Systems and methods for disease prediction

Abstract

A system (100) and method (200) for predicting the presence of a disease in a subject is disclosed. In one aspect, a system (100) includes one or more processing units (101), a medical database (112) coupled to the one or more processing units (101), and a disease prediction module (110). The module (110) is configured to receive a plurality of parameters associated with a subject. Additionally, the module (110) is configured to determine a threshold of sensitivity and / or specificity associated with the trained machine learning model. Further, the module (110) is configured to predict the presence of a disease in the subject using the trained machine learning model based on the desired sensitivity and specificity and the plurality of parameters associated with the subject, and to output the prediction on the output unit (105).

Description

[0001] CROSS-REFERENCE TO RELATED APPLICATIONS

[0002] This application claims priority or benefit under 35 U.S.C. §119 to U.S. Provisional Application No. 63 / 268,435 filed on February 24, 2022, Indian Patent Application No. 202231010035 filed on February 24, 2022, and Indian Patent Application No. 202331002562 filed on January 12, 2023, all of which are incorporated herein by reference in their entirety. Technical Field

[0003] The present disclosure relates generally to the fields of medicine, subject health, sample analysis, and more particularly to the field of prediction of the presence of a disease in a subject. Background Art

[0004] Prediction of disease is important not only in ensuring timely treatment of patients, but also in managing the resources and logistics involved in the disease management process. For example, the epidemic caused by the SARS-CoV-2 virus has increased the burden on health care systems across the world. The number of individuals who are required to undergo testing for the detection of the SARS-CoV-2 virus has accelerated, resulting in a shortage of infectious disease test kits such as reverse transcriptase polymerase chain reaction (RT-PCR) test kits, delays in providing test results to individuals, and delays in providing timely medical support to patients when the SARS-CoV-2 virus is not suspected. In such a scenario, it may be important to predict the presence of one or more pathogens (such as viruses, bacteria, etc. such as SARS-CoV-2, etc.) in a patient.

[0005] Currently, without RT-PCR testing or antigen testing, there is no reliable means by which the presence of a pathogen such as a virus (e.g., SARS-CoV-2 virus) in a subject can be efficiently predicted. There is a continuing need for improved systems and methods for predicting the presence of a pathogen such as a virus or bacteria in a subject and / or detecting increases in infection rates in a population to improve utilization of medical resources. Summary of the invention

[0006] A system for predicting the presence of a disease in a subject is disclosed. In one aspect of the disclosure, the system includes a processing unit and a memory. Additionally, the memory includes a disease prediction module, which is configured to receive a plurality of parameters associated with the subject and determine a threshold of sensitivity and / or specificity associated with a trained machine learning model. Additionally, the module is configured to predict the presence of a disease in a subject and the plurality of parameters associated with the subject using the trained machine learning model, and output the prediction on an output unit.

[0007] In another aspect of the present disclosure, a method for predicting the presence of a disease in a subject is disclosed. The method includes receiving a plurality of parameters associated with the subject. Additionally, the method includes determining a threshold for sensitivity and / or specificity associated with the trained machine learning model. Still further, the method includes predicting the presence of the disease in the subject using the trained machine learning model based on the sensitivity and specificity of the disease and the plurality of parameters associated with the subject. Further, the method includes outputting the prediction on an output unit.

[0008] In an embodiment, the present disclosure includes an article of manufacture comprising a non-transitory computer-readable medium having encoded thereon instructions configured to cause one or more processors to perform a method comprising predicting the presence of a disease according to the present disclosure.

[0009] In an embodiment, the present disclosure includes a non-transitory computer-readable medium having encoded thereon instructions configured to cause one or more processors to perform a method comprising predicting the presence of a disease according to the present disclosure.

[0010] In embodiments, the present disclosure encompasses predicting the presence of a disease in a subject according to the present disclosure, and subsequently treating the subject in need thereof.

[0011] In an embodiment, the present disclosure includes a system for predicting the presence of a disease in a population. In one aspect of the present disclosure, the system includes a processing unit and a memory. Additionally, the memory includes a disease prediction module, which is configured to receive a plurality of parameters associated with one or more subjects and determine a threshold of sensitivity and / or specificity associated with a trained machine learning model. Additionally, the module is configured to predict the presence of a disease in one or more subjects or subject populations and the plurality of parameters associated with the subjects or subject populations using a trained machine learning model, and output the prediction on an output unit.

[0012] This summary is provided to introduce selected concepts in a simplified form that are further described below in the following description. It is not intended to identify features or essential features of the claimed subject matter. Furthermore, the claimed subject matter is not limited to implementations that solve any or all disadvantages noted in any part of this disclosure. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] The invention is further described hereinafter with reference to the illustrated embodiments shown in the accompanying drawings, in which:

[0014] Figure 1 A system for predicting the presence of a disease in a subject according to an embodiment is illustrated.

[0015] Figure 2 A process flow for predicting the presence of a disease in a subject according to an embodiment of the present disclosure is illustrated.

[0016] Figure 3 Receiver operating characteristic (ROC) curves for selecting thresholds for sensitivity and specificity associated with a trained machine learning model for a demand scenario according to an embodiment of the present disclosure are illustrated.

[0017] Figure 4 Depicted is the ROC curve of the 48-parameter ensemble model calculated using the test set of the present disclosure.

[0018] Figure 5 Depicted is model interpretability: features arranged in descending order of importance as computed by the MIMIC LightGBM explainer.

[0019] Fig. 6A , Figure 6B and Figure 6C Use case embodiments such as in an epidemic scenario are depicted, as well as use cases for an endemic scenario at different regions (1, 2, and 3) of the ROC curve. DETAILED DESCRIPTION

[0020] Hereinafter, the systems, methods and products for implementing the embodiments of the present disclosure are described in detail. Various embodiments are described with reference to the accompanying drawings, wherein the same reference numerals are always used to refer to the same elements. In the following description, for the purpose of explanation, many specific details are set forth to provide a thorough understanding of one or more embodiments. It may be apparent that such embodiments can be practiced without these specific details. In other examples, well-known materials or methods are not described in detail to avoid unnecessarily obscuring the embodiments of the present disclosure. Although the present disclosure allows various modifications and alternative forms, its specific embodiments are shown by way of example in the accompanying drawings and will be described in detail herein. However, it should be understood that it is not intended to limit the present disclosure to the specific form disclosed, but on the contrary, the present disclosure is to cover all modifications, equivalents and substitutes that fall within the spirit and scope of the present disclosure. The disclosed embodiments provide systems and methods for predicting diseases in subjects.

[0021] Advantages of embodiments of the present disclosure include excellent and robust prediction of diseases caused by one or more pathogens to ensure, among other things, timely treatment of patients or subjects in need, and improved management of resources and logistics involved in the disease management process. Embodiments of the present disclosure reduce the burden on health care systems across the world, prevent shortages of test kits, and shorten the duration of waiting for individual test results, and shorten the waiting period for medical support for patients when the presence of a pathogen is not suspected.

[0022] COVID-19 testing and antigen testing using RT-PCR (reverse transcriptase-polymerase chain reaction) have been massively expanded since the start of the COVID-19 epidemic, however, there are multiple scenarios in which predictive algorithms based on CBC / Diff can be applied in the clinical value chain in both epidemic and endemic scenarios. First, predictive algorithms based on CBC / Diff can serve as substitutes for such tests in countries where testing is unavailable or in short supply. Second, rapid marking by highly sensitive hematology-based machine learning (ML) algorithms (e.g., see algorithm A described below, with a threshold of 0.41) can reduce the burden of testing in epidemic scenarios. Third, algorithms with high negative predictive values ​​(NPV) (e.g., see algorithm A, with a threshold of 0.33) can reduce the number of patients who undergo testing before procedures such as surgery, thereby reducing the time to care for patients. Fourth, in endemic scenarios, algorithms with high specificity (e.g., see algorithm A, with a threshold of 0.67) can be used to mark asymptomatic individuals, who can then be tested to confirm infection. Thus, the same ML algorithm, such as algorithm A, with different classification thresholds, such as threshold values ​​of, for example, 0.4 to 1.0, 0.4-0.97, 0.4 to 0.95, 0.4-0.7, 0.4-0.6, 0.4-0.5, or 0.4-0.45, 0.80-0.95, 0.99-1.0, may be used in multiple scenarios described herein.

[0023] Specific Definitions

[0024] As utilized in accordance with the present disclosure, unless otherwise indicated, the following terms should be understood to have the following meanings. Unless otherwise apparent from the context, (i) the terms "a" and "an" may be understood to mean "at least one"; (ii) the terms "including" and "comprising" may be understood to cover itemized components or steps, whether presented by themselves or with one or more additional components or steps; and (iii) where a range is provided, the endpoints are included.

[0025] The use of the term "at least one" will be understood to include one and any number more than one, including but not limited to 2, 3, 4, 5, 10, 15, 20, 30, 40, 50, 100, etc. The term "at least one" can extend up to 100 or 1000 or more, depending on the term to which it is attached; in addition, the number of 100 / 1000 is not considered to be limiting, as higher limits can also produce satisfactory results. In addition, the use of the term "at least one of X, Y, and Z" will be understood to include X alone, Y alone, and Z alone, as well as any combination of X, Y, and Z. The use of ordinal terms (i.e., "first," "second," "third," "fourth," etc.) is only for the purpose of distinguishing between two or more items, and is not meant to imply any order or sequence or importance of one item relative to another, or any order of addition, for example.

[0026] The use of the term "or" in the claims is intended to mean the inclusive "and / or" unless explicitly indicated to refer to only alternatives or unless the alternatives are mutually exclusive. For example, the condition "A or B" is satisfied by any of the following: A is true (or present) and B is false (or absent), A is false (or absent) and B is true (or present), and both A and B are true (or present).

[0027] As used herein, any reference to "one embodiment," "an embodiment," "some embodiments," "an example," "for example," or "an example" means that a particular element, feature, structure, or characteristic described in conjunction with the embodiment is included in at least one embodiment. For example, the appearance of the phrase "in some embodiments" or "an example" in various places in the specification does not necessarily all refer to the same embodiment. Further, all references to one or more embodiments or examples are to be interpreted as non-limiting with respect to the claims.

[0028] Throughout this application, the term "approximately" is used to indicate a value including a change in the inherent error of a composition / device / equipment, a method for determining the value, or a change present in a research subject. For example, but not by way of limitation, when utilizing the term "approximately", a specified value can change from a specified value by plus or minus 20%, or 15%, or 12%, or 11%, or 10%, or 9%, or 8%, or 7%, or 6%, or 5%, or 4%, or 3%, or 2%, or 1%, because such a change is suitable for performing the disclosed method and is understood by those of ordinary skill in the art. In an embodiment, the term "approximately" refers to a value similar to a reference value in the context when used herein with reference to a value. Generally speaking, those skilled in the art who are familiar with the context will appreciate the relevant degree of variation covered by "approximately" in this context. For example, in some embodiments, the term "about" may encompass a range of values ​​within 25%, 20%, 19%, 18%, 17%, 16%, 15%, 14%, 13%, 12%, 11%, 10%, 9%, 8%, 7%, 6%, 5%, 4%, 3%, 2%, 1% or less of the reference value.

[0029] As used herein, the term "or a combination thereof" refers to all permutations and combinations of the items listed before the term. For example, "A, B, C, or a combination thereof" is intended to include at least one of the following: A, B, C, AB, AC, BC, or ABC, and if order is important in a particular context, BA, CA, CB, CBA, BCA, ACB, BAC, or CAB. Continuing with this example, explicitly included are combinations containing repetitions of one or more items or terms, such as BB, AAA, AAB, BBC, AAABCCCC, CBBAAA, CABABB, and the like. Those skilled in the art will understand that there is typically no limit on the number of items or terms in any combination, unless otherwise apparent from the context.

[0030] As used herein, the term "CBC / Diff" or "blood cell count with differential" refers to a measure of the number of red blood cells, white blood cells, and platelets in the blood, including different types of white blood cells (neutrophils, lymphocytes, monocytes, basophils, and eosinophils). In embodiments, the amount of hemoglobin (the substance that carries oxygen in the blood) and the hematocrit (the amount of whole blood composed of red blood cells) are also measured.

[0031] As used herein, "diagnostic test" is a step or series of steps that are performed or have been performed to obtain information, which is useful in determining whether a patient has a disease, disorder or condition and / or classifying a disease, disorder or condition into a phenotypic category or any category that is important for the prognosis of a disease, disorder or condition or for the possible response to the treatment (general treatment or any specific treatment) of a disease, disorder or condition. Similarly, "diagnosis" refers to providing any type of diagnostic information, including but not limited to whether a subject is likely to have or develop a disease, disorder or condition, the state, stage or characteristics of a disease, disorder or condition as manifested in a subject, information related to the nature or classification of a tumor, information related to prognosis and / or information useful in selecting appropriate treatment or additional diagnostic tests. The choice of treatment may include selecting a specific therapeutic agent or other treatment modality, such as surgery, radiation, etc., a choice about whether to retain or deliver therapy, a choice related to a dosing regimen (e.g., the frequency or level of one or more doses of a specific therapeutic agent or therapeutic agent combination), etc. The choice of additional diagnostic tests may include more specific tests for a given disease, disorder or condition.

[0032] As used herein, the terms "disease" and "disorder" are used interchangeably to refer to a condition in a subject, including a deleterious deviation from the normal structural or functional state of an organism. Non-limiting examples of diseases / disorders include a subject with one or more viral infections, one or more bacterial infections, one or more fungal or parasitic infections, or sepsis.

[0033] As used herein, the term "viral infection" means that a cell or subject is invaded by a virus, the virus is propagated and / or present in a cell or subject. In one embodiment, a viral infection is an "active" infection, i.e., an infection in which the virus replicates in a cell or subject. Such an infection is characterized in that the virus spreads to other cells, tissues and / or organs from cells, tissues and / or organs initially infected by the virus. An infection can also be a latent infection, i.e., an infection in which the virus does not replicate.

[0034] As used herein, the term "bacterial infection" means the invasion of a cell or a subject by bacteria, the proliferation and / or presence of bacteria in a cell or a subject.

[0035] As used herein, the term "fungal infection" means the invasion of a cell or a subject by a fungus, the proliferation and / or presence of a fungus in a cell or a subject.

[0036] As used herein, the term "pathogen infection" means the invasion of a cell or a subject by a pathogen (such as a bacterial, fungal, parasitic or viral pathogen), the proliferation and / or presence of a pathogen in a cell or a subject.

[0037] As used herein, the term "sample" refers to a biological sample obtained or derived from a human subject as described herein. In some embodiments, a biological sample includes a biological tissue or fluid. In some embodiments, a biological sample may include blood; blood cells; tissue or fine needle biopsy samples; body fluids containing cells; free floating nucleic acids; cerebrospinal fluid; lymph; tissue biopsy specimens; surgical specimens; other body fluids, secretions and / or excretions; and / or cells therefrom. In some embodiments, a biological sample includes cells obtained from an individual, such as a human or animal subject. In some embodiments, the obtained cells are or include cells from an individual from which the sample is obtained. In some embodiments, a sample is a "primary sample" obtained directly from a source of interest by any appropriate means. For example, in some embodiments, a primary biological sample is obtained by a method selected from a biopsy (e.g., fine needle aspiration or tissue biopsy), a surgical procedure, or a collection of body fluids (e.g., blood). In some embodiments, the sample is cardiac tissue obtained from a subject. In some embodiments, as will be clear from the context, the term "sample" refers to a preparation obtained by processing (e.g., by removing one or more components of a primary sample and / or by adding one or more reagents to a primary sample). For example, filtration is performed using a semipermeable membrane. As another example of sample processing, the sample can be a plasma sample treated with an anticoagulant selected from the group consisting of EDTA, heparin, and citrate. As another example of sample processing, the sample can be processed to isolate one or more proteins (e.g., by capturing proteins using one or more antibodies). A "processed sample" can include, for example, nucleic acids or polypeptides extracted from a sample or obtained by subjecting a primary sample to techniques such as amplification or reverse transcription of mRNA, isolation and / or purification of specific components.

[0038] As used herein, when used to refer to SARS CoV-2, the term "SARS CoV-2 variant" or "virus variant" or simply "variant" (as will be apparent from the context) is a SARS CoV-2 virus that includes one or more genetic changes relative to a SARS CoV-2 reference strain or one or more dominant virus variants that are already circulating in a population. In some embodiments, the SARS CoV-2 variant includes one or more genomic mutations in the S gene sequence. In some embodiments, the SARS CoV-2 variant includes one or more genomic mutations relative to a reference genome sequence or a portion thereof. In some embodiments, the SARS CoV-2 variant includes one or more genomic mutations in the S gene sequence relative to the S gene sequence of the reference genome. In some embodiments, the reference genome sequence corresponds to the genome sequence of the Wuhan-Hu1 strain (the first gene sequence identified) or the USA-WA1 / 2020 strain (the first identified in the United States) or a portion thereof. In some embodiments, the SARS CoV-2 variant includes one or more genomic mutations relative to one or more dominant virus variants circulating in a population. In some embodiments, the SARS CoV-2 variant includes one or more genomic mutations in the S gene sequence relative to the S gene sequence of the dominant virus variant circulating in the population.

[0039] As used herein, the term "subject" refers to an organism, such as a mammal (e.g., human). In some embodiments, a human subject is an adult, teenager, or child subject. In some embodiments, the subject is at least 50 years old, at least 55 years old, at least 60 years old, at least 65 years old, at least 70 years old, at least 75 years old, or at least 80 years old. In some embodiments, the subject suffers from a disease, disorder, or condition, such as a disease, disorder, or condition that can be treated as provided herein. In some embodiments, the subject is susceptible to a disease, disorder, or condition; in some embodiments, susceptible subjects tend to and / or show an increased risk of developing a disease, disorder, or condition (compared to the average risk observed in a reference subject or population). In some embodiments, the subject shows one or more symptoms of a disease, disorder, or condition. In some embodiments, the subject does not show specific symptoms (e.g., clinical manifestations of the disease) or characteristics of a disease, disorder, or condition. In some embodiments, the subject does not show any symptoms or characteristics of a disease, disorder, or condition. In some embodiments, the subject is a patient. In some embodiments, the subject is an individual to whom diagnosis and / or therapy is being and / or has been administered. In some embodiments, the human subject is an adult, teenager, or child subject. In some embodiments, the subject is at risk of SARSCoV-2 viral infection. In some embodiments, the subject is at risk for viral infection, bacterial infection, sepsis, or a combination thereof. In some embodiments, the subject has been exposed to or is suspected of having a COVID-19 infection. In some embodiments, the subject is susceptible to COVID-19; in some embodiments, susceptible subjects tend to develop COVID-19 and / or show an increased risk of developing COVID-19 (compared to the average risk observed in a reference subject or population). In some embodiments, the subject shows one or more symptoms of SARS CoV-2 infection. In some embodiments, the subject does not show specific symptoms or characteristics of COVID-19. In some embodiments, the subject does not show any symptoms or characteristics of COVID-19 (i.e., asymptomatic). In some embodiments, the subject is a patient. In some embodiments, the subject is an individual to whom diagnosis and / or therapy is being and / or has been administered.

[0040] As used herein, the term "substantially" means that the event or situation described subsequently occurs completely, or the event or situation described subsequently occurs in a large range or degree. For example, when associated with a particular event or situation, the term "substantially" means that the event or situation described subsequently occurs in at least 80% of the time or at least 85% of the time or at least 90% of the time or at least 95% of the time. The term "substantially adjacent" can mean that two items are 100% adjacent to each other, or two items are closely adjacent to each other but not 100% adjacent to each other, or a part of one of the two items is not 100% adjacent to another item, but is closely adjacent to another item.

[0041] As used herein, the term "threshold value" refers to a value (or values) used as a reference for obtaining information about a measurement result and / or classifying a measurement result, such as a measurement result obtained in an experiment. The threshold value can be determined based on one or more control samples. The threshold value can be determined before, simultaneously, or after performing a measurement of interest. In some embodiments, the threshold value can be a range of values. In some embodiments, the threshold value can be a value (or range of values) reported in a related field (e.g., a value found in a standard table).

[0042] The term "treatment" or "treating" of a subject includes applying or administering a compound to a subject for the purpose of delaying, slowing, stabilizing, curing, healing, mitigating, alleviating, altering, remedying, less exacerbating, improving, improving, or affecting a disease or condition, a symptom of a disease or condition, or a risk (or susceptibility) for a disease or condition. The term "treatment" refers to any indication of success in the treatment or improvement of an injury, pathology, or condition, including any objective or subjective parameter, such as mitigation; relief; reducing the rate of worsening; reducing the severity of the disease; stabilizing, alleviating symptoms, or making an injury, pathology, or condition more tolerable for a subject; slowing the rate of degeneration or decline; making the endpoint of degeneration less debilitating; or improving the physical or mental health of a subject. In an embodiment, the term "treatment" means reducing or improving the progression, severity, and / or duration of COVID-19, or improving one or more symptoms of COVID-19 caused by the administration of one or more therapies (e.g., one or more therapeutic agents). In a specific embodiment, the term "treatment" means improving a measurable physical parameter of COVID-19. In embodiments, the term "treat" means reducing or ameliorating the progression, severity and / or duration of a viral, bacterial or fungal infection, or ameliorating one or more symptoms of a viral, bacterial or fungal infection caused by the administration of one or more therapies (e.g., one or more therapeutic agents). In specific embodiments, the term "treat" means ameliorating a measurable physical parameter of a viral, bacterial or fungal infection. In embodiments, the term "treat" means reducing or ameliorating the progression, severity and / or duration of sepsis, or ameliorating one or more symptoms of sepsis caused by the administration of one or more therapies (e.g., one or more therapeutic agents). In specific embodiments, the term "treat" means ameliorating a measurable physical parameter of sepsis. In embodiments, "treat" changes the natural or presenting state of a subject.

[0043] As used herein, the term "therapeutically effective amount" means when applied to a subject for the treatment or prevention of a specific disorder, disease or condition, the amount of the compound sufficient to achieve such treatment or prevention of the disorder, disease or condition. Dosage and therapeutically effective amount can vary, for example, depending on a variety of factors, including the activity of the specific agent used, the age, body weight, general health, sex and diet of the subject, the time of administration, the route of administration, the rate of excretion and any drug combination (if applicable), the practitioner's desired effect of the compound on the subject and the properties of the compound (e.g., bioavailability, stability, efficacy, toxicity, etc.), and the (one or more) specific disorders suffered by the subject. In addition, the therapeutically effective amount administered intravenously can depend on the blood parameters of the subject, for example, blood lipid examination, insulin level, blood sugar or liver metabolism. The therapeutically effective amount will also vary according to the severity of the disease state, organ function or potential disease or complication. Such an appropriate dose can be determined using any available test. When one or more of the compound or therapeutic agent is to be applied to humans, for example, a doctor can first prescribe a relatively low dose, then increase the dose until an appropriate response is obtained.

[0044] As used herein, the term "viral disease" refers to a pathological condition caused by the presence of a virus in a cell or in a subject, or by the invasion of a virus into a cell or a subject.

[0045] As used herein, the term "influenza virus disease" refers to a pathological state caused by the presence of influenza virus (e.g., influenza A or B virus) in a cell or in a subject or by the invasion of a cell or subject by influenza virus. In a specific embodiment, the term refers to a respiratory disease caused by influenza virus.

[0046] As used herein, the term "COVID-19 disease" refers to a pathological state caused by the presence of a SARS-CoV-2 virus (e.g., a SARS-CoV-2 variant) in a cell or in a subject, or by the invasion of a cell or subject by a SARS-CoV-2 virus or its variant. In a specific embodiment, the term refers to a respiratory disease caused by a coronavirus such as a SARS-CoV-2 virus or its variant.

[0047] The term "sepsis" has been used to describe a wide variety of clinical conditions associated with systemic manifestations of inflammation accompanying infection. Identifying sepsis has been a particularly challenging diagnostic problem due to clinical similarities to inflammatory responses secondary to non-infectious etiologies. The definition of sepsis may change over time and may also vary across various clinical / hospital systems (e.g., Singer M, Deutschman CS, Seymour CW, Shankar-Hari M, Annane D, Bauer M, Bellomo R, Bernard GR, Chiche JD, Coopersmith CM, Hotchkiss RS. The third international consensus definitions for sepsis and septic shock (Sepsis-3). JAMA. 2016 Feb 23;315(8):801-10. Bone RC, Balk RA, Cerra FB, et al. American College of Chest Physicians / Society of Critical Care Medicine Consensus Conference: definitions for sepsis and organ failure and guidelines for the use of innovative therapies in sepsis. Crit Care Med. 1992;20(6):864-874. Levy MM, Fink MP, Marshall JC, et al. International Sepsis DefinitionsConference.2001SCCM / ESICM / ACCP / ATS / SIS International Sepsis DefinitionsConference, Intensive Care Med.2003;29(4):530-538. Centers for Disease Control and Prevention.Hospital toolkit for adult sepsis surveillance.2018", see, e.g., www.cdc.gov / sepsis / pdfs / Sepsis-Surveillance-Toolkit-Mar-2018_508.pdf).All such definitions of sepsis, without limitation to the examples cited above, are intended for purposes of this application.

[0048] Detailed Description of Certain Embodiments

[0049] Figure 1 is a block diagram of a system 100 in which an embodiment may be implemented, for example, as a system 100 for predicting the presence of a disease, the system 100 being configured to perform the processes as described therein. Figure 1 In the embodiment, the system 100 includes a processing unit 101, a memory 102, a storage unit 103, an input unit 104, a bus 106, an output unit 105 and a network interface 107.

[0050] As used herein, processing unit 101 means any type of computing circuit, such as but not limited to a microprocessor, a microcontroller, a complex instruction set computing microprocessor, a reduced instruction set computing microprocessor, a very long instruction word microprocessor, an explicitly parallel instruction computing microprocessor, a graphics processor, a digital signal processor, or any other type of processing circuit. Processing unit 101 may also include an embedded controller, such as a general or programmable logic device or array, an application specific integrated circuit, a single chip computer, and the like.

[0051] The memory 102 may be a volatile memory and a non-volatile memory. The memory 102 may be coupled for communication with the processing unit 101. The processing unit 101 may execute instructions and / or codes stored in the memory 102. A variety of computer-readable storage media may be stored in the memory 102 and accessed from the memory 102. The memory 102 may include any suitable element for storing data and machine-readable instructions, such as a read-only memory, a random access memory, an erasable programmable read-only memory, an electrically erasable programmable read-only memory, a hard drive, a removable media drive for handling compact disks, digital video disks, floppy disks, cassettes, memory cards, and the like. In the present embodiment, the memory 102 includes a disease prediction module 110 stored in the form of machine-readable instructions on any of the above-mentioned storage media, and may communicate with the processor 101 and be executed by the processor 101. When executed by the processor 101, the disease prediction module 110 enables the processor 101 to predict the presence of a disease in a subject. In Figure 2 The method steps executed by the processor 101 to implement the above-mentioned functions are described in detail.

[0052] The storage unit 103 may be a non-transitory storage medium storing a medical database 112. The medical database 112 is a patient data repository including blood parameters maintained by a health care service provider. The input unit 104 may include an input component capable of receiving an input signal, such as a keyboard, a touch-sensitive display, a camera (such as a camera receiving gesture-based input), etc. The bus 106 serves as an interconnection between the processing unit 101, the memory 102, the storage unit 103, the input unit 104, the output unit 105, and the network interface 107.

[0053] One of ordinary skill in the art will appreciate that Figure 1 The hardware depicted in the can vary for a particular implementation. For example, other peripherals, such as optical drives and the like, local area network (LAN) / wide area network (WAN) / wireless (e.g., Wi-Fi) adapters, graphics adapters, disk controllers, input / output (I / O) adapters, can also be used in addition to or in place of the depicted hardware. The depicted examples are provided for purposes of explanation only and are not meant to imply architectural limitations with respect to the present disclosure.

[0054] The system 100 according to an embodiment of the present disclosure includes an operating system that employs a graphical user interface. The operating system permits multiple display windows to be presented simultaneously in the graphical user interface, wherein each display window provides an interface to a different application or a different instance of the same application. The cursor in the graphical user interface can be manipulated by the user through a pointing device. The positioning of the cursor can be changed, and / or events such as clicking a mouse button can be generated to actuate a desired response. If suitably modified, one of various commercial operating systems, such as Microsoft Windows TM A version of , which is a product of Microsoft Corporation located in Redmond, Wash. The operating system is modified or created according to the present disclosure as described.

[0055] Additionally, a non-transitory computer-readable medium containing executable instructions is provided, which when executed causes a processor to perform operations including methods as provided herein. In an embodiment, the present disclosure includes an article of manufacture, such as a system or component thereof including a non-transitory computer-readable medium having instructions encoded thereon, the instructions being configured to cause one or more processors to perform a method including the steps of predicting the presence of a disease according to the present disclosure. In an embodiment, the present disclosure includes a non-transitory computer-readable medium having instructions encoded thereon, the instructions being configured to cause one or more processors to perform a method including the steps of predicting the presence of a disease according to the present disclosure.

[0056] Figure 2The process flow 200 for predicting the presence of a disease in a subject according to another embodiment is illustrated.At step 201, multiple parameters associated with a subject are received.Parameters associated with a subject can be obtained from a complete blood cell count / difference test performed for the subject.In particular, parameters can include red blood cell (RBC) parameters, white blood cells (WBC) and platelet parameters.RBC parameters include hemoglobin level, hematocrit, RBC size, hemoglobin level (normal, high, low) in a single RBC, mean blood volume, mean blood hemoglobin concentration, normal RBC, hemoglobin concentration ≥28g / dL and ≤41g / dL RBC quantity, hemoglobin volume ≥60fL and ≤120fL RBC quantity, hemoglobin distribution width, RBC mean blood volume, RBC volume distribution width, RBC total count, hemoglobin content distribution width and hemoglobin content mean value.Parameters based on RBC size include RBC counts below 60fL and above 120fL. Additional parameters are also considered, such as cell hemoglobin concentration mean value, RBC quantity with hemoglobin concentration greater than 41g / dL, and RBC quantity with hemoglobin concentration less than 28g / dL. Parameters associated with the subject can be obtained from the complete blood cell count / difference test performed for the subject. In particular, the parameters can include red blood cell (RBC) parameters, white blood cells (WBC) and platelet parameters. In an embodiment, the RBC parameters are pre-selected, and include hemoglobin level, hematocrit, RBC size, hemoglobin level (normal, high, low) in individual RBC, mean blood volume, mean blood hemoglobin concentration, normal RBC, RBC quantity with hemoglobin concentration ≥28g / dL and ≤41g / dL, RBC quantity with hemoglobin volume ≥60fL and ≤120fL, hemoglobin distribution width, RBC mean blood volume, RBC volume distribution width, RBC total count, hemoglobin content distribution width and hemoglobin content mean value and one or more of its combination.

[0057] WBC parameters include WBC types, such as lymphocytes, neutrophils, basophils, monocytes, eosinophils, polymorphonuclear cells, large unstained cells (LUC), blasts and platelets. Additionally, WBC count, WBC percentage, neutrophil and lymphocyte ratio, unstained large cells and lymphocyte ratio, WBC quantity and WBC maturation parameters (such as Delta neutrophil index) of indication pseudobasophils are also considered. The parameter indication inflammation and immune activation associated with the experimenter. Therefore, the parameter can be used to determine the existence of infection in the experimenter. In an embodiment, the WBC parameters are preselected. In an embodiment, the WBC parameters include one or more of WBC type (such as lymphocytes, neutrophils, basophils, monocytes, eosinophils, polymorphonuclear cells, large unstained cells (LUC), blasts, and platelets), WBC count, WBC percentage, neutrophil to lymphocyte ratio, large unstained cell to lymphocyte ratio, number of WBCs indicating pseudobasophilia, WBC maturation parameters (such as Delta Neutrophil Index), and combinations thereof.

[0058] In the embodiments, platelet parameters are included and contemplated. Non-limiting platelet parameters include those set forth in platelet threshold, platelet volume, platelet count, and platelet percentage.

[0059] At step 202, the threshold or threshold value of sensitivity and specificity associated with the trained machine learning model is determined for the desired use case. In an embodiment, the threshold value is predetermined. In an embodiment, the trained machine learning model is an integration of decision tree-based methods. For example, an additional tree classifier and a light gradient boosting machine (LightGBM) can be used as a machine learning model. The integration of the decision tree-based model is trained using hematology data associated with multiple subjects. The hematology data includes WBC and RBC data obtained from a complete blood count / difference test associated with multiple subjects. The machine learning model is trained using 5-fold cross validation and optimized for sensitivity. The additional tree classifier is an ensemble learning technique that improves accuracy and controls overfitting by learning randomized decision trees on different data subsamples. The light gradient boosting machine (LightGBM) is an enhancement method for decision tree learning, in which continuous feature values ​​are discretized into intervals for faster training and reduced memory usage. Different classifiers (ag) with associated hyperparameters used in the ensemble are specified below. The weights of each classifier in the ensemble are in the same order as listed [2, 1, 1, 2, 1, 1, 1].

[0060] a.ExtraTreesClassifier(bootstrap=False,class_weight="balanced″,

[0061] criterion="gini",max_features="log2″,mmin_samples_leaf=0.47421052631578947,min_samples_split=0.15052631578947367,n_estimators=600,oob_score=False)

[0062] b.LightGBM Classifier(boosting_type="gbdt″,

[0063] colsample_bytree=0.6933333333333332,learning_rate=0.05789894736842106,subsample_for_bin=190,max_depth=3,min_child_weight=9,min_data_in_leaf=

[0064] 0.03793724137931035,min_split_gain=0.15789473684210525,n_estimators=50,

[0065] num_leaves=230,reg_alpha=0.2631578947368421,reg_lambda=

[0066] 0.7894736842105263,subsample=0.3963157894736842)

[0067] c.ExtraTreesClassifier(bootstrap=False,class_weight="balanced″,

[0068] criterion="gini",max_features=0.2,min_samples_leaf=0.01,

[0069] min_samples_split=0.01,n_estimators=200,oob_score=False)

[0070] d.ExtraTreesClassifier(bootstrap=True, class_weight="balanced", criteria="gini", max_features=0.4, min_samples_leaf=0.01, min_samples_split=0.15052631578947367, n_estimators=10, oob_score=True)

[0071] e.ExtraTreesClassifier(bootstrap=False, class_weight="balanced", criteria="gini", max_features="log2", min_samples_leaf=0.01, min_samples_split=0.056842105263157895, n_estimators=400, oob_score=False)

[0072] f.ExtraTreesClassifier(bootstrap=False, class_weight="balanced", criteria="gini", max_features=0.8, min_samples_leaf=0.01, min_samples_split=0.15052631578947367, n_estimators=25, oob_score=False)

[0073] g.ExtraTreesClassifier(Default parameters).

[0074] (As used herein, the term "algorithm A" refers to the above algorithm).

[0075] In an embodiment, the threshold of sensitivity and / or specificity associated with a trained machine learning model is determined based on the demand scenario associated with a health care provider. For example, in an embodiment, the demand scenario associated with a health care provider includes reducing the RT-PCR test burden, reducing the infectious disease test burden, replacing the RT-PCR test and / or determining at least one of the needs of the subject test. In the demand scenario of reducing the RT-PCR test burden, the algorithm can be used at a fairly high sensitivity (0.80-0.95) so that most positives will be predicted as positive, and some negatives will be predicted as positive. By selecting the desired probability threshold, the RT-PCR test can be avoided in samples predicted to be negative by the algorithm. The method can significantly reduce the RT-PCR test burden in many laboratories while missing a small amount of positives. This use case can be useful in epidemic scenarios. Similarly, for the demand scenario of RT-PCR test substitution, the algorithm can be used at a very high sensitivity (0.99-1.0) so that all positives will be predicted as positive, and many negatives will be predicted as positive. In this high sensitivity scenario, the RT-PCR test can be avoided in samples predicted to be negative without missing any true positives. This method can reduce the RT-PCR test burden in many laboratories / hospitals without missing any positives. This use case can be useful in epidemic scenarios. In demand scenarios that determine the need for subject testing, the algorithm can be used at very high specificity (0.99-1.0) so that all negatives will be predicted as negative, but some positives will be marked as positive. The method can predict positives when there is no suspected positive, while the positive markers for negatives are minimized or non-existent. This use case can be useful in endemic disease scenarios in emergency rooms.

[0076] At step 203, the presence of the disease in the subject is predicted using a trained machine learning model. Such predictions are based on the sensitivity and specificity of the disease and multiple parameters related to the subject. The binary classification model generates the probability that the input is to be classified into a category (in this case, COVID positive or negative). The final positive / negative callout is based on the selected threshold. The default threshold in such a model can be 0.5 (i.e., if the probability of the model output positive prediction is ≥0.5 for any input, the input is marked as positive, and if the probability of the positive prediction is <0.5, it is marked as negative). Depending on the scenario, the threshold is adjusted to different values ​​so that the output from the model can be curated. For example, if a higher positive prediction rate (fewer false positives) is required, a higher threshold of sensitivity and specificity is selected. However, if a higher negative prediction (fewer false negatives) is required, a lower threshold of sensitivity and specificity is selected. At step 204, the prediction is output on the output unit 105.

[0077] For example, the prediction can be displayed on a display unit. In an alternative embodiment, the method includes using a trained machine learning model to determine the need for hospitalization of a subject based on multiple parameters associated with the subject. The model is trained using parameters associated with the multiple subjects who tested positive for the disease, wherein the parameters associated with the multiple subjects are obtained within 2-3 days of the subject testing positive for the disease. Additionally, the model is also trained using health data associated with the subject, which can be observed and collected within 7-10 days after the subject tested positive for the disease.

[0078] In yet another embodiment, the method may include predicting the occurrence of long COVID-19 in a subject. The model is trained using parameters associated with the plurality of subjects who tested positive for the disease and health data associated with the subjects, which may be observed and collected over the next few months after the subject tested positive for the disease.

[0079] Figure 3 A receiver operating characteristic curve or ROC curve 300 is illustrated according to an embodiment for selecting thresholds for sensitivity and specificity associated with a trained machine learning model for a demand scenario. The ROC curve is a graphical plot that illustrates the diagnostic ability of a binary classification model as its discrimination threshold changes. For various thresholds, the ROC plots a curve of the true positive rate (sensitivity) versus the false positive rate (false alarm probability / 1-specificity). The diagonal line divides the ROC space. Points above the diagonal represent good classification results (better than random); points below the line represent bad results (worse than random).

[0080] treat

[0081] In an embodiment, the present disclosure includes methods of treating a subject in need by determining that the subject is a subject in need (such as having a pathogenic infection or sepsis) and then treating the subject. For example, the methods of the present disclosure can be applied to diagnose a patient as positive for a pathogenic infection, sepsis, a viral infection such as COVID-19, and then administer a therapeutic agent or compound to a subject in need in a therapeutically effective amount. For example, a physician can administer any drug, therapeutic agent, or biologic in a therapeutically effective amount suitable for treating a pathogenic infection or disease state following a diagnosis according to the present disclosure. A non-limiting example of an agent suitable for treating a viral infection (such as COVID-19) includes Brand remdesivir (remdesvir) (100MG injection antiviral drug for the treatment of patients hospitalized for COVID-19). See also, for example, international patent application WO 2017 / 049060 A1 and U.S. Patent US20210395345, entitled "Methods for treating or preventing sars-cov-2infectionsand covid-19with anti-sars-cov-2spike glycoprotein antibodies". The therapeutically effective amount of the reagent provided can be determined by a physician based on the subject's response, comorbidities, and the like. In an embodiment, the therapeutic compound is administered to a subject in a diseased state in an effective amount. In an embodiment, a therapeutically effective amount of a compound is administered to a subject in need in an amount sufficient to change the subject's natural or presented state.

[0082] In an embodiment, the subject may be in a sick state due to infection with an RNA virus. Non-limiting examples of RNA viruses include segmented, single-stranded, negative-sense RNA viruses (e.g., single-stranded retrovirales); non-segmented, single-stranded, positive-sense RNA viruses (single-negative viruses); non-segmented, single-stranded, positive-sense RNA viruses (e.g., coronaviruses); double-sense RNA viruses (e.g., bunyaviruses and arenaviruses, such as Lassa virus, Junin virus, Machupi virus, and lymphocytic choroiditis virus); and double-stranded RNA viruses (e.g., reoviruses). Other viruses that cause disease include segmented, single-stranded, negative-sense RNA viruses; non-segmented, single-stranded, negative-sense RNA viruses; non-segmented, single-stranded, positive-sense RNA viruses; or double-stranded RNA viruses (e.g., reoviruses). Additional RNA viruses include: picornaviruses, togaviruses (e.g., Sindbis virus), flaviviruses, and coronaviruses. The virus can be any type, species, and / or strain of picornaviruses, togaviruses, flaviviruses, and coronaviruses. Additional pathogenic viruses include: respiratory syncytial virus (RSV), influenza virus (influenza A, B or C), human metapneumovirus (HMPV), rhinovirus, parainfluenza virus, SARS coronavirus, human immunodeficiency virus (HIV), hepatitis virus (A, B, C), Ebola virus, herpes virus, rubella, smallpox major and smallpox minor.

[0083] In certain embodiments, the therapeutic composition is administered to a subject who has been diagnosed with a disease caused by a viral infection, for example, the patient has been infected with respiratory syncytial virus (RSV), influenza virus (influenza A, influenza B, or influenza C), human metapneumovirus (HMPV), rhinovirus, parainfluenza virus, SARS coronavirus, human immunodeficiency virus (HIV), hepatitis virus (A, B, C), Ebola virus, herpes virus, rubella, smallpox major, and / or smallpox minor.

[0084] In certain embodiments, the disease treated according to the methods described herein is a disease caused by a bacterial infection. Non-limiting examples of pathogenic bacteria include: Streptococcus pneumoniae, Mycobacterium tuberculosis, Chlamydia pneumoniae, Bordetella pertussis, Mycoplasma pneumoniae, Haemophilus influenzae, Moraxella catarrhalis, Legionella, Pneumocystis jiroveci, Chlamydia psittaci, Chlamydia trachomatis, Bacillus anthracis and Francisella tularensis, Borrelia burgdorferi, Salmonella, Yersinia pestis, Shigella, Escherichia coli, Corynebacterium diphtheriae and Treponema pallidum.

[0085] In certain embodiments, the composition is administered to a patient who has been diagnosed with a disease caused by a bacterial infection, for example, the patient has been infected with Streptococcus pneumoniae, Mycobacterium tuberculosis, Chlamydia pneumoniae, Bordetella pertussis, Mycoplasma pneumoniae, Haemophilus influenzae, Moraxella catarrhalis, Legionella, Pneumocystis, Chlamydia psittaci, Chlamydia trachomatis, Bacillus anthracis and Francisella tularensis, Borrelia burgdorferi, Salmonella, Yersinia pestis, Shigella, Escherichia coli, Corynebacterium diphtheriae and / or Treponema pallidum.

[0086] In certain embodiments, the composition is administered to a patient diagnosed with a disease caused by a fungal infection, for example, the patient has been infected with Blastomyces, Parasporum, Sporothrix, Cryptococcus, Candida, Aspergillus, Histoplasma, Cryptococcus, Dipolaris, Cladosporium, Cladosporium, Desmodium, Exophiala, Chromoblastoma, Phiala, Xylem, Ochracus, Rhinobacterium, Dendrocalamus, and / or Exophiala.

[0087] In certain embodiments, the composition is administered to a patient who has been diagnosed with a disease caused by a yeast infection, for example, a patient who has been infected with Lactobacillus acidophilus, Staphylococcus, Brettanomyces, Sporobacterium bulleri, Sporobacterium bulleri, Candida, Pseudomonas, Cleospora, Cryptococcus, Pseudomonas, Debaryomyces, Debaryomyces, Dekkera, Diplocystis, Endosporium, Pseudomonas ... and / or Zygosaccharomyces.

[0088] In certain embodiments, the composition is administered to a subject infected with a parasite, e.g., a patient already infected with Babesia, Cryptosporidium, Entamoeba histolytica, Leishmania, Giardia lamblia, Plasmodium, Toxoplasma, Trichomonas, Trypanosoma, Ascaris, Tapeworm, Hookworm, Brugia, Fasciola, Trichinella, Schistosoma, Taeniasis, Bedbug, Lice, and / or Sarcoptes. See U.S. Patent Publication No. 20170321192 for additional information (incorporated herein by reference in its entirety).

[0089] Exemplary Numbering Embodiments

[0090] Embodiment 1. A system (100) for predicting the presence of a disease in a subject, the system (100) comprising: one or more processing units (101); a medical database (112) coupled to the one or more processing units (101), the medical database (112) comprising patient data; and a disease prediction module (110) configured to: receive a plurality of parameters associated with a subject; determine a threshold of sensitivity and / or specificity associated with a trained machine learning model; predict the presence of a disease in a subject using the trained machine learning model based on the model's specific threshold and the plurality of parameters associated with the subject; and output a prediction on an output unit (105).

[0091] Embodiment 2: A system (100) for predicting the presence of a disease in a subject, the system (100) comprising: one or more processing units (101); a medical database (112) coupled to the one or more processing units (101), the medical database (112) comprising patient data; and a disease prediction module (110), configured to: receive a plurality of parameters associated with a subject; determine a threshold of sensitivity and / or specificity associated with a trained machine learning model; use the trained machine learning model to predict the presence of a disease in a subject based on a specific threshold of the model and the plurality of parameters associated with the subject; and output a prediction on an output unit (105), wherein the threshold or threshold value of sensitivity and / or specificity associated with the trained machine learning model is determined based on a demand scenario associated with a healthcare provider.

[0092] Embodiment 3. A system (100) for predicting the presence of a disease in a subject, the system (100) comprising: one or more processing units (101); a medical database (112) coupled to the one or more processing units (101), the medical database (112) comprising patient data; and a disease prediction module (110) configured to: receive a plurality of parameters associated with a subject; determine a threshold for sensitivity and / or specificity associated with a trained machine learning model; use the trained machine learning model to predict the presence of a disease in a subject based on a specific threshold or threshold value of the model and the plurality of parameters associated with the subject; and output a prediction on an output unit (105), wherein the threshold or threshold value for sensitivity and / or specificity associated with the trained machine learning model is determined based on a demand scenario associated with a healthcare provider, and wherein the demand scenario associated with the healthcare provider includes at least one of reducing the burden of RT-PCR testing, replacing RT-PCR testing, and / or determining the need for testing of the subject.

[0093] Embodiment 4. A system (100) for predicting the presence of a disease in a subject, the system (100) comprising: one or more processing units (101); a medical database (112) coupled to the one or more processing units (101), the medical database (112) comprising patient data; and a disease prediction module (110) configured to: receive a plurality of parameters associated with the subject; determine a threshold of sensitivity and / or specificity associated with a trained machine learning model; use the trained machine learning model to predict the presence of a disease in the subject based on the specific threshold of the model and the plurality of parameters associated with the subject; and output a prediction on an output unit (105), wherein the plurality of parameters associated with the subject comprise red blood cell (RBC) parameters and white blood cell (WBC) parameters.

[0094] Embodiment 5. A system (100) for predicting the presence of a disease in a subject, the system (100) comprising: one or more processing units (101); a medical database (112) coupled to the one or more processing units (101), the medical database (112) comprising patient data; and a disease prediction module (110) configured to: receive a plurality of parameters associated with the subject; determine a threshold of sensitivity and / or specificity associated with a trained machine learning model; predict the presence of a disease in the subject using the trained machine learning model based on the specific threshold of the model and the plurality of parameters associated with the subject; and output the predicted disease to an output unit (105). The method comprises the step of: predicting a disease state of the present invention, wherein the plurality of parameters associated with the subject include red blood cell (RBC) parameters and white blood cell (WBC) parameters, and wherein the RBC parameters include one or more of the following: hemoglobin level, hematocrit, RBC size, hemoglobin level in individual RBCs, mean corpuscular volume, mean corpuscular hemoglobin concentration, normal RBCs, the number of RBCs with a hemoglobin concentration ≥28 g / dL and ≤41 g / dL, the number of RBCs with a hemoglobin volume ≥60 fL and ≤120 fL, hemoglobin distribution width, RBC mean corpuscular volume, RBC volume distribution width, RBC total count, hemoglobin content distribution width, and hemoglobin content mean.

[0095] Embodiment 6. A system (100) for predicting the presence of a disease in a subject, the system (100) comprising: one or more processing units (101); a medical database (112) coupled to the one or more processing units (101), the medical database (112) comprising patient data; and a disease prediction module (110) configured to: receive a plurality of parameters associated with the subject; determine a threshold of sensitivity and / or specificity associated with a trained machine learning model; use the trained machine learning model to predict the presence of a disease in the subject based on the specific threshold of the model and the plurality of parameters associated with the subject; and output a prediction on an output unit (105), wherein the plurality of parameters associated with the subject comprise red blood cell (RBC) parameters and white blood cell (WBC) parameters, wherein the WBC parameters comprise one or more of the following: WBC type, WBC count, WBC percentage, neutrophil to lymphocyte ratio, large unstained cell to lymphocyte ratio, number of WBCs indicative of pseudobasophilia, and WBC maturation parameters.

[0096] Embodiment 7. A system (100) for predicting the presence of a disease in a subject, the system (100) comprising: one or more processing units (101); a medical database (112) coupled to the one or more processing units (101), the medical database (112) comprising patient data; and a disease prediction module (110) configured to: receive a plurality of parameters associated with the subject; determine a threshold of sensitivity and / or specificity associated with a trained machine learning model; use the trained machine learning model to predict the presence of a disease in the subject based on the model's specific threshold and the plurality of parameters associated with the subject; and output a prediction on an output unit (105), wherein the plurality of parameters associated with the subject include a red blood cell (RBC) parameter and white blood cell (WBC) parameters, and wherein the RBC parameters include one or more of the following: hemoglobin level, hematocrit, RBC size, hemoglobin level in individual RBCs, mean corpuscular volume, mean corpuscular hemoglobin concentration, normal RBCs, the number of RBCs with a hemoglobin concentration ≥28 g / dL and ≤41 g / dL, the number of RBCs with a hemoglobin volume ≥60 fL and ≤120 fL, hemoglobin distribution width, RBC mean corpuscular volume, RBC volume distribution width, RBC total count, hemoglobin content distribution width, and hemoglobin content mean, wherein the system further includes a cellular hemoglobin concentration mean, a number of RBCs with a hemoglobin concentration greater than 41 g / dL, and a number of RBCs with a hemoglobin concentration less than 28 g / dL.

[0097] Embodiment 8. A system (100) for predicting the presence of a disease in a subject, the system (100) comprising: one or more processing units (101); a medical database (112) coupled to the one or more processing units (101), the medical database (112) comprising patient data; and a disease prediction module (110) configured to: receive a plurality of parameters associated with a subject; determine a threshold of sensitivity and / or specificity associated with a trained machine learning model; use the trained machine learning model to predict the presence of a disease in a subject based on the specific threshold of the model and the plurality of parameters associated with the subject; and output a prediction on an output unit (105), wherein the trained machine learning model is an integration of decision tree-based methods.

[0098] Embodiment 9. A system (100) for predicting the presence of a disease in a subject, the system (100) comprising: one or more processing units (101); a medical database (112) coupled to the one or more processing units (101), the medical database (112) comprising patient data; and a disease prediction module (110) configured to: receive a plurality of parameters associated with the subject; determine a threshold of sensitivity and / or specificity associated with a trained machine learning model; use the trained machine learning model to predict the presence of a disease in the subject based on the specific threshold of the model and the plurality of parameters associated with the subject; and output a prediction on an output unit (105), wherein the disease prediction module (110) is further configured to use the trained machine learning model to predict the need for hospitalization for the subject based on the plurality of parameters associated with the subject.

[0099] Embodiment 10. A system (100) for predicting the presence of a disease in a subject, the system (100) comprising: one or more processing units (101); a medical database (112) coupled to the one or more processing units (101), the medical database (112) comprising patient data; and a disease prediction module (110) configured to: receive multiple parameters associated with the subject; determine a threshold of sensitivity and / or specificity associated with a trained machine learning model; use the trained machine learning model to predict the presence of a disease in the subject based on the specific threshold of the model and the multiple parameters associated with the subject; and output the prediction on an output unit (105), wherein the disease prediction module (110) is further configured to predict the occurrence of long COVID-19 in the subject.

[0100] Example 11. A method (200) for predicting the presence of a disease in a subject, the method (200) comprising: receiving a plurality of parameters associated with the subject; determining a threshold of sensitivity and / or specificity associated with a trained machine learning model; using the trained machine learning model to predict the presence of the disease in the subject based on the specific threshold of the model and the plurality of parameters associated with the subject; and outputting the prediction on an output unit.

[0101] Example 12. A method (200) for predicting the presence of a disease in a subject, the method (200) comprising: receiving a plurality of parameters associated with the subject; determining a threshold of sensitivity and / or specificity associated with a trained machine learning model; using the trained machine learning model to predict the presence of the disease in the subject based on a specific threshold of the model and the plurality of parameters associated with the subject; and outputting a prediction on an output unit, wherein the threshold of sensitivity and / or specificity associated with the trained machine learning model is determined based on a demand scenario associated with a healthcare provider.

[0102] Example 13. A method (200) for predicting the presence of a disease in a subject, the method (200) comprising: receiving a plurality of parameters associated with the subject; determining a threshold of sensitivity and / or specificity associated with a trained machine learning model; using the trained machine learning model to predict the presence of the disease in the subject based on the specific threshold of the model and the plurality of parameters associated with the subject; and outputting a prediction on an output unit, wherein the threshold of sensitivity and / or specificity associated with the trained machine learning model is determined based on a demand scenario associated with a healthcare provider, and wherein the demand scenario associated with the healthcare provider includes at least one of reducing the burden of RT-PCR testing, replacing RT-PCR testing, and / or determining a need for subject testing.

[0103] Example 14. A method (200) for predicting the presence of a disease in a subject, the method (200) comprising: receiving a plurality of parameters associated with the subject; determining a threshold of sensitivity and / or specificity associated with a trained machine learning model; using the trained machine learning model to predict the presence of a disease in the subject based on the model's specific threshold and the plurality of parameters associated with the subject; and outputting the prediction on an output unit, wherein the disease is COVID-19.

[0104] Example 15. A method (200) for predicting the presence of a disease in a subject, the method (200) comprising: receiving a plurality of parameters associated with the subject; determining thresholds of sensitivity and / or specificity associated with a trained machine learning model; using the trained machine learning model to predict the presence of the disease in the subject based on the model's specific thresholds and the plurality of parameters associated with the subject; and outputting the prediction on an output unit, further comprising: using the trained machine learning model to predict the need for hospitalization for the subject based on the plurality of parameters associated with the subject; and predicting the occurrence of long COVID-19 in the subject.

[0105] Example 16. A method (200) for predicting the presence of a disease in a subject, the method (200) comprising: receiving a plurality of parameters associated with the subject; determining a threshold of sensitivity and / or specificity associated with a trained machine learning model; using the trained machine learning model to predict the presence of a disease in the subject based on the model's specific threshold and the plurality of parameters associated with the subject; and outputting the prediction on an output unit, wherein the disease is COVID-19.

[0106] Example 17. A method (200) for predicting the presence of a disease in a subject, the method (200) comprising: receiving a plurality of parameters associated with the subject; determining thresholds of sensitivity and / or specificity associated with a trained machine learning model; using the trained machine learning model to predict the presence of the disease in the subject based on the model's specific thresholds and the plurality of parameters associated with the subject; and outputting the prediction on an output unit, further comprising: using the trained machine learning model to predict the need for hospitalization of the subject based on the plurality of parameters associated with the subject; and predicting the occurrence of a viral, bacterial, or fungal infection in the subject.

[0107] Embodiment 18. An article of manufacture such as a system or component thereof including a non-transitory computer-readable medium having instructions encoded thereon, the instructions configured to cause one or more processors to perform the methods of embodiments 15-17 above.

[0108] Embodiment 19. In an embodiment, the present disclosure includes a non-transitory computer-readable medium having instructions encoded thereon, the instructions being configured to cause one or more processors to execute a method comprising: predicting the presence of a disease in a subject, the method (200) comprising: receiving a plurality of parameters associated with the subject; determining a threshold of sensitivity and / or specificity associated with a trained machine learning model; using the trained machine learning model to predict the presence of the disease in the subject based on the specific threshold of the model and the plurality of parameters associated with the subject; and outputting the prediction on an output unit, further comprising: using the trained machine learning model to predict the need for hospitalization of the subject based on the plurality of parameters associated with the subject; and predicting the occurrence of a viral, bacterial, or fungal infection in the subject.

[0109] Embodiment 20. In an embodiment, the present disclosure includes a non-transitory computer-readable medium having instructions encoded thereon, the instructions being configured to cause one or more processors to execute a method comprising: predicting the presence of a disease in a subject, the method (200) comprising: receiving a plurality of parameters associated with the subject; determining a threshold of sensitivity and / or specificity associated with a trained machine learning model; using the trained machine learning model to predict the presence of the disease in the subject based on the specific threshold of the model and the plurality of parameters associated with the subject; and outputting the prediction on an output unit, further comprising: using the trained machine learning model to predict the need for hospitalization of the subject based on the plurality of parameters associated with the subject; and predicting the occurrence of sepsis in the subject.

[0110] Embodiment 21: An article of manufacture comprising a non-transitory computer-readable medium having instructions encoded thereon, the instructions configured to cause one or more processors to perform a method comprising: predicting the presence of a disease according to an embodiment of the present disclosure.

[0111] Embodiment 22. A non-transitory computer-readable medium having encoded thereon instructions configured to cause one or more processors to perform a method comprising: predicting the presence of a disease according to at least one embodiment of the present disclosure.

[0112] Embodiment 23. A method of treating a subject in need thereof comprises predicting the presence of a disease in a subject according to embodiments of the present disclosure, and subsequently treating the subject in need thereof. In embodiments, the treating comprises administering to the subject a therapeutically effective amount of the composition.

[0113] Embodiment 24. A method of treating a subject in need thereof, comprising predicting the presence of a disease in a subject, the method (200) comprising: receiving a plurality of parameters associated with the subject; determining thresholds of sensitivity and / or specificity associated with a trained machine learning model; using the trained machine learning model to predict the presence of a disease in a subject based on the model's specific thresholds and the plurality of parameters associated with the subject; and outputting the prediction on an output unit, and subsequently treating the subject in need thereof. In embodiments, the treatment comprises administering a therapeutically effective amount of a composition to the subject. In embodiments, the disease is characterized as a viral disease, a bacterial disease, COVID-19, long Covid, etc.

[0114] Example 25: In an embodiment, the present disclosure includes a system for predicting the presence of a disease in a population. In one aspect of the present disclosure, the system includes a processing unit and a memory. Additionally, the memory includes a disease prediction module, which is configured to receive a plurality of parameters associated with one or more subjects and determine a threshold of sensitivity and / or specificity associated with a trained machine learning model. Additionally, the module is configured to use the trained machine learning model to predict the presence of a disease in one or more subjects or a population of subjects and the plurality of parameters associated with the subjects or a population of subjects, and output the prediction on an output unit.

[0115] Example

[0116] Example 1

[0117] This example describes a COVID-19 detection algorithm based on CBC / Diff hematology data. In an embodiment, the present disclosure includes a multi-parameter machine learning algorithm to run on, among other things, a hematology instrument and provide COVID-19 markers along with CBC / Diff data. In an embodiment, in an epidemic or endemic situation, the algorithm can be run at different sensitivity and specificity values ​​based on the expected use case. The training data used in the algorithm is based on an unvaccinated population from 2020. Additional data from more varied positive and negative individuals can be used to improve the robustness of the algorithm.

[0118] Hematology tests such as CBC / Diff (complete blood count with differential) are commonly performed for overall health monitoring. In an embodiment, the present disclosure uses a blood test kit from Siemens Healthineers The CBC / Diff results of the 2120i Hematology System are combined with a multi-parameter machine learning algorithm for rapid marking of COVID-19 (Coronavirus Disease 2019) positives. Diagnostic data generation and running the algorithm on the data all occur on the same instrument, ensuring easy data access. Additionally, the algorithm can also be run on a connected laboratory information system. Such an algorithm can be valuable when access to COVID testing is limited, and can be used to predict COVID-19 in unsuspected individuals, thus prompting confirmatory testing. Here, various epidemic and endemic applications for the algorithm are demonstrated.

[0119] Methods. CBC / Diff data of 309 COVID-19 positive and 245 negative were obtained from the 2020 pre-vaccination cohort from Mumbai, India. Using this data, a machine learning classification model was trained using 5-fold cross validation to distinguish COVID-19 positive and negative.

[0120] Results. The final ML model was an ensemble of decision tree-based methods with 74% accuracy on the test set and an AUROC of 0.79. The most notable features included eosinophil and lymphocyte counts, which had been reported earlier. The model was able to more accurately predict hospitalized COVID-19 patients (92% correct classification) than non-hospitalized patients (70% correct classification).

[0121] Conclusion. A predictive algorithm, method, and system for rapid COVID-19 flagging using data from routine CBC testing is provided. The algorithm is useful in epidemic or endemic scenarios as an alternative to testing, reducing testing burden and flagging unsuspected positives.

[0122] discuss

[0123] Hematology data provides a snapshot of the patient's health status and represents the host's response to infection. Additionally, hematology tests such as CBC / Diff are often prescribed before any disease is suspected, and are therefore an ideal platform for marking and stratifying possible infections or health conditions, thus enabling earlier intervention and streamlining different medical resources (e.g., see Barnes PW, McFadden SL, Machin SJ, Simson E. "The international consensus group for hematology review: suggested criteria for action following automated CBC and WBC differential analysis", Laboratory hematology: official publication of the International Society for Laboratory Hematology 2005; 11 2: 83-90). Here, Siemens Healthineers The CBC / Diff test performed by the 2120i Hematology System is combined with a multi-parameter AI algorithm to flag COVID-19 positivity in individuals.

[0124] Studies have reported hematologic characteristics associated with COVID-19 positivity and prognosis, such as eosinophilia and lymphocytopenia (see, e.g., Frater JL, Zini G, d'Onofrio G, Rogers HJ, “COVID-19 and the clinical hematology laboratory,” International Journal of Laboratory Hematology 2020;42:11-8; Guan Wj, Ni Zy, Hu Y, Liang Wh, Ou Cq, He J-x, Liu L, et al., “Clinical characteristics of coronavirus disease 2019 in China,” New England journal of medicine 2020;382:1708-20; Henry BM, De Oliveira MHS, Benoit S, Plebani M, Lippi G., “Hematologic, biochemical and immune biomarker abnormalities associated with severe illness and mortality in coronavirus disease 2019 (COVID-19): a meta-analysis,” Clinical Chemistry 2020;43:11-20; andLaboratory Medicine (CCLM) 2020;58:1021; "The association between severe COVID-19 and low platelet count: evidence from 31observational studies involving 7613participants" by Jiang SQ, Huang QF, Xie WM, Lv C, QuanX-Q, British journal of haematology2020:e29-e33; and Kermali M, Khalsa "The role of biomarkers in diagnosis of COVID-19--A systematic review" by RK, Pillai K, Ismail Z, Harky A., Lifesciences 2020;254:117788).Some hematological parameters also showed mixed trends in different studies, probably due to the population dependence of parameter values. However, it is difficult for doctors to identify COVID-19 positivity by evaluating a single parameter or even a combination of parameters in isolation. Multiparameter machine learning (ML) algorithms enable modeling of complex relationships between multiple predictive parameters and generate predictions more accurately and efficiently. Multiparameter studies that combine parameters measured on multiple instruments (e.g., see Kukar M,. G, Vovko T, Podnar S, P, Brvar M, Zalaznik M et al., "COVID-19 diagnosis by routine blood tests using machine learning", Scientific reports 2021; 11: 1-9), the present disclosure uses multiple hematological parameters measured / derived using a single instrument for prediction. This eliminates the need to aggregate data from multiple sources and allows easy access to data without any missing values, which is a common problem in such multi-parameter algorithms.

[0125] Since the beginning of the epidemic, COVID-19 testing and antigen testing using RT-PCR (reverse transcriptase-polymerase chain reaction) have been massively expanded, however, in both epidemic and endemic scenarios, there are multiple use cases where predictive algorithms based on CBC / Diff can be applied to the clinical value chain. First, it can replace testing in countries that deal with the unavailability or shortage of such tests. Secondly, in epidemic scenarios, rapid marking by highly sensitive hematology-based machine learning (ML) algorithms can reduce the testing burden. Third, algorithms with high negative predictive values ​​(NPV) can reduce the number of patients who undergo testing before procedures such as surgery, thereby reducing the time of patient care. Fourth, in endemic scenarios, algorithms with high specificity can be used to mark asymptomatic individuals, who are then tested to confirm infection. The same ML algorithm with different classification thresholds can be used in multiple scenarios as described above.

[0126] Hematology-based multi-parameter ML algorithms can be used to flag COVID within seconds to minutes. Reduced time to results can avoid unnecessary isolation of negatives and quickly access care for patients who need to undergo procedures such as surgery in epidemic scenarios, and can also be used to flag unsuspected positives during endemic scenarios.

[0127] The present disclosure includes a multi-parameter machine learning algorithm to predict COVID-19 positivity in individuals using CBC / Diff data derived from a single instrument. In an embodiment, the multi-parameter ML algorithm predicted COVID-19 positivity in individuals with 74% accuracy and an AUROC (area under the receiver operating characteristic curve) of 0.79. Further, in an embodiment, the algorithm predicted positive hospitalized patients with 91.7% accuracy, giving us confidence in the performance of our algorithm and the potential of the selected parameters.

[0128] Materials and methods

[0129] Dataset

[0130] CBC / Diff data were collected from two labs installed at Dr. Jariwalla Laboratory & Diagnostics in Mumbai (India). 2120i instrument is obtained for this analysis. The disclosure includes a retrospective study of the population of 309 positive and 245 COVID-19 negative individuals determined using RT-PCR tests. Only anonymous data are sent to researchers, and the anonymous data include information on the individual sex and age of COVID-19 positive individuals (whether they are hospitalized / admitted to ICU / survived or dead) and disease outcomes. During the first wave of infection in India (May 2020-January 2021), before any vaccination activities (drive), COVID-19 positive and negative were measured simultaneously. The COVID-19 negative included in this study is an individual who received medical attention due to other illnesses during the first wave of the epidemic. Among the 309 positives, 60 were hospitalized at the time of CBC / Diff test, 29 of which were in ICU, and 53 survived from infection. The demographics and blood cell counts of the COVID-19 positive and negative individuals included in this study are summarized in Table 1.

[0131] Table 1 Demographics and blood counts of COVID-19 – positive and negative individuals used in this study.

[0132]

[0133] Parameter selection

[0134] The starting data set consisted of all parameters in the ADVIA 2120i CBC / Diff raw data set. Various criteria were used to eliminate parameters, such as when the median value between positive and negative was not significantly different (Kruskal-Wallis H-test was performed using box plots and p-values), parameters that only measured instrument characteristics, etc. Redundant and highly correlated parameters (Spearman correlation coefficient > 0.9) were removed. Finally, a set of 48 parameters was identified (see Figure 5 ), which includes both raw and calculated patient parameters. All 48 parameters have values ​​and there are no missing values. When training the algorithm, no patient demographic information or information identifying the source instrument (instrument 1 or 2) was included.

[0135] algorithm

[0136] After parameter identification, the classification model was trained using Microsoft Azure AutoML (https: / / www.microsoft.com / en-us / research / project / automl / ) to distinguish between positive and negative. 80% of the data (247 positive, 196 negative) was used to train the model using 5-fold cross validation, and the remaining 20% ​​of the data (62 positive, 49 negative) was used as a test set. Azure AutoML uses the dataset to automatically train many different models, and model training can be optimized based on different metrics. When optimized for sensitivity, an excellent model for the dataset is a voting ensemble of multiple tree-based classifiers. The different classifiers (ag) with associated hyperparameters used in the ensemble are specified below. The weights of each classifier in the ensemble are in the same order as listed [2, 1, 1, 2, 1, 1, 1].

[0137] h.ExtraTreesClassifier(bootstrap=False, class_weight="balanced",

[0138] criterion="gini", max_features="log2", min_samples_leaf=0.47421052631578947,

[0139] min_samples_split=0.15052631578947367,n_estimators=600,oob_score=False)

[0140] i.LightGBM Classifier(boosting_type="gbdt″,

[0141] colsample_bytree=0.6933333333333332,learning_rate=0.05789894736842106,

[0142] subsample_for_bin=190,max_depth=3,min_child_weight==9,min_data_in_leaf=

[0143] 0.03793724137931035,min_split_gain=0.15789473684210525,n_estimators=50,

[0144] num_leaves=230,reg_alpha=0.2631578947368421,reg_lambda=

[0145] 0.7894736842105263,subsample=0.3963157894736842)

[0146] j.ExtraTreesClassifier(bootstrap=False,class_weight="balanced″,

[0147] criterion="gini",max_features=0.2,min_samples_leaf=0.01,min_samples_split=0.01,

[0148] n_estimators=200,oob_score=False)

[0149] k.ExtraTreesClassifier(bootstrap=True,class_weight="balanced",

[0150] criterion="gini",max_features=0.4,min_samples_leaf=0.01,min_samples_split=

[0151] 0.15052631578947367,n_estimators=10,oob_score=True)

[0152] I.ExtraTreesClassifier(bootstrap=False, class_weight="balanced",

[0153] criterion="gini", max_features="log2", min_samples_leaf=0.01, min_samples_split=

[0154] 0.056842105263157895,n_estimators=400,oob_score=False)

[0155] m.ExtraTreesClassifier(bootstrap=False,class_weight="balanced",

[0156] criterion="gini", max_features=0.8, mmin_samples_leaf=0.01, min_samples_split=

[0157] 0.15052631578947367,n_estimators=25,oob_score=False)

[0158] n.ExtraTreesClassifier(Default parameters).

[0159] The trained ensemble model is downloaded from Microsoft Azure. This gives excellent flexibility when using the model, such as when evaluating model performance in test sets and deployment.

[0160] result

[0161] The performance of the model is evaluated based on different metrics, including prediction accuracy, AUROC, sensitivity, and specificity. Next, the feature importance obtained on the training data using the MIMIC LightGBM explainer is presented. The MIMICLightGBM explainer uses a LightGBM proxy model to approximate a complex black-box ensemble model obtained from Azure AutoML to obtain feature / parameter importance values. A comparison of the results from our 48-parameter model with those of other models learned using the most important parameters is performed.

[0162] Different use cases of the algorithm.

[0163] Model Evaluation

[0164] The entire data set was divided into a separate training set (80% of the total data) and a test set (the remaining 20% ​​of the data). The model was trained on the training set using 5-fold cross validation, and the model performance was evaluated on the test set. The test data set consisted of 62 positives and 49 negatives. The model performed on the test data with an accuracy of 74%, a sensitivity of 0.73, a specificity of 0.74, and an AUROC of 0.79 (all metrics were reported using the default classification threshold of 0.5). Figure 4 The ROC (Receiver Operating Curve) curve calculated using the test data is shown. Since the positives and negatives in the test are balanced, the AUROC is a good performance measure for validation.

[0165] Of the 62 positives in the test set, 12 were hospitalized, while the remaining 50 were not hospitalized. It was observed that hospitalized positives were predicted more accurately than non-hospitalized positives (Table 2).

[0166] Table 2

[0167]

[0168] This gives increased confidence in the selected parameters.The effects of age and gender on the algorithm performance were also observed (see Supplementary Tables 1 and 2 below), and it was found that the algorithm predictions showed no bias for age or gender.

[0169] Feature Importance

[0170] Here, model explanations obtained using the MIMIC LightGBM explainer on the training data are presented. Figure 5 Features are shown in decreasing order of importance value as calculated by the model interpreter.

[0171] Comparison with models generated using top parameters

[0172] The importance of the selected 48 parameters is demonstrated by learning two additional classification models using the top 20 and top 10 most important features obtained from the MIMIC explainer ( Figure 5 ). Table 3 summarizes the performance of the three models on the test set.

[0173] Table 3. Comparison of the prediction accuracy of models built using different parameter subsets.

[0174]

[0175] It was observed that the performance of the top 20 parameter model was lower than that of the 48 parameter model in both overall accuracy and sensitivity (or percentage of positives predicted accurately). The prediction accuracy of the top 10 parameter model further decreased, especially the specificity (or percentage of negatives predicted accurately). These results show that all 48 parameters are indeed important in distinguishing positives from negatives.

[0176] Use Cases

[0177] The hematology-based ML algorithm for COVID-19 markers can be useful in multiple use cases in epidemic and endemic scenarios. In embodiments, the classification thresholds can be adjusted to leverage the same predictive model useful in different settings. This disclosure illustrates two use cases for epidemic scenarios, and another use case for endemic scenarios ( Figures 6A-6C ).

[0178] In countries facing inaccessibility or scarcity of COVID tests in epidemic scenarios, AI-based COVID-19 markers using CBC / Diff data can serve as a low-cost and rapid alternative. Fig. 6A As shown in , the algorithm can be used to predict this use case at a sensitivity of 0.94 and a specificity of 0.39 (point 1 in the ROC curve). Depending on the sensitivity chosen, the algorithm can capture most of the positives. This result can be used as an additional point in clinical decision making in the absence of confirmatory testing.

[0179] In another application in an epidemic scenario, the algorithm can be used at high sensitivity to reduce the infection testing burden. This may be particularly useful in high-volume laboratories. The rapid turnaround time for CBC / Diff results on the ADVIA 2120i analyzer (approximately one minute) combined with the algorithm run time (a few seconds) provides a faster alternative to RT-PCR or antigen testing. At point 2 in the ROC curve ( Fig. 6A ), the NPV of the algorithm is 1, where a negative is predicted with high certainty and does not need to be tested and can be exempted from quarantine. Individuals predicted to be positive by the algorithm will need to follow up with COVID-specific testing. The method can reduce the testing burden by approximately 10%.

[0180] The algorithm can also be used in endemic scenarios to drive testing of individuals with fever in the Emergency Department (ED) when COVID-19 is not suspected. An algorithm operating at (point 3 on the ROC curve) high specificity and a positive predictive value (PPV) of 1 can predict COVID-19 positivity with high certainty in individuals who would otherwise have been missed and recommend them for COVID testing. Importantly, at a PPV of 1, no negatives will be flagged for testing, eliminating false alarms.

[0181] discuss

[0182] Embodiments provide a multi-parameter machine learning algorithm for using the The parameters obtained from the CBC / Diff test performed on the 2120i Hematology System are used to flag COVID-19 positive in individuals. The CBC / Diff test is a routinely performed blood test and can help detect a variety of health conditions. There are several advantages to coupling the results of the CBC / Diff test with an AI algorithm for marking infectious diseases. First, all parameters are acquired on a single instrument, which eliminates the need to aggregate data from multiple sources. Second, such an algorithm can serve as a faster alternative to standard virus-specific tests as results can be obtained in minutes. Third, machine learning algorithms can help learn complex relationships between multiple parameters and therefore improve the accuracy of diagnosis.

[0183] There are several studies reporting altered hematological parameters in the presence of COVID-19 infection (e.g., see Henry BM, De Oliveira MHS, Benoit S, Plebani M, Lippi G, “Hematologic, biochemical and immune biomarker abnormalities associated with severe illness and mortality in coronavirus disease 2019 (COVID-19): a meta-analysis,” Clinical Chemistry and Laboratory Medicine (CCLM) 2020; 58: 1021-8; Kermali M, Khalsa RK, Pillai K, Ismail Z, Harky A, “The role of biomarkers in diagnosis of COVID-19--A systematic review,” Life sciences 2020; 254: 117788; Lippi G, Plebani M, “Laboratory abnormalities in patients with COVID-2019 infection,” Clinical chemistry and laboratory medicine (CCLM) 2020; 58: 1131-4; Yip "Temporal changes in immune blood cell parameters in COVID-19 infection and recovery from severe infection" by CY, Yap ES, De Mel S, Teo WZ, Lee CT, Kan S, Lee MC, et al., British Journal of Hematology 2020:33-6.Another study used a linear combination of blood cell count parameters to determine a COVID-19 prognostic score (e.g., see Linssen J, Ermens A, Berrevoets M, Seghezzi M, Previtali G, Russcher H, Verbon A, et al., “A novel haemocytometric COVID-19 prognostic score developed and validated in an observational multicentreEuropean hospital-based study,” Elife; 9: e63195).

[0184] In contrast, multi-parameter machine learning algorithms are able to learn complex nonlinear relationships between multiple parameters and improve the overall performance of diagnostic tests.

[0185] Embodiments include an ensemble of tree-based algorithms for predicting COVID-19 positive or negative status using hematology data. In an embodiment, the classifier predicts COVID-19 positivity with an accuracy of 74% on the entire test set and 91.7% on hospitalized patients. The better performance on hospitalized individuals gives excellent confidence in the parameter selection process of the present disclosure. As described in Section 3.4, algorithm embodiments can be advantageously used in different settings in epidemic and endemic scenarios, including using it as a faster alternative to RT-PCR testing, reducing the testing burden in epidemics, and detecting COVID-19 in unsuspected asymptomatic individuals. Another study (see Kukar M, G, Vovko T, Podnar S, P, Brvar M, Zalaznik M et al., "COVID-19 diagnosis by routine blood tests using machine learning", Scientific reports 2021; 11: 1-9) have used machine learning algorithms to use routine blood parameters for COVID-19 diagnosis, but those parameters need to be collected from multiple different tests, which contrasts with the embodiments of the present disclosure that use parameters obtained from a single instrument. In addition, Kukar et al. included a skewed data set (3% positive, 97% negative). Although the prevalence of COVID-19 was 3% at the time of the publication, it is believed that AUROC is not the correct metric for severely skewed data sets, and perhaps some other metrics (such as the area under the precision-recall curve (AUPRC)) may be a better metric for reporting cross-validation results. On the other hand, the present disclosure includes a retained test data set (20% of the total data) with a balanced number of positives and negatives to report AUROC and other metrics.

[0186] In an embodiment, the algorithm used described herein includes a total of 48 parameters, including the percentage of various leukocyte types, such as lymphocytes, neutrophils, basophils, monocytes, eosinophils, polymorphonuclear cells, large unstained cells (LUC) and blasts. In an embodiment, parameter set also includes platelet count and several red blood cell (RBC) parameters, such as hemoglobin level, mean corpuscular volume (MCV), cell hemoglobin concentration mean value (g / dL), red blood cell number with low, normal and high hemoglobin concentration and volume, hemoglobin distribution width (HDW) (it is the standard deviation of hemoglobin concentration (HC) histogram, and expressed in g / dL), red blood cell distribution width (RDW), total RBC count. In an embodiment, parameter set also includes some original instrument measurements unique to ADVIA 2120i analyzer. In an embodiment, in addition to the parameters acquired using the ADVIA 2120i, two derived parameters are included in the analysis: the neutrophil to lymphocyte ratio (NLR) and the LUC to lymphocyte ratio. LUC is a parameter specific to the ADVIA 2120i analyzer.

[0187] Some parameters used in the algorithms of the present disclosure have been widely reported to be important in COVID-19 diagnosis and / or prognosis.Although eosinophilia has been identified in the literature as an important feature of COVID-19 patients (e.g., see Lippi G, Plebani M, “Laboratory abnormalities in patients with COVID-2019 infection”, Clinical chemistry and laboratory medicine (CCLM) 2020; 58: 1131-4), lymphocytes have been observed to be important biomarkers in the diagnosis and prognosis of COVID-19 (e.g., see Henry BM, De Oliveira MHS, Benoit S, Plebani M, Lippi G, “Hematologic, biochemical and immune biomarker abnormalities associated with severe illness and mortality in coronavirus disease 2019 (COVID-19): a meta-analysis”, Clinical Chemistry and Laboratory Medicine (CCLM) 2020; 58: 1021-8; Kermali M, Khalsa RK, Pillai K, Ismail Z, Harky A, “The role of biomarkers in diagnosis of COVID-19--A meta-analysis”, Clinical Chemistry and Laboratory Medicine (CCLM) 2020; 58: 1021-8; Kermali M, Khalsa RK, Pillai K, Ismail Z, Harky A, “The role of biomarkers in diagnosis of COVID-19--A meta-analysis”, Clinical Chemistry and Laboratory Medicine (CCLM) 2020; 58: 1021-8; Kermali M, Khalsa RK, Pillai K, Ismail Z, Harky A, “ systematic review", Life sciences 2020;254:117788; Lippi G, Plebani M's "Laboratory abnormalities in patients with COVID-2019infection", Clinicalchemistry and laboratory medicine (CCLM) 2020;58:1131-4; Khartabil T, Russcher H, van der Ven A, De Rijke Y's "A summary of the diagnostic and prognostic value of hemocytometry markers in COVID-19patients”, Criticalreviews in clinical laboratory sciences 2020;57:415-31.Another study reported lower percentages of monocytes, eosinophils, and basophils in severe cases. Leukopenia was observed in 33.7% of hospitalized patients at the time of admission (e.g., see Guan Wj, Ni Zy, Hu Y, Liang Wh, Ou Cq, He Jx, Liu L et al., "Clinical characteristics of coronavirus disease 2019 in China", New England journal of medicine 2020; 382: 1708-20).The prevalence of thrombocytopenia has been widely reported in patients with COVID-19, particularly in more severe patients (e.g., see Henry BM, De Oliveira MHS, Benoit S, Plebani M, Lippi G, “Hematologic, biochemical and immune biomarker abnormalities associated with severe illness and mortality in coronavirus disease 2019 (COVID-19): a meta-analysis,” Clinical Chemistry and Laboratory Medicine (CCLM) 2020;58:1021-8; Jiang SQ, Huang QF, Xie WM, Lv C, Quan XQ, “The association between severe COVID-19 and low platelet count: evidence from 31 observational studies involving 7613 participants,” British journal of hematology 2020:e29-e33; Kermali M, Khalsa RK, Pillai K, Ismail Z, Harky A, “The role of biomarkers in diagnosis of COVID-19--A systematic review", Lifesciences2020;254:117788; Lippi G, Plebani M's "Laboratory abnormalities inpatients with COVID-2019infection", Clinical chemistry and laboratory medicine (CCLM) 2020;58:1131-4; and Lippi G, Plebani M, Henry "Thrombocytopenia is associated with severe coronavirus disease 2019 (COVID-19) infections: a meta-analysis" by BM, Clinica chimica acta2020;506:145-8).Some studies have also reported the impact of COVID-19 on RBC parameters (such as hemoglobin, RDW, and MCV), which are also used in the algorithms of the present disclosure (e.g., see Lippi G, Plebani M, “Laboratory abnormalities in patients with COVID-2019 infection,” Clinical chemistry and laboratory medicine (CCLM) 2020; 58: 1131-4; Khartabil T, Russcher H, van der Ven A, De Rijke Y, “A summary of the diagnostic and prognostic value of hemocytometry markers in COVID-19 patients,” Critical reviews in clinical laboratory sciences 2020; 57: 415-31; and Yu H, Li D, Deng Z, Yang Z, Cai J, Jiang L, Wang K, et al., “Total protein as a biomarker for predicting coronavirus disease-2019 pneumonia,” Preprint with The Lancet 2).

[0188] The present disclosure demonstrates an embodiment for detecting COVID-19 positivity using hematological parameters derived from a single instrument. This is a specific case study in cases of co-existing COVID-19 positives and negatives from Mumbai, India during the first wave of infection in India. The results are excellent.

[0189] This study also highlights the use of common non-specific tests such as CBC / Diff to flag infection and alert doctors to the patient's underlying health condition. This is possible due to the unique feature of hematology instruments that measure a large number of blood parameters simultaneously, the values ​​of which may vary depending on the patient's health condition. In embodiments, such embedded algorithms within hematology instruments can be useful as an early warning system for doctors and aid in patient management.

[0190] Supplement to Example 1

[0191] The training data used for algorithm development here was skewed in terms of gender, containing roughly twice as many males as females. The positive population contained 201 males and 108 females, while the negative population had 116 males and 129 females. Of these, the test set consisted of 43 males and 19 females from the COVID-19 positive population and 22 males and 27 females from the negative population, respectively. However, model performance across males and females was consistent, with prediction accuracy of approximately 74% for both, thus indicating no gender-related bias in the predictions (Supplementary Table 1).

[0192]

[0193] In order to understand the impact of age on algorithm performance, the prediction accuracy of the embodiments of the present disclosure was compared across test sets of different age groups (Supplementary Table 2).

[0194]

[0195] Variations in prediction accuracy were observed across different age groups. This may be due to the small and unequal number of individuals in the test set across different age groups, making it difficult to find out whether age affects algorithm performance. However, the prediction performance of the algorithm for the 40-80 age group (which corresponds to the majority of the data) was fairly consistent.

[0196] The foregoing examples have been provided for the purpose of explanation only and are in no way to be construed as limitations on the invention disclosed herein. Although the present invention has been described with reference to various embodiments, it should be understood that the words used herein are words of description and illustration, rather than restrictive words. Further, although the present invention has been described with reference to specific means, materials, and embodiments, the present invention is not intended to be limited to the details disclosed herein; on the contrary, the present disclosure extends to all functionally equivalent structures, methods, and uses, such as within the scope of the appended claims. A person skilled in the art who benefits from the teachings of this specification may make various modifications thereto, and may make changes in its various aspects without departing from the scope and spirit of the present invention.

[0197] Example 2

[0198] Examples of Prophecy

[0199] Hematology tests such as CBC / Diff (complete blood count with differential) are commonly performed for general health monitoring. In an embodiment, the present disclosure will be from Siemens Healthineers The CBC / Diff results of the 2120i Hematology System are combined with a multi-parameter machine learning algorithm for rapid flagging of pathogenic disease positivity, such as viral, bacterial, or fungal disease positivity. Diagnostic data generation and running the algorithm on the data occur on the same instrument to ensure easy data access. Additionally, the algorithm can also be run on a connected laboratory information system. Such an algorithm is valuable when access to pathogen testing is limited and is used to predict the disease status of unsuspected individuals, prompting confirmatory testing. Various epidemiological and endemic applications of the algorithm are demonstrated.

[0200] Methods. Pathogen-positive and -negative CBC / Diff data were obtained from a pre-selected population. Using the data obtained from them, a machine learning classification model was trained using 5-fold cross validation to distinguish between positive and negative.

[0201] Results. The final ML model was an ensemble of decision tree-based methods with excellent accuracy and suitable AUROC on the test set. Significant features including eosinophil and lymphocyte counts were reported earlier. The model was able to more accurately predict patients who were hospitalized compared to those who were not.

[0202] Conclusion. A predictive algorithm, method and system for rapid pathogenic disease marking using data from routine CBC testing is provided. The algorithm is useful in epidemic or endemic scenarios as an alternative to testing, reducing testing burden and marking unsuspected positives.

[0203] Example 3

[0204] Examples of Prophecy

[0205] Hematology tests such as CBC / Diff (complete blood count with differential) are commonly performed for overall health monitoring. In an embodiment, the present disclosure uses a blood test kit from Siemens Healthineers The CBC / Diff results from the 2120i Hematology System are combined with a multi-parameter machine learning algorithm for rapid flagging of sepsis positivity. Diagnostic data generation and running the algorithm on the data occur on the same instrument to ensure easy data access. Additionally, the algorithm can also be run on a connected laboratory information system. Such an algorithm is valuable when access to sepsis testing is limited and is used to predict sepsis status in unsuspected individuals, prompting confirmatory testing. Various epidemiological and endemic applications of the algorithm are demonstrated.

[0206] Methods. Pathogen-positive and -negative CBC / Diff data were obtained from a pre-selected population. Using the data obtained from them, a machine learning classification model was trained using 5-fold cross validation to distinguish between positive and negative.

[0207] Results. The final ML model was an ensemble of decision tree-based methods with excellent accuracy and suitable AUROC on the test set. Significant features including eosinophil and lymphocyte counts were reported earlier. The model was able to more accurately predict hospitalization patients compared to nonhospitalization patients.

[0208] Conclusion. A predictive algorithm, method and system for rapid sepsis marking using data from routine CBC testing is provided. The algorithm is useful in epidemic or endemic scenarios as an alternative to testing, reducing testing burden and marking unsuspected positives.

[0209] Therefore, according to the present disclosure, there have been provided systems, articles, and methods of production and use thereof that fully satisfy the purposes and advantages set forth above. Although the present disclosure has been described in conjunction with the specific figures, experiments, results, and language set forth above, it is apparent that many substitutions, modifications, and variations will be apparent to those skilled in the art. Therefore, it is intended to include all such substitutions, modifications, and variations that fall within the spirit and broad scope of the present disclosure.

[0210] All references cited herein are hereby incorporated by reference in their entirety.

Claims

1. A system (100) for predicting the presence of a disease in a subject, the system (100) include: one or more processing units (101); a medical database (112) coupled to the one or more processing units (101), the medical database (112) comprising patient data; and The disease prediction module (110) is configured to: receiving a plurality of parameters associated with a subject; determining a threshold for sensitivity and / or specificity associated with the trained machine learning model; using the trained machine learning model to predict the presence of a disease in a subject based on specific thresholds of the model and the plurality of parameters associated with the subject; as well as The prediction is output at an output unit (105).

2. The system (100) of claim 1, wherein a threshold for sensitivity and / or specificity associated with the trained machine learning model is determined based on a need scenario associated with a healthcare provider.

3. A system (100) according to claim 2, wherein the demand scenario associated with the healthcare provider includes at least one of the following: reducing the infectious disease testing burden or the RT-PCR testing burden, replacing the infectious disease testing burden, replacing RT-PCR testing and / or determining the need for subject testing.

4. The system (100) of claim 1, wherein the plurality of parameters associated with the subject include red blood cell (RBC) parameters, platelet parameters, and white blood cell (WBC) parameters.

5. The system (100) according to claim 4, wherein the RBC parameters include one or more of the following: hemoglobin level, hematocrit, RBC size, hemoglobin level in a single RBC, mean corpuscular volume, mean corpuscular hemoglobin concentration, normal RBCs, the number of RBCs with a hemoglobin concentration ≥28 g / dL and ≤41 g / dL, the number of RBCs with a hemoglobin volume ≥60 fL and ≤120 fL, hemoglobin distribution width, RBC mean corpuscular volume, RBC volume distribution width, total RBC count, hemoglobin content distribution width and hemoglobin content mean.

6. The system (100) of claim 4, wherein the WBC parameters include one or more of the following: WBC type, WBC count, WBC percentage, neutrophil to lymphocyte ratio, large unstained cell to lymphocyte ratio, number of WBCs indicating pseudobasophilia, and WBC maturation parameters.

7. The system (100) according to claim 5 further includes an average value of cellular hemoglobin concentration, a number of RBCs with a hemoglobin concentration greater than 41 g / dL, and a number of RBCs with a hemoglobin concentration less than 28 g / dL.

8. The system (100) of claim 1, wherein the trained machine learning model is pre-selected.

9. The system (100) of claim 1, wherein the disease prediction module (110) is further configured to use the trained machine learning model to predict the stratification / severity or need for hospitalization of the subject based on the multiple parameters associated with the subject.

10. The system (100) of claim 1, wherein the disease prediction module (110) is further configured to predict the occurrence of long COVID-19 in a subject.

11. A method (200) for predicting the presence of a disease in a subject, the method (200) include: receiving a plurality of parameters associated with a subject; determining a threshold for sensitivity and / or specificity associated with the trained machine learning model; using the trained machine learning model to predict the presence of a disease in a subject based on specific thresholds of the model and the plurality of parameters associated with the subject; and Output the prediction on the output unit.

12. The method (200) of claim 11, wherein a threshold for sensitivity and / or specificity associated with the trained machine learning model is determined based on a need scenario associated with a healthcare provider.

13. The method (200) of claim 12, wherein the demand scenario associated with the healthcare provider includes at least one of: reducing the burden of RT-PCR or other infectious disease testing, replacing such testing, and / or determining the need for subject testing.

14. The method (200) of claim 11, wherein the disease is COVID-19.

15. The method (200) of claim 11, further comprising: include: using the trained machine learning model, predicting a need for hospitalization of the subject based on the plurality of parameters associated with the subject; and Predicting the occurrence of long-term COVID-19 in subjects.

16. An article of manufacture such as a system or component thereof comprising a non-transitory computer-readable medium having instructions encoded thereon, the instructions being configured to cause one or more processors to perform a method comprising: predicting the presence of a disease in a subject, the method (200) include: receiving a plurality of parameters associated with a subject; determining a threshold for sensitivity and / or specificity associated with the trained machine learning model; using the trained machine learning model to predict the presence of a disease in a subject based on specific thresholds of the model and the plurality of parameters associated with the subject; and Output the prediction on the output unit.

17. A non-transitory computer-readable medium having instructions encoded thereon, the instructions being configured to cause one or more processors to perform a method (200) of predicting the presence of a disease in a subject, the method (200) include: receiving a plurality of parameters associated with a subject; determining a threshold for sensitivity and / or specificity associated with the trained machine learning model; using the trained machine learning model to predict the presence of a disease in a subject based on specific thresholds of the model and the plurality of parameters associated with the subject; and Output the prediction on the output unit.

Images