Systems and methods for predicting and detecting postoperative complications

Abstract

Disclosed herein are computer-implemented methods and systems for assessing a risk value of a target patient, the method comprising: receiving, at a server, target patient data via a connected device; at the server, one or more risk values associated with the target patient data are estimated using one or more risk assessment models. The target patient data may include historical data, patient population level data, and real-time data. The risk assessment model may include one or more of the following: a historical data risk assessment model, a real-time data risk assessment model or a combined risk assessment model. Methods for training such models are described. The described systems and methods preferably enable continuous patient monitoring based on patient's historical data and continuous real-time physiological data.

Description

[0001] CROSS-REFERENCE TO RELATED APPLICATIONS

[0002] This application claims priority to U.S. Provisional Patent Application No. 63 / 357,370, filed on June 30, 2022, the entire disclosure of which is incorporated herein by reference. Technical Field

[0003] The present disclosure relates to systems, methods and devices for monitoring, predicting and detecting different forms of postoperative complications. The present invention relates to patient monitoring and risk assessment, and more particularly, to computer-implemented systems and methods for monitoring and assessing patient risk, particularly postoperative patient risk. Background Art

[0004] Surgical procedures may use open and minimally invasive techniques on a user, such as a patient, in order to identify and treat a pathological condition or improve body function. Surgery performed for a variety of reasons has an inherent risk of postoperative complications, such as bleeding, infection, and leakage.

[0005] One of the most dangerous complications of surgery is the complication known as anastomotic leak. Anastomotic leaks can occur after an anastomosis is performed where two organs are surgically connected, and are most common in gastrointestinal surgery. Anastomotic leaks result in leakage of luminal contents into the peritoneal cavity, which can lead to a range of fatal complications. This often involves a form of severe sepsis, peritonitis, and can result in death.

[0006] Using traditional techniques, it can take an average of three to seven days to diagnose a leak. This is extremely dangerous, especially considering that every hour of delay leads to a significant increase in the risk of morbidity and mortality for the patient.

[0007] Every year, 70 million major abdominal surgeries (MAS) are performed worldwide. MAS include pancreatic, hepatobiliary and colorectal surgeries with primary anastomosis. These surgeries have a complication rate of 30-60%, of which 20% have significant adverse effects, requiring invasive treatment and enhanced patient monitoring. The occurrence of such complications can have serious consequences, both short-term and long-term, including considerable morbidity and mortality for affected patients.

[0008] Postoperative complications may further include, but are not limited to, hemorrhage, postoperative leakage, ischemia, infection, and sepsis.

[0009] Typically, medical providers wait for clinical factors such as abdominal pain, fever, and tachycardia to develop before making a diagnosis. Existing technologies for detecting postoperative complications such as anastomotic leaks can be non-specific, inefficient, time-consuming, expensive, and / or lack the ability to provide real-time detection of complications.

[0010] Additionally, even with the large amount of patient data available, current technology cannot provide care providers with accurate predictions of whether a patient will develop a postoperative complication.

[0011] Current patient monitoring systems may not be aware of historical data and often rely on physician assessments, which can be prone to bias. These monitoring systems may not be configured to receive continuous real-time data from patients, such as biosignal data or post-operative data (including but not limited to vital sign measurements).

[0012] Alternatively, prior art patient monitoring systems may rely on static models that may not be continually updated based on newly acquired patient data.

[0013] Such models are usually trained on data that may have missing values. Common approaches to overcome the missing data problem include missing value imputation or completely removing missing value entries from the dataset.

[0014] Removing patients with missing values ​​limits the available number of patients for training, and if the number of imputed values ​​is greater than the number of available values, the imputation may be problematic and may be non-trivial for the complete set of sensor measurements.

[0015] US20220361823 discloses a wearable blood pressure biosensor, system, and method for short-term blood pressure prediction of average blood pressure for a future time period. The system may include a model trained to predict future systolic blood pressure values, diastolic blood pressure values, and / or trends. The model can be trained based on data related to the user's personalized physical characteristics. However, these models do not receive data from multiple sensors that receive data from multiple patients, but rather, they are personalized for the user, so these models are developed from scratch for each new user.

[0016] US10463312 discloses an embodiment of a method and system for predicting patient mortality. The method includes classifying historical data into a first category and a second category. The method also includes determining a first test parameter and a second test parameter based on at least one of the sample data of a first patient and the historical data corresponding to at least one of the first category and the second category. The method also includes determining a probability score based on the cumulative distribution of at least one of the first test parameter and the second test parameter. The method also includes classifying the sample data into one of the first category and the second category based on the probability score. In addition, the method includes predicting the death of the first patient based on at least the classification of the sample data of the first patient. The limitation of this death prediction method is that it does not associate biological data with a given timestamp and essentially cannot provide real-time predictions. It is configured to predict the death of a patient based on static, measured biological data, and does not associate the time of the measured data with the probability of death because the model is trained under static physiological conditions.

[0017] US20220238235 discloses a computer-implemented method for providing patient outcome tracking, which may include generating an event trigger for a patient, wherein during the duration of the patient's recovery, when the patient is performing a postoperative activity related to the patient's recovery, the event trigger may correspond to a value of a patient biomarker above or below a threshold. The computer-implemented method may include receiving actual patient biomarker data from the patient's patient sensor system while the patient is performing postoperative activities. If the actual patient biomarker data includes a value above or below a threshold when the patient is performing postoperative activities, the method may include triggering an event trigger. The method may include generating a notification alarm corresponding to the event trigger. Outside the hospital, the patient sensor system includes medical wearable devices such as watches and headbands, which means that the method is not configured to receive sensor information related to the patient's continuous flow of biological fluids after surgery, but rather to receive sensor information related to externally measured data such as heart rate. In this sense, it can monitor the patient's recovery based on these external factors, but may not determine surgical-specific postoperative risks such as leakage or infection based on continuous measurements of biomarkers from online devices.

[0018] A new method and system have been developed to collect continuous biomarker data from patients to detect postoperative complications earlier than the standard of care (SOC). This system includes a novel method that continuously calculates the risk probability of a specific patient developing a postoperative complication at a specific time point. This enables healthcare providers to detect postoperative complications earlier than the SOC, thereby improving patient outcomes.

[0019] Despite significant advances in postoperative care and the development of a variety of novel monitoring technologies, to date, there are no devices or technologies specifically designed for early detection or prediction of postoperative complications after surgery. Therefore, systems and methods that can perform continuous patient monitoring and risk assessment using a patient's historical data, real-time data, or a combination of both may be needed. Summary of the invention

[0020] A simplified summary of the general inventive concepts described herein is given below to provide a basic understanding of some aspects of the present disclosure. This summary is not an extensive review of the present disclosure. It is not intended to limit the key or important elements of the embodiments of the present disclosure, or to depict the scope of the present disclosure beyond the scope explicitly or implicitly described by the following description and claims.

[0021] In some cases, the incidence of leakage from surgical procedures can range between 1% and 40%. The reasons why anastomotic leaks occur are still under investigation, and a definitive cause has not yet been determined. However, there are risk factors associated with a higher incidence, such as, but not limited to, age, gender, organ tension, ischemia, medical history, and surgical errors.

[0022] Before and during anastomotic leaks, physiological changes may occur, such as tissue ischemia followed by necrosis. These changes may manifest themselves as subtle changes in the patient's vital sign parameters that may not be detected by current standards of care.

[0023] The object of the present invention is to provide a system and method for predicting and detecting postoperative complications.

[0024] The present disclosure provides systems, methods and devices for analyzing body fluids and cavity fluids, including but not limited to peritoneal fluid, peritoneal drainage fluid, pleural drainage fluid, gastric juice, feces, bile, urine, amniotic fluid, dialysate, sebum or blood. Changes and trends of specific analytes and biological characteristics in fluids can be continuously monitored. Examples of these characteristics and analytes include but are not limited to pH, lactate, electrolytes, impedance, conductivity, dissolved oxygen, dissolved CO2, temperature, inflammatory markers, enzymes, bacterial proteins, RNA or lipids. The systems, methods and devices disclosed herein can be used for various diagnostic applications, such as but not limited to postoperative leakage, ischemia, infection and sepsis.

[0025] In some embodiments, sensors such as biosensors can be placed on catheters that can be inserted into the body and can enable fluids to be injected into or withdrawn from the body. The catheter can be placed near the surgical site to monitor the biofluid environment near the area. Fluids can be directly sensed locally without the need for negative pressure, or negative pressure can be used to help drive the fluid through the catheter. Any number of sensors can be placed on the surface of the catheter so that it is in direct contact with the biofluid around the area of ​​interest, such as the sutures in the case of anastomosis, for example. Sensors can also be placed inside a catheter, balloon, pump, or any conduit that can collect fluid.

[0026] In another embodiment, the sensor can be mounted in a system that can be connected to a catheter. The catheter can be placed near the surgical site to monitor the peritoneal fluid environment near the area. The system can be an extension of an existing catheter system. The system can be attached at any time when the catheter is placed or later.

[0027] Conveniently, early management using the techniques disclosed herein to address complications and resolve leaks can significantly mitigate the risks associated with such complications.

[0028] Prior art for handling complications may involve the use of interventional radiology techniques to handle the patient's existing complications. In the case of a leak, this may involve techniques such as placement of a drain, placement of a stent, reinforcement of staples and sutures. These interventions may be performed endoscopically without the need to perform a second surgical procedure. Monitoring the user's status using the systems and techniques disclosed herein may allow for more effective treatment planning and earlier intervention when complications recur.

[0029] Additionally, as more patients move to outpatient monitoring settings, the technology disclosed herein may allow for home monitoring of the postoperative journey. Furthermore, the technology disclosed herein may allow a user to continuously monitor the status of a patient's surgical procedure, which may be an improvement over existing diagnostic tests that obtain samples at specific time points that may not indicate the patient's status.

[0030] According to one aspect of the present invention, a computer-implemented method for monitoring drainage fluid from a patient is provided, the method comprising: receiving biosignal data from one or more biosensors coupled to the patient, the sensors being in communication with the drainage fluid and storing pH data in a memory; calculating a cumulative average of the percentage of data points in a current postoperative time period that are below a predetermined baseline threshold; if the cumulative average exceeds the predetermined baseline threshold, marking the current postoperative time period as "high risk", otherwise, indicating it as "low risk".

[0031] According to another aspect of the present invention, a system for monitoring a user is provided, the system comprising: a processor; a memory in communication with the processor, the memory storing instructions, which, when executed by the processor, cause the processor to perform the above method.

[0032] According to one aspect, a monitoring device is provided, comprising: an input port attachable for fluid communication with a catheter, the catheter being inserted into a user's body to receive fluid from the user's body; an output port generally parallel to the input port and connected to a fluid reservoir fluid; a fluid channel defining fluid communication between the input port and the output port; and a biosensor communicating with a computing device for continuously measuring biosignal data of the fluid in the fluid channel, the biosensor comprising an electrode pair.

[0033] In some embodiments, the computing device is used to determine a condition of the user based at least in part on bio-signal data.

[0034] In some embodiments, the biosensor comprises an impedance sensor for detecting the conductivity of the fluid in the fluid channel.

[0035] In some embodiments, the biosensor comprises a pH sensor for detecting the pH level of the fluid in the fluid channel.

[0036] In some embodiments, the biosensor comprises at least one of a lactate sensor, an amylase sensor, a urea sensor, or a creatinine sensor.

[0037] In some embodiments, the device further comprises a flow sensor for continuously determining the flow rate of the fluid in the fluid channel over time.

[0038] In some embodiments, the device further comprises a light-based sensor comprising a light emitter and a light receiver for detecting the transmission of light through the fluid in the fluid channel.

[0039] In some embodiments, the light-based sensor is configured to detect the color of the fluid based at least in part on the detected wavelength.

[0040] In some embodiments, the device further comprises a temperature sensor for detecting the temperature of the fluid in the fluid channel.

[0041] In some embodiments, the biosensor is disposed on a substrate in fluid communication with the fluidic channel.

[0042] In some embodiments, the electrode pairs are positioned sequentially along the length of the fluid channel.

[0043] According to another aspect, a computer-implemented method for monitoring a user is provided, the method comprising: continuously receiving biosignal data from a biosensor in communication with a fluid; determining a condition of the user based at least in part on the biosignal data; and predicting a future occurrence of a complication based at least in part on the condition of the user.

[0044] In some embodiments, the method further includes receiving a profile of the user, the user profile including information related to a surgical procedure performed on the user, wherein a future occurrence of a complication is predicted based at least in part on the user profile.

[0045] In some embodiments, the method further comprises updating a profile of the user based at least in part on the biosignal data.

[0046] In some embodiments, the method further comprises continuously receiving flow data from a flow sensor connected to a fluid flow from a body of a user; and determining a flow rate of the fluid based at least in part on the flow data, wherein the condition of the user is determined based at least in part on the flow rate.

[0047] In some embodiments, the method further comprises determining a change in flow rate of the fluid over time and a change in the biosignal data over time, and predicting a future occurrence based at least in part on the change in the flow rate and the change in the biosignal data.

[0048] In some embodiments, the traffic data is received in near real time.

[0049] In some embodiments, the biosignal data is received in near real time.

[0050] In some embodiments, the method further includes receiving light data associated with transmission of light through the fluid from a light-based sensor in fluid communication with the fluid.

[0051] In some embodiments, the method further comprises determining a color of the fluid based at least in part on the light data.

[0052] In some embodiments, the method further comprises receiving temperature data of the fluid from a temperature sensor in fluid communication with the fluid. In some embodiments, the method further comprises adjusting the biosignal data based at least in part on the temperature data.

[0053] In some embodiments, the method further comprises determining a risk factor for the user based on a cross-correlation with trends of bio-signal data of other users.

[0054] In some embodiments, the condition of the user is based at least in part on determining whether the biosignal data is within a threshold range.

[0055] According to another aspect, a system for monitoring a user is provided, the system comprising: a processor; a memory in communication with the processor, the memory storing instructions which, when executed by the processor, cause the processor to perform the method described herein.

[0056] Current SOCs rely on a combination of standardized management guidelines and the personal experience of healthcare providers. There has been active research into the use of various risk factors and novel biomarkers for early detection of postoperative complications.

[0057] Embodiments of the present invention include methods for calculating a continuous, real-time numerical score based on the pH of the drainage fluid that correlates to the patient's probability of developing a postoperative complication. These methods are novel and do not exist in current SOCs. The methods enable healthcare providers (HCPs) to obtain real-time status of their patients to monitor more closely, intervene, or discharge earlier than SOCs.

[0058] The method can integrate more than one biomarker data or other sources of data available to the HCP (eg, EMR / EHR, other sensors or medical devices) to give a more accurate probability score.

[0059] The method can be deployed in firmware or in the cloud and updated / improved as needed or as newly acquired data becomes available.

[0060] The method can be adapted to detect different types of postoperative complications.

[0061] The method can be used with pH data from any source or device, including continuous monitoring devices or point measurements from benchtop pH meters.

[0062] The methods may also be used with other biomarker or biosignal data from any source or device, including online and / or continuous monitoring devices.

[0063] According to one aspect of the present invention, a computer-implemented method for evaluating a risk value of a target patient is disclosed, the method comprising: receiving target patient data at a server via a connected device; at the server, using one or more risk assessment models to estimate one or more risk values ​​associated with the target patient data.

[0064] In one embodiment, the target patient data includes one or more of the following: historical data, patient population level data, and real-time data.

[0065] In one embodiment, the one or more risk assessment models include one or more of the following: a historical data risk assessment model, a real-time data risk assessment model, and a combined risk assessment model, wherein the combined risk assessment model includes one or more risk assessment models.

[0066] In one embodiment, the historical data risk assessment model is trained by the following steps: receiving historical data corresponding to multiple patients at a server comprising one or more processors and a memory, the one or more processors comprising one or more of a mapping engine and a standardization engine, the historical data of each of the multiple patients corresponding to an indication of whether the patient has developed complications; generating preprocessed historical data by the following steps: mapping the historical data of each of the multiple patients into a numerical value via the mapping engine; standardizing the historical data of each of the multiple patients via the standardization engine; performing regression analysis on the preprocessed data, the regression analysis being used to determine the relationship between the historical data and the risk value.

[0067] In one embodiment, the real-time data risk assessment model is trained by the following steps: receiving time-stamped data corresponding to signal data measured by multiple sensors coupled to multiple patients at a server comprising one or more processors and a memory, the one or more processors comprising one or more of a filtering and enhancement engine and a standardization engine, the time-stamped data for each of the multiple patients corresponding to an indication of whether the patient has developed a complication; generating preprocessed time-stamped data by the following steps: filtering and enhancing the time-stamped data for each of the multiple patients via the filtering and enhancement engine; standardizing the time-stamped data for each of the multiple patients via the standardization engine; performing one or more regression analyses on the preprocessed time-stamped data, the one or more regression analyses being used to determine one or more relationships between the real-time data and the risk value.

[0068] According to one embodiment of the present invention, the historical data includes one or more of the following items of the target patient or target patient group: preoperative risk factors, medical records, surgical history, personal health indicators and surgical parameters.

[0069] According to one embodiment of the present invention, the real-time data includes sensor data from one or more sensors that continuously measure signals associated with a physiological condition of the target patient.

[0070] In one embodiment, the method further comprises notifying a user of the risk value.

[0071] In one embodiment, the risk value is estimated continuously.

[0072] According to one embodiment of the present invention, the risk value includes the probability of the target patient developing postoperative complications.

[0073] According to one aspect of the present invention, a system for assessing the risk value of a target patient is disclosed, the system comprising: a server, one or more processors and a memory, the one or more processors being communicatively coupled to a database, the database comprising historical data and time-stamped data from a plurality of patients; the one or more processors being configured to receive the target patient data via a connected device; and the memory comprising instructions which, when executed by the one or more processors, configure the server to perform the following operations: receiving the target patient data at the server; preprocessing or processing the target patient data via a processor; and using one or more risk assessment models to estimate one or more risk values ​​associated with the target patient data.

[0074] According to one embodiment of the present invention, the target patient data includes one or more of the following: historical data, patient population level data, and real-time data.

[0075] According to one embodiment of the present invention, the one or more risk assessment models include one or more of the following: a historical data risk assessment model, a real-time data risk assessment model and a combined risk assessment model, and the combined risk assessment model includes one or more risk assessment models.

[0076] According to one embodiment of the present invention, the historical data risk assessment model is trained through the following steps: receiving historical data corresponding to multiple patients at a server comprising one or more processors and a memory, the one or more processors comprising one or more of a mapping engine and a standardization engine, the historical data of each of the multiple patients corresponding to an indication of whether the patient has developed complications; generating preprocessed historical data through the following steps: mapping the historical data of each of the multiple patients into a numerical value via the mapping engine; standardizing the historical data of each of the multiple patients via the standardization engine; performing regression analysis on the preprocessed data, the regression analysis being used to determine the relationship between the historical data and the risk value.

[0077] According to one embodiment of the present invention, the real-time data risk assessment model is trained by the following steps: receiving time-stamped data corresponding to signal data measured by multiple sensors coupled to multiple patients at a server comprising one or more processors and memories, the one or more processors comprising one or more of a filtering and enhancement engine and a standardization engine, the time-stamped data of each patient in the multiple patients corresponding to an indication of whether the patient has a complication; generating pre-processed time-stamped data by the following steps: filtering and enhancing the time-stamped data of each patient in the multiple patients via the filtering and enhancement engine; standardizing the time-stamped data of each patient in the multiple patients via the standardization engine; performing one or more regression analyses on the pre-processed time-stamped data, the one or more regression analyses being used to determine one or more relationships between the real-time data and a risk value. According to one embodiment of the present invention, the risk value comprises the probability of a target patient having a postoperative complication.

[0078] According to an embodiment of the present invention, the system further comprises a display system for displaying the risk value.

[0079] According to one aspect of the present invention, a non-transitory computer-readable storage medium is disclosed, which includes instructions. When executed by a computer, the instructions cause the computer to perform the following steps: receiving target patient data via a connected device; preprocessing or processing the target patient data via a processor; and using one or more risk assessment models to estimate one or more risk values ​​associated with the target patient data.

[0080] According to one embodiment of the present invention, the target patient data includes one or more of the following: historical data, patient group level data, and real-time data.

[0081] Other features will become apparent from the accompanying drawings in conjunction with the following description.

[0082] Advantages and features of the present invention include convenient early management using the techniques disclosed herein to treat complications, such as anastomotic leaks, which can significantly reduce the risks associated with such complications. The advantages and features of the present invention will be better understood by referring to the following embodiments and claims in conjunction with the accompanying drawings, in which like elements are identified by like symbols.

[0083] Brief description of the attached figure

[0084] With reference to the accompanying drawings, several embodiments of the present disclosure are provided by way of example only, in which:

[0085] Figure 1is a representative ROC curve, where the maximum Youden Index threshold is highlighted with a black dot.

[0086] Figure 2 Baseline pH thresholds for each hour are shown.

[0087] Figure 3 Mean pH values ​​are shown.

[0088] Figure 4 Continuous learning of a method according to one embodiment is shown.

[0089] Figure 5 A method for determining the probability of a postoperative complication after initialization of the method is shown.

[0090] Figure 6 The method used to assess patient risk values ​​is described.

[0091] Figure 7 A system for assessing a risk value for a patient is described.

[0092] Figure 8 The method used to train risk assessment models on historical data is described.

[0093] Fig. 9 Methods for training risk assessment models on real-time data are described.

[0094] Fig.10 The method used to estimate the patient risk value is described. DETAILED DESCRIPTION

[0095] Various embodiments and aspects of the present specification will be described with reference to the following details. The following description and accompanying drawings are illustrative of the specification and should not be construed as limiting the specification. Many specific details are described to provide a thorough understanding of the various embodiments of the present specification. However, in some cases, well-known or conventional details are not described in order to provide a concise discussion of the embodiments of the present specification.

[0096] Various equipment and processes will be described below to provide examples of embodiments of the systems disclosed herein. The embodiments described below do not limit any claimed embodiments, and any claimed embodiments may cover processes or equipment different from those described below. The claimed embodiments are not limited to equipment or processes having all the features of any of the following equipment or processes, or equipment or processes having features common to multiple or all of the following equipment or processes. The following equipment or processes may not be embodiments of any claimed subject matter.

[0097] In addition, numerous specific details are set forth to provide a thorough understanding of the embodiments shown herein. However, it will be understood by those skilled in the relevant art that the embodiments described herein may be practiced without these specific details. In other cases, well-known methods, procedures, and components are not described in detail to avoid obscuring the embodiments described herein.

[0098] In this specification, an element may be described as being "configured to" perform one or more functions or "configured for" such functions. Generally speaking, an element configured to perform or configured for performing a function is capable of performing the function, is adapted to perform the function, is adapted to perform the function, is operable to perform the function, or is otherwise capable of performing the function.

[0099] It should be understood that for the purpose of this specification, the expressions "at least one of X, Y, and Z" and "one or more of X, Y, and Z" can be interpreted as only X, only Y, only Z, or any combination of two or more items X, Y, and Z (e.g., XYZ, XY, YZ, ZZ, etc.). Similar logic can be applied to any occurrence of "at least one of..." and "one or more of..." expressions for two or more items.

[0100] In some embodiments, the sensor may include electrochemical or solid-state sensors of different forms, including but not limited to potential, voltammetry, conductivity, capacitance, current or ion-sensitive field effect transistor (ISFET). In some embodiments, the sensor may be a piezoelectric or microelectromechanical system (MEMS). Depending on the type of sensor used, the sensor may include terminals connected to an active electrode, a counter electrode, a reference electrode or a pseudo-reference electrode. The sensor may be of different types, including but not limited to pH sensors, ion-sensitive sensors, temperature sensors, lactate sensors, electrolyte sensors, impedance sensors, light-based sensors, microbial sensors, protein sensors, DNA-based sensors, carbohydrate sensors, enzyme sensors, oxygen sensors such as PO2 (oxygen partial pressure) sensors, amylase sensors, urea sensors, creatinine sensors, motion sensors (such as accelerometers), pressure sensors and flow sensors.

[0101] The sensors may be connected in series or in parallel, and may be positioned sequentially, for example, along the length of a fluid channel.

[0102] In some embodiments, the sensor may include one or more temperature sensors, such as thermistors or resistance temperature detectors.

[0103] In use, the temperature sensor may experience a change in resistance associated with a change in temperature. Thus, the temperature can be determined by determining the resistance of the thermistor by stimulating it with a current and measuring the voltage (or vice versa).

[0104] In some embodiments, the sensor may include a pH sensor that is electrochemical in nature, enabling biological analytes to be converted into electrical signals that can then be measured, monitored, and analyzed to determine if postoperative complications are occurring. An interdigitated electrode (active electrode, counter electrode, and reference electrode) system can be fabricated on a biocompatible substrate. The electrodes can be made of biocompatible materials such as gold, platinum, titanium, and silver, and then functionalized with active polyaniline (PANI), polyaniline / polyurethane (PAIN / PU), polyurethane polymers, or other suitable layers. In one example, the m-biosensor has a size of 500pm×500pm, enabling it to be placed on or embedded in a catheter to monitor changes in pH over time.

[0105] In some embodiments, the sensor may include a light-based sensor, such as a photoelectric sensor, which utilizes a light emitter or combination of a light source and a detector in the ultraviolet to infrared spectrum to measure the light absorption or transmission characteristics of the fluid. Single or multiple wavelength radiation may be used.

[0106] In some embodiments, the sensor comprises an impedance sensor, typically operated with alternating current (AC) excitation, which can be used to assess the state of the user. The impedance sensor can include an electrode pair and include excitation and readout circuitry.

[0107] Assessing postoperative risks

[0108] The systems, methods and devices disclosed herein can be used to monitor, detect and predict different forms of postoperative complications that may occur after surgery, such as leakage. Embodiments may include sensing and diagnostic devices that utilize sensors on, for example, catheters or online devices to detect or predict the presence of luminal fluid, for example, when leakage occurs.

[0109] In some embodiments, the systems, methods, and devices disclosed herein include sensors placed at locations proximate to a surgical site, such as biosensors that can be used to sense biosignal data, enabling monitoring of analytes in biological fluids that may indicate surgical leaks.

[0110] Biosignal data may be received from a prior art sensing system, such as the sensing system described in U.S. Publication No. 2022 / 0265175, which is owned by the assignee of the present application and is incorporated herein by reference in its entirety.

[0111] In some embodiments, the sensor may include an electrochemical or solid-state sensor having different forms, including but not limited to potential, voltammetry, conductivity, capacitance, current or ion-sensitive field effect transistor (ISFET). In some embodiments, the sensor may be a piezoelectric or microelectromechanical system (MEMS). Depending on the type of sensor used, the sensor may include terminals connected to an active electrode, a counter electrode, a reference electrode or a pseudo-reference electrode. The sensor may be of different types, including but not limited to pH sensors, ion-sensitive sensors, temperature sensors, lactate sensors, electrolyte sensors, impedance sensors, fluid sensors, light-based sensors, microbial sensors, protein sensors, inflammation sensors, carbohydrate sensors, enzyme sensors, oxygen sensors such as PO2 (oxygen partial pressure) sensors, amylase sensors, urea sensors, creatinine sensors, pressure sensors and flow sensors.

[0112] The sensors may be connected in series or in parallel, and may be positioned sequentially, for example, along the length of a fluid channel.

[0113] In some embodiments, the sensor may include a temperature sensor, such as a thermistor.

[0114] In use, a thermistor may experience a change in resistance that is related to a change in temperature. Thus, the resistance of the thermistor, and hence the temperature, can be determined by exciting it with a current and measuring the voltage (or vice versa).

[0115] Temperature sensors can be used to account for many artifacts and error sources in biosignal measurements. Temperature sensors can be used to compensate or adjust for temperature-dependent signals from other sensors, such as impedance and pH. A rise in fluid temperature detected by a temperature sensor can indicate the inflow of new fluid, since biological fluids tend to have a higher temperature relative to the ambient temperature.

[0116] The principles of thermal mass fluid transport can be exploited to measure fluid flow rate using a temperature sensor array and a heating element.

[0117] In some embodiments, the sensor may include a flow sensor such as a flow meter to measure, for example, the volume or mass flow rate of a fluid such as a liquid or a gas within a user's body.

[0118] In some embodiments, the sensor may include a pH sensor that is electrochemical in nature, enabling biological analytes to be converted into electrical signals that can then be measured, monitored, and analyzed to determine if postoperative complications are occurring. An interdigitated electrode (active electrode, counter electrode, and reference electrode) system can be fabricated on a biocompatible substrate. The electrodes can be made of biocompatible materials such as gold, platinum, titanium, and silver, and then functionalized with active polyaniline (PANI), polyaniline / polyurethane (PAIN / PU), polyurethane polymers, or other suitable layers. In one example, the m-biosensor has a size of 500pm×500pm, enabling it to be placed on a catheter to monitor changes in pH over time.

[0119] The pH sensor can be formed from a conductive polymer made from aniline monomers. The sensitivity of a suitable conductive polymer to pH levels can allow it to be used as a pH sensitive component in a pH sensor.

[0120] The pH sensor can be calibrated and / or controlled by a potentiostat, specifically an electronic device that controls the potential difference and current of a three-electrode system consisting of a working electrode (WE), a reference electrode (RE) and a counter electrode (CE). Such electronic instrumentation has many applications and can be used to make pH sensors, such as using cyclic voltammetry (CV), chronoamperometry and chronopotentiometry.

[0121] The pH sensor may be configured to detect pH values ​​that are within thresholds or boundaries, or that deviate from such boundaries.

[0122] In some embodiments, the sensor may include a light-based sensor, such as a photoelectric sensor, which utilizes a light emitter or combination of a light source and a detector in the ultraviolet to infrared spectrum to measure the light absorption or transmission characteristics of the fluid. Single or multiple wavelength radiation may be used.

[0123] Light-based sensors may include a combination of light emitters and detectors in the ultraviolet to infrared spectrum and are used to measure the light absorption or transmission characteristics of a fluid.

[0124] Light absorption or transmission characteristics can indicate changes in body and cavity fluids, including but not limited to protein composition and concentration, pH, conductivity, inflammatory markers, and cellular activity due to complications or disease onset. This also enables the measurement of the color of the fluid, which can indicate bleeding (red), bile leakage (yellow-green), fecal leakage (brown), gastric leakage (green), urine leakage (yellow), and other fluids of specific colors.

[0125] In some embodiments, single wavelength or multi-wavelength radiation can be used. Changes in the absorption or transmission characteristics of the fluid detected within a specific light band or wavelength make it possible to measure the color of the fluid. Since serous fluids (e.g., peritoneal fluid and pleural fluid) are usually light yellow, changes in color can indicate bleeding (red), bile leakage (yellow-green), fecal leakage (brown), gastric leakage (green), urine leakage (yellow) or other fluids of a specific color.

[0126] In some embodiments, a light-based sensor may include a combination of a light emitter and detector in the ultraviolet to infrared spectrum to measure the scattering of light by the fluid, thereby measuring the turbidity of the fluid. Serous fluids are generally clear in appearance and have low turbidity. An increase in turbidity, such as measured by an increase in light measured at right angles by a light detector, may indicate white blood cells and microorganisms in the fluid, which may be due to infection.

[0127] Light-based sensors can include multiple light sources and receivers. For example, a single broadband light source can be used in combination with multiple photodiodes of specific frequency bands (e.g., red, green, and blue). In this way, the absorption / transmission characteristics of the fluid can be measured in as many frequency bands as there are photodetectors. Similarly, multiple light sources can be used in combination with a single broadband photodetector, whereby each light source is turned on continuously and the transmitted light is measured accordingly by the photodetector. Finally, light sources and photodetectors can also utilize dynamic filters to enable the emission or detection of light of specific frequency bands instead of multiple light sources or photodetectors.

[0128] In some embodiments, the sensor comprises an impedance sensor, typically operated with alternating current (AC) excitation, which can be used to assess the state of the user. The impedance sensor can include an electrode pair and include excitation and readout circuitry.

[0129] In some embodiments, the impedance sensor can be configured to perform AC excitation within well-defined and constant fluid geometry (confined by a channel or enclosure), thereby enabling determination of normalized impedance (or specific impedance) and admittance.

[0130] In some embodiments, the impedance of the fluid may be measured over a range of frequencies (Hz to MHz) to separate the contributions of the individual electrolytes and infer the ionic composition of the fluid.The user condition may be based at least in part on the ionic composition of the fluid.

[0131] The measured impedance value can be converted to determine the conductivity of the fluid (eg, the real part of the impedance). The conductivity can reveal characteristics of the fluid itself and therefore may be of direct clinical value.

[0132] For example, conductivity can indicate the intrinsic characteristics and composition of the analyte.

[0133] Impedance can be affected by fluid volume and geometry, so the measured impedance can be used to locate and track particles and bubbles in the fluid channel.

[0134] In some embodiments, impedance sensors can be used to account for many artifacts and error sources in biosignal measurements.

[0135] In some embodiments, an impedance sensor can be used to detect rapid and sharp increases in impedance beyond the range of body fluids, which may indicate the presence of bubbles in the channel. Bubbles are a challenge for catheter-based measurements because they cause artifacts in the readings.

[0136] In some embodiments, an impedance sensor can be used to detect a sudden increase in impedance, which may indicate the presence and amount of inhomogeneous material and particles (eg, blood clots, fibrin).

[0137] In some embodiments, impedance sensors can be used to detect blood clotting (generally characterized by a sudden increase in impedance followed by a slower but sustained increase in impedance), and therefore the presence of blood and the risk of channel blockage.

[0138] In some embodiments, an array of impedance sensors placed along a channel can be used to detect and track bubbles, inhomogeneous materials, and / or particles as they pass through the channel using the techniques described herein.

[0139] In some embodiments, the sensor may include an amylase sensor.

[0140] When in use, the systems, methods and devices can monitor trends and changes in physical and chemical biomarkers, which can include but are not limited to pH, temperature, fluid flow, pressure, lactate, lactic acid, nitrate, glucose, alkali ions, oxygen, bicarbonate, inflammatory proteins, bacterial proteins and other biomarkers associated or related, for example, to leakage.

[0141] A single sensor or an array of sensors can be placed along the catheter wall, within a dedicated lumen, or in an in-line device, which enables the device to detect and monitor if a leak occurs.

[0142] In some embodiments, the catheter can be used as a carrier for sensors to monitor internal chambers of the body, such as the peritoneal cavity or the pleural cavity, without applying any negative pressure. The catheter can be connected to a balloon or mechanical pump to apply negative pressure to promote drainage of fluids. The catheter can also be connected to a fluid source, such as a saline solution, to perform therapeutic and diagnostic functions, such as dialysis or irrigation.

[0143] In some embodiments, multiple sensors may be spaced apart along the length of the catheter. Multiple sensors placed along the catheter may allow sensing of multiple areas and tracking the spatial progression of a leak.

[0144] The catheter may be formed of a tube having a hollow or solid body and made of a medical grade material, such as a suitable polymer. In some embodiments, the catheter may be a flexible substrate.

[0145] In some embodiments, the catheter may be formed from a material having low friction.

[0146] The catheter may have different designs, wherein the cross section of the catheter may be cylindrical, rectangular, flat or T-shaped, and the catheter may have a single lumen or multiple lumens.

[0147] In some embodiments, the sensor can be placed in a reservoir that can collect fluid from the user's body. The reservoir can include elements such as a balloon, a pump, or other container that can hold biological fluids. The sensor placed in the reservoir can be used simultaneously with the sensor placed in the catheter. This can allow more sensors to be used to determine a variety of different conditions or postoperative complications, such as fluid leakage, infection, inflammation, or other dangerous complications.

[0148] A sensor such as a biosensor can be connected to a monitor such as an electronic data acquisition system (DAQ), which can be located inside or outside the user's body and can continuously process data obtained from the sensor. The connection can be established via different methods, including but not limited to wires and connectors, which can be embedded in the catheter body or at least one lumen designed to allow the wires and connectors to pass through it. The connection can also be established wirelessly by transmitting the in vivo data obtained from the biosensor to a receiver placed outside the body via a transmission system.

[0149] In some embodiments, each of the plurality of sensors independently communicates with the monitor.

[0150] The monitor may have a screen so that the readings can be viewed directly on the device. The monitor may also use various visual or audio sequences, such as small LEDs or alarm sounds, to signal various events.

[0151] The data acquired by the monitors may also be communicated to a computer system via a wired or wireless medium to enable further analysis and visualization. The communicated data may be processed, raw, or aggregated.

[0152] In some embodiments, the data collected by the monitor can be analyzed to identify trends associated with the occurrence of different complications. This can be performed by evaluating single or multiple data sets acquired from one or more sensors over time to diagnose and determine the stage of occurrence of complications.

[0153] If one or more sensors exhibit biological trends associated with a surgical leak, an alarm signal may be sent from the monitor to a computer-based system, thereby enabling a user to determine appropriate medical action.

[0154] In one example, a slow drop in local pH may indicate a small amount of leakage or poor blood supply to the wound site. If a simultaneous slow increase in lactate concentration is observed, it may indicate a lack of blood supply (i.e., ischemia). If the lactate concentration is stable, it indicates a slow leak.

[0155] In another example, a sharp drop in pH may indicate a large leak. If the pH returns to its baseline, it may indicate that the wound is healing, but there is still leakage. If the pH continues to drop, or remains at a low level, it may indicate a severe leak that the body may have difficulty recovering from.

[0156] The systems and methods disclosed herein can perform monitoring, detection and diagnosis, and prediction. For example, monitoring can present data sensed by sensors such as biosensors. Through algorithms, detection and diagnosis can detect the user's condition and / or determine a diagnosis, such as a leak, the type of leak, and where the leak is coming from, for example, with an associated confidence level. Prediction can use sensor data to check for different trends and process signals to, for example, predict leaks that may occur in the future with an associated confidence level. Therefore, embodiments of the systems and methods disclosed herein can identify the physiological differences between leak occurrence and leak precursors.

[0157] The systems and methods disclosed herein can be used to perform clinical functions. In one example, the catheter system can be connected to mechanical elements that, in addition to their diagnostic functions, can also apply negative pressure to enable fluid to be drained from the user's body. Such clinical functions can all be performed at a location within the user's body, such as within the GI tract or within the peritoneal cavity.

[0158] The technology for applying negative pressure may include, but is not limited to, a balloon, a mechanical pump, a vacuum system, or other devices capable of sucking fluid, for example, from the body to the outside. In some embodiments, the drained fluid can aid in diagnostic applications by causing a constant flow of fluid through the sensor. In some embodiments, clinical functions can be performed by pumping fluid into the user's body.

[0159] As used herein, the term "body fluid" may refer to fluids originating from within the human body, fluids excreted or secreted by the body (e.g., blood, gastric fluid, and peritoneal fluid), and similar fluids. In a broad sense, the term "luminal fluid" refers to a subset of body fluids present in the lumen, intestines, blood vessels, tubular organs, and many other membrane-bound organs, such as gastric fluid, intestinal fluid, feces, urine, bile, and other similar fluids.

[0160] The terms "biomarker" and "aptamer" as used herein may refer to molecules, substances, and chemical or physical properties that can be measured or detected as biological signals in body fluids. They include, but are not limited to, pH, temperature, electrolyte concentration, fluid flow rate, pressure, lactate, lactic acid, nitrate, alkali ions, inflammatory proteins, bacterial proteins, specific cells, molecules, genes, gene products, enzymes, hormones, inflammatory proteins, and glucose.

[0161] As used herein, the terms "biosensor" and "sensor" may refer to a device or system that detects or responds to a biomarker or biological signal, thereby converting these signals into a measurable electrical signal. Biosensors and sensors as used herein may include, but are not limited to, pH sensors, lactate sensors, amylase sensors, lactate sensors, glucose sensors, temperature sensors, pressure sensors, enzyme sensors, protein sensors, biosensors, ion sensors, electrolyte sensors, impedance sensors, conductivity sensors, flow sensors, and other forms of electrochemical and solid-state sensors.

[0162] In some embodiments, signal data received from sensor devices can be aggregated, for example, over time and associated with a user to generate a profile of the user. In some embodiments, user profile information, such as signal data associated with one or more users, can be applied to machine learning techniques to develop a model for such signal data.

[0163] The user profile may include information about the user, such as information related to the surgical procedure performed on the user, as well as the date and time of the surgical procedure, the location of the surgical procedure, the date and time of insertion of the sensor device, the location of insertion of the sensor device, the user's age, height, weight, medical history, condition or disease (e.g., diabetes), medications currently or in the past used by the user, or other current or historical factors related to the user, the surgical procedure, or the device details.

[0164] In some embodiments, the user profile includes a list of medications taken by the user. This can be used to identify potential sources of error caused by medications altering the thresholds of one or more biosignals measured using the sensors described herein.

[0165] In some embodiments, the user profile contains procedures performed on the patient. Such a list of procedures can be used to further analyze the list of potential complications that the user may suffer from given the risks of each procedure. In addition, such a list of procedures can be used to identify the anatomy of biological fluids near the procedure location.

[0166] In some embodiments, the information related to the surgical procedure performed on the user includes the date and time of the surgical procedure. In addition, the surgical date and time can be used to analyze the user's condition when a complete timeline of recovery of a given user is given.

[0167] In some embodiments, user profile data can be entered by the user, a healthcare facility, or can be entered by a healthcare professional, for example, a surgeon can enter information related to the surgical procedure performed and details about the sensor devices used postoperatively (e.g., operating parameters, number and type of sensors, etc.).

[0168] In some embodiments, the user profile may be automatically generated, for example, from a health record indicating surgical details or the user's electronic health record. These records may be received from a computing device in communication with system 100.

[0169] In some embodiments, a user profile may include information identifying factors associated with a user condition determined from collected signal data.

[0170] In one example, a user condition that occurs within a brief period of time indicated by the sensor data (which indicates, for example, a brief spike) may be ignored because it does not indicate that particular condition, but rather is an anomaly.

[0171] In some embodiments, future occurrences of complications such as anastomotic leaks may be predicted based at least in part on when the user condition occurred and the length of time the user condition occurred.

[0172] A machine learning algorithm may be applied to previously acquired signal data associated with a user condition. For example, pattern recognition may be performed on previously acquired signal data associated with a particular user condition. Machine learning may produce a user condition classification model trained by previously acquired signal data.

[0173] Further description of such models will be described in more detail below.

[0174] Leaks can be predicted by analyzing changes in fluid flow around the sensor of the sensor device. For example, how quickly the flow changes occur can indicate how fast the leak is flowing.

[0175] In another example, leakage can be predicted based on the accumulation of lactate detected by the sensor.

[0176] In another example, a leak may be predicted based on oxygen consumption detected by a sensor.

[0177] In another example, leakage can be predicted based on detected pH changes and can include analyzing why the pH changed to distinguish different causes or conditions for such pH changes.

[0178] In some embodiments, the future occurrence of anastomotic leakage can be predicted based on the user's condition being above or below a predetermined threshold. Such a threshold can be, for example, a pH value.

[0179] As described above, a secondary condition or second user condition may be determined from the sensor data. Future occurrences of leakage may be predicted based at least in part on the second user condition. The second user condition may also indicate a risk factor or risk level for a leakage condition.

[0180] In some embodiments, the risk model may determine a confidence level that a leak has occurred based on a combined analysis of the user profile and the processed biosignals. Algorithm inputs / outputs may be contextualized based on the likelihood of a leak occurring using weighting factors based on the user profile and current condition. These weights may be dynamic over time and updated as the user's condition is updated. In one example, if the user has undergone bariatric surgery, a higher weight may be applied to a gastric leak detection algorithm compared to a colorectal leak detection algorithm.

[0181] In some embodiments, predictive analysis of signals such as biosignals from biosignal sensors, such as from a sensor device or an online monitoring device, may include diagnosis, such as identifying the nature of a leak or disease by examining symptoms monitored by the sensor.

[0182] In some embodiments, the classification condition or risk level of predicting future occurrences of leakage may be based on signal data, user conditions, and user profiles.The generated data may include the classification condition or risk level.

[0183] Machine learning algorithms may be applied to previously acquired signal data, user profile data, and user condition data. For example, pattern recognition may be performed on previously acquired signal data associated with a particular leakage prediction.

[0184] The data associated with the prediction of a future occurrence of a leak may include a notification of the prediction.

[0185] In some embodiments, signal data may be collected to establish trends across multiple patients, and cross-correlation techniques may be used to identify similarities between patient data and previous patients, and matches or correlations may indicate risk factors.

[0186] Advantageously, better predictions may be possible because data acquired over time has better fidelity, for example, acquired continuously as fluid flows through the sensor device, and in one example, acquired in real time or near real time.

[0187] An exemplary application of the systems and methods disclosed herein can be demonstrated by observing a patient who has colorectal cancer and needs to have the tumor removed. The surgeon may decide to perform an anastomosis after the tumor is removed from the body. If the surgeon believes that a paracolic gutter is the most likely area to collect fluid from a leak, he or she can place a catheter with a biosensor in the paracolic gutter. The catheter system they use can be equipped with fluid, pH, and lactate sensors. After the catheter is placed, the surgeon may also choose to use absorbable sutures to keep the catheter in place. The catheter can then be connected to a monitor placed outside the body. The catheter can also be connected to a balloon, which will apply negative pressure to drain fluid from the peritoneum.

[0188] The monitor will then confirm that a connection has been established with the biosensor, notifying the user and surgeon that the patient's condition looks normal. The patient can then stay in the hospital overnight and be discharged the next day. The patient can be discharged with the monitor and catheter. Three days after surgery and after the patient is discharged, the biosensor can detect clinically relevant pH changes and a simultaneous increase in lactate concentration. The monitor can then send a signal to the patient to seek medical attention, or the system can delay the signal to wait for more significant changes. The data can be transmitted wirelessly to a medical institution, where specialists can review the data and can also decide whether to have the patient come to a medical institution. The system can then detect significant pH changes and relatively high fluid flow in the abdominal cavity. The monitor can then alert the patient, notifying them that a leak has been detected and that they need to seek medical intervention immediately. The medical institution can also be notified. After the patient is in the hospital, the medical institution can view the data obtained from the biosensor and make clinical decisions to help the patient before complications worsen. The surgeon can also decide to take corrective medical actions, including but not limited to reoperating on the patient. The system can be used again after the corrective action is completed.

[0189] Conveniently, the sensor device described herein can be removed from the user or patient by a nurse in an outpatient setting, such as at the user's home, and does not require the user to undergo additional surgical procedures. In one example, this may occur ten to twenty days after surgery.

[0190] In some embodiments, integrating the sensor with the catheter (and particularly the drainage tube) enables the sensor device to be removed in an outpatient setting.

[0191] The method calculates a continuous, real-time numerical score based on the pH of the drainage fluid that correlates with the patient's probability of developing a postoperative complication such as a leak.

[0192] In clinical trials, multiple sets of pH measurements are obtained from patients (i.e., from patient biofluids) either continuously or at discrete time intervals. The data for each patient will be labeled "leakage" (AL) or "no leakage" (nAL). Continuous data can be subdivided into time intervals of any length. For this example, the intervals are 1 hour intervals. The determination of the baseline threshold is performed using a receiver operating characteristic curve (ROC) analysis based on the postoperative hourly pH from the clinical data.

[0193] Figure 1 ROC analysis of the data is illustrated. Essentially, each hour of data is examined and a baseline pH value for that hourly interval is determined via ROC analysis, thereby enabling differentiation of AL and nAL patient data. The baseline is based on the maximum area under the curve (AUC) of the ROC curve, or the highest possible sum of sensitivity and specificity. In the representative ROC curve, the maximum Youden Index threshold is highlighted with a black dot.

[0194] Figure 2 The baseline cutoff point for each interval is specified. This step is repeated until all available data have been processed. The result will be a list of baseline cutoff points for each interval (in Figure 2 ). Each point represents a pH threshold that, when applied to the data for that hourly interval after surgery, would discriminate between patients with AL and nAL, for example, by maximizing the sum of sensitivity and specificity in the data or by some other method that can estimate the maximum true positive rate with the minimum false positive rate.

[0195] exist Figure 3 In the Figures, mean pH values ​​are presented as a function of postoperative hours.

[0196] Figure 4 Embodiments of methods are described in which a continuous, real-time numerical score is calculated based on the pH of the drainage fluid that is correlated with the patient's probability of developing a postoperative complication.

[0197] The method enables healthcare providers (HCPs) to obtain real-time status of their patients, allowing for closer monitoring, intervention, or earlier discharge than is possible with the SOC.

[0198] The method includes continuous learning, such as Figure 4 The method is based on clinical trial data sets and Figure 2 The baseline threshold (OT) and baseline threshold cumulative average (OTCA) were initially established based on a dataset 414 from a clinical trial based on biomarker statistics for patients with complications (e.g., leaks) versus patients without complications.

[0199] The method is based on data collected in memory 412 from patients using this embodiment (see Figure 2 ) and whether they develop complications to continue learning.

[0200] The method measures one or more biomarker (pH is used as an example to describe the embodiment, but the method is not limited to pH) data points at regular intervals (eg, every 5 minutes). The data points are stored in memory 412.

[0201] The method calculates a baseline threshold (OT) 404 for pH data points using a receiver operating characteristic curve (ROC) analysis.

[0202] The method determines the percentage of data points in each interval that are below the baseline threshold calculation value 406. A cumulative average (CA) of the percentages below the baseline threshold calculation value is calculated 408 for each interval.

[0203] A baseline threshold for cumulative mean calculation (OTCA) was calculated 410 using receiver operating characteristic curve (ROC) analysis.

[0204] Figure 5 A method of determining a post-operative complication probability 502 after the method is initialized is depicted.

[0205] The method measures one or more biomarkers at regular intervals (eg, every 5 minutes) during a current period of time post-operatively.

[0206] The method calculates the percentage of data points below OT in the current time interval after surgery and calculates CA 506 for the percentage of data points below OT.

[0207] If CA exceeds OTCA 508, then the current post-operative hour interval is identified as "high risk" 512. Otherwise, it is marked as "low risk" 510.

[0208] If the number of consecutive post-operative hours marked as "high risk" exceeds a predetermined limit (e.g., 10), the method notifies the HCP by sending a message, generating a noise on the device, etc. that the patient is at high risk for suffering or developing a post-operative complication 516.

[0209] If a patient develops a complication, the accumulated dataset for this patient will be marked as a complication to improve the accuracy of the CA and OTCA thresholds.

[0210] The postoperative time period can be 30 minutes, 1 hour, half a day, etc.

[0211] Instead of CA, an IIR (Infinite Impulse Response) or FIR (Finite Impulse Response) time filter may be used to calculate the OTCA, such as a moving average filter, a Gaussian window filter, or an exponential filter.

[0212] Other non-linear statistical methods such as the median or mode may be used to determine the OT.

[0213] Other preoperative, intraoperative, and postoperative data (eg, American Society of Anesthesiologists (ASA) scores, demographic information, surgical details, etc.) can also be integrated into the method.

[0214] The method can combine more than one biomarker data or other source data available to the HCP (e.g., EMR / EHR, heart rate, other sensors or medical devices) to give a more accurate probability score. Different weights can be applied to different biomarkers to adjust their impact on the output risk / prediction. Other metadata of the patient can be integrated into the method. Statistical confidence can be added to the measured probability.

[0215] The method can be deployed in firmware and executed on a processor or cloud server that allows for updates and improvements. The method can be adapted to detect different types of postoperative complications. The method can be used with pH data from any source.

[0216] Generating a model for assessing postoperative risk

[0217] The disclosed embodiments may include a computer-implemented method for predicting a patient's postoperative risk value based on analyzing continuous patient data and patient historical data with reference to an independent risk model including a plurality of patient real-time data and risk values ​​and a plurality of patient historical data and risk values.

[0218] In some embodiments, the computer-implemented method may be used in conjunction with the target patient's pre-existing medical records, electronic medical records (EMR), pre-operative patient data, and the like.

[0219] As in the previous section, biosignal data can be received from a prior art sensing system, such as the sensing system described in U.S. Publication No. 2022 / 0265175, which is owned by the assignee of the present application and is incorporated herein by reference in its entirety.

[0220] Generally, for purposes of this disclosure, a target patient is one who is being monitored for complications and / or risks associated with a pre-operative, post-operative, or intra-operative condition.

[0221] In addition, multiple patients can provide time-stamped data and historical data to train the various risk models disclosed herein. Such data can be obtained from clinical trials, medical records, etc.

[0222] In some embodiments, the computer-implemented method can be used in conjunction with an online device that can continuously measure real-time data of a target patient or multiple patients through sensors, and the real-time data is sent via a network to a server containing one or more risk assessment models to provide a risk value for the target patient, or to train one or more risk assessment models to assess risk based on real-time data.

[0223] In-line devices may include devices attached to, embedded within, or placed along a catheter.

[0224] In some embodiments, the computer-implemented methods may be used in conjunction with implantable medical devices, desktop devices, point-of-care (POC), and handheld devices.

[0225] In some embodiments, the sensor may include electrochemical or solid-state sensors of different forms, including but not limited to potential, voltammetry, conductivity, capacitance, current or ion-sensitive field effect transistor (ISFET). In some embodiments, the sensor may be a piezoelectric or microelectromechanical system (MEMS). Depending on the type of sensor used, the sensor may include terminals connected to an active electrode, a counter electrode, a reference electrode or a pseudo-reference electrode. The sensor may be of different types, including but not limited to pH sensors, ion-sensitive sensors, temperature sensors, lactate sensors, electrolyte sensors, impedance sensors, light-based sensors, microbial sensors, protein sensors, DNA-based sensors, carbohydrate sensors, enzyme sensors, oxygen sensors such as PO2 (oxygen partial pressure) sensors, amylase sensors, urea sensors, creatinine sensors, motion sensors (such as accelerometers), pressure sensors and flow sensors.

[0226] The sensors may be connected in series or in parallel, and may be positioned sequentially, for example, along the length of a fluid channel.

[0227] In some embodiments, the sensor may include one or more temperature sensors, such as a thermistor or a resistance temperature detector.

[0228] In use, the temperature sensor may experience a change in resistance associated with a change in temperature. Thus, the temperature can be determined by determining the resistance of the thermistor by stimulating it with a current and measuring the voltage (or vice versa).

[0229] In some embodiments, the sensor may include a pH sensor that is electrochemical in nature, enabling biological analytes to be converted into electrical signals that can then be measured, monitored, and analyzed to determine if postoperative complications are occurring. An interdigitated electrode (active electrode, counter electrode, and reference electrode) system can be fabricated on a biocompatible substrate. The electrodes can be made of biocompatible materials such as gold, platinum, titanium, and silver, and then functionalized with active polyaniline (PANI), polyaniline / polyurethane (PAIN / PU), polyurethane polymers, or other suitable layers. In one example, the m-biosensor has a size of 500pm×500pm, enabling it to be placed on or embedded in a catheter to monitor changes in pH over time.

[0230] In some embodiments, the sensor may include a light-based sensor, such as a photoelectric sensor, which utilizes a light emitter or combination of a light source and a detector in the ultraviolet to infrared spectrum to measure the light absorption or transmission characteristics of the fluid. Single or multiple wavelength radiation may be used.

[0231] In some embodiments, the sensor comprises an impedance sensor, typically operated with alternating current (AC) excitation, which can be used to assess the state of the user. The impedance sensor can include an electrode pair and include excitation and readout circuitry.

[0232] Advantageously, according to various embodiments, the systems and methods described below act as a universal interface that enables the use of data generated by different types of sources. Thus, the systems and methods can be configured to work with existing systems that measure patient biosignals.

[0233] Figure 6 A computer-implemented method 600 for assessing a patient's risk value is illustrated.

[0234] In general, a computer-implemented method 600 for assessing a risk value of a target patient 620 includes first receiving one or more of patient real-time data 602 and patient historical data 606 from a target patient 620 at a server 612 via a connection device such as a network 618. The method also includes estimating a risk value 616 of a measurement signal of the target patient at the server 612 via a combined risk assessment model 610, the combined risk assessment model 610 including one or more real-time data risk assessment models 604 and one or more historical data risk assessment models 608. The method also includes displaying or notifying a user of the risk value 616 via a display system 614.

[0235] The display system may include a computer, a smart phone, a physical printout, etc., as well as any other means of displaying data known in the art. The display system may be connected to the server via a network, a cable, or the like.

[0236] Notifying the user may include notification sounds, visual notifications, or the like generated by tactile, visual, auditory, or digital notification systems.

[0237] Alternatively, the risk value may be retained without being displayed or notified to the user 616 .

[0238] Risk assessment models 604 and 608 may preferably be trained separately.

[0239] Training the risk assessment models 604, 608 typically includes sending data to a server, associating known or measured data with a numerical value or indication of whether a complication has occurred (e.g., by calibration, mapping, etc.), standardizing the data (e.g., by normalization, or subtracting the mean and dividing by the standard deviation), and performing one or more regression analyses on the standardized data to obtain a model that associates data points with risk factors.

[0240] For example, the real-time data risk assessment model 604 can be trained using continuous sensor data from a plurality of sensors that measure biological signals from a first plurality of patients. The continuous sensor data can be sent, for example, from an online device 622 that can continuously monitor a patient 620, sending the patient real-time data 602 to a server 612 via a connection device, such as a network 618, a cable, a USB device, or the like.

[0241] Alternatively, continuous sensor data may be sent from an implantable device, a desktop device, a POC device, a handheld device, and / or a medical wearable device (ie, a smart watch).

[0242] The training data may include associating sensor data with postoperative time such that the time-tagged data may be associated with an indication of whether a patient among a plurality of patients suffered a complication such that a regression curve may be generated associating the sensor data with the probability of suffering a complication at a given postoperative time.

[0243] For example, the time-stamped data used to train the real-time risk assessment model 604 may include pH and postoperative time for each of a plurality of patients. For example, the time-stamped data may be classified based on whether it corresponds to a patient who developed a complication. For example, the time-stamped data may be classified based on when a complication occurred in a given patient. The time-stamped data may be classified by a combination of the above methods and any other methods known in the art.

[0244] The historical data risk assessment model 608 may be trained using historical data from a second plurality of patients, including preoperative risk factors. The second plurality of patients may overlap with patients from the first plurality of patients.

[0245] Historical Data Risk assessment models may additionally or alternatively be trained using historical data at the patient or population level.

[0246] The historical data may include, but is not limited to, one or more of the following for the target patient or target patient population: preoperative risk factors (eg, demographic information, weight, gender, age, etc.), medical records, surgical history, etc.

[0247] For example, the historical data risk assessment model 608 may be trained with preoperative data of multiple patients, such as their ASA scores.

[0248] Training the combined risk assessment model 610 may include mapping decision rules associated with the two individual models 604, 608 to risk values ​​616. It may also include decision making rules based on data availability within the two individual models 604, 608.

[0249] For example, the combined risk assessment model 610 may include a decision making engine that selects to use one or both of the risk assessment models 604, 608 before calculating the risk value 616. Some patients may not have continuous patient real-time data 602, in which case only patient historical data 606, including preoperative data or medical records 624, may be input into the method 600. In this case, the decision making engine may choose to assess risk based only on the historical data risk assessment model 608.

[0250] This enables partially complete patient data to be used in risk assessment model training and validation (eg, when information about a patient's preoperative risk factors is available but their sensor data is not, and vice versa).

[0251] This is an alternative to more common approaches such as imputing missing values ​​or removing patients with missing values.

[0252] Removing patients with missing values ​​limits the available number of patients for training, and if the number of imputed values ​​is greater than the number of available values, the imputation may be problematic and may be non-trivial for the complete set of sensor measurements.

[0253] Furthermore, models 604, 608, and 610 may be continuously updated by training with target patient data.

[0254] Typically, but not necessarily, risk assessment models 604, 608, and 610 are trained using machine learning algorithms or techniques.

[0255] The machine learning algorithm or technique may include, for example, a deep learning architecture, such as a deep belief network (DBN), a stacked autoencoder (SAE), a convolutional neural network (CNN), or a recurrent neural network (RNN). Other examples include, but are not limited to: a TTRBM model or similar methods using a restricted Boltzmann machine (RBM), a social restricted Boltzmann machine (SRBM), a fuzzy restricted Boltzmann machine (FRBM), a deep belief network (DBN); AE, FAE, GAE, DAE, BAE models using a statistically adjusted terminal use (SAE) model; the following models may be used, such as AlexNet, ResNet, Inception, VGG16, ECNN models of CNN; bidirectional recurrent neural networks (BiRNN), long short-term memory (LSTM) networks, and gated recurrent units (GRU) of RNNs may also be used. Other techniques specific to time series modeling may be used, including, but not limited to, dynamic time warping, change point detection, and autoregressive integrated moving average (ARIMA) models.

[0256] In some embodiments, other types of algorithms may also be relied upon, such as physics-based mathematical calculations and basic multivariate linear regression models, which are used in conjunction with or complement these architectures and learning algorithms. They may also include the cumulative average (CA) method described above, such as Figure 4 and Figure 5 exemplified.

[0257] In one non-limiting example, the method 600 can estimate a risk value 616 for post-operative anastomotic leakage.

[0258] The combined risk assessment model 610 uses the real-time data risk assessment model 604 and the historical data risk assessment model 608 to estimate a continuous real-time risk value 616 for a patient to develop anastomotic leakage, typically after a surgical anastomosis. In this embodiment, the historical data risk assessment model 608 contains a "preoperative risk factor model" and the real-time data risk assessment model 604 contains a pH model. This can preferably enable users, including healthcare providers, to obtain the real-time status of their patients for closer monitoring, intervention, or earlier discharge than the current standard of care.

[0259] For the target patient 620, its historical data 702 is sent to the server 612 via the network 618 and preprocessed. In this case, the historical data 702 includes preoperative risk factors, such as ASA scores of 1, 2, 3, 4 or 5, and surgical methods, such as laparoscopy and robotic methods, laparotomy. Preprocessing can include determining preoperative risk factors (such as ASA scores) based on the patient's medical records. Preprocessing can include mapping or specifying preoperative risk factors as numerical values. Historical data can also include the patient's personal health indicators, such as gender, height, weight or past surgical operations, past surgical complications, surgical parameters (including but not limited to duration, blood loss, etc.) and vital signs data (heart rate, respiratory rate, oxygen saturation, etc.).

[0260] The pre-processed historical data may then be input into a historical data risk assessment model 608, which uses the normalized values ​​calculated for the data used to train the model 608. The historical data risk assessment model 608 then preferably estimates a risk value 616 based on the preoperative risk factors to estimate the probability of anastomotic leak risk using a regression model that relates the probability of leak to the given preoperative risk factors.

[0261] For the same patient, the patient real-time data 602 may be pre-processed at a first time point or a first time period, in which case the real-time data is a pH sensor measurement obtained from a pH sensor device that may be in fluid communication with the patient 620. For example, the pH sensor device may continuously measure the pH of the patient's body fluid. Pre-processing the pH sensor measurement may include taking an average of the pH measurement over a time period. This may be achieved by filtering and enhancing the raw data using a filtering and enhancement engine.

[0262] The pre-processed patient real-time data 602 may then be input into a real-time data risk assessment model 604, which in this case is trained using pH sensor data.

[0263] The real-time data risk assessment model 604 can estimate the risk value based on the pH value at a first time point or a first time period, use the pH average value calculated via one or more trained pH models to generate a probability estimate of the risk of anastomotic leakage at a given time, and use one or more regression models that associate leakage with a given pH value at a certain time / time period after surgery to generate a probability estimate of the risk of anastomotic leakage.

[0264] Alternatively or additionally, it is possible to Figure 4-5 The CA method is used to estimate the risk value.

[0265] For the same patient, at desired time intervals (eg, every given hour after surgery), the combined risk assessment model 610 may preferably use the two model risk values ​​to classify the patient's anastomotic leakage risk based on a set of rules corresponding to the two risk models.

[0266] In one non-limiting example, a patient's risk of anastomotic leakage can be based on the following decision rules: "very high" if the probability estimate of the preoperative risk factor model is above 0.5 and the probability estimate of the pH model is above 0.75; "high" if the probability estimate of the preoperative risk factor model is above 0.5 and the probability estimate of the pH model is between 0.5 and 0.75; "low" if 1) the probability estimate of the preoperative risk factor model is above 0.5 and the probability estimate of the pH model is below 0.5, or 2) the probability estimate of the preoperative risk factor model is below 0.5 and the probability estimate of the pH model is above 0.5; "very low" if the probability estimate of the preoperative risk factor model is below 0.5 and the probability estimate of the pH model is below 0.5.

[0267] Figure 7 A 700 system for assessing a patient's risk value is described.

[0268] Typically, the system includes a server having one or more processors and a memory, wherein the one or more processors are communicatively coupled to a database 706 via a network, cable, or the like, the database including historical data 702 from a plurality of patients and time-stamped data 716 from a plurality of sensors measuring biological signals from a plurality of patients. The one or more processors are also communicatively coupled to target patient real-time data 712 and target patient historical data 714 via one or more networks, cables, or the like.

[0269] The memory includes instructions that, when executed by one or more processors, configure the server 704 to:

[0270] 1) Receive one or more of target patient real-time data 712 and target patient historical data 714 at the server 704. For the target patient real-time data 712, the server may be configured to continuously receive the data 712 in real time. Alternatively, the server may be configured to continuously receive 712 at specific time intervals before, after, or during a surgical operation.

[0271] 2) Pre-processing or processing one or more of data 712 and 714 via a processor.

[0272] 3) Using the first model, estimate a first risk value 616 associated with the target patient 710. For example, the first model may include a historical data risk assessment model 608, which may be trained using historical data 702 from a database 706. The historical data risk assessment model 608 may give a risk value 616 based on preoperative risk factors.

[0273] 4) Estimate a second risk value associated with the target patient 710 using a second model. For example, the second model may include a real-time data risk assessment model 604, which may be trained using time-stamped data 716 from a database 706. The real-time data risk assessment model 604 may provide a risk value 616 based on real-time sensor measurements corresponding to the physiological condition of the target patient 710 at a given time. The second model may estimate the risk value 616 continuously, such as at specific time intervals after surgery.

[0274] 5) Estimate a final risk value 616 associated with the target patient 710 using the combined risk assessment model 610, and thereby determine a risk assessment 708 based on the final risk value 616. The risk assessment 708 may be determined based on a set of rules corresponding to the values ​​616 estimated in steps 3 and 4. The risk assessment 708 is determined once, continuously in real time, or continuously at specific time intervals.

[0275] The system may also include a display system 614, such as a smart phone or computer, that is communicatively coupled to the server. The one or more processors may be configured to send one or more of the first, second, or third risk values ​​616 and / or the risk assessment 708 to the display system to notify the user of the risk value. The system 700 may alternatively or further include an auditory notification system such that a noise associated with the risk value 616 or the risk assessment 708 notifies the user of the risk value 616.

[0276] The risk assessment 708 may be in the form of "high", "low", "medium", for example to inform the risk of postoperative leakage. It may be in the form of a percentage. It may be sent to the patient, their healthcare provider, or other designated user.

[0277] Risk assessment 708 may include error and / or confidence intervals resulting from statistics associated with the various models and the quantity and quality of data in database 706 .

[0278] The risk assessment 708 may be used to inform a post-operative treatment plan for the target patient 710 and / or automatically update the post-operative treatment plan.

[0279] The combined risk assessment model 610 can be trained based on one or more risk assessment models. The historical data risk assessment model 608 can be trained based on various types of historical data, including but not limited to preoperative risk factors such as ASA scores, surgical procedures, patient medical records (i.e., EMRs), preoperative physical conditions, etc. The real-time data risk assessment model 604 can be trained based on various types of time-stamped data 716, including biosensor data, etc.

[0280] Figure 8 A schematic diagram showing a method of training a historical data risk assessment model 800 using historical data 702 is illustrated.

[0281] The method includes sending historical data for a plurality of patients and an indication of whether the patient suffers from a complication associated with the historical data to a server 612, preprocessing the historical data 702 by mapping the historical data into numerical values ​​with a mapping engine 802, normalizing the data with a normalization engine 804, and inputting the mapped and normalized data into a historical data risk assessment model 608. The model 608 includes one or more regression models that are trained, for a plurality of patients, by associating the preprocessed historical data of a given patient with an indication of whether the given patient suffers from a complication.

[0282] In the example of anastomotic leakage risk assessment, training the historical data risk assessment model 608 may include:

[0283] Preprocessing of historical data via mapping engine 802 and normalization engine 804:

[0284] 1) Collect historical data 702 from multiple patients, in which case, for each patient in the available training data, the historical data 702 includes preoperative risk factors such as ASA score and surgical approach (laparotomy, laparoscopy, robotic approach), as well as an indication of whether the patient associated with the historical data 702 suffered complications.

[0285] 2) Mapping the ASA scores into numerical values ​​(e.g., ASA I: 1, ASA II: 2, ASA III: 3, ASA IV: 4, ASA V: 5) and mapping the surgical approaches into numerical values ​​(e.g., laparoscopy and robotic approaches: 0, laparotomy: 1) via a mapping engine 802, and standardizing the set of training values ​​to a mean of 0 and a standard deviation of 1 via a standardization engine 804.

[0286] Training historical data risk assessment model 608:

[0287] 3) Train one or more regression models with numerical values ​​and associated indications of complications. For example, a logistic regression model (using L2 penalty and balanced class weights) can be trained to predict a risk value 616 in the form of the probability that a patient has anastomotic leak given a numerical surgical approach and ASA score.

[0288] The above is an example of using preoperative data to train the historical data risk assessment model 608 to give a preoperative risk factor historical data risk value 806 .

[0289] The historical data risk assessment model 608 can also be trained using the patient's personal health indicators, such as gender, height, weight or past surgical procedures, past surgical complications, surgical parameters (including but not limited to duration, blood loss, etc.) and vital signs data (heart rate, respiratory rate, oxygen saturation, etc.).

[0290] Fig. 9 A schematic diagram showing a method for training a real-time data risk assessment model 604 using time-tagged data 716 is illustrated.

[0291] The method includes sending time-stamped data 902 for a plurality of patients to a server, typically by receiving signals from sensors coupled to the patients, and an indication of whether the patients have complications associated with the time-stamped data 716; preprocessing the time-stamped data 716, which may include different techniques: filtering and enhancing the signals received from the sensors via a filtering and enhancement engine 904, depending on the nature of the time-stamped data 716; normalizing the data with a normalization engine 906; and inputting the filtered, enhanced, and normalized data into a real-time data risk assessment model 604. Filtering the data into time-based averages may include reducing the sensor measurements to larger / smaller averages (e.g., every 30 minutes) or aggregating using statistics such as median / mean.

[0292] Model 604 includes one or more regression models that are trained for multiple patients by associating time-stamped data of a given patient with an indication of whether the given patient has a complication. When a target patient is evaluated in real time, the postoperative time and the real-time sensor measurements can be input into the real-time data risk assessment model 604 to estimate a real-time data risk value 908 of whether the target patient is likely to have a complication at a given postoperative time, which can be displayed on the display system 614.

[0293] In the example of anastomotic leakage risk assessment, training the real-time data risk assessment model 604 may include:

[0294] The pH data is pre-processed via filtering and enhancement engine 904:

[0295] 1) Use pH and EC measurements (these measurements come from continuous measurements of the online device 622 in communication with the patient fluid and outside expected boundaries, as well as any measurements from the calibration fluid) to remove erroneous sensor measurements.

[0296] 2) For patients who experienced anastomotic leak, all pH measurements after their diagnosis were removed (this was only performed for the patients used to train the model).

[0297] 3) Reduce the filtered pH signal to a 15 minute average. This time frame may be smaller or larger depending on the desired data resolution.

[0298] The pH data is normalized via normalization engine 804:

[0299] 4) For a given hour post-surgery, calculate the average of the pH sensor readings for the 24 hours prior to that hour post-surgery and normalize the average calculated from the training set to have a mean of 0 and a standard deviation of 1 (if predictions are to be made for new patients, use the normalized values ​​calculated from the training data).

[0300] Training real-time data risk assessment model 604:

[0301] 5) Using the average of the standardized pH sensor readings for each patient in the available training data, train one or more regression models, such as a logistic regression model (using L2 penalty and balanced class weights) to predict a risk value 616 in the form of the probability that the patient has anastomotic leakage given the average pH value for that hour postoperatively. In some embodiments, multiple models may be trained for specific time intervals, such as for each hour postoperatively for which a risk value prediction is made.

[0302] The above is an example of using continuous sensor measurements related to the patient's real-time condition to train the real-time data risk assessment model 604 to continuously estimate the post-operative risk factor risk value 616.

[0303] The real-time data risk assessment model 604 can also be trained using data from different sensors that measure in real time or in a point measurement manner (e.g., blood gas analyzers, etc.), quantitative and qualitative data obtained by healthcare professionals (e.g., heart rate, heart rate variability, blood pressure, SpO2, temperature, ECG, etc.), color monitoring of patient fluids (blood, urine, bile, etc.), or any other relevant physiological data related to the patient's condition.

[0304] Fig.10 is an exemplary process flow diagram of a method 1000 for estimating a risk value for a patient.

[0305] Method 1000 starts in step 1002 .

[0306] In step 1004, the system acquires historical data from the target patient. In step 1012, the system begins to continuously acquire real-time data from the patient. Typically, steps 1004 and 1012 are performed simultaneously, but they can be performed sequentially, or in the case of a patient for whom one form of data is not available, only one of these steps can be performed.

[0307] In step 1006, the historical information is preprocessed and processed using the historical data risk assessment model 608. This step generally involves estimating a risk value from the processed historical data using the risk model 608. In step 1014, the real-time data is continuously preprocessed and processed using the real-time data risk assessment model 604. This step generally involves estimating a risk value from the real-time data using the risk model 604. Typically, steps 1006 and 1014 are performed simultaneously, but they may be performed sequentially, or in the case of patients for whom one form of data is not available, only one of these steps may be performed.

[0308] In step 1008, the processed historical data or estimated risk value from step 1006 is sent to the combined risk assessment model. Simultaneously or separately, in step 1016, the processed real-time data or estimated risk value from step 1014 is sent to the combined risk assessment model.

[0309] In step 1010 , the combined risk assessment model estimates a risk assessment for the patient using the risk values ​​sent from the individual models in steps 1008 and 1016 .

[0310] In step 1018, the method ends.

[0311] Method 1000 may also include continuously assessing risk, ie, continuously repeating steps 512 - 1010 using continuously generated real-time data.

[0312] It may also include notifying or displaying the risk assessment in the form of a visual, audible or tactile notification.

[0313] In some embodiments, different machine learning algorithms or techniques may be used alone or in combination in the above-mentioned different engines.

[0314] The machine learning algorithm or technique may include, for example, a deep learning architecture, such as a deep belief network (DBN), a stacked autoencoder (SAE), a convolutional neural network (CNN), or a recurrent neural network (RNN). Other examples include, but are not limited to: a TTRBM model or similar method using a restricted Boltzmann machine (RBM), a social restricted Boltzmann machine (SRBM), a fuzzy restricted Boltzmann machine (FRBM), a deep belief network (DBN); AE, FAE, GAE, DAE, BAE models using a statistically adjusted terminal use (SAE) model; the following models may be used, such as AlexNet, ResNet, Inception, VGG16, ECNN models of CNN; bidirectional recurrent neural networks (BiRNN), long short-term memory (LSTM) networks, and gated recurrent units (GRU) of RNNs may also be used. Other techniques specific to time series modeling may be used, including, but not limited to, dynamic time warping, change point detection, and ARIMA models.

[0315] In some embodiments, other types of algorithms may also be relied upon, such as physics-based mathematical calculations and basic multivariate linear regression models, which are used in conjunction with or in complement to these architectures and learning algorithms.

[0316] Many of the functional units described in this specification are labeled "engines" to more particularly emphasize their implementation independence. For example, an engine can be implemented as a hardware circuit containing custom VLSI circuits or gate arrays, off-the-shelf semiconductors such as logic chips, transistors, or other discrete components. An engine can also be implemented in a programmable hardware device such as a field programmable gate array, programmable array logic, a programmable logic device, etc.

[0317] The engine may also be implemented in software to be executed by various types of processors. The executable code of the recognition engine may, for example, contain one or more physical or logical blocks of computer instructions, which may, for example, be organized as objects, procedures, or functions. However, the executable code of the recognition engine need not be physically located in the same location, but may contain different instructions stored in different locations, which, when logically combined together, constitute the engine and achieve the stated purpose of the module.

[0318] In fact, the engine of executable code can be a single instruction or multiple instructions, and can even be distributed on several different code segments, different programs, and several storage devices. Similarly, operating data can be identified and described in the engine here, and can be embodied in any suitable form and organized in any suitable type of data structure. The operating data can be collected as a single data set, or can be distributed in different locations, including different storage devices, and can exist at least partially as electronic signals on a system or network. In the case where the engine or part of the engine is implemented in software, the software portion is stored on one or more computer-readable storage media.

[0319] It should be understood that the models described herein can be trained using a variety of data sources.

[0320] Other models may be integrated with the preoperative risk factors and pH models (eg, vital signs models), used in conjunction with, or as an alternative to the above models.

[0321] More complex individual models can be developed to be used in conjunction with or as an alternative to the above models, for example, a preoperative risk factor model that also considers blood loss during surgery, or a preoperative risk factor model that predicts probabilities using a random forest model.

[0322] Filtering the data into time-based averages may include reducing the sensor measurements into larger / smaller averages (eg, every 30 minutes) or aggregating using statistics such as median / mean.

[0323] Alternative approaches that integrate predictions from individual models may be used (e.g., into a logistic regression model instead of using thresholds for classification).

[0324] The systems and methods described herein may also include features such as prompts to request input of additional data at key moments to update the model. For example, prompting for HR at 6 hours post-operatively, which can be used to update the risk assessment.

[0325] The combined risk assessment model or algorithm can be deployed in firmware or in the cloud and updated / improved as needed. It can be further tuned to detect different types of postoperative complications.

[0326] The combined risk assessment model may be used with an online device to continuously receive data in real time from a biosensor or sensors in fluid communication with the patient, or adapted to work with other sensor systems, or adapted to receive manual input of data.

[0327] The present disclosure includes systems with processors to provide various functions to process information and determine results based on inputs. Typically, the processing can be implemented with a combination of hardware and software elements. Hardware aspects may include a combination of operably coupled hardware components, including a microprocessor, a logic circuit, a communication / network port, a digital filter, a memory, or a logic circuit. The processor may be adjusted to perform operations specified by a computer executable code, which may be stored on a computer readable medium.

[0328] The steps of the method described herein can be implemented via a suitable programmable processing device or an onboard field programmable gate array (FPGA) or a digital signal processor (DSP) that executes software or stored instructions. Generally, the physical processor and / or machine used for any processing or evaluation in the embodiments of the present disclosure may include one or more networked or non-networked general-purpose computer systems, microprocessors, field programmable gate arrays (FPGAs), digital signal processors (DSPs), microcontrollers, etc., which are programmed according to the teachings of the above-mentioned exemplary embodiments and understood by the technicians in the fields of computers and software. As understood by the technicians in the field of software, programmers of ordinary technology can easily manufacture appropriate software based on the teachings of the exemplary embodiments. In addition, as understood by the technicians in the field of electricity, the devices and subsystems of the exemplary embodiments can be implemented by manufacturing application-specific integrated circuits. Therefore, the exemplary embodiments are not limited to any specific combination of hardware circuits and / or software.

[0329] Exemplary embodiments of the present invention may include software stored on any one or a combination of computer-readable media to control the devices and subsystems of the exemplary embodiments, process data and signals, and enable the devices and subsystems of the exemplary embodiments to interact with human users, etc. Such software may include, but is not limited to, device drivers, firmware, operating systems, development tools, application software, etc. Such computer-readable media may also include computer program products of embodiments of the present invention to perform all or part of the processing performed in the implementation scheme (if the processing is distributed). The computer code means of exemplary embodiments of the present invention may include any suitable interpretable or executable code device, including, but not limited to, scripts, interpretable programs, dynamic link libraries (DLLs), complete executable programs, etc.

[0330] Common forms of computer readable media may include, for example, disks, flash memory, RAM, PROM, EPROM, FLASH-EPROM, or any other suitable storage chip or medium that can be read by a computer or processor.

[0331] While particular embodiments and applications of the present disclosure have been illustrated and described, it should be understood that the disclosure is not limited to the precise construction and composition disclosed herein and that various modifications, changes, and variations may be apparent from the foregoing description without departing from the spirit and scope of the disclosure.

Claims

1. A computer-implemented method for assessing a risk value of a target patient, the method comprising: receiving, at a server via a connection device, target patient data; One or more risk assessment models are used at the server to estimate one or more risk values ​​associated with the target patient data. 2 . The computer-implemented method of claim 1 , wherein the target patient data comprises one or more of: historical data, patient population level data, and real-time data.

3. A computer-implemented method according to claim 2, wherein the one or more risk assessment models include one or more of the following: a historical data risk assessment model, a real-time data risk assessment model, and a combined risk assessment model, and the combined risk assessment model includes one or more risk assessment models.

4. The computer-implemented method of claim 3, wherein the historical data risk assessment model is trained by: receiving, at a server comprising one or more processors and a memory, historical data corresponding to a plurality of patients, the one or more processors comprising one or more of a mapping engine and a normalization engine, the historical data for each patient in the plurality of patients corresponding to an indication of whether the patient experienced a complication; The preprocessed historical data is generated by the following steps: mapping, via the mapping engine, the historical data for each of the plurality of patients into a numerical value; normalizing, via the normalization engine, historical data for each of the plurality of patients; A regression analysis is performed on the preprocessed data, the regression analysis being used to determine the relationship between the historical data and the risk value.

5. The computer-implemented method of claim 3, wherein the real-time data risk assessment model is trained by: receiving, at a server comprising one or more processors and a memory, time-stamped data corresponding to signal data measured by a plurality of sensors coupled to a plurality of patients, the one or more processors comprising one or more of a filtering and enhancement engine and a normalization engine, the time-stamped data for each patient in the plurality of patients corresponding to an indication of whether the patient experienced a complication; The pre-processed time-stamped data is generated by the following steps: filtering and enhancing, via the filtering and enhancing engine, the time-stamped data for each of the plurality of patients; normalizing, via the normalization engine, the time-stamped data for each patient in the plurality of patients; One or more regression analyses are performed on the pre-processed time-stamped data, the one or more regression analyses being used to determine one or more relationships between the real-time data and the risk value.

6. The computer-implemented method of claim 2, wherein the historical data comprises one or more of the following for a target patient or a target patient population: preoperative risk factors, medical records, surgical history, personal health indicators, and surgical parameters.

7. The computer-implemented method of claim 2, wherein the real-time data comprises sensor data from one or more sensors that continuously measure signals associated with a physiological condition of the target patient.

8. The computer-implemented method of claim 1, further comprising notifying a user of the risk value.

9. The computer-implemented method of claim 7, wherein the risk value is estimated continuously.

10. The computer-implemented method of claim 1, wherein the risk value comprises a probability of the target patient developing a postoperative complication.

11. A system for assessing a risk value of a target patient, the system comprising: a server, one or more processors and memory, the one or more processors being communicatively coupled to a database containing historical and time-stamped data from a plurality of patients; The one or more processors are configured to receive target patient data via a connection device; and, The memory includes instructions that, when executed by the one or more processors, configure the server to: receiving the target patient data at the server; preprocessing or processing the target patient data via a processor; One or more risk assessment models are used to estimate one or more risk values ​​associated with the target patient data.

12. The system of claim 11, wherein the target patient data comprises one or more of: historical data, patient population level data, and real-time data.

13. The system according to claim 11, wherein the one or more risk assessment models include one or more of the following: a historical data risk assessment model, a real-time data risk assessment model, and a combined risk assessment model, wherein the combined risk assessment model includes one or more risk assessment models.

14. The system according to claim 13, wherein the historical data risk assessment model is trained by the following steps: receiving, at a server comprising one or more processors and a memory, historical data corresponding to a plurality of patients, the one or more processors comprising one or more of a mapping engine and a normalization engine, the historical data for each patient in the plurality of patients corresponding to an indication of whether the patient experienced a complication; The preprocessed historical data is generated by the following steps: mapping, via the mapping engine, the historical data for each of the plurality of patients into a numerical value; normalizing, via the normalization engine, historical data for each of the plurality of patients; A regression analysis is performed on the preprocessed data, the regression analysis being used to determine the relationship between the historical data and the risk value.

15. The system according to claim 13, wherein the real-time data risk assessment model is trained by the following steps: receiving, at a server comprising one or more processors and a memory, time-stamped data corresponding to signal data measured by a plurality of sensors coupled to a plurality of patients, the one or more processors comprising one or more of a filtering and enhancement engine and a normalization engine, the time-stamped data for each patient in the plurality of patients corresponding to an indication of whether the patient experienced a complication; The pre-processed time-stamped data is generated by the following steps: filtering and enhancing, via the filtering and enhancing engine, the time-stamped data for each of the plurality of patients; normalizing, via the normalization engine, the time-stamped data for each patient in the plurality of patients; One or more regression analyses are performed on the pre-processed time-stamped data, the one or more regression analyses being used to determine one or more relationships between the real-time data and the risk value.

16. The system of claim 11, wherein the risk value comprises a probability of the target patient developing a postoperative complication.

17. The system of claim 11, further comprising a display system for displaying the risk value.

18. A non-transitory computer-readable storage medium comprising instructions which, when executed by a computer, cause the computer to perform the following steps: receiving target patient data via a connected device; preprocessing or processing the target patient data via a processor; One or more risk assessment models are used to estimate one or more risk values ​​associated with the target patient data.

19. The non-transitory computer-readable storage medium of claim 18, wherein the target patient data comprises one or more of: historical data, patient population level data, and real-time data.

Images