Systems and methods for detecting acute kidney disease

Abstract

In some embodiments, a method of training a machine-learning (ML) model for use in detecting acute kidney injury (AKI) is provided that includes (1) obtaining a dataset of detected biomarkers; (2) selecting a plurality of routine biomarkers from the dataset; and (3) training an ML model to identify AKI based on the plurality of routine biomarkers by employing the plurality of routine biomarkers as inputs and presence of AKI and absence of AKI as target values. Numerous other embodiments are provided.

Description

SYSTEMS AND METHODS FOR DETECTING ACUTE KIDNEY DISEASECROSS-REFERENCE TO RELATED APPLICATIONS

[0001] The present patent application claims priority to United States Serial No. 63 / 578,380, filed on August 24, 2023, the entire content of which is hereby incorporated herein by reference.FIELD

[0002] The present application relates to medical testing and more particularly to systems and methods for detecting acute kidney disease.BACKGROUND

[0003] Acute kidney injury (AKI) is a sudden and often reversible loss of kidney function that represents a major public health burden. AKI has high morbidity and mortality rates, particularly in elderly individuals and those with underlying comorbidities such as diabetes, hypertension, and cardiovascular disease. If left untreated, AKI may result in permanent kidney damage and increased mortality rates.

[0004] AKI is usually a secondary condition that can be caused by a variety of conditions and factors which damage the kidneys and affect their ability to function properly, such as sepsis, cancer, infection, pregnancy, certain medications, aberrant glucose metabolism, decreased blood flow to the kidneys, and many more. AKI is characterized by a rapid decrease in kidney function, which is usually detected by a decrease in urine output and / or a rise in serum creatinine. The underlying pathophysiology may include damage to the kidney's functional units, leading to decreased production of urine and accumulation of waste products in the blood.

[0005] As stated, AKI is diagnosed primarily based on changes in serum creatinine levels and urine output. However, these markers are not always reliable, and AKI can be difficult to diagnose early, particularly in patients with underlying kidney disease or other comorbidities. A need therefore exists for improved systems and methods for detecting acute kidney disease, and in particular acute kidney injury.SUMMARY

[0006] In some embodiments, a method of training a machine-learning (ML) model for use in detecting acute kidney injury (AKI) includes obtaining a dataset of detected biomarkers; selecting a plurality of routine biomarkers from the dataset; and trainingan ML model to identify AKI based on the plurality of routine biomarkers by employing the plurality of routine biomarkers as inputs and presence of AKI and absence of AKI as target values.

[0007] In some embodiments, a method of training an ML model for use in detecting AKI includes obtaining a dataset of detected biomarkers; selecting a plurality of routine biomarkers from the dataset; training an ML model to identify AKI based on the plurality of routine biomarkers by employing the plurality of routine biomarkers as inputs and presence of AKI and absence of AKI as target values; determining a subset of the plurality of routine biomarkers that are altered in the presence of AKI; and retraining the ML model based on the subset of the plurality of routine biomarkers.

[0008] In some embodiments, a method of training an ML model for use in detecting AKI includes obtaining a dataset of detected biomarkers; selecting a plurality of routine biomarkers from the dataset; obtaining one or more additional biomarkers determined for patients suspected of having AKI; training an ML model to identify AKI based on the plurality of routine biomarkers and the one or more additional biomarkers, the ML model employing (1 ) the plurality of routine biomarkers and the one or more additional biomarkers as inputs; and (2) presence of AKI and absence of AKI as target values.

[0009] In some embodiments, a method of diagnosing AKI includes obtaining routine biomarkers for a patient at a hospital; inputting the routine biomarkers for the patient through a first ML model to determine an initial AKI score, the first ML model trained using routine biomarkers as inputs and presence of AKI and absence of AKI as target values; based on the initial AKI score, determining if the patient is at an elevated risk of AKI; and if the patient is at an elevated risk of AKI, recommending at least one additional test for the patient.

[0010] Other features and aspects of the present invention will become more fully apparent from the following detailed description, the appended claims, and the accompanying drawings.BRIEF DESCRIPTION OF THE DRAWINGS

[0011] FIG. 1A illustrates an example flow diagram of a method of training and deploying a machine-learning (ML) model based on routine biomarker data in accordance with one or more embodiments.

[0012] FIG. 1 B illustrates an example computer in which the method of FIG. 1A may be implemented in accordance with one or more embodiments.

[0013] FIG. 1 C illustrates an example flow diagram of a method of training and deploying a staged ML model based on routine biomarker data and additional biomarker data in accordance with one or more embodiments.

[0014] FIG. 1 D illustrates an embodiment of a computer in which the method of FIG. 1 C may be implemented in accordance with one or more embodiments.

[0015] FIG. 2 illustrates a flowchart of a method of training an ML model for identifying acute kidney injury (AKI) in accordance with one or more embodiments.

[0016] FIG. 3 illustrates a flowchart of another method of training an ML model for identifying AKI in accordance with one or more embodiments.

[0017] FIG. 4 illustrates a flowchart of a method of diagnosing AKI in accordance with one or more embodiments.

[0018] FIG. 5A illustrates an example first ML model in accordance with one or more embodiments.

[0019] FIG. 5B illustrates an example second ML model in accordance with one or more embodiments.

[0020] FIG. 6A illustrates a first AKI detection system in accordance with one or more embodiments.

[0021] FIG. 6B illustrates a second AKI detection system in accordance with one or more embodiments.DETAILED DESCRIPTION

[0022] Independent of the grammatical term usage, individuals with male, female or other gender identities are included within the term.

[0023] As described above, accurately identifying AKI is difficult and the consequences of a failed diagnosis may be severe. In accordance with embodiments provided herein, a machine-learning (ML) model is trained to identify AKI using routine biomarkers commonly measured in hospital emergency rooms, outpatient or inpatient clinics, doctors’ offices, walk-in laboratories, large commercial laboratories, or the like. By relying on routine biomarkers, the trained ML model may be able to quickly determine a patient’s risk of AKI in most clinical settings, allowing medical staff to start treatment earlier and / or order follow-up testing to confirm a diagnosis. In at least some embodiments, use of a single-stage ML model trained on routine biomarkers wasfound to have approximately half the false positive rate of more targeted, manual AKI testing.

[0024] As used herein, “routine” biomarkers refer to common biomarkers determined for patients during hospital-related visits such as to emergency rooms, inpatient and outpatient clinics, doctors’ offices, or laboratories associated with hospitals. Example routine tests include blood pressure measurements, weight measurements, urine tests, blood tests such as a lipid panel, complete blood count (CBC), basic metabolic panel (BMP), and comprehensive metabolic panel (CMP), or the like. Example routine biomarkers may include creatinine, BUN, BUN / creatinine ratio, chloride, red blood cell width (RDW), sodium, albumin, hematocrit, platelet count, red or white blood cell counts, neutrophil, lymphocyte, eosinophil, blood pressure, weight, height, urine pH, urine volume, urine glucose, 24-hour urine analysis, other biomarkers associated with macroscopic or microscopic urinalysis, bilirubin, prostatespecific antigen, and / or other biomarkers routinely used for assessing the health of a patient (e.g., in a clinical or similar setting). As used herein, routine biomarkers do not include biomarkers that are detectable by more specialized tests such as cystatin C, interleukin-18 (IL-18), calprotectin, carcinoembryonic antigen, alpha-fetoprotein, tumor mutation burden, tumor infiltrating lymphocytes, immunoreactive trypsinogen, neurofilament light chain, insulin-like growth factor binding protein 7 (IGFBP7), livertype fatty acid binding protein (L-FABP), tissue inhibitor of metalloproteinases 2 (TIMP2), a combination of IGFBP7 and TIMP2, fractional excretion of sodium, fractional excretion of urea, combinations of such biomarkers, etc.

[0025] In some embodiments, routine biomarkers may be examined to determine their relevance to AKI detection and to eliminate redundant biomarkers (e.g., using maximum relevance minimum redundancy (MRMR) feature selection or another suitable algorithm). Thereafter, an ML model may be trained by employing a plurality of selected routine biomarkers as inputs and presence of AKI and absence of AKI as target values. Once trained, in one or more embodiments, the most relevant routine biomarkers for AKI detection may be identified and an additional retraining of the ML model may be performed (e.g., to further increase model prediction accuracy). Such an ML model may be easily updated with additional biomarkers relevant to AKI detection when suitable biomarker tests are identified.

[0026] In another example implementation, a multi-stage ML model may be employed for AKI detection. A first ML model may be trained on a selection of routinebiomarkers to provide an initial AKI score (e.g., an indication of at least one of absence of AKI, presence of AKI, and probability of having AKI). A second ML model may be trained on routine biomarkers as well as biomarkers more specifically associated with AKI such as cystatin C, neutrophil gelatinase-associated lipocalin (NGAL), IL-18, IGFBP7, calprotectin, L-FABP, TIMP2, a combination of IGFBP7 and TIMP2, fractional excretion of sodium, fractional excretion of urea, combinations of two or more of such biomarkers, etc. If the initial AKI score from the first L model indicates that a patient likely has AKI (e.g., based on a probability exceeding a predetermine threshold, for example), one or more additional tests relevant to AKI may be recommended for the patient. The second ML model may then employ biomarkers from the additional tests (and / or one or more routine biomarkers for the patient) to provide an updated AKI score.

[0027] These and other embodiments of the invention are described below with reference to FIGS. 1A-6B.

[0028] FIG. 1A illustrates an example flow diagram 100a of a method of training and deploying a machine-learning model based on routine biomarker data in accordance with embodiments provided herein. With reference to FIG. 1A, and as described in further detail below, the method of flow diagram 100a includes obtaining a detected biomarkers dataset 102 (e.g., from a hospital, a laboratory, a healthcare data provider, etc.). The detected biomarkers dataset 102 is then employed to generate a training dataset (routine biomarkers dataset 104). For example, in some embodiments, the detected biomarker dataset 102 is examined and routine biomarkers are identified. Routine biomarkers may be examined to determine their relevance to AKI detection and to eliminate redundant biomarkers (e.g., using maximum relevance minimum redundancy (MRMR) feature selection or another suitable algorithm). A plurality of the routine biomarkers may then be employed to form the routine biomarkers dataset 104. Thereafter, the routine biomarkers dataset 104 may be employed to train a machine-learning (ML) model 106. The trained ML model 106 may be used within a clinical setting as a deployed ML model 108 to predict whether a patient has AKI and / or a probability of the patient having AKI based on routine biomarkers measured for the patient.

[0029] FIG. 1 B illustrates an example computer 120 in which the method of FIG. 1 A may be implemented in accordance with one or more embodiments. With reference to FIG. 1 B, computer 120 includes a processor 122 coupled to a memory 124. Memory124 may include detected biomarkers dataset 102, routine biomarkers (e.g., training) dataset 104, and ML model 106. Memory 124 may also include one or more programs 126 for carrying out the methods described herein when executed by processor 122, such as examining detected biomarkers dataset 102 for routine biomarkers, identifying and / or eliminating redundant routine biomarkers, creating routine biomarkers dataset 104 based on detected biomarkers dataset 102, and the like. In some embodiments, processor 122, executing one or more of programs 126, may train ML model 106 based on routine biomarkers dataset 104. Memory 124 may include multiple memory units and / or types of memory. In some embodiments, all or a portion of memory 124 may be external to and / or remote from computer 120. Additionally, in some embodiments, multiple processors may be employed.

[0030] FIG. 1 C illustrates an example flow diagram 100b of a method of training and deploying a staged machine-learning model based on routine biomarker data and additional biomarker data in accordance with embodiments provided herein. With reference to FIG. 1 C, and as described in further detail below, the method of flow diagram 100b includes obtaining a detected biomarkers dataset 102 (e.g., from a hospital, a laboratory, a healthcare data provider, etc.). The detected biomarkers dataset 102 is then employed to generate a training dataset (routine biomarkers dataset 104). Thereafter, the routine biomarkers dataset 104 is employed to train a first ML model 106 (e.g., as described above).

[0031] The method of flow diagram 100b also includes obtaining additional biomarkers for patients suspected of having AKI. In some embodiments, additional biomarkers such as routine biomarkers known to be associated with AKI and / or other biomarkers more specifically associated with AKI such as cystatin C, IL-18, IGFBP7, calprotectin, L-FABP, TIMP2, etc., may be added to the training dataset of biomarkers to create a routine and additional biomarkers dataset 110. Thereafter, the routine and additional biomarkers dataset 1 10 is employed to train a second machine-learning (ML) model 112.

[0032] The trained, first ML model 106 and the trained, second ML model 112 may be used within a clinical setting (as deployed, first and second ML models) to predict whether a patient has an elevated risk of AKI and / or a probability of the patient having AKI based on routine and / or additional biomarkers measured for the patient. For example, if the first ML model 106 indicates that a patient has an elevated risk of AKI, additional testing may be performed on the patient to generate additional biomarkersthat are fed into the second ML model 112 to obtain an updated AKI score as described further below.

[0033] FIG. 1 D illustrates an embodiment of computer 120 in which the method of FIG. 1 C may be implemented in accordance with one or more embodiments. With reference to FIG. 1 D, in addition to the components, datasets and programs of FIG. 1 B, computer 120 may include the routine and additional biomarkers dataset 110 and the second machine learning model 112. Program 126 may be updated to include the additional functionality associated with training and implementing the second ML model 112.

[0034] FIG. 2 illustrates a flowchart of a method 200 of training a machine-learning model for identifying acute kidney injury (AKI) in accordance with one or more embodiments. With reference to FIG. 2, method 200 includes, in block 202, obtaining a dataset of detected biomarkers. The detected biomarkers dataset may be obtained from a hospital, a laboratory, a healthcare data provider such as Dandelion Health, Inc. of New York, NY, Prognos Health, Inc. of New York, NY, or the like, for example. FIGS. 1A and 1 B illustrate an example detected biomarkers dataset 102.

[0035] In some embodiments, the detected biomarkers dataset may contain biomarkers determined for at least one of a plurality of hospitals and a plurality of laboratories. In other embodiments, the detected biomarkers dataset may contain only biomarkers determined for a specific hospital (e.g., biomarkers determined for at least one of an emergency room, an in-patient clinic, an outpatient clinic, a doctor’s office, and a laboratory affiliated with the specific hospital).

[0036] After the dataset of detected biomarkers is obtained, in block 204, method 200 includes selecting a plurality of routine biomarkers from the dataset. For example, routine biomarkers from within the dataset of detected biomarkers (e.g., detected biomarkers dataset 102) may be selected such as creatinine, BUN, BUN / creatinine ratio, chloride, red blood cell width (RDW), sodium, albumin, hematocrit, platelet count, red or white blood cell counts, neutrophil, lymphocyte, eosinophil, blood pressure, weight, urine pH, urine volume, urine glucose, bilirubin, prostate-specific antigen, and / or other biomarkers routinely used for assessing the health of a patient. The plurality of routine biomarkers selected from the dataset of detected biomarkers may be used to form a training dataset (e.g., routine biomarkers dataset 104 of FIGS. 1A and 1 B). In some embodiments, routine biomarkers may be examined to determine their relevance to AKI detection and to eliminate redundant biomarkers (e.g., usingmaximum relevance minimum redundancy (MRMR) feature selection or another suitable algorithm). In this manner, an optimized training dataset of routine biomarkers may be obtained (e.g., the routine biomarkers dataset 104).

[0037] After the plurality of routine biomarkers is selected (e.g., after routine biomarkers dataset 104 has been formed), in block 206, method 200 includes training an ML model to identify AKI based on the plurality of routine biomarkers by employing the plurality of routine biomarkers as inputs and presence of AKI and absence of AKI as target values. For example, routine biomarkers dataset 104 (FIG. 1A) may be employed to train ML model 106 using routine biomarkers from the routine biomarkers dataset 104 as inputs and presence of AKI and absence of AKI as target values. In some embodiments, the ML model 106 may also be trained with demographic information (e.g., age, sex, gender, etc.).

[0038] Example ML algorithms that may be employed within ML model 106 and / or deployed ML model 108 include tree-based ML algorithms such as decision trees, gradient boosted trees (e.g., XGBoost or LightGBM), random forests, etc. Other ML algorithms may be used such as a deep learning algorithm (e.g., a deep neural network (DNN)), a convolutional neural network (CNN), a region-based CNN (R-CNN), a fully convolutional neural network (FCN), a region-based FCN (R-FCN), etc. Other suitable neural networks may be employed. Example architectures include Inception, ResNet, ResNeXt, DenseNet, or the like, although other architectures may be employed.

[0039] In some embodiments, after training, ML model 106 may be fine tuned and / or validated. For example, method 200 may include, in block 208, determining a subset of the plurality of routine biomarkers that are altered in the presence of AKI, and in block 210, retraining the ML model based on the subset of the plurality of routine biomarkers. In one or more embodiments, Shapley additive explanations (SHAP) feature importance, decision tree feature importance, permutation feature importance, or a similar technique may be employed to identify the most relevant biomarkers employed during training (e.g., those that impact an AKI diagnosis). These identified biomarkers may then be employed to update the training of the ML model 106 by retraining the ML model 106 using the identified biomarkers (e.g., such as by giving the identified biomarkers more weight or omitting biomarkers with low importance).

[0040] Once the ML model 106 has been trained, it may be deployed and used in a clinical setting (as deployed ML model 108). For example, a hospital may employdeployed ML model 108 to quickly screen patients for AKI using biomarkers obtained for patients from routine blood and / or urine tests.

[0041] In at least some embodiments, the output of deployed ML model 108 may be an AKI score for a patient indicative of the presence or absence of AKI (or probability of having AKI). Further, in one or more embodiments, the deployed ML model 108 may determine and / or output which biomarkers of the patient contributed to the AKI score (e.g., by using SHAP feature importance, permutation feature importance, or a similar technique to identify the most relevant biomarkers contributing to the AKI score).

[0042] FIG. 3 illustrates a flowchart of another method 300 of training an ML model for identifying AKI in accordance with one or more embodiments. With reference to FIG. 3, method 300 includes, in block 302, obtaining a dataset of detected biomarkers (e.g., detected biomarkers database 102 of FIGS. 1A-1 D such as a biomarkers database from a hospital, a laboratory, a healthcare data provider, or the like).

[0043] After the dataset of detected biomarkers is obtained, in block 304, method 300 includes selecting a plurality of routine biomarkers from the dataset. For example, routine biomarkers from within the detected biomarkers dataset 102 may be selected as described above with reference to block 204 of method 200 (FIG. 2).

[0044] Method 300 further includes, in block 306, obtaining one or more additional biomarkers determined for patients suspected of having AKI. In some embodiments, additional biomarkers such as routine biomarkers known to be associated with AKI and / or other biomarkers more specifically associated with AKI such as cystatin C, IL- 18, IGFBP7, calprotectin, L-FABP, TIMP2, etc., may be added to the training dataset of biomarkers (e.g., routine biomarkers dataset 104).

[0045] Once the training dataset has been formed, method 300 includes, in block 308, training an ML model (e.g., ML model 106 of FIG. 1A) to identify AKI based on the plurality of routine biomarkers and the one or more additional biomarkers by employing the plurality of routine biomarkers and the one or more additional biomarkers as inputs, and presence of AKI and absence of AKI as target values.

[0046] In some embodiments, after training, ML model 106 may be fine tuned and / or validated. For example, a subset of the plurality of routine biomarkers or additional biomarkers contributing to the AKI score may be determined using SHAP feature importance, permutation feature importance, or a similar feature importance technique. These identified biomarkers may then be employed to update the trainingof the ML model 106. In some embodiments, the ML model 106 may also be trained with demographic information.

[0047] FIG. 4 illustrates a flowchart of a method 400 of diagnosing AKI in accordance with one or more embodiments. With reference to FIG. 4, method 400 includes, in block 402, obtaining routine biomarkers for a patient at a hospital. For example, a routine blood test and / or urine test may be performed on a patient in a clinical setting such as an emergency room, inpatient or outpatient clinic, doctor’s office, or laboratory associated with a hospital. Example routine tests include CBC, BMP, CMP, urine volume, urine pH, etc.

[0048] Method 400 includes, at block 404, inputting the routine biomarkers for the patient through a first ML model to determine an initial AKI score, the first ML model trained using routine biomarkers as inputs and presence of AKI and absence of AKI as target values. For example, routine biomarkers for a patient may be fed through first ML model 106 (FIGS. 1C and 1 D) to produce an initial AKI score.

[0049] FIG. 5A illustrates an example first ML model 500a in accordance with embodiments provided herein. The first ML model 500a may be trained using routine biomarkers as inputs and presence of AKI and absence of AKI as target values. In response to routine biomarkers from a patient, first ML model 500a outputs an initial AKI score (e.g., indication of a positive or negative AKI diagnosis, a probability of AKI, etc.). While shown as a tree-based model, it will be understood that other ML models may be employed.

[0050] Method 400 further includes, in block 406, based on the initial AKI score, determining if the patient is at an elevated risk of AKI. In some embodiments, this may include comparing the initial AKI score to a predetermined and / or user specified threshold. If the AKI score exceeds the threshold, the patient may be determined to be at an elevated risk of AKI. Selection of the threshold for such a determination may be based on numerous criteria such as hospital requirements, risk factors of the patient such as comorbidities, age, etc., patient history, or the like. Merely for illustrative purposes an example threshold may be any nonzero probability such that an AKI score of greater than 0% (e.g., alone or in combination with other factors such as patient comorbidities, age, etc.) may trigger a sufficiently elevated risk of AKI.

[0051] If an elevated risk of AKI is not present, method 400 proceeds to block 408 and a negative AKI diagnosis is provided (e.g., to medical staff); otherwise, if an elevated risk of AKI is determined, method 400 proceeds to block 410.

[0052] In block 410, method 400 includes recommending at least one additional test for the patient. For example, one or more tests for specific routine biomarkers or non-routine biomarkers relevant to AKI detection may be recommended, such as tests for urinary TIM P2 and IGFBP7, plasma Cystatin C, etc.

[0053] In some embodiments, method 400 includes, in block 412, obtaining one or more biomarkers based on the one or more additional tests recommended for the patient. Thereafter, in block 414, method 400 includes inputting the one or more biomarkers (based on the at least one additional test for the patient) to a second ML model (e.g., second ML model 112 in FIGS. 1C and 1 D) to determine an updated AKI score. The second ML model may be trained using (1 ) routine biomarkers and one or more additional biomarkers for patients suspected of having AKI as inputs; and (2) presence of AKI and absence of AKI as target values.

[0054] FIG. 5B illustrates an example second ML model 500b in accordance with embodiments provided herein. The second ML model 500b may be trained using routine biomarkers and one or more additional biomarkers for patients suspected of having AKI as inputs (e.g., AKI related biomarkers such as urinary TIMP2 and IGFBP7, plasma Cystatin C, etc.) and presence of AKI and absence of AKI as target values. In response to routine biomarkers and / or one or more additional biomarkers from a patient, second ML model 500b outputs an updated AKI score (e.g., an indication of a positive or negative AKI diagnosis, a probability of AKI, etc.). While shown as a treebased model, it will be understood that other ML models may be employed.

[0055] As described above, in some embodiments, a multi-stage machine-learning approach for AKI diagnosis may be provided. Such an approach may employ feature importance exploration during model training and model explanation generation (e.g., on a single patient level for model prediction). For example, SHAP feature importance or a similar technique may be employed to identify the most relevant biomarkers during training as well as to identify feature biomarkers to explain model predictions to tested patients. In one or more embodiments, to improve the training dataset of routine biomarkers, MRMR or a similar technique may be employed to identify relevant biomarkers and eliminate redundant biomarkers from the training set.

[0056] In an example implementation, a first-stage ML model (e.g., first ML model 106, 500a or 606) may examine routine biomarkers (e.g., blood biomarkers) with or without a baseline (e.g., a baseline creatinine value). Feature importance techniques may be employed post training to confirm or update biomarker (e.g., feature) selection.The results from the first-stage ML model may be used to increase the pretest probability for a second-stage ML model (e.g., second ML model 112, 500b or 612) trained on biomarkers more relevant to AKI detection (e.g., specific routine biomarkers or non-routine urine or blood biomarkers such as urinary TIMP2 and IGFBP7 and plasma Cystatin C). Thus, in some embodiments, the results of the first-stage ML model may be used to recommend one or more additional tests for a patient, and biomarkers from the additional tests may be fed to the second-stage ML model to generate a more accurate AKI prediction. As with the first-stage ML model, contributions of individual biomarkers to an individual patient's test results may be determined for explaining model results to patients (e.g., using SHAP feature importance or similar techniques).

[0057] Embodiments of the present invention allow for fast initial screening (e.g., via a first-stage ML model) for AKI based on a subset of routine blood or other biomarkers with or without baseline data. Such an examination of multidimensional data may be performed faster and more accurately with the trained ML models described herein than may be performed by clinicians. Trained ML models may quickly recognize multiple patterns in multidimensional data, and feature importance analysis may identify a subset of routine biomarkers relevant for AKI diagnosis. For cases with elevated AKI probability, additional, more-specific tests may be ordered without delay for use in the more AKI-targeted second-stage ML model (e.g., a model with higher sensitivity and specificity to false positive and false negative diagnoses).

[0058] As stated, in some embodiments, the routine biomarkers used to train the first and second ML models may be from the hospital using the ML models. Further, in some embodiments, the one or more additional biomarkers (for patients suspected of having AKI) used to train the second ML model may be from the hospital using the ML models.

[0059] In one or more embodiments, the computer 120 and / or processor 122 (and / or processor 604) may recommend treatment based on an AKI score, such as blood pressure reduction and / or electrolytes in response to an elevated AKI score. Further, in some embodiments, computer 120 and / or processor 122 (and / or processor 604) may recommend additional testing based on an AKI score. In yet other embodiments, a recommendation of additional testing may be based on an AKI score in combination with one or more risk factors of a patient (e.g., one or more of age, blood pressure, kidney disease and diabetes).

[0060] FIG. 6A illustrates a first AKI detection system 600a provided in accordance with one or more embodiments. With reference to FIG. 6A, first AKI detection system 600a includes a first analyzer 602a configured to determine one or more biomarkers of a patient blood sample (e.g., creatinine, BUN, BUN / creatinine ratio, chloride, red blood cell width (RDW), sodium, albumin, hematocrit, platelet count, red orwhite blood cell counts, neutrophil, lymphocyte, eosinophil, etc.). Resultant biomarkers are fed to a processor 604 and run through a deployed ML model 606 stored in a memory 608. For example, ML model 606 may be similar to the first ML model 106 (FIG. 1A). ML model 606 provides an AKI score relating to the presence, absence, and / or probability of AKI based on the biomarkers from the patient sample. A program 610 executed by processor 604 may receive biomarkers from first blood analyzer 602a (e.g., directly, from medical staff or via a central computer system), use the ML model 606 to determine an AKI score, and / or make recommendations (e.g., regarding treatment, further testing, etc.).

[0061] FIG. 6B illustrates a second AKI detection system 600b provided in accordance with one or more embodiments. With reference to FIG. 6B, second AKI detection system 600b includes a first analyzer 602a configured to determine one or more biomarkers of a patient blood sample. Resultant biomarkers are fed to processor 604 and run through first ML model 606 stored in memory 608. For example, first ML model 606 may be similarto first ML model 106 (FIG. 1 C). First ML model 606 provides an initial AKI score relating to the presence, absence, and / or probability of AKI based on the biomarkers from the patient sample. A program 610 executed by processor 604 may receive biomarkers from first analyzer 602a (e.g., directly, from medical staff or via a central computer system), use the first ML model 606 to determine an initial AKI score and / or make recommendations. For example, in some embodiments, if the AKI score indicates an elevated risk of AKI, the program 610 may communicate that additional tests are recommended for the patient.

[0062] In some embodiments, one or more additional analyzers (only a second analyzer 602b is shown in FIG. 6B) may be employed to analyze a urine, blood or other sample from the patient and determine one or more additional biomarkers (e.g., urinary TIMP2 and IGFBP7 and / or plasma Cystatin C). Resultant additional biomarkers may be fed to processor 604 and run through a second ML model 612 stored in memory 608. For example, second ML model 612 may be similar to second ML model 112 (FIG. 1C). Second ML model 612 may provide an updated AKI scorerelating to the presence, absence, and / or probability of AKI based on the biomarkers from the additional patient sample(s). Program 610 executed by processor 604 may receive biomarkers from second analyzer 602b and / or other analyzers not shown (e.g., directly, from medical staff or via a central computer system), use the second ML model 612 to determine an updated AKI score based on the additional biomarkers, and / or make recommendations.NON-LIMITING ILLUSTRATIVE EMBODIMENTS

[0063] Illustrative embodiment 1. A method of training a machine-learning (ML) model for use in detecting acute kidney injury (AKI), comprising: obtaining a dataset of detected biomarkers; selecting a plurality of routine biomarkers from the dataset; and training an ML model to identify AKI based on the plurality of routine biomarkers by employing the plurality of routine biomarkers as inputs and presence of AKI and absence of AKI as target values.

[0064] Illustrative embodiment 2. The method of illustrative embodiment 1 , wherein the dataset contains biomarkers determined for at least one of a plurality of hospitals and a plurality of laboratories.

[0065] Illustrative embodiment 3. The method according to one of the preceding embodiments, wherein the dataset contains biomarkers determined for a specific hospital.

[0066] Illustrative embodiment 4. The method according to one of the preceding embodiments wherein the biomarkers for the specific hospital include biomarkers determined for at least one of an emergency room, an in-patient clinic, an outpatient clinic, a doctor’s office and a laboratory affiliated with the specific hospital.

[0067] Illustrative embodiment 5. The method according to one of the preceding embodiments wherein selecting the plurality of routine biomarkers includes at least one of identifying relevant biomarkers and removing redundant biomarkers.

[0068] Illustrative embodiment 6. The method according to one of the preceding embodiments further comprising: determining a subset of the plurality of routine biomarkers that are altered in the presence of AKI; and retraining the ML model based on the subset of the plurality of routine biomarkers.

[0069] Illustrative embodiment 7. The method according to one of the preceding embodiments wherein training the ML model further comprises training the ML model with demographic information.

[0070] Illustrative embodiment 8. The method according to one of the preceding embodiments further comprising deploying the ML model for use in a hospital.

[0071] Illustrative embodiment 9. The method according to one of the preceding embodiments further comprising: employing the deployed ML model on a patient sample from a patient within the hospital.

[0072] Illustrative embodiment 10. The method according to one of the preceding embodiments further comprising determining an AKI score indicative of the presence or absence of AKI using the ML model.

[0073] Illustrative embodiment 11 . The method according to one of the preceding embodiments wherein the AKI score is determined without baseline biomarkers for the patient.

[0074] Illustrative embodiment 12. The method according to one of the preceding embodiments wherein the AKI score is determined with baseline biomarkers for the patient.

[0075] Illustrative embodiment 13. The method according to one of the preceding embodiments wherein the AKI score indicates at least one of absence of AKI, presence of AKI, and probability of having AKI.

[0076] Illustrative embodiment 14. The method according to one of the preceding embodiments further comprising determining which biomarkers of the patient contributed to the AKI score.

[0077] Illustrative embodiment 15. The method according to one of the preceding embodiments further comprising recommending at least one of blood pressure reduction and electrolytes based on the AKI score.

[0078] Illustrative embodiment 16. The method according to one of the preceding embodiments further comprising recommending additional testing based on the AKI score.

[0079] Illustrative embodiment 17. The method according to one of the preceding embodiments wherein recommending additional testing based on the AKI score includes recommending additional testing based on the AKI score in combination with one or more risk factors of the patient.

[0080] Illustrative embodiment 18. The method according to one of the preceding embodiments wherein the one or more risk factors include at least one of age, blood pressure, kidney disease and diabetes.

[0081] Illustrative embodiment 19. The method according to one of the preceding embodiments further comprising employing a biomarker from the additional testing as an input to a second ML model, the second ML model trained using: routine biomarkers and one or more additional biomarkers determined by the hospital for patients suspected of having AKI as inputs; and presence of AKI and absence of AKI as target values.

[0082] Illustrative embodiment 20. The method according to one of the preceding embodiments wherein the one or more additional biomarkers determined by the hospital for patients suspected of having AKI are from non-routine tests.

[0083] Illustrative embodiment 21 . The method according to one of the preceding embodiments further comprising determining a second AKI score indicative of the presence or absence of AKI using the second ML model.

[0084] Illustrative embodiment 22. A method of training a machine-learning (ML) model for use in detecting acute kidney injury (AKI), comprising: obtaining a dataset of detected biomarkers; selecting a plurality of routine biomarkers from the dataset; training an ML model to identify AKI based on the plurality of routine biomarkers by employing the plurality of routine biomarkers as inputs and presence of AKI and absence of AKI as target values; determining a subset of the plurality of routine biomarkers that are altered in the presence of AKI; and retraining the ML model based on the subset of the plurality of routine biomarkers.

[0085] Illustrative embodiment 23. The method according to one of the preceding embodiments wherein the dataset contains biomarkers determined for a specific hospital.

[0086] Illustrative embodiment 24. A method of training a machine-learning (ML) model for use in detecting acute kidney injury (AKI), comprising: obtaining a dataset of detected biomarkers; selecting a plurality of routine biomarkers from the dataset; obtaining one or more additional biomarkers determined for patients suspected of having AKI; training an ML model to identify AKI based on the plurality of routine biomarkers and the one or more additional biomarkers, the ML model employing: the plurality of routine biomarkers and the one or more additional biomarkers as inputs; and presence of AKI and absence of AKI as target values.

[0087] Illustrative embodiment 25. The method according to one of the preceding embodiments further comprising: determining a subset of the plurality of routinebiomarkers that are altered in the presence of AKI; and retraining the ML model based on the subset of the plurality of routine biomarkers.

[0088] Illustrative embodiment 26. The method according to one of the preceding embodiments wherein retraining the ML model further comprises retraining the ML model based on the one or more additional biomarkers.

[0089] Illustrative embodiment 27. The method according to one of the preceding embodiments wherein the dataset contains biomarkers determined for a specific hospital.

[0090] Illustrative embodiment 28. The method according to one of the preceding embodiments further comprising deploying the ML model for use in a hospital.

[0091] Illustrative embodiment 29. The method according to one of the preceding embodiments further comprising: employing the deployed, ML model on a patient sample from a patient within the hospital.

[0092] Illustrative embodiment 30. The method according to one of the preceding embodiments further comprising: determining an AKI score indicative of the presence or absence of AKI from the ML model.

[0093] Illustrative embodiment 31. A method of diagnosing acute kidney injury (AKI) comprising: obtaining routine biomarkers for a patient at a hospital; inputting the routine biomarkers for the patient through a first machine-learning (ML) model to determine an initial AKI score, the first ML model trained using routine biomarkers as inputs and presence of AKI and absence of AKI as target values; based on the initial AKI score, determining if the patient is at an elevated risk of AKI; and if the patient is at an elevated risk of AKI, recommending at least one additional test for the patient.

[0094] Illustrative embodiment 32. The method according to one of the preceding embodiments further comprising: obtaining one or more biomarkers based on the at least one additional test for the patient; and inputting the one or more biomarkers based on the at least one additional test for the patient to a second ML model to determine an updated AKI score, the second ML model trained using: routine biomarkers and one or more additional biomarkers for patients suspected of having AKI as inputs; and presence of AKI and absence of AKI as target values.

[0095] Illustrative embodiment 33. The method according to one of the preceding embodiments wherein the routine biomarkers used to train the first and second ML models are from the hospital.

[0096] Illustrative embodiment 34. The method according to one of the preceding embodiments wherein the routine biomarkers used to train the first and second ML models are from a different hospital.

[0097] Illustrative embodiment 35. The method according to one of the preceding embodiments wherein the one or more additional biomarkers for patients suspected of having AKI used to train the second ML model are from the hospital.

[0098] Illustrative embodiment 36. The method according to one of the preceding embodiments wherein the one or more additional biomarkers for patients suspected of having AKI used to train the second ML model are from a different hospital.

[0099] Illustrative embodiment 37. The method according to one of the preceding embodiments further comprising recommending at least one of blood pressure reduction and electrolytes based on the AKI score.

[0100] The foregoing description discloses only example embodiments of the invention. Modifications of the above disclosed apparatus and methods which fall within the scope of the invention will be readily apparent to those of ordinary skill in the art.

[0101] Accordingly, while the present invention has been disclosed in connection with example embodiments thereof, it should be understood that other embodiments may fall within the spirit and scope of the invention, as defined by the following claims.

Claims

WHAT IS CLAIMED IS:

1. A method of training a machine-learning (ML) model for use in detecting acute kidney injury (AKI), comprising: obtaining a dataset of detected biomarkers; selecting a plurality of routine biomarkers from the dataset; and training an ML model to identify AKI based on the plurality of routine biomarkers by employing the plurality of routine biomarkers as inputs and presence of AKI and absence of AKI as target values.

2. The method of claim 1 wherein the dataset contains biomarkers determined for at least one of a plurality of hospitals and a plurality of laboratories.

3. The method of claim 1 wherein the dataset contains biomarkers determined for a specific hospital.

4. The method of claim 3 wherein the biomarkers for the specific hospital include biomarkers determined for at least one of an emergency room, an in-patient clinic, an outpatient clinic, a doctor’s office and a laboratory affiliated with the specific hospital.

5. The method of claim 1 wherein selecting the plurality of routine biomarkers includes at least one of identifying relevant biomarkers and removing redundant biomarkers.

6. The method of claim 1 further comprising: determining a subset of the plurality of routine biomarkers that are altered in the presence of AKI; and retraining the ML model based on the subset of the plurality of routine biomarkers.

7. The method of claim 1 wherein training the ML model further comprises training the ML model with demographic information.

8. The method of claim 1 further comprising deploying the ML model for use in a hospital.

9. The method of claim 8 further comprising: employing the deployed ML model on a patient sample from a patient within the hospital.

10. The method of claim 9 further comprising determining an AKI score indicative of the presence or absence of AKI using the ML model.11 . The method of claim 10 wherein the AKI score is determined without baseline biomarkers for the patient.

12. The method of claim 10 wherein the AKI score is determined with baseline biomarkers for the patient.

13. The method of claim 10 wherein the AKI score indicates at least one of absence of AKI, presence of AKI, and probability of having AKI.

14. The method of claim 10 further comprising determining which biomarkers of the patient contributed to the AKI score.

15. The method of claim 10 further comprising recommending at least one of blood pressure reduction and electrolytes based on the AKI score.

16. The method of claim 10 further comprising recommending additional testing based on the AKI score.

17. The method of claim 16 wherein recommending additional testing based on the AKI score includes recommending additional testing based on the AKI score in combination with one or more risk factors of the patient.

18. The method of claim 17 wherein the one or more risk factors include at least one of age, blood pressure, kidney disease and diabetes.

19. The method of claim 16 further comprising employing a biomarker from the additional testing as an input to a second ML model, the second ML model trained using: routine biomarkers and one or more additional biomarkers determined by the hospital for patients suspected of having AKI as inputs; and presence of AKI and absence of AKI as target values.

20. The method of claim 19 wherein the one or more additional biomarkers determined by the hospital for patients suspected of having AKI are from non-routine tests.21 . The method of claim 19 further comprising determining a second AKI score indicative of the presence or absence of AKI using the second ML model.

22. A method of training a machine-learning (ML) model for use in detecting acute kidney injury (AKI), comprising: obtaining a dataset of detected biomarkers; selecting a plurality of routine biomarkers from the dataset; training an ML model to identify AKI based on the plurality of routine biomarkers by employing the plurality of routine biomarkers as inputs and presence of AKI and absence of AKI as target values; determining a subset of the plurality of routine biomarkers that are altered in the presence of AKI; and retraining the ML model based on the subset of the plurality of routine biomarkers.

23. The method of claim 22 wherein the dataset contains biomarkers determined for a specific hospital.

24. A method of training a machine-learning (ML) model for use in detecting acute kidney injury (AKI), comprising: obtaining a dataset of detected biomarkers; selecting a plurality of routine biomarkers from the dataset;obtaining one or more additional biomarkers determined for patients suspected of having AKI; training an ML model to identify AKI based on the plurality of routine biomarkers and the one or more additional biomarkers, the ML model employing: the plurality of routine biomarkers and the one or more additional biomarkers as inputs; and presence of AKI and absence of AKI as target values.

25. The method of claim 24 further comprising: determining a subset of the plurality of routine biomarkers that are altered in the presence of AKI; and retraining the ML model based on the subset of the plurality of routine biomarkers.

26. The method of claim 25 wherein retraining the ML model further comprises retraining the ML model based on the one or more additional biomarkers.

27. The method of claim 24 wherein the dataset contains biomarkers determined for a specific hospital.

28. The method of claim 24 further comprising deploying the ML model for use in a hospital.

29. The method of claim 28 further comprising: employing the deployed, ML model on a patient sample from a patient within the hospital.

30. The method of claim 29 further comprising: determining an AKI score indicative of the presence or absence of AKI from the ML model.31 . A method of diagnosing acute kidney injury (AKI) comprising: obtaining routine biomarkers for a patient at a hospital;inputting the routine biomarkers for the patient through a first machinelearning (ML) model to determine an initial AKI score, the first ML model trained using routine biomarkers as inputs and presence of AKI and absence of AKI as target values; based on the initial AKI score, determining if the patient is at an elevated risk of AKI; and if the patient is at an elevated risk of AKI, recommending at least one additional test for the patient.

32. The method of claim 31 further comprising: obtaining one or more biomarkers based on the at least one additional test for the patient; and inputting the one or more biomarkers based on the at least one additional test for the patient to a second ML model to determine an updated AKI score, the second ML model trained using: routine biomarkers and one or more additional biomarkers for patients suspected of having AKI as inputs; and presence of AKI and absence of AKI as target values.

33. The method of claim 32 wherein the routine biomarkers used to train the first and second ML models are from the hospital.

34. The method of claim 32 wherein the routine biomarkers used to train the first and second ML models are from a different hospital.

35. The method of claim 33 wherein the one or more additional biomarkers for patients suspected of having AKI used to train the second ML model are from the hospital.

36. The method of claim 33 wherein the one or more additional biomarkers for patients suspected of having AKI used to train the second ML model are from a different hospital.

37. The method of claim 31 further comprising recommending at least one of blood pressure reduction and electrolytes based on the AKI score.

Images