Chronic kidney disease (CKD) machine learning prediction system, method, and apparatus

Abstract

To provide an appropriate chronic kidney disease (CKD) machine learning prediction system, method, and apparatus.SOLUTION: A chronic kidney disease ("CKD") machine learning prediction system is disclosed. The example system is configured to provide a prediction of whether a patient may progress to a next stage of CKD and / or whether the patient may need to start dialysis urgently. The machine learning algorithms include dynamic, multifactorial predictive algorithms that are programmed to consider clinical, pharmacological, and extra-clinical factors that adversely impact kidney function. The predictions provided by the machine learning system convey information to clinicians for improving CKD treatment before the disease worsens. In some instances, the predictions may be used for a treatment plan, a dialysis treatment, and / or a renal replacement therapy ("RRT").SELECTED DRAWING: Figure 1

Description

[Background technology]

[0001] Chronic kidney disease ("CKD") is a serious and often debilitating medical condition suffered by millions of individuals worldwide each year. Individuals with kidney disease have damaged kidneys that are unable to filter blood at all, or at least at sufficient levels, to remove toxins from the individual's blood. Individuals suffering from kidney disease or kidney failure can no longer maintain fluid and mineral balance or excrete their daily metabolic load. Toxic end products of nitrogen metabolism (urea, creatinine, uric acid, calcium, phosphorus, sodium, potassium, and others) can accumulate in the individual's blood and tissues. Some patients with kidney disease or kidney failure may also suffer from high / low blood pressure and low red blood cell counts. Often, kidney disease is a chronic condition that worsens over time, leading to complete kidney failure (i.e., end-stage renal disease ("ESRD")) or death.

[0002] As the world's population improves its overall standard of living, more individuals are able to consume foods and beverages and live lifestyles that lead to CKD. Some studies estimate that as many as 10% of the world's population has some form of CKD. Overall, the global burden of CKD is caused not only by the increasing number of individuals with ESRD requiring renal replacement therapy ("RRT"), but also by the increasing prevalence of conditions associated with the development of CKD. Currently, individuals receiving RRT consume the majority of healthcare resources to treat CKD. Therefore, individuals with less severe CKD are often not treated or only minimally treated, which ultimately leads to the worsening of CKD to the point where they eventually require RRT. Efforts are being made by healthcare providers to control predisposing conditions in individuals susceptible to CKD or by individuals suffering from early-onset CKD to delay and / or avoid the progression to ESRD.

[0003] Currently, individuals are assessed for CKD by monitoring their estimated glomerular filtration rate ("GFR"), which indicates the amount of blood passing through an individual's glomeruli (the microfiltration functions in the kidneys) per minute. GFR is typically calculated by a blood creatinine test, taking into account the individual's age, body size, and sex. Generally, patients with a GFR of less than 90 mL / min are considered to be suffering from CKD. Proteinuria or albuminuria, a condition characterized by the presence of greater than normal amounts of protein (e.g., albumin) in the urine, can also indicate the onset of CKD if the condition persists for three months.

[0004] After a patient is assessed as having CKD, a healthcare provider estimates the patient's potential CKD progression timeline and determines potential treatment. Early detection of CKD is crucial because it allows appropriate preventative treatment to be prescribed before any CKD exacerbation occurs through worsening complications. For example, a patient with a predicted slow progression may be treated with lifestyle and dietary changes in addition to medication. However, a patient with a predicted rapid progression may need to undergo more intensive clinical treatment, such as initiating RRT.

[0005] Currently, healthcare providers assess an individual's rate of progression through periodic blood creatinine testing and urine analysis. This involves administering blood tests to the individual every few weeks or months, which is a burden to the healthcare provider and the individual. In some cases, healthcare providers or individuals do not have the capacity to perform periodic blood tests to assess CKD progression. As a result of these known problems, some individuals may progress more rapidly than initially estimated, and any preventative treatment may be too late or ineffective by the time the individual is reassessed.

[0006] Therefore, there is a need for a CKD clinician diagnostic tool that provides an accurate prediction of an individual's CKD progression and / or the likelihood that an individual will need to begin dialysis urgently. Summary of the Invention [Means for solving the problem]

[0007] Chronic kidney disease ("CKD") machine learning prediction systems, methods, and devices are disclosed. Exemplary machine learning prediction systems, methods, and devices are configured to predict a patient's CKD progression and / or the urgency with which the patient will need to start dialysis or RRT in the future. In some embodiments, separate machine learning models are used to predict CKD progression and estimate the patient's need for urgent dialysis initiation.

[0008] The disclosed machine learning prediction systems, methods, and devices provide more information, allowing clinicians to make more informed patient care decisions. While knowing a patient's GFR and / or urinary albumin / creatinine rate / level is useful in determining a patient's current CKD stage, the data often does not indicate the rate of progression through CKD stages or the urgency with which the patient will need to begin dialysis. Instead, other factors or characteristics may be more indicative of the rate of CKD progression and / or the urgency with which dialysis will be initiated. The algorithms disclosed herein use machine learning to model and use classified patient factors / characteristics to determine a patient's CKD progression prediction and the likelihood of urgency in needing dialysis. The classified factors / characteristics are readily available from the patient's medical records. The factors / characteristics may include gender, race, age, body mass index ("BMI"), blood pressure, creatinine level, GFR, hemoglobin level, and / or albumin level. The factors / traits may also include diagnosed causes of CKD, including hypertension, diabetes, obstructive uropathy, glomerulonephritis / autoimmune disease, polycystic kidney disease, chronic tubulointerstitial nephritis, or chronic pyelonephritis. The factors / traits may further include a medical history of hypertension, diabetes, cardiac ischemia, congestive heart failure, or cerebrovascular disease, etc.

[0009] In some cases, the disclosed machine learning prediction systems, methods, and devices are configured to calculate derived factors / signatures from available patient factors / signatures. The derived factors / signatures may include ratios of factors such as albumin / creatinine ratio. The derived factors / signatures may also include determining a patient's current or past CKD stage based on their GFR and / or albumin level.

[0010] Together, the factors / traits and derived factors / traits are associated with positive / negative outcomes related to CKD stage progression, rapidity of CKD stage progression, and the urgency of need to initiate dialysis for a population of patients with known CKD outcomes. The associations are used to determine the probability or likelihood that patients with similar factors / traits will have similar outcomes.

[0011] As disclosed herein, the machine learning prediction system, method, and device compares the characteristics of the patient under analysis with the classified factors / characteristics of known patients represented in the prediction algorithm / model. The probability of the classified factors / characteristics being approximately comparable to the characteristics of the patient under analysis is reported as the predicted CKD outcome. Clinicians may use the reported CKD outcome for treatment planning purposes to slow CKD progression and / or determine the need for urgent dialysis.

[0012] In some embodiments, the disclosed machine learning prediction systems, methods, and devices include a CKD progression prediction algorithm or model. As disclosed herein, the CKD progression prediction algorithm or model is configured to provide a likelihood or probability that a patient will progress to the next CKD stage within a specified time frame. In some embodiments, the CKD progression algorithm or model includes an ensemble machine learning algorithm configured to determine the likelihood that a patient will transition to a new CKD stage and the length of time that the patient may transition to the new CKD stage. The model or algorithm is configured to compare the patient's physiological data, demographic data, medical history, and other identified characteristics / factors with a modeled classifier trained using known patient CKD progression data. Based on the comparison, the model determines the closest matching predictive decile and outputs a percentage and time frame for that decile. In some alternative embodiments, the CKD progression model may average or weighted average the patient's comparison to one or more deciles to estimate the CKD stage progression likelihood and time frame.

[0013] Additionally or alternatively, the disclosed machine learning prediction systems, methods, and devices include a CKD urgent dialysis initiation prediction algorithm or model. As disclosed herein, the CKD progression urgent dialysis initiation prediction algorithm or model is configured to provide a likelihood or probability that a patient will require dialysis within a specified time frame. The model or algorithm is configured to compare the patient's physiological data, demographic data, medical history, and other identified characteristics / factors with a modeled classifier trained using known patient CKD urgent dialysis initiation data. Based on the comparison, the model or algorithm determines the closest matching predictive decile and outputs a percentage and time frame for that decile. In some alternative embodiments, the CKD urgent dialysis initiation model may average or weight the patient's comparison to one or more deciles to estimate the likelihood that the patient will need to start dialysis within a discrete time frame.

[0014] The disclosed machine learning prediction systems, methods, and devices of the present disclosure are applicable to fluid delivery for, for example, plasma exchange, hemodialysis ("HD"), hemofiltration ("HF"), hemodiafiltration ("HDF"), and continuous renal replacement therapy ("CRRT") therapies. The disclosed machine learning prediction systems, methods, and devices described herein are also applicable to peritoneal dialysis ("PD"), intravenous drug delivery, and nutritional fluid delivery. These modalities may be referred to herein collectively, or generally and individually, as medical fluid delivery or therapy.

[0015] As described in detail below, the disclosed CKD machine learning prediction system, method, and apparatus may operate within a comprehensive healthcare platform that may include many machines, patients, clinicians, physicians, maintenance personnel, electronic medical record ("EMR") databases, websites, resource planning systems, and business intelligence, with many different types of devices. The disclosed CKD machine learning prediction system, method, and apparatus are configured to operate seamlessly within the entire system without violating its rules and protocols.

[0016] In a first aspect of the present disclosure, which may be combined with any other aspects enumerated herein unless otherwise specified in light of the disclosure herein and without limiting the disclosure in any way, a system for estimating a patient's chronic kidney disease ("CKD") progression includes a memory device that stores patient characteristic data about a patient undergoing analysis, the patient characteristic data including demographic / physiological data, an early CKD stage, a diagnosed cause of CKD, and medical history. The system also includes an ensemble machine learning algorithm configured to predict progression to a next stage of CKD and a time frame for progression of the next stage of CKD, the ensemble machine learning algorithm containing predictive decile classifiers each including a percentage of known patients who progressed from a moderate CKD stage to a next moderate or severe CKD stage over a discrete time frame. The system further includes an analysis processor communicatively coupled to the memory device. The analysis processor, in conjunction with the ensemble machine learning algorithm, is configured to classify the patient undergoing analysis into a matched predictive decile that is closest to the patient's early CKD stage by comparing the classification of the patient characteristic data provided in the ensemble machine learning algorithm with the patient characteristic data of the patient under analysis, determine the probability that the patient undergoing analysis will progress to the next moderate or severe CKD stage for each discrete time frame based on the closest matched predictive decile, and display, via a user interface, the percentage likelihood that the patient undergoing analysis will progress to the next moderate or severe CKD stage over the discrete time frame.

[0017] According to a second aspect of the present disclosure, which may be used in combination with any other aspect enumerated herein unless otherwise stated, the demographic / physiological data includes at least one of gender, race, age, body mass index, blood pressure, creatinine level, glomerular filtration rate ("GFR"), hemoglobin level, or albumin level.

[0018] According to a third aspect of the present disclosure, which may be used in combination with any other aspect recited herein unless otherwise stated, the diagnosed causes of CKD include at least one of hypertension, diabetes, obstructive uropathy, glomerulonephritis / autoimmune disease, polycystic kidney disease, chronic tubulointerstitial nephritis, or chronic pyelonephritis.

[0019] According to a fourth aspect of the present disclosure, which may be used in combination with any other aspect recited herein unless otherwise stated, the medical history includes at least one of hypertension, diabetes, cardiac ischemia, congestive heart failure, or cerebrovascular disease.

[0020] According to a fifth aspect of the present disclosure, which may be used in combination with any other aspect recited herein unless otherwise stated, the percentage of known patients who have progressed from one moderate CKD stage to a next moderate or severe CKD stage is determined using patient population data, including patient characteristic data, known CKD progression data, and clinical trial discontinuation results.

[0021] According to a sixth aspect of the present disclosure, which may be used in combination with any other aspect recited herein unless otherwise stated, the trial discontinuation outcome comprises at least one of dialysis therapy, renal replacement therapy ("RRT"), death, kidney transplant, or palliative care.

[0022] According to a seventh aspect of the present disclosure, which may be used in combination with any other aspect recited herein unless otherwise stated, the known CKD progression data identifies stage progression based on changes in estimated glomerular filtration rate ("GFR") associated with different moderate or severe CKD stages, or a change of at least 25% in estimated GFR from previously known GFR.

[0023] According to an eighth aspect of the present disclosure, which may be used in combination with any other aspect recited herein unless otherwise stated, the early stage of CKD in a patient is based on at least one of the patient's estimated GFR or the length of time the patient has experienced proteinuria.

[0024] According to a ninth aspect of the present disclosure, which may be used in combination with any other aspect recited herein unless otherwise stated, the discrete time frame comprises at least one of 30 days, 60 days, 90 days, 120 days, 180 days, and 360 days.

[0025] According to a tenth aspect of the present disclosure, which may be used in combination with any other aspect recited herein unless otherwise stated, moderate or severe CKD stages include stage 3A with a GFR of 45-59 mL / min, stage 3B with a GFR of 30-44 mL / min, stage 4 with a GFR of 15-29 mL / min, and stage 5 with a GFR of less than 15 mL / min.

[0026] According to an eleventh aspect of the present disclosure, which may be used in combination with any other aspects recited herein unless otherwise stated, an ensemble machine learning algorithm includes predictive decile classifiers each including a percentage of known patients who progressed from one mild CKD stage to the next moderate or severe CKD stage over a discrete time frame, wherein the early CKD stages include at least one of stage 1 with a GFR greater than 90 mL / min, stage 2 with a GFR of 60-89 mL / min, stage 3A with a GFR of 45-59 mL / min, stage 3B with a GFR of 30-44 mL / min, or stage 4 with a GFR of 15-29 mL / min.

[0027] According to a twelfth aspect of the present disclosure, which may be used in combination with any other aspect recited herein unless otherwise stated, the user interface is displayed on a clinician computer.

[0028] According to a thirteenth aspect of the present disclosure, which may be used in combination with any other aspects enumerated herein unless otherwise stated, a system for estimating the likelihood that a patient suffering from chronic kidney disease ("CKD") will need to urgently start dialysis includes a memory device that stores patient characteristic data regarding the patient undergoing analysis, the patient characteristic data including demographic / physiological data, early CKD stage, diagnosed cause of CKD, and medical history. The system also includes a machine learning algorithm configured to predict the likelihood that the patient undergoing analysis will need urgent initiation of dialysis, the machine learning algorithm containing predictive decile classifiers each including a percentage of known patients requiring urgent initiation of dialysis over discrete time frames. The system further includes an analysis processor communicatively coupled to the memory device. The analysis processor, in conjunction with the ensemble machine learning algorithm, is configured to classify the patient undergoing analysis into a matched prediction group that most closely matches the patient's early CKD stage by comparing the classification of patient characteristic data provided in the machine learning algorithm with the patient characteristic data of the patient under analysis, determine a probability that the patient undergoing analysis will be in need of urgent initiation of dialysis over a discrete time frame based on the most closely matched prediction decile, and display via a user interface the percentage likelihood that the patient undergoing analysis will be in need of urgent initiation of dialysis over a discrete time frame.

[0029] According to a fourteenth aspect of the present disclosure, which may be used in combination with any other aspect enumerated herein unless otherwise stated, the demographic / physiological data includes at least one of gender, race, age, body mass index, blood pressure, creatinine level, glomerular filtration rate ("GFR"), hemoglobin level, or albumin level.

[0030] According to a fifteenth aspect of the present disclosure, which may be used in combination with any other aspects recited herein unless otherwise stated, the diagnosed causes of CKD include at least one of hypertension, diabetes, obstructive uropathy, glomerulonephritis / autoimmune disease, polycystic kidney disease, chronic tubulointerstitial nephritis, or chronic pyelonephritis.

[0031] According to a sixteenth aspect of the present disclosure, which may be used in combination with any other aspect recited herein unless otherwise stated, the medical history includes at least one of hypertension, diabetes, cardiac ischemia, congestive heart failure, or cerebrovascular disease.

[0032] According to a seventeenth aspect of the present disclosure, which may be used in combination with any other aspect recited herein unless otherwise stated, the percentage of known patients who progressed from one CKD stage to the next CKD stage was determined using patient population data, including patient characteristic data, known CKD progression data, and clinical trial discontinuation results.

[0033] According to an eighteenth aspect of the present disclosure, which may be used in combination with any other aspect recited herein unless otherwise stated, CKD stages include stage 1 with a GFR of greater than 90 mL / min, stage 2 with a GFR of 60-89 mL / min, stage 3A with a GFR of 45-59 mL / min, stage 3B with a GFR of 30-44 mL / min, stage 4 with a GFR of 15-29 mL / min, and stage 5 with a GFR of less than 15 mL / min.

[0034] According to a nineteenth aspect of the present disclosure, which may be used in combination with any other aspect enumerated herein unless otherwise stated, the analysis processor is configured to receive an indication to start dialysis treatment and prepare dialysis treatment for the patient.

[0035] According to a twentieth aspect of the present disclosure, which may be used in combination with any other aspect recited herein unless otherwise stated, the system further includes a dialysis machine configured to administer dialysis therapy to the patient.

[0036] In the twenty-first aspect of the present disclosure, any of the structures and functionality disclosed in connection with FIGS. 1-8 may be combined with any other structure and functionality disclosed in connection with FIGS. 1-8.

[0037] In light of the present disclosure and the above aspects, it is therefore an advantage of the present disclosure to provide a CKD machine learning algorithm configured to provide a prediction regarding a patient's CKD progression over time.

[0038] It is another benefit of the present disclosure to provide a CKD machine learning algorithm configured to provide a prediction regarding a patient's need for urgent initiation of dialysis or other RRT.

[0039] It is a further benefit of the present disclosure to provide a clinician or other healthcare provider with information for clinician diagnosis and treatment that indicates a prediction of a patient's CKD progression over time and / or the patient's need for urgent initiation of dialysis.

[0040] It is yet a further benefit of the present disclosure to provide improved patient outcomes from the onset of CKD detection to slow disease progression.

[0041] Additional features and benefits will be described in and apparent from the following detailed description and figures. The features and benefits described herein are not all-inclusive; in particular, many additional features and benefits will become apparent to those skilled in the art in view of the figures and description. Moreover, it is not necessary for any particular embodiment to have all of the benefits enumerated herein; it is expressly contemplated that each advantageous embodiment may be separately claimed. It should also be noted that the language used herein has been chosen primarily for readability and instructional purposes, rather than to limit the scope of the inventive subject matter. The present invention provides, for example, the following. (Item 1) 1. A system for estimating chronic kidney disease ("CKD") progression in a patient, the system comprising: a memory device that stores patient characteristic data regarding a patient undergoing analysis, the patient characteristic data including demographic / physiological data, early stage of CKD, diagnosed cause of CKD, and medical history; an ensemble machine learning algorithm configured to predict progression to a next stage of CKD and a time frame for said progression to said next stage of CKD, the ensemble machine learning algorithm containing predictive decile classifiers each comprising a percentage of known patients who progressed from a moderate CKD stage to a next moderate or severe CKD stage over discrete time frames; an analytics processor communicatively coupled to the memory device, the analytics processor in conjunction with the ensemble machine learning algorithm; Classifying the patient undergoing the analysis into a matched predictive decile that is closest to the patient's early CKD stage by comparing the classification of patient characteristic data provided in the ensemble machine learning algorithm with the patient characteristic data of the patient under analysis; determining the probability that the patient undergoing said analysis will progress to the next moderate or severe CKD stage for each of said discrete time frames based on said closest matched predictive decile; displaying, via a user interface, a percentage likelihood that the patient undergoing said analysis will progress to said next moderate or severe CKD stage over said discrete time frame; an analysis processor configured to: A system comprising: (Item 2) Item 10. The system of item 1, wherein the demographic / physiological data includes at least one of gender, race, age, body mass index, blood pressure, creatinine level, glomerular filtration rate ("GFR"), hemoglobin level, or albumin level. (Item 3) 3. The system of item 1 or 2, wherein the diagnosed cause of CKD includes at least one of hypertension, diabetes, obstructive uropathy, glomerulonephritis / autoimmune disease, polycystic kidney disease, chronic tubulointerstitial nephritis, or chronic pyelonephritis. (Item 4) 3. The system of item 1 or 2, wherein the medical history includes at least one of hypertension, diabetes, cardiac ischemia, congestive heart failure, or cerebrovascular disease. (Item 5) 5. The system of item 1 or 4, wherein the percentage of known patients who progressed from one moderate CKD stage to the next moderate or severe CKD stage is determined using patient characteristic data, known CKD progression data, and patient population data including trial discontinuation results. (Item 6) 6. The system of item 5, wherein the trial discontinuation outcome includes at least one of dialysis therapy, renal replacement therapy ("RRT"), death, kidney transplant, or palliative care. (Item 7) Item 6. The system of item 5, wherein the known CKD progression data identifies stage progression based on changes in estimated glomerular filtration rate ("GFR") associated with different moderate or severe CKD stages, or a change in the estimated GFR of at least 25% from a previously known GFR. (Item 8) 8. The system of claim 1 or 7, wherein the early CKD stage of the patient is based on at least one of the patient's estimated GFR or the length of time the patient has experienced proteinuria. (Item 9) 8. The system of claim 1 or 7, wherein the discrete time frame comprises at least one of 30 days, 60 days, 90 days, 120 days, 180 days, and 360 days. (Item 10) 8. The system of item 1 or 7, wherein the moderate or severe CKD stages include stage 3A with a GFR of 45 to 59 mL / min, stage 3B with a GFR of 30 to 44 mL / min, stage 4 with a GFR of 15 to 29 mL / min, and stage 5 with a GFR of less than 15 mL / min. (Item 11) the ensemble machine learning algorithm includes predictive decile classifiers, each containing the percentage of known patients who progressed from one mild CKD stage to the next moderate or severe CKD stage over a discrete time frame; The early CKD stage includes at least one of stage 1 with a GFR of more than 90 mL / min, stage 2 with a GFR of 60-89 mL / min, stage 3A with a GFR of 45-59 mL / min, stage 3B with a GFR of 30-44 mL / min, or stage 4 with a GFR of 15-29 mL / min. Item 1. The system of item 1. (Item 12) 2. The system of claim 1, wherein the user interface is displayed on a clinician computer. (Item 13) 1. A system for estimating the likelihood that a patient with chronic kidney disease ("CKD") will need to begin dialysis urgently, the system comprising: a memory device that stores patient characteristic data regarding a patient undergoing analysis, the patient characteristic data including demographic / physiological data, early stage of CKD, diagnosed cause of CKD, and medical history; a machine learning algorithm configured to predict the likelihood that a patient undergoing the analysis will be in need of urgent initiation of dialysis, the machine learning algorithm containing predictive decile classifiers each comprising a percentage of known patients requiring urgent initiation of dialysis over discrete time frames; an analytics processor communicatively coupled to the memory device, the analytics processor in conjunction with the ensemble machine learning algorithm; Classifying the patient undergoing the analysis into a matched prediction group that most closely matches the early stage of CKD for the patient by comparing the classification of patient characteristic data provided in the machine learning algorithm with the patient characteristic data for the patient under analysis; determining the probability that the patient undergoing said analysis will be in need of urgent initiation of dialysis over said discrete time frame based on the closest matched prediction decile; displaying, via a user interface, a percentage likelihood that the patient undergoing said analysis will be in need of urgent initiation of said dialysis over said discrete time frame; an analysis processor configured to: A system comprising: (Item 14) Item 14. The system of item 13, wherein the demographic / physiological data includes at least one of gender, race, age, body mass index, blood pressure, creatinine level, glomerular filtration rate ("GFR"), hemoglobin level, or albumin level. (Item 15) Item 15. The system of item 14, wherein the diagnosed cause of CKD includes at least one of hypertension, diabetes, obstructive uropathy, glomerulonephritis / autoimmune disease, polycystic kidney disease, chronic tubulointerstitial nephritis, or chronic pyelonephritis. (Item 16) 16. The system of item 14 or 15, wherein the medical history includes at least one of hypertension, diabetes, cardiac ischemia, congestive heart failure, or cerebrovascular disease. (Item 17) 16. The system of item 14 or 15, wherein the percentage of known patients who progressed from one CKD stage to the next CKD stage is determined using patient characteristic data, known CKD progression data, and patient population data including clinical trial discontinuation results. (Item 18) 16. The system of item 14 or 15, wherein the CKD stages include stage 1 with a GFR of more than 90 mL / min, stage 2 with a GFR of 60 to 89 mL / min, stage 3A with a GFR of 45 to 59 mL / min, stage 3B with a GFR of 30 to 44 mL / min, stage 4 with a GFR of 15 to 29 mL / min, and stage 5 with a GFR of less than 15 mL / min. (Item 19) The analysis processor receiving an indication to start dialysis treatment; To prepare dialysis treatment for said patient; Item 15. The system of item 14, configured to perform the following: (Item 20) 20. The system of claim 19, further comprising a dialysis machine configured to administer the dialysis treatment to the patient. [Brief explanation of the drawings]

[0042] [Figure 1] FIG. 1 is a schematic diagram of a CKD machine learning prediction system including a model generator and an analysis processor, according to an example embodiment of the present disclosure.

[0043] [Figure 2] FIG. 2 is a flow diagram of an exemplary procedure for creating the CKD predictive machine learning algorithm disclosed herein, according to an exemplary embodiment of the present disclosure.

[0044] [Figure 3] FIG. 3 is a diagram of exemplary patient characteristic data received by the model generator of FIG. 1, according to certain exemplary embodiments of the present disclosure.

[0045] [Figure 4] FIG. 4 is a graph of probability data associated with a positive outcome of a CKD staging prediction machine learning algorithm, according to an exemplary embodiment of the present disclosure.

[0046] [Figure 5] FIG. 5 is a diagram of exemplary patient characteristic data received by the analysis processor of FIG. 1 according to an exemplary embodiment of the present disclosure.

[0047] [Figure 6] FIG. 6 is a diagram of a user interface displayed via an application on a clinician device showing machine learning model output from the analysis processor of FIG. 1 according to an exemplary embodiment of the present disclosure.

[0048] [Figure 7] FIG. 7 is a diagram illustrating a process flow for a clinician using an application to input treatment parameters for programming a medical device based on the machine learning model output of FIG. 6, according to an exemplary embodiment of the present disclosure.

[0049] [Figure 8] FIG. 8 is a flow diagram of an exemplary procedure for analyzing patient characteristic data via the CKD predictive machine learning models disclosed herein, according to certain exemplary embodiments of the present disclosure. DETAILED DESCRIPTION OF THE INVENTION

[0050] Detailed Description Disclosed herein are CKD machine learning prediction systems, methods, and devices. Exemplary CKD machine learning prediction systems, methods, and devices are configured to provide predictions about whether a patient will progress to the next stage of CKD and / or whether the patient will need to begin dialysis urgently. The machine learning algorithms disclosed herein include dynamic multifactorial prediction algorithms that are programmed to consider clinical, pharmacological, and additional clinical factors that adversely affect kidney function. The predictions provided by the machine learning systems, methods, and devices inform clinicians to improve CKD treatment before the disease worsens. In some cases, the predictions may be used to select treatment plans, dialysis treatments, and / or RRT.

[0051] In this specification, references are made to machine learning algorithms and models, and the terms are used interchangeably. As disclosed, the machine learning algorithms and models are configured to receive certain patient factors / characteristics, which are processed and compared with the classified factors / characteristics to determine the probability or likelihood of a positive outcome. The algorithms or models are defined by one or more machine-readable instructions stored in a memory device. The algorithms and models are also defined by factor / characteristic evaluation adjustment parameters / weights / correlation indices created during the creation of the algorithms or models. The adjustment parameters / weights / correlation indices are also stored in the memory device. Execution of the one or more machine-readable instructions by a processor causes operations to be performed using the stored adjustment parameters / weights / correlation indices. These operations allow an analysis of the patient characteristics of a specified patient to be processed through the exemplary machine learning algorithms and models to provide a predicted outcome.

[0052] Also referred to as machine learning model deciles of positive outcomes. As disclosed herein, a machine learning model / algorithm classifies / orders known patients into 10 groups by CKD stage. For each decile of CKD stage, the model / algorithm determines the probability of a positive outcome for CKD progression and / or CKD urgent dialysis initiation. The probabilities are determined over a range of discrete time frames, such as those with a positive outcome within 30 days, 60 days, 90 days, 120 days, 180 days, 360 days, etc., for that decile of CKD stage. In other examples, different ranges / classifications may be used. For example, classification may be performed in a non-uniform manner based on natural boundaries between known patient characteristics / factors. For example, deciles 8-10 disclosed herein may be partitioned into additional groups for a larger resolution, where there is more outcome variation compared to deciles 1-5, which may be combined into a single group assuming general outcome uniformity for known patient outcomes.

[0053] As provided herein, exemplary systems, methods, and devices provide more accurate predictions compared to known clinical methods for treating CKD. For example, the Kidney Disease International Guidelines ("KDIGO") recommend classifying CKD by the patient's level of albuminuria according to the underlying etiology. This definition and classification has generally been accepted and implemented worldwide, despite known limitations in the current equation used to calculate a patient's glomerular filtration rate ("GFR") from serum creatinine, which can result in overestimation, particularly among patients with a GFR above 60 mL / min ("min"). Current clinical practice involves assessing a patient's progression of CKD through periodic estimation of their GFR, which is based on the assumption of a predictable long-term decline. However, recent clinical trials have shown that certain acute events, medications, and sudden changes in blood pressure can lead to fluctuations in a patient's GFR trajectory, thus gradually undermining the expected rate of renal function decline.

[0054] The exemplary systems, methods, and devices disclosed herein provide a unique assessment of factors contributing to CKD progression and conditions that may affect the trajectory of GFR decline in a patient. References are made herein to CKD stages. Table 1 below provides KDIGO definitions of different stages of CKD, which are based on the patient's estimated GFR and the length of time the patient has experienced proteinuria. Rapid progression of CKD is defined as an absolute annual decline in GFR of ≥ 5 ml / min per year, accompanied by at least a GFR < 90 ml / min. [Table 1]

[0055] Exemplary predictive CKD machine learning algorithms disclosed herein are configured to assess the likelihood of a patient progressing from their current CKD stage to the next CKD stage. Thus, the predictive CKD machine learning algorithm provides an assessment of the degree of progression between each of the stages listed in Table 1. In some embodiments, the predictive CKD machine learning algorithm may only provide an assessment for moderate or severe stages 3A-5 or 5D. In addition to determining whether a patient will progress to the next CKD stage, the predictive CKD machine learning algorithm is configured to determine the rate or time frame of progression. In some instances, rate may be defined as the likelihood of progression within a discrete time frame, such as 30 days, 60 days, 90 days, 120 days, 180 days, and / or 360 days. The predictive CKD machine learning algorithms disclosed herein may also provide an assessment of a patient's risk of emergency dialysis initiation, which refers to the emergency initiation of dialysis for ESRD patients without previously established functional vascular access or a peritoneal dialysis ("PD") catheter. As disclosed herein, the progression likelihood and rate can be combined in an ensemble machine learning algorithm (e.g., a CKD stage progression prediction model), while the risk of urgent dialysis initiation is determined by a second machine learning algorithm (e.g., a CKD urgent dialysis initiation prediction model). I. CKD machine learning prediction system

[0056] FIG. 1 is a schematic diagram of a CKD machine learning prediction system 100 according to an example embodiment of the present disclosure. The example system 100 includes a CKD management server 102 configured to create / update predictive machine learning algorithms as disclosed herein and provide patient predictions using the algorithms. The CKD management server 102 includes a model generator 104 configured to generate the predictive machine learning algorithms as disclosed herein. The CKD management server 102 also includes an analysis processor 106 configured to apply patient characteristic data about the patient under analysis to one or more predictive machine learning algorithms to assess or predict the patient's CKD progression, potentially progression rate, and probability of requiring urgent dialysis initiation. While both are shown as being part of the CKD management server 102, in other embodiments, the model generator 104 may be separate from the analysis processor 106. For example, the model generator 104 may be provided in a back-end server, while the analysis processor 106 is provisioned as a cloud-based service available to clinician devices.

[0057] It should be understood that the operations described in connection with the model generator 104 and the analysis processor 106 may be implemented using one or more computer programs or components. The component programs may be provided as a series of computer instructions on any computer-readable medium, including random access memory (“RAM”), read-only memory (“ROM”), flash memory, magnetic or optical disks, optical memory, or other storage medium. The instructions may be configured to be executed by a processor of the management server 102, which, when executing the series of computer instructions, performs or facilitates the performance of all or part of the disclosed methods and procedures.

[0058] 1 , the model generator 104 is communicatively coupled to a known patient data source 110, which may include a memory device that stores known patient characteristic data 112 for modeling. The model generator 104 partitions the received characteristic data into training data 112a to train and / or create the predictive machine learning algorithms disclosed herein. The model generator 104 also partitions the received characteristic data 112 into test data 112b to test the accuracy of the predictive machine learning algorithms disclosed herein. The received data 112 is further partitioned into validation data 112c to validate the predictive machine learning algorithms disclosed herein.

[0059] The model generator 104 is also communicatively coupled to a clinical objectives source 114, which may include a memory device that stores clinical objectives for the model. In some embodiments, the clinical objectives source 114 may include converting the clinical objectives into machine learning objectives 116. The model generator 104 uses the machine learning objectives 116 and the training data 112a to create one or more predictive machine learning algorithms, shown as a CKD stage progression prediction model 118a and a CKD urgent dialysis initiation prediction model 118b. In the illustrated embodiment, the machine learning objectives 118 include a first objective for providing a CKD stage progression probability or likelihood, a second objective for providing a rate of CKD progression, and a third objective for providing a probability or likelihood that urgent dialysis initiation will be required within a defined time frame. The CKD stage progression prediction model 118a achieves the progression and rate objectives as an ensemble model. The CKD urgent dialysis initiation prediction model 118b achieves an urgent dialysis initiation objective. In some embodiments, the model generator 104 tests different combinations of objectives and models to identify the optimal approach to achieving the stated objectives.

[0060] 2 is a flow diagram of an example procedure 200 for creating the CKD predictive machine learning algorithm disclosed herein, according to an example embodiment of the present disclosure. While procedure 200 is described with reference to the flow diagram illustrated in FIG. 2 , it should be understood that many other ways of implementing the steps associated with procedure 200 may be used. For example, the order of many of the blocks may be changed, certain blocks may be combined with other blocks, and many of the described blocks may be optional. In an embodiment, the number of blocks may be varied based on the type of data preprocessing and filtering and / or the machine learning model developed. The actions described in procedure 200 are defined by one or more instructions stored in a memory device and may be implemented across multiple devices, including, for example, model generator 104.

[0061] The exemplary procedure 200 begins when the model generator 104 receives known patient characteristic data 112, for example, from a known patient data source 110 (block 202). The known patient data source 110 may include one or more electronic medical record ("EMR") databases located at a doctor's office or hospital, storing electronic information about patients. Table 2 below shows an example of known patient characteristic data 112 received by the model generator 104. In the illustrated example, data about 7,131 patients is received and used to create the CKD machine learning models disclosed herein. The known patient data may include, for each patient, GFR, creatinine level, hemoglobin level, and / or albumin level, which may be determined or estimated from patient blood tests. The known patient data may also include blood pressure, body temperature, etc. [Table 2-1] [Table 2-2] [Table 2-3]

[0062] 3 is a schematic diagram of exemplary patient characteristic data 112 received by the model generator 104, according to an exemplary embodiment of the present disclosure. The patient characteristic data 112 may include demographic data such as age, gender, and race. The patient characteristic data 112 may also include physiological data such as blood pressure, BMI, body temperature, weight, GFR, creatinine level, hemoglobin level, and albumin level. In some cases, the patient characteristic data 112 may include an early CKD stage. Otherwise, the model generator 104 may determine the patient's CKD stage from the GFR and / or albumin data. The patient characteristic data 112 may further include a diagnosed cause of CKD, including hypertension, diabetes, obstructive uropathy, glomerulonephritis / autoimmune disease, polycystic kidney disease, chronic tubulointerstitial nephritis, or chronic pyelonephritis. Additionally, the patient characteristic data 112 may include a medical history of hypertension, diabetes, cardiac ischemia, congestive heart failure, or cerebrovascular disease. Figure 3 also shows that the patient characteristic data 112 may include the patient's last known outcome, including dialysis treatment or RRT, completion of treatment, death, kidney transplant, and palliative care. It should be understood that less or additional patient characteristic data 112 may also be used by the model generator 104.

[0063] The known patient characteristic data 112 represents patients at different stages of CKD at which they received medical care and periodic monitoring. The characteristic data 112 includes timestamps provided for clinical activities, including vital sign measurements, laboratory values, pharmacological interventions, hospitalizations for emergency dialysis initiation, appointment dates, and procedures (including hemodialysis and peritoneal dialysis).

[0064] Returning to FIG. 2 , after receiving the data, the model generator 104 is configured to filter the characteristic data 112 according to defined criteria (block 204). For example, the model generator 104 may retain only data related to patients aged 18-80 years at the time of their first treatment for CKD, patients who have reached stage 3 or 4 CKD, and / or patients for whom at least three months, six months, one year, or two years of data are available. In some embodiments, the model generator 104 may filter patient characteristic data 112 related to patients who have reached stage 5 CKD (ESRD) and have received at least three months of follow-up and dialysis treatment. Additionally, the model generator 104 may filter patient characteristic data 112 related to patients with at least three separate GFR measurements.

[0065] After filtering, the model generator 104 is configured to create a data distribution of the filtered data 112 (block 206). Distributions of the characteristic data 112, such as GFR, blood pressure, weight, BMI, creatinine level, hemoglobin level, and / or albumin level, are created, examined, and compared to normal or expected behavior for that type of variation (clinical or administrative). The comparison may reveal data errors, missing data, and data exhibiting other abnormal behavior that should be addressed before modeling. The model generator 104 may also exclude patients with data outside the normal distribution (block 208). Additionally, the model generator 104 may provide missing data using time-stamped medical records from which the characteristic data 112 was received. The model generator 104 may also analyze the structure and aggregation of the characteristic data 112 by identifying variable formats, the nature of variables, and data dependencies between variables. For example, the model generator 104 may determine that the albumin / creatinine ratio is useful for patient classification for CKD progression. Additionally, the model generator 104 may determine a CKD stage (including early CKD stage) for the patient based on the GFR and / or albumin level.

[0066] As shown in FIG. 2, the model generator 104 partitions the processed patient characteristic data 112 into different subsets (block 210). For example, subsets may be included for training data, validation data, and testing data, with patients (and their corresponding data) being assigned to one of the three subsets. The model generator 104 also determines derived data (e.g., engineered variables) from the patient characteristic data 112. The derived data may include calculating ratios between certain data, such as albumin / creatinine ratios. The derived data may also include determining a patient's CKD stage at a certain time based on GFR and / or albumin levels.

[0067] The model generator 104 then correlates the positive and negative outcomes with the distribution of the training data (e.g., data 112a) (block 212). The classification of positive and negative outcomes is based on machine learning objectives 116. With respect to CKD stage progression, a positive outcome comprises characteristic data 112 corresponding to progression from one CKD stage to the next. The model generator 104 creates a classification for each CKD stage. In some instances, the model generator 104 may create a classification from stage 3A or stage 3B to stage 5. The model generator 104 identifies a positive outcome for stage progression based solely on GFR and / or when a known patient's GFR has changed by at least 25% from a previous GFR measurement.

[0068] With respect to CKD stage velocity, the model generator 104 may create and / or use a patient trajectory chart (from the characteristic data 112) that accounts for changes in GFR over time. A positive outcome is determined based on velocity between known CKD stage progressions, which is determined based on GFR measurements as discussed above. With respect to urgent dialysis initiation outcome, a positive result is based on an indication for the patient to begin dialysis treatment.

[0069] For positive outcomes, the model generator 104 also determines a time frame for each positive outcome (block 214). This includes, for each patient, sampled patient data at a point during their medical history. The sampled patient data up to the sampling point is fed into a machine learning algorithm to generate a prediction. If the patient experiences a positive outcome, the model generator 104 calculates a time frame based on the generated prediction and the positive outcome. The model generator 104 creates a classification of time frames for combining patient data to calculate the probability of a positive outcome for each time frame. In some embodiments, the discrete time frames include 30 days, 60 days, 90 days, 120 days, 180 days, and 360 days.

[0070] In one example, a known patient A is sampled at a certain date corresponding to a midpoint in their treatment. Patient A's patient data up to that date is analyzed through a machine learning algorithm to determine, for example, a predicted probability of progression from Stage 3B to Stage 4 CKD. The algorithm may provide a 45-day estimate. The model generator 104 compares the prediction to the actual known outcome for patient A; in this example, progression from Stage 3B to Stage 4 CKD occurred at 60 days. In this example, the model generator 104 refines the machine learning algorithm based on the difference between the predicted 45 days and the actual 60 days. Thus, over at least the 60-day time frame, patient A had a positive progression rate from Stage 3B to Stage 4 CKD of 100% to 0% prior to the 60-day time frame. Patient A's probabilities are combined with other patients to provide estimates for the entire training dataset across different time frames.

[0071] In some cases, the model generator 104 resamples the training patient data 112a multiple times to refine the machine learning model. For example, for patient A, the patient may be sampled at a first date / time, a second subsequent date / time, and a third date / time to refine the machine learning algorithm. After the model and / or algorithm are created and / or refined, the model generator 104 is configured to perform validation using a subset 112b of patient characteristic data 112 separated from the training data 112a (block 216). The model generator 104 is configured to generate predictions using the validated data and then compare the predictions to actual known outcomes to determine statistical accuracy. Statistics may include positive predictive value, factor / characteristic sensitivity, F1-score, and / or area under the receiver operating characteristic ("ROC") curve.

[0072] The model generator 104 determines whether the machine learning algorithm is accurate by analyzing the statistics (block 218). If the algorithm is not accurate to within a defined accuracy (e.g., 95% accurate), the example procedure 200 returns to block 202 to refine the algorithm or create a new machine learning algorithm using the same and / or different known patient characteristic data 112. However, if the machine learning algorithm is accurate, the model generator 104 deploys the machine learning algorithm 118 (block 220). This may include providing the CKD stage progression prediction model 118a (e.g., the first machine learning algorithm) and / or the CKD urgent dialysis initiation prediction model 118b (e.g., the first machine learning algorithm) to the analysis processor 106. The example procedure 200 then ends. It should be understood that in some instances, the model generator 104 may update the machine learning algorithm as new training data becomes available. II. CKD stage progression prediction model implementation

[0073] This section discusses the properties and accuracy of the CKD staging prediction model 118a. As shown in Tables 3 and 4 below, the exemplary model 118a demonstrates discriminatory performance in identifying risk of progression over different discrete time frames (corresponding to potential clinical follow-up periods), as illustrated by the positive predictive value, sensitivity, F1-score, and area under the ROC curve. [Table 3-1] [Table 3-2]

[0074] As shown in Table 4, the model output is grouped by deciles (as the average of different CKD stages) to illustrate the discrimination of patients with a higher probability of progression from one CKD stage to another and make the model more viable. A closer look at the decile analysis of the staging prediction model shows that the model is capable of segmenting patients across the entire range of risk. For example, as the decile increases, the percentage of patients with advanced stage also increases. Higher deciles tend not only to have a higher rate of staging, but also to have more rapid staging. [Table 4]

[0075] 4 is a graph 400 of the probability data shown in Table 4, according to an exemplary embodiment of the present disclosure. Graph 400 shows that as the decile increases, the percentage of patients with CKD progression increases for each of the 30-, 60-, 90-, 120-, 180-, and 360-day time frames. Furthermore, graph 400 shows that for each decile, the probability of progression increases over time. However, the greatest increase in probability occurs for patients in the highest decile group (deciles 7-10), who are initially more susceptible to progression.

[0076] The exemplary CKD staging prediction model 118a was compared to the known KDIGO two-factor model. The KDIGO model provides guidelines for how often patients should be assessed for CKD. The KDIGO model includes four different recommendations for the number of medical appointments a patient should receive per year based on a combination of GFR and albumin / creatinine ratio ("ACR"). KDIGO provides a risk prediction model that correlates patients with a higher number of recommended medical appointments to a higher risk level prediction.

[0077] In current clinical practice, the KDIGO two-factor model outputs the number of times per year a patient should be assessed to appropriately treat their current level of kidney disease, based on a cross-sectional survey of the patient's GFR level and albumin-to-creatinine ratio (ACR). The two-factor model presents several limitations. Not only do simpler models utilizing only two factors present limitations, but GFR, one of those two factors, presents its own limitations. Creatinine-based GFR estimating equations tend to generate overestimations of true GFR in patients with nephrotic syndrome and uncertainty regarding whether CKD is present when there is a substantial deviation from normal values ​​due to confounding by age, sex, race, and creatinine production.

[0078] A comparative analysis of the CKD staging prediction model 118a and the two-factor KDIGO model demonstrates the strength of model 118a and the inherent operability it offers clinicians. Within the test data, laboratory measurements for determining the recommended number of interventions were available within 14 days of the prediction for many of the known sampled patients. For these samples, each recommended intervention group was divided to show results by decile from the staging prediction model, as shown below in Table 5. Examination of this data reveals that the recommended number of interventions appears correlated with the risk of staging progression. However, when divided by staging prediction model deciles, it is shown that each level of recommended intervention contains patients from different deciles with different propensity for staging progression. For example, for the three recommended intervention categories, it is shown that the categories contain patients from all different deciles with different rates of staging progression according to the deciles. [Table 5]

[0079] In addition to the above, Table 6 below provides a more direct comparison between the KDIGO two-factor model and the CKD staging prediction model 118a by comparing the F1-scores for those samples where the recommended practice is known. Each time frame is considered, the CKD staging prediction model 118a significantly outperforms the two-factor model. [Table 6]

[0080] Tables 5 and 6 above demonstrate that the dynamic multifactor CKD staging prediction model 118a provides more meaningful risk differentiation than the KDIGO two-factor model, especially among patients with values ​​in the mid-range. Focusing on patients recommended three outpatient visits per year, patients from all different deciles with different staging rates were grouped together by the KDIGO two-factor model according to deciles. Following the guidelines of the KDIGO model, all of these patients would have been treated similarly by undergoing three assessments over the course of a year. However, following the guidelines of the CKD staging prediction model 118a, 25% of patients falling into the three visit categories were identified by the exemplary model 118a as being at very low risk (deciles 1-4), while over 40% of patients were identified as being at high risk for staging progression (deciles 8-10).

[0081] Thus, decile analysis reveals that the exemplary CKD staging prediction model 118a more precisely stratifies patients in a manner that will guide physicians toward the best level of care for each patient. Resource utilization will be more efficient in that those patients in deciles 1-4, who are recommended three assessments, will be treated with fewer visits. Clinical care will improve for those patients in higher deciles because they will be treated more frequently. Patients in decile 10, if assessed three times per year as recommended by the KDIGO two-factor model, will already have progressed in disease stage before their next visit (within 120 days). III. CKD emergency dialysis initiation prediction model implementation

[0082] This section discusses the properties and accuracy of the CKD Urgent Dialysis Initiation Prediction Model 118b. The CKD Urgent Dialysis Initiation Prediction Model 118b demonstrated high performance in predicting the risk of urgent dialysis initiation across different time frames of potential clinical follow-up, as shown below in Table 7. The high sensitivity and PPV values ​​indicate that clinicians have a high probability of identifying potential urgent dialysis initiation candidates as short as 30 days and can take appropriate anticipatory steps, such as having a catheter placed for PD or ordering a home HD machine. [Table 7]

[0083] The 90-day prevalence (percent of samples with a positive outcome) for the CKD urgent dialysis initiation prediction model 118 is 4.4%. Decile analysis demonstrates that nearly all of these urgent initiation patients are identified within the top decile of risk, as shown below in Table 8. Machine learning metrics indicate that the positive predictive value and F1-score can be even higher than implied by the decile analysis when clustered on the riskiest portion within the top decile, but a trade-off in sensitivity will be reached. [Table 8] IV. CKD machine learning usage example

[0084] 1 , the analysis processor 106 of the management server 102 receives the CKD stage progression prediction model 118a and / or the CKD urgent dialysis initiation prediction model 118b from the model generator 104. The analysis processor 106 uses the models 118 to provide clinical decision support to clinicians treating patients with CKD. The analysis processor 106 may store the models in a memory device 130.

[0085] In some embodiments, the analysis processor 106 hosts a website or other internet-accessible interface, such as an application programmable interface ("API"), that allows the clinician device 132 to submit patient characteristics and receive predicted outcomes. The clinician device 132 may include an application 134, such as a web browser or "app," for accessing the analysis processor 106.

[0086] In some examples, the clinician device 132 and the analysis processor 106 may be connected to a system hub (not shown). Alternatively, the system hub may be included as part of the analysis processor 106 and include a service portal, an enterprise resource planning system, a web portal, a business intelligence portal, a HIPAA-compliant database, and an electronic medical record database.

[0087] A web page or form provided by the analysis processor 106 may prompt the clinician for patient characteristic data 136. In other examples, the application 134 may allow the clinician to specify a patient identifier, which causes the application 134 to transmit information from the patient's EMR (as patient characteristic data 136) to the analysis processor 106.

[0088] 5 is a schematic diagram of exemplary patient characteristic data 136 received by analysis processor 106, according to an exemplary embodiment of the present disclosure. Patient characteristic data 136 may include demographic data such as age, gender, and race. Patient characteristic data 136 may also include physiological data such as blood pressure, BMI, body temperature, weight, GFR, creatinine level, hemoglobin level, and albumin level. In some cases, patient characteristic data 136 may include an early CKD stage. Otherwise, analysis processor 106 may determine the patient's CKD stage from the GFR and / or albumin data. Patient characteristic data 136 may also include a diagnosed cause of CKD, including hypertension, diabetes, obstructive uropathy, glomerulonephritis / autoimmune disease, polycystic kidney disease, chronic tubulointerstitial nephritis, or chronic pyelonephritis. Additionally, the patient characteristic data 136 may include a medical history of hypertension, diabetes, cardiac ischemia, congestive heart failure, or cerebrovascular disease.

[0089] It should be understood that less or additional patient characteristic data 136 may also be used by the analysis processor 106. For example, the analysis processor 106 may be configured to analyze characteristic data 136 of patients having only a small amount of data to submit to the machine learning model 118. The analysis processor 106 may cause an error message to be transmitted to the clinician device 132 if a sufficient amount of patient characteristic data 136 is not provided (e.g., missing GFR data).

[0090] After receiving the data 136, the analysis processor 106 performs a CKD prediction analysis using the CKD stage progression prediction model 118a and / or the CKD urgent dialysis initiation prediction model 118b. To perform the analysis, the analysis processor 106 may classify the patient undergoing the analysis to the closest matching prediction for the patient's early CKD stage. To perform this operation, the analysis processor 106 compares the classification 112 of the patient characteristic data provided in the individual model 118 with the patient characteristic data 136 of the patient under analysis. This includes identifying a current CKD stage as a starting point for the model 118. This identification may include comparing modeled factors / characteristics (including derived factors / characteristics) for each patient factor / characteristic within the same CKD stage. The model 118 may, for example, assign the patient to one or more deciles based on the comparison. For each model 118, the analysis processor 106 uses the probability of a positive outcome to determine the percentage likelihood (or probability) that the patient undergoing the analysis will progress to the next CKD stage (or need to urgently start dialysis) over a discrete time frame, for example, based on the closest matching prediction decile.

[0091] The analysis processor 106 generates a report 138 that provides a predicted positive outcome for the patient under analysis for the modeled discrete time frame. The analysis processor 106 may display information from the report 138 in a user interface, such as a web page or interface of an application 134 on the clinician device 132. FIG. 6 is a schematic illustration of a user interface 600 displayed via an application 134 on the clinician device 132, showing information from the report 138, according to certain exemplary embodiments of the present disclosure. In some embodiments, a clinician, via the analysis processor 106, may also use the interface 600 to specify a patient identifier or provide patient characteristic data for generating the report 138.

[0092] The exemplary user interface 600 includes a patient identifier and at least some of the patient characteristic data 136, including GFR and albumin level. The user interface 600 also includes at least some information related to the processing of the patient characteristic data in the model 118, including an estimated CKD stage and a predicted decile. The user interface 600 further includes a summary of outputs from the machine learning model 118. A first output 602 provides the rate and probability of progression from CKD Stage 3A to CKD Stage 3B over a discrete time frame. A second output 604 provides the probability that the patient under analysis will need to urgently start dialysis within a defined time frame. A clinician reviews the first output 602 and the second output 604 to determine potential treatments for the patient to slow the progression of the patient's CKD.

[0093] In some embodiments, analysis processor 106 may display options 606 for prescribing treatments on user interface 600. In one example, analysis processor 106 may determine recommended treatments for selection based on the patient's CKD stage, probability of CKD progression, estimated rate of CKD progression, and probability of requiring urgent dialysis initiation. For example, analysis processor 106 may provide options for medication and / or lifestyle changes to a patient in CKD stages 3A or 3B, with a progression probability of less than 25% and a need for urgent dialysis of less than 10%. By comparison, analysis processor 106 may be configured to provide recommendations for PD therapy or critical care ("CC") therapy if the patient is in CKD stage 5, has a greater than 50% probability of progressing to stage 5 within 180 days, and / or has a greater than 35% change in CKD progression probability of requiring urgent dialysis within 90 days.

[0094] For illustrative purposes (not related to the data in outputs 602 and 604), user interface 600 includes option 606 for prescribing PD therapy and / or CC therapy to the patient. Selection of PD therapy causes analysis processor 106 to display a form or web page via application 134 for entering PD prescription parameters, including, for example, glucose levels, treatment duration, treatment frequency, treatment dialysis dose, UF expected to be removed, etc. In some instances, selection of the PD therapy option may also allow a clinician to schedule a medical procedure to insert a catheter into the patient.

[0095] 7 shows a schematic diagram in which a clinician uses application 134 to enter treatment parameters 702, which are transmitted to analysis processor 106. Receipt of treatment parameters 702 may cause analysis processor 106 to remotely program or create a therapy program 704 for medical device 706. Analysis processor 106 may provide therapy program 704 after medical device 706 has been identified and / or configured for the patient under analysis.

[0096] The prescribed therapy, prescription, or therapy program 704 corresponds to one or more parameters that define how the medical device 706 should operate to administer therapy to the patient. For peritoneal dialysis therapy, the parameters may specify the amount (or rate) of fresh dialysis fluid to be pumped into the patient's peritoneal cavity, the amount of time the fluid should remain in the patient's peritoneal cavity (i.e., dwell time), and the amount (or rate) of spent dialysis fluid and ultrafiltration ("UF") to be pumped or drained from the patient after the dwell cycle is complete. For therapy involving multiple cycles, the parameters may specify the fill, dwell, and drain per cycle and the total number of cycles to be performed during the course of therapy (one therapy provided per day, or separate therapy provided during the day and night). Additionally, the parameters may specify the date / time / day of the week (e.g., schedule) on which the therapy should be administered by the medical fluid delivery machine. Furthermore, the parameters of a prescribed therapy may specify the total volume of dialysis fluid to be administered per therapy, as well as concentration levels of the dialysis fluid, such as glucose levels.

[0097] The medical device 706 of FIG. 7 may include a renal failure therapy machine for treating renal failure or reduced kidney function. Through dialysis, the renal failure machine removes waste, toxins, and excess water from the patient that would otherwise be removed by normally functioning kidneys. For peritoneal dialysis, the medical device 706 infuses a dialysate, also called a dialysis fluid, into the patient's peritoneal cavity via a catheter. The dialysis fluid contacts the peritoneal membrane of the peritoneal cavity. Waste, toxins, and excess water pass from the patient's bloodstream, through the peritoneal membrane, and into the dialysis fluid due to diffusion and osmosis; i.e., an osmotic gradient occurs across the membrane. An osmotic agent in the dialysis fluid provides the osmotic gradient. Spent or depleted dialysis fluid is pumped out of the patient, removing the waste, toxins, and excess water from the patient. This cycle may be repeated, for example, multiple times.

[0098] Various types of peritoneal dialysis therapies exist, including continuous ambulatory peritoneal dialysis ("CAPD"), automated peritoneal dialysis ("APD"), and tidal flow dialysis, and continuous flow peritoneal dialysis ("CFPD"). CAPD is a manual dialysis treatment. Here, the patient manually connects an implanted catheter to a drain to allow spent or spent dialysate fluid to drain from the patient's peritoneal cavity. The patient then connects the catheter to a bag of fresh dialysis fluid to infuse fresh dialysis fluid into the patient through the catheter. The patient disconnects the catheter from the fresh dialysis fluid bag and allows the dialysis fluid to dwell in the peritoneal cavity, where transfer of waste, toxins, and excess water occurs. After a dwell period, the patient repeats the manual dialysis procedure, for example, four times per day, with each treatment lasting approximately one hour. Manual peritoneal dialysis requires a significant amount of time and effort from the patient and leaves room for improvement.

[0099] Automated peritoneal dialysis ("APD") is similar to CAPD in that the dialysis treatment involves drain, fill, and dwell cycles. However, APD machines typically perform the cycles automatically while the patient sleeps. APD machines relieve patients from having to manually perform treatment cycles and from having to transport supplies during the day. APD machines fluidly connect an implanted catheter to a source or bag of fresh dialysis fluid and to a fluid drain. The APD machine pumps fresh dialysis fluid from the dialysis fluid source, through the catheter, and into the patient's peritoneal cavity. APD machines also allow the dialysis fluid to dwell within the cavity, allowing transfer of waste, toxins, and excess water to occur. The source may include multiple sterile dialysis fluid bags.

[0100] APD machines pump spent or depleted dialysate from the peritoneal cavity, through the catheter, and to a drain. As with the manual process, several drain, fill, and dwell cycles occur during dialysis. A "final fill" occurs at the end of APD, and the fluid remains in the patient's peritoneal cavity until the next treatment.

[0101] Another type of renal failure therapy that can be performed by medical device 706 is hemodialysis ("HD"), which generally uses diffusion to remove waste products from a patient's blood. A diffusion gradient occurs across a semi-permeable dialyzer between the blood and an electrolyte solution called dialysate or dialysis fluid to cause diffusion.

[0102] Hemofiltration ("HF") is an alternative renal replacement therapy that relies on the convective transport of toxins from a patient's blood. HF is accomplished by adding replacement or replacement fluid (typically 10–90 liters of such fluid) to the extracorporeal circuit between treatments. The replacement fluid and fluid accumulated by the patient between treatments are ultrafiltered over the course of HF treatment, providing a convective transport mechanism that is particularly beneficial in removing medium and large molecules. (In hemodialysis, small amounts of waste products are removed with the fluid obtained during the dialysis session; however, the solute drag from the removal of the ultrafiltrate is not sufficient to provide convective clearance.)

[0103] Hemodiafiltration ("HDF") is a treatment modality that combines convective and diffusive clearance. HDF uses dialysis fluid flowing through a dialyzer, similar to standard hemodialysis, to provide diffusive clearance. In addition, replacement fluid is provided directly to the extracorporeal circuit to provide convective clearance.

[0104] Most HD (HF, HDF) treatments are performed in centers. There is a trend toward home hemodialysis ("HHD") today, in part because HHD can be performed daily, offering therapeutic benefits over in-center hemodialysis treatments, which are typically performed two or three times a week. Studies have shown that frequent treatments remove more toxins and waste products than patients who receive less frequent, but perhaps longer, treatments. Patients who receive more frequent treatments do not experience as many downcycles compared to in-center patients who accumulate two or three days' worth of toxins prior to treatment. In some areas, the nearest dialysis center may be many miles from the patient's home, and treatment time, including the entire procedure, may take up most of the day. HHD can be performed overnight or during the day while the patient is relaxing, working, or otherwise productive.

[0105] The embodiments described in connection with medical device 706 are applicable to any medical fluid delivery system that delivers medical fluids, such as blood, dialysis fluid, replacement fluid, or intravenous medications ("IV"). The embodiments are particularly well suited for renal failure therapies, such as all forms of hemodialysis ("HD"), hemofiltration ("HF"), hemodiafiltration ("HDF"), continuous renal replacement therapy ("CRRT"), and peritoneal dialysis ("PD"), which are collectively or generally and individually referred to herein as prescription therapies or programs. The medical fluid delivery machine may alternatively be a drug delivery or nutritional fluid delivery device, such as a large-volume peristaltic-type pump or syringe pump. The machines described herein may also be used in home settings.

[0106] FIG. 8 is a flow diagram of an example procedure 800 for analyzing patient characteristic data 136 via the CKD predictive machine learning model 118 disclosed herein, according to an example embodiment of the present disclosure. While procedure 800 is described with reference to the flow diagram illustrated in FIG. 8 , it should be understood that many other ways of implementing the steps associated with procedure 800 may be used. For example, the order of many of the blocks may be changed, certain blocks may be combined with other blocks, and many of the described blocks may be optional. In an embodiment, the number of blocks may be varied based on the type of data preprocessing and filtering and / or the machine learning model developed. The actions described in procedure 800 are defined by one or more instructions stored in a memory device and may be implemented across multiple devices, including, for example, the analysis processor 106.

[0107] The exemplary procedure 800 begins when the analysis processor 106 receives patient characteristic data 136 via the application 134 on the clinician device 132 (block 802). The data 136 may be received via one or more APIs of the analysis processor 106 that are linked to inputs of the CKD stage progression prediction model 118a and / or the CKD urgent dialysis initiation prediction model 118b. In some embodiments, the analysis processor 106 determines derived characteristic data from the patient characteristic data, such as the patient's CKD stage and / or albumin / creatinine ratio (block 804). The analysis processor 106 identifies the patient's current CKD stage, which is used as an input to the CKD stage progression prediction model 118a and / or the CKD urgent dialysis initiation prediction model 118b for comparison with data classified in the same CKD stage (block 806).

[0108] The exemplary analysis processor 106 then processes the patient characteristic data 136, derived data, and / or the patient's CKD stage in the CKD stage progression prediction model 118a and / or the CKD urgent dialysis initiation prediction model 118b to identify the closest matching classification category or decile (block 808). As part of the comparison, the analysis processor 106 matches each patient characteristic to the same classified characteristic and uses one or more best-fit analyses to determine the classification for the patient under analysis. For example, the patient's blood pressure, GFR, BMI, gender, age, and albumin value are compared to the distributions for different classifications to determine their distance from a normal distribution or mean value. Differences may be summed for each characteristic or factor, and the category or decile corresponding to the lowest difference is selected for the patient. In other cases, the analysis processor 106 uses a weighted averaging routine to compile probabilities from different classification categories for each factor, such that the probabilistic outcome is a combined mixture of different classification categories based on their closeness to the patient's characterization data or factors.

[0109] The analysis processor 106 uses the matching and / or comparison to determine outcome probabilities for the patient under analysis (block 810). This includes determining the rate and stage progression probabilities from the CKD stage progression prediction model 118a and / or the probability that the patient will require dialysis from the CKD urgent dialysis initiation prediction model 118b. The models 118a and 118b generate probabilities over defined discrete time frames, including, for example, 30 days, 60 days, 90 days, 120 days, 180 days, 360 days, etc.

[0110] Analysis processor 106 then generates report 138 using the output from models 118a and 118b (block 812). Analysis processor 106 causes report 138 to be displayed in the user interface of application 134 on clinician device 132 (block 814). Analysis processor 106 may then determine whether a treatment prescription is received (block 816). If a treatment prescription is not received, example procedure 800 ends until CKD analysis is needed for another patient or again for the same patient. However, if a treatment prescription is received, analysis processor 106 causes treatment to be prescribed (block 818). This may include transmitting instructions for a dialysis machine or other medical device, instructions for catheter placement, medication instructions, and / or instructions for an application to assist the patient in making lifestyle changes. The instructions may also include a message to cause the dialysis machine or other medical device to begin treatment. The exemplary procedure 800 continues until a CKD analysis is required for another patient or again for the same patient. V. Predictive CKD machine learning model performance

[0111] As shown above, the multi-factor machine learning models 118a and 118b exhibit excellent predictive capabilities. The models 118a and 118b are not only able to utilize time-dependent data, such as laboratory values ​​that change over time, but also are able to consider as many characteristic features as the dataset presents to assess patient risk. Numerous factors and patient characteristics are considered by the models 118 when generating the algorithms. Different factors emerged as most influential in determining patient risk for each model. For example, GFR, creatinine, blood pressure, and BMI, among other top inputs, were considered in identifying patient risk for the CKD stage progression prediction model 118. Meanwhile, factors such as hemoglobin, albumin, and creatinine emerged at the top of the list for the CKD urgent dialysis initiation prediction model 118b.

[0112] The output of the CKD staging prediction model 118 can be used by the analysis processor 106 to guide clinicians to the level of care where the patient will derive the greatest benefit from delaying their progression to the next CKD stage. As can be seen in Table 4, the patients the model places in higher deciles of predicted risk progressed in stage more rapidly. 88 percent of patients predicted to progress in stage within 120 days actually did so. Thus, clinicians using the CKD staging prediction model 118 have a high level of confidence when treating patients based on their risk level. These patients require earlier and more frequent outpatient visits to address their symptoms and slow disease progression whenever possible.

[0113] Furthermore, it has been determined that the CKD staging prediction model 118a is highly robust in handling missing or incomplete data because it is based on many factors. Even when recommended care data is unknown due to missing ACR values, the CKD staging prediction model 118a continues to effectively differentiate risk. The decile analysis described above, discussed in connection with Tables 5 and 6, demonstrates predicted CKD staging rates with greater accuracy, allowing physicians to treat higher-risk patients more aggressively and refrain from using resources unnecessarily to assess lower-risk patients.

[0114] The CKD Urgent Dialysis Initiation Prediction Model 118b demonstrates accurate identification of patients at high risk for urgent initiation. As can be seen from the above chart in Table 8, 41% of patients are predicted to be at high risk for urgent dialysis initiation (decile 10) within 30 to 90 days. Because the model exhibits high sensitivity and PPV, care providers have a high probability of identifying potential urgent dialysis initiation candidates in as little as 30 days and can take appropriate anticipatory steps. Urgent, unscheduled dialysis treatments can be up to 20 times more costly than regularly scheduled treatments. Therefore, a reduction in the number of urgent treatments, along with improvements in patient care, will result in cost savings. VI. conclusion

[0115] It should be understood that various changes and modifications to the present preferred embodiments described herein will be apparent to those skilled in the art. Such changes and modifications can be made without departing from the spirit and scope of the present subject matter and without diminishing its intended benefits. Accordingly, it is intended that such changes and modifications be covered by the appended claims.

Claims

[Claim 1] The invention described in this specification.

Images