Method and apparatus for detecting condition from physiological data

Abstract

A computerized system for measuring and / or detecting human responses or conditions based on data from wearable sensors worn in a natural free living environment. Based on the measurement and / or detection, various actions may be taken. Physiological data is acquired and instructions are transmitted to the vaccine manufacturer to alter the composition and / or dose of the vaccine.

Description

[0001] Cross - Reference to Related Applications

[0002] This application claims the priority of U.S. Provisional Application 63 / 392,218, filed on July 26, 2022, which is incorporated herein by reference in its entirety. Technical Field

[0003] The present invention generally relates to detecting human inflammatory processes and taking actions based on such detections. Background Art

[0004] Inflammatory responses occur in the human body and can be considered biological responses of the human immune system, which are characterized by the activation of signaling pathways. Inflammatory responses can be triggered by a variety of factors, including pathogens, damaged cells, toxic compounds, vaccines, and other immune system activators. In some cases, inflammatory responses can be considered due to adverse conditions (e.g., due to diseases), but in other cases, inflammatory responses can be considered positive (e.g., a positive result of receiving a vaccine). Brief Description of the Drawings

[0005] Figure 1 A block diagram of a system for detecting human inflammatory processes and taking actions based on such detections, including various embodiments according to the present invention;

[0006] Figure 2 A flowchart of a method for detecting human inflammatory processes and taking actions based on such detections, including various embodiments according to the present invention;

[0007] Figure 3 A flowchart of a general method for creating inflammatory markers using residuals, including various embodiments according to the present invention;

[0008] Figure 4 A flowchart of a method for detecting human inflammatory processes and taking actions based on such detections, including various embodiments according to the present invention; and

[0009] Figure 5 Is an illustration of a safety monitoring system of the present invention for remotely monitoring a patient in a home environment.

[0010] The elements in the figures are shown for simplicity and clarity and are not necessarily drawn to scale. For example, the dimensions and / or relative positioning of some of the elements in the figures may be exaggerated relative to other elements to assist in improving the understanding of the various embodiments of the present teachings. Additionally, common but well-known elements that are useful or necessary in commercially viable embodiments are often not depicted so as to render the understanding of these different embodiments of the present teachings clearer. Certain acts and / or steps may be described or depicted in a particular order of occurrence, and those skilled in the art will understand that such specificity as to order is not actually required. Detailed Description

[0011] The methods described herein provide a computerized system for measuring and / or detecting adverse conditions or side effects (e.g., inflammatory responses) of a human based on data from a wearable sensor worn in a natural free-living environment. Based on the measurement and / or detection, various actions can be taken. "Inflammation" or "inflammatory response" means a biological response of the immune system characterized by the activation of signaling pathways that can be triggered by a variety of factors, including pathogens, damaged cells, toxic compounds, vaccines, and other immune system activators. It should also be understood that these methods can be used to detect positive effects on a person, such as the efficacy of a vaccine. In other words, the methods provided herein can be used to determine inflammation-related physiological changes in a person, whether the change is characterized as positive or negative.

[0012] Measuring a person's side effects or conditions such as an inflammatory response is useful for a variety of different reasons. To name a few examples, measuring an inflammatory response is useful when safety monitoring (e.g., in the case of clinical care or a clinical trial); a patient or participant can be sent home after treatment is administered; and / or side effects manifested through immune system activation must be identified and prevented. Measuring an inflammatory response can also be used to detect the reactogenicity of a vaccine in order to record its effectiveness or level of action (e.g., immunogenicity) and its safety in a representative population. Additionally, measuring an inflammatory response is useful when determining whether a medical intervention has the desired response or when determining whether a patient with an insufficient response should have their treatment escalated or an additional booster round administered.

[0013] Measuring adverse or positive side effects or conditions such as an inflammatory response can also be used to monitor patients in a home environment who are vulnerable to infection (such as sepsis or an infection at the site of a recent surgery). Measuring an inflammatory response can also be used to differentiate the course of an inflammatory condition (such as an autoimmune disease like rheumatoid arthritis) and an individual's response to therapeutic intervention.

[0014] In the methods provided herein, data is obtained from a human subject using sensors. In some aspects, the sensor data from the sensors includes continuous vital sign data such as heart rate, respiratory rate, core temperature, skin temperature, and activity and their statistical derivatives, which are measured from a wearable sensor during activities of daily living. The sensor data is collected and sent to a computerized platform, such as a platform in the cloud, where the data is synthesized into an analysis for observing a patient's inflammatory response.

[0015] The methods provided herein generate a personalized (also referred to as "individualized") baseline model of serial vital sign behavior, typically based on vital signs collected from an individual over a day to several days. "Serial vital sign behavior" refers to the interactive or cross-correlated nature between multiple vital signs over time and in response to efforts through activities of daily living. By using the baseline model, the natural variations between individuals in a population can be controlled, which otherwise may interfere with the detection of the inflammatory response signal. The baseline model represents the way in which the serial changes in an individual's vital signs occur due to the unconstrained activities of daily life. In one embodiment, the model is an autoassociative model or an autoencoder, which is restricted to identifying the coupled patterns of serial vital sign changes in an individual and estimating the expected vital sign behavior, as described in more detail herein.

[0016] In many of these embodiments, physiological data is collected from at least one wearable sensor worn by a patient during a pre-treatment interval. An individualized estimator model is created based on the physiological data collected from the patient prior to treatment. The model is capable of estimating the expected behavior of a physiological variable in response to receiving new physiological data from at least one wearable sensor worn by the patient.

[0017] During a post-treatment interval, additional physiological data is collected from at least one wearable sensor worn by the patient. The individualized estimator model is used to generate an estimate of the expected value of the post-treatment physiological data. The post-treatment physiological data is compared with its estimate, and it is determined when a predefined effect pattern exists, at least in part based on this comparison. When the predefined effect pattern exists, an action is determined and executed. The action is one or more of the following: triggering an electronic questionnaire to be presented to the patient; providing the patient with instructions to take a measurement; providing the patient with instructions to contact the patient's clinician; triggering a ticket in a call center system to queue a call to the patient; creating a prompt in an application on the patient's phone to contact the clinician; providing the patient with instructions related to classifying the effect; transmitting a control signal to control a medical device associated with treating the patient; transmitting instructions to a vaccine manufacturer to change the composition and / or dosage of a vaccine; and transmitting instructions to a treating clinician to change the composition and / or dosage of a treatment. Other examples of actions are possible.

[0018] In some examples, the effect is an adverse side effect. In other examples, the effect is positive, such as an indication of the potency of a vaccine.

[0019] In other examples, the physiological data includes heart rate data, respiratory rate data, core temperature data, skin temperature data, and activity data. Other examples are possible.

[0020] In other aspects, data collected from a patient in a free-living physiological state is used to train an individualized estimator model, where the patient's inflammatory state is stable and not expected to change in the free-living physiological state. In other aspects, the comparison is performed by determining the residuals between the estimated values and the physiological data. In some aspects, the residuals are synthesized into a single score, which is a scalar index.

[0021] In other aspects, the individualized estimator model includes a neural network or a neural network system. Other examples of models and modeling methods are possible.

[0022] In other embodiments of these embodiments, a system for monitoring a patient's drug treatment effect includes at least one wearable sensor and a control circuit. The at least one wearable sensor is worn by the patient during a pre-treatment interval and is configured to collect physiological data from the patient.

[0023] The control circuit is coupled to the at least one wearable sensor and is configured to create an individualized estimator model based on the physiological data collected from the patient before treatment, the model being capable of estimating a physiological variable in response to receiving new physiological data from the at least one wearable sensor worn by the patient. The at least one wearable sensor collects additional physiological data from the at least one wearable sensor worn by the patient during a post-treatment interval.

[0024] The control circuit is further configured to: generate an estimated value of the post-treatment physiological data using the individualized estimator model; compare the post-treatment physiological data measured by the sensor with its estimated value, and determine when a predefined effect pattern exists at least in part based on the comparison; and when the predefined effect pattern exists, determine and perform an action. The action is one or more of the following: trigger an electronic questionnaire to be presented to the patient; provide the patient with instructions to take a measurement; provide the patient with instructions to contact the patient's clinician; trigger a ticket in a call center system to queue a call to the patient; create a prompt in an application on the patient's phone to contact the clinician; provide the patient with instructions related to classifying the effect; transmit a control signal to control a medical device associated with treating the patient; and transmit instructions to a vaccine manufacturer to change the composition and / or dosage of a vaccine. Other examples of actions are possible.

[0025] In other aspects, baseline models are trained using data collected from an individual when in a free-living physiological state, in which the individual's inflammatory state is stable and not expected to change. For safety monitoring during clinical trials or clinical care, training data is collected prior to the administration of a drug or other therapeutic intervention that involves an inflammatory response. To detect vaccine action, training data is collected prior to receiving the vaccine. For infection detection, training data is preferably collected prior to surgery or at least immediately after discharge, when an infection is not yet likely to have progressed to a systemic-level response. In the case of an autoimmune disease, training data is collected prior to the initiation of a new therapy that can reduce systemic inflammation.

[0026] In some aspects, at least one diurnal cycle is used to train or personalize the model, and preferably data over several days. After learning or personalizing from the initial data, the model switches from being used in a learning (or training) mode to being utilized in a monitoring mode and can then generate estimates of vital signs in response to receiving new data from sensors. The estimated values of the vital signs are then compared to the actual measured values of the vital signs to produce a residual, which is the estimated value subtracted from the measured value. The residual can be negative or positive, depending on whether the estimated value is higher or lower than the measured value. Additionally, the residuals can be synthesized into a single score, a scalar index of the change in behavioral vital signs compared to the behavior during baseline modeling. In some aspects, such a score can be a number between 0 and 1, where 0 indicates that the vital sign behavior is similar or identical to what was initially learned, and 1 indicates that the behavior of the vital signs has become significantly different from that during training.

[0027] In other aspects, the methods provided herein rely on patterns or signatures of residuals as examples of an inflammatory response. Several patterns can be tested simultaneously, alternately, to find different ways in which an inflammatory response might manifest itself in the signature. At the same time, all signatures must meet other conditions. Conversely, signatures can be defined that indicate a reduction in the inflammatory response. This is applied when the individual was inflamed during the baseline period. In other aspects, the methods provided herein identify the persistence of any one signature or group of signatures within a time window characteristic of an inflammatory response. If the single instantaneous value of a residual or any pattern or signature does not persist, it may not represent a true physiological response.

[0028] The analysis described in this method can be aggregated from many individuals to help inform drug or vaccine development, or can be presented to clinicians, patients, and / or other clinical decision-makers for taking action on a patient's condition. Such action can be to contact the patient who may be away from the clinical environment to determine if the patient can confirm symptoms. In other examples, the action can be to automatically send a survey to the patient via a smartphone provided as part of the system for the patient to answer questions that confirm or deny symptoms. In other examples, the action can be to perform or have the patient perform other measurements with manually used instruments to provide more data to confirm or disconfirm suspicion of inflammation. In other examples, the action is to administer or have the patient self-administer an intervention that remedies the source of inflammation, whether an infectious agent or an autoimmune response, such as appropriately taking antibiotics or taking drugs against the immune response. In other examples, the action can be to control a call center ticket queuing system to create a new ticket, where the call center is staffed with trained clinicians who can contact the patient. Given the difficulty in finding time for patients and clinicians to talk, another example includes activating or displaying a button on an application on the patient's smartphone that enables the patient to automatically initiate a call to a clinician at the call center, who is trained to handle the patient's condition, and the patient can click the button when in a situation where discussion is permitted.

[0029] A wearable sensor can include any device capable of recording signals representative of heart rate, temperature, activity, respiratory rate, and other behavioral and physiological parameters. Statistical derivatives of these values, such as heart rate variability, can also be used to generate a baseline model. An exemplary sensor includes a chest-worn adhesive patch that is capable of collecting continuous electrocardiogram (ECG) signals, acceleration, and temperature.

[0030] Generally, monitoring signs of acute immune system activation in at-risk populations has clinical significance and value. Additionally, in some individuals, detecting evidence of chronic inflammation counteracting a therapeutic intervention designed to reduce inflammation is valuable. In some aspects, the present method is directed to such situations where it is important to detect changes in systemic inflammation that can be life-threatening (e.g., cytokine release syndrome or infection) or can be used to optimize the efficacy and safety of therapies (e.g., vaccines or biomedical products), which can occur at any time, systemically, and at various severities, and for which continuous monitoring is necessary or beneficial. In this sense, "acute" means over a period of time measured in hours to weeks, as opposed to a long-term inflammatory process measured over months to years.

[0031] Taking one of these examples, immunotherapy is a cancer treatment method by which a patient's own immune system is recruited to attack cancer. An example of immunotherapy is CAR T-cell therapy, which involves engineering a person's own T-cells to have a special receptor called a chimeric antigen receptor (CAR). Such receptors help T-cells recognize and attack cancer cells.

[0032] More specifically, in the first step, the patient donates blood to obtain T-cells. In the second step, in the laboratory, the T-cells are genetically engineered to encode a tumor-specific CAR. In the third step, the CAR T-cells multiply. In the fourth step, tests are conducted to ensure the safety and purity of the manufactured CAR T-cells, which are then cryopreserved. In the fifth step, the CAR T-cells are injected into the patient. In the sixth step, the CAR T-cells search for cancer cells, bind to them, and destroy them.

[0033] A side effect of cancer immunotherapy is cytokine release syndrome. Cytokines are small proteins involved in messaging to coordinate immune responses in the body. However, in cytokine release syndrome, a large amount of cytokines are released, which can be harmful to organs and, in severe cases, cause death. Cytokine release syndrome (CRS) can develop within a time frame of approximately 3 to 14 days after the administration of immunotherapy. Due to the potential risks, the therapy is administered in a clinical setting and the patient remains in the hospital. This makes the treatment very expensive. Advantageously, the method provided herein allows the patient to be sent home with a monitoring system that detects early signs of CRS, such that if CRS is detected, the clinician can take immediate action. This reduces the cost compared to previous systems.

[0034] At different stages of the acute phase, the typical known symptoms of CRS include fever, decreased blood pressure, and decreased SpO2. An increase in heart rate may occur in the early stage; heart rate variability and respiration may also change.

[0035] Using the method provided herein, and before the administration of CAR T-cells (the fifth step above), the patient is provided with a wearable sensor kit to wear the sensor at home for approximately 1 - 4 weeks, 24 hours a day. This can also occur before the first step above, and the timing can depend on other steps taken to treat the cancer or prepare for treatment.

[0036] Data during this time period prior to step 5 above is used to train the personalized baseline model. The patient is then taken to the clinic for treatment. CAR T-cells are infused into the patient. The patient is sent home with a wearable sensor and may also be sent home with other manual measurement devices such as a blood pressure cuff. Data from the patient is monitored and evaluated according to the methods provided herein. The system scans the input data for one or more markers of CRS or general inflammatory response, as described in more detail below.

[0037] If a test is performed, the test is provided to the clinician via one of several notification methods, including sending an alert on a web-enabled clinical portal. Push alerts can be sent to other medical monitoring systems. Push alerts can also be sent as a text message or to the clinician's email address. A push notification to the patient's smartphone can be provided, which indicates to contact their clinical care provider; or by clicking on a newly generated button in the application to automatically connect to the clinician call center. Push notifications can be sent to the patient via the smartphone asking the patient to answer a survey question about their status, which answer is then pushed or sent to the clinician. Push notifications can be sent to the patient on the smartphone asking the patient to manually take their temperature or manually take their SpO2 or manually take their blood pressure and enter it into the smartphone application (app), which entry is then pushed or sent to the clinician.

[0038] Another potential use of the methods provided herein relates to vaccination. Vaccination activates the innate immune system, triggering the synthesis of inflammatory cytokines that are crucial for initiating antigen-specific adaptive immune responses. The physical manifestation of this inflammation is called reactogenicity, which has historically been tracked only through symptom surveys. Limited studies directly measuring inflammatory blood biomarkers have not only found significant inter-individual differences in this inflammatory response, but have also found strong correlations between this response and both systemic symptoms and humoral immune responses. Due to the lack of any scalable method to measure an individual's response to a vaccine, for most people, the ultimate measure of the adequacy of their vaccine-induced immune protection is whether they experience a breakthrough infection and its severity.

[0039] Furthermore, objective evidence of an individual's inflammatory response to a vaccine could help design safer and better-tolerated vaccines. In an analysis of adverse events reported in placebo-controlled COVID-19 vaccine trials, the limitations of the current gold standard for safety tracking (subjective surveys) were highlighted, which analysis found that more than 50% of the reported systemic adverse events could be attributed to "nocebo" responses.

[0040] In vaccine-related use cases, the methods provided herein include the monitoring system described below. In some aspects, an individual will initiate monitoring before receiving a vaccine and continue for 5 - 10 days afterwards.

[0041] In clinical trials for vaccine development, the clinical trials can incorporate wearable sensors to better assess the variability of the inflammatory responses to various vaccine doses across different age groups in a larger population, thus informing the optimal dosing strategies for optimizing efficacy and safety. The variability of each study participant in the study cohort can be evaluated by monitoring before vaccine administration to establish a multivariate baseline in the context of daily activities. The vaccine will be administered. During the period across the expected responses, continuous monitoring will be conducted. The degree of the individual's inflammatory response will be evaluated by the following methods and the baseline will be compared with the post-vaccination period. Statistical analysis can be performed on the study results.

[0042] To identify the unique responses of individuals after vaccination, many known and unknown factors can affect a person's response to a vaccine. Known factors include age, gender, and comorbidities, especially conditions / treatments that affect the immune system. Individuals who do not exhibit the expected inflammatory response may undergo further testing (e.g., blood levels of vaccine-induced antibodies) or have a booster dose scheduled. For some individuals who do not experience any symptoms after vaccination, the appropriate changes detected by the monitoring system can provide reassurance.

[0043] Another example of a situation where this method can be employed involves the early detection of sepsis or other manifestations of infection. The primary function of the immune system is to protect the body from viral and bacterial pathogens. This response causes inflammation, which, in severe cases, manifests as fever. However, fever (e.g., temperature > 38°C) is a lagging indicator of infection as it is based on population metrics and is only a single measure of the inflammation-induced changes.

[0044] For many individuals, such as immunocompromised individuals, the earliest possible detection of infection can be life-saving. For others, early detection can prevent re-hospitalization and significant morbidity. For individuals at high risk, such as after surgery, especially after transplantation, proactive monitoring of the earliest signs of immune system activation and inflammation may potentially trigger early testing of blood cultures and / or more intensive monitoring in a healthcare facility.

[0045] In this use case, when it is known that the patient does not have sepsis, a baseline of multivariate vital sign behavior is established from monitoring data of unconstrained activities from daily life. Subsequently, monitoring is performed using the same sensors within a time window of concern for sepsis risk, and then a baseline model is used to provide a comparison with the monitoring data such that when compared with the multivariate baseline described below, markers of early signs of inflammation accompanying sepsis are revealed. When an alert is received, clinical staff can then intervene with appropriate steps to prevent the full development of sepsis.

[0046] Aspects of the operation and implementation of the method are now described. The inflammatory process affects measurable vital signs in a marked way. For example, an increase in temperature, an increase in heart rate (HR), an increase in respiratory rate (RR), and a decrease in heart rate variability (HRV) can be the result of inflammation.

[0047] While long window averages can be used for post hoc detection of inflammation, it is advantageous to detect the inflammatory process on the time scale (i.e., hours) on which the inflammatory process can unfold emergently. Over this time period, long window averages are not very useful. The challenge is to detect relative changes in one or more of the above aspects (a) for an individual and (b) against the backdrop of normal variations caused by free-living activities and behavior.

[0048] The methods provided herein establish a personalized multivariate dynamic baseline for detecting relative anomalies from the normal dynamic behavior of these vital signs. These methods then measure whether the residuals are positive or negative and in this way provide evidence of the above markers. For example, it can be determined whether the measured HR or RR is higher than its assumed value; whether the temperature is higher than its assumed value; or whether the HRV is lower than its expected value.

[0049] In some aspects, the priority of these changes is determined and observed. In some examples, HRV is the most important factor and this change is considered in some aspects to be the earliest evidence specific to inflammation. The rise in body temperature lags slightly, the respiratory rate is under a certain degree of autonomic control and is more difficult to measure reliably, and the heart rate can be abnormally high and is less specific to inflammation (i.e., other causes can result in this). Accordingly, the physiology of these four factors and particularly HRV is utilized. While evidence can be provided through all of these vital signs, some of these methods focus on at least detecting the relative decline in HRV estimated by a multivariate model.

[0050] The vital signs used by the method include signs characterizing the performance of the cardiorespiratory system. As described below, several vital signs can be used.

[0051] Heart rate (HR) can be used. HR is typically expressed as a count of heartbeats per minute, but can be determined on a beat-by-beat basis by measuring the time interval between QRS peaks of an ECG and then converted to an average HR over a time window (such as one minute) to yield a minute rate. The minute average heart rate can be determined as a trimmed mean of the heartbeat intervals within a minute, which in turn provides the number of heartbeats per minute. Just to give other examples, this can also be done via a PPG waveform or ballistocardiogram.

[0052] Respiratory rate (RR) can also be used. RR can be measured from multiple sources, including movement from the chest wall measured by an accelerometer in an adhesive patch; via respiratory sinus arrhythmia, manifested as a slight slowing and speeding up of the heart rate with each breath; or by measuring the envelope of the ECG QRS peaks, which varies with changes in thoracic gas volume.

[0053] Total activity (ACT) can also be used. Total ACT can be measured as the standard deviation of the vector magnitude of the accelerometer bias. However, other quantifications of movement are available, such as "activity counts", which is a method of quantifying vibration sensor movement that has been used for a long time.

[0054] Heart rate variability (HRV) can also be utilized. HRV can be measured in multiple ways using the heart rate within a time window. HRV can involve the standard deviation of "normal" R-R intervals (SDNN); the standard deviation of all R-R intervals (SDRR), the standard deviation of the average NN intervals for each 5-minute epoch within a 24-hour window (SDANN); the root mean square of successive differences in R-R intervals (RMSSD); and / or spectral HRV bands; or Poincaré HRV dispersion parameters.

[0055] Skin temperature (TEMP) can also be used, which is typically measured by a skin-facing thermistor that is part of the sensor device. Additionally, core body temperature (TEMPc) can be used. TEMPc can be predicted based on two thermistors on the sensor, one facing the skin and the other pointing to the ambient air outside the sensor. Then, the core temperature is a function of the heat flux through one side of the sensor with two thermistors and out the other side.

[0056] ECG waveform variations can be used. For example, the QT interval is measured as the time difference between the start or peak of the QRS complex of the ECG and the end of the T wave on the ECG, or can be measured by proxy, such as the time interval to the peak of the T wave.

[0057] Since contemporaneous "snapshots" facilitate the multivariate modeling of the relationships between vital signs, it is preferable to provide window statistics of the vital signs (e.g., average heart rate, maximum activity, 95th percentile temperature) above a common window assigned the same timestamp. In other words, each input for downstream modeling will include a vector that includes data for each vital sign at the time point represented by the vector. An example is a vector of the one-minute averages of each of HR, HRV, RR, ACT, TEMPc, and QT.

[0058] As described elsewhere in this document, the method uses a personalized estimator model. In some aspects, the personalized estimator model is any multivariate model that provides an estimate of the expected value of one or more vital sign values measured from a patient as output. By way of example, such models can be generated by decision tree / random forest function approximators; similarity-based models; and neural networks.

[0059] In some aspects, training data is required that is obtained from a patient prior to monitoring for an adverse event (e.g., CRS or sepsis). Once training is complete, in response to the input of new measurement data from the patient, the personalized model is used to generate an estimate. The input includes new readings of multivariate vital sign data from a wearable sensor. The output from the model includes an estimate of the input, which is then used to generate a residual, i.e., the measured value minus the estimate, as described below.

[0060] Similarity-based modeling (SBM) is an autoassociative or autoencoding pattern reconstruction technique. At any given time point, a set of readings from all variables can be considered an input pattern (analogous to pixels in an image). The autoassociative estimation process is then based on the reconstruction of the input pattern from the learned patterns used to generate the SBM model. SBM essentially attempts to reproduce the input pattern based on the information stored in the training data. Reproduction can occur only when a linear combination of the training data that matches the input pattern can be determined. This is possible when the input pattern represents the multivariate behavior in the training data. If the input pattern does not represent the training data behavior, one or more of the estimated pattern elements will not closely match the corresponding input elements. In effect, the estimate reflects what SBM believes each pattern element should be based on the information present in the model training data and the input pattern. The residual pattern (the difference between the input pattern and the estimated pattern) highlights where there are deviations from the training data within the pattern. These differences tend to be subtle and can therefore accumulate over time, driving the decision-making process.

[0061] The mathematics behind the SBM method focuses on the application of a "similarity operation" on pairs of observation vectors and the manipulation of a "state" matrix D that contains a set of historical training vectors (input patterns). The number of columns in D is equal to the number of representative training vectors (M), and the number of rows is equal to the number of data sources included in each vector (L). The set of measurements to be made at a given time n j is defined as the training vector x(n j )

[0062] x(n j ) = [x 1 (n j ) x 2 (n j ) x 3 (n j ) … x L (n j )] T (1)

[0063] where x i (n j ) is the measurement from data source i at time n j , then the state matrix D is given by:

[0064] D = [x(n 1 ) x(n 2 ) x(n 3 ) … x(n M )]. (2)

[0065] The result of the similarity operation on two observation vectors is a similarity score (scalar). The similarity operation is non-linear, but can be extended to matrix operations where a scalar similarity score is provided for each combination of two vectors stored in two matrices of appropriate dimensions. Thus, given an input vector (or pattern) x in that contains a single reading from each of the L data sources, the vector x est of the corresponding estimated data source values is determined from (3) to (5). Using a linear combination of the training vectors in D, the estimate of each variable in the input forms a "reconstructed" input pattern.

[0066] x est = D · w (3)

[0067] Here, w is a set of weighting factors derived from the following equation.

[0068]

[0069] The similarity operation is denoted by the symbol Indications are given to avoid confusion with the Kronecker product. Generally, the vectors stored in D are selected in such a way that they attempt to evenly span the dynamic range of the monitored process. However, in many applications, including human health monitoring, this may not always be feasible without creating a very large D matrix because the number of feasible operating regimes is very large. In practice, a large D matrix generally results in an SBM model that overfits the input data. When the model overfits the monitored data, the ability of the SBM to perform early anomaly detection is weakened. In this case, the estimation follows or tracks the progress abnormally rather than deviating from it over time. One way to address this problem is to generate several SBM models, each covering a subset of all operating regimes. Another way is to modify the SBM algorithm so that it is restricted to the current operating regime at any given time point.

[0070] Provide localization. The localized SBM method relies on dynamically generating the D matrix to characterize only the local behavior of the system at a given point in time. The idea is to select a subset of data from a much larger superset matrix H that characterizes the entire dynamic range of the monitored system to define the state matrix D(t) that is most relevant to the current input pattern. Then, as shown in (6) to (8), the estimation is generated based only on these selected, currently relevant vectors. This process is repeated for each new input vector. The weighting factor thus becomes,

[0071] D(t) = {H|F(H,x in (t))} (6)

[0072]

[0073] where F(·,·) in (6) is the relevant vector selection process for the given input vector x in (t) and the reference data matrix H. Finally, the residual vector at time t is given by (10).

[0074] r(t) = x in (t) - x est (t) (8)

[0075] A random forest method can also be used. The "loop" configuration can act like an autoencoder / autoassociative model, where for M variables, we have M RFs, each RF producing a single output and together representing M estimates of the M inputs. Using data collected from pre-treatment data of patients, each RF is trained to be personalized for the patient. Neural networks can be used to implement the personalized models provided herein. When trained with sufficient examples, neural networks can be very effective in non-linear pattern recognition, non-linear encoding of key features from the original input patterns, and inference function approximators. In practice, the challenge with the use cases envisioned herein is that using instrumentation for patients with sensors during the pre-treatment period may not generate sufficient raw vital sign data to effectively train a neural network from scratch to make personalized estimates of a patient's vital signs. Thus, in one method according to the present invention, a first neural network is developed as an "encoding" neural network that takes as input a time series of a first slice of vital sign data from one or more days from any person and outputs a multi-element encoding vector to represent the personalized cardiorespiratory behavior of that person. Second, a set of one or more estimator neural networks is developed to take as input the encoding vector and a second slice (non-overlapping with the first slice) of vital sign data from the person in order to estimate the expected values of the second slice of vital sign data. The set includes individual neural networks, each neural network estimating one vital sign given the input of all other vital signs: for example, one member of the set estimates heart rate given the input of activity, HRV, respiration, temperature, and the encoding vector, while a second member of the set estimates HRV given the input of activity, heart rate, respiration rate, temperature, and the encoding vector, and so on. Thus, each set member makes an inferential estimate of a vital sign not present in its input.

[0076] Before being deployed as a personalized estimator, these networks are pre-trained using a large amount of vital sign data from at least dozens or hundreds of individuals. Each training sample includes a first slice and a second slice of vital sign data from a unique person in the training set. The first slice of data is input into an encoding neural network, which produces an encoded vector. The encoded vector plus the second slice of data is input into a set of estimator networks, which correspondingly generate each of their respective vital sign estimates as outputs, and these vital sign estimates do not appear as inputs, that is, they are inferred estimates. These inferred outputs are compared with the actual known vital signs from the second slice to determine the estimation error, which is backpropagated in a cost function to train both the estimator set and the encoding network. This is done on a large number of first and second slices of matched data, such that the encoding network becomes effective at generating an encoded vector that accurately encodes the cardiorespiratory behavior of anyone whose data is input into the encoding network; and given the input of the encoded vector representing a person and the second slice of data from the person, the estimator networks become effective at making accurate inferred estimates of their respective inferred vital signs in the second slice of data from the person. The input to the encoding network can be constructed, for example, as one minute of vital sign data for several days; the input to the estimator network can be constructed as a combination of the encoded vector produced by the first network and a time series vital sign data window that allows for the timely monitoring of the inflammatory response. For example, the window can be a 3-hour window; or a 24-hour sliding window with a 3-hour slide. However, for clarity, the pre-training of these networks does not involve any person who is experiencing an inflammatory response. Instead, the data involved in the pre-training is such that the first slice and the second slice of data from a person can be expected to represent unaltered physiological functions. In other words, given the encoding of the first slice, these networks are being trained to accurately estimate what the second slice of data should be.

[0077] Once pre-trained, these neural networks become effective in making personalized estimates as follows. Pre-treatment data from a patient is input into the encoding network to generate an encoded vector. This is the personalized step because the encoded vector now customizes the estimator networks to estimate the expected vital sign behavior specific to the patient. Thus, as an illustration, when a window of post-treatment vital sign data excluding heart rate is input into a member of the set that estimates heart rate, given the encoded vector from pre-treatment plus the other non-heart rate vital signs from the post-treatment time window, it will estimate what the expected heart rate should be. Then, this estimated heart rate can be compared with the actually measured heart rate from a sensor to create the aforementioned heart rate residual. Other members of the set can do the same for other vital signs. In this way, the residuals of all vital signs required to detect or quantify the inflammatory response (as a change relative to normal health) can be provided by the neural networks on a personalized basis.

[0078] The method provided herein utilizes residuals. A residual is equal to each measured vital sign value minus the corresponding estimated value. For example, if the average heart rate over one minute is used as a monitoring parameter, the measured value of the contemporaneous vital sign value is input into the model, and the model outputs a heart rate value, which is the expected or desired heart rate of the person given the other vital signs as a set. Then, the residual of the heart rate is the measured heart rate minus the estimated heart rate.

[0079] Thus, the residuals themselves include time series, one for each vital sign in the model. The value and sign (positive or negative) of the residual have a specific meaning.

[0080] For example, a high residual (positive, >0) means that the measured value is higher than the model expected for the normal physiology of the patient. If the residual is low (<0), then the measured value is lower than the value expected for the normal physiology of the patient. The residuals can then be examined in the detection rules described below.

[0081] In addition to the individual residuals, a single scalar index of overall disorder of vital sign behavior can be generated, which combines the normalized magnitudes of the residuals for each measurement. An example of a single scalar index is the multivariate health index described in USPN 8620591, which is incorporated herein by reference in its entirety and is also referred to as the multivariate change index (MCI) as it will be referred to herein.

[0082] In summary, the personalized baseline model is used to generate estimated values for at least a portion of the training dataset, and the relative magnitudes of the residuals are used to scale the typical expected deviation distribution of the residuals from the baseline under normal (non-disordered) vital sign behavior. When new data observations are input during the monitoring mode, the resulting residuals can be compared to this distribution, thereby generating a single score ranging from 0 to 1, which is related to the likelihood that the residual vector is a member of this expected distribution. When the residual vector is well within the typical distribution (as characterized by the training data scaling results), then the output of the MCI is closer to 0 (low likelihood of different vital sign manifestations), while if the residual vector is closer to the tail of the multidimensional distribution, the MCI is closer to 1 (high likelihood of behavioral change). In this way, the MCI is a measure of the likelihood of disordered vital sign behavior, regardless of the person's underlying activity (i.e., whether they are sleeping, watching TV, running, etc.).

[0083] Thus, the MCI is a time series that has a data point for each input observation of a vital sign estimated by the personalized model. The value of the MCI itself can be used as part of the rule pattern provided by the method described herein.

[0084] Other methods that provide a single scalar index for a multi-dimensional vector of residual data can be used as an alternative to the MCI described above. For example, Euclidean distance is a well-known method for measuring the distance of multi-dimensional points from a distribution of points and can be used as follows: (a) Residuals from training samples (without inflammation) are normalized to a mean of 0 and a standard deviation of 1; (b) The "center" in the multi-dimensional space of residuals from the training samples is found; (c) A range of distances from this center is defined such that the maximum normal likelihood (set to 1) is at zero distance and the minimum normal likelihood (set to 0) is defined at some maximum distance equal to a multiple of the standard deviation (the maximum value being the larger distance); (d) Each residual vector of the test observations is normalized and its distance is calculated to map to the normal likelihood range of 0 to 1. Euclidean distance assumes a spherical distribution; other techniques, such as Mahalanobis distance, may work better with relevant variables such as vital signs used herein. However, when mapping multi-variate residual vectors from test data to the expected distribution of normal training data, the principle is the same, to provide a likelihood within the range of 0 to 1 that can be mapped to whether the residual vector is abnormal or not and the degree of abnormality, just like the MCI technique. When applied to residuals from personalized modeling of expected values, all such methods that produce a single scalar change index are considered signals available within the rules of the present invention. It should be understood that in the present disclosure, when "MCI" is used herein to detect or quantify an inflammatory response, according to the present invention, other single scalar indices can be substituted.

[0085] The concept of a rule pattern (e.g., implemented as computer code) for inflammation detection according to the methods provided herein is that there are markers of inflammation in the residuals as a time series. As discussed herein, given the free-living environment of an individual's physiology and individual behavior, the marker pattern involves one or more relative changes. For example, a lower-than-expected HRV, a higher-than-expected HR, a higher-than-expected temperature, and / or a higher-than-expected respiratory rate can all be such relative changes.

[0086] For example, during inflammation, a reasonable assumption is that the heart rate increases beyond the normal expected value. The personalized model provides the ability to see an elevated heart rate above the normal heart rate by masking the changes in the raw heart rate value due to normal activities of daily life, and this elevation is seen as a positive residual value. The same is true for the respiratory rate.

[0087] In addition, the rule can be gated by the magnitude of the MCI, which quantifies the overall disorder of the vital signs system. When measuring vital signs in a free-living environment, the personalized model can have difficulty isolating changes to one or another vital sign - the model may suffer from "spillover" where a deviation in one variable seeps into the estimate of another variable and the latter variable appears abnormal. Trivial or minute deviations in the above-mentioned vital signs may not produce a reliable signal of inflammation; by increasing the requirement for at least a minimum overall amount of disorder between vital signs, the MCI ensures a more reliable detection.

[0088] The rule for detecting a pattern can also have a non-residual signal. For example, the core temperature is only above the threshold because the human body temperature normally remains within a tight bound of homeostasis and is independent of daily life activities (assuming it drops during sleep and rises during exercise, but these changes are relatively small). Alternatively, as part of the rule, the absolute value of the MCI can be tested.

[0089] Generally, the rule elements for detecting any pattern can usually be selected from several factors. To name just a few examples, these factors include the directionality of the residual, such as positive or negative; the magnitude of the residual; the magnitude of the MCI; the duration or persistence of the pattern; and / or any filter applied to the absolute value of the vital sign or condition.

[0090] The output of the rule can be a modification of the time series, such as the MCI, to modulate its value when the rule is true or false, or to trigger an alarm when the rule is satisfied. Other examples are possible.

[0091] In one example of rule operation, in the presence of a large enough MCI, the operating rule will check for a negative HRV residual (a decrease in HRV relative to the expected value).

[0092] In another example of rule operation, the rule will check to ensure that the MCI is high enough and the HRV residual is not positive, and then, given a gating condition, check for any one of a positive HR residual, a positive RR residual, a positive temperature residual, or a negative HRV residual.

[0093] In yet another example of rule operation, the rule will transform the MCI value to zero based on the rule being false (no inflammation pattern detected). For example, the MCI becomes an inflammatory MCI (iMCI) such that its value is retained when inflammation is recognized or cleared when inflammation is not recognized. This makes the MCI specific to inflammation, and the "iMCI" is essentially a univariate marker of inflammation.

[0094] In yet another example of rule operation, the rule processing then compares the iMCI with a threshold to trigger a detection alarm when the iMCI exceeds the threshold.

[0095] In yet another example of rule operation, the rule uses a window of iMCI values to create an average over the window to smooth the signal, and uses the smoothed signal for quantization or trigger detection.

[0096] In yet another example of rule operation, the rule performs an additional transformation on the iMCI values, where any highest value is "latched" at that value until the next highest value is encountered, or a value low enough is encountered (e.g., keep the signal high until the signal drops by at least a certain amount, indicating a significant reduction in signs of inflammation).

[0097] Reference is now made Figure 1 , to an example of a system 100 for determining and acting on a person's inflammatory response. The system includes a first person 102, a second person 104, and a third person 106. A first sensor 108 is worn by the first person 102, a second sensor 110 is worn by the second person 104, and a third sensor 112 is worn by the third person 106. An electronic network 114 is coupled to the sensors 108, 110, and 112, and to a control circuit 116.

[0098] A memory 118 is coupled to the control circuit 116. The memory 118 stores a first personalized estimator model 120, a second personalized estimator model 122, and a third personalized estimator model 124. The electronic network 114 is also coupled to a machine 126, an electronic device 128, and a manufacturer 130. Other devices or systems may be coupled to the network 116.

[0099] In one example, the first person 102, the second person 104, and the third person 106 are patients. They may be receiving treatment for certain diseases, or may simply be interested in monitoring their own health. In some examples, they may be participating in a clinical study.

[0100] The first sensor 108, the second sensor 110, and the third sensor 112 can be any type of sensor or sensor arrangement (including multiple sensors) wearable by the persons 102, 104, and 106. The sensors can be placed at any convenient location on the human body, such as the torso or wrist. In some aspects, a sensor patch attached to the torso is used and is worn continuously by the person, enabling the measurement of several values. These values can include heart rate and HRV from a single-lead ECG, movement quantification (activity) from a 3-axis accelerometer, respiratory rate from a movement derivative, or amplitude / frequency modulation of the HR, and / or skin temperature at the sensor site. As described above, the above vital signs can be calculated from the waveform data of the ECG and the 3-axis accelerometer. For example, the ECG can be sampled at 125 Hz, and the sampling rate of the accelerometer can be 15 Hz or higher. Other examples are possible.

[0101] By way of example only, the electronic network 114 is any type of electronic communication network or combination of networks, such as a wireless network, the Internet, a local area network, a wide area network, etc.

[0102] The control circuit 116 is any type of processing device, processor, controller, and includes all types of electronic controllers, microcontrollers, servers, or microprocessors, by way of example. The control circuit 116 may also include an internal memory that stores computer instructions that are executed to implement the functions, rules, or other operations described herein. The control circuit 116 and the memory 118 may be arranged at a central processing device or a central call center.

[0103] The memory 118 is any type of storage device or database or combination of devices. By way of example only, the memory 118 may be a read-only memory, a random access memory, a programmable read-only memory, or a combination of these and other types of electronic memories.

[0104] The first personalized estimator model 120, the second personalized estimator model 122, and the third personalized estimator model 124 are configured to estimate physiological variables in response to receiving new physiological data from the sensors 108, 110, and 112 worn by the persons 102, 104, and 106. The personalized estimator models 120, 122, and 124 are any multivariate models that provide an estimate of the expected value of one or more vital sign values as output given measurements from sensors on a patient. By way of example only, such models may be generated by decision tree / random forest function approximators; similarity-based models; and neural networks.

[0105] In one example, the machine 126 is a medical device utilized by one or more of the persons 102, 104, and 106. For example, the machine 126 may dispense medications, monitor the persons 102, 104, or 106, or perform some other function. An electronic control signal 125 may be sent from the control circuit 116 to the machine 126 to control the machine 126 and / or aspects of the operation of the machine 126. The control signal 125 may activate the machine 126, deactivate the machine 126, control the operating parameters of the machine 126 (e.g., by way of example only, speed, amount of medication dispensed, operation of a display at a screen on the machine 126, information displayed on the display, or how often the machine 126 monitors the persons 102, 104, or 106).

[0106] By way of example only, the electronic device 128 can be a smart phone, a laptop computer, a personal computer, a cellular phone, etc. The electronic signal 127 can be sent by the control circuit 116, which can be an electronic message, a control signal, or any other type of electronic signal. For example, the electronic device 128 can be used by one of the persons 102, 104, or 106 and is used to display an alert, an instruction, or other information to the persons 102, 104, or 106. In other examples, the electronic device 128 can be utilized by medical personnel (e.g., a doctor, a nurse, a hospital, or a therapist) treating the persons 102, 104, or 106. Although only one electronic device 128 is shown, it should be understood that there can be multiple devices and that they are operated by different individuals or institutions. Further, the electronic device 128 can communicate with the control circuit 116 and, through the control circuit 116, with other devices or systems coupled to the network 114. The electronic device 128 itself can include a processing device or control circuit to perform the described operations.

[0107] In one example, the manufacturer 130 is a vaccine manufacturer. The control circuit 116 can send a control signal or other electronic instructions 129 to the manufacturer 130. These instructions 129 can automatically cause or inform the manufacturer 130 to change the process. In one example, the manufacturer 120 is a vaccine manufacturer and the electronic instructions 129 cause the manufacturer 130 to change the composition, dosage, or some other characteristic of the vaccine (or drug) being administered to one or more of the persons 102, 104, and 106. In some aspects, this can automatically control the machines and / or processes (at the facility of the manufacturer 130) creating the vaccine or drug. In other examples, the electronic instructions include a warning or other message to the manufacturer 130, where a change to the vaccine or drug is recommended or proposed. The manufacturer 130 can utilize an electronic receiver, transmitter, transceiver, memory, database, server, processor, control circuit, display, computer, and / or other electronic devices (and combinations of these devices) to perform these functions.

[0108] Prior to treatment of persons 102, 104, and 106, models 120, 122, and 124 are trained. When the models are neural networks, it should be understood that the physical structures of these neural networks vary. For example, the weights, layers, or other structures of the neural networks change from a first structure or state to a second structure or state. It should also be understood that the structures of the resulting trained models 120, 122, and 124 are unique to each other. Each trained model 120, 122, and 124 is trained specifically based on data from the corresponding person. Trained model 120 is trained only based on data from person 102, trained model 122 is trained only based on data from person 104, and trained model 124 is trained only based on data from person 106. Thus, these models are completely different from each other, are unique, are personal to the unique person, and cannot be considered general computing resources. Applying the same data to different models does not necessarily produce the same or similar results.

[0109] During monitoring or execution (after the training phase), control circuit 116 applies input data from a specific person 102, 104, and 106 to models 120, 122, and 124 associated with the patient. In some aspects, control circuit 116 may select or retrieve the appropriate model 120, 122, or 124 for the specific data to be received and processed from memory 118. In an example, the data may indicate the person from whom the data originated, or a technician may electronically inform control circuit 116 of the origin of certain data such that control circuit 116 can select the correct model.

[0110] Applying the data to the model produces an estimate of the data. The estimate may be for one or more parameters such as heart rate, heart rate variability, respiratory rate, activity, temperature, etc. The estimate is compared to the actual data (on a parameter-by-parameter basis), and a difference is obtained. This difference is the residual, and the residuals can be determined over time and their patterns analyzed by control circuit 116 as described elsewhere in this document. Control circuit 116 may use the analysis results to determine and execute actions. As described above, these actions may include controlling machine 126, sending instructions to electronic device 128, and / or sending instructions to manufacturer 130. It should be understood that only machine 126, electrical device 128, and manufacturer 130 are shown here, but control circuit 116 may control or notify other devices. It should also be understood that the communication between control circuit 116 and these devices may also be two-way communication, that is, the devices may send instructions or other electronic information to control circuit 116 for other purposes.

[0111] Now refer to Figure 2, which describes an example of a method for detecting a person's inflammatory response. In step 202, physiological data is collected from at least one wearable sensor worn by a patient during a pre-treatment interval.

[0112] In step 204, an individualized estimator model is created based on the physiological data collected from the patient before treatment. This model is capable of estimating physiological variables in response to receiving new physiological data from at least one wearable sensor worn by the patient. The personalized / individualized estimation model is any multivariate model that provides an estimate of the expected value of one or more vital sign values measured from the patient as output. Just to name a few examples, such models can be approximated by decision tree / random forest function approximators; similarity-based models; and neural networks.

[0113] In step 206, during a post-treatment interval, additional physiological data is collected from at least one wearable sensor worn by the patient. The sensor can be any type of sensor or sensor arrangement (including multiple sensors) that can be worn by a person and can be placed at any convenient location on the human body, such as the torso or wrist.

[0114] In step 208, the individualized estimator model is used to generate estimates of the post-treatment physiological data. In some aspects, these estimates are multiple and separate estimates of parameters such as heart rate, heart rate variability, respiratory rate, activity, or temperature.

[0115] In step 210, the post-treatment physiological data is compared with its estimates, and it is determined when a predefined effect pattern exists based at least in part on this comparison. A difference can be obtained between each of the parameters or variables.

[0116] In step 212, when a predefined effect pattern exists, an action is determined and executed. The action is one or more of the following: triggering an electronic questionnaire to be presented to the patient; providing the patient with instructions to take a measurement; providing the patient with instructions to contact the patient's clinician; triggering a ticket in a call center system to queue a call to the patient; creating a prompt in an application on the patient's phone to contact the clinician; providing the patient with instructions related to classifying the effect; transmitting a control signal to control a medical device associated with treating the patient; and transmitting instructions to a vaccine manufacturer to change the composition and / or dosage of a vaccine. Other examples of actions are possible.

[0117] Now refer to Figure 3 , which describes an overview of determining and utilizing residuals according to the methods provided herein. It should be understood that reference Figure 3The described steps may be implemented as computer instructions executed on a processing device or control circuit. At step 302, receive vital sign values from sensor readings from a monitored patient. At step 304, filter multivariate observations of the vital signs to eliminate unacceptable or poor-quality data or activity states that interfere with inflammation detection. Various criteria may be used to determine whether the data is unacceptable.

[0118] The individualized estimator model 306 (trained as described elsewhere herein to be personalized for a specific person) receives the filtered data (filtered at step 304). As mentioned and described elsewhere herein, an individualized estimator model is any multivariate model that provides an estimate of the expected value of one or more vital sign values measured from a patient as output. To name just a few examples, such models may be generated by decision tree / random forest function approximators; similarity-based models; and neural networks.

[0119] (Filtered at step 304) The data may include heart rate data, HRV data, respiratory rate data, core temperature data, skin temperature data, and activity data obtained from sensors deployed on a person. Other examples are possible.

[0120] Once the data (filtered at step 304) is applied to the model 306, the model 306 correspondingly generates an estimate at step 307. The estimate generated at step 307 represents what the model believes the data should be. In some aspects, separate estimates of the heart rate data, HRV data, respiratory rate data, core temperature data, skin temperature data, and activity data are obtained.

[0121] The difference operator 308 obtains the difference between the input data (filtered at step 304) and the estimate (generated at step 307) to produce a residual 310. The residual 310 encodes how each of the measured vital signs differs from the expectation based on pre-treatment data from the patient. In some aspects, separate differences of the heart rate data, HRV data, respiratory rate data, core temperature data, skin temperature data, and activity data are obtained.

[0122] In step 312, the residuals 310 can also be combined into a single scalar value of the overall change, a time series change index (such as the MCI described elsewhere in this document, or an alternative form thereof). Then, in step 314, the pattern representing the vital sign disorder caused by inflammation can be applied to the residuals. The pattern in some aspects is implemented as a composite rule set (e.g., and can be physically implemented as computer code or software), and the change index as well as the measured vital signs can also be utilized in rule evaluation to increase the residuals 310. In some aspects, the change index is useful for capturing overall changes that are sufficiently significant, while the measured values can also play a role in confirming commonly understood inflammation indicators (e.g., temperature). All the residuals, change indices, and measured values are time series data, such that applying the rule set to detect the inflammation pattern also produces a time series, and represents an inflammation marker or biomarker at the instant time point in step 316.

[0123] Given that inflammation may be developing and may manifest inconsistently in vital signs, it can become important to integrate the instant inflammation biomarkers over time (in step 318) to produce a signal of the persistent evidence of the inflammatory response. The integration can be a time window statistic, or can be a latching process (e.g., as described elsewhere in this document, sometimes using or comparing thresholds) combined with the integration of the area under the curve and performed in step 320. The result determined in step 320 can be used to determine various actions in step 322. These actions are described elsewhere in this document.

[0124] Now referring to Figure 4 , an example of a method for determining the effects of inflammation and taking actions on these effects is described. Steps 402 - 410 are part of a model training or learning process, and the remaining steps are part of a monitoring process (occurring after the training or learning process is completed).

[0125] In step 402, one or more sensors are placed on and worn by a patient, and data is obtained. In some aspects, a sensor patch attached to the torso is used and continuously worn by the patient, enabling the measurement of several variables. These variables can include heart rate and HRV from a single - lead ECG, movement quantification (activity) from a 3 - axis accelerometer, respiratory rate from a moving derivative, or amplitude / frequency modulation of HR, and / or skin temperature at the sensor site.

[0126] In step 404, the vital signs can be calculated based on the waveform data of the ECG and the 3 - axis accelerometer. For example, the ECG can be sampled at 125 Hz, and the sampling rate of the accelerometer can be 15 Hz or higher.

[0127] Heart rate can be calculated on a beat-by-beat basis. Respiratory rate can be calculated at a sampling rate of 5 seconds, although higher and lower sampling rates are acceptable. Activity can be calculated as a vector magnitude measure of movement (vibration) from all 3 axes, typically at 1 Hz, in any of the various measurement units known in the art. Temperature measurements are typically taken at 1 Hz, although lower sampling rates commensurate with the expected rate of change of the human body or environmental conditions are acceptable. Generally, all variables should be measured at a rate of at least once per minute, and preferably at the higher rates described.

[0128] In step 406, vital signs (e.g., HR, RR, HRV, activity, and temperature) are then statistically summarized on a one-minute windowing basis. In some aspects, this is a 10% trimmed mean of the minute values. However, other statistics are acceptable and can be used, such as median, nth percentile, mode, maximum, or minimum.

[0129] In step 408, due to motion artifacts and occasional loss of sufficient signal-to-noise ratio, a signal quality index (SQI) is used to identify when the vital signs are reliable for subsequent treatment. In some aspects, the signal quality index can be scaled from zero (e.g., indicating a poor or unacceptable signal) to one (e.g., indicating a perfect signal), and can be evaluated using the ECG waveform. In methods for providing such SQI, a deep neural network can be reliably trained to output an SQI that exhibits a high correlation with human expert evaluation of ECG trace availability. For example, the SQI input can be a 10-second-long ECG window, containing a range of approximately 7 to approximately 30 heartbeats (depending on heart rate), and can be evaluated at a 5-second period (e.g., 5-second overlap). Other methods of generating SQI using the ECG waveform are known to those skilled in the art. Given an SQI ranging from 0 to 1, the threshold for excluding data from processing can be set to <0.8. Other examples are possible.

[0130] In addition to SQI, activity level can also be used as a filter for whether to use data for processing. The absolute value of the activity level can be applied, and data exceeding this value will not be used. This provides a simple way to avoid bad data and does not cause a large loss of data for processing, as the level can be set such that the monitored person rarely exhibits movement at that level during the day. The activity threshold for excluding data from processing can be set at a level in movement units corresponding to robust walking.

[0131] After filtering the one-minute windowed average of the vital signs using SQI and the activity threshold, what remains are samples of HR, RR, HRV, activity, and temperature that can be collected from the monitored person and used to train a personalized baseline model.

[0132] At step 410, and as described elsewhere herein, a variety of methods can be used to create a baseline model (an individualized estimator model) that, after training, can generate estimates for comparison with measurements. In some aspects, a similarity-based model (SBM) is used. Such a model can be generated from one or two diurnal cycles of a person's one-minute continuous data (1440 samples per day); however, more days of data can be used if the use case permits. For example, if a patient will receive therapy at a later scheduled date, a week's worth of samples can be collected to build a personalized model.

[0133] Once the model is trained, it enters or is used in a monitoring mode at step 412, in which it generates estimates in response to the input of each new sample of one-minute multivariate vital sign data obtained from the individual being monitored. This occurs continuously during all activities of daily life, subject to the filters described above. When each multivariate reading of HR, RR, HRV, activity, and temperature is obtained, it is input into the model to generate an estimate of the expected one-minute value for each variable.

[0134] At step 414, a residual is generated by subtracting the estimated value from the measured one-minute value. For example, in the case where the estimated one-minute heart rate is lower than the measured value, the residual is positive, meaning that the person's actual heart rate is higher than expected. A residual is generated for each vital sign parameter, and the residuals occur at the same sampling rate of once per minute.

[0135] At step 416, the residuals are combined into a multivariate change index (MCI) that quantifies the overall distribution between the set of measured vital signs and the set of estimated vital signs. The MCI calculation is described elsewhere herein. In some aspects, the MCI is scaled between 0 (meaning no abnormal behavior) and 1 (meaning high confidence of abnormal behavior). The MCI is also generated at a sampling rate of 1 minute, with each input observation.

[0136] Using the available one-minute vital signs, residuals, and MCI as inputs, an inflammation-specific detection pattern is applied in the form of rules. At step 418, the rule outputs a modified MCI for inflammation, designated as iMCI for the purposes of this article. In one example, the rule logic follows these steps:

[0137] Initialize iMCI = 0

[0138] If ((HRV RES <= 0) and (Temp RES >= 0) and (MCI >= 0.1)), then:

[0139] If ((HRRES > 0.5) or (RR RES(> 0.5) or (Temp RES (> 0.2) or (HRV RES (<- 0.025)), then:

[0140] Set iMCI = MCI

[0141] Wherein:

[0142] HR RES Is measured in beats per minute.

[0143] RR RES Is measured in breaths per minute.

[0144] Temperature is measured in degrees Fahrenheit.

[0145] HRV RES Is measured in seconds.

[0146] This rule basically zeros out the MCI value, unless the evidence has inflammatory characteristics that make iMCI specific to inflammation. The first part of the rule provides a gate for relatively high HRV or relatively low temperature or overall lack of change (MCI). In other words, the rule fundamentally looks for sufficient disorder quantified in MCI, but only if that disorder is not due to higher-than-expected HRV or lower-than-expected temperature. Once this initial gating condition is met, the second part of the rule looks for any of the possible inflammatory positive evidence, which comes in the form of higher-than-expected HR, or higher-than-expected RR, or higher-than-expected temperature or lower-than-expected HRV, and determines the minute inflammatory MCI (M_iMCI) at step 420. If the presence level of any one or more of those evidence forms exceeds their respective thresholds, then iMCI is set to the value of MCI.

[0147] M_iMCI is a time series of values that are also at a one-minute sampling rate. At step 422, the persistence of the evidence of inflammatory physiological disorder is examined by summing or averaging the iMCI values above the window, which is also used to smooth the signal from transient noise. In some aspects, this window is 3 hours and is evaluated every 15 minutes on average (2 hours 45 minutes overlap). Other statistical characterizations of the windowed iMCI are possible.

[0148] In other aspects, and at step 422, when the time series values continuously exceed a "latch" threshold until a low enough value sequence drops below the threshold, the one-minute iMCI time series is "latched" to the highest value, at which point iMCI is unlocked from the higher value and remains unlocked at the lower value. Then at step 424, a 3-hour sum or average of the latched version of iMCI is taken.

[0149] Latch can be understood through the integer toy example: Given a time series of instantaneous values as shown in the first series below, a latch threshold of 6, and a persistence requirement for two samples, after the first series of 6 and 7, the series is latched to the highest value until the original series drops below the value of 6 for at least two consecutive samples, at which point the samples are unlocked, as shown in the second series below:

[0150] [4,6,5,5,6,7,8,7,6,7,5,3,1,2,1]

[0151] [4,6,5,5,6,7,8,8,8,8,8,8,1,2,1]

[0152] By itself, 1-minute iMCI or 3-hour windowed (summed or averaged) iMCI or latched iMCI with a 15-minute sampling rate can be used as a quantitative indicator of a suspected inflammatory response. This helps to quantify trends or evaluate changes from before to after treatment in a drug trial. A drug study of a patient can examine the change in windowed iMCI / latched iMCI from before to after treatment administration and compare the responses of different groups (e.g., treatment group vs. control group). Any of 1-minute iMCI, 1-minute latched iMCI, windowed iMCI, or windowed latched iMCI can be recorded on a per-unit-time basis (e.g., daily) to quantify the overall relative inflammatory response present in the physiological function of any patient or trial participant.

[0153] An action can be taken at step 426. In cases where it is useful for detecting and escalating an inflammation notification (e.g., an acute inflammation attack or cytokine release syndrome), the windowed iMCI / latched iMCI can also be compared to a threshold, and if the value exceeds the threshold, an acute level of inflammation is identified and an alert is provided to the clinician and / or patient to take measures to mitigate the acute response. In some aspects, when the value of windowed, latched iMCI exceeds 0.25, a notification can be triggered. This threshold is used to distinguish an acute level of inflammation (requiring intervention) from an expected level of inflammation (not requiring classification). For example, in the case of cancer treatment using the power of the human immune system, this can be useful. A moderate level of inflammation is expected because the treatment is effective against cancer, but there is also a risk of an acute inflammation attack, such as a potentially fatal cytokine "storm".

[0154] In other aspects, the threshold for escalating or sending an acute inflammation alert is applied to the difference between samples in the windowed iMCI / latched iMCI. In other words, the sudden rise of consecutive 15-minute samples of the 3-hour windowed value is examined. If a rise at least as large as the threshold is seen, regardless of the absolute value, an alert is triggered.

[0155] Turning to Figure 5, shows a safety monitoring system for a patient undergoing medical treatment (e.g., for a patient who may have side effects of cytokine release syndrome, which may be dangerous to the patient). In some aspects, the system provides a safety net for the patient after the patient leaves the urgent care environment and moves to a home environment away from the urgent care facility. The system includes a wearable torso patch sensor 505 that transmits data via a Bluetooth radio (or some other communication technology or protocol) to a smartphone 510 owned by the patient. The sensor data is transmitted by an application (e.g., software application) on the smartphone to a data center (also referred to as a central call center) via a public data infrastructure 515, which can be the Internet or other mobile data infrastructure. The data center includes a processor unit (or control circuit) 520 connected to a data storage memory 525. The central call center or data center can be a doctor's office, hospital, acute care facility, hospital group, clinic, any one or more healthcare facilities, or can be associated with them, or can be a data processing center located at a geographical location different from the healthcare facility. Other locations are possible. The processor 520 and the associated data storage memory 525 can be physically located at the central call center, but can also be located at other locations, such as the cloud or another healthcare facility.

[0156] The patient's pre-treatment sensor data prior to previous treatment has been used to train a personalized multivariate estimator model 530 stored in the memory 525 and is accessible by the processor 520 for computational steps that include generating an estimate of the expected value of the data measured by the sensor 505 and comparing it to generate residuals and univariate scores, such as MCI. The residuals and MCI are then used to detect markers of a progressive inflammatory process in the patient's body.

[0157] Buttons can be used to initiate operations on the smartphone 510. When a button press, depression, or actuation is detected, the processor 520 transmits instructions to an application on the smartphone to prompt the patient with one or more actions embodied in screen button features: button 540, which is used to fill out a survey on the smartphone that can confirm or deny symptoms related to inflammation; button 545, which automatically connects the patient to a clinician responsible for managing the patient, which can include a call center of trained personnel with those responsibilities; and button 550, which is used to perform and report additional measurements using an instrument the patient has at home, such as a thermometer for obtaining core body temperature, a blood pressure measurement device, and / or a pulse oximeter. In some aspects and as described above, buttons 540, 545, and 550 can be displayed on the screen of the smartphone 510. The user can press or otherwise actuate these buttons by touching and pressing the screen in the appropriate area where the buttons are displayed (e.g., using their finger, cursor, or stylus). Alternatively, buttons 540, 545, and 550 can be physical buttons on the smartphone 510 that are physically pressed by the user.

[0158] More specifically, and during some operations, the processor 520 obtains sensor data 570, applies the sensor data to generate an estimate at step 572, uses a difference operator 574 to obtain the difference between the sensor data 570 and the estimate (obtained at step 572) to obtain a residual, and then determines MCI at step 576, as described herein. At step 578, rules can be used to determine an inflammation pattern, and an action is taken at step 580. These steps are described in more detail elsewhere in this document.

[0159] For example, with respect to button 540, pressing the button causes a survey to be displayed on the screen of the smartphone 510. The survey can include several pages that the patient flips through (each page on a separate screen). As the patient flips through the survey, they can answer questions. In one example, they can type information and / or check boxes depending on the nature and format of the survey. In another example, the patient can verbally input information into the smartphone 510. That is, once a question is posed, the patient can say the answer, the microphone on the smartphone 510 can receive the spoken information, and the information can be received by the processor or control circuitry of the smartphone for further processing.

[0160] Regarding button 545, for example, a patient can be identified by a patient number, code, and / or other suitable identifier. Once button 545 is pushed, the application transmits an electronic message including the code to a call center (or other facility). In these respects, pressing button 545 can establish a telephone, video, and / or other type of electronic communication connection or link with a central call center. Each patient (via their patient number or code) can be associated with a specific clinician. Thus, and in one example, a control circuit 520 at the central call center (associated with the clinician) can receive the patient code and / or number, identify the clinician associated with that number or code, and then establish a link between the patient and the associated clinician.

[0161] For each individual clinician, the central call center can store separate and distinct protocols for contacting or reaching different clinicians. Each of these can be stored as a separate program or computer instructions (e.g., stored at memory 525), which are executed when needed (e.g., by control circuit 520). For example, for a first clinician, a text message is automatically generated instructing the clinician to call the patient. In another example, an automatic call (or other electronic communication) is made to the clinician, and when the clinician confirms, a telephone link or connection is directly established between the clinician and the patient. It should be understood that the execution of such computer instructions automatically controls electronic communication appliances by setting switches, routing messages between various networks and components of these networks such as gateways, and otherwise controls various electronic devices within infrastructure 515, for example. Once the appropriate clinician is identified, the control circuit 520 of the central call center can identify, retrieve, and then execute the appropriate program to contact that clinician.

[0162] Security or verification protocols can also be automatically and / or manually executed at the central call center to verify the identity of the patient. For example, control circuit 520 can check the patient code against a database of valid patient codes. In other respects, security questions can be presented to the patient on a smartphone before the process can continue. In other words, a correct response to these questions may be required from the patient to continue the process.

[0163] More specifically, the patient's responses can be sent to the central call center, and the control circuit 520 at the central call center can compare the patient's responses with the responses stored at the central call center (e.g., at memory 525). A correct match allows the establishment of a link and / or the exchange of information. An incorrect match can interrupt processing operations regarding the patient. In these respects, any program attempting to establish a link can be stopped and / or the electronic message can be set back to smartphone 510, notifying the patient that they did not pass the security check.

[0164] In other examples, the biometric information can indicate the identity of the patient. In some aspects, the patient can place their finger on the screen of the smartphone 510 (or some other device that obtains fingerprints connected to the smartphone 510), and this information is sent to a central call center. The control circuit 520 compares the fingerprint with the stored fingerprints archived at the central call center.

[0165] In other examples, the initiation of the process can cause the retrieval of the patient's insurance information. In some examples, the insurance information can be stored at the central call center, but in other examples, it can be stored at an external source. If needed, another electronic connection (e.g., automatically established by the control circuit 520) can be established between the central call center and the external source to access, retrieve, and / or query the insurance information. The insurance information can be used to determine whether to continue the process or how to continue. In some examples, depending on the nature of the insurance information, the process and any program being executed to implement the process can be electronically stopped by the control circuit 520 at the central call center. The infrastructure 515 can be used to establish any connections required to obtain information from the external source.

[0166] Regarding the button 550, the report can be manually typed or captured by an application on the smartphone via Bluetooth from the measuring device. In some aspects, the patient can first establish a communication link between the smartphone 510 and the additional instrument. For example, the smartphone in the setup mode can attempt to detect the presence of the additional device. This can be achieved using many different protocols, where once detected, the link is established. For example, the smartphone 510 can send a signal to the additional device, and the additional device can respond. Alternatively, the additional device can broadcast a signal detected by the smartphone 510. Once the link is established, the smartphone 510 can receive information from the additional device periodically or continuously.

[0167] Measurements made by the patient or survey answers provided by the patient are automatically uploaded by the application to a data center, which notifies the clinician of the detected inflammation pattern and the information provided by the patient in response thereto. Using this system has the great advantage of providing continuous monitoring of the patient in a home environment without the cost of full - day admission to an urgent care facility. Additionally, the personalized model - based detection of early warning signs of inflammation escalation, coupled with automatic actions pushed to the patient, provides a more actionable context for the clinician to make decisions, where any detection is escalated to the clinician, thus improving the timeliness of intervention with assurance.

[0168] The methods provided herein are advantageous for various reasons. For example, these methods are individualized and provide better sensitivity and specificity. These methods learn about a person's physiological performance from data, such that when inflammation affects the physiology, changes are highlighted.

[0169] In addition, the methods provided herein are objective. They do not rely on subjective feelings as do symptom reports.

[0170] In addition, the methods provided herein do not require particular reliance on the patient to actively perform tasks to collect data. The patient only needs to wear a sensor. In contrast, other methods that attempt to detect severe inflammatory responses include asking the patient to regularly measure their temperature. Relying on the patient to do something at regular intervals is always problematic.

[0171] The method uses an individualized model tailored to a specific patient. The method also looks for evidence of adverse side effects, such as patterns in the differences between the model estimates and the values actually measured from sensors on the patient.

[0172] The method provides instructions to the patient to address side effects. These instructions can be instructions to fill out a questionnaire with symptom confirmation or severity, instructions to visit a clinic for testing or treatment, instructions to take ameliorating medications, and / or instructions to perform manual measurements with another device.

[0173] The model can be characterized as (a) “dynamic” such that the input data to the model plays a role in how the estimate is arrived at, or (b) as a “multivariate” model. An “estimate” is a static average; the difference from that static value is considered an “individualized” residual, which can be tested for “markers” (large differences) of inflammation.

[0174] Those skilled in the art will understand that various other modifications, changes, and combinations can be made to the above embodiments without departing from the scope of the invention, and such modifications, changes, and combinations are considered to be within the scope of the concept of the invention.

Claims

1. A method for monitoring the effectiveness of a patient's drug treatment, comprising the steps of: collecting physiological data from at least one wearable sensor worn by the patient during a pre-treatment interval; creating an individualized estimator model based on the physiological data collected from the patient before treatment, the model being capable of estimating a physiological variable in response to receiving new physiological data from the at least one wearable sensor worn by the patient; collecting additional physiological data from the at least one wearable sensor worn by the patient during a post-treatment interval; generating an estimate of the post-treatment physiological data using the individualized estimator model; comparing the post-treatment physiological data with its estimate and determining when a predefined effect pattern exists based at least in part on the comparison; and when the predefined effect pattern exists, determining and performing an action, the action being one or more of the following: triggering an electronic questionnaire to be presented to the patient; providing the patient with instructions to take a measurement; providing the patient with instructions to contact the patient's clinician; triggering a ticket in a call center system to queue a call to the patient; creating a prompt in an application on the patient's phone to contact the clinician; providing the patient with instructions related to classifying the effect; transmitting a control signal to control a medical device associated with treating the patient; transmitting instructions to a vaccine manufacturer to change the composition and / or dosage of a vaccine.

2. The method according to claim 1, wherein the effect is an adverse side effect.

3. The method according to claim 1, wherein the effect is an indication of the efficacy of a vaccine.

4. The method according to claim 1, wherein the physiological data includes heart rate data, respiratory rate data, core temperature data, skin temperature data, and activity data.

5. The method according to claim 1, wherein the individualized estimator model is trained using data collected from the patient in a free-living physiological state, in which the patient's inflammatory state is stable and not expected to change.

6. The method according to claim 1, wherein the comparison includes determining a residual between the estimate and the physiological data.

7. The method according to claim 6, wherein the residuals are synthesized into a single score, the score being a scalar index.

8. The method according to claim 1, wherein the individualized estimator model includes a neural network.

9. A system for monitoring the effectiveness of a patient's drug treatment, the system comprising: at least one wearable sensor worn by the patient during a pre-treatment interval, the at least one wearable sensor being configured to collect physiological data from the patient; a control circuit coupled to the at least one wearable sensor, the control circuit being configured to: create an individualized estimator model based on the physiological data collected from the patient before treatment, the model being capable of estimating a physiological variable in response to receiving new physiological data from the at least one wearable sensor worn by the patient; wherein the at least one wearable sensor collects additional physiological data during a post-treatment interval from the at least one wearable sensor worn by the patient; wherein the control circuit is further configured to: generate an estimate of the post-treatment physiological data using the individualized estimator model; compare the post-treatment physiological data with its estimate and determine when a predefined effect pattern exists at least in part based on the comparison; and when the predefined effect pattern exists, determine and perform an action, the action being one or more of the following: provide the patient with instructions related to classifying the effect; transmit a control signal to control a medical device associated with treating the patient; transmit instructions to a vaccine manufacturer to change the composition and / or dosage of a vaccine.

10. The system of claim 9, wherein the effect is an adverse side effect.

11. The system of claim 9, wherein the effect is an indication of the potency of a vaccine.

12. The system of claim 9, wherein the physiological data includes heart rate data, respiratory rate data, core temperature data, skin temperature data, and activity data.

13. The system of claim 9, wherein the control circuit trains the individualized estimator model using data collected from the patient when in a free-living physiological state, in which the patient's inflammatory state is stable and not expected to change.

14. The system of claim 9, wherein the control circuit is configured to perform the comparison by determining a residual between the estimate and the physiological data.

15. The system of claim 14, wherein the control circuit is configured to synthesize the residual into a single score, the score being a scalar index.

16. The system of claim 9, wherein the individualized estimator model includes a neural network.

Images