US20130330829A1 - Prediction of acute kidney injury from a post-surgical metabolic blood panel

Info

Abstract

Description

Claims

US20130330829A1

Publication number: US20130330829A1
Application number: US13/911,236
Authority: US
Inventors: Sevag G. Demirjian
Original assignee: Cleveland Clinic Foundation
Current assignee: Cleveland Clinic Foundation
Priority date: 2012-06-06
Filing date: 2013-06-06
Publication date: 2013-12-12
Also published as: WO2013184868A1; US10281455B2; US20160377595A1

Systems and methods are provided for predicting the likelihood of acute kidney injury. An input interface is configured to receive a plurality of features derived from the results of a post-surgical metabolic blood panel and either a pre-surgical metabolic blood panel or a perisurgical metabolic blood panel. A predictive model is configured to calculate a parameter representing a likelihood of acute kidney injury from the plurality of features. A user interface is configured to provide the calculated parameter to a user in a human comprehensible form.

RELATED APPLICATION

This application claims the benefit of U.S. Provisional Patent Application Ser. No. 61/656,258, filed Jun. 6, 2012, the entirety of which is hereby incorporated by reference for all purposes.

FIELD OF THE INVENTION

The present invention relates to systems and methods for diagnosing illness in a living being and, in particular, is directed to systems and methods for predicting the occurrence of acute kidney injury from parameters derived from a post-surgical metabolic blood panel.

BACKGROUND OF THE INVENTION

Acute kidney injury (AKI) is a rapid loss of kidney function, which in its severe form, is associated with significant morbidity and mortality. Its causes are numerous and include low blood volume from any cause, exposure to substances harmful to the kidney, and obstruction of the urinary tract. As a result, AKI can be a complication from a surgical procedure, particular cardiac surgery. AKI may lead to a number of complications, including metabolic acidosis, high potassium levels, uremia, changes in body fluid balance, and effects to other organ systems. Management includes supportive care, such as renal replacement therapy, as well as treatment of the underlying disorder, and the effectiveness of treatment of AKI increases greatly with timely diagnosis.

SUMMARY OF THE INVENTION

In accordance with an aspect of the present invention, a non-transitory computer readable medium stores machine executable instructions for predicting the likelihood of acute kidney injury. An input interface is configured to receive a plurality of features derived from the results of a post-surgical metabolic blood panel and either a pre-surgical metabolic blood panel or a perisurgical metabolic blood panel. A predictive model is configured to calculate a parameter representing a likelihood of acute kidney injury from the plurality of features. A user interface is configured to provide the calculated parameter to a user in a human comprehensible form.
In accordance with another aspect of the present invention, a method is provided for predicting acute kidney injury from a surgical procedure. A first blood serum sample is isolated from a blood sample drawn from a patient before an end of the surgical procedure. A first creatinine level is determined from the first blood serum sample. A second blood serum sample is isolated from a blood sample drawn from the patient after the end of the surgical procedure. A second creatinine level is determined from the second blood serum sample. From a predictive model, a parameter is calculated representing a likelihood of acute kidney injury to the patient from at least the difference between the first creatinine level and the second creatinine level. The calculated parameter is displayed to a user.
In accordance with still another aspect of the present invention, a diagnostic system includes a processor and a non-transitory computer readable medium storing machine executable instructions executable by the processor to predict a likelihood of acute kidney injury. The instructions include an input interface configured to receive a plurality of features derived from the results of a post-surgical metabolic blood panel and either a pre-surgical metabolic blood panel or a perisurgical metabolic blood panel. The plurality of features include a difference in a serum creatinine level between the pre-surgical metabolic blood panel or the perisurgical metabolic blood panel and the post-surgical metabolic blood panel. A predictive model is configured to calculate a parameter representing a likelihood of acute kidney injury from the plurality of features. A user interface configured to provide the calculated parameter to a user in a human comprehensible form.

BRIEF DESCRIPTION OF THE DRAWINGS

The foregoing and other features of the present invention will become apparent to those skilled in the art to which the present invention relates upon reading the following description with reference to the accompanying drawings, in which:

FIG. 1 illustrates a system for predicting acute kidney injury with parameters derived from a post-surgical metabolic blood panel in accordance with an aspect of the present invention;

FIG. 2 illustrates one example of a system for predicting acute kidney injury in accordance with an aspect of the present invention;

FIG. 3 illustrates a method for predicting acute kidney injury (AKI) in accordance with an aspect of the present invention; and

FIG. 4 illustrates a computer system that can be employed to implement systems and methods described herein.

DETAILED DESCRIPTION OF THE INVENTION

FIG. 1 illustrates a system 10 for predicting the likelihood of acute kidney injury in accordance with an aspect of the present invention. Specifically, the illustrated system 10 utilizes parameters derived from a standard post-surgical metabolic blood panel to accurately predict the likelihood of an acute kidney injury, defined for the purposes of this application as a rapid loss of kidney function resulting in a need for treatment via dialysis. It will be appreciated that the illustrated system 10 can be implemented as dedicated hardware, software instructions stored on one or more computer readable media, or a combination of hardware and software. In the illustrated system 10, a software implementation is used, with a general purpose processor 12 executing machine readable instructions stored on a non-transitory computer readable medium 20.
The instructions can include an input interface 22 configured to receive a plurality of features derived from the results of a post-surgical metabolic blood panel. For example, the input interface 22 can comprise machine readable instructions for providing an input screen to a user to allow the user to enter the plurality of features via an appropriate input device (not shown), such as a keyboard, mouse, or microphone. In an alternative implementation, the input interface 22 is configured to interact with a medical database or other automated data source to extract the plurality of features from recorded results of one or more post-surgical metabolic blood panels.
In accordance with an aspect of the present invention, the plurality of features comprise various parameters derived from the determined concentrations of various substances of interest within a patient's blood following surgery. These parameters are calculated from the received post-surgical metabolic blood panel data at a feature extractor 24. For example, the substances of interest can include two or more of serum creatinine, blood urea nitrogen, potassium, sodium, bicarbonate, and albumin. The derived parameters can include, for example, the concentrations themselves, a rate of change of a given concentration since the surgery, exponential powers of the concentrations or associated rates of change, and products of concentrations and/or rates of change. For example, the derived parameters can include the concentrations of potassium, sodium, bicarbonate, and albumin, the rate of change (per hour) of serum creatinine and blood urea nitrogen, and one or more cubic terms derived from each of the rate of change of the serum creatinine and the concentrations of sodium and bicarbonate.
The parameters derived for the plurality of features are provided to a predictive model 26, which is configured to calculate a parameter representing a likelihood of acute kidney injury from the plurality of features. It will be appreciated that the predictive model 26 can be implemented as any appropriate classification or regression model, such as a polynomial model provided via least squares regression procedure, an artificial neural network, a statistical classifier, a support vector machine, or other, similar model. In the illustrated implementation, each of the plurality of features has an associated weight, and the predictive model 26 provides a linear combination of the plurality of features. It will be appreciated, however, that since the features themselves can be non-linear functions of the measured concentrations and rates of change, the results of the predictive model can be a non-linear function of the measured concentrations and rates of change. The output of the predictive model 26 represents the likelihood of an acute kidney injury requiring dialysis. An output interface 28 is configured to provide the calculated parameter to a user in a human comprehensible form. Specifically, the output interface 28 interacts with a display, printer, speaker, or other appropriate output device (not shown) to provide the calculated likelihood of acute kidney injury to a user.
FIG. 2 illustrates one example of a system 50 for predicting acute kidney injury in accordance with an aspect of the present invention. Specifically, the system 50 uses laboratory data from a first postoperative metabolic panel to calculate a change in serum creatinine (ΔCr) and blood urea nitrogen (ΔBUN) compared to values from a preoperative or perioperative metabolic blood panel. Each of ΔCr, and ΔBUN can be normalized by the time interval from surgery to the first laboratory draw. The system can also utilize serum potassium, bicarbonate, sodium, and albumin levels from the same metabolic panel to further refine the results. The predictor variables were limited to those available in routine metabolic panel which is frequently checked in the perioperative period. The illustrated system 50 can be implemented as a networked computer, comprising a processor 52, input device 54, display 56, and a non-transitory computer readable medium 60. In one implementation, the system 50 can be implemented as an online calculator accessed via a remote computer via the Internet. In such a system, it will be appreciated that the input device 54 and display 56 are located remotely from the non-transitory medium 60.
During a training process, records can be retrieved from an electronic health records (EHR) database via a database interface 62, and predictor variables and the outcomes can be extracted from these records at an associated feature extractor 63. One or more predictive models 64 can be generated at a regression engine 65 using features from these metabolic panels and the provision of dialysis within two weeks of surgery (or discharge/death if sooner), as a primary outcome measure. In the illustrated implementation, a least squares regression is performed using restricted cubic splines with either four knot locations, at the fifth, thirty-fifth, sixty-fifth, and ninety-fifth percentiles, or five knot locations, at the fifth, twenty-seven and five-tenths, fiftieth, seventy-two and five-tenths, and ninety-fifth percentiles to allow for flexible nonlinear associations with the primary outcome measure using logistic regression. The training set can be limited to fifteen events per predictor degree of freedom, inclusive of the degrees spent due to use of cubic splines, to avoid overfitting the model. In this implementation, the result of each model can be represented as a plurality of sets of at least three parameters for each feature, with a linear term and plurality of parameters representing the non-linear (e.g., cubic) terms of the fitted curve.
One example of a predictive model uses a difference between the pre/perioperative serum creatinine as the sole feature. In another example, this difference is normalized according to a length of time between the end of the surgery and the performance of the postsurgical metabolic panel, for example, as a number of hours. Using the normalized value allows for the rate of change of the creatinine to be measured, which the inventor has determined to be a better predictor of the decline in renal function. In a third model, the normalized change in serum creatinine can be used along with the preoperative serum creatinine to further refine the model.
For a fourth model, readily available and frequently tested early postoperative basic metabolic panel parameters can be used to provide an extended model. Specifically, the extended model includes preoperative serum creatinine, the normalized change in creatinine and BUN from preoperative values, and serum sodium, potassium, bicarbonate, and albumin. The inclusion of the additional elements in the panel such as potassium, sodium, and bicarbonate also independently contribute to the predictive capacity of the model because they capture clinically relevant downstream complications of renal injury in a multidimensional approach, such as electrolyte perturbations, and acid-base derangements. Serum albumin adjusts for volume shifts that typically occur during surgery, which account not only for the volume burden but also influence the rate and magnitude of change of all elements in the metabolic panel. Table 1 provides example values for each of these four models as determined by the inventor.

Model 1
Linear term	−5.03	(<.001)
1^stcubic term	38.42	(<.001)
2^ndcubic term	−138.03	(<.001)

P, linear^b	<.001
P, nonlinear^c	<.001
Model 2

Linear term	−16.61	(<.001)
1^stcubic term	−198.86	(.007)
2^ndcubic term	2486.9	(<.001)
3^rdcubic term	−7563.4	(<.001)

P, linear^b	<.001
P, nonlinear^c	<.001
Model 3

Linear term	−9.94	(<.001)	−1.56	(.11)
1^stcubic term	−33.5	(.65)	24.41	(<.001)
2^ndcubic term	1261.8	(.006)	−55.20	(<.001)
3^rdcubic term	−4527.0	(<.001)

P, linear^b	<.001	<.001
P, nonlinear^c	<.001	<.001
Model 4

Linear term	−4.31	(.10)	−0.79	(.45)	−0.318	(.001)	−0.129	(<.001)
1^stcubic term	52.09	(.50)	19.22	(.003)	−1.006	(.37)	0.267	(0.03)
2^ndcubic term	649.12	(.17)	−43.84	(.002)	12.374	(.46)	−0.601	(.32)
3^rdcubic term	−3077.35	(.005)			−22.213	(.56)

P, linear^b	<.001	<.001	<.001	<.001
P, nonlinear^c	<.001	.003	.07	<.001

	Potassium	Bicarbonate	Albumin	C statistic

Model 1				0.768
Linear term
1^stcubic term
2^ndcubic term
P, linear^b
P, nonlinear^c
Model 2				0.779
Linear term
1^stcubic term
2^ndcubic term
3^rdcubic term
P, linear^b
P, nonlinear^c
Model 3				0.786
Linear term
1^stcubic term
2^ndcubic term
3^rdcubic term
P, linear^b
P, nonlinear^c
Model 4

Linear term	−0.728	(.03)	−1.1243	(<.001)	−1.344	(<.001)	0.899
1^stcubic term	1.170	(.48)	0.0761	(.51)	1.007	(.22)
2^ndcubic term	−2.121	(.68)	0.2344	(.72)	−3.948	(.49)
3^rdcubic term

P, linear^b	.02	<.001	<.001
P, nonlinear^c	.10	<.001	.16

During operation, a user can use the input device 54 to provide values for the features via an input interface 66. These values are then provided to the predictive model 64 which calculates a parameter representing the likelihood of acute kidney injury requiring dialysis as a weighted non-linear combination of the plurality of features based upon the set of linear and non-linear weights associated with each feature. The resulting parameter can be provided to a display via a display interface 68 to inform the user of the likelihood that a patient represented by the features will require dialysis.
It will be appreciated that the models 64 are based solely on routine metabolic panel that can be obtained within hours of surgery, allowing for early diagnosis of acute kidney injury. The extended model also incorporates the time lapse from the injury event to the laboratory draw, without posing any time restrictions. The use of readily available, standardized, and affordable predictors from routine metabolic panel makes the predictive model easily applicable to any patient. While the example given in Table 1 represents patients undergoing cardiac surgery, the model could easily be extended, with appropriate validation, to virtually any non-cardiac setting. Finally, the model has is sufficiently sophisticated to capture complex non-linear relationships between the parameters associated with the routine metabolic blood panel and
FIG. 3 illustrates a method 100 for predicting acute kidney injury (AKI) in accordance with an aspect of the present invention. Specifically, the method uses changes in basic metabolic panel early after surgery to predict severe AKI requiring dialysis within a two week period following surgery. At 102, blood drawn for a preoperative or perioperative metabolic blood panel is mechanically separated to isolate a first serum sample. For example, a blood sample can be drawn from the patient and spun in a centrifuge to separate out blood serum. At 104, a preoperative or perioperative level of serum creatinine is determined from the first blood serum sample. It will be appreciated, however, that other parameters can be extracted from the preoperative or perioperative metabolic blood panel, for example, a blood urea nitrogen level.
At 106, blood drawn for a post-operative metabolic blood panel is mechanically separated to isolate a second serum sample. At 108, a post-operative level of serum creatinine is determined from the second blood serum sample. It will be appreciated, however, that other parameters can be extracted from the preoperative or perioperative metabolic blood panel, for example, levels for blood urea nitrogen, sodium, potassium, bicarbonate, and albumin. At 110, a parameter representing a likelihood of acute kidney injury to the patient is calculated from at least a difference between the preoperative or perioperative creatinine level and the post-operative creatinine level. In one implementation, this difference can be normalized according to the length of a period of time (e.g., in hours) between the end of the surgical procedure and the drawing of the post-operative metabolic blood panel. It will be appreciated that other features can be used to refine the calculation including any or all the sodium level, potassium level, bicarbonate level, and albumin level determined from the post-operative blood serum sample, the creatinine level from the first serum sample, and a difference between the blood urea nitrogen level from the first serum sample and the blood urea nitrogen level from the second serum sample normalized by the period of time between the end of the surgical procedure and the drawing of the post-operative metabolic blood panel. At 112, the calculated parameter is provided to a user at an associated display. If the parameter is within a range associated with a severe risk of acute kidney injury, the user can begin appropriate treatment.
FIG. 4 illustrates a computer system 200 that can be employed to implement systems and methods described herein, such as based on computer executable instructions running on the computer system. The user may be permitted to preoperatively simulate the planned surgical procedure using the computer system 200 as desired. The computer system 200 can be implemented on one or more general purpose networked computer systems, embedded computer systems, routers, switches, server devices, client devices, various intermediate devices/nodes and/or stand alone computer systems. Additionally, the computer system 200 can be implemented as part of the computer-aided engineering (CAE) tool running computer executable instructions to perform a method as described herein.
The computer system 200 includes a processor 202 and a system memory 204. Dual microprocessors and other multi-processor architectures can also be utilized as the processor 202. The processor 202 and system memory 204 can be coupled by any of several types of bus structures, including a memory bus or memory controller, a peripheral bus, and a local bus using any of a variety of bus architectures. The system memory 204 includes read only memory (ROM) 206 and random access memory (RAM) 208. A basic input/output system (BIOS) can reside in the ROM 206, generally containing the basic routines that help to transfer information between elements within the computer system 200, such as a reset or power-up.
The computer system 200 can include one or more types of long-term data storage 210, including a hard disk drive, a magnetic disk drive, (e.g., to read from or write to a removable disk), and an optical disk drive, (e.g., for reading a CD-ROM or DVD disk or to read from or write to other optical media). The long-term data storage 210 can be connected to the processor 202 by a drive interface 212. The long-term data storage 210 components provide nonvolatile storage of data, data structures, and computer-executable instructions for the computer system 200. A number of program modules may also be stored in one or more of the drives as well as in the RAM 208, including an operating system, one or more application programs, other program modules, and program data.
A user may enter commands and information into the computer system 200 through one or more input devices 222, such as a keyboard or a pointing device (e.g., a mouse). These and other input devices are often connected to the processor 202 through a device interface 224. For example, the input devices can be connected to the system bus by one or more a parallel port, a serial port or a universal serial bus (USB). One or more output device(s) 226, such as a visual display device or printer, can also be connected to the processor 202 via the device interface 224.
The computer system 200 may operate in a networked environment using logical connections (e.g., a local area network (LAN) or wide area network (WAN) to one or more remote computers 230. A given remote computer 230 may be a workstation, a computer system, a router, a peer device or other common network node, and typically includes many or all of the elements described relative to the computer system 200. The computer system 200 can communicate with the remote computers 230 via a network interface 232, such as a wired or wireless network interface card or modem. In a networked environment, application programs and program data depicted relative to the computer system 200, or portions thereof, may be stored in memory associated with the remote computers 230.
It will be understood that the above description of the present invention is susceptible to various modifications, changes and adaptations, and the same are intended to be comprehended within the meaning and range of equivalents of the appended claims. The presently disclosed embodiments are considered in all respects to be illustrative, and not restrictive. The scope of the invention is indicated by the appended claims, rather than the foregoing description, and all changes that come within the meaning and range of equivalence thereof are intended to be embraced therein.

ΔCr/hour^a

Preoperative Cr

ΔBUN/hour^a

Having described the invention, the following is claimed:

1. A non-transitory computer readable medium storing machine executable instructions for predicting the likelihood of acute kidney injury comprising:

an input interface configured to receive a plurality of features derived from the results of a post-surgical metabolic blood panel and one of a pre-surgical metabolic blood panel and a perisurgical metabolic blood panel;

a predictive model configured to calculate a parameter representing a likelihood of acute kidney injury from the plurality of features; and

a user interface configured to provide the calculated parameter to a user in a human comprehensible form.

2. The non-transitory computer readable medium of claim 1, wherein the plurality of features includes a change in a creatinine level between the one of the pre-surgical metabolic blood panel and the perisurgical metabolic blood panel and the post-surgical blood panel.

3. The non-transitory computer readable medium of claim 2, wherein the change in the creatinine level is normalized by a time elapsed between a surgery and the post-surgical metabolic blood panel.

4. The non-transitory computer readable medium of claim 1, wherein the plurality of features includes a sodium level from the post-surgical metabolic blood panel.

5. The non-transitory computer readable medium of claim 1, wherein the plurality of features includes a potassium level from the post-surgical metabolic blood panel.

6. The non-transitory computer readable medium of claim 1, wherein the plurality of features includes a bicarbonate level from the post-surgical metabolic blood panel.

7. The non-transitory computer readable medium of claim 1, wherein the plurality of features includes a creatinine level from the post-surgical metabolic blood panel.

8. The non-transitory computer readable medium of claim 1, wherein the plurality of features includes an albumin level from the post-surgical metabolic blood panel.

9. The non-transitory computer readable medium of claim 1, wherein the plurality of features includes a change in a blood urea nitrogen level between the one of the pre-surgical metabolic blood panel and the perisurgical metabolic blood panel and the post-surgical blood panel.

10. The non-transitory computer readable medium of claim 9, wherein the change in the blood urea nitrogen level is normalized by a time elapsed between a surgery and the post-surgical metabolic blood panel.

11. The non-transitory computer readable medium of claim 1, wherein the predictive model is configured to calculate the parameter representing a likelihood of acute kidney injury as a weighted combination of the plurality of features.

12. The non-transitory computer readable medium of claim 11, wherein a first feature of the plurality of features is a non-linear function of a second feature of the plurality of features.

13. A method for predicting acute kidney injury associated with a surgical procedure comprising:

isolating a first blood serum sample from a blood sample drawn from a patient before an end of the surgical procedure;

determining at least a first creatinine level from the first blood serum sample;

isolating a second blood serum sample from a blood sample drawn from the patient after the end of the surgical procedure;

determining at least a second creatinine level from the second blood serum sample;

calculating, from a predictive model, a parameter representing a likelihood of acute kidney injury to the patient from at least a difference between the first creatinine level and the second creatinine level; and

displaying the calculated parameter to a user.

14. The method of claim 13, wherein the difference between the first creatinine level is normalized according to a length of time between the end of the surgical procedure and drawing of the blood sample from the patient after the end of the surgical procedure.

15. The method of claim 14, wherein calculating the parameter representing the likelihood of acute kidney injury to the patient comprises calculating the parameter from the normalized difference between the first creatinine level and the second creatinine level and the first creatinine level.

16. The method of claim 15, further comprising:

determining a first blood urea nitrogen level from the first blood serum sample; and

determining a second blood urea nitrogen level from the second blood serum sample;

wherein calculating the parameter representing the likelihood of acute kidney injury to the patient comprises calculating the parameter from the normalized difference between the first creatinine level and the second creatinine level and a difference between the first second blood urea level and the second second blood urea level, normalized by the length of time.

17. The method of claim 16, further comprising determining each of a sodium level, a potassium level, a bicarbonate level, and an albumin level from the second blood serum sample, and calculating the parameter representing the likelihood of acute kidney injury to the patient comprises calculating the parameter from the normalized difference between the first creatinine level and the second creatinine level, the normalized difference between the first blood urea nitrogen level and the second blood urea nitrogen level, the first creatinine level, the sodium level, the potassium level, the bicarbonate level, and the albumin level.

18. A diagnostic system for predicting acute kidney injury associated with a surgical procedure comprising:

a processor; and

a non-transitory computer readable medium storing machine executable instructions executable by the processor to predict a likelihood of acute kidney injury, the instructions comprising:

an input interface configured to receive a plurality of features derived from the results of a post-surgical metabolic blood panel and one of a pre-surgical metabolic blood panel and a perisurgical metabolic blood panel, the plurality of features including at least a difference in a serum creatinine level between the one of the pre-surgical metabolic blood panel and the perisurgical metabolic blood panel and the post-surgical metabolic blood panel;

19. The diagnostic system of claim 18, wherein the input interface is configured to receive an input from a remote user via an Internet connection, and the user interface is configured to provide the calculated parameter to a display associated with the remote user via the Internet connection.

20. The diagnostic system of claim 18, wherein the predictive model is configured to calculate the parameter representing a likelihood of acute kidney injury as a weighted non-linear combination of a ratio of the difference in a serum creatinine level and a period of time between an end of the surgical procedure and a drawing of the post-surgical metabolic blood panel, a ratio of a difference between a presurgical blood urea nitrogen level and a post-surgical blood urea nitrogen level to the period of time, the presurgical creatinine level, a post-surgical sodium level, a post-surgical potassium level, a post-surgical bicarbonate level, and a post-surgical albumin level.