Determining the likelihood of kidney failure

Abstract

Prediction time t p A computer-implemented method for determining a likelihood of renal failure in a patient within an amount of time Δt is provided, the method comprising receiving input data, the input data comprising a recent creatinine level c R Or recent eGFR eGFR R (a) an initial creatinine level c0, and a time t0 or time interval ΔT0=t p -t0, (b) the initial estimated glomerular filtration rate (eGFR), eGFR0, and the time t0 or time interval ΔT0=t p -t0, (c) each time t i Multiple previous creatinine level measurements taken at i and (d) statistical parameters derived from linear regression of multiple past creatinine level measurements for each time t i Multiple historical eGFR values ​​determined in i and applying a machine learning model to the input data to generate an output indicative of the likelihood of renal failure within a given amount of time Δt. Corresponding training methods and systems are also provided.

Description

[Technical field]

[0001] The present invention relates to a computer-implemented method and system for determining the likelihood of renal failure within an amount of time Δt. The computer-implemented method uses machine learning models. [Background technology]

[0002] Chronic kidney disease (herein "CKD") is a condition in which a patient's kidneys do not function properly. CKD is a common condition that affects a huge number of people worldwide, especially the elderly. CKD is typically caused by other conditions, primarily hypertension and diabetes, which, along with many others, result in increased strain on the kidneys.

[0003] Early stage CKD is often not easily detected, however, when it is detected early, a combination of lifestyle changes and medications can be recommended and patients generally have a good prognosis.

[0004] When advanced to the end stage, CKD often eventually leads to renal failure, and patients then undergo either dialysis or kidney transplantation. However, even within the late stage patient population, CKD can progress at different rates, and often, patients with slow progression who do not require specialist care but can be treated by general practitioners are not identified and are unnecessarily referred to specialists, thus placing a higher burden on the healthcare system.

[0005] CKD is often detected by creatinine measurements during routine health examinations. However, CKD is often not formally diagnosed by physicians in patients with creatinine results indicative of CKD. Therefore, it is desirable to provide a method for reliably detecting CKD early, especially early CKD suspected to progress rapidly, so that physicians can make appropriate recommendations as early as possible. Summary of the Invention

[0006] Broadly speaking, the present invention provides a computer-implemented method for determining the likelihood of renal failure within a given timescale, which utilizes a machine learning model and parameters representative of the patient's renal function. The inventors have identified a set of features that, when input into an appropriate machine learning model, can reliably predict the likelihood of renal failure within a given timeframe. In the context of the present invention, the term "renal failure" should be understood to be synonymous with end-stage renal disease, or the point at which dialysis or transplantation is required to maintain life, i.e., when the kidneys can no longer adequately filter blood for the patient. In our data, this classification is preferably done by physician diagnosis code.

[0007] More specifically, the first aspect of the present invention provides a method for predicting a time t p

[0013] A computer-implemented method for determining a likelihood of renal failure in a patient within an amount of time Δt is provided, the computer-implemented method comprising: receiving input data, the input data being a recent creatinine level c R Or recent eGFReGFR R and, (a) An initial creatinine level, c0, and the time, t0, or time interval, ΔT0=t, at which the initial creatinine level, c0, was measured. p -t0, (b) The initial estimated glomerular filtration rate (eGFR), eGFR0, and the time t0 or time interval ΔT0=t at which the initial eGFR was determined. p -t0, (c) At each time t i Multiple previous creatinine level measurements taken at i a statistical parameter derived from a linear regression of said plurality of past creatinine level measurements with respect to (d) At each time t i Multiple historical eGFR values ​​determined in i a statistical parameter derived from linear regression of the plurality of past eGFR values ​​with respect to receiving the signal; and and applying a machine learning model to the input data to generate an output indicative of the likelihood of renal failure within the given amount of time Δt.

[0008] The machine learning model of the present invention may be a gradient boosted decision tree algorithm or a neural network model. It should be understood that the machine learning model applied to the input data is a trained machine learning model configured to generate an output indicative of the likelihood of renal failure within a given amount of time Δt based on the input data.

[0009] In the context of the present invention, the term "neural network" is used to refer to a machine learning model (or equivalently, an algorithm) composed of artificial neurons or nodes. Neural networks are sometimes called artificial neural networks because they aim to mimic the neuronal structure of the brain. Positive weights reflect excitatory connections and negative weights reflect inhibitory connections. All inputs are modified by the weights and summed, called a linear combination. An activation function may be used to control the amplitude of the output. In implementations of the first aspect of the present invention, various types of neural networks may be used. For example, the neural network may include a multi-layer perceptron (or MLP).

[0010] MLPs are trained on a dataset to find the function f(·):R m →R owhere m is the number of input dimensions and o is the number of output dimensions. Given a set of features X=x1, x2, ..., xn and a target y, a nonlinear function approximator for either classification or regression can be learned. The MLP may comprise an input layer, one or more hidden layers, and an output layer. The MLP may include 2, 3, 4, 5, 6, 7, 8, 9, or 10 hidden layers. Each hidden layer may include 50 or more nodes, 100 or more nodes, 200 or more nodes, 300 or more nodes, 400 or more nodes, or 500 or more nodes. In some cases, the number of nodes in each hidden layer is a power of 2. In one embodiment, the MLP may include 4 hidden layers, each layer having 256 nodes.

[0011] In a neural network model, each node may be associated with an activation function (or transfer function) that ultimately generates the node's output. A variety of activation functions may be used in implementations of the present invention, including linear activation function, sigmoid activation function, hyperbolic tangent activation function, rectified linear unit (ReLU) activation function, leaky ReLU activation function, parameterized ReLU activation function, exponential linear unit (ELU) activation function, swish activation function, or softmax activation function. Another activation function that the inventors have found to be effective is the miss activation function, defined as follows:

number

[0012] Hyperbolic tangent (i.e. ln(1+e x The argument of )) is sometimes called a softplus function.

[0013] Further details regarding how the neural network model may be trained are provided later in this application.

[0014] The computer implemented invention according to the first aspect of the present invention essentially relies on the use of data indicative of changes in creatinine levels of eGFR over time as input to a machine learning algorithm which then returns the probability of renal failure within a specified time. In some cases, additionally or alternatively, input features relating to cystatin-c may be used in the computer implemented method according to the first aspect of the present invention (i.e. rather than or in addition to creatinine).

[0015] In some implementations of options (a) and (b), prior to applying the machine learning model, the computer-implemented method further comprises: R , the initial creatinine level c0, and the time interval between the two, or the most recent eGFR R , the initial eGFR eGFR0, and the time interval between the two to calculate a slope over time. This slope may form part of the input data for the machine learning model. If the input data includes statistical parameters for a linear regression, the linear regression is preferably a linear fit obtained using least squares regression.

[0016] In some cases, the "statistical parameters" referred to above may be the slope or gradient of the linear regression (over time), the intercept of the linear regression (i.e., on the y-axis or on the axis representing the creatinine level or eGFR value), the error of the linear regression (with respect to the sum of the residuals), the number of points considered when constructing the linear regression, and the variance of the linear regression. The input data may include one, all, or any subset of these statistical parameters. Alternatively or additionally, rather than a statistical parameter derived from a linear regression, the computer-implemented input data may include another statistical parameter, for example, a historical mean of creatinine levels or eGFR values, a historical variance of creatinine levels or eGFR values, a historical standard deviation of creatinine levels or eGFR values, or a historical median of creatinine levels or eGFR values. Here, "history" refers to a statistical parameter that covers multiple past measurements. Other types of historical statistical parameters may also be used (again, alternatively or additionally).

[0017] A recent creatinine measurement c (as described in more detail in the "Experimental Results" section of this patent application) R And the use of a machine learning model on a minimum set of input features including one or more of (a)-(d) above has been found to provide a reliable determination of likelihood of renal failure. In particular, the computer-implemented method of the first aspect of the present invention has been found to provide a reliable determination of likelihood of renal failure for all stages of CKD (i.e., 1-5). This is advantageous over prior art methods of determining likelihood of renal failure in CKD patients.

[0018] A "recent" creatinine measurement should be understood to represent a creatinine measurement taken after any "historical" measurement. The term "recent" is not necessarily intended to specify a time frame in which the measurement should be taken, and the term is used only as a label. In some cases, a recent creatinine measurement may correspond to the most recent creatinine measurement available for a subject patient. A recent creatinine measurement may be taken after a predicted time t pThe linear regression may correspond to measurements of the patient's creatinine level at 1000 rpm. When one or more statistical parameters of the linear regression are used, the linear regression may include recent creatinine measurements. However, in some cases, the linear regression may not cover recent creatinine measurements (i.e., the linear regression may cover all of the multiple creatinine measurements except for the most recent one, which is a "recent creatinine measurement" within the meaning of this application). The selection of points considered in constructing the linear regression may include all points measured. Alternatively, the selection of points may include only points based on measurements since CKD diagnosis, or points going back a predetermined amount of time (e.g., 1 year, 2 years, 3 years, 4 years, 5 years or more).

[0019] In some cases, the computer-implemented method may include calculating a recent eGFR value from a recent creatinine measurement. The input data may then further include a recent eGFR value. Alternatively, in some cases, the input data may include a recent eGFR value instead of a recent creatinine measurement. Specifically, the computer-implemented invention may include calculating a recent eGFR value based on a recent creatinine (or cystatin-c) measurement, the patient's age, and optionally one or more of the following: the patient's sex, the patient's race, the patient's body size (e.g., in terms of body mass index or body surface area), the patient's blood urea nitrogen measurement, and the patient's serum albumin measurement. In some cases, rather than calculating an eGFR value, the computer-implemented method may further include receiving an eGFR value from an external source.

[0020] Related to the previous point, in some cases the input data may include (a) an initial creatinine level, c0, and a time, t0, or a time interval, ΔT0=t p -t0, and (b) the initial estimated glomerular filtration rate (eGFR), eGFR0, and the time t0 at which the initial eGFR was determined, or the time interval ΔT0=t p-t0, or both. In these cases, the initial eGFR value eGFR0 may be calculated from additional patient data including the initial creatinine (or cystatin-c) level c0 and age, and optionally one or more of the patient's gender, the patient's race, the patient's body size, the patient's blood urea nitrogen measurement, and the patient's serum albumin measurement. In these cases, the machine learning model may be applied to both inputs (a) and (b), or in some cases, only to input (b), which indirectly includes information from input (a).

[0021] Similarly, in some cases, the input data may be (c) i Multiple previous creatinine level measurements taken at i and (d) statistical parameters derived from linear regression of multiple past creatinine level measurements for each time t i Multiple historical eGFR values ​​determined in i In these cases, the eGFR may include both a statistical parameter derived from a linear regression of multiple past eGFR values. i Each of the following is a past creatinine level (or cystatin-c) measurement: i and additional patient data including age, and optionally one or more of the patient's sex, the patient's race, the patient's body size (e.g., with respect to body mass index and / or body surface area, example body sizes include overweight, obese and very obese), the patient's blood urea nitrogen measurement, and the patient's serum albumin measurement. As discussed above, in these cases, the machine learning model may be applied to both inputs (c) and (d), or in some cases, only to input (d), which indirectly includes information from input (c).

[0022] The more detailed the input data to the machine learning model, the more reliable the prediction. With this in mind, in some cases, it may be preferred that the inputs include two or more inputs (a)-(d), three or more inputs (a)-(d), or in some cases all four inputs (a)-(d). Of these combinations, the set of inputs including (a) and (c) or (b) and (d) is preferred as it provides two different types of information, namely information derived from a linear regression of the creatinine level or eGFR, and information about the initial creatinine level or eGFR, and the time it was taken. However, it should be noted that advantageous results compared to known prediction techniques may be obtained with any combination of inputs (a)-(d).

[0023] In some cases, the input data may include additional features to those mentioned above. Specifically, the input data may further include one or more of age, sex, race, albumin to creatinine ratio, serum albumin, serum cystatin-c, serum phosphate, serum bicarbonate, serum calcium, hemoglobin, glycated hemoglobin, blood urea nitrogen, number of acute kidney injury events, systolic blood pressure, diastolic blood pressure, resting heart rate, diabetic status, hypertensive status, and CKD diagnosis status. In this specification, when the above features refer to the name of a chemical or other species, it should be understood to mean, as appropriate, the level or measurement of the concentration of that species in, for example, blood, serum, or other body fluid. In the computer-implemented method according to the first aspect of the present invention, the input data may include one, two, three, four, five, six, seven, eight, nine, ten, eleven, twelve, thirteen, fourteen, fifteen, sixteen, seventeen, or eighteen of the additional features listed at the beginning of this paragraph.

[0024] In certain cases, the additional characteristics may include age.

[0025] In another specific case, the additional features may include albumin to creatinine ratio, hemoglobin, glycated hemoglobin, systolic blood pressure, CKD diagnosis status, gender, serum albumin, and patient gender. The list may further include blood urea nitrogen. It should be noted that in some cases, various ones of these features may be used to calculate the eGFR value. In these cases, the values ​​of these features may be considered both in this calculation and as input to the machine learning model. Alternatively, if features are used to calculate the eGFR, they may not form input features (explicitly and / or directly) to the machine learning model.

[0026] In other cases, additional features may include historical creatinine or eGFR test density, which have been shown to be highly predictive. As used herein, "historical creatinine or eGFR test density" refers to the frequency (e.g., points per year) of creatinine or eGFR measurements in a patient's medical history, which is an effective indicator of the extent to which a patient is monitored by the health care system.

[0027] The computer-implemented method according to the first aspect of the invention has been shown to be particularly useful for predicting the onset of renal failure over a timescale of 1 to 5 years. That is, the amount of time Δt is preferably 1 to 5 years. Alternatively, the computer-implemented method is preferably used to determine the likelihood of renal failure in a patient within the next 1 to 5 years. The prediction timescale may be variable, that is, a clinician or other user may select the value of Δt for which they wish to obtain the likelihood of renal failure. Thus, the input data may further include a value of Δt.

[0028] Previously, determining the likelihood of renal failure was possible (or at least reliable) only in later stages of CKD. However, the computer-implemented invention of the first aspect of the present invention has been found to provide more reliable results than known methods for end-stage CKD patients. Furthermore, the computer-implemented method of the first aspect of the present invention can also provide reliable results for early stage CKD patients, or patients who have not been diagnosed with CKD at all. Thus, in some cases, the patient may have been diagnosed with stage 1 or stage 2 CKD, or the patient may not have been diagnosed with CKD at all. Alternatively, the patient may have stage 3 (including 3a and 3b) to 5 CKD. The stages of CKD are defined as follows: - Stage 1 (G1) - Normal eGFR above 90ml / min, but other tests have detected signs of kidney damage - Stage 2 (G2) - A slightly reduced eGFR of 60-89ml / min, with other signs of kidney damage - Stage 3a (G3a) - eGFR between 45 and 59 ml / min - Stage 3b (G3b) - eGFR between 30 and 44 ml / min - Stage 4 (G4) - eGFR between 15 and 29 ml / min - Stage 5 (G5) - eGFR less than 15ml / min, meaning the kidneys have lost almost all of their function.

[0029] The output of the present invention is the likelihood of renal failure within time Δt. The output may include a probability of renal failure within time Δt. The probability may be presented as a value between 0 and 1, with 0 indicating zero likelihood and 1 indicating certain occurrence of renal failure within time Δt. Similarly, the likelihood may be presented in the form of a percentage. Alternatively or additionally, the output of the computer-implemented method may include a plot showing how the likelihood varies with the value of Δt, for example in the form of a graph with the likelihood on the y-axis and the value of Δt on the x-axis. Of course, it should be emphasized that such a plot is only reliable at the time it is generated, and the probability may change as, for example, the patient's data changes over time. Alternatively, the output may include a score (e.g., 0 to 10) that does not directly reflect the probability, but correlates with the probability. This may be obtained, for example, by multiplying the probability by 10. In an alternative case, the computer-implemented method may further include calculating a predicted time at which renal failure is most likely to occur based on the output of the machine learning model.

[0030] As explained throughout this application, one of the main objectives of the present invention is to enable the determination of whether a given patient is a "slow progressor" or a "fast progressor" with respect to CKD. Making this prediction earlier allows for better shaping of the patient's treatment plan. Thus, the computer-implemented method of the present invention may further comprise the step of determining whether the patient is a fast or slow progressor based on the output of the machine learning model. This determination may comprise comparing the output of the machine learning model (or a value representative thereof, or a value calculated therefrom) to a threshold. Then, if the output (or value) is greater than (or equal to or greater than) the threshold, the patient is determined to be a fast progressor. And, if the output (or value) is less than (or equal to or less than) the threshold, the patient is determined to be a slow progressor. In some cases, "greater than" and "less than" may be interchanged, where the values ​​correspond, for example, to the inverse or negative values ​​representing the likelihood. The value of the threshold may be based on the value of Δt and / or the stage of CKD of the patient in question.

[0031] For example, for a value of Δt in 5 years (i.e. the output of the machine learning model represents the likelihood that the patient will suffer from renal failure within 5 years), for patients in stages 3 to 5, the threshold may range from 0.050 to 0.080, preferably 0.055 to 0.070, more preferably 0.060 to 0.065, more preferably even about 0.020. In one embodiment, the threshold may be 0.062.

[0032] The threshold for determining the rate of progression for stage 1 or stage 2 patients may be slightly different than for stage 3 to 5 patients. For example, the threshold may be in the range of 0.070 to 0.100, preferably 0.075 to 0.090, and more preferably even about 0.080. In one embodiment, the threshold may be 0.081.

[0033] For a Δt of 2 years, the values ​​may be slightly different. For example, for patients with stages 3 to 5, the threshold may be in the range of 0.020 to 0.050, preferably 0.025 to 0.040, more preferably 0.030 to 0.035, and more preferably still about 0.030. In one embodiment, the threshold may be 0.032.

[0034] For stage 1 or 2 patients, the threshold value may range from 0.010 to 0.040, preferably 0.015 to 0.030, more preferably even about 0.020. In one embodiment, the threshold value may be 0.021.

[0035] Having a lower threshold favors false positive predictions, which are more appropriate for patients in stages 3-5, and it is better to err on the side of caution and exercise more caution. Having a higher threshold favors false negative predictions, which are more appropriate for patients in stages 1-2, as the results of screening should not overload the healthcare system with asymptomatic, slow or non-progressing disease.

[0036] More generally, the threshold may be generated when or after the machine learning model is trained. Specifically, after the model is trained on the training data (see the second aspect of the invention below), the computer implemented invention may further comprise determining a threshold for stage 3-5 patients and / or stage 1-2 patients based on the training data. The threshold is preferably determined based on a specificity or sensitivity threshold. Specifically, the threshold is preferably determined such that when the machine learning model is applied to the training data using that threshold, the output meets a predefined specificity or sensitivity threshold. For example, the specificity threshold may be 75%, 80%, 85%, or preferably 90%, or more preferably 95%. Similarly, the sensitivity threshold may be 75%, 80%, 85%, or preferably 90%, or more preferably 95%. Preferably, the sensitivity threshold is used for stage 3-5 patients. Also preferably, the specificity threshold is used for stage 1-2 patients. As used herein, a sensitivity of 90% may be understood to mean that the computer-implemented method of the first aspect of the invention correctly identifies 90% of rapid progression patients as fast progressors. As used herein, a specificity of 90% may be understood to mean that the computer-implemented method of the first aspect of the invention can correctly identify 90% of slow progression patients as slow progressors.

[0037] The above steps relate to the use of probabilities on a scale of 0 to 1. However, it can be appreciated that similar decisions can be made based on, for example, probabilities or scores that correlate with probabilities.

[0038] A first aspect of the invention provides a computer-implemented method. Related aspects of the invention may, for example, provide a data processing apparatus configured to perform the computer-implemented method of the first aspect of the invention. Other related aspects include a computer program product comprising instructions that, when executed by a computer, cause the computer to perform the computer-implemented method of the first aspect of the invention. Another aspect may provide a computer-readable storage medium having the computer program product stored thereon.

[0039] A first aspect of the present invention relates to the use of a machine learning model to determine the likelihood of renal failure within a time Δt. A second related aspect of the present invention provides a computer-implemented method for generating such a model. In the computer-implemented method provided by the first aspect of the present invention, it can be understood that the machine learning model can be generated using the computer-implemented method of the second aspect of the present invention.

[0040] Specifically, the second aspect of the present invention is a method for predicting a time t p A computer-implemented method for generating a machine learning model configured to determine a likelihood of renal failure in a patient within a given amount of time Δt, comprising receiving training data, the training data including a plurality of data sets representing a plurality of patients, each data set including input data and output data; 番目 For the data set, the input data is the recent creatinine level c j,R (Optionally, the time t j,R ) or recent eGFReGFR j,R (Optionally, the time t j,R ) and (a) historical creatinine level c j,H , and the time t at which it was obtained j,H , (b) Historical eGFReGFR j,H , and the time t at which it was obtained j,H , (c) each time t ij (d) statistical parameters derived from a linear regression of the multiple past creatinine level measurements for the multiple past creatinine levels measured at each time t i Multiple historical eGFR values ​​determined in i and statistical parameters derived from linear regression of a plurality of past eGFR values ​​with respect to the time t j and c j,R The interval between the times of measurement Δt j and training a machine learning model using the training data.

[0041] The computer-implemented method of the second aspect of the invention allows the generation of a machine learning model that can be used in the computer-implemented method of the first aspect of the invention. Prior to the step of training the machine learning model, the computer-implemented invention may further include a data augmentation step. This refers to a step in which the amount of data is artificially expanded to increase the amount of training data and therefore the quality of the training of the machine learning model. In particular, a set of data from a single patient may be used to generate the two or more input data, i.e., multiple "snapshots" may be taken from each patient's data to provide additional input data items. Alternatively, the multiple data sets may include one or more clusters of data sets, each cluster including multiple input data items and a respective multiple corresponding output data items, the input data items and output data items in each cluster corresponding to data obtained at different times or different timescales for the same patient. In other words, the training data may include multiple clusters, each cluster corresponding to a respective patient and including an input data item and a corresponding output data item, each input / output pair corresponding to measurements obtained at different times or different timescales.

[0042] The data augmentation techniques described above may be particularly useful in enhancing the amount of data available for patients with relatively rare conditions. Furthermore, by using the computer-implemented methods outlined above, the machine learning model can "learn" the characteristics of patients at different stages of time, e.g., different lengths of time before renal failure occurs. This helps to avoid training bias. One such condition is end-stage renal disease (ESRD), which is the final, permanent stage of CKD in which renal function has deteriorated to the extent that the kidneys can no longer provide sufficient filtration of blood for the patient. Generally, patients with ESRD can only survive if they undergo regular dialysis or transplantation. Because this is relatively rare, it may be desirable to focus data augmentation efforts on patients suffering from this condition. Thus, in some cases, the only patients for which there are relevant clusters that include multiple input data items and corresponding output data items are those diagnosed with ESRD. This essentially increases the availability of ESRD training data, which is rarer than non-ESRD data, and may result in improved learning by the machine learning model.

[0043] It may be necessary to consider "competing risks of death," i.e., the possibility that a patient whose kidneys would not function at some point may die before then for an entirely different reason. Without considering competing risks of death, non-ESRD patients are only used to make predictions within the available data; i.e., if only 2 more years of data are available, one cannot be sure that a patient will not decline within 5 years.

[0044] In the spirit of competing risks of mortality, a non-ESRD patient who dies in, say, 2 years can be used as an example of “not declining within 5 years” because the mortality event can be considered “never reaching ESRD in the future.”

[0045] So the training data changes slightly because we can use non-ESRD patients near the end of the data if they die, and the testing changes slightly because we can test the algorithm on those examples too.

[0046] The training data above includes only a data set with an indication of the time when renal failure actually occurred. However, in some cases, the patient's kidneys do not fail. To improve the performance of the machine learning model, it is useful for the training data to further include data on patients who do not suffer from renal failure. Thus, the training data may include additional pairs of data, k 番目 For further pairs, the input data is the recent creatinine level c k,R and one or more of the following: (e) historical creatinine level c k,H , and the time t at which it was obtained k,H , (f) Historical eGFR eGFR k,H , and the time t at which it was obtained k,H , (g) each time t ki Multiple previous creatinine levels measured at ki (h) a statistical parameter determined from a linear regression of a plurality of past creatinine level measurements for each time t ki Multiple historical eGFR values ​​determined in ki and a statistical parameter derived from linear regression of a plurality of past eGFR values ​​for the respective values ​​of eGFR, k,R From the time of measurement of Δt k By including training data in which renal failure did not occur, the machine learning model is likely to be trained more thoroughly and the resulting prognoses more reliable.

[0047] A further aspect of the present invention provides a renal failure likelihood determination system, comprising a processor configured to execute the computer-implemented method of the first and / or second aspects of the present invention. A further aspect of the present invention may provide a computer program comprising instructions that, when executed by a computer (or its processor), cause the computer (or its processor) to execute the computer-implemented method of the first and / or second aspects of the present invention. In some cases, the "computer" may be a renal failure likelihood determination system according to the previous aspect of the present invention. A further aspect of the present invention may provide a computer-readable medium comprising the computer program.

[0048] The present invention includes combinations of the described embodiments and preferred features unless such combinations are expressly not permitted or explicitly avoided. [Brief description of the drawings]

[0049] Embodiments of the present invention will now be described with reference to the accompanying drawings.

[0050] [Figure 1] 1 shows a system for predicting the likelihood of renal failure. [Diagram 2] 4 is a flow chart illustrating a high level method that may be performed by a likelihood determination module. [Figure 3A] 1 illustrates various types of input data for a machine learning model that may be used to determine a patient's likelihood of renal failure. [Figure 3B] 1 illustrates various types of input data for a machine learning model that may be used to determine a patient's likelihood of renal failure. [Figure 3C] 1 illustrates various types of input data for a machine learning model that may be used to determine a patient's likelihood of renal failure. [Figure 3D] 1 illustrates various types of input data for a machine learning model that may be used to determine a patient's likelihood of renal failure. [Figure 4] 1 shows an example of an output plot providing a prognosis over time. [Diagram 5] 1 is a flowchart illustrating a high level method for training a machine learning model used to determine a patient's likelihood of renal failure. [Figure 6A] 1 illustrates various types of training input data that may be used to train a machine learning method used to determine a patient's likelihood of renal failure. [Figure 6B] 1 illustrates various types of training input data that may be used to train a machine learning method used to determine a patient's likelihood of renal failure. [Figure 6C] 1 illustrates various types of training input data that may be used to train a machine learning method used to determine a patient's likelihood of renal failure. [Figure 6D] 1 illustrates various types of training input data that may be used to train a machine learning method used to determine a patient's likelihood of renal failure. [Figure 7] 1 illustrates a data augmentation process. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0051] Aspects and embodiments of the present invention will now be described with reference to the accompanying figures. Further aspects and embodiments will be apparent to those skilled in the art. All documents mentioned in this specification are incorporated herein by reference.

[0052] FIG. 1 shows an example of a renal failure probability determination system 100 that can be used to execute, for example, the computer-implemented method according to the first and / or second aspects of the present invention. The renal failure probability determination system 100 comprises an interface module 101, a processor 102, and a memory 104. The processor 102 includes a training module 106 and a probability determination module 108. The memory 104 includes a machine learning model 109, which may be, for example, a gradient boosted decision tree algorithm 110 and / or a neural network model 111 and training data 112. It can be understood that other types of machine learning models can also be used in the context of the present invention. FIG. 1 includes both a gradient boosted decision tree algorithm 110 and a neural network model 111. It should be noted that this is merely an example, and that it is by no means a requirement of the present invention that both of these features are included (indeed, different machine learning models 109) can be used.

[0053] Figure 2 is a flow chart showing the high level steps performed in a computer implemented method according to a first aspect of the present invention. In a first step S20, input data is received, for example, by an interface module 101 of the renal failure probability determination system 100, which acts as an interface through which information or data may be received from other external devices (not shown). Input data that may be received in step S20 is shown in Figures 3A-3D. In each case, the input data may be a recent creatinine measurement c R and / or recent eGFReGFR R R, as well as additional data that can be used to estimate trends in a patient's renal function. More specifically, FIG. 3A shows that the input data is an initial creatinine level c0 and a time t0, or time interval ΔT0=t p -t0 and case (a) are shown. FIG. 3B shows that the input data is the initial estimated glomerular filtration rate (eGFR), eGFR0, and the time t0 or time interval ΔT0=tp -t0 and case (b) are shown. - Figure 3C shows the input data at each time t i Multiple previous creatinine level measurements taken at i 3C shows case (c) with statistical parameters derived from linear regression of multiple past creatinine levels for the most recent creatinine measurement points c. In FIG. 3C, line L represents a linear regression with a negative slope s. In FIG. 3C, line L represents a linear regression with a negative slope s for the most recent creatinine measurement points c. R However, in an alternative embodiment, the linear regression is calculated based on a set of points including the most recent creatinine measurement points c R It is assumed that the - Figure 3D shows that the input data is i Multiple previous eGFR values ​​determined by eGFR i 3D, the line L represents the linear regression with a negative slope s, and the recent eGFR value eGFR R However, in an alternative embodiment, the linear regression is calculated based on a set of points including the most recent eGFR value eGFR R It is assumed that the

[0054] The input data may further include various other measurements or information as outlined earlier in this patent application.

[0055] In step S22 of Figure 2, the machine learning model 109 is applied to the input data, which generates an output in step S24 indicative of the likelihood that the patient will suffer from renal failure within time Δt. Step S22 may be performed by the likelihood determination module 108 of the processor 102, with the machine learning model 109 retrieved from memory 104. For example, referring to the plots shown in Figures 3A-3D, we show what is meant by time Δt. In the computer-implemented method of the first aspect of the present invention, the likelihood determination process is performed based on the likelihood of the patient suffering from renal failure within time t p(where p stands for "prediction", but it should be emphasized that this is only a label). In some cases, p is the time t when the most recent creatinine measurement or eGFR value was obtained or determined. R The likelihood estimation step may then be the same as the predicted time t p or the most recent measurement time t R and determining whether renal failure has occurred at a later time Δt, which is illustrated in Figures 3A-3D. The computer-implemented method according to the first aspect of the invention has been shown to be effective in predicting the likelihood of renal failure in a patient for values ​​of Δt in the range of 1-5 years, although larger ranges such as 1-10 years are also envisaged (as described in more detail below).

[0056] The output generated in step S24 may include a likelihood in the form of a probability on a scale of 0 to 1, or a percentage likelihood. In step S26, the output may be sent to a client device (not shown) for display to a user, such as a clinician.

[0057] In alternative cases, steps S22 and S24 may be performed multiple times for different values ​​of Δt. In these cases, the output generated in step S24 may include a plot of the determined probabilities for various values ​​of Δt, thereby showing changes to the patient's outcome over time. An example of such a plot is shown in FIG. 4.

[0058] We now discuss the operation of the training module 106 of the processor 102, and more specifically, the types of training data 112 that may be used to train the likelihood estimation model 110. An exemplary flow chart is shown in FIG. 5. In a first step S50, training data 112 is received. This training data may be stored in the memory 104 of the likelihood determination system 100. The training data may include multiple data sets, each data set including input data and output data. Broadly speaking, the input data represents values ​​(and so on) for one or more features that are input to the machine learning model 109 in step S22 of FIG. 2. Of course, actual patient data cannot be used to assign data associated with a given patient having a likelihood of renal failure, as this is not information that can be obtained from the patient. Thus, the output data may include an indication of either when renal failure occurred for that patient, or if renal failure did not occur, and an indication of the time that renal failure did not occur (or relatedly, the time interval between the most recent measurement of creatinine level or eGFR value and the time the data was obtained). Four types of training data will now be described with reference to FIGS. 6A-6D. It should be emphasized that the training data may take other specific forms. The training data comprises a plurality of j data sets, and Figures 6A-6D show diagrammatically the input data received in each data set. This will be apparent to one skilled in the art from the drawings. - Figure 6A shows the input data for the recent creatinine level c R and the time t at which it was obtained R , as well as historical creatinine levels H and the time t at which it was obtained H The data shows the case (a) including (t H ,c H ) and (t R ,c R ) of at least two points in time. This data can be used to train a machine learning model 109 that takes input data as shown in FIG. 3A above, where the time t of either a renal failure event (for cases) or a non-renal failure event (for controls) is j tp Equivalent to +Δt. - Figure 6B shows that the input data is R and the time t at which it was obtained R , as well as historical eGFReGFR H and the time t at which it was obtained H The data shows the case (b) including (t H ,eGFR H ) and (t R ,eGFR R ) of at least two points in time. This data can be used to train a machine learning model 109 that takes input data such as that shown in FIG. 3B above, where the time t of either a renal failure event (for cases) or a non-renal failure event (for controls) is j t p Equivalent to +Δt. -The data in Figure 6C shows that the input data is the recent creatinine level c R and the time t at which it was obtained R and each time t i and a statistical parameter derived from a linear regression L of the multiple past creatinine level measurements. The data may optionally include a creatinine level c i and the associated time t i In a simple case, the input data from FIG. 6C may further include one or more statistical parameters and, optionally, (t R ,c R) in the form of a data set representing the time t of either a renal failure event (for cases) or a non-renal failure event (for controls). In a simple case, this input data can be used to train a machine learning model 109 that takes input data such as that shown in FIG. 3C, since it provides input data in the form of statistical parameters (e.g., the slope s or intercept of the line L, although it should be noted that the data may not actually include the line L, but only the raw points, and the statistical parameters may be calculated during the training process). From the data shown in FIG. 6C (and indeed similar data), various additional data points can also be extracted to perform data augmentation, as will be described in more detail later in this application. Here, the time t of either a renal failure event (for cases) or a non-renal failure event (for controls) is taken as the input data. j t p Equivalent to +Δt. -The data in Figure 6D shows that the input data is the recent eGFReGFR R and the time t at which it was obtained R and each time t i Multiple previous eGFR values ​​determined by eGFR i 6D , and a statistical parameter derived from a linear regression of multiple past eGFR values ​​for t t . In the simple case, the input data from FIG. 6D includes one or more statistical parameters (e.g., the slope s or intercept of the line L, although note that the data may not actually include the line L, but only the raw points, and the statistical parameters may be calculated during the training process), and optionally (t R ,eGFR R ) in the form of a dataset representing the time course of a renal failure event (cases) or a non-renal failure event (controls). In a simple case, this input data provides input data in the form of statistical parameters, and can therefore be used to train a machine learning model 109 that takes input data such as that shown in FIG. 3D. As with FIG. 6C, various additional data points can also be extracted from the data shown in FIG. 6D (and indeed similar data) to perform data augmentation, as will be described in more detail later in this application. Here, j t p Equivalent to +Δt.

[0059] 6A-6D show an example of input data. The training data further includes output data. This may take two forms: first, for patients with renal failure, j For example, t R t (i.e., the time of the most recent creatinine measurement for the eGFR value determination). Alternatively, the absolute time of renal failure may be provided. In other cases, the patient may not be suffering from renal failure. In this case, the output data is provided from a time t k may include an indication of

[0060] Before describing the training of the machine learning model 109 in more detail, it will be described how the training data 112 can be augmented. By performing data augmentation, a larger amount of training data 112 is generated, which itself results in better training of the machine learning model 109. To illustrate this principle, refer to the plot shown in FIG. 7, which contains the same points as FIG. 6C, but in this case the linear regression L has been removed and the time t of renal failure is j From such a single data set, various data sets can be extracted that can be used to train a machine learning model 109, such as one that can be used on the data shown in Figures 3A and 3C. - The "real" latest creatinine points (c) are used to generate additional training data for training the algorithm that takes input data as shown in FIG. 3A, as shown by examples (i) and (ii) of FIG. 7 (black data points form the input data, white points are not taken into account). R ,t R ) is used, (c i ,t i ) points can be treated as history points or original points. The time interval to renal failure can also be easily calculated. i ,t i) points can be used as the nearest point (adjusting the interval to renal failure accordingly). In this way, much more data may be obtained from a single data set. The multiple data sets that may be obtained may be referred to as a cluster of data sets. In example (ii), the earliest point is used, and an earlier point than in example (i) is used as the creatinine measurement (hence, t p and t p +Δt are shifted to an earlier time point while maintaining the value of Δt). - To generate additional training data for training an algorithm that takes input data as shown in Fig. 3C, as shown by examples (iii) and (iv) in Fig. 7 (black data points form the input data, white points are not considered), the situation here is similar, except that the focus is on the statistical parameters derived from a linear regression built from multiple points. In this case, (c i ,t i ) points and a statistical parameter can be calculated for each of these subsets. The interval to time of renal failure can then be adjusted as necessary. Multiple data sets that may be obtained may be referred to as a cluster of data sets. In example (iv), the last two creatinine measurements are excluded from consideration (hence, t p and t p +Δt are shifted to an earlier time point, preserving the value of Δt).

[0061] Of course, equivalent steps may be performed for machine learning models 109 that focus on eGFR rather than creatinine levels, and for models that take input data that includes both eGFR and creatinine level information. As previously described in this application, this may be particularly useful for generating larger amounts of training data 112 for rare types of patients, such as those suffering from ESRD.

[0062] After suitable training data 112 is obtained (and optionally augmented), a training step takes place in step S52. At a high level, the gradient boosted decision tree algorithm 110 may be trained by iteratively reducing a loss function (e.g., cross-entropy) obtained by a series of weak learners (e.g., "stumps" that are single-split decision trees). More information on training gradient boosted decision tree algorithms can be found in Chen & Guestrin, "XGBoost: A Scalable Tree Boosting System" (2016) arXiv:1603.02754, focusing on XGBoost, Prokorenkova et al., "CatBoost: Unbiased Boosting with Categorical Features" (2019) arXiv:1706.09516, focusing on CatBoost, and Ke, Guolin et al., "LightGBM: Highly Efficient Gradient Boosting Decision Trees". Advances in Neural Information Processing Systems 30 (2017):3146-3154, focusing on LightGBM. For the avoidance of doubt, all these publications are incorporated herein by reference.

[0063] A neural network model 111, such as a multi-layer perceptron, may be trained using the Adam optimizer for multi-target cross-entropy loss.

[0064] After sufficient training in step S52, in step S54 a complete (i.e. trained) machine learning model 109 is output. This algorithm may then be used to perform the computer-implemented method according to the first aspect of the invention.

[0065] Now that the training and operation of the machine learning model 109 has been described, in the next section we present evidence of the effectiveness of such algorithms, specifically the gradient boosted decision tree algorithm 110 and the neural network model 111.

[0066] Experimental Methods and Results - Gradient Boosted Decision Trees A. Training Data To develop and train the CKD risk prediction, real-world data (RWD) was obtained from a database. The database includes longitudinal electronic health records (EHR) and medical insurance claims. A relevant subset of over 250,000 patients was used. A total of 49 laboratory parameters, vital signs, demographics, and diagnostic codes were extracted from this database to serve as input features for the gradient boosted decision tree algorithm utilized in an embodiment of the present invention. Chronic kidney disease (CKD) patients were identified by searching for ICD9 and ICD10 codes related to CKD (585 and N18, respectively) and with at least one measurement of serum creatinine in the EHR. CKD staging (stages 1-5) was performed based on estimated glomerular filtration rate (eGFR) values ​​recalculated using the FAS equation (see Pottel, Hans et al. "An estimated glomerular filtration rate equation for the full age spectrum". Nephrology, dialysis, transplantation: Official publication of the European Dialysis and Transplant Association - European Renal Association vol.31,5(2016):798-806.doi:10.1093 / ndt / gfv454). Patients with renal failure were identified based on medical claims data and diagnosis codes by searching for claims or diagnosis codes related to dialysis or kidney transplantation, and by searching for consistently low eGFR values.

[0067] B. Validation Data To test the trained model with unseen data, patient information was obtained from a second database: in total, data from over 650,000 relevant patients was available, and the data was preprocessed similarly to the database used for training.

[0068] C. Algorithm Training CKD risk prediction also uses gradient boosted decision tree models (in this case the CatBoost implementation of Prokhorenkova et al. (2019) (Prokorenkova et al. "CatBoost: Unbiased Boosting with Categorical Features" (2019) arXiv:1706.09516)) that have been found to perform well on tabular data such as patient data extracted from RWD databases. As explained with reference to Figure 7 of this application, the availability of longitudinal data allowed for various ways of aggregating or transforming the temporal feature data (e.g., the single measurement closest to the prediction time point, the variance of measurements over a certain time interval, the trend represented by the slope of a linear regression over a certain time interval, etc.). This feature engineering process increased the number of available features to a total of 87. Thousands of feature combinations were systematically evaluated and a core set of particularly significant features was selected. The inclusion of additional features could further improve the predictive performance, but it was shown that the magnitude of improvement decreased with each additional feature added.

[0069] The performance of CKD risk was also evaluated by using 3-fold cross-validation on the training data. In parallel, a reference method (the so-called "Kidney Failure Risk Equation", KFRE; see Tangri, Navdeep et al. "Predictive Model for Progression of Chronic Kidney Disease to Renal Failure", JAMA vol.305,15(2011):1553-9.doi:10.1001 / jama.2011.451) was evaluated on the same dataset and performance was compared by area under the receiver operating characteristic curve (AUROC or AUC).

[0070] D. Algorithm Verification To test the performance of the CKD-risk algorithm on a second, independent dataset and to compare the results with a reference method (KFRE) in an unbiased manner, the CKD-risk algo was trained on the full training data and tested on the validation data. In parallel, KFRE was also applied to the validation data and the prediction results were compared by AUROC.

[0071] E. Results "Creatinine" (corresponding to recent creatinine level) and "Creatinine_Slope" were identified as a Core Feature Set "Core Set 1" that covers options (c) and (d) of claim 1. Additional beneficial features, hereinafter collectively referred to as "add'l features", are as follows: -Albumin (albumin level) -Last_ACR (Last albumin / creatinine ratio) -CKD Diag (CKD diagnosis status) -HbA1c (glycosylated hemoglobin) -Hemoglobin (hemoglobin level) -BP systolic (systolic blood pressure) -patient_gender (patient gender)

[0072] For Δt=5 year prediction, the results are shown in the table below. These results were obtained from patients with stages 3-5 CKD. [Table 1]

[0073] Core Set 2 refers to a combination of recent creatinine levels and past creatinine values ​​(and implicitly, the time interval between the two). This corresponds to options (a) and (b) of claim 1, with the results shown below. The KFRE values ​​are as in the previous table, again for Δt=5 years, stage 3-5 patients. [Table 2]

[0074] From this, it can be seen that the use of the currently provided predictive model results in better performance than KRFE in all settings. Note that improvement is demonstrated when the feature set includes only the core set of features, and further improvement is achieved when it further includes age and the additional features listed above. It is expected that similar results will be obtained by substituting the eGFR value above with the creatinine measurement, since the two are approximately proportional to each other.

[0075] As mentioned above, the present invention is not only highly effective for predicting the probability of renal failure in patients with CKD stages 3-5, but is also useful for making predictions for patients with CKD stages 1 or 2. The results of these experiments are shown in the table below. [Table 3]

[0076] In the above table, it can be seen that there are no comparative examples based on KFRE calculations, since KFRE is designed to work only with stages 3 to 5. The values ​​shown in the table are AUC values ​​obtained using a gradient boosted decision tree algorithm as described above.

[0077] Experimental Method and Results - Neural Network Model Instead of a gradient boosted decision tree algorithm, we used a similar method to evaluate the performance of the neural network model. Specifically, a multilayer perceptron with four layers of 256 nodes using misactivation was trained using the Adam optimizer for multi-target cross-entropy loss, respectively. The neural network model was trained on a UK-based training dataset containing data on approximately 850,000 patients with CKD. Data were obtained for Δt = 3 years and Δt = 5 years. The data was cross-validated using the same UK-based training dataset, which gave slightly better results.

[0078] A similar set of results was obtained for gradient boosted decision trees.

[0079] "Creatinine" (corresponding to recent creatinine level) and "Creatinine_Slope" were identified as a Core Feature Set "Core Set 1" that covers options (c) and (d) of claim 1. Additional beneficial features, hereinafter collectively referred to as "add'l features", are as follows: -Albumin (albumin level) -Last_ACR (Last albumin / creatinine ratio) -CKD Diag (CKD diagnosis status) -HbA1c (glycated hemoglobin level) -Hemoglobin (hemoglobin level) -BP systolic (systolic blood pressure) -patient_gender (patient gender)

[0080] For Δt=3 and 5 year predictions, the results are shown in the table below. These results were obtained from patients with stages 3-5 CKD. [Table 4]

[0081] Core Set 2 refers to a combination of recent creatinine levels and past creatinine values ​​(and implicitly, the time interval between the two). This corresponds to options (a) and (b) of claim 1, with the results shown below. KFRE values ​​are as in the previous table, again for Δt=3 and 5 years, and for patients in stages 3-5. [Table 5]

[0082] The results in the table below are similar, with Δt=3 and 5 years, but obtained for stage 1 and 2 patients. [Table 6]

[0083] In the above table, it can be seen that there are no comparative examples based on KFRE calculations, since KFRE is designed to work only with stages 3 to 5. The values ​​shown in the table are the AUC values ​​obtained using the neural network model as described above.

[0084] General description of the application The features disclosed in the foregoing description, or in the following claims, or in the accompanying drawings, and expressed in a specific form or in terms of means for performing a disclosed function or a method or process for obtaining a disclosed result, may be utilized, individually or in any combination of such features, as appropriate, to realize the invention in its diverse forms.

[0085] While the present invention has been described in conjunction with the exemplary embodiments set forth above, many equivalent modifications and variations will be apparent to those skilled in the art given this disclosure. Accordingly, the exemplary embodiments of the present invention set forth above are considered to be illustrative and not limiting. Various changes may be made to the described embodiments without departing from the spirit and scope of the present invention.

[0086] For the avoidance of doubt, any theoretical explanations provided herein are provided for the purpose of enhancing the understanding of the reader, and the inventors do not wish to be bound by any of these theoretical explanations.

[0087] Any section headings used herein are for organizational purposes only and are not to be construed as limiting the subject matter described.

[0088] Throughout this specification, including the claims which follow, unless the context requires otherwise, the words "comprise" and "include", as well as variations such as "comprises", "comprising" and "including", are understood to mean the inclusion of a stated integer or step or group of integers or steps but not the exclusion of any other integer or step or group of integers or steps.

[0089] It should be noted that, as used in this specification and the appended claims, the singular forms "a," "an," and "the" include plural referents unless the context clearly dictates otherwise. Ranges may be expressed herein as from "about" one particular value and / or to "about" another particular value. When such a range is expressed, another embodiment includes from the one particular value and / or to the other particular value. Similarly, when values ​​are expressed as approximations, by use of the antecedent "about," it will be understood that the particular value forms another embodiment. The term "about" with respect to numerical values ​​is arbitrary and may mean, for example, + / - 10%.

Claims

1. Predicted time t p 1. A computer-implemented method for determining a patient's likelihood of renal failure within an amount of time Δt, comprising: receiving input data, the input data being a recent creatinine level c R or recent eGFR eGFR R and, (a) Initial creatinine level c 0 , and initial creatinine level c 0 The time t at which 0 or time interval ΔT 0 = t p -t 0 Either (b) Initial estimated glomerular filtration rate (eGFR) eGFR 0 , and the time t at which the initial eGFR was determined 0 or time interval ΔT 0 = t p -t 0 Either (c) Each time t i Multiple previous creatinine level measurements measured at i a statistical parameter derived from a linear regression of the plurality of past creatinine level measurements with respect to (d) Each time t i A plurality of past eGFR values ​​determined in i a statistical parameter derived from linear regression of the plurality of past eGFR values ​​with respect to receiving the signal, the signal including one or more of: applying a machine learning model to the input data to generate an output indicative of the likelihood of renal failure within the given amount of time Δt; 10. A computer-implemented method comprising:

2. The computer-implemented method of claim 1 , wherein the machine learning model comprises a gradient boosted decision tree algorithm or a neural network model.

3. 3. The computer-implemented method of claim 1, wherein the statistical parameters include one or more of a slope over time, an error calculated from a sum of residuals, an intercept, a number of points considered in constructing the linear regression, and a variance.

4. the input data includes (b) and / or (d); The eGFR value eGFR 0 Or each eGFR value eGFR 0 is the corresponding creatinine level c 0 and further patient data including one or more of age, sex, race, body size, blood urea nitrogen measurement, and serum albumin measurement.

3. The computer-implemented method of claim 1 or 2.

5. 3. The computer-implemented method of claim 1 or 2, wherein the input data further comprises one or more of age, albumin to creatinine ratio, serum albumin, serum cystatin-c, serum phosphate, serum bicarbonate, serum calcium, hemoglobin, glycated hemoglobin, blood urea nitrogen, number of acute kidney injury events, systolic blood pressure, diastolic blood pressure, resting heart rate, diabetic status, hypertension status, CKD diagnosis status, and patient gender.

6. The input data is Recent creatinine level R and, The patient's age and Initial creatinine level c 0 , and initial creatinine level c 0 The time t at which 0 or time interval ΔT 0 = t p -t 0 Any of the following, and Each time t i Multiple past creatinine level measurements taken at i With respect to the slope of the linear regression over time, s, and one or more of The computer-implemented method of claim 5 , comprising:

7. The input data is Recent creatinine level R and, Albumin to creatinine ratio, Serum albumin, Hemoglobin and Glycated hemoglobin and systolic blood pressure and CKD diagnostic status; The patient's gender and Initial creatinine level c 0 , and initial creatinine level c 0 The time t at which 0 or time interval ΔT 0 = t p -t 0 Any of the following, and Each time t i Multiple past creatinine level measurements taken at i With respect to the slope of the linear regression over time, s, and one or more of The computer-implemented method of claim 5 , comprising:

8. The computer-implemented method of claim 6 , wherein the input data further comprises blood urea nitrogen.

9. the amount of time Δt is between 1 and 10 years; or the input data further includes a value for Δt selectable by a user of the computer-implemented method.

3. The computer-implemented method of claim 1 or 2.

10. 3. The computer-implemented method of claim 1 or 2, further comprising determining whether the patient is a fast progressor or a slow progressor based on the output of the machine learning model.

11. The patient has been diagnosed with stage 1 or stage 2 chronic kidney disease (CKD). The patient has been diagnosed with stage 3, stage 4, or stage 5 CKD; or The patient was not diagnosed with CKD.

3. The computer-implemented method of claim 1, wherein:

12. Predicted time t p 1. A computer-implemented method for generating a machine learning model configured to determine a likelihood of renal failure in a patient within a given amount of time Δt, comprising: receiving training data, the training data including a plurality of data sets representing a plurality of patients, each data set including input data and output data, and for a j-th data set, the input data including a recent creatinine level c j,R or recent eGFR eGFR j,R and, (a) Historical creatinine level c j,H , and the time t j,H , (b) Historical eGFR eGFR j,H , and the time t j,H , (c) Each time t ij statistical parameters derived from a linear regression of the plurality of past creatinine level measurements for the plurality of past creatinine levels measured at (d) Each time t i A plurality of past eGFR values ​​determined in i and statistical parameters derived from linear regression of a plurality of past eGFR values ​​with respect to the time t j and C j,R The interval between the time of measurement Δt j receiving, including displaying, training the machine learning model using the training data; 10. A computer-implemented method comprising:

13. 13. The computer-implemented method of claim 12, wherein the plurality of datasets can include one or more clusters of datasets, each cluster including a plurality of input data items and a respective plurality of corresponding output data items, the input data items and the output data items within each cluster corresponding to data acquired for the same patient at different times or over different timescales.

14. 14. The computer-implemented method of claim 13, wherein the patients for whom there are related clusters of the dataset include patients diagnosed with end-stage renal disease (ESRD).

15. The training data includes a further plurality of pairs of data, and for the kth further pair: The input data is the most recent creatinine level c k,R and, (e) Historical creatinine level c k,H , and the time t k,H , (f) Historical eGFR eGFR k,H , and the time t k,H , (g) Each time t ki Multiple past creatinine levels measured at ki a statistical parameter determined from a linear regression of the plurality of past creatinine level measurements with respect to (h) Each time t ki A plurality of past eGFR values ​​determined in ki Statistical parameters derived from linear regression of multiple past eGFR values ​​with respect to and one or more of: The output data is k,R From the time of measurement of Δt k 15. The computer-implemented method of claim 12, further comprising an indication that no renal failure has occurred within the interval.

16. 15. The computer-implemented method of claim 1 or 2, wherein the machine learning model is generated using the computer-implemented method of claim 12 to 14.

17. Predicted time t p A renal failure likelihood determination system configured to determine the likelihood of renal failure in a patient within a time amount Δt, comprising a processor configured to perform the method of claim 1 or 2.

Images