., computer-readable) storage medium, which may be read and executed by one or more processors. A machine-readable storage medium may be embodied as any storage device, mechanism, or other physical structure for storing or transmitting information in a form readable by a machine (e.g., a volatile or non-volatile memory, a media disc, or other media device).

[0046] In the drawings, some structural or method features may be shown in specific arrangements and / or orderings. However, it should be appreciated that such specific arrangements and / or orderings may not be required. Rather, in some embodiments, such features may be arranged in a different manner and / or order than shown in the illustrative figures. Additionally, the inclusion of a structural or method feature in a particular figure is not meant to imply that such feature is required in all embodiments and, in some embodiments, may not be included or may be combined with other features.

[0047] Referring now to FIG. 1, an illustrative system 100 for integrated point-of-care testing with adverse pregnancy outcome prediction includes a computing device 102 which may be coupled to an analytical device 104. In use, the computing device 102 collects feature data for a patient including demographic data and health data. As described further below, the health data includes routine clinical data including routing clinical laboratory tests, which are commonly performed during prenatal care. In some embodiments, some or all of the health data may be generated or otherwise received from the analytical device 104, which may be a point of care instrument, a laboratory instrument, or other analytical device as described below. After being generated and in some embodiments pre-processed, the feature data is provided to a pre-trained predictive model, which predicts a risk of adverse pregnancy outcome such as preeclampsia, other hypertensive disorders of pregnancy, preterm delivery, intrauterine growth restriction (IUGR), and / or fetal demise. The predictive model may be used in both acute clinical and routine prenatal care settings. Based on the predicted risk, the computing device 102 generates one or more recommendations which are presented to the patient and / or clinician. Accordingly, the system 100 may predict risk of developing preeclampsia or other complications in pregnant women before clinical symptoms manifest.

[0048] Thus, the system 100 provides an Al -based, data-driven predictive model that integrates easily obtainable patient demographic information, longitudinal blood pressure measurements, and routine laboratory test results to provide early and accurate indications of riskDocket No. 85007-428789 and recommendations for response. Accordingly, the system 100 may improve accessibility compared to existing systems, particularly in low-resource or community healthcare settings, leading to increased opportunities for early risk stratification and timely intervention. Additionally, unlike conventional tools, the disclosed predictive model does not depend on specialized biomarkers or sophisticated equipment, enabling broad applicability and consistent performance across diverse healthcare environments. By accurately identifying pregnant persons at high risk for developing preeclampsia and other complications before clinical onset, the disclosed technologies support earlier clinical decisions, closer monitoring, and targeted preventive strategies, ultimately improving maternal and fetal outcomes and reducing healthcare costs.

[0049] Additionally, the technologies disclosed herein can be utilized in various healthcare settings, including hospitals, outpatient clinics, community health centers, and even remote or resource-limited areas. The disclosed technologies can be incorporated into existing electronic health record (EHR) systems (e.g., Epic), where risk scores are automatically updated as new patient data becomes available. The disclosed technologies can be integrated into POCT devices or mobile health applications, allowing for broader accessibility and enabling patients to engage in self-monitoring under medical guidance. In home-based settings, the disclosed technologies can be paired with advanced blood pressure devices to provide real-time risk updates, empowering patients to seek timely medical attention when necessary.

[0050] The computing device 102 may be embodied as any type of device capable of performing the functions described herein. For example, the computing device 102 may be embodied as, without limitation, a workstation, a desktop computer, a laptop computer, a server, a rack-mounted server, a blade server, a network appliance, a web appliance, a tablet computer, a smartphone, a consumer electronic device, a distributed computing system, a multiprocessor system, and / or any other computing device capable of performing the functions described herein. Additionally, in some embodiments, the computing device 102 may be embodied as a “virtual server” formed from multiple computing devices distributed across a network and operating in a public or private cloud. Accordingly, although the computing device 102 is illustrated in FIG. 1 as embodied as a single computing device, it should be appreciated that the computing device 102 may be embodied as multiple devices cooperating together to facilitate the functionality describedDocket No. 85007-428789 below. The system 100 thus may implement a scalable architecture that can handle increases in user numbers and data volume without degradation in performance.

[0051] As shown in FIG. 1, the illustrative computing device 102 includes a processor 120, an I / O subsystem 122, memory 124, a data storage device 126, and communication circuitry 128. Of course, the computing device 102 may include other or additional components, such as those commonly found in a workstation computer (e.g., various input / output devices), in other embodiments. Additionally, in some embodiments, one or more of the illustrative components may be incorporated in, or otherwise form a portion of, another component. For example, the memory 124, or portions thereof, may be incorporated in the processor 120 in some embodiments.

[0052] The processor 120 may be embodied as any type of processor or compute engine capable of performing the functions described herein. For example, the processor may be embodied as a single or multi-core processor(s), digital signal processor, microcontroller, or other processor or processing / controlling circuit. Similarly, the memory 124 may be embodied as any type of volatile or non-volatile memory or data storage capable of performing the functions described herein. In operation, the memory 124 may store various data and software used during operation of the computing device 102 such as operating systems, applications, programs, libraries, and drivers. The memory 124 is communicatively coupled to the processor 120 via the VO subsystem 122, which may be embodied as circuitry and / or components to facilitate input / output operations with the processor 120, the memory 124, and other components of the computing device 102. For example, the I / O subsystem 122 may be embodied as, or otherwise include, memory controller hubs, input / output control hubs, firmware devices, communication links (i.e., point-to- point links, bus links, wires, cables, light guides, printed circuit board traces, etc.) and / or other components and subsystems to facilitate the input / output operations. In some embodiments, the I / O subsystem 122 may form a portion of a system-on-a-chip (SoC) and be incorporated, along with the processor 120, the memory 124, and other components of the computing device 102, on a single integrated circuit chip.

[0053] The data storage device 126 may be embodied as any type of device or devices configured for short-term or long-term storage of data such as, for example, memory devices and circuits, memory cards, hard disk drives, solid-state drives, or other data storage devices. The communication circuitry 128 of the computing device 102 may be embodied as any communication circuit, device, or collection thereof, capable of enabling communications betweenDocket No. 85007-428789 the computing device 102 and remote devices. The communication circuitry 128 may be configured to use any one or more communication technology (e.g., wireless or wired communications) and associated protocols (e.g., Ethernet, Bluetooth®, Wi-Fi®, WiMAX, etc.) to effect such communication.

[0054] As shown in FIG. 1, the computing device 102 may include a display 130. The display 130 may be embodied as any type of display capable of displaying digital images, text, or other information, such as a liquid crystal display (LCD), a light emitting diode (LED), a plasma display, a cathode ray tube (CRT), or other type of display device. In some embodiments, the display 130 may be coupled to a touch screen to allow user interaction with the computing device 102.

[0055] The medical analytical device 104 may be embodied as any device capable of performing analytical tests on patient samples, such as blood samples, urine samples, or other samples. As described further below, the analytical device 104 may be designed to perform hematology testing, chemistry testing, or other tests corresponding to significant features associated with the predictive model executed by the computing device 102. The illustrative analytical device 104 includes one or more measurement modules 142 and a sample handler 144. The measurement module 142 may perform hematology testing using one or more analytical technologies such as electrical impedance measurement, photometry, or other measurements. Additionally or alternatively, the measurement module 142 may perform chemistry testing including one or more analytic technologies such as ion-selective electrodes, photometry, or enzymatic reactions. Accordingly, in some embodiments the analytical device 104 may also include a reagent management module including storage and fluid management systems for enzymes, substrates, wash fluids, waste, and / or other reagents or materials. The sample handler 144 is configured to receive and process one or more patient samples, such as blood samples (e.g., whole blood, serum, venipuncture or capillary blood sample, etc.), urine samples, or other patient samples. The analytical device 104 may be embodied as a laboratory -based instrument, a point- of-care instrument, or other analytical instrument. Additionally, as described further below, in some embodiments, the computing device 102 may be in communication with multiple medical imaging devices 104, which may be geographically or otherwise distributed.

[0056] As shown in FIG. 1, in some embodiments, the computing device 102 and the medical analytical device 104 may be configured to transmit and receive data with each otherDocket No. 85007-428789 and / or other devices of the system 100 over a network 106. The network 106 may be embodied as any number of various wired and / or wireless networks. For example, the network 106 may be embodied as, or otherwise include, a wired or wireless local area network (LAN), a wired or wireless wide area network (WAN), and / or a publicly-accessible, global network such as the Internet. As such, the network 106 may include any number of additional devices, such as additional computers, routers, and switches, to facilitate communications among the devices of the system 100.

[0057] Although illustrated as a separate computing device 102 and analytical device 104, it should be understood that in some embodiments some or all of the functionality of the computing device 102 and the analytical device 104 may be included in a single, integrated device, such as a point-of-care analytical device 104 with an integrated computing device 102. Accordingly, in some embodiments the analytical device 104 may include components and devices commonly found in a workstation or similar computing device, such as a processor, an I / O subsystem, a memory, a data storage device, a communication subsystem, and / or a display or other VO devices. Those individual components of an integrated analytical device 104 may be similar to the corresponding components of the computing device 102, the description of which is applicable to the corresponding components of the analytical device 104 and is not repeated herein so as not to obscure the present disclosure.

[0058] Referring now to FIG. 2, in the illustrative embodiment, the computing device 102 establishes an environment 200 during operation. The illustrative environment 200 includes a training data manager 202, a model trainer 204, an adverse pregnancy outcome predictive model 206, a model analyzer 210, a sample analyzer 212, a feature data manager 214, a risk monitor 220, and a recommendation engine 222. The various components of the environment 200 may be embodied as hardware, firmware, software, or a combination thereof. As such, in some embodiments, one or more of the components of the environment 200 may be embodied as circuitry or a collection of electrical devices (e.g., training data manager circuitry 202, model trainer circuitry 204, adverse pregnancy outcome predictive model circuitry 206, model analyzer circuitry 210, sample analyzer circuitry 212, feature data manager circuitry 214, risk monitor circuitry 214, and / or recommendation engine circuitry 222). It should be appreciated that, in such embodiments, one or more of those components may form a portion of the processor 120, the I / O subsystem 122, and / or other components of the computing device 102.Docket No. 85007-428789

[0059] The training data manager 202 is configured to collect a training dataset 226, which includes predetermined categories of demographic information and predetermined categories of health data associated with a training population of pregnant persons. The predetermined categories of demographic information may include age, race, pregravid body mass index (BMI), smoking status, alcohol use, nulliparity, gestational age at delivery, gravidity, number of fetuses, baby weight, delivery method, medication use, and medical history. The predetermined categories of health data may include hematology test results, blood chemistry test results, urine chemistry test results, urine protein analysis test results, or blood pressure data. In some embodiments, the training dataset 226 includes age, race, pregravid BMI, smoking status, alcohol use, nulliparity, gestational age at delivery, gravidity, number of fetuses, baby weight, delivery method, medication (aspirin, anti-hypertensive medications, insulin, heparin) use, history of chronic hypertension, chronic kidney diseases, diabetes mellitus, auto-immune diseases or stroke, blood pressure and laboratory tests (Hematology tests: white blood cell count (WBC), red blood cell count (RBC), Ppatelet count, hemoglobin, hematocrit, mean cell volume (MCV), mean cell hemoglobin (MCH), mean cell hemoglobin concentration (MCHC), red cell distribution width (RDW), mean platelet volume (MPV), nucleated RBC (NRBC), neutrophil count, lymphocyte count, monocyte count, eosinophil count, basophil count; Chemistry test results: aspartate aminotransferase (AST), alanine aminotransferase (ALT), creatinine level, blood urea nitrogen (BUN), potassium level, alkaline phosphatase (ALP), sodium level, total protein level, albumin level, carbon dioxide, anion gap, globulin, calcium level, bilirubin total, bilirubin direct, estimated glomerular filtration rate (eGFR), glucose, and chloride level; Urinalysis: urine color, urine pH, urine nitrite, urine bilirubin, urine ketones, urine appearance, urine glucose, urine blood, urine protein, urine creatinine, urine gravidity, urine leukocyte esterase). In some embodiments, the predetermined categories of health data include lab test results, sFlt-1 test results, P1GF test results, sFIt-l / PIGF ratio test results, or other biomarkers associated with the patient.

[0060] In some embodiments, the training data manager 202 is further configured to preprocess the training dataset 226. Preprocessing training dataset 226 may include encoding categorical variables, performing missing value imputation, removing outliers, or performing feature value normalization. In some embodiments, preprocessing the training dataset 226 includes enriching the training dataset 226 with one or more statistical features or temporal features based on the demographic information or the health data.Docket No. 85007-428789

[0061] The model trainer 204 is configured to train the adverse pregnancy outcome predictive model 206 to predict a risk of adverse pregnancy outcome with the training dataset 226. The adverse pregnancy outcome predictive model 206 is a machine learning model, and is illustratively a binary classification model using gradient-boosted decision trees. In some embodiments, the predictive model 206 may be embodied as a random forest model, a deep learning model, a decision tree model, a logistic regression model, or a gradient-boosted tree model. In other embodiments, the model 206 may be embodied as any other deep learning model suitable for prediction tasks as described herein. Accordingly, the model 206 includes multiple weights 208, which are adjusted during model training. The adverse pregnancy outcome may include preeclampsia, hypertensive disorders of pregnancy, preterm delivery, IUGR, or stillbirth, to train the predictive model comprises to train the predictive model in response to preprocessing of the first training data. Training the predictive model 206 may include performing nested cross- validation with multiple patient-level folds of the training dataset 226. In some embodiments, training the predictive model 206 may include training multiple predictive models 206 for multistage cascading inference.

[0062] The model analyzer 210 is configured to identify one or more most predictive features of the training dataset 226 in response to training the predictive model 206. The model analyzer 210 may be configured to evaluate performance of the predictive model 206 in response training the predictive model 206.

[0063] The feature data manager 214 is configured to receive feature data 224 associated with a patient. The feature data 224 includes predetermined categories of demographic information and predetermined categories of health data associated with the patient. The predetermined categories of demographic information may include maternal characteristics, age, race, pregravid BMI, smoking status, alcohol use, nulliparity, gestational age at delivery, gravidity, number of fetuses, baby weight, delivery method, medication use, and medical history. In some embodiments, the computing device 102 further includes an input device such as a touchscreen, mouse, keyboard, or other input device configured to receive the predetermined categories of demographic information. The predetermined categories of health data may include hematology test results, blood chemistry test results, urine chemistry test results, urine protein analysis test results, or blood pressure data. In some embodiments, the feature data 224 associated with the patient may include blood pressure, ALP, pregravid BMI, gravidity, number of fetuses, age,Docket No. 85007-428789 globulin, creatinine level, carbon dioxide, albumin, total protein level, platelet count, race, nulliparity, RBC, MCV, MPV, RDW, MCH, MCHC, glucose, potassium level, WBC, hemoglobin, hematocrit, anion gap, BUN, AST, ALT and total bilirubin. In some embodiments, the predetermined categories of health data may include lab test results, sFlt-1 test results, P1GF test results, sFIt-l / PIGF ratio test results, or other biomarkers associated with the patient. In some embodiments, one or more of those operations may be performed by one or more sub-components, such as an input manager 216 or an electronic health record (EHR) system connector 218.

[0064] In some embodiments, the sample analyzer 212 is configured to analyze a blood sample collected from the patient to generate blood sample test results. In those embodiments, the predetermined categories of health data include the blood sample test results. The blood sample may be collected by venipuncture or capillary blood collection. In some embodiments, the sample analyzer 212 is configured to analyze a urine sample collected from the patient to generate urine sample test results. In those embodiments, the predetermined categories of health data include the urine sample test results. The sample analysis may be performed by the analytical device 104, which may be integrated with the computing device 102, for example as illustrated in FIG. 5 and described further below. Additionally or alternatively, the environment 200 may not include a sample analyzer 212.

[0065] The risk monitor 220 is configured generate a predicted risk of adverse pregnancy outcome by an input of the feature data 224 to the trained adverse pregnancy outcome predictive model 206. The predictive model 206 is trained as described above. The training may be performed by the computing device 102 and / or another device, and the trained predictive model 206 may be deployed to the computing device 102. The risk monitor 220 is further configured to determine a predicted risk score based on the predicted risk of adverse pregnancy outcome. The adverse pregnancy outcome may include preeclampsia, hypertensive disorders of pregnancy, preterm delivery, IUGR, or stillbirth. As described above, the trained predictive model 206 may include a random forest model, a deep learning model, a decision tree model, a logistic regression model, or a gradient-boosted tree model. In some embodiments, the trained predictive model 206 may include multiple trained predictive models 206. In those embodiments, generating the predicted risk of adverse pregnancy outcome includes performing multi-stage cascading inference with the multiple trained predictive models 206. In some embodiments, the risk monitor 220 is further configured to preprocess the feature data 224. The feature data 224 may be input to theDocket No. 85007-428789 predictive model 206 in response to preprocessing the feature data 224. Preprocessing the feature data 224 may include encoding categorical variables, performing missing value imputation, removing outliers, or performing feature value normalization.

[0066] The recommendation engine 222 is configured to generate a recommendation as a function of the predicted risk score based on one or more predetermined recommendation rules and to provide the recommendation to the patient. The recommendation may be indicative of frequency or location of maternal and fetal monitoring from routine monitoring to emergency monitoring. In some embodiments, the recommendation may be indicative of a recommended medication or a recommendation to start labor.

[0067] Although FIG. 2 illustrates the environment 200 as including all of the training data manager 202, the model trainer 204, the adverse pregnancy outcome predictive model 206, the model analyzer 210, the sample analyzer 212, and the risk monitor 220, it should be understood that in some embodiments one or more of those components may be established by different computing devices 102, and the operations of those components may be distributed in space and / or in time. For example, in an embodiment the training data manager 202, the model trainer 204, and the model analyzer 210 may be established by a computing device 102 configured to perform model training, for example in a cloud computing environment. Continuing that example, the sample analyzer 212 and the risk monitor 220 may be established by a computing device 102 configured to perform model inference, for example on a workstation in a clinical environment, or integrated in a point-of-care analytical device 104. Continuing that example further, the adverse pregnancy outcome predictive model 206 may be established by a separate computing device 102 to perform inference, for example in a cloud computing environment. Of course, other configurations of computing devices are possible in other environments.

[0068] Referring now to FIG. 3, in use, the computing device 102 may execute a method 300 for training the adverse pregnancy outcome predictive model 206. It should be appreciated that, in some embodiments, the operations of the method 300 may be performed by one or more components of the environment 200 of the computing device 102 as shown in FIG. 2. The method 300 begins with block 302, the computing device 102 collects a training dataset 226 including maternal static and time-series features. In block 304, the computing device 102 collects demographic data. In block 306, the computing device 102 collects routine clinical data including health data such as blood pressure and lab test results (e.g., biochemical marker test results).Docket No. 85007-428789

[0069] In an illustrative embodiment, training data 226 may be collected using preeclampsia as an example. Other complications and other adverse pregnancy outcomes may be applied via similar methods. In the illustrative embodiment a comprehensive training dataset 226 may be collected that includes maternal static and time-series features including blood pressure and biochemical markers from routine lab tests. The collected categories of variables / features included in the training dataset 226 were included in a Pregnancy Risk Prediction Panel, and are listed in Table 1, below. The training data may be collected from a diverse population to ensure accuracy and applicability across different demographics.Docket No. 85007-428789Docket No. 85007-428789Table 1. Pregnancy Risk Prediction Panel.

[0070] In block 308, the computing device 102 preprocesses the training dataset 226. Preprocessing may include cleaning, normalizing, feature scaling, or otherwise preparing the training dataset 226 for use in model training. In some embodiments, in block 310 the computing device 102 may determine missing values using an imputation technique such as ^-nearest neighbor (KNN) imputation, median-based imputation, or another imputation technique. In some embodiments, in block 312, the computing device 102 may encode categorical variables, such as race, smoking status, nulliparity, or other features.

[0071] In some embodiments, in block 314, the computing device 102 may determine statistical and temporal features for each variable. Input features may be constructed using both static variables and longitudinal clinical measurements extracted from the observation window. Static features include maternal demographics and obstetric history available before or early in pregnancy, while dynamic features may be derived from time-series laboratory tests and vital signs collect up to gestational week t_”obs.”

[0072] Static features include maternal characteristics, age, race, pregravid BMI, smoking status, alcohol use, nulliparity, gestational age at delivery, gravidity, number of fetuses, baby weight, delivery method, medication use, and medical history. These variables are selected based on their established associations with pregnancy adverse outcome risk in clinical literature.

[0073] For each variable X, a comprehensive set of statistical and temporal features we may be computed to capture the shape, trend, and complexity of its trajectory during the observation window. Specifically, the following transformations will be applied:

[0074] First-order statistics: Mean, standard deviation, minimum, maximum, and median.

[0075] Trend metrics: Slope of linear regression over time; baseline change rate defined as

[0076] Rolling window statistics: Rolling mean, rolling max, and rolling median computed over fixed-length daily sequences or observation-indexed positions.Docket No. 85007-428789

[0077] Exponentially weighted means (EWM) to emphasize recent values, defined recursively as:EWMt= axt+ (1 - a)EWMt-t, a G (0,1) (2).

[0078] Shape and distribution features: Interquartile range (IQR), skewness, kurtosis, and entropy of the observed values.

[0079] Autocorrelation: Lag-1 autocorrelation to quantify periodic structure in the measurements.

[0080] The number of measurements per feature vary across patients and variables. To mitigate sparsity, values may be aggregated using time-agnostic descriptors and allow variablelength sequences. In a secondary analysis, the use of learned representations from deep sequence models, including one-dimensional convolutional neural networks (1D-CNN), recurrent neural networks (LSTM), and attention-based encoders may be explored. These models are trained to compress each variable’s raw time series into a fixed-length embedding, which can be concatenated with the hand-crafted features.

[0081] In block 316, the computing device 102 trains the adverse pregnancy outcome predictive model 206 using the preprocessed training dataset 226. The computing device 102 may use any appropriate machine learning training algorithm to train the model 206. Illustratively, the predictive model 206 is a binary classification model trained using gradient-boosted decision trees implemented via XGBoost. This choice is motivated by the model’s ability to handle tabular data with heterogeneous feature scales, missing values, and complex nonlinear interactions, while maintaining interpretability through feature importance analysis.

[0082] As indicated by Table 1, the trained predictive model 206 relies on routine clinical data, which eliminates the need for new or unique biomarkers, thereby simplifying the testing process and reducing the burden on patients and healthcare systems. Unlike other models that may require specific or additional biomarkers, the disclosed predictive model 206 model exclusively uses standard clinical lab results and maternal characteristics. This makes the predictive model 206 more universally applicable and easier to adopt in standard clinical practice. Additionally, the predictive model 206 has the ability to dynamically integrate and analyze diverse data points (maternal characteristics along with multiple laboratory results) in real-time, which provides a holistic view of the patient’s condition, leading to potentially more accurate predictions. Additionally or alternatively, as described above, the disclosed predictive model 206 is not limitedDocket No. 85007-428789 to predicting preeclampsia. The disclosed technologies can also be applied to identify the risk of other pregnancy-related complications, including other hypertensive disorders of pregnancy, preterm delivery, IUGR, and stillbirth.

[0083] In some embodiments, in block 318 the computing device 102 may perform nested cross-validation, which ensures robust and unbiased performance estimation. In the illustrative embodiment, the outer loop included 5-fold cross-validation, where the training dataset 226 was split into five patient-level folds. In each round, one-fold is held out for evaluation, and the remaining four folds are used for training and hyperparameter tuning. The inner loop used a 4- fold split on the training dataset 226 to identify optimal hyperparameters via grid search, minimizing log loss on the validation set. The final reported metrics are averaged across the five outer test folds. All preprocessing steps, including imputation, feature scaling (if applied), and model training, are performed within each training fold to avoid data leakage. Such multi-fold cross-validation may more robustly estimate the performance of the model 206 given a relatively small dataset in the training data 226.

[0084] In some embodiments, in block 320 the computing device 102 may mitigate bias toward the negative class. Given the pronounced class imbalance in occurrence of preeclampsia or the other complications, several strategies may be applied to mitigate bias toward the negative class. In some embodiments, class-weighted loss functions may be used in XGBoost, where the positive class weight is set proportional to the inverse prevalence:Additionally or alternatively, in some embodiments, a two-stage cascade modeling approach may be used, in which an initial classifier with high sensitivity is used to identify a high-risk subset, followed by a more specific second-stage model to refine predictions. This strategy aims to improve positive predictive value (PPV) while maintaining acceptable recall. Hyperparameters such as maximum tree depth, learning rate, and number of estimators can be tuned during inner cross-validation. Missing values can be handled natively by the XGBoost algorithm, which leams optimal default directions for missing branches during training. To benchmark model performance, logistic regression, random forest, and multilayer perceptron models may be trained using the same input features.Docket No. 85007-428789

[0085] In block 322, the computing device 102 identifies most-predictive features of the training dataset 226. The most predictive features may be identified and selected from a broad range of routinely collected clinical data, including selecting one or more most predictive features from the pregnancy risk prediction panel shown above in Table 1. These predictive features may include demographic characteristics (e.g., maternal age, BMI, gravidity, parity, smoking status), serial blood pressure measurements, and laboratory values from the laboratory tests listed above. Feature selection is guided by statistical significance and predictive relevance using advanced machine learning algorithms, allowing the model to dynamically prioritize variables that contribute meaningfully to the accurate prediction of preeclampsia. In an illustrative embodiment, it was determined that blood pressure, ALP, pregravid BMI, gravidity, number of fetuses, age, globulin, creatinine, carbon dioxide, albumin, total protein level, platelet count, race, nulliparity, RBC, MCV, MPV, RDW, MCH, MCHC, glucose, potassium level, WBC, hemoglobin, hematocrit, anion gap, BUN, AST, ALT and total bilirubin are the most significantly different between preeclampsia and normal individuals by univariate test (urine features were not included). Of course, it should be understood that these significant features could be different in other embodiments, for example by involving more data and Al-machine learning model optimization. In some embodiments, in block 324 the computing device 102 may adjust predictions by the predictive model 206 based on progression of pregnancy.

[0086] In block 326, the computing device 102 evaluates the trained adverse pregnancy outcome predictive model 206. A grid of prediction tasks may be constructed by varying the observation window endpoint and the prediction window length A. This design allows for assessment of how predictive signal evolves across pregnancy and to identify optimal time windows for early risk stratification. Then, a combination of threshold-independent and thresholdspecific evaluation metrics may be used. Performance metrics include the area under the receiver operating characteristic curve (AUROC) and the area under the precision-recall curve (AUPRC). These metrics can be computed on the held-out fold in each outer cross-validation round and average across the five folds.

[0087] In addition to AUROC and AUPRC, several clinically relevant threshold-specific metrics can be reported:

[0088] Sensitivity at fixed specificity. Sensitivity can be calculated at a pre-specified specificity threshold, such as 80%, to simulate high-specificity screening.Docket No. 85007-428789

[0089] Positive predictive value (PPV) at fixed sensitivity. The likelihood that a model- flagged patient would truly develop preeclampsia can be measured, when sensitivity was constrained to match clinical screening levels (e.g., 80%).

[0090] Fl score'. The harmonic means of precision and recall, providing a balance between false positives and false negatives.

[0091] Calibration curves: The agreement between predicted probabilities and observed event rates can be assessed using reliability plots and binned calibration analysis.

[0092] Let y " i denote the predicted risk score for patient i, and y_i£{0, 1 } the ground-truth label. The AUROC is defined as:The AUPRC is given by:Prec dTPR (5).

[0093] The optimal decision threshold for each model may be computed using the Youden index:Thresholdoptiinal= argmax(Sensitivity(T) + Specificity(r) — 1) (6).

[0094] All metrics are averaged over the five outer cross-validation folds, and the mean and standard deviation are reported. External validation may be conducted using data from multiple independent sites to assess the generalizability and robustness of the predictive model 206 across diverse populations and clinical settings.

[0095] After training and evaluating the predictive model 206, the method 300 loops back to block 302, in which additional training data 226 may be collected and / or refined, and the predictive model 206 may continue to be trained, fine-tuned, or otherwise refined. The disclosed model 206 may be continuously updated and refined as more data becomes available or as further research identifies new predictive factors. This iterative process helps maintain the accuracy and relevancy of the model 206. Accordingly, the use of advanced machine learning techniques allows the predictive model 206 to continuously improve its accuracy and predictive capabilities as it processes more data, making it more reliable over time.

[0096] The trained model 206 may be deployed into production, for example by being deployed to a workstation computing device 102, deployed to a cloud computing server computing device 102, deployed to an integrated point of care instrument, or otherwise deployed for use.Docket No. 85007-428789

[0097] Referring now to FIG. 4, in use, the computing device 102 may execute a method 400 for predicting adverse pregnancy outcomes using the adverse pregnancy outcome predictive model 206. It should be appreciated that, in some embodiments, the operations of the method 400 may be performed by one or more components of the environment 200 of the computing device 102 as shown in FIG. 2. The method 400 begins with block 402, in which the computing device 102 receives feature data 224 for a patient. The computing device 102 may implement robust data security measures to protect sensitive patient information such as feature data 224, for example complying with healthcare regulations (e.g., HIPAA). Feature data 224 may be received via a user interface provided by the computing device 102. This user interface may be integrated into existing hospital or clinic IT systems. The user interface is designed to be user-friendly for healthcare providers or other clinicians to input patient data and receive risk assessments. For resource limited areas, POCT solutions can be provided. In some embodiments, in block 404 the computing device 102 may connect to an electronic health record (EHR) system or other data repository that includes patient feature data 224. For example, the disclosed system may be fully integrated into an existing EHR system, such as Epic. As described further below, by using patients’ demographic information, serial blood pressure measurements, and longitudinal routine laboratory testing results, without requiring any additional devices, the disclosed system 100 can generate and update a risk score. This score can be continuously displayed to both patients and obstetric providers, supporting timely and proactive patient management.

[0098] In block 406, the computing device 102 receives demographic data or other maternal characteristic data for the patient. The demographic data may be received, for example, from the EHR system or from another data repository that includes the demographic information concerning the patient. Additionally or alternatively, the demographic data may be input directly with the computing device 102, for example during a patient interview or other interaction with the patient. The basic demographic information collected may be based on attributes of the trained predictive model 206. For example, the demographic information may include the significant features identified during model training and evaluation, as described above in connection with block 322 of FIG. 3. For example, in the illustrative embodiment the significant features from maternal characteristics or other demographic data include age, gestational age, race, smoking status, nulliparity, and BMI.Docket No. 85007-428789

[0099] In some embodiments, in block 408 the computing device 102 may receive health data related to the patient. The health data may include clinical or laboratory data including hematology and chemistry test results related to the patients. The health data may include results from current tests and / or historical (e.g., time series) test results. Similar to the demographic information, the health data mya include significant features identified during model training and evaluation, as described above in connection with block 322 of FIG. 3. For example, in the illustrative embodiment the significant features from health data include blood pressure, ALP, pregravid BMI, gravidity, number of fetuses, age, globulin, creatinine level, carbon dioxide, albumin, total protein level, platelet count, race, nulliparity, RBC, MCV, MPV, RDW, MCH, MCHC, glucose, potassium level, WBC, hemoglobin, hematocrit, anion gap, BUN, AST, ALT and total bilirubin. In some embodiments, the health data may be received from a remote device such as an EHR system. In some embodiments, the health data may be received from an analytical device 104, which may be located remote from the computing device 102 or at another location. In some embodiments, the health data may be provided directly to the computing device 102, for example being input via a touch screen or other input device. In some embodiments, in block 410 the computing device 102 may receive pregnancy risk prediction panel data related to the patient. For example, the computing device 102 may receive any features or combination of features listed in the Pregnancy Risk Prediction Panel shown in Table 1, above.

[0100] In some embodiments, in block 412, the computing device 102 may generate analytical health data for the patient. For example, in some embodiments the computing device 102 may be integrated with or otherwise coupled to an analytical device 104 or other point of care system. The analytical device 104 may perform one or more tests (e.g., hematology tests, chemical tests, or other tests) to determine one or more health data features. Those health data features may include any features or combination of features listed in the Pregnancy Risk Prediction Panel shown in Table 1, above. In some embodiments, in block 414 the computing device 102 may collect and analyze a blood sample from the pregnant person. The blood sample may be collected by venipuncture for laboratory testing or capillary blood collection for a POCT device. In some embodiments, in block 416 the computing device 102 may collect and analyze a urine sample.

[0101] In block 418, the computing device 102 pre-processes the patient feature data 224, which includes demographic data and health data related to the patient. As described above, the feature data 224 may include data from an EHR system or other existing data source along withDocket No. 85007-428789 health data from one or more current tests and / or historical tests. Preprocessing the feature data 224 may include cleaning, normalizing, feature scaling, or otherwise preparing the feature data 224 for use with the predictive model 206. The feature data 224 may be preprocessed using one or more processes similar to the preprocessing performed on the training dataset 226 as described above in connection with block 308 of FIG. 3. For example, the computing device 102 may determine missing values using imputation techniques (e.g., KNN or median-based imputation), encode categorical variables, determine statistical and temporal features for each variable, or otherwise preprocess the feature data 224.

[0102] In block 420, the computing device 102 predicts risk of adverse pregnancy outcome by inputting the preprocessed patient feature data 224 to the trained adverse pregnancy outcome predictive model 206. As described above, the predictive model 206 may be embodied as a binary classifier, which outputs either a positive class (e.g., preeclampsia or other adverse outcome present) or a negative class (e.g., preeclampsia or other adverse outcome not present). The predictive model 206 may also output a confidence value or other score associated with the classification. The predictions and importance of different data points considered by the model 206 may be adjusted as the pregnancy progresses, for example as it gets closer to the time when preeclampsia or other complications typically shows up. In block 422, the computing device 102 determines a predicted risk score based on a predicted risk of adverse outcome provided by the predictive model 206. The risk score may indicate the likelihood of developing preeclampsia or other complications within the next 1 to 12 weeks. This scoring can be applied during either the antepartum or postpartum period, based on when the blood sample was taken. In some embodiments, the computing device 102 may perform multi-stage cascaded inference with multiple trained predictive models 206. For example, as described above, an initial classifier with high sensitivity may be used to identify a high-risk subset, followed by a more specific second- stage model to refine predictions.

[0103] In block 426, the computing device 102 generates one or more specific alerts or recommendations based on the predicted risk score. The recommendations may be generated as a function of the predicted risk score, based on one or more predetermined recommendation rules. Depending on the risk score, the results may include warnings of high risk, suggestions for more frequent check-ups, medication suggestions like aspirin, hospitalization, or recommendations to start labor. For example, for extremely high risk of preeclampsia patients, the computing deviceDocket No. 85007-428789102 may suggest that intensive monitoring and admission to hospital immediately once there are clear clinical features indicating preeclampsia. For relative medium risk patients, the computing device 102 may suggest enhanced monitoring and retest is required after 2 weeks, or immediately if the clinical situation changes. In those situations, the medium risk patient may also be hospitalized if clear preeclampsia or other complications features are identified, for example in subsequent testing. Continuing that example, for very low risk patients, the computing device 102 may suggest keeping the routine prenatal care check, and also retesting should be considered if the clinical situation changes.

[0104] In block 428, the computing device 102 presents the generated one or more recommendations to the patient and / or clinician. The results, including alerts and / or recommendations are shared with the pregnant person, potentially with advice on what to do next, such as closer monitoring, medical treatment options, or timing labor. Accordingly, the system 100 may be used to predict the risk of developing preeclampsia or other complications in pregnant women before clinical symptoms manifest. By integrating this predictive tool into clinical workflows, obstetric providers can identify high-risk patients early, enabling proactive monitoring, timely administration of preventive therapies (such as low-dose aspirin), and optimized delivery planning to minimize complications. The illustrative system 100 provides real-time processing, with immediate scoring, enabling prompt clinical decision-making. Additionally, by providing autonomous functionality, there may be no need for expert interpretation of the predictive model’s output. The system 100 is designed to be user-friendly and provide clear, actionable results, making it accessible to a wide range of healthcare providers, and not just specialists in preeclampsia or obstetrics. Accordingly, the point-of-care capabilities and autonomous functionality allow the disclosed system 100 to be used in resource limited areas without technical and clinical expertise. Additionally, training may be provided for healthcare providers on how to use the risk score effectively in clinical settings, including understanding the output and how it should influence clinical decisions.

[0105] After predicting risk and presenting recommendations, the method 400 loops back to block 402, in which the computing device 102 continues to receive feature data 224 and predict outcomes for the same patient and / or for additional patients. As described above, the predictive model 206 may be updated and improved by providing new and relevant data to make its predictions more accurate.Docket No. 85007-428789

[0106] Referring now to FIG. 5, diagram 500 illustrates one potential embodiment of an integrated point-of-care (POCT) instrument 502 that combines features of the computing device 102 and the analytical device 104. As shown, the illustrative POCT instrument 502 is a bench-top design including one or more measurement modules 142 and a sample handler 144, along with computational features such as a processor 120 and an integrated display 130. The POCT instrument 502 may perform one or more operations described above in connection with the method 400, including receiving feature data for the patient, generating analytical health data based on one or more patient samples, predicting risk of adverse pregnancy outcome using a pretrained predictive model 206, and presenting recommendations, for example using the display 130. The pre-trained predictive model 206 may be stored and executed locally by the integrated POCT instrument 502 or, in some embodiments, may be executed remotely for example in a cloud computing environment or other remote environment. The particular analytical tests performed by the POCT instrument 502 may be selected based on the significant features identified for the predictive model 206 as described above in connection with block 322 of FIG. 3. This approach eliminates the need for complex laboratory testing and IT infrastructure, allowing for easy and accessible risk score calculation at or near the point of care. A device prototype for such a POCT instrument 502 may be tested in the lab before field deployment, to ensure it meets the required standards for accuracy and repeatability. Then, pilot testing may be conducted in real-world clinical settings to gather feedback.

[0107] Additionally or alternatively, in some embodiments, one or more operations described above in connection with the method 400 may be performed by a mobile application, for example executed by a mobile computing device, tablet computing device, or other relatively low cost and accessible computing device. In those embodiments, the mobile application may perform operations including receiving feature data for the patient, receiving analytical health data for the patient (e g., historical health data, laboratory testing, or other health data), predicting risk of adverse pregnancy outcome using a pretrained predictive model 206, and presenting recommendations, for example using the display 130. The pre-trained predictive model 206 may be stored and executed locally by mobile device executing the mobile application, in some embodiments, may be executed remotely for example in a cloud computing environment or other remote environment. Similar to an integrated POCT device 502, using a mobile application alsoDocket No. 85007-428789 eliminates the need for complex laboratory testing and IT infrastructure, allowing for easy and accessible risk score calculation at or near the point of care.

[0108] Accordingly, the disclosed system 100 is space-efficient, due to its digital nature and ability to integrate with existing healthcare systems without needing additional hardware or significant storage space. Both large-scale instruments and point-of-care testing (POCT) devices can be applied to accommodate different space / clinical scenarios.

[0109] Similarly, the system 100 allows for integration of laboratory testing and Al. As described above, the necessary medical and / or lab features that an analytical device needs to measure may be selected according to the requirements of the predictive model 206. Large-scale and / or POCT testing devices may be integrated with the predictive model.

[0110] Additionally or alternatively, in some embodiments an adverse pregnancy outcome predictive model 206 can be developed without including laboratory data in the training dataset 226 and / or the feature data 224. Instead, in those embodiments the predictive model 206 may rely on demographic and blood pressure features. The predictive model 206 and risk calculator may be embodied into an advanced blood pressure monitoring device, which may enable at-home monitoring. By providing real-time risk scores, such an integrated device empowers patients to self-monitor and receive timely guidance on whether medical evaluation or hospital visits are necessary.

Claims

Docket No. 85007-428789CLAIMS:

1. A computing device for predicting health risk for a pregnant person, the computing device comprising: a trained predictive model; a feature data manager configured to receive feature data associated with a patient, wherein the feature data comprises predetermined categories of demographic information and predetermined categories of health data associated with the patient; and a risk monitor configured to (i) generate a predicted risk of adverse pregnancy outcome by an input of the feature data to the trained predictive model, and (ii) determine a predicted risk score based on the predicted risk of adverse pregnancy outcome.

2. The computing device of claim 1, wherein the adverse pregnancy outcome comprises preeclampsia, other hypertensive disorders of pregnancy, preterm delivery, intrauterine growth restriction, or stillbirth.

3. The computing device of claim 1, wherein the predetermined categories of demographic information comprises maternal characteristics.

4. The computing device of claim 1, wherein the predetermined categories of demographic information comprises age, race, pregravid body mass index (BMI), smoking status, alcohol use, nulliparity, gestational age at delivery, gravidity, number of fetuses, baby weight, delivery method, medication use, and medical history.

5. The computing device of claim 1, wherein the predetermined categories of health data comprises hematology test results.Docket No. 85007-4287896. The computing device of claim 1, wherein the predetermined categories of health data comprises chemistry test results.

7. The computing device of claim 1, wherein the predetermined categories of health data comprises urinalysis results.

8. The computing device of claim 1, wherein the predetermined categories of health data comprises routine blood test results.

9. The computing device of claim 1, wherein the predetermined categories of health data comprises urine protein analysis test results.

10. The computing device of claim 1, wherein the feature data associated with the patient comprises age, race, pregravid body mass index (BMI), smoking status, alcohol use, nulliparity, gestational age at delivery, gravidity, number of fetuses, baby weight, delivery method, medication (aspirin, anti -hypertensive medications, insulin, heparin) use, history of chronic hypertension, chronic kidney diseases, diabetes mellitus, auto-immune diseases or stroke, blood pressure and laboratory tests (Hematology tests: white blood cell count (WBC), red blood cell count (RBC), platelet count, hemoglobin, hematocrit, mean cell volume (MCV), mean cell hemoglobin (MCH), mean cell hemoglobin concentration (MCHC), red cell distribution width (RDW), mean platelet volume (MPV), nucleated RBC (NRBC), neutrophil count, lymphocyte count, monocyte count, eosinophil count, basophil count; chemistry test results: aspartate aminotransferase (AST), alanine aminotransferase (ALT), creatinine level, blood urea nitrogen (BUN), potassium level, alkaline phosphatase (ALP), sodium level, total protein level, albumin level, carbon dioxide, anion gap, globulin, calcium level, bilirubin total, bilirubin direct, estimated glomerular fdtration rate (eGFR), glucose, and chloride level; Urinalysis: urine color, urine pH,Docket No. 85007-428789 urine nitrite, urine bilirubin, urine ketones, urine appearance, urine glucose, urine blood, urine protein, urine creatinine, urine gravidity, urine leukocyte esterase).

11. The computing device of claim 1, wherein the predetermined categories of health data comprises lab test results.

12. The computing device of claim 1, wherein the predetermined categories of health data comprises sFlt-1 test results, P1GF test results, sFIt-l / PIGF ratio test results, or other biomarkers associated with the patient.

13. The computing device of claim 1, wherein the predetermined categories of health data comprises blood pressure data.

14. The computing device of claim 1, further comprising an input device configured to receive the predetermined categories of demographic information.

15. The computing device of claim 1, further comprising a sample analyzer configured to analyze a blood sample collected from the patient to generate blood sample test results, wherein the predetermined categories of health data comprises the blood sample test results.

16. The computing device of claim 15, wherein the blood sample is collected by venipuncture or capillary blood collection.

17. The computing device of claim 1, further comprising a sample analyzer to analyze a urine sample collected from the patient to generate urine sample test results, wherein the predetermined categories of health data comprises the urine sample test results.Docket No. 85007-42878918. The computing device of claim 1, wherein the trained predictive model comprises a random forest model, a deep learning model, a decision tree model, a logistic regression model, or a gradient-boosted tree model.

19. The computing device of claim 1, wherein the trained predictive model comprises a plurality of trained predictive models, and wherein to generate a predicted risk of adverse pregnancy outcome comprises to perform multi-stage cascading inference with the plurality of trained predictive models.

20. The computing device of claim 1, wherein: the risk monitor is further configured to preprocess the feature data; wherein to input the feature data comprises to input the feature data in response to preprocessing of the feature data.

21. The computing device of claim 20, wherein to preprocess the feature data comprises to encode categorical variables, to perform missing value imputation, to remove outliers, or to perform feature value normalization.

22. The computing device of claim 1, further comprising a recommendation engine configured to: generate a recommendation as a function of the predicted risk score based on a one or more predetermined recommendation rules; and provide the recommendation to the patient.Docket No. 85007-42878923. The computing device of claim 22, wherein the recommendation is indicative of frequency or location of maternal and fetal monitoring from routine monitoring to emergency monitoring.

24. The computing device of claim 22, wherein the recommendation is indicative of a recommended medication or a recommendation to start labor.

25. The computing device of claim 1, wherein the computing device comprises a point of care testing (POCT) device.

26. The computing device of claim 1, wherein the computing device comprises a healthcare facility information technology system.

27. The computing device of claim 1, wherein the computing device comprises a cloudbased system.

28. A computing device for pregnancy health risk model training, the computing device comprising: a predictive model; a data manager configured to collect a first training data set, wherein the first training data set comprises predetermined categories of demographic information and predetermined categories of health data associated with a training population of pregnant persons; and a model trainer configured to train the predictive model to predict a risk of adverse pregnancy outcome with the first training data set.Docket No. 85007-42878929. The computing device of claim 28, wherein the adverse pregnancy outcome comprises preeclampsia, other hypertensive disorders of pregnancy, preterm delivery, intrauterine growth restriction, or stillbirth.

30. The computing device of claim 28, wherein the predetermined categories of demographic information comprises maternal characteristics.

31. The computing device of claim 28, wherein the predetermined categories of demographic information comprises age, race, first-trimester body mass index (BMI), smoking status, nulliparity, gestational age, and medical history.

32. The computing device of claim 28, wherein the predetermined categories of health data comprises hematology test results.

33. The computing device of claim 28, wherein the predetermined categories of health data comprises chemistry test results.

34. The computing device of claim 28, wherein the predetermined categories of health data comprises urinalysis test results.

35. The computing device of claim 28, wherein the predetermined categories of health data comprises routine blood test results.

36. The computing device of claim 28, wherein the predetermined categories of health data comprises urine protein analysis test results.

37. The computing device of claim 28, wherein the first training data set comprises age, race, pregravid body mass index (BMI), smoking status, alcohol use, nulliparity, gestational ageDocket No. 85007-428789 at delivery, gravidity, number of fetuses, baby weight, delivery method, medication (aspirin, antihypertensive medications, insulin, heparin) use, history of chronic hypertension, chronic kidney diseases, diabetes mellitus, auto-immune diseases or stroke, blood pressure and laboratory tests (Hematology tests: white blood cell count (WBC), red blood cell count (RBC), platelet count, hemoglobin, hematocrit, mean cell volume (MCV), mean cell hemoglobin (MCH), mean cell hemoglobin concentration (MCHC), red cell distribution width (RDW), mean platelet volume (MPV), nucleated RBC (NRBC), neutrophil count, lymphocyte count, monocyte count, eosinophil count, basophil count; Chemistry test results: aspartate aminotransferase (AST), alanine aminotransferase (ALT), creatinine level, blood urea nitrogen (BUN), potassium level, alkaline phosphatase (ALP), sodium level, total protein level, albumin level, carbon dioxide, anion gap, globulin, calcium level, bilirubin total, bilirubin direct, estimated glomerular filtration rate (eGFR), glucose, and chloride level; Urinalysis: urine color, urine pH, urine nitrite, urine bilirubin, urine ketones, urine appearance, urine glucose, urine blood, urine protein, urine creatinine, urine gravidity, urine leukocyte esterase).

38. The computing device of claim 28, wherein the predetermined categories of health data comprises lab test results.

39. The computing device of claim 28, wherein the predetermined categories of health data comprises sFlt-1 test results, P1GF test results, sFIt-l / PIGF ratio test results, or other biomarkers associated with the patient.

40. The computing device of claim 28, wherein the predetermined categories of health data comprises blood pressure data.Docket No. 85007-42878941. The computing device of claim 28, wherein the trained predictive model comprises a random forest model, a deep learning model, a decision tree model, a logistic regression model, or a gradient-boosted tree model.

42. The computing device of claim 28, wherein: the data manager is further configured to preprocess the first training data; and to train the predictive model comprises to train the predictive model in response to preprocessing of the first training data.

43. The computing device of claim 42, wherein to preprocess the first training data comprises to encode categorical variables, to perform missing value imputation, to remove outliers, or to perform feature value normalization.

44. The computing device of claim 42, wherein to preprocess the first training data comprises to enrich the first training data with a statistical feature or a temporal feature based on the demographic information or the health data.

45. The computing device of claim 28, wherein to train the predictive model comprises to perform nested cross-validation with a plurality of patient-level folds of the first training data.

46. The computing device of claim 28, wherein to train the predictive model comprises to train a plurality of predictive models for multi-stage cascading inference.

47. The computing device of claim 28, further comprising a model analyzer configured to identify a most predictive feature of the first training data in response to training of the predictive model.Docket No. 85007-42878948. The computing device of claim 28, further comprising a model analyzer configured to evaluate performance of the predictive model in response to training of the predictive model.

49. A method for predicting health risk for a pregnant person, the method comprising: receiving, by a computing device, feature data associated with a patient, wherein the feature data comprises predetermined categories of demographic information and predetermined categories of health data associated with the patient; inputting, by the computing device, the feature data to a trained predictive model to generate a predicted risk of adverse pregnancy outcome; and determining, by the computing device, a predicted risk score based on the predicted risk of adverse pregnancy outcome.

50. The method of claim 49, wherein the adverse pregnancy outcome comprises preeclampsia, other hypertensive disorders of pregnancy, preterm delivery, intrauterine growth restriction, or stillbirth.

51. The method of claim 49, wherein the predetermined categories of demographic information comprises maternal characteristics.

52. The method of claim 49, wherein the predetermined categories of demographic information comprises age, race, pregravid body mass index (BMI), smoking status, alcohol use, nulliparity, gestational age at delivery, gravidity, number of fetuses, baby weight, delivery method, medication use, and medical history.

53. The method of claim 49, wherein the predetermined categories of health data comprises hematology test results.Docket No. 85007-42878954. The method of claim 49, wherein the predetermined categories of health data comprises chemistry test results.

55. The method of claim 49, wherein the predetermined categories of health data comprises urinalysis results.

56. The method of claim 49, wherein the predetermined categories of health data comprises routine blood test results.

57. The method of claim 49, wherein the predetermined categories of health data comprises urine protein analysis test results.

58. The method of claim 49, wherein the feature data associated with the patient comprises age, race, pregravid body mass index (BMI), smoking status, alcohol use, nulliparity, gestational age at delivery, gravidity, number of fetuses, baby weight, delivery method, medication (aspirin, anti -hypertensive medications, insulin, heparin) use, history of chronic hypertension, chronic kidney diseases, diabetes mellitus, auto-immune diseases or stroke, blood pressure and laboratory tests (Hematology tests: white blood cell count (WBC), red blood cell count (RBC), platelet count, hemoglobin, hematocrit, mean cell volume (MCV), mean cell hemoglobin (MCH), mean cell hemoglobin concentration (MCHC), red cell distribution width (RDW), mean platelet volume (MPV), nucleated RBC (NRBC), neutrophil count, lymphocyte count, monocyte count, eosinophil count, basophil count; Chemistry test results: aspartate aminotransferase (AST), alanine aminotransferase (ALT), creatinine level, blood urea nitrogen (BUN), potassium level, alkaline phosphatase (ALP), sodium level, total protein level, albumin level, carbon dioxide, anion gap, globulin, calcium level, bilirubin total, bilirubin direct, estimated glomerular filtration rate (eGFR), glucose, and chloride level; Urinalysis: urine color, urine pH, urine nitrite, urine bilirubin, urineDocket No. 85007-428789 ketones, urine appearance, urine glucose, urine blood, urine protein, urine creatinine, urine gravidity, urine leukocyte esterase).

59. The method of claim 49, wherein the predetermined categories of health data comprises lab test results.

60. The method of claim 49, wherein the predetermined categories of health data comprises sFlt-1 test results, P1GF test results, sFIt-l / PIGF ratio test results, or other biomarkers associated with the patient.

61. The method of claim 49, wherein the predetermined categories of health data comprises blood pressure data.

62. The method of claim 49, further comprising receiving, by the computing device, the predetermined categories of demographic information with an input device of the computing device.

63. The method of claim 49, further comprising: collecting a blood sample from the patient; and analyzing the blood sample to generate blood sample test results, wherein the predetermined categories of health data comprises the blood sample test results.

64. The method of claim 63, wherein collecting the blood sample comprises collecting the blood sample by venipuncture or capillary blood collection.

65. The method of claim 49, further comprising: collecting a urine sample from the patient; andDocket No. 85007-428789 analyzing the urine sample to generate urine sample test results, wherein the predetermined categories of health data comprises the urine sample test results.

66. The method of claim 49, wherein the trained predictive model comprises a random forest model, a deep learning model, a decision tree model, a logistic regression model, or a gradient-boosted tree model.

67. The method of claim 49, wherein generating the predicted risk of adverse pregnancy outcome comprises performing multi-stage cascading inference with a plurality of trained predictive models, wherein the trained predictive model comprises the plurality of trained predictive models.

68. The method of claim 49, further comprising: preprocessing, by the computing device, the feature data; wherein inputting the feature data comprises inputting the feature data in response to preprocessing the feature data.

69. The method of claim 68, wherein preprocessing the feature data comprises encoding categorical variables, performing missing value imputation, removing outliers, or performing feature value normalization.

70. The method of claim 49, further comprising: generating, by the computing device, a recommendation as a function of the predicted risk score based on a one or more predetermined recommendation rules; and providing, by the computing device, the recommendation to the patient.Docket No. 85007-42878971. The method of claim 70, wherein the recommendation is indicative of frequency or location of maternal and fetal monitoring from routine monitoring to emergency monitoring.

72. The method of claim 70, wherein the recommendation is indicative of a recommended medication or a recommendation to start labor.

73. The method of claim 49, wherein the computing device comprises a point of care testing (POCT) device.

74. The method of claim 49, wherein the computing device comprises a healthcare facility information technology system.

75. The method of claim 49, wherein the computing device comprises a cloud-based system.

76. A method for pregnancy health risk model training, the method comprising: collecting, by a computing device, a first training data set, wherein the first training data set comprises predetermined categories of demographic information and predetermined categories of health data associated with a training population of pregnant persons; and training, by the computing device, a predictive model to predict a risk of adverse pregnancy outcome using the first training data set.

77. The method of claim 76, wherein the adverse pregnancy outcome comprises preeclampsia, other hypertensive disorders of pregnancy, preterm delivery, intrauterine growth restriction, or stillbirth.Docket No. 85007-42878978. The method of claim 76, wherein the predetermined categories of demographic information comprises maternal characteristics.

79. The method of claim 76, wherein the predetermined categories of demographic information comprises age, race, pregravid body mass index (BMI), smoking status, alcohol use, nulliparity, gestational age at delivery, gravidity, number of fetuses, baby weight, delivery method, medication use, and medical history.

80. The method of claim 76, wherein the predetermined categories of health data comprises hematology test results.

81. The method of claim 76, wherein the predetermined categories of health data comprises chemistry test results.

82. The method of claim 76, wherein the predetermined categories of health data comprises urinalysis results.

83. The method of claim 76, wherein the predetermined categories of health data comprises routine blood test results.

84. The method of claim 76, wherein the predetermined categories of health data comprises urine protein analysis test results.

85. The method of claim 76, wherein the first training data set comprises age, race, pregravid body mass index (BMI), smoking status, alcohol use, nulliparity, gestational age at delivery, gravidity, number of fetuses, baby weight, delivery method, medication (aspirin, antihypertensive medications, insulin, heparin) use, history of chronic hypertension, chronic kidneyDocket No. 85007-428789 diseases, diabetes mellitus, auto-immune diseases or stroke, blood pressure and laboratory tests (Hematology tests: white blood cell count (WBC), red blood cell count (RBC), platelet count, hemoglobin, hematocrit, mean cell volume (MCV), mean cell hemoglobin (MCH), mean cell hemoglobin concentration (MCHC), red cell distribution width (RDW), mean platelet volume (MPV), nucleated RBC (NRBC), neutrophil count, lymphocyte count, monocyte count, eosinophil count, basophil count; Chemistry test results: aspartate aminotransferase (AST), alanine aminotransferase (ALT), creatinine level, blood urea nitrogen (BUN), potassium level, alkaline phosphatase (ALP), sodium level, total protein level, albumin level, carbon dioxide, anion gap, globulin, calcium level, bilirubin total, bilirubin direct, estimated glomerular filtration rate (eGFR), glucose, and chloride level; Urinalysis: urine color, urine pH, urine nitrite, urine bilirubin, urine ketones, urine appearance, urine glucose, urine blood, urine protein, urine creatinine, urine gravidity, urine leukocyte esterase).

86. The method of claim 76, wherein the predetermined categories of health data comprises lab test results.

87. The method of claim 76, wherein the predetermined categories of health data comprises sFlt-1 test results, P1GF test results, sFIt-l / PIGF ratio test results, or other biomarkers associated with the patient.

88. The method of claim 76, wherein the predetermined categories of health data comprises blood pressure data.

89. The method of claim 76, wherein the trained predictive model comprises a random forest model, a deep learning model, a decision tree model, a logistic regression model, or a gradient-boosted tree model.Docket No. 85007-42878990. The method of claim 76, further comprising: preprocessing, by the computing device, the first training data; wherein training the predictive model comprises training the predictive model in response to preprocessing the first training data.

91. The method of claim 90, wherein preprocessing the first training data comprises encoding categorical variables, performing missing value imputation, removing outliers, or performing feature value normalization.

92. The method of claim 90, wherein preprocessing the first training data comprises enriching the first training data with a statistical feature or a temporal feature based on the demographic information or the health data.

93. The method of claim 76, wherein training the predictive model comprises performing nested cross-validation with a plurality of patient-level folds of the first training data.

94. The method of claim 76, wherein training the predictive model comprises training a plurality of predictive models for multi-stage cascading inference.

95. The method of claim 76, further comprising identifying, by the computing device, a most predictive feature of the first training data in response to training the predictive model.

96. The method of claim 76, further comprising evaluating, by the computing device, performance of the predictive model in response to training the predictive model.

97. A computing device comprising: a processor; andDocket No. 85007-428789 a memory having stored therein a plurality of instructions that when executed by the processor cause the computing device to perform the method of any of claims 49-96.

98. One or more machine readable storage media comprising a plurality of instructions stored thereon that in response to being executed result in a computing device performing the method of any of claims 49-96.

99. A computing device comprising means for performing the method of any of claims 49-96.

Images