Predictive machine learning model for preeclampsia using artificial neural networks

Abstract

A method may include generating and / or using a predictive machine learning classifier including one or more artificial neural networks. The classifier is configured to output a prediction related to the onset of preeclampsia (e.g., premature premature preeclampsia) during the current pregnancy of the patient based on the health feature and the one or more DNA metrics. The health characteristics and DNA indicators, such as total cell free DNA (cfDNA) and fetal fraction (FF), can be obtained during conventional and non-invasive or minimally invasive prenatal screening. The predictive machine learning classifier may be generated by applying a deep learning technique to data regarding the subjects in the queue. The data may include characteristics corresponding to the outcome of past pregnancy, health indicators, and one or more cfDNA measurements. Early pregnancy risk assessment of preterm preeclampsia can identify patients who are most likely to benefit from prophylactic treatment regimens using lowest or low levels of intervention.

Description

[0001] CROSS-REFERENCE TO RELATED APPLICATIONS

[0002] This patent application claims priority to U.S. Provisional Patent Application No. 63 / 324,469 filed on March 28, 2022 and U.S. Provisional Patent Application No. 63 / 400,380 filed on August 23, 2022, each of which is incorporated herein by reference in its entirety. Technical Field

[0003] The present technology generally relates to the use of artificial intelligence to predict health outcomes for pregnant subjects, and more specifically, to machine learning models for identifying patients at risk for developing preeclampsia during a current pregnancy using patient characteristics, such as those collected non-invasively or minimally invasively during routine visits to a healthcare provider. Background Art

[0004] Pre-eclampsia is a major cause of maternal mortality and morbidity. Early pregnancy risk assessment for preterm pre-eclampsia (<37 weeks gestation) can identify patients most likely to benefit from low-dose aspirin. Low-dose aspirin started between 12 and 16 weeks of gestation has been shown to reduce the risk of early-onset pre-eclampsia in high-risk patients. Accurate early pregnancy risk assessment for pre-eclampsia can identify patients most likely to benefit from aspirin started between 12 and 16 weeks of gestation (when evidence of aspirin's effectiveness is available). Risk assessment is also important to guide appropriate pregnancy care pathways and monitoring.

[0005] Current U.S. Preventive Services Task Force (USPSTF) guidelines have limited clinical ability to identify patients at risk for preeclampsia, and more accurate early pregnancy risk assessment has been identified as a priority. Previous approaches to predicting preeclampsia have used patient demographics and clinical characteristics, ultrasound measurements, and biomarkers. However, the more expensive, difficult to obtain, invasive, and / or unconventional the health data required to predict preterm preeclampsia, the more likely it is that at least some high-risk patients will not be identified because such health data are less commonly available. Summary of the invention

[0006] Various potential embodiments of the disclosed method relate to a method for applying a predictive machine learning classifier by a computing system to generate a prediction of a patient's occurrence of a pre-eclampsia condition during the current pregnancy, such as preterm pre-eclampsia or term pre-eclampsia, or other pregnancy outcomes. The predictive machine learning classifier may include one or more artificial neural networks. Generating a prediction may include obtaining patient data by at least one of the following: (A) performing non-invasive prenatal screening (NIPS) on the patient, and / or (B) receiving data from a health record system by a computing system through a communication network. Patient data may include (i) a set of health characteristics, and / or (ii) one or more cell-free DNA (cfDNA) indicators. As used herein, a cfDNA indicator is any measurement, derivation, or calculation value based on data generated from one or more cfDNA analyses. Examples of cfDNA indicators include, but are not limited to, total cfDNA, fetal fraction (FF), SNP data from cfDNA screening, etc. Generating a prediction may include feeding patient data to a predictive machine learning classifier to obtain a prediction of the occurrence of pre-eclampsia in a patient during the current pregnancy. The method may include providing a prediction for the care of a patient by a computing system. Providing a prediction may include at least one of the following: transmitting the prediction by a computing system to at least one of a health record system and / or a computing device associated with a healthcare provider via a communication network. The predictive machine learning classifier may have been trained by applying one or more machine learning techniques to data about subjects in a cohort of pregnant subjects. In various potential implementations, one or more machine learning techniques may include, for example, deep learning, logistic regression, random forests, and / or gradient boosting. For each pregnant subject, the data may include a plurality of the following items: (A) a first feature set indicating one or more outcomes of one or more previous pregnancies of the pregnant subject; (B) a second feature set including one or more health indicators of the pregnant subject; and / or (C) a third feature set based on one or more cfDNA measurements of the pregnant subject. As used herein, health indicators (interchangeably used with health features or health data) include values ​​(measured or derived) associated with health quality or health level, and include, but are not limited to, blood pressure, body mass index (BMI), weight, biomarkers (e.g., blood levels of various compounds), and the like. As used herein, patient characteristics (used interchangeably with patient data) may include or may be health indicators and / or other information about a patient that does not necessarily indicate health quality or health level by itself. As a non-limiting example, sex at birth may be a patient characteristic.

[0007] In various embodiments, all patient data used to obtain predictions can be collected non-invasively during one or more routine medical visits of the patient. In various embodiments, the patient data used to obtain predictions can exclude uterine artery Doppler and / or other ultrasound data. In various embodiments, pre-eclampsia is preterm pre-eclampsia or early-onset pre-eclampsia. In various embodiments, the method may include generating a training data set based on a first, second and / or third feature set by a computing system. In various embodiments, the method may include training a prediction machine learning classifier using data about subjects in a pregnant subject cohort by a computing system. In various embodiments, the prediction machine learning classifier may include, for example, any combination of one or more rectified linear units (reLu), sigmoid linear units (SiLU), softmax functions, etc. In various embodiments, the prediction machine learning classifier may include a hidden layer having the same dimension as the input layer. In various embodiments, the prediction machine learning classifier may include multiple hidden layers and may have different dimensions. In various embodiments, one or more deep learning techniques may include using one or more regularization terms (e.g., L2 regularization terms).

[0008] Various potential embodiments of the disclosed method relate to a method comprising generating a predictive machine learning classifier for pre-eclampsia in a pregnant patient. The predictive machine learning classifier may include one or more artificial neural networks. Generating the predictive machine learning classifier may include: generating a training data set by a computing system using data about subjects in a cohort of pregnant subjects. For each pregnant subject, the training data set may include a plurality of the following items: (A) a first feature set indicating one or more outcomes of one or more previous pregnancies of the subject; (B) a second feature set corresponding to one or more health indicators of the subject; and / or (C) a third feature set corresponding to one or more cfDNA indicators. The method may include applying one or more deep learning techniques to the training data set by a computing system to generate a predictive machine learning classifier. The predictive machine learning classifier may be configured to receive patient health characteristics and one or more cfDNA indicators as inputs, and output a prediction of the occurrence of pre-eclampsia. Patient health characteristics or other patient data may be obtained from or via, for example, an electronic medical record system. The method may include providing a predictive machine learning classifier by a computing system to generate patient-specific indicators for individual patients. Providing the predictive machine learning classifier may include at least one of: (i) transmitting the predictive machine learning classifier by the computing system to a second computing system via a communication network, and / or (ii) storing the predictive machine learning classifier in a non-transitory computer-readable storage medium that can be retrieved by the healthcare provider for subsequent use. The second computing system may be, for example, an electronic medical record system.

[0009] In various example implementations, the predictive machine learning classifier may include one or more implicit rectified linear units (reLu). In various implementations, the predictive machine learning classifier may include a hidden layer having the same dimension as the input layer. In various implementations, applying one or more deep learning techniques may include using one or more regularization terms. In various implementations, one or more regularization terms may include an L2 regularization term. In various implementations, the method may include using a predictive machine learning classifier to assess a patient during the current pregnancy by: performing non-invasive prenatal screening (NIPS) on the patient; feeding data to a predictive machine learning classifier to obtain a prediction of preeclampsia occurring during the current pregnancy, the data including data from NIPS; and / or using the prediction to provide medical care to the patient. In various implementations, all data provided to the predictive machine learning classifier to obtain a prediction can be collected in a non-invasive or minimally invasive manner (e.g., blood draw for NIPS or biomarkers) during one or more routine medical visits of the patient. In various embodiments, the method may further include using a predictive machine learning classifier to evaluate the patient during the current pregnancy by: receiving data corresponding to one or more previous pregnancies, one or more health characteristics, and one or more cfDNA indicators of the patient from a health record system; feeding the data to the predictive machine learning classifier to obtain a prediction of the occurrence of preterm preeclampsia during the current pregnancy; and / or using the prediction to provide medical care to the patient. In various embodiments, all data input into the predictive machine learning classifier to obtain the prediction can be collected in a non-invasive or minimally invasive manner during one or more routine medical visits of the patient. In various embodiments, preterm preeclampsia can be early-onset preterm preeclampsia.

[0010] Various potential embodiments of the disclosed method relate to a computing system configured to apply a predictive machine learning classifier to generate a prediction of a patient's occurrence of premature pre-eclampsia during the current pregnancy. The predictive machine learning classifier may include one or more artificial neural networks. The computing system may include one or more processors configured to receive patient data including a set of health features and one or more cfDNA measurements through a communication network during the patient's current pregnancy. One or more processors may be configured to feed patient data to the predictive machine learning classifier to obtain a prediction of the occurrence of premature pre-eclampsia in the patient during the current pregnancy. Data used for and / or fed to the predictive machine learning classifier may be collected from or via one or more electronic health record systems. One or more processors may be configured to provide predictions for the care of patients. Providing a prediction may include transmitting the prediction to a health record system (e.g., an electronic medical record) and / or at least one of a computing device associated with a healthcare provider through a communication network. The predictive machine learning classifier may have been trained by applying one or more deep learning techniques to data about subjects in a cohort of pregnant subjects. For each pregnant subject, the training data may have included a plurality of the following: (A) a first feature set indicating one or more outcomes of one or more previous pregnancies of the pregnant subject; (B) a second feature set corresponding to one or more health indicators of the pregnant subject; and / or (C) a third feature set corresponding to one or more cell-free DNA indicators of the pregnant subject.

[0011] In various embodiments, all patient data for obtaining predictions may have been collected in a non-invasive manner during one or more routine medical visits of the patient. In various embodiments, ultrasound data may be excluded from the patient data for obtaining predictions. In various embodiments, premature pre-eclampsia is early-onset premature pre-eclampsia. In various embodiments, one or more processors are configured to generate a training data set and apply deep learning techniques to the training data set to generate a prediction machine learning classifier. In various embodiments, one or more processors are configured to receive at least a subset of health data from a health record system via a communication network. In various embodiments, one or more processors are configured to receive at least a subset of health data via a software application running on a computing device associated with a healthcare provider.

[0012] Various embodiments of the present disclosure may be directed to processes performed using the apparatus and / or systems disclosed herein.

[0013] Various embodiments of the present disclosure may be directed to computing systems and / or devices for performing the processes disclosed herein. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] Figure 1 Example systems for implementing the disclosed methods are described in accordance with various potential embodiments.

[0015] Figure 2 Example modeling and prediction processes are described according to various potential embodiments.

[0016] Figure 3 Shown is a simplified block diagram of a representative server system and client computer system that can be used to implement certain embodiments of the present disclosure.

[0017] Figure 4 Data for generating and validating example models according to potential embodiments of the disclosed methods are described.

[0018] Figure 5 An example production workflow according to a potential embodiment of the disclosed method is presented.

[0019] Figure 6 Describe the feature significance and relative contribution to the final prediction of preterm pre-eclampsia according to potential embodiments of the disclosed method. The upper figure (A) corresponds to the linear neural network (LNN) model, and the lower figure (B) corresponds to the nonlinear neural network (NLNN) model. The average contribution and 95% confidence interval are shown. The size and direction of the bar represent the contribution of each single feature to the definition of preterm pre-eclampsia positive patients.

[0020] Figure 7 Describe the feature significance of the simplified model with cfDNA features deleted according to the potential embodiment of the disclosed method and the relative contribution to the final prediction of preterm preeclampsia. The upper figure (A) corresponds to the simplified LNN model, and the lower figure (B) corresponds to the simplified NLNN model. The average contribution and 95% confidence interval are shown. The size and direction of the bar represent the contribution of each single feature to the definition of preterm preeclampsia positive patients.

[0021] The foregoing and other features of the present disclosure will become apparent from the following description and the appended claims in conjunction with the accompanying drawings. It should be understood that these drawings depict only several embodiments according to the present disclosure and therefore should not be considered to limit the scope of the present disclosure, which will be described with additional specificity and detail through the use of the accompanying drawings. DETAILED DESCRIPTION

[0022] It should be appreciated that certain aspects, modes, embodiments, variations and features of the methods of the present invention are described below at different levels of detail in order to provide a substantial understanding of the technology of the present invention. It should be understood that the present disclosure is not limited to a particular use, method, device or system, each of which can of course vary. It should also be understood that the terminology used herein is for the purpose of describing specific embodiments only and is not intended to be limiting.

[0023] Lower-intervention preventive treatments (such as low-dose aspirin) tend to be associated with reduced risk of complications or adverse outcomes and reduced costs and are therefore generally preferred over higher-intervention treatments. Patient characteristics known to be associated with an increased risk of preeclampsia are routinely collected as part of prenatal care. Additionally, early pregnancy maternal and fetal placental cell-free DNA (cfDNA) results may be associated with an increased risk of developing preeclampsia and are typically provided as part of noninvasive prenatal screening (NIPS), which is the primary screening test for fetal chromosomal abnormalities in the United States.

[0024] In potential embodiments, an artificial neural network model is generated and / or used to predict preterm preeclampsia (preeclampsia at <37 weeks of gestation) and early-onset preeclampsia (preeclampsia at <34 weeks of gestation) using routinely available and / or non-invasively collected patient characteristics, such as health data available at the first prenatal visit and data from cfDNA screening.

[0025] refer to Figure 1 In a potential implementation, the system 100 may include a computing system 110 (which may be or may include one or more computing devices, co-located or remote from each other), a condition detection system 160, a clinical information system (CIS) 170 (interchangeably used with an electronic medical record (EMR) system or an electronic health record (EHR) system), a platform 175, and a treatment system 180. The computing system 110 (e.g., one or more computing devices) may be used to control and / or exchange signals and / or data with the condition detection system 160, CIS 170, platform 175, and / or treatment system 180, either directly or via another component of the system 100. In certain embodiments, the computing system 110 may be used to control and / or exchange data or other signals with the condition detection system 160, CIS 170, platform 175, and / or treatment system 180. The computing system 110 may include one or more processors and one or more volatile and non-volatile memories for storing captured, collected, recorded, and / or generated computing code and data.

[0026] The computing system 110 may include a controller 112 configured to exchange control signals with the condition detection system 160, the CIS 170, the platform 175, the treatment system 180, and / or any components thereof, thereby allowing the computing system 110 to control, for example, the collection of patient data such as test results, the capture of images, the acquisition of signals through sensors, the positioning or repositioning of subjects and patients, the recording or obtaining of other subject or patient information, and the application of therapy.

[0027] The transceiver 114 allows the computing system 110 to exchange readings, control commands, and / or other data with the condition detection system 160, CIS 170, platform 175, and / or treatment system 180, or parts thereof, wirelessly or via wires, directly or via a network protocol. One or more user interfaces 116 allow the computing device 110 to receive user input (e.g., via a keyboard, touch screen, microphone, camera, etc.) and provide output to the user (e.g., via a display screen, audio speakers, etc.). The computing device 110 may additionally include one or more databases 118 for storing, for example, data collected from one or more systems or devices, signals collected via one or more sensors, biomarker signatures, etc. In some implementations, the database 118 (or portions thereof) may alternatively or additionally be part of another computing device that is co-located or remote and communicates with the computing device 110, the condition detection system 160, CIS 170, platform 175, and / or treatment system 180, or parts thereof (e.g., via "cloud computing").

[0028] Condition detection system 160 may include testing system 162, which may be or may include, for example, any system or device that involves, for example, analyzing a sample (e.g., blood, plasma, or urine), recording patient data, and / or obtaining laboratory or other test results (e.g., DNA data). Imaging system 164 may be any system or device for capturing imaging data, such as a magnetic resonance imaging (MRI) scanner, a positron emission tomography (PET) scanner, a single photon emission computed tomography (SPECT) scanner, a computed tomography (CT) scanner, a fluoroscopic scanner, and / or other imaging devices and / or sensors. Sensor 166 may detect, for example, the position or movement of a patient, an organ, a tissue, a physiological reading such as vital capacity or cardiac activity / signals, or other states and / or conditions of a patient.

[0029] The treatment system 180 may include a treatment unit 180, which may be or may include, for example, a radiation source for external beam therapy (e.g., an orthovoltage X-ray machine, a cobalt-60 machine, a linear accelerator, a proton beam machine, a neutron beam machine, etc.) and / or one or more other treatment devices. Sensors 184 may be used by the treatment system 180 to assess and guide treatment (e.g., by detecting the level of emitted radiation, the condition or state of the patient, or other states or conditions). In various implementations, the components of the system 100 may be rearranged or integrated in other configurations. For example, the computing system 110 (or its components) may be integrated with one or more of the condition detection system 160, the treatment system 180, and / or its components. The condition detection system 160, the treatment system 180, and / or its components may be directed to a platform 175, on which a patient, subject, or sample may be located (in order to test the sample, image the subject, treat or treat the subject, detect activity and / or movement of the subject in a stress test, etc.). In various embodiments, the platform 175 can be movable (e.g., using any combination of motors, magnets, etc.) to allow positioning and repositioning of the sample and / or subject (such as fine adjustment to compensate for subject or patient motion, or positioning the patient to scan different regions of interest). The platform 175 can include its own sensors to detect the condition or state of the sample, patient, and / or subject.

[0030] The computing system 110 may include a tester and imager 120 configured to direct, for example, laboratory tests, image capture, and acquisition of test and / or imaging data. The tester and imager 120 may include an image generator that may convert or transform raw imaging data from the condition detection system 160 into a usable medical image or into another form to be analyzed. The computing system 110 may include a test and image analyzer 122 configured to, for example, analyze raw test results to generate relevant indicators, identify features in images or imaging data, or otherwise utilize tests or test data, and / or images or imaging data.

[0031] The data acquisition unit 124 can retrieve, collect or otherwise obtain various data for, for example, training a model using data about a subject, or applying a trained model using data about a patient. The data acquisition unit 124 can, for example, obtain test data stored in the CIS 170. The interaction unit 126 can interact with a user (e.g., a patient and / or a healthcare provider) (e.g., via the user interface 116) to obtain information required by the model or provide information to the user. In some embodiments, the data acquisition unit 124 can obtain data from the user via the interaction unit 126.

[0032] The machine learning model trainer 130 ("ML model trainer") can be configured to train the machine learning model used herein, as further discussed below. The ML model trainer can include a training data set generator that processes various data to generate one or more training data sets to be used, for example, to train the model discussed herein. In certain embodiments, as more data and outcomes become available, the model retrainer 134 can further train the previously trained model to improve or otherwise update or revise the model. The ML model trainer can apply various machine learning techniques to the training data set. The machine learning modeler 140 ("ML modeler") can be configured to apply the trained machine learning model to specific patient data. The feature selector 142 can be configured to select which features to feed to the model to obtain a prediction (e.g., the likelihood of developing term pre-eclampsia, pre-term pre-eclampsia, and / or early-onset pre-eclampsia). The feature selector 142 can select features based on, for example, which features were previously used to train the ML model to be used, the parameters of the model, and / or the influence of the features in more predicted outcomes (e.g., feature coefficients). The feature selector 142 can select features based on which data (e.g., test results) are available for the patient in addition or alternatively. For example, if the data corresponding to a feature in the "1st layer" feature set (e.g., the "first five" most influential features with the five highest coefficients) is not available for the patient, two features in the "2nd layer" feature set (e.g., features with the 6th to 10th highest coefficients) can be selected instead. The feature extractor 144 can obtain the values ​​of the selected features from various data sources. The output and reporting unit 150 can transmit the model or its prediction (to, for example, a healthcare professional, a patient, etc.), or generate and provide a report based on the model or its prediction. The report can include, for example, the model output (e.g., prediction) together with information about the basis of the prediction, and guidance and / or suggestions and subsequent steps based on the prediction, so as to identify the factors that have the greatest impact on low or high probability or other prognosis for clinicians, and guide clinicians and patients.

[0033] refer to Figure 2 , shows an example process 200 according to various potential embodiments. Various elements of process 200 may be implemented by or via system 100 or components thereof. If a model is not already available (e.g., in database 118), or if an additional model is to be generated or updated by training or retraining using new training data, process 200 may be generated from model training (in Figure 2 ), the model training may be performed by or via computing system 110. Alternatively, if a trained model is already available, process 200 may begin by using (applying) the model to a patient (on the left side of Figure 2). If a suitable trained model is available, prediction can be achieved by or via computing system 110. In various embodiments, process 200 may include model training (e.g., steps 210-225) and subsequent prognosis prediction (e.g., steps 250-270).

[0034] At 210, health data related to the subject can be obtained. This can be obtained by or via, for example, the condition detection system 160 and / or the CIS 170 for a cohort of, for example, 10,000, 11,000, 12,000, 13,000, 14,000, 15,000 or more subjects. For example, this can represent, for example, multiple years of treatment (e.g., multiple pregnancies. The health data should have enough data (e.g., across multiple pregnancies) to be able to capture and explain various features and build a generalizable model. The laboratory test or imaging data can be transformed, converted, or otherwise analyzed by the test and image analyzer 122. Step 215 involves extracting feature values ​​corresponding to the subjects in the cohort from (or based on) the health data.

[0035] At 220, one or more data sets (e.g., training data sets) may be generated using the extracted feature values ​​corresponding to the subject's health data. This may be performed, for example, by or via the ML model trainer 130, or more specifically by or via the training and test data set generator 132. At 225, one or more data sets may be used to train and test a model for making predictions. The model may be stored (e.g., in a database 118) for subsequent use. Process 200 may end (290), or may proceed to step 250 for prognostic prediction. (As indicated by the dashed line from step 225 to step 265, the model may then be used to generate and use predictions.)

[0036] At 250, the patient's health data may be obtained and analyzed (e.g., by or via condition detection system 160 and / or CIS 170). The test data may correspond to various laboratory tests performed on the patient's samples. At 255, a set of features may be selected (e.g., by or via feature selector 142), and at 260, values ​​of the features may be extracted from the patient's health data (e.g., by or via feature extractor 144).

[0037] At 265, feature values ​​extracted from the patient health data can be input into a predictive model (e.g., a machine learning classifier) ​​to predict the patient's prognosis. At 270, the predicted prognosis and the factors that form the basis of the prognosis can be used to care for the patient. For example, the predicted prognosis can be used to plan potential outcomes and / or identify potential preventive care. Additionally or alternatively, the predicted prognosis can be used to evaluate the patient's health and its changes or trends, such as whether the patient is deteriorating or improving, as indicated by changes in the prognosis over time from the ML modeler as new tests are performed or otherwise as new health data becomes available.

[0038] Process 200 may end (290), or return to step 250 (eg, after running another test or administering treatment) for follow-up planning based on changes in the patient's condition.

[0039] The various operations described herein may be implemented on computer systems having various configurations. Figure 3 A simplified block diagram of a representative server system 300 (e.g., computing system 110) and a client computer system 314 (e.g., computing system 110, condition detection system 160, CIS 170, and / or treatment system 180) that can be used to implement various embodiments of the present disclosure is shown. In various embodiments, server system 300 or a similar system can implement the services or servers described herein or portions thereof. Client computer system 314 or a similar system can implement the clients described herein.

[0040] Server system 300 may have a modular design including multiple modules 302 (eg, blades in a blade server embodiment); although two modules 302 are shown, any number may be provided. Each module 302 may include a processing unit 304 and a local memory 306.

[0041] Processing unit 304 may include a single processor, which may have one or more cores, or multiple processors. In some embodiments, processing unit 304 may include a general purpose main processor and one or more dedicated coprocessors, such as a graphics processor, a digital signal processor, etc. In some embodiments, some or all of processing units 304 may be implemented using custom circuits, such as application specific integrated circuits (ASICs) or field programmable gate arrays (FPGAs). In some embodiments, such integrated circuits execute instructions stored on the circuit itself. In other embodiments, processing unit 304 may execute instructions stored in local memory 306. Any type of processor in any combination may be included in processing unit 304.

[0042] The local memory 306 may include volatile storage media (e.g., conventional DRAM, SRAM, SDRAM, etc.) and / or non-volatile storage media (e.g., magnetic disks or optical disks, flash memory, etc.). The storage media incorporated into the local memory 306 may be fixed, removable, or upgradeable as needed. The local memory 306 may be physically or logically divided into various subunits, such as system memory, read-only memory (ROM), and permanent storage. The system memory may be a read-write memory device or a volatile read-write memory, such as a dynamic random access memory. The system memory may store some or all of the instructions and data required by the processing unit 304 at runtime. The ROM may store static data and instructions required by the processing unit 304. The permanent storage may be a non-volatile read-write memory device that may store instructions and data even when the module 302 is powered off. The term "storage medium" as used herein includes any medium in which data may be stored indefinitely (experienced rewriting, electrical interference, power loss, etc.) and does not include carrier waves and transient electronic signals transmitted wirelessly or through wired connections.

[0043] In some embodiments, local memory 306 may store one or more software programs to be executed by processing unit 304, such as an operating system and / or programs implementing various server functions, or any system or apparatus described herein.

[0044] "Software" generally refers to a sequence of instructions that, when executed by processing unit 304, causes server system 300 (or portions thereof) to perform various operations, thereby defining one or more specific machine embodiments of the operations for executing and running software programs. Instructions may be stored as firmware residing in read-only memory and / or program code stored in non-volatile storage media, which may be read into volatile working memory for execution by processing unit 304. Software may be implemented as a single program or as a collection of separate programs or program modules that interact as needed. Processing unit 304 may retrieve program instructions to be executed and data to be processed from local memory 306 (or non-local memory described below) to perform the various operations described above.

[0045] In some server systems 300, multiple modules 302 may be interconnected via a bus or other interconnect 308, forming a local area network that supports communications between modules 302 and other components of server system 300. Interconnect 308 may be implemented using a variety of technologies, including server racks, hubs, routers, and the like.

[0046] Wide area network (WAN) interface 310 can provide data communication capabilities between the local area network (interconnect 308) and a larger network such as the Internet. Conventional or other active technologies can be used, including wired (e.g., Ethernet, IEEE 802.3 standards) and / or wireless technologies (e.g., Wi-Fi, IEEE 802.11 standards).

[0047] In some embodiments, local memory 306 is intended to provide working memory for processing unit 304, providing fast access to programs and / or data to be processed while reducing the amount of traffic on interconnect 308. Storage for larger amounts of data can be provided on a local area network by one or more mass storage subsystems 312 that can be connected to interconnect 308. Mass storage subsystem 312 can be based on magnetic, optical, semiconductor or other data storage media. Direct attached storage, storage area networks, network attached storage, etc. can be used. Any data store or other data collection described herein as being generated, consumed or maintained by a service or server can be stored in mass storage subsystem 312. In some embodiments, additional data storage resources can be accessed via WAN interface 310 (possibly with increased latency).

[0048] The server system 300 can operate in response to requests received via the WAN interface 310. For example, one of the modules 302 can implement monitoring functions and assign discrete tasks to other modules 302 in response to received requests. Conventional work allocation techniques can be used. When the request is processed, the result can be returned to the requester via the WAN interface 310. Such operations can usually be automated. Further, in some embodiments, the WAN interface 310 can connect multiple server systems 300 to each other, providing a scalable system capable of managing a large number of activities. Conventional or other techniques for managing server systems and server farms (a collection of collaborative server systems) can be used, including dynamic resource allocation and reallocation.

[0049] The server system 300 may interact with various user-owned or user-operated devices via a wide area network such as the Internet. Examples of user-operated devices are Figure 3 3 is shown as client computing system 314. Client computing system 314 can be implemented as, for example, a consumer device such as a smartphone, other mobile phone, tablet computer, wearable computing device (e.g., smart watch, glasses), desktop computer, laptop computer, etc.

[0050] The client computing system 314 may communicate via the WAN interface 310. The client computing system 314 may include conventional computer components, such as a processing unit 316, a storage device 318, a network interface 320, a user input device 322, and a user output device 324. The client computing system 314 may be a computing device implemented in a variety of form factors, such as a desktop computer, a laptop computer, a tablet computer, a smartphone, other mobile computing devices, a wearable computing device, and the like.

[0051] Processor 316 and storage device 318 may be similar to processing unit 304 and local memory 306 described above. Appropriate devices may be selected based on the requirements to be placed on client computing system 314; for example, client computing system 314 may be implemented as a "thin" client with limited processing capabilities or as a high-performance computing device. Client computing system 314 may be equipped with program code executable by processing unit 316 to implement various interactions with server system 300 of the message management service, such as accessing messages, performing actions on messages, and other interactions described above. Some client computing systems 314 may also interact with the messaging service independently of the message management service.

[0052] The network interface 320 may provide a connection to a wide area network (e.g., the Internet) to which the WAN interface 310 of the server system 300 is also connected. In various embodiments, the network interface 320 may include a wired interface (e.g., Ethernet) and / or a wireless interface implementing various RF data communication standards such as Wi-Fi, Bluetooth, or cellular data network standards (e.g., 3G, 4G, 5G, LTE, etc.).

[0053] User input device 322 may include any device (or devices) via which a user may provide a signal to client computing system 314; client computing system 314 may interpret the signal as indicating a particular user request or information. In various embodiments, user input device 322 may include any or all of a keyboard, touch pad, touch screen, mouse or other pointing device, scroll wheel, click wheel, dial, button, switch, keypad, microphone, etc.

[0054] The user output device 324 may include any device via which the client computing system 314 can provide information to a user. For example, the user output device 324 may include a display to display images generated by or transmitted to the client computing system 314. The display may incorporate various image generation technologies, such as a liquid crystal display (LCD), a light emitting diode (LED) including an organic light emitting diode (OLED), a projection system, a cathode ray tube (CRT), etc., and supporting electronic devices (e.g., digital-to-analog or analog-to-digital converters, signal processors, etc.). Some embodiments may include devices such as a touch screen that serves as both an input device and an output device. In some embodiments, other user output devices 324 may be provided in addition to or in place of a display. Examples include indicator lights, speakers, tactile "display" devices, printers, etc.

[0055] Some embodiments include electronic components, such as microprocessors, storage and memory, that store computer program instructions in computer-readable storage media. Many features described in this specification may be implemented as processes designated as a group of program instructions encoded on a computer-readable storage medium. When these program instructions are executed by one or more processing units, they cause the processing units to perform the various operations indicated in the program instructions. Examples of program instructions or computer codes include machine codes such as those produced by a compiler and files comprising high-level codes executed by a computer, electronic component or microprocessor using an interpreter. Through appropriate programming, processing units 304 and 316 may provide various functions for server system 300 and client computing system 314, including any functions performed by a server or client described herein, or other functions associated with a message management service.

[0056] It should be understood that the server system 300 and the client computing system 314 are illustrative, and variations and modifications are possible. The computer system used in conjunction with the embodiments of the present disclosure may have other capabilities not specifically described herein. Further, although the server system 300 and the client computing system 314 are described with reference to specific frames, it should be understood that these frames are defined for convenience of description and are not intended to imply a specific physical arrangement of components. For example, different frames may but need not be located in the same facility, in the same server rack, or on the same motherboard. In addition, the frames do not need to correspond to physically different components. The frames may be configured to perform various operations, for example, by programming a processor or providing appropriate control circuits, and various frames may or may not be reconfigurable depending on how the initial configuration is obtained. The embodiments of the present disclosure may be implemented in various devices including electronic devices implemented using any combination of circuits and software.

[0057] Example Method

[0058] Analysis: A secondary analysis of the Single Nucleotide Polymorphism (SNP)-Based Microdeletion and Aneuploidy Registry (SMART) was performed. SMART was designed to evaluate the performance of SNP-based NIPS in a general population of pregnant women screened as part of clinical care. Twenty-one centers in six countries participated between April 2015 and January 2019. The study was limited to singleton pregnancies, ≥9 weeks of gestation, and maternal age ≥18 years. Data on patient characteristics, NIPS results, serum screening results, ultrasound, and pregnancy outcomes were collected by local study coordinators.

[0059] Study population: Reference Figure 4 , a total of 20,887 women participated in SMART. For this analysis, pregnancies were excluded if there was a fetal chromosomal abnormality or major structural abnormality, termination of pregnancy, missing data on pregnancy outcomes, participant withdrawal from the study (n=1604), or if data on preeclampsia status or gestational age at delivery were missing (n=365). Participants were further excluded if they had missing data on any of the variables required for model development. The final cohort used for model development included 17,520 participants: 671 with preeclampsia and 16,849 without preeclampsia. Of the participants with preeclampsia, 251 developed preterm preeclampsia, 72 developed early-onset preeclampsia, and 420 developed term preeclampsia. Preeclampsia is a pregnancy-related condition that includes high blood pressure with or without organ dysfunction (e.g., liver and kidney dysfunction), symptoms (e.g., headache and visual disturbances), and seizures in severe cases (“eclampsia”). The presence of “preeclampsia” was based on the clinical diagnosis of preeclampsia in the participant’s medical record.

[0060] Patient characteristics: Patient characteristics examined in the analysis included maternal age, body mass index (BMI), maternal height, maternal weight, country, race / ethnicity, parity, use of in vitro fertilization, chronic hypertension, pregestational diabetes, and smoking during pregnancy. Characteristics obtained from SNP-based NIPS included total cfDNA and fetal fraction (FF). Patient characteristics were compared between the groups of patients with and without preeclampsia. Comparisons of continuous data were performed using the Wilcoxon rank-sum test, and comparisons of independent categorical data were performed using the chi-square test. P values ​​were adjusted for multiple comparisons using the Holm method.

[0061] Outcome measures: The primary outcome was preterm preeclampsia, defined as preeclampsia with delivery <37 weeks' gestation. Secondary outcomes were early-onset preeclampsia (preeclampsia with delivery <34 weeks' gestation) and term preeclampsia (preeclampsia with delivery ≥37 weeks' gestation).

[0062] Features in the model: Variables entered into the prediction model included 13 patient characteristics available in the SMART study data and identified as risk factors for developing preeclampsia based on the existing medical literature (Table 1). Two example features obtained from SNP-based cfDNA screening (total cfDNA and FF) were also included. FF was estimated by using a probabilistic model for SNPs considered homozygous for the maternal allele. This probabilistic model used maximum likelihood to determine the fetal fraction value that maximized the likelihood function based on the observed minor allele frequency. Two features obtained from SNP-based cfDNA screening were also included: total cfDNA (maternal and fetal) and FF. Six of the 15 features were coded as continuous variables; the remaining 10 features were coded as binary (Table 1).

[0063] Table 1: Description, coded name and type of input variables used in the prediction model. Description with coded name

[0064]

[0065]

[0066] Including race and ethnicity in the prediction model is likely to perpetuate racial inequality. Therefore, race and ethnicity were not included in the model development of this example embodiment. However, in view of recent studies supporting the importance of race in the prediction of pre-eclampsia, a sensitivity analysis was performed to evaluate the impact of incorporating race / ethnicity on the accuracy of model prediction. Ethnic information was added to the feature set in Table 1 as a binary marker. Four racial groups (blacks, Asians, Latinos, and whites) were modeled using the neural network model development and cross-validation protocols described in this method section. For example, for the black group, a binary marker was added to the feature list in Table 1. For samples listed as "black" ethnicity, the marker was set to one, and for all other samples, the marker was set to zero. Each group was explored in turn in this way, and the receiver operating characteristic area under the curve (AUROC) performance index was calculated for each iteration. In some specific implementations, a general protocol based on stratified, repeated k-fold cross validation (SRKFCV) can be used, for example, to train and test (validate) the model.

[0067] Machine Learning Models for Classification: First, a linear neural network (LNN) model was considered as a “reference classifier”. The LNN model is a neural network architecture that directly connects input features to a single sigmoid output neuron. In addition, multiple feed-forward dense neural network models (NLNN models) were considered, gradually adding one or more hidden layers of rectified linear unit (reLu) units to the LNN model. The performance of the classifier was evaluated based on the following multiple indicators: AUC score, sensitivity, positive predictive value, and negative predictive value at different values ​​of the screening positivity indicator.

[0068] Training and internal validation: Reference Figure 5 , the example training and validation process 500 includes a series of steps discussed below. Figure 5 All neural network models discussed in the example implementation are implemented and optimized in keras.tensorflow using binary cross entropy as the loss function. The network parameters are optimized by stochastic gradient descent with a constant learning rate of 1e-2 and zero momentum.

[0069] At block 505, the model parameter accumulator and the score accumulator are initialized. The model parameter accumulator and the score accumulator are data structures in which the results are stored at each iteration of the loop discussed below (blocks 515 to 550). At block 510, a variable i is initialized to indicate an iteration (here, the first iteration) of a series of steps that can be looped until a value (n) is reached.

[0070] At box 515, there is a stratified split where the data set is randomly split into a training data set and a validation data set. In the example implementation, the split produces a ratio of 0.9 and 0.1 corresponding to training and validation, respectively (i.e., 90% for training and 10% for testing). Here, the random split is also stratified according to the outcome.

[0071] At box 520, a training data set and a test data set are generated, wherein steps are taken to account for imbalanced data. In the study population, only 1.4% develop preterm preeclampsia, so a machine learning model applied to such a data set is typically expected to over-predict the majority class (i.e., patients who do not have preterm preeclampsia) and be relatively inefficient in predicting "rare" events associated with the minority class (i.e., patients who have preterm preeclampsia). In view of this, after the training and validation split (515), oversampling is applied to the positive samples in the training set via uniform random sampling with replacement, resulting in a training set with an equal number of positive and negative samples.

[0072] At block 525, example process 500 includes steps of neural network scaling and noise injection. A standard scaler is fit to the oversampled training dataset to appropriately scale the training and validation datasets. Gaussian noise with a mean equal to zero and a standard deviation equal to 1e-1 is added to the normalized training dataset. The standard scaler and Gaussian noise are only applied to the continuous input features (Table 1). Noise injection is used both as a form of regularization and data augmentation.

[0073] At block 530, the model is fit on the oversampled training dataset. Then, using stochastic gradient descent and a constant learning rate of 1e-2, the fitting process is performed over 100 epochs and a mini-batch size of 32.

[0074] At block 535, relevant performance metrics are calculated based on the validation dataset. Specifically, in addition to the sensitivity, positive predictive value, and negative predictive value for a fixed screening positive rate, the prediction performance of the two classifiers is evaluated by calculating the area under the receiver operating characteristic (AUC) curve.

[0075] At block 540, the model parameters are stored in the model parameter accumulator, and the scores are stored in the score accumulator. To repeat the steps in the loop n times, at 545, it is determined whether i (initially equal to 1) is equal to n. If not (e.g., if i < n), then at block 550, i is incremented upward, and process 500 loops back to block 515. If so (e.g., if i = n), then process 500 ends at block 555.

[0076] Internal validation: To minimize or otherwise reduce the bias inherent in a single instance of the random split at block 515, blocks 515 through 530 are repeated 200 times (n = 200) in example process 500. This Monte Carlo cross-validation generates 200 estimates for each of the performance metrics of interest, corresponding to 200 different random training-validation splits. Thus, the metrics collected are actually random variables and can be analyzed as such. For each metric, the mean and its 95% confidence interval are considered to evaluate the "expected behavior" of the prediction model.

[0077] Network regularization: In addition to the above noise injection (block 525), L2 penalties with magnitudes varying between 0.0 and 1e-2 are introduced in the example model via the kernel_regularization term to introduce regularization.

[0078] Hyperparameter optimization: Additionally, an incremental (albeit partial) grid-based hyperparameter search is performed. First, the number of epochs and the batch size are considered and fixed. Second, the number and size of the hidden layers are explored, as well as three regularization methods: the magnitude of the L2 penalty in kernel regularization, the percentage of layer dropout, and the standard deviation of Gaussian noise injection in the training dataset.

[0079] Model performance: Model performance was assessed by expected area under the curve (AUC) scores and sensitivities at fixed screen-positive rates of 10%, 15%, and 20% for the primary and secondary outcomes.

[0080] Characteristics of the study population: Again refer to Figure 4 Of the 17,520 participants included in the final analysis, 671 (3.8%) developed preeclampsia. Of these, 251 (1.4%) developed preterm preeclampsia, 72 (0.4%) developed early-onset preeclampsia, and 420 (2.4%) developed term preeclampsia. The characteristics of the 17,520 participants (671 with preeclampsia and 16,849 without preeclampsia) included in model development are shown in Table 2. The maternal age distribution was similar in the group with and without preeclampsia (34.3 years vs. 34.7 years; p = 0.022). In the preeclampsia group, BMI greater than 30 kg / m 2 Results: A significantly higher proportion of participants who developed preeclampsia (40.8% vs. 18.8%; p<0.001) were in the U.S. Among participants, nulliparous, black, or with a history of preterm birth, stillbirth, cesarean section, chronic hypertension, or pregestational diabetes were overrepresented in the group that developed preeclampsia. The median gestational age at the time of cfDNA measurement was 12.6 weeks, and n=16,653 (95.1%) patients had cfDNA measurement at <20 weeks of gestation. cfDNA parameters also differed significantly in the groups that developed preeclampsia. In the preeclampsia group, total cfDNA was significantly higher (median, 362.3 DNA copies / ml vs. 339.0 DNA copies / ml; p<0.001), and FF was significantly lower (median, 7.5% vs. 9.4%; p<0.001) (Table 2).

[0081] Table 2: Baseline demographic and clinical characteristics by pre-eclampsia for participants with complete data (modeling cohort)

[0082]

[0083]

[0084]

[0085]

[0086] Model Selection and Performance: Example Implementations A relatively large number of dense feedforward neural network NLNN models were analyzed and optimized using the above protocol. In more detail, NLNN models with the number of hidden layers varying from 1 to 5 and the number of neurons varying from 2 to 1024 were considered. On these NLNN models, the example implementations tested different regularization strategies, such as L2 kernel regularization, dropout layers, and noise injection in the training set. In some example implementations, the NLNN models performed similarly to linear neural network models.

[0087] In an exemplary implementation, the NLNN model can be a neural network with one hidden layer consisting of 15 reLU units (so that the hidden layer has the same dimension as the input) and with L2 regularization implemented as the kernel_regularization term of keras.

[0088] Table 3 gives the expected AUC score of each outcome and the sensitivity of fixed screening positive rate 10%, 15% and 20%.The numbers shown are the mean values ​​obtained from the optimization protocol described in the method, and represent the expected behavior of the model to unseen data.LNN and NLNN model classifiers have similar prediction performance across each of the pre-eclampsia outcomes (early-onset, premature and full-term pre-eclampsia).The AUC scores of premature, premature and full-term pre-eclampsia are 0.80, 0.79 and 0.71 for the NLNN model, and 0.80, 0.78 and 0.71 (table 3) for the LNN model.The sensitivity to the main outcome (premature pre-eclampsia) is the highest, when the screening positive rate is 15%, the NLNN model predicts 59% of the positive cases, and the LNN model predicts 58% of the positive cases (table 3).In some specific implementations, no differences are observed in the prediction performance of LNN and NLNN models for secondary outcomes (tables 4 and 5).

[0089] Table 3: Sensitivity and AUC scores and 95% confidence intervals of LNN (linear neural network) and NLNN (non-linear neural network) models for predicting preterm preeclampsia (<37 weeks), early-onset preeclampsia (<34 weeks) and term preeclampsia (>=37 weeks)

[0090]

[0091]

[0092] Table 4: Positive predictive value (PPV) and 95% confidence interval of LNN (linear neural network) and NLNN (non-linear neural network) models for predicting preterm pre-eclampsia (<37 weeks), early-onset pre-eclampsia (<34 weeks) and term pre-eclampsia (>=37 weeks). The prevalence of preterm pre-eclampsia, early-onset pre-eclampsia and term pre-eclampsia was 0.015, 0.005 and 0.04, respectively.

[0093]

[0094]

[0095] Table 5: Negative predictive values ​​of LNN (linear neural network) and NLNN (nonlinear neural network) models for predicting the primary outcome (preterm preeclampsia) and secondary outcomes (early-onset preeclampsia and term preeclampsia). The prevalence of preterm preeclampsia, early-onset preeclampsia, and term preeclampsia were 0.015, 0.005, and 0.04, respectively.

[0096]

[0097]

[0098] To evaluate the contribution of cfDNA features to the performance of the example models in various implementations, additional training validation runs were performed on a simplified version of the dataset in which the FF and total cfDNA features were omitted. In these neural network models, the number of units in the hidden layer was rescaled to 13 to match the dimensionality of the input layer. In the exemplary implementation, for early-onset preeclampsia, removing total cfDNA and FF features from the model was associated with a decrease in sensitivity of approximately 7% at a screen-positive rate of 15% for both the LNN and NLNN models (Table 6).

[0099] Table 6: Model performance of LNN and NLNN models with and without cfDNA features. Models omitting the cfDNA feature are named "Simplified". "Full" refers to models trained and optimized for all 15 features in Table 2.

[0100]

[0101]

[0102] The contribution of race-related information was evaluated by another set of models (see Methods for details), and the AUC scores and their 95% confidence intervals in the exemplary implementation were estimated (the results are listed in Table 7). After comparing these results with the results listed in Tables 3, 4, and 5 (where race was not considered), race-related information did not appear to contribute significantly to the predictive performance of the model.

[0103] Table 7: Expected AUC scores for linear and nonlinear neural network (LNN and NLNN) models extended to account for ethnicity information.

[0104]

[0105]

[0106] Feature Importance: In the exemplary implementation, and with reference to Figure 6 and Figure 7 The input features that contributed most to the predictions of the LNN and NLNN models were maternal weight, chronic hypertension, pre-gestational diabetes, previous preterm birth, and previous stillbirth. In both models, parity and increased height were strongly negatively associated with preterm pre-eclampsia. When the cfDNA feature was omitted from the model, the remaining 13 features were adjusted to compensate for their absence ( Figure 7 ). Of note, a larger relative contribution of the IVF signature (in vitro fertilization) was observed for early-onset preeclampsia, whereas a smaller relative contribution of the predictive signature associated with pregestational diabetes (BDIABTYP_PRE) was observed after omitting the cfDNA input variable. It is worth mentioning that the use of matrix multiplication (see the Methods section) to evaluate the contribution of input variables to the NLNN model in the presence of reLu latent units is reasonable, based on the fact that in the example optimized model, all parameters of the “hidden” activation function in the optimized model are non-negative. Figure 7 A similar bar graph constructed based on the results obtained from a simplified model omitting the cfDNA signature is shown, demonstrating how the relative contributions of the remaining 13 signatures “adapt” to compensate for the absence of the missing cfDNA signature. The same graph for term pre-eclampsia shows a similar trend, while for the early-onset pre-eclampsia outcome, a larger relative contribution of the IVF signature (in vitro fertilization) and a smaller relative contribution of the predictive signature associated with pre-pregnancy diabetes (BDIABTYP_PRE) are observed.

[0107] The exemplary implementations discussed above show that, in various embodiments, both linear and nonlinear neural network models using patient characteristics collected at the first prenatal visit and features available for SNP-based cfDNA screening (total cfDNA and FF) can be used to predict preterm preeclampsia with good performance levels (expected AUC score = 0.80, 58% sensitivity for a 15% screening positive rate). These exemplary models are able to use only routinely collected clinical data (mostly binary data), which has practical and economic advantages for clinical implementation.

[0108] In various alternative implementations, mean arterial pressure, when added to the model in addition to the other features discussed, or in place of one or more of the features discussed, can significantly improve the predictive performance for preterm preeclampsia. Blood pressure is not collected as part of SMART, but is collected early in pregnancy as part of routine clinical practice and can be incorporated into alternative implementations of the disclosed model.

[0109] In the disclosed analysis, cfDNA features (total cfDNA and FF) were all associated with the risk of pre-eclampsia. In various embodiments, FF and cfDNA were particularly helpful for model performance of early-onset pre-eclampsia. This finding of biomarker performance of early-onset disease may reflect the different pathogenic processes of early-onset and late-onset pre-eclampsia.

[0110] Preferably, example implementations of the disclosed methods develop predictive models based solely on easily acquired, routinely collected clinical data and cfDNA data from non-invasive prenatal screening, all of which are part of routine clinical care in early pregnancy in the U.S. Without the need for additional biomarkers (e.g., maternal serum biomarkers) and ultrasound (e.g., uterine artery Doppler), implementations of the disclosed methods have the advantage of maximizing routinely collected input features without adding the expense of biomarkers and ultrasound that are not part of the clinical pathway for current pregnancy care in the U.S. Thus, the disclosed methods can be incorporated into clinical care with minimal resource use and minimal inconvenience to patients.

[0111] Although the present disclosure has been described with respect to specific embodiments, those skilled in the art will recognize that a variety of modifications are possible. For example, although specific examples of rules (including trigger conditions and / or result actions) and processes for generating suggested rules are described, other rules and processes may be implemented. Embodiments of the present disclosure may be implemented using a variety of computer systems and communication technologies, including but not limited to the specific examples described herein.

[0112] Embodiments of the present disclosure can be implemented using any combination of dedicated components and / or programmable processors and / or other programmable devices. The various processes described herein can be implemented on the same processor or on different processors in any combination. When a component is described as being configured to perform certain operations, this configuration can be completed, for example, by designing an electronic circuit to perform the operation, by programming a programmable electronic circuit (such as a microprocessor) to perform the operation, or any combination thereof. In addition, although the above-described embodiments can refer to specific hardware and software components, it will be appreciated by those skilled in the art that different combinations of hardware and / or software components can also be used, and specific operations described as being implemented in hardware can also be implemented in software, and vice versa.

[0113] Computer programs incorporating various features of the present disclosure may be encoded and stored on a variety of computer-readable storage media; suitable media include disks or tapes, optical storage media such as compact disks (CDs) or DVDs (digital versatile disks), flash memory, and other non-transitory media. Computer-readable media encoded with program code may be packaged with a compatible electronic device, or the program code may be provided separately from the electronic device (e.g., via Internet download or as a separately packaged computer-readable storage medium).

[0114] Therefore, although the disclosure has been described with respect to particular embodiments, it should be understood that the disclosure is intended to cover all modifications and equivalents within the scope of the appended claims.

[0115] As used herein, a "routine" prenatal visit to a health care provider is a "checkup" type visit to a health care provider for the purpose of maintaining health and wellness. The purpose of a routine visit is to prevent adverse health outcomes, not to treat existing symptoms or illness. Routine prenatal visits are recommended for the general pregnant population, not for a patient's specific health concerns.

[0116] The following include potential embodiments of the disclosed methods, which are representative of other examples that are not intended to be limiting in any way:

[0117] Embodiment A1: A method for generating a prediction of a patient's development of pre-eclampsia during a current pregnancy by applying a predictive machine learning classifier by a computing system, the predictive machine learning classifier comprising one or more artificial neural networks, wherein generating the prediction comprises: obtaining patient data by at least one of the following: (A) performing non-invasive prenatal screening (NIPS) on the patient, or (B) receiving data from a health record system by the computing system via a communication network, wherein the patient data comprises (i) a set of health characteristics, and (ii) one or more cell-free DNA (cfDNA) indicators; feeding the patient data to the predictive machine learning classifier to obtain a prediction of the patient's development of pre-eclampsia during the current pregnancy; and the computing system Providing the prediction for use in the care of the patient, wherein providing the prediction comprises at least one of: transmitting, by the computing system, the prediction to at least one of the health record system or a computing device associated with a healthcare provider over the communications network; wherein the predictive machine learning classifier is trained by applying one or more deep learning techniques to data about subjects in a cohort of pregnant subjects, the data comprising, for each pregnant subject: (A) a first feature set indicating one or more outcomes of one or more prior pregnancies of the pregnant subject; (B) a second feature set comprising one or more health indicators of the pregnant subject; and (C) a third feature set based on one or more cfDNA measurements of the pregnant subject.

[0118] Embodiment A2: The method of Embodiment A1, wherein all of the patient data used to obtain the prediction is collected non-invasively during one or more routine medical visits to the patient.

[0119] Embodiment A3: The method of Embodiment A1 or A3, wherein the patient data used to obtain the prediction does not include ultrasound data.

[0120] Embodiment A4: The method according to any one of Embodiments A1 to A3, wherein the pre-eclampsia condition is early-onset preterm pre-eclampsia.

[0121] Embodiment A5: The method according to any one of embodiments A1 to A4, further comprising generating, by a computing system, a training data set based on the first, second and third feature sets.

[0122] Embodiment A6: The method of any one of Embodiments A1 to A5, further comprising training, by a computing system, a predictive machine learning classifier using data about subjects in a cohort of pregnant subjects.

[0123] Embodiment A7: A method according to any one of Embodiments A1 to A6, wherein the predictive machine learning classifier comprises one or more rectified linear units (reLu).

[0124] Embodiment A8: A method according to any one of Embodiments A1 to A7, wherein the predictive machine learning classifier includes a hidden layer having the same dimension as the input layer.

[0125] Embodiment A9: A method according to any one of Embodiments A1 to A8, wherein applying the one or more deep learning techniques includes using one or more regularization terms.

[0126] Embodiment B1: A method, the method comprising generating a predictive machine learning classifier for pre-eclampsia in a pregnant patient, the predictive machine learning classifier comprising one or more artificial neural networks, wherein generating the predictive machine learning classifier comprises: generating a training data set by a computing system using data about subjects in a cohort of pregnant subjects, wherein for each pregnant subject, the training data set comprises: (A) a first feature set indicating one or more outcomes of one or more previous pregnancies of the subject; (B) a second feature set corresponding to one or more health indicators of the subject; and (C) a third feature set corresponding to one or more cfDNA indicators; the computing system converting the one or more deep The method further comprises applying a deep learning technique to the training data set to generate the predictive machine learning classifier, wherein the predictive machine learning classifier is configured to receive patient health characteristics and one or more cfDNA indicators as input, and output a prediction of the occurrence of the pre-eclampsia condition; and providing the predictive machine learning classifier by the computing system to generate patient-specific indicators for individual patients, wherein providing the predictive machine learning classifier comprises at least one of the following: (i) transmitting the predictive machine learning classifier by the computing system to a second computing system via a communication network, or (ii) storing the predictive machine learning classifier in a non-transitory computer-readable storage medium that can be obtained by a healthcare provider for subsequent use.

[0127] Embodiment B2: The method according to embodiment B1, wherein the predictive machine learning classifier includes one or more implicit rectified linear units (reLu).

[0128] Embodiment B3: A method according to embodiment B1 or B2, wherein the predictive machine learning classifier includes a hidden layer having the same dimension as the input layer.

[0129] Embodiment B4: The method according to any one of Embodiments B1 to B3, wherein applying the one or more deep learning techniques includes using one or more regularization terms.

[0130] Embodiment B5: The method of any one of Embodiments B1 to B4, wherein the one or more regularization terms include an L2 regularization term.

[0131] Embodiment B6: The method of any one of Embodiments B1 to B5, further comprising using the predictive machine learning classifier to evaluate a patient during a current pregnancy by: performing non-invasive prenatal screening (NIPS) on the patient; feeding data to the predictive machine learning classifier to obtain a prediction of the occurrence of pre-eclampsia during the current pregnancy, the data comprising data from NIPS; and using the prediction to provide medical care to the patient, wherein all of the data provided to the predictive machine learning classifier to obtain the prediction is collected non-invasively during one or more routine medical visits to the patient.

[0132] Embodiment B7: A method according to any one of Embodiments B1 to B6, further comprising using the predictive machine learning classifier to evaluate the patient during the current pregnancy by: receiving data corresponding to one or more previous pregnancies, one or more health characteristics, and one or more cfDNA indicators of the patient from a health record system; feeding the data to the predictive machine learning classifier to obtain a prediction of the occurrence of pre-eclampsia during the current pregnancy; and using the prediction to provide medical care to the patient, wherein all of the data input into the predictive machine learning classifier to obtain the prediction is collected non-invasively during one or more routine medical visits to the patient.

[0133] Embodiment B8: The method of any one of Embodiments Bl to B7, wherein the preterm pre-eclampsia condition is early-onset preterm pre-eclampsia.

[0134] Embodiment C1: A computing system configured to apply a predictive machine learning classifier to generate a prediction of a patient developing a pre-eclampsia condition during a current pregnancy, the predictive machine learning classifier comprising one or more artificial neural networks, the computing system comprising one or more processors, the one or more processors configured to: receive patient data over a communication network during the current pregnancy of the patient, the patient data comprising a set of health characteristics and one or more cfDNA measurements; feed the patient data to the predictive machine learning classifier to obtain a prediction of the patient developing a pre-eclampsia condition during the current pregnancy; and provide the prediction for the patient's care, wherein providing the prediction comprises transmitting the prediction to at least one of a health record system or a computing device associated with a healthcare provider over a communication network; wherein the predictive machine learning classifier is trained by applying one or more deep learning techniques to data about subjects in a cohort of pregnant subjects, and for each pregnant subject, the training data comprises: (A) a first feature set indicating one or more outcomes of one or more previous pregnancies of the pregnant subject; (B) a second feature set corresponding to one or more health indicators of the pregnant subject; and (C) a third feature set corresponding to one or more cell-free DNA indicators of the pregnant subject.

[0135] Embodiment C2: The computing system of Embodiment C1, wherein all of the patient data used to obtain the prediction is collected non-invasively during one or more routine medical visits of the patient.

[0136] Embodiment C3. The computing system of Embodiment C1 or C2, wherein the preterm pre-eclampsia condition is early-onset preterm pre-eclampsia.

[0137] Embodiment C4: A computing system according to any one of embodiments C1 to C3, wherein the one or more processors are further configured to generate the training data set and apply the deep learning technology to the training data set to generate the predictive machine learning classifier.

[0138] Embodiment C5: The computing system of any one of Embodiments C1 to C4, wherein the one or more processors are further configured to receive at least a subset of the health data from the health record system via the communication network.

[0139] Embodiment C6: The computing system of any one of Embodiments C1 to C5, wherein the one or more processors are further configured to receive at least a subset of the health data via a software application running on the computing device associated with the healthcare provider.

[0140] Embodiment D1: A method for developing a machine learning model that can be used to predict the occurrence of preterm preeclampsia, comprising: identifying, by one or more processors, data for each of a plurality of subjects including maternal age, maternal weight, gestational age, total cell-free DNA, and at least three of the fetal score percentiles of the fetal score distribution relative to the maternal weight and gestational age corresponding to the sample, the data further comprising a first marker indicating that the subject has developed preterm preeclampsia or one of a second marker indicating that the subject has not developed preterm preeclampsia; and training a machine learning model using the identified data by the one or more processors to generate a value indicating the risk of the subject developing preterm preeclampsia.

[0141] Embodiment D2: A method according to Embodiment D1, wherein the machine learning model includes at least one of a regularized logistic regression model or a feed-forward neural network.

[0142] Embodiment D3: A method according to embodiment D1 or D2, wherein the model comprises a binary classifier.

[0143] Embodiment D4. A method according to any one of embodiments D1 to D3, wherein a stratified, repeated k-fold cross-validation (SRKFCV) method is used to train and validate the model.

[0144] Embodiment D5. A method according to any one of embodiments D1 to D4, wherein the machine learning model is trained in the following manner: a) initializing a model parameter accumulator and a score accumulator; b) assigning a first part of the data set to a training data set, and assigning a second part of the data set to a validation data set; c) oversampling the minority class in the training data set to generate an oversampled training data set d) oversampling the minority class in the validation data set to generate an oversampled validation data set; e) fitting the machine learning model on the oversampled training data set; f) generating a predicted outcome on the oversampled validation data set; g) determining an accuracy score and an area under the curve (AUC) score based on the predicted outcome; and h) storing one or more model parameters in the model parameter accumulator and the accuracy score and AUC score in the score accumulator in one or more data structures.

[0145] Embodiment D6. The method of embodiment D5, wherein steps b) to h) are repeated a predetermined number of times.

[0146] Embodiment D7. A method according to embodiment D5, wherein the machine learning model is logistic regression and the regularization constant is a hyperparameter.

[0147] Embodiment D8. A method according to embodiment D5, wherein the machine learning model is a feedforward neural network, and the regularization constant, activation function, number and size of hidden layers, and cost function are optimized.

[0148] Embodiment E1. A method for predicting the occurrence of preterm preeclampsia, comprising: identifying, by one or more processors, input data for a subject including at least three of maternal age, maternal weight, gestational age, total cell-free DNA, and fetal score percentiles of a fetal score distribution relative to the maternal weight and gestational age of the sample; in response to inputting the input data into a machine learning model trained using the data to generate a value indicating the risk of the subject developing preterm preeclampsia, determining, by one or more processors, a value indicating the risk of the subject developing preterm preeclampsia; and providing, by one or more processors, an output based on the value for presentation.

[0149] Embodiment E2: The method of Embodiment E1, wherein the machine learning model comprises at least one of a regularized logistic regression model or a feed-forward neural network.

[0150] Embodiment E3: The method of Embodiment E1 or E2, wherein the model comprises a binary classifier.

[0151] Embodiment E4: The method according to any one of Embodiments E1 to E3, wherein a stratified, repeated k-fold cross-validation (SRKFCV) method is used to train and validate the model.

[0152] Embodiment E5: A method according to any one of Embodiments E1 to E4, wherein the machine learning model is trained in the following manner: a) initializing a model parameter accumulator and a score accumulator; b) assigning a first part of the data set to a training data set, and assigning a second part of the data set to a validation data set; c) oversampling the minority class in the training data set to generate an oversampled training data set; d) oversampling the minority class in the validation data set to generate an oversampled validation data set; e) fitting the machine learning model on the oversampled training data set; f) generating a predicted outcome on the oversampled validation data set; g) determining an accuracy score and an area under the curve (AUC) score based on the predicted outcome; and h) storing one or more model parameters in the model parameter accumulator and the accuracy score and AUC score in the score accumulator in one or more data structures.

[0153] Embodiment E6: The method according to Embodiment E5, wherein steps b) to h) are repeated a predetermined number of times.

[0154] Embodiment E7: A method according to Embodiment E5, wherein the machine learning model is logistic regression and the regularization constant is a hyperparameter.

[0155] Embodiment E8: A method according to Embodiment E5, wherein the machine learning model is a feedforward neural network, and the regularization constant, activation function, number and size of hidden layers, and cost function are optimized.

[0156] Embodiment F: A computing system comprising one or more processors configured to implement any of Embodiments A1-B8 or D1-E8.

[0157] Embodiment E: A non-transitory computer-readable storage medium having instructions configured to cause one or more processors of a computing system to implement any of Embodiments A1 to B8 or D1 to E8.

[0158] The technology of the present invention is not limited to the specific embodiments described in this application, and these specific embodiments are intended to be separately illustrated as a single aspect of the technology of the present invention. Without departing from the spirit and scope of the technology of the present invention, many modifications and changes can be made to the technology of the present invention, which will be apparent to those skilled in the art. According to the above description, in addition to those methods and devices listed herein, functionally equivalent methods and devices within the scope of the technology of the present invention will be apparent to those skilled in the art. Such modifications and changes are intended to fall within the scope of the technology of the present invention. It should be understood that the technology of the present invention is not limited to specific methods, reagents, compounds, compositions or biological systems, and of course, the specific methods, reagents, compounds, compositions or biological systems may vary. It should also be understood that the terms used herein are only for the purpose of describing specific embodiments and are not intended to be limited.

[0159] In addition, where features or aspects of the disclosure are described in terms of Markush groups, those skilled in the art will recognize that the disclosure is also thereby described in terms of any individual member or subgroup of members of the Markush group.

[0160] As will be understood by those skilled in the art, for any and all purposes, particularly with respect to providing a written description, all ranges disclosed herein also encompass any and all possible sub-ranges and combinations of sub-ranges thereof. Any listed range can be easily considered to fully describe and enable the same range to be decomposed into, for example, at least equal halves, thirds, quarters, fifths, tenths, etc. As a non-limiting example, each range discussed herein can be easily divided into a lower third, a middle third, and an upper third, etc. As will be understood by those skilled in the art, all terms, such as "until", "at least", "greater than", "less than", and etc., include the numbers listed and refer to the ranges that can be subsequently decomposed into sub-ranges as described above. Finally, as will be understood by those skilled in the art, a range includes each individual member. Thus, for example, a group having 1-3 cells refers to a group having 1, 2, or 3 cells. Similarly, a group having 1-5 cells refers to a group having 1, 2, 3, 4, or 5 cells, etc.

[0161] All patents, patent applications, provisional applications, and publications mentioned or cited herein are hereby incorporated by reference in their entirety, including all figures and tables, to the extent they are not inconsistent with the explicit teachings of this specification.

Claims

1. A method of applying a predictive machine learning classifier by a computing system to generate a prediction of a patient developing preterm pre-eclampsia during a current pregnancy, the predictive machine learning classifier comprising one or more artificial neural networks, wherein generating the prediction comprises: Obtaining patient data by at least one of: (A) performing non-invasive prenatal screening (NIPS) on the patient, or (B) receiving data from a health record system by the computing system via a communication network, wherein the patient data includes (i) a set of health characteristics, and (ii) one or more cell-free DNA (cfDNA) indicators; feeding the patient data to the predictive machine learning classifier to obtain a prediction of the patient developing the preterm pre-eclampsia during the current pregnancy; as well as providing, by the computing system, the prediction for use in the care of the patient, wherein providing the prediction comprises at least one of: transmitting, by the computing system, the prediction over the communications network to at least one of the health record system or a computing device associated with a healthcare provider; wherein the predictive machine learning classifier is trained by applying one or more deep learning techniques to data regarding subjects in a cohort of pregnant subjects, the data comprising, for each pregnant subject: (A) a first set of features indicative of one or more outcomes of one or more prior pregnancies of the pregnant subject; (B) a second feature set comprising one or more health indicators of the pregnant subject; and (C) a third feature set based on one or more cfDNA measurements of the pregnant subject.

2. The method of claim 1, wherein all of the patient data used to obtain the prediction (i) is collected non-invasively during one or more routine medical visits to the patient and (ii) does not include ultrasound data.

3. The method of claim 1, wherein the preterm preeclampsia is early-onset preterm preeclampsia. 4 . The method of claim 1 , further comprising generating, by the computing system, a training data set based on the first, second, and third feature sets.

5. The method of claim 1, further comprising training, by the computing system, the predictive machine learning classifier using the data about the subjects in the cohort of pregnant subjects.

6. The method of claim 1, wherein the predictive machine learning classifier comprises (i) one or more rectified linear units (reLu), and (ii) a hidden layer having the same dimension as an input layer, wherein the one or more deep learning techniques comprise using one or more regularization terms.

7. A method comprising generating a predictive machine learning classifier for preterm preeclampsia in a pregnant patient, the predictive machine learning classifier comprising one or more artificial neural networks, wherein generating the predictive machine learning classifier comprises: A training data set is generated by a computing system using data about subjects in a cohort of pregnant subjects, the training data set comprising, for each pregnant subject: (A) a first set of features that are indicative of one or more outcomes of one or more prior pregnancies of the subject; (B) a second set of features corresponding to one or more health indicators of the subject; and (C) a third feature set corresponding to one or more cfDNA indicators; applying, by the computing system, one or more deep learning techniques to the training data set to generate the predictive machine learning classifier, wherein the predictive machine learning classifier is configured to receive as inputs a patient health characteristic and one or more cfDNA indicators, and to output a prediction of the occurrence of preterm preeclampsia; and The predictive machine learning classifier is provided by the computing system to generate a patient-specific indicator for an individual patient, wherein providing the predictive machine learning classifier comprises at least one of: (i) transmitting the predictive machine learning classifier by the computing system to a second computing system via a communication network, or (ii) storing the predictive machine learning classifier in a non-transitory computer-readable storage medium that is accessible by a healthcare provider for subsequent use.

8. The method of claim 7, wherein the predictive machine learning classifier comprises one or more implicit rectified linear units (reLu).

9. The method of claim 7, wherein the predictive machine learning classifier comprises a hidden layer having the same dimension as an input layer.

10. The method of claim 7, wherein applying the one or more deep learning techniques comprises using one or more regularization terms. The method of claim 10 , wherein the one or more regularization terms include an L2 regularization term.

12. The method of claim 7, further comprising using the predictive machine learning classifier to assess a patient during a current pregnancy by: performing non-invasive prenatal screening (NIPS) on the patient; feeding data to the predictive machine learning classifier to obtain a prediction of the occurrence of preterm pre-eclampsia during the current pregnancy, the data comprising data from the NIPS; and and providing medical care to the patient using the prediction, wherein all of the data provided to the predictive machine learning classifier to obtain the prediction is collected non-invasively during one or more routine medical visits by the patient.

13. The method of claim 7, further comprising using the predictive machine learning classifier to assess a patient during a current pregnancy by: receiving data corresponding to one or more prior pregnancies, one or more health characteristics, and one or more cfDNA indicators of the patient from a health record system; feeding the data to the predictive machine learning classifier to obtain a prediction of the occurrence of preterm pre-eclampsia during the current pregnancy; as well as and providing medical care to said patient using said prediction, wherein all of said data input into said predictive machine learning classifier to obtain said prediction is collected non-invasively during one or more routine medical visits by said patient.

14. The method of claim 1, wherein the preterm pre-eclampsia is early-onset preterm pre-eclampsia.

15. A computing system configured to apply a predictive machine learning classifier to generate a prediction of a patient developing preterm pre-eclampsia during a current pregnancy, the predictive machine learning classifier comprising one or more artificial neural networks, the computing system comprising one or more processors configured to: receiving, over a communication network, patient data during the current pregnancy of the patient, the patient data comprising a set of health characteristics and one or more cfDNA measurements; feeding the patient data to the predictive machine learning classifier to obtain a prediction of the patient developing the preterm pre-eclampsia during the current pregnancy; as well as providing the prognosis for use in the care of the patient, wherein providing the prognosis comprises transmitting the prognosis over the communication network to at least one of a health record system or a computing device associated with a health care provider; wherein the predictive machine learning classifier is trained by applying one or more deep learning techniques to data about subjects in a cohort of pregnant subjects, and for each pregnant subject, the training data comprises: (A) a first set of features that are indicative of one or more outcomes of one or more prior pregnancies of the pregnant subject; (B) a second feature set corresponding to one or more health indicators of the pregnant subject; and (C) a third feature set corresponding to one or more cell-free DNA indicators of the pregnant subject.

16. The computing system of claim 15, wherein all of the patient data used to obtain the prediction is collected non-invasively during one or more routine medical visits of the patient.

17. The computing system of claim 15, wherein the preterm pre-eclampsia is early-onset preterm pre-eclampsia.

18. The computing system of claim 15, wherein the one or more processors are further configured to generate the training dataset and apply the deep learning technique to the training dataset to generate the predictive machine learning classifier.

19. The computing system of claim 15, wherein the one or more processors are further configured to receive at least a subset of the health data from the health records system via the communication network.

20. The computing system of claim 15, wherein the one or more processors are further configured to receive at least a subset of the health data via a software application running on the computing device associated with the healthcare provider.

Images