System and method for predicting dry weight of a haemodialysis patient

Field

The present disclosure relates to haemodialysis, and more particularly to dry weight estimation in haemodialysis patients.

Background

In the absence of working kidneys, dialysis patients are unable to excrete excess fluid accumulated through eating and drinking. Dialysis acts as an artificial kidney to remove this fluid, however, there is a limit to how much fluid can be removed safely in a single session. Excess fluid accumulated in between dialysis sessions contributes to multi-organ morbidity and mortality. It is important for patients and dialysis providers to know a patient’s ideal fluid status i.e. the ‘dry weight’, or the weight at which they are neither too overhydrated nor dehydrated. The dry weight is the person’s normal weight without any extra fluid in the body. By targeting dry weight at the end of dialysis, the amount of excess fluid a patient carries in between dialysis sessions can be minimised.

Dry weight needs to be assessed periodically in order to ensure its accuracy and to prevent excess fluid accumulation in patients. However, patients’ weights change during the course of months or years due to fluctuations in muscle or fat mass. Current methods of periodically assessing the dry weight are either time-consuming and inaccurate, involving a trial-and-error clinical approach, or cumbersome and costly, involving devices such as bioimpedance and lung ultrasound. Bioimpedance measurements using the Body Composition Monitor (BCM) is a method of assessing fluid status that is performed pre-dialysis, adding an additional 10-15 mins to dialysis treatment time. Lung Ultrasound is an evaluation of the amount of extracellular fluid present in the thorax and has been used to categorise a patient’s fluid status and guide reductions in dry weight. However, this requires proficient personnel to periodically examine patient’s pre-dialysis and is thus cumbersome to administer. The Relative Blood volume method uses haematocrit estimated throughout a dialysis session to determine a patient’s relative blood volume. However, attempts to use this method to guide fluid removal on dialysis have been unsuccessful.

The difficulty in assessing and maintaining fluid status in haemodialysis patients is reflected in the high prevalence of chronic fluid overload. Algorithms that use machine learning methods for dry weight estimation have used clinical dry weight and post-dialysis weight as proxies for a patient’s ideal fluid status, and are trained to predict these targets. However, neither of these targets have been reproducibly validated as measurements of a patient’s ideal fluid status.

An existing dry weight algorithm use a Sparse Laplacian regularized Random Vector Functional Link model consisting of 7 input features (age, gender, systolic blood pressure, diastolic blood pressure, BMI, heart rate and years on dialysis) to target clinical dry weight prediction using data from 476 haemodialysis patients augmented with 10 cross fold validation. The model had a R squared= 0.9501 and root mean square error (RMSE) of 1.3136 kg. Similar performances are found with a multiple Laplacian- regularised radial basis function neural network model in the same cohort of 476 patients. In both studies, the limits of agreement remained unacceptably wide for clinical applicability (-4.4 to 4.3 % of dry weight).

Another existing dry weight algorithm creates a multi-layer perceptron model using simulated data and a limited number of input variables, including relative blood volume, bioimpedance and blood pressure. The output is a correction factor to apply to clinical dry weight that was tested in 14 patients. Lack of transparency with methods limits conclusions on this model. Yet another existing algorithm is unable to predict clinical overhydration status accurately using machine learning algorithms in a retrospective dataset of 1672 patients. The prediction accuracy for clinical overhydration status, defined as a gap between the pre-dialysis weight and clinical dry weight is less than 40% using 3 different ML methods. This model unsuccessfully targets clinical overhydration, which is subject to error.

Yet another existing algorithm includes creating a time-series based regression method using weights and ultrafiltration to predict post-dialysis weight with a Mean Absolute Error (MAE) of 0.17 kg ± 0.04 kg. However, post-dialysis weight is not an objective measurement of dry weight.

In view of the above, there is therefore an unresolved and unfulfilled need in the art for a method for an automated, accurate algorithm that outputs a validated dry weight measurement on a monthly basis and is efficient and low-cost. Such an algorithm would support the uptake of home haemodialysis therapies by allowing dry weight assessments to be performed at home.

Summary

The present invention relates to a system and method, as set out in the appended claims, to predict a dry weight of a haemodialysis patient.

In an aspect of the present invention, there is provided a method for predicting dry weight of a haemodialysis patient which includes receiving haemodialysis data pertaining to a plurality of previous haemodialysis sessions and a current haemodialysis session, determining first or second set of haemodialysis features from the received data, forming first or second linear regression models for predicting pre-OH index or post-OH index of the patient, based on the first or second set of haemodialysis features respectively, wherein the pre-OH index and post-OH index are the amount of fluid carried by a patient pre-dialysis and post-dialysis respectively. The method further includes predicting the pre-OH index or post-OH index in the current session based on the first or second regression models respectively, and calculating the dry weight of the patient by subtracting the predicted pre-OH index or post-OH index, from patient’s pre-dialysis or post-dialysis weight respectively.

In an embodiment of the present invention, the first set of haemodialysis features includes a body mass index (BMI), an average of a first slope from the Relative Body Volume (RBV) slopes calculated over a 10 min period of the current session, a standard deviation of inter-dialytic weight gain divided by number of hours between two consecutive sessions, a difference between a current and a weekly averaged post diastolic blood pressure, a standard deviation of the pre-dialysis weight over two week period, a difference between a current and weekly averaged pre-systolic blood pressure, and an area under the RBV curve with 100% RBV as the reference axis.

In an embodiment of the present invention, the second set of haemodialysis features includes a standard deviation of maximum ultra-filtration rate in 2 weekly average period, an average ultra-filtration rate in 2 week period, a difference in current post systolic and pre-systolic blood pressure, a pulse rate at the start of the current session, an inter-dialytic weight divided by post-weight, an ultra-filtration goal in millimetres, an average of pulse rate over the duration of the current session, and a minimum value of the negative RBV slopes calculated over 10 minute period in the current session.

In an embodiment of the present invention, the forming the first and or second linear regression models comprises checking multi-collinearity in the first or second regression variables, using variance inflation factor (VIF), splitting the haemodialysis data obtained from the plurality of previous haemodialysis sessions and the current haemodialysis session into test data and training data, training the first or second linear regression models using the training data, and calculating an error metrics of the trained first or second linear regression models by predicting the pre-OH index or post- OH index on the test data.

In an embodiment of the present invention, each of the first and second regression variables produce a co-efficient concordant with a degree of influence over the pre-OH index or post-OH index respectively.

In an embodiment of the present invention, the first or second linear regression models are trained to predict bioimpedance measurements of the fluid carried by the patient using Body Composition Monitor (BCM) measurements.

In an embodiment of the present invention the method comprises the steps of predicting a pre-dialysis fluid overload of a haemodialysis patient which includes receiving haemodialysis data pertaining to a two-week moving average of previous haemodialysis sessions based on a third set of haemodialysis features, and predicting the pre-OH index using the third regression model.

In an embodiment of the present invention, the third set of haemodialysis features includes a body mass index (BMI), an average difference between maximum diastolic BP and pre-dialysis diastolic BP (mmHg), an Area Under the curve (AUC) variability of RBV calculated using average real variability (%), an average of difference of pre-dialysis systolic blood pressure and post-dialysis systolic blood pressure of previous session (mmHg), an average of time elapsed since last dialysis, an average of difference of pre-dialysis diastolic blood pressure and post-dialysis diastolic blood pressure of previous session (mmHg). In another aspect of the present invention, there is provided a system to predict a dry weight of a haemodialysis patent. The system includes a processing unit, a non-transitory memory means operably coupled to the processing unit, at least one dialysis unit operably coupled to the processing unit, the memory means has a plurality of instructions stored thereon which configures the processing unit to receive haemodialysis data pertaining to a plurality of previous haemodialysis sessions and a current haemodialysis session, determine first or second set of haemodialysis features from the received data, form first or second linear regression models for predicting pre-OH index or post-OH index of the patient, based on the first or second set of haemodialysis features respectively, wherein the pre-OH index and post-OH index are the amount of fluid carried by a patient pre-dialysis and post-dialysis respectively, predict the pre-OH index or post-OH index in the current session based on the first or second regression models respectively; and calculate the dry weight of the patient by subtracting the predicted pre- OH index or post-OH index, from patient’s pre-dialysis or post-dialysis weight respectively.

Various embodiments of the present invention provide a method of dry weight assessment that relies on existing haemodialysis data, rather than in-centre clinical assessments, and would support the uptake of home haemodialysis therapies by allowing dry weight assessments to be performed at home. Said method allows an accurate prediction of dry weight by using specific, engineered variables that are associated with fluid status and combining averaged data from the preceding 1/2/4 weeks with single session in order to accurately predict the OH index, targeting the BCM OH index and thus removing multi-collinearity with weight, using the OH index as a predictive endpoint to calculate dry weight. The approach of determining dry weight by predicting bioimpedance determined excess fluid pre and post dialysis and the approach of combining single session with features created from 2- or 4-week averages of data to predict excess fluid status are the primary distinguishing features of the present invention.

There is also provided a computer program comprising program instructions for causing a computer program to carry out the above method which may be embodied on a record medium, carrier signal or read-only memory.

The present invention hence provides a robust solution to problems identified in the art. Other advantages and additional novel features of the present invention will become apparent from the subsequent detailed description.

Brief Description of Drawings

The present invention will be more clearly understood from the following description of an embodiment thereof, given by way of example only, with reference to the accompanying drawings, in which

FIG.1 is a block diagram of a system for predicting dry weight of haemodialysis patient, in accordance with an embodiment of the present invention;

FIG. 2 is a flowchart illustrating a method for predicting dry weight of haemodialysis patient, in accordance with an embodiment of the present invention; and

FIG.3 is a diagram demonstrating the time-series points used to create moving averages of haemodialysis data.

Detailed Description of Drawings

FIG.1 is a block diagram of a system 100 for predicting dry weight of haemodialysis patient, in accordance with an embodiment of the present invention. The system 100 includes at least one dialysis unit 102 for performing haemodialysis of a patient, a processing unit 104 operably coupled to the at least one dialysis unit 102 for receiving haemodialysis data therefrom, and a non-transitory memory 106 means operably coupled to the processing unit 104. In the context of the present invention, the processing unit 104 may represent a computational platform that includes components that may be in a server or another computer system, and execute, by way of a processor (e.g., a single or multiple processors) or other hardware described herein. These methods, functions and other processes may be embodied as machine-readable instructions stored on a computer-readable medium, which may be non-transitory, such as hardware storage devices (e.g., RAM (random access memory), ROM (read-only memory), EPROM (erasable, programmable ROM), EEPROM (electrically erasable, programmable ROM), hard drives, and flash memory). Based on the instructions stored in the non-transitory memory 106, the processing unit 104 is configured to execute a method of predicting a dry weight of the patient based on the haemodialysis data. The method has been illustrated in detail with reference to FIG. 2.

Referring to FIG.2, at step 202, the processing unit 104 receives haemodialysis data from at least one dialysis unit 102, pertaining to a plurality of previous haemodialysis sessions over a predefined period, and a current haemodialysis session. The pre-defined period may be 2 or 4 weeks.

At step 204, the processing unit 104 determines first or second set of haemodialysis features from the received data. The first set of haemodialysis features is used to form a first regression model for calculating a pre-OH index, and is illustrated in Table I. The pre-OH index is the amount of fluid carried by a patient pre-dialysis, i.e. before the current haemodialysis session.

Further, the second set of haemodialysis features is used to form a second regression model for calculating a post-OH index, and is illustrated in Table II. The post-OH index is the amount of fluid carried by a patient post-dialysis, i.e. after the current haemodialysis session.

Table II

Based on Tables I and II, it can be seen that the additional variables may be calculated using 1 /2/4 week moving averages (shown in FIG.3) of current variables in order to account for fluctuations over time, as well as variables that are based on the subtraction of weekly averages from targeted sessions parameters. Dry weight is a dynamic target, and an algorithm that uses dry weight needs to consider preceding data in order to make future predictions.

In an embodiment of the present invention, the determining the first or second set of features includes cleaning the data, removing nulls, outliers etc., and creating new features. The determining further includes averaging all the features in the data over specified number of previous weeks, and merging the averaged features with individual targeted session features. A novel feature of the proposed invention is the use of features created from 2-week or 4-week averages of preceding data combined with the data from a single (targeted) haemodialysis session, as well as a set of unique variables associated with excess fluid. At step 206, the processing unit 104 forms first or second linear regression models for predicting pre-OH index or post-OH index of the patient, based on the first or second set of haemodialysis features respectively.

It is to be noted that the linear regression attempts to model the relationship between two or more explanatory variables and a response variables by fitting a linear equation to the observed dataset.

Where: i is the number of samples, y is the target variable, x are the predictor variables, k is the number of predictor variables,

[3 is the coefficients, ei is the residual error

In an embodiment of the present invention, the forming the first and or second linear regression models first includes checking multi-collinearity in the first or second regression variables, using variance inflation factor (VIF) method (threshold of less than 5). VIF is a measure of the amount of multicollinearity in a set of multiple regression variables. Mathematically, the VIF for a regression model variable is equal to the ratio of the overall model variance to the variance of a model that includes only that single independent variable. This ratio is calculated for each independent variable for forming the respective linear regression model. A high VIF indicates that the associated independent variable is highly collinear with the other variables in the model. Further the features may be selected using either of below methods: RFECV, Forward selection and backward elimination. The first and second regression models may contain a number of unique features that have been created as predictors of volume excess and depletion. In an embodiment of the present invention, the proposed method may use up to unique 143 engineered variables that are associated with fluid status in order to improve prediction accuracy. Also, additional features may be calculated using variability measures, differences, confidence intervals and inter-dialytic calculations. Also, the linear regression model has the advantage of interpretability, as each engineered input variable produces a co-efficient concordant with the degree of influence over the predictor variable, the overhydration index.

Once the features are selected, the haemodialysis data may be split into test data and training data using group split cross-validation technique where the test data may be standardized based on training data. Group split allows for splitting the data into groups of patients rather than haemodialysis sessions and cross-validation allows for resampling the data in a way that uses different portions of the data to test and train in multiple iterations. In an example, the haemodialysis data may be split by group of patients into 5-folds where 1 portion out of 5 may be used for testing and rest 4 may be used for training the models in each iteration.

In the first and second linear regression models, each of the first and second regression variables produce a co-efficient concordant with a degree of influence over the pre-OH index or post-OH index respectively. In an embodiment of the present invention, the first or second linear regression models may be trained to predict bioimpedance measurements of the fluid carried by the patient using Body Composition Monitor (BCM) measurements, which is validated method of fluid assessment compared to clinical methods. The trained models use a more accurate prediction endpoint, the BCM overhydration index. In an embodiment of the present invention, the trained first or second regression models may be used for predicting the pre-OH index or post-OH index on the standardized test data and calculating corresponding error metrics - MAE and RMSE (kg).

At step 208, the processing unit 104 predicts the pre-OH index or post-OH index in the current session based on the first or second regression models respectively. At step 210, the processing unit 104 calculates the dry weight of the patient by subtracting the predicted pre-OH index or post-OH index, from patient’s pre-dialysis or post-dialysis weight respectively.

EXPERIMENTAL RESULTS

The experimental results of the first linear regression model used for predicting pre-OH index are illustrated in below Table III Table III

The experimental results of the second linear regression model used for predicting post-OH index are illustrated in below Table IV

Table IV

In contrast to methods of estimating clinical dry weight, the proposed method has been validated in multicentre studies against gold standard dilution methods of volume assessment in both healthy volunteers and haemodialysis patients. By using an objective measure of fluid status that is based on extracellular volume rather than weight, the proposed linear regression models have a more accurate prediction target and avoids multicollinearity when performing dry weight predictions.

In an embodiment of the present invention, the first or second linear regression models may be used to make automated predictions of dry weight in the dialysis unit. The models could be automated from the dialysis online portal and applied at monthly to 3 monthly intervals in order to ensure the accuracy of patients’ dry weights. Said models that output dry weight on a monthly basis would be clinically useful, low-cost, and rapidly scalable, and can be implemented as a software product that integrates with a dialysis online portal system/ clinical decision support system to provide automated, periodic dry weight updates, in addition to advice on implementation of these updates. Excess fluid in patients may be minimised by providing updated assessments of dry weight to patients and clinical staff. Clinical staff can use the algorithm output to guide ultrafiltration goals and reduce fluid overload over time. Patients can use the algorithm output as a baseline dry weight in order to assess deviations in their fluid status in between dialysis status. Patients undergoing home haemodialysis treatments rather than in-centre haemodialysis treatments can use the algorithm to guide ultrafiltration goals.

In an aspect of the present invention, there is provided a method for predicting the pre-dialysis fluid overload of a haemodialysis patient which includes receiving haemodialysis data pertaining to a two-week moving average of previous haemodialysis sessions based on a third set of haemodialysis features, wherein the pre-OH index is the amount of fluid carried by a patient pre-dialysis. The method includes predicting the pre-OH index using the third regression model. As this method does not use haemodialysis features from the same day of the prediction, it can be used to pre-dialysis to tailor treatments.

In an embodiment of the present invention, the third set of haemodialysis features is illustrated in Table V.

Table V

The third set of haemodialysis features includes a body mass index (BMI), an average difference between the maximum diastolic BP and the pre- dialysis diastolic BP (mmHg), the Area Under the curve (AUC) variability of RBV calculated using average real variability (%), an average of the predialysis systolic blood pressure minus the post-dialysis systolic blood pressure of the previous session (mmHg), an average of the hours since the last dialysis, average of the pre-dialysis diastolic blood pressure minus the post-dialysis diastolic blood pressure of the previous session (mmHg).

The experimental results of the third linear regression model used for predicting pre-OH index using weekly averages only are illustrated in below Table VI

Table VI

Although the present invention has been described with reference to specific embodiments, this description is not meant to be construed in a limiting sense. Various modifications of the disclosed embodiments, as well as alternate embodiments of the subject matter, will become apparent to persons skilled in the art upon reference to the description of the subject matter. It is therefore contemplated that such modifications can be made without departing from the spirit or scope of the present invention as defined.

Further, a person ordinarily skilled in the art will appreciate that the various illustrative method steps described in connection with the embodiments disclosed herein may be implemented using electronic hardware, or a combination of hardware and software. To clearly illustrate this interchangeability of hardware and a combination of hardware and software, various illustrations and steps have been described above, generally in terms of their functionality. Whether such functionality is implemented as hardware, or a combination of hardware and software depends upon the design choice of a person ordinarily skilled in the art. Such skilled artisans may implement the described functionality in varying ways for each particular application, but such obvious design choices should not be interpreted as causing a departure from the scope of the present invention.

The embodiments in the invention described with reference to the drawings comprise a computer apparatus and/or processes performed in a computer apparatus. However, the invention also extends to computer programs, particularly computer programs stored on or in a carrier adapted to bring the invention into practice. The program may be in the form of source code, object code, or a code intermediate source and object code, such as in partially compiled form or in any other form suitable for use in the implementation of the method according to the invention. The carrier may comprise a storage medium such as ROM, e.g. a memory stick or hard disk. The carrier may be an electrical or optical signal which may be transmitted via an electrical or an optical cable or by radio or other means.

In the specification the terms "comprise, comprises, comprised and comprising" or any variation thereof and the terms include, includes, included and including" or any variation thereof are considered to be totally interchangeable and they should all be afforded the widest possible interpretation and vice versa.

The invention is not limited to the embodiments hereinbefore described but may be varied in both construction and detail.