Method and system for postdialytic determination of dry weight

Abstract

A method for determining the dry weight of a patient after dialysis therapy, wherein the patient's blood volume is monitored and blood volume values are output. The blood volume values are recorded and evaluated for a predetermined period of time after reaching an ultrafiltration volume appropriately predetermined for the patient, wherein the dry weight of the patient then is determined on the basis of the rate of change of the blood volume during the predetermined period of time.

Claims

1. A method for evaluating dry weight of a patient after ultrafiltration in a dialysis therapy, comprising the steps of: recording blood volume values for a predetermined period of time after reaching an ultrafiltration volume predetermined for the patient in the dialysis therapy; evaluating whether the dry weight of the patient is reached, is high, or is low by applying a neuronal network to the recorded blood volume values, the neuronal network trained with known input-output pairs being blood volume rebound values and remaining ultrafiltration blood volume values, respectively; and adapting a subsequent dialysis therapy for the patient based on the dry weight evaluation using the recorded blood volume values.

2. The method of claim 1, further comprising the steps of: calculating hypervolemia or hypovolemia of the patient based on the recorded blood volume values; and continuing the dialysis therapy with ultrafiltration when hypervolemia of the patient is calculated and infusing a physiologic salt solution as a bolus when hypovolemia of the patient is calculated.

3. A computer program comprising code mediums for generating the steps according to claim 1 when said program is run on a computer system.

4. The method of claim 1, further comprising the step of training the learning means on the basis of data pairs from the rate of change of blood volume and the ultrafiltration volume of other patients.

5. The method of claim 1, further comprising the step of initiating termination of the dialysis therapy after expiry of a predetermined therapy duration.

6. The method of claim 1, further comprising the step of initiating continuation of the dialysis therapy without ultrafiltration until expiry of a predetermined therapy duration if a predefined dry weight has been reached before expiry of the predetermined therapy duration.

7. The method of claim 1, further comprising the step of establishing information about a hydration condition of the patient based on the blood volume values.

8. The method of claim 1, further comprising the step of continuing dialysis therapy with ultrafiltration if hypervolemia of the patient is calculated.

9. The method of claim 1, further comprising the step of terminating the dialysis therapy when a predefined dry weight for the patient is reached.

10. The method of claim 1, further comprising: checking at a predetermined time period before the end of a therapy is reached whether the desired ultrafiltration volume is reached, until the ultrafiltration is reached: carrying out the therapy with ultrafiltration if the desired ultrafiltration volume is not reached, carrying out the therapy without ultrafiltration if the desired ultrafiltration volume is reached; and storing blood volume values at the end of the therapy to record blood volume values evaluated to determine the dry weight.

Description

BRIEF DESCRIPTION OF THE DRAWINGS

(1) The invention is best understood from the following detailed description when read in connection with the accompanying drawings. Included in the drawings are the following figures:

(2) FIG. 1 shows a flow diagram of a method for determining the dry weight according to a first embodiment;

(3) FIG. 2 shows a schematic architecture of a neuronal network for use in the present invention;

(4) FIG. 3 shows a table including exemplary results of various cycles of a neuronal network; and

(5) FIG. 4 shows another table including exemplary results of various cycles of different neuronal networks having different input levels and hidden levels.

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

(6) Hereinafter, preferred embodiments of the present invention shall be described by way of the example of an adaptive measuring system for determining/judging the dry weight of a patient.

(7) Accordingly, it is intended to make an as exact statement as possible about the dry weight with the blood volume values collected during and after the dialysis. The blood volume rebound, i.e. the increase in the blood volume after termination of the therapy, is used to make a statement about reaching the dry weight and the degree of a possible hypervolemia or hypovolemia. In general, it is applicable that the higher the increase in the blood volume in the wake of therapy, the higher the degree of hypervolemia of the patient and the higher the still tolerable ultrafiltration volume/the still tolerable ultrafiltration quantity.

(8) By way of a suitable processing of the blood volume values, the increase in the blood volume upon termination of the therapy can be automatically evaluated and an exact statement about reaching the dry weight can be made. For this, the blood volume values after reaching the given/predetermined ultrafiltration value are recorded for a certain period of time and are evaluated with the aid of a learning means (e.g. a neuronal network) based on an algorithm, for example. The learning means then calculates/establishes the still missing or the superfluous ultrafiltration volume so that the physician may appropriately adapt his/her next therapies.

(9) According to the embodiments described hereinafter, a neuronal network is provided as an example of the learning means which evaluates the increase in the rebound so as to determine the required ultrafiltration volume for reaching the dry weight. The neuronal network may be trained, for example, with the aid of data of clinical studies (carried out before) or other predetermined training data. The training thus constitutes a prerequisite for the definition of an established neuronal network for determining or judging the dry weight. For said data in the case of stable dialysis patients having a known dry weight the ultrafiltration is stopped at a particular point in time before the end of therapy at a known remaining ultrafiltration volume and the blood volume rebound is recorded for 15 minutes (e.g. with the help of a hematocrit, toxin, ultrasound or substance concentration sensor which is either integrated in the dialysis machine or external). The pairs of data obtained in this way are transmitted to the network for training. The neuronal network learns from the training which blood volume rebound corresponds to which volume quantity. Following the studies, the trained neuronal network is implemented in the dialysis machine.

(10) FIG. 1 illustrates a flow diagram of a (control) method for determining the dry weight according to a first embodiment. The method may be implemented e.g. as a computer program whose code mediums (instructions etc.) generate the following steps, when the program is run on a computer system.

(11) In step 101 a dialysis therapy including blood volume monitoring is carried out. The therapy parameters (e.g. time of therapy) are adapted so that the required ultrafiltration volume is reached a predetermined period (e.g. 15 min) before the actual end of therapy. In step 102 it is checked whether or not the time of therapy minus the predetermined period has been reached already. When the therapy time minus the predetermined period is reached, the cycle proceeds along the yes branch (J) to step 103. Otherwise, the cycle returns to step 101 along the no branch (N).

(12) In step 103 it is checked whether or not the desired ultrafiltration volume is reached. Unless the ultrafiltration volume is reached, the cycle proceeds along the no branch (N) to step 104 and the therapy is continued until the ultrafiltration volume is reached or is terminated and the excessive volume will be withdrawn during the next therapy.

(13) If it is determined in step 103, on the other hand, that the ultrafiltration volume has been reached, the cycle proceeds along the yes branch (J) to step 105 and the therapy is carried out without any ultrafiltration up to the end of the therapy time. During the residual time of the predetermined period (e.g. 15 min) the blood volume monitoring is continued.

(14) In the following step 106 it is checked whether the therapy time has been reached. If this is not the case, the cycle returns to step 105 along the no branch (N). Otherwise, the cycle proceeds along the yes branch (J) to step 107 where the blood volume values, more exactly speaking the blood volume rebound, are stored for evaluating the dry weight and are evaluated by the neuronal network. After this evaluation, the neuronal network checks in the subsequent step 108 whether the dry weight has been reached. If in step 108 it is determined, based on the empirical values of the neuronal network learnt by training, that the dry weight has not yet been reached, the cycle proceeds along the no branch (N) to step 109 where it is checked whether hypervolemia or hypovolemia of the patient is given. Furthermore, the neuronal network establishes exact information about the volume quantity of the hydration condition. If in step 109 hypervolemia is calculated, the cycle proceeds along the yes branch (J) to step 110 where the patient is automatically continued being ultrafiltrated or a physician (system user) is informed about the hypervolemia.

(15) If, on the other hand, in step 109 hypovolemia is calculated, the cycle proceeds along the no branch (N) to step 111 where e.g. automatically an infusion of a physiologic salt solution is triggered as a bolus or the physician (system user) is informed about the hypovolemia.

(16) If it is determined in step 108 that the dry weight has been reached, the cycle proceeds along the yes branch (J) to step 112 and the process is terminated. The therapy thus can be terminated.

(17) According to the embodiment, a neuronal network that is individual for each patient is provided which can be individually adapted to the patient. According to the same principle of training as afore-mentioned, data pairs of rebound ultrafiltration quantity can be established for the respective patient and can be transmitted to the neuronal network for training. This is only possible, however, when the patient has reached a stable condition already and therefore the dry weight is largely known, as otherwise no statement can be made about the remaining ultrafiltration volume. If this is the case, as in the afore-described studies the ultrafiltration can be stopped from a particular point in time and the blood volume rebound can be recorded until the end of therapy. The pairs of data of the individual patient obtained in this way are input to the system for training so that the neuronal network is continuously capable of furnishing precise statements about the fluid balance of the respective patient.

(18) Another variant for the neuronal network that is individual for each patient is a continuous training for which input information is required from the medical staff. The staff members adjust an ultrafiltration volume and wait for the blood volume rebound at the end of therapy. Before, the system was trained by the data of the clinical studies. The blood volume rebound is evaluated with the aid of said neuronal network and the operator is requested to judge whether the displayed statement is correct. Unless this is the case, the operator is requested to inform the system about the hydration condition of the patient and about his/her possible adaptations to the ultrafiltration volume. Said data are stored again and are made available to another neuronal network, the personal neuronal network, for training. In this way, the personal neuronal network is automatically adjusted to the respective patient over time and with the aid of the operator.

(19) Since moreover even when making use of the network an operator may continue to be requested to check the result, the network can be newly adapted and trained at any time.

(20) The afore-described solutions promise high precision by the individual training data. Even if the weight is varying, the ratio of the blood volume rebound to the hydration condition remains unaffected.

(21) Hereinafter the applicability of the embodiments shall be discussed in detail. For this purpose, a neuronal network was chosen which evaluates blood volume values and, on the basis thereof, calculates the dry weight.

(22) FIG. 2 illustrates a schematic architecture of a neuronal network for use in the embodiments.

(23) A neuronal network may be implemented as a software solution which is applied for identifying patterns and courses, for fitting curves and for further problems. It consists of an input layer 201 which may contain any number of input parameters 20. This layer is followed by one or more hidden layers 202 which are linked to each other via different activating functions (e.g. sigmoid function). Each hidden layer 202 may have a variable number of neurons 23. Finally, the last hidden layer 202 leads to the output layer 203 which outputs one or more output parameters 24. It is important to mention that, depending on the selected number of hidden layers 202 and neurons 23, the expected results and the capability of the network may strongly vary (which can be determined with the aid of the mean square error (MSE) between the calculated output and the actual output). With a large number of layers and neurons, respectively, overfitting may easily occur, whereas a low number does not offer sufficient space for exactly weighting the output.

(24) Each neuronal network passes two steps: training and validation. For the training, there are different training algorithms such as e.g. the back-propagation process, which shall not be discussed in detail here, however. In the training known input-output pairs are transmitted to the network. The network attempts to achieve the expected output via the different activating functions to which the neurons are linked. Accordingly, the neuronal network weights the different connections 22 between the neurons 23 of the different layers in a differently strong way, until a certain error tolerance or a maximum number of training cycles has been reached. Validation is performed after the training in that again known input-output pairs are transmitted to the network and the latter compares the correct outputs to the ones calculated by itself.

(25) With an approach validation with the aid of therapy data a neuronal network was drafted, trained and validated with the help of 48 therapy data. Said therapy data comprised blood volume, ultrafiltration and blood pressure values of anonymous dialysis patients. For the training, the neuronal network requires blood volume rebound values that are missing in the therapy data. For this reason, the blood volume rebound was processed from the individual therapy data as follows.

(26) Changes in the blood volume mainly occur from the difference of ultrafiltration and refilling. This simplified consideration allows to calculate the refilling back on the basis of the blood volume and ultrafiltration data of the therapies. The calculation shall not be discussed here in detail, however. The refilling calculated in this way then was adapted (fitted) to the desired period of time after therapy taking the blood volume rebound behavior known from literature into account. With the aid of the fitted refilling curves the blood volume rebounds then could be calculated.

(27) The training was carried out with 30 out of 48 of the afore-mentioned therapy data. In the following, the best possible network parameters were determined for evaluation of the processed blood volume rebounds. For this purpose, the input parameters were transmitted to different network configurations, were trained and validated. In so doing, three different respective input parameters (blood volume rebound, blood volume rebound plus the last blood volume value of the therapy, blood volume values of the complete therapy) are encountering one to three respective hidden layers each having three to ten neurons. Each configuration was trained ten times. After completing the training cycle, the neuronal network with the best performance out of 11970 individual trainings, for example, was obtained.

(28) For the processed therapy data, a total of six of the afore-described training cycles were run so that a total of 71820 individual trainings took place.

(29) FIG. 3 illustrates a table including exemplary results of the best outputs of different cycles of a neuronal network.

(30) Accordingly, it is evident that five of the six cycles encountered the blood volume rebound as input parameter and three hidden layers, wherein only one cycle achieved the best performance with a blood volume rebound and the last blood volume value as input parameter. This cycle also had shown the best output of 27.5 ml.sup.2. It is clearly evident, however, that also the neuronal networks with an output of 30.6 ml.sup.2 to 41.5 ml.sup.2 furnished very good results. Thus, the difference of the performance (not shown in FIG. 3) in the neuronal network with the worst output of 41.5 ml.sup.2 is higher by only 1 ml than in the network with the best output of 27.5 ml.sup.2. For this reason, only the blood volume rebound can be incorporated in the neuronal network as input parameter.

(31) FIG. 4 shows another table including exemplary results of different cycles of various neuronal networks having different input levels and hidden levels.

(32) A first neuronal network 1 with a blood volume rebound as input parameter and three hidden layers, a fourth neuronal network 4 with a blood volume rebound as input parameter and three hidden layers, a fifth neuronal network 5 with a blood volume rebound and the last blood volume value as input parameter and three hidden layers, and a sixth neuronal network 6 with a blood volume rebound as input parameter and three hidden layers furnish the neuronal network having the best performance, wherein a second neuronal network 2 with a blood volume rebound as input parameter and one hidden layer and a third neuronal network 3 with blood volume values of the entire therapy as input parameter and three hidden layers provide an example of a neuronal network of poor performance. The representation of the results of the second neuronal network 2 and of the third neuronal network 3 serve for illustration of the fact that a neuronal network with a certain number of hidden layers and input parameters need not always show good performance.

(33) For implementing the afore-mentioned invention, at the beginning patients' data are necessary. These data can be collected from different dialysis centers and can be processed within the scope of a clinical study (carried out before). Within the scope of said study, stable patients having a known dry weight are involved. With said patients the therapy is stopped, before the end of therapy, at a known residual amount of ultrafiltration volume and the blood volume rebound is recorded over a particular period of time. If possible, also data are detected in which the ultrafiltration volume is even increased in a defined manner.

(34) In this way, various cases of hypervolemia and hypovolemia may be incorporated in the trainings data and the network training will be capable of furnishing better results even for extreme cases. The data obtained in this way (blood volume rebound and ultrafiltration volume) now are transmitted to a neuronal network, the main network.

(35) These results then form the basis of determining the dry weight with the aid of neuronal networks.

(36) Summing up, a system/an apparatus and a (control) method for determining the dry weight of a patient after a dialysis therapy are described, with the blood volume of the patient being monitored and blood volume values being output. The blood volume values are recorded and evaluated for a predetermined period of time after reaching an ultrafiltration volume appropriately predetermined for the patient, wherein the dry weight of the patient is then determined on the basis of the rate of change of the blood volume during a predetermined period of time.