Machine learning-based safety controller

By applying machine learning-based safety control system in a medical environment, identifying the shifts and high-risk periods of clinicians, the problem of difficult time detection of harmful behaviors is solved, effectively preventing drug transfer and mishandling, and improving the medical environment security and resource management efficiency.

CN113507914BActive Publication Date: 2024-11-15CAREFUSION 303 INC
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202080017942.4
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Priority Date
2019-01-07
Filing Date
2020-01-07
Publication Date
2024-11-15
Estimated Expiration
2040-01-07

AI Technical Summary

Technical Problem

In a medical environment, harmful behaviors (such as drug transfer and mishandling) are difficult to detect and prevent in a timely manner, resulting in loss of resources and patient harm.

Method used

Using a machine learning-based security control system, the clinical physician's activity transaction data is analyzed, shifts and high-risk periods that may be involved in hazardous behaviors are identified, protection workflows are activated to prevent hazardous behaviors, and relevant evidence is collected.

Benefits of technology

Effectively identify and prevent harmful behaviors, reduce drug transfer and mishandling, and improve the safety and resource management efficiency of the medical environment.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN113507914B_ABST
    Figure CN113507914B_ABST
Patent Text Reader

Abstract

A method may include identifying shifts associated with a clinician by applying a machine learning model, the machine learning model being trained to identify one or more shifts associated with the clinician based on a series of transaction records associated with the clinician. Based at least on the shifts associated with the clinician, the clinician may be identified as potentially engaging in a hazardous behavior. A protection workflow is activated in response to determining that the clinician is potentially engaging in a hazardous behavior. The protection workflow may be configured to prevent the clinician from engaging in a hazardous behavior and to collect evidence associated with the hazardous behavior. Related methods and articles are also disclosed.
Need to check novelty before this filing date? Find Prior Art

Description

[0001] CROSS-REFERENCE TO RELATED APPLICATIONS

[0002] This application claims priority to U.S. Provisional Application No. 62 / 789,337, entitled “MACHINE LEARNING BASED SHIFTTTRACKING” and filed on January 7, 2019, the disclosure of which is incorporated herein by reference in its entirety. Technical Field

[0003] The subject matter described herein relates generally to machine learning and, more particularly, to machine learning-based techniques for detecting and preventing compromised behavior. Background Art

[0004] Hazardous behavior that results in loss of resources and harm to patients and clinicians may occur in various medical settings. Diversion may be an example of hazardous behavior, where instead of administering a substance to a patient and / or discarding the substance, a clinician retains the substance for unauthorized personal use and / or personal gain. High-value medical supplies and controlled medications (e.g., opiates, opioids, narcotics, and / or the like) may be particularly susceptible to diversion. Other examples of hazardous behavior may include the unintentional mishandling of a substance. For example, a clinician may seize, administer, and / or discard an incorrect type and / or quantity of medication. Summary of the invention

[0005] Systems, methods, and articles of manufacture, including computer program products, for machine learning-based security controllers are provided. In some exemplary embodiments, a system is provided that includes at least one processor and at least one memory. The at least one memory may include program code that provides operations when executed by at least one processor. The operations may include: identifying a first shift associated with a first clinician by at least applying a machine learning model, the machine learning model being trained to identify one or more shifts associated with the first clinician based at least on a series of transaction records associated with the first clinician; determining that the first clinician may be involved in hazardous behavior based at least on the first shift associated with the first clinician; and activating a protection workflow in response to determining that the first clinician may be involved in hazardous behavior.

[0006] In some variations, one or more features disclosed herein including the following features may optionally be included in any feasible combination. A first shift associated with a first clinician may be compared to a second shift associated with the first clinician and / or a third shift associated with a second clinician. A determination may be made based at least on the comparison that the first clinician may have engaged in a hazardous behavior.

[0007] In some variations, the first shift may be compared to the second shift and / or the third shift by comparing at least a first plurality of transaction records included in the first shift associated with the first clinician and a second plurality of transaction records included in the second shift associated with the first clinician and / or a third shift associated with the second clinician. In response to detecting one or more anomalous transaction records included in the first plurality of transaction records, it may be determined that the first clinician may have engaged in a compromising behavior.

[0008] In some variations, the first shift may be compared to the second shift and / or the third shift associated with the first clinician based at least on the first shift having one or more attributes in common with the second shift associated with the first clinician and / or the third shift associated with the second clinician.

[0009] In some variations, a first shift associated with the first clinician and a third shift associated with the second clinician may be compared based at least on the first clinician and the second clinician having one or more attributes in common.

[0010] In some variations, one or more high-risk periods during which the first clinician may engage in hazardous behavior may be identified based on at least a first shift associated with the first clinician.A protective workflow may be activated during the one or more high-risk periods.

[0011] In some variations, the one or more high risk periods may include at least one of a first amount of time after the start of the first shift and a second amount of time before the end of the first shift.

[0012] In some variations, the one or more high risk periods may include a portion of the first shift that exceeds a threshold amount of time.

[0013] In some variations, the machine learning model may include a probabilistic machine learning model that may be trained to determine the probability of the first clinician being on duty and being off duty based at least on a series of transaction records associated with the first clinician.

[0014] In some variations, a first transition from the off-duty state to the on-duty state may correspond to the beginning of a first shift associated with the first clinician. A second transition from the on-duty state to the off-duty state may correspond to the end of the first shift associated with the first clinician.

[0015] In some variations, the probabilistic machine learning model may be trained using reinforcement learning techniques including: Q-learning, Monte Carlo, state-action-reward-state-action (SARSA), deep Q network (DQN), deep deterministic policy gradient (DDPG), asynchronous actor-critic algorithm (A3C), trust region policy optimization (TRPO), and / or proximal policy optimization (PPO).

[0016] In some variations, the series of transaction records may include one or more transaction records generated in response to the first clinician interacting with the access control system.The transaction record may indicate scanning of an access token at an entry point to the treatment facility via the access control system.

[0017] In some variations, the series of transaction records may include one or more transaction records generated in response to the first clinician interacting with the dispensing system.The transaction record may indicate accessing a dispensing cabinet associated with the dispensing system to retrieve a substance.

[0018] In some variations, the series of transaction records may include one or more transaction records generated in response to a first clinician interacting with an electronic medical record (EMR) system. The transaction record may indicate that a substance was administered to a patient identified in the EMR system.

[0019] In some variations, the protection workflow may include generating an alert identifying the first clinician as potentially engaging in compromising behavior.

[0020] In some variations, the protection workflow may include activating one or more monitoring devices in response to the first clinician interacting with the medical device and / or isolating a substance accessed by the first clinician.

[0021] In some variations, protecting the workflow may include configuring one or more medical devices to prevent a first clinician from retrieving, administering, and / or discarding a substance without authorization by a second clinician.

[0022] In another aspect, a method for machine learning-based security control is provided. The method may include: identifying a first shift associated with a first clinician by at least applying a machine learning model, the machine learning model being trained to identify one or more shifts associated with the first clinician based at least on a series of transaction records associated with the first clinician; determining that the first clinician may be engaged in a hazardous behavior based at least on the first shift associated with the first clinician; and activating a protection workflow in response to determining that the first clinician may be engaged in a hazardous behavior.

[0023] In some variations, one or more of the features disclosed herein including the following features can optionally be included in any feasible combination. The method can also include: comparing a first shift associated with a first clinician and a second shift associated with the first clinician and / or a third shift associated with a second clinician; and determining that the first clinician may have engaged in a hazardous behavior based at least on the comparison.

[0024] In some variations, the first shift may be compared to the second shift and / or the third shift by comparing at least a first plurality of transaction records included in the first shift associated with the first clinician and a second plurality of transaction records included in the second shift associated with the first clinician and / or a third shift associated with the second clinician. In response to detecting one or more anomalous transaction records included in the first plurality of transaction records, it may be determined that the first clinician may have engaged in a compromising behavior.

[0025] In some variations, the first shift may be compared to the second shift and / or the third shift associated with the first clinician based at least on the first shift having one or more attributes in common with the second shift associated with the first clinician and / or the third shift associated with the second clinician.

[0026] In some variations, a first shift associated with the first clinician and a third shift associated with the second clinician may be compared based at least on the first clinician and the second clinician having one or more attributes in common.

[0027] In some variations, the method may further include: identifying, based at least on a first shift associated with the first clinician, one or more high-risk periods during which the first clinician may engage in hazardous behavior; and activating a protective workflow during the one or more high-risk periods.

[0028] In some variations, the one or more high risk periods may include at least one of a first amount of time after the start of the first shift and a second amount of time before the end of the first shift.

[0029] In some variations, the one or more high risk periods may include a portion of the first shift that exceeds a threshold amount of time.

[0030] In some variations, the machine learning model may include a probabilistic machine learning model that may be trained to determine the probability of the first clinician being on duty and being off duty based at least on a series of transaction records associated with the first clinician.

[0031] In some variations, a first transition from the off-duty state to the on-duty state may correspond to the beginning of a first shift associated with the first clinician. A second transition from the on-duty state to the off-duty state may correspond to the end of the first shift associated with the first clinician.

[0032] In some variations, the probabilistic machine learning model may be trained using reinforcement learning techniques including: Q-learning, Monte Carlo, state-action-reward-state-action (SARSA), deep Q network (DQN), deep deterministic policy gradient (DDPG), asynchronous actor-critic algorithm (A3C), trust region policy optimization (TRPO), and / or proximal policy optimization (PPO).

[0033] In some variations, the series of transaction records may include one or more transaction records generated in response to the first clinician interacting with the access control system.The transaction record may indicate scanning of an access token at an entry point to the treatment facility via the access control system.

[0034] In some variations, the series of transaction records may include one or more transaction records generated in response to the first clinician interacting with the dispensing system.The transaction record may indicate accessing a dispensing cabinet associated with the dispensing system to retrieve a substance.

[0035] In some variations, the series of transaction records may include one or more transaction records generated in response to a first clinician interacting with an electronic medical record (EMR) system. The transaction record may indicate that a substance was administered to a patient identified in the EMR system.

[0036] In some variations, the protection workflow may include generating an alert identifying the first clinician as potentially engaging in compromising behavior.

[0037] In some variations, the protection workflow may include activating one or more monitoring devices in response to the first clinician interacting with the medical device and / or isolating a substance accessed by the first clinician.

[0038] In some variations, protecting the workflow may include configuring one or more medical devices to prevent a first clinician from retrieving, administering, and / or discarding a substance without authorization by a second clinician.

[0039] In another aspect, a computer program product is provided, the computer program product comprising a non-transitory computer-readable medium storing instructions. The instructions may cause operations when executed by at least one data processor. The operations may include: identifying a first shift associated with a first clinician by at least applying a machine learning model, the machine learning model being trained to identify one or more shifts associated with the first clinician based at least on a series of transaction records associated with the first clinician; determining that the first clinician may have engaged in a hazardous behavior based at least on the first shift associated with the first clinician; and activating a protection workflow in response to determining that the first clinician may have engaged in a hazardous behavior.

[0040] In another aspect, a device for machine learning-based security control is provided. The device may include: a component for identifying a first shift associated with a first clinician by at least applying a machine learning model, the machine learning model being trained to identify one or more shifts associated with the first clinician based at least on a series of transaction records associated with the first clinician; a component for determining that the first clinician may be involved in a hazardous behavior based at least on the first shift associated with the first clinician; and a component for activating a protection workflow in response to determining that the first clinician may be involved in a hazardous behavior.

[0041] Implementations of the current subject matter may include methods consistent with the descriptions provided herein and articles including tangibly embodied machine-readable media, which are operable to cause one or more machines (e.g., computers, etc.) to generate operations that implement one or more of the features described. Similarly, a computer system is also described, which may include one or more processors and one or more memories connected to one or more processors. The memory that may include a non-volatile computer-readable or machine-readable storage medium may include one or more programs, encode one or more programs, store, etc., and the one or more programs cause one or more processors to perform one or more operations described herein. The computer-implemented method consistent with one or more implementations of the current subject matter may be implemented by one or more data processors present in a single computing system or multiple computing systems. Such multiple computing systems may be connected via one or more connections including, for example, to a connection on a network (e.g., the Internet, a wireless wide area network, a local area network, a wide area network, a wired network, etc.), via a direct connection between one or more computing systems in multiple computing systems, etc., and data and / or commands or other instructions may be exchanged.

[0042] Details of one or more variations of the subject matter described herein are set forth in the accompanying drawings and the following description. Other features and advantages of the subject matter described herein will be apparent from the description and drawings, as well as from the claims. Although certain features of the subject matter disclosed herein are described for illustrative purposes with respect to applying machine learning models to detect and prevent harmful behavior, it should be readily understood that such features are not intended to be limiting. The claims appended to this disclosure are intended to define the scope of the protected subject matter. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate certain aspects of the subject matter disclosed herein and, together with the description, help explain some principles associated with the disclosed embodiments.

[0044] Figure 1 depicts a system diagram illustrating an example of a security control system according to some exemplary embodiments;

[0045] Figure 2A depicts a flow chart illustrating an example of a process for machine learning-based security control according to some exemplary embodiments;

[0046] Figure 2B depicts a flow chart illustrating another example of a process for machine learning-based security control according to some exemplary embodiments;

[0047] Figure 3depicts a block diagram illustrating a computing system according to some exemplary embodiments;

[0048] Figure 4 depicts examples of medical devices according to some exemplary embodiments; and

[0049] Figure 5 Another example of a medical device is depicted in accordance with some exemplary embodiments.

[0050] Indeed, similar reference numerals indicate similar structures, features or elements. DETAILED DESCRIPTION

[0051] In a medical institution, hazardous behavior may occur at various times. For example, medical supplies and medications may be susceptible to diversion and / or unintentional mishandling at any point during the transportation, receipt, storage, distribution, administration and / or disposal of the material. Due to the lack of adequate custody supervision during the transportation, receipt, storage, distribution, administration and disposal of prescription analgesics, the prescription analgesics may be particularly susceptible to diversion and unintentional mishandling. For example, distribution cabinets at medical institutions may be accessible to multiple clinicians. In addition, different clinicians may be responsible for the distribution, administration and disposal of medications. Therefore, hazardous behavior is usually detected only after the fact that the hazardous behavior has caused losses and / or damage. Delayed detection of hazardous behavior may further prevent the identification of the clinician who caused the hazardous behavior.

[0052] In some exemplary embodiments, the security controller can be configured to detect and prevent hazardous behavior, including by analyzing transactional data representing the activities of individual clinicians. For example, the security controller can apply a machine learning model that is trained to identify one or more shifts of the clinician based at least on a series of transactional data associated with the clinician. In addition, the security controller can identify one or more high-risk periods during which the clinician may engage in hazardous behavior based at least on one or more shifts associated with the clinician, the one or more high-risk periods including, for example, shift changes, portions of a shift that exceed a threshold amount of time, and / or the like. In response to identifying one or more high-risk periods, the security controller can activate a protection workflow that is configured to prevent the clinician from engaging in hazardous behavior during the high-risk period and is configured to collect evidence related to the hazardous behavior.

[0053] In some exemplary embodiments, the safety controller may also be configured to identify the first clinician as potentially engaging in a hazardous behavior by comparing a first shift associated with the first clinician to at least a second shift associated with the first clinician and / or a third shift associated with the second clinician. The safety controller may detect an anomaly indicating that the first clinician has engaged in a hazardous behavior, such as diverting and / or mishandling a substance, based at least on the comparison. In response to detecting the hazardous behavior, the safety controller may activate a protection workflow configured to prevent the clinician from continuing to engage in the hazardous behavior during the high risk period and configured to collect evidence related to the hazardous behavior.

[0054] In some exemplary embodiments, the protection workflow may include a safety controller generating and sending an alarm. The alarm may cause the device to adjust one or more functions. For example, the alarm may cause the device to present a human-perceptible indication (e.g., visual, tactile, auditory message transceiving) via a user interface, a tactile motor, or an audio speaker / amplifier. As another example, the alarm may change an operation that may be activated via the device or cause the circuit system of the device to be reconfigured to change the operating mode (e.g., lock / unlock, power off, etc.). The alarm may identify clinicians who may be involved in hazardous behaviors such as transfer and / or mishandling of materials. Alternatively and / or additionally, the protection workflow may include limiting and / or preventing clinicians from performing certain activities (e.g., picking up, applying, and / or discarding certain substances) during certain high-risk periods (e.g., shift changes, shifts exceeding a threshold amount of time, and / or the like). The protection workflow may also include a safety controller activating one or more monitoring devices (e.g., video cameras, static image cameras, audio recorders, and / or the like) at the medical device whenever a clinician accesses a medical device (e.g., a distribution cabinet, an infusion pump, a discard station, and / or the like). The protection workflow may also include the security controller configuring the medical management device to isolate substances accessed by the clinician.

[0055] Figure 1 A system diagram is depicted that illustrates an example of a security control system 100 according to some exemplary embodiments. Figure 1 , the security system 100 may include a security controller 110, a client 120, and one or more data systems 130. Figure 1As shown, the security controller 110, the client 120, and one or more data systems 130 can be connected in a communication manner via a network 140. For example, the security controller 110 can be accessed by the client 120 as a cloud-based service (e.g., software as a service (SaaS) and / or the like). Alternatively, the security controller 110 can be at least partially embedded and / or implemented in one or more data systems such as access control systems 135a, distribution systems 135b, electronic medical records (EMR) systems 135c, and / or the like. That is, the security controller 110 can be at least partially embedded and / or implemented in a medical device such as a distribution cabinet, an infusion pump, a waste station, and / or the like. Therefore, at least some functions of the security controller 110 can be accessed via one or more data systems 130. In addition, the security controller 110 can be updated and / or configured to serve and / or update portions of the corresponding data system 130. Client 120 may be a processor-based device, including, for example, a smartphone, a tablet computer, a wearable device, a desktop computer, a laptop computer, a workstation, and / or the like. In addition, network 140 may be any wired and / or wireless network, including, for example, a public land mobile network (PLMN), a local area network (LAN), a virtual local area network (VLAN), a wide area network (WAN), the Internet, and / or the like.

[0056] In some exemplary embodiments, the security controller 110 can be configured to identify shifts associated with one or more clinicians including, for example, the first clinician 150a, the second clinician 150b, and / or the like. For example, the security controller 110 can identify shifts associated with each of the first clinician 150a and the second clinician 150b based on transaction records from the one or more data systems 130. Referring again to Figure 1 , the data system 130 may include an access control system 135a, a distribution system 135b, and an electronic medical record (EMR) system 135c. While treating the first patient 160a and / or the second patient 160b, the first clinician 150a and / or the second clinician 150b may interact with the data system 130 and trigger the generation of one or more corresponding transaction records. The security controller 110 may apply a machine learning model 115 to at least a portion of the transaction records generated by the data system 130, the machine learning model being trained to at least identify a first shift associated with the first clinician 150a and / or a second shift associated with the second clinician 150b.

[0057] For example, the first clinician 150a interacting with the access control system 135a, such as by scanning an access tag at an entry point in the treatment facility, can trigger the generation of a first transaction record, the first transaction record including at least one identifier of the first clinician 150a and a timestamp indicating when the first clinician 150a interacted with the access control system 135a. The first clinician 150a dispensing a substance from a dispensing cabinet for the first patient 160a can trigger the generation of a second transaction record, the second transaction record including: a timestamp, a clinician identifier of the first clinician 150a, a device identifier of the dispensing cabinet, a patient identifier of the first patient 160a, an identifier of a substance retrieved from the dispensing cabinet, an amount of a substance retrieved from the substance cabinet, and / or the like.

[0058] Alternatively and / or additionally, the first clinician 150a's use of a disposal station to discard unused substances from the first patient 160a can trigger the generation of a third transaction record, the third transaction record comprising: a timestamp, a clinician identifier of the first clinician 150a, a clinician identifier of a second clinician 150b who witnessed the disposal, a device identifier of the disposal station, a patient identifier of the first patient 160a, an identifier of the discarded substance, an amount of the discarded substance, and / or the like. As used herein, "discarding" of a substance can refer to the disposal of a substance in accordance with institutional guidelines and / or government regulations. For example, proper disposal of a prescription pain medication may require that the controlled substance be collected in a designated container (e.g., a disposal station) in the presence of one or more witnesses.

[0059] It should be appreciated that one or more data systems 130 may also provide non-transactional data. For example, the electronic medical record system 135c may store a plurality of electronic medical records (EMRs), each of which includes a patient's history, including, for example, past opioid use and / or the like. Although a time-stamped transaction record may be generated as a result of the first clinician 150a and / or the second clinician 150b interacting with the electronic medical record system 135c to create, update, and / or retrieve an electronic medical record for the first patient 160a and / or the second patient 160b, the electronic medical record itself is not a transactional record.

[0060] To further illustrate, Figures 4 to 5 Examples of medical devices that may be part of data system 130 are depicted. For example, Figure 4An example of a dispensing cabinet 400 according to some exemplary embodiments is depicted. The dispensing of a substance from the dispensing cabinet 400 by a clinician may trigger the generation of a transaction record including: a timestamp, a clinician identifier of the clinician, a device identifier of the dispensing cabinet, a patient identifier of a patient prescribed the substance, an identifier of the substance retrieved from the dispensing cabinet, an amount of the substance retrieved from the substance cabinet, a location identifier, and / or the like. Additional details related to the dispensing cabinet 400 are described in U.S. Patent No. 8,195,328, which is commonly owned and incorporated herein by reference in its entirety.

[0061] Figure 5 An example of an infusion pump 500 according to some exemplary embodiments is depicted. The use of the infusion pump 500 by a clinician to administer a substance to a patient may trigger the generation of a transaction record including: a timestamp, a clinician identifier of the clinician, a device identifier of the infusion pump, a patient identifier of the patient receiving the substance, an identifier of the substance administered to the patient, an amount of the substance administered to the patient, a location identifier, and / or the like. Additional details related to the infusion pump 500 are described in commonly owned U.S. Patent No. 9,227,025, which is incorporated herein by reference in its entirety.

[0062] As noted, the security controller 110 can identify at least a first shift associated with the first clinician 150a and a second shift associated with the second clinician 150b based on transaction records from the data system 130. As used herein, a shift can refer to a period of time during which a clinician is on duty. It should be appreciated that a single clinician can be associated with shifts of different lengths. In addition, a single shift can include one or more inactive periods (e.g., short breaks, meetings, and / or the like) during which the clinician does not generate any transaction records. Therefore, the security controller 110 may not be able to correctly identify a shift associated with the first clinician 150a or the second clinician 150b by applying a deterministic rule that identifies a shift based on the first transaction record of the day and the last transaction record of the day.

[0063] Alternatively, according to some exemplary embodiments, the security controller 110 may apply a machine learning model 115 that may be trained to identify at least a first shift associated with a first clinician 150a and a second shift associated with a second clinician 150b based on transaction records from one or more data systems 130. For example, the machine learning model 115 may be trained to determine the start time and / or end time of at least a first shift associated with the first clinician 150a based at least on transaction records associated with the first clinician 150a. Alternatively and / or additionally, the machine learning model 115 may be trained to determine the start time and / or end time of at least a second shift associated with the second clinician 150b based at least on transaction records associated with the second clinician 150b. The machine learning model 115 may include an input layer for receiving a vector of values ​​representing transaction records. The machine learning model 115 may include an output layer for providing one or more output values ​​identifying shift information. The output value may include a probability or accuracy value for the confidence level of identifying a given shift information output. For example, the machine learning model 115 may receive an encoded version of a transaction from a previous time period. The output layer of the machine learning model 115 may include twenty-four output values, each representing an hour of the day. Each of the twenty-four output values ​​may be associated with a value from 0 to 1, where 0 is the least likely shift boundary and 1 is a nearly certain shift boundary. The machine learning model 115 may include an additional layer connecting the input layer to the output layer. The additional layer may be referred to as a hidden layer and includes weights that can be trained using historical transaction and shift data to identify shift boundaries for subsequently received transactions.

[0064] In some exemplary embodiments, the machine learning model 115 can be a probabilistic machine learning model that is trained to determine the probability that the clinician is on duty or off duty based on a series of transaction records related to the clinician. For example, the machine learning model 115 can be a hidden Markov model. In addition, the machine learning model 115 can be trained using reinforcement learning techniques that train the machine learning model 115 to maximize the cumulative reward. Examples of reinforcement learning techniques include: Q learning, Monte Carlo, state-action-reward-state-action (SARSA), deep Q network (DQN), deep deterministic policy gradient (DDPG), asynchronous actor-critic algorithm (A3C), trust region policy optimization (TRPO), proximal policy optimization (PPO) and / or the like.

[0065] The machine learning model 115 may be defined by a tuple (S, A), where S may represent a finite set of states {S1, S2}, where a first state S1 corresponds to the clinician being on duty and a second state S2 corresponds to the clinician being off duty. In addition, A may represent a finite set of actions that the clinician can perform {A1, A2, ..., A n That is, the finite action set {A1, A2, ..., A n} may correspond to different types of transaction records from the data system 130. For example, A may include a first action A1 corresponding to a clinician scanning an access tag at an entry point to a treatment facility, a second action A2 corresponding to a clinician accessing a dispensing cabinet to retrieve a substance, and a third action A3 corresponding to the clinician administering the substance to a patient.

[0066] When the clinician is in one of the two states {S1, S2}, the clinician performs an action {A1, A2, ..., A n} can trigger a state transition. For example, when the clinician is in the first state S1 corresponding to the clinician being at work, the clinician performs actions {A1, A2, ..., A n} can trigger a state transition to a first state S1 corresponding to the clinician still being on duty or a second state S2 corresponding to the clinician being off duty. Alternatively and / or additionally, when the clinician is in the second state S2 corresponding to the clinician being off duty, the clinician performs actions {A1, A2, ..., A n} can trigger a state transition to a second state S2 corresponding to the clinician still off duty or a first state S1 corresponding to the clinician on duty. It should be understood that a state transition from the second state S2 to the first state S1 (corresponding to the clinician switching from off duty to on duty) can indicate the start of the clinician's shift. In contrast, a state transition from the second state S2 to the first state S1 (corresponding to the clinician switching from off duty to on duty) can indicate the start of the clinician's shift.

[0067] As noted, the machine learning model 115 may be trained using reinforcement learning techniques. For example, the machine learning model 115 may be trained using Q-learning, which is an example of a reinforcement learning technique. The quality Q of the combination of actions performed on the fields of the first software application 145B may be expressed as the following equation (1):

[0068]

[0069] Before starting the training of the machine learning model 115, the value Q can be initialized to an arbitrary value. The training of the machine learning model 115 can be an iterative process of updating the value Q. For example, at each time t, an action a can be selected. t To trigger to the new state st+1 The corresponding reward r can be observed t , and the value Q can be updated. The iterative update of the value Q can be expressed by the following equation (2):

[0070] Q new (s t , a t )←(1-α)·Q(s t , a t )+α·(r t +γ·max a Q(s t+1 , a)) (2)

[0071] Among them, r t It can be expressed for the current state s t observed reward, α can represent the learning rate (e.g., 0≤α≤1), γ can represent the reduction factor, Q(s t , a t ) can represent the value Q from the previous iteration, and max a Q(s t+1 , a) may correspond to the estimated optimal future value of Q. Referring to equation (2), it should be understood that (r t +γ·max a Q(s t+1 , a)) can correspond to the learning value. In addition, the reward r t may correspond to the cumulative reward associated with all state transitions up to time t.

[0072] In some exemplary embodiments, the machine learning model 115 can be trained based on training data, the training data including a series of transaction records that form a ground truth shift. For example, the ground truth shift can be an actual shift associated with the first clinician 150a and / or the second clinician 150b. Alternatively and / or additionally, the ground truth shift can be synthesized based on actual shifts associated with the first clinician 150a and / or the second clinician 150b and additional data including, for example, medical protocols, generalized observations, and / or the like.

[0073] In some exemplary embodiments, training of the machine learning model 115 may be performed offline and / or in real time. For example, the safety controller 110 may detect a deviation between the output of the machine learning model 115 and a transaction record received from the data system 130. An example of a deviation may include the data system 130 generating a transaction record that is inconsistent with the output of the machine learning model 115, the output of the machine learning model 115 indicating the start or end of a shift associated with the first clinician 150a and / or the second clinician 150b. The safety controller 110 may respond to the deviation by at least triggering a retraining of the machine learning model 115. For example, the machine learning model 115 may be retrained offline or in real time based on additional training data, the additional training data including a baseline truth shift adjusted to include the transaction record. In instances where the security controller 110 is at least partially embedded and / or implemented within one or more data systems 130, the retrained machine learning model 115 may be propagated to one or more data systems 130 (e.g., an access control system 135a, a distribution system 135b, an electronic medical record (EMR) system 135c, and / or the like) as part of routine service and / or software updates at the one or more data systems 130.

[0074] In some exemplary embodiments, the safety controller 110 can be configured to detect one or more high-risk periods based on one or more shifts associated with the first clinician 150a and / or the second clinician 150b, during which the first clinician 150a and / or the second clinician 150b may engage in hazardous behaviors including, for example, transfer and / or mishandling of materials. Examples of high-risk periods can include shift changes (e.g., n time amounts before the end of a shift and / or n time amounts after the start of a shift), portions of a shift that exceed a threshold time amount, and / or the like. The safety controller 110 can be configured to activate a protective workflow in response to detecting one or more high-risk periods. For example, the safety controller 110 can activate a protective workflow during one or more high-risk periods.

[0075] The protection workflow can be configured to prevent the first clinician 150a and / or the second clinician 150b from engaging in hazardous behavior during the high-risk period. For example, the protection workflow can include: the security controller 110 generates one or more electronic alerts (e.g., emails, push notifications, short message service (SMS) messages, and / or the like) and sends the electronic alerts to the client 120 associated with the facility monitor, the electronic alerts identifying the first clinician 150a and / or the second clinician 150b as potentially engaging in hazardous behavior during the high-risk period.

[0076] Alternatively and / or additionally, the protection workflow may include preventing the first clinician 150a and / or the second clinician 150b from performing certain activities (e.g., picking up, administering, and / or discarding certain substances) during high-risk periods. For example, the protection workflow may include: the security controller 110 configures one or more medical devices (e.g., a dispensing cabinet, an infusion pump, a discarding station, and / or the like) to limit and / or deny interaction with certain substances. The security controller 110 may configure the medical device to prevent the first clinician 150a and / or the second clinician 150b from picking up and / or administering certain substances without the authorization of the third clinician 150c. For example, as part of the protection workflow, the dispensing cabinet may be configured to prevent the first clinician 150a and / or the second clinician 150b from accessing a container containing a prescription analgesic without the authorization of the third clinician 150c.

[0077] In some exemplary embodiments, the protection workflow can be configured to collect evidence of hazardous behavior. For example, the protection workflow can include: the security controller 110 activates one or more monitoring devices (e.g., video cameras, still image cameras, audio recorders, and / or the like) at the medical device whenever the first clinician 150a and / or the second clinician 150b access the medical device (e.g., a distribution cabinet, an infusion pump, a waste station, and / or the like). Alternatively and / or additionally, the protection workflow can include: the security controller 110 configures the medical device to isolate the substances accessed by the first clinician 150a and / or the second clinician 150b.

[0078] In some exemplary embodiments, the safety controller 110 can also be configured to: identify the first clinician 150a as possibly having engaged in a hazardous behavior by comparing at least a first shift associated with the first clinician 150a and at least a second shift associated with the first clinician 150a and / or a third shift of the second clinician 150b. It should be appreciated that the safety controller 110 can selectively compare two or more shifts. That is, the safety controller 110 can compare two or more shifts that are similar due to one or more common attributes, and the one or more common attributes include, for example, duration, time of day, institution, department, patient, and / or the like. In addition, the safety controller 110 can compare shifts associated with different clinicians, and the clinicians have the same and / or similar user types, and the user types include, for example, doctors, pharmacists, nurse managers, nurse practitioners, and / or the like.

[0079] The safety controller 110 can detect the occurrence of a hazardous behavior that occurred during a first shift associated with the first clinician 150a by comparing at least the transaction records included in the first shift associated with the first clinician 150a with the transaction records included in the second shift associated with the first clinician 150a and / or the third shift associated with the second clinician 150b. In doing so, the safety controller 110 can detect the presence of one or more anomalies including one or more abnormal transaction records. The presence of anomalies in the first shift associated with the first clinician 150a can indicate that the first clinician 150a has engaged in hazardous behavior including, for example, diverting and / or mishandling substances.

[0080] The safety controller 110 may activate a protection workflow in response to identifying the first clinician 150a as possibly having engaged in a hazardous behavior. In some exemplary embodiments, the protection workflow may be configured to prevent future occurrences of the hazardous behavior and / or to collect evidence associated with the hazardous behavior. For example, the protection workflow may include the safety controller 110 generating and sending one or more electronic alerts (e.g., emails, push notifications, short message service (SMS) messages, and / or the like) identifying the first clinician 150a as possibly having engaged in a hazardous behavior. The one or more electronic alerts may be sent to a client 120 associated with an institutional monitor.

[0081] In some exemplary embodiments, the protection workflow may also include: the security controller 110 activates one or more monitoring devices (e.g., video cameras, still image cameras, audio recorders, and / or the like) at the medical device whenever the first clinician 150a accesses the medical device (e.g., a dispensing cabinet, an infusion pump, a waste station, and / or the like). Alternatively and / or additionally, the security controller 110 may configure the medical device to isolate substances taken by the first clinician 150a. For example, the security controller 110 may activate a monitoring device at the waste station in response to the first clinician 150a accessing the waste station to dispose of unused substances. In addition, the security controller 110 may configure the waste station to isolate unused substances returned by the first clinician 150a, for example, into a designated container that is separate from containers that hold unused substances returned by other clinicians.

[0082] Figure 2A A flow chart is depicted that illustrates another example of a process 200 for machine learning-based security control according to some exemplary embodiments. Figure 1 and Figure 2A , the process 200 may be performed by the safety control system 100 , for example, by the safety controller 110 .

[0083] At 202, the security controller 110 can identify shifts associated with the clinician by at least applying a machine learning model that is trained to identify one or more shifts associated with the first clinician based at least on a series of transaction records associated with the clinician. For example, the security controller 110 can apply the machine learning model 115 to identify one or more shifts associated with the first clinician 150a. As noted, the machine learning model 115 can be trained to identify one or more shifts associated with the first clinician 150a based at least on a series of transaction records associated with the first clinician 150a.

[0084] In some exemplary embodiments, each transaction record in a series of transaction records can be generated in response to the first clinician 150a interacting with one or more data systems 130, which can include, for example, an access control system 135a, a distribution system 135b, and / or an electronic medical record system 135c. For example, a first transaction record can be generated in response to the first clinician 150a interacting with the access control system 135a by scanning an access tag at an entry point in the treatment facility. In addition, a second transaction record can be generated in response to the first clinician 150a interacting with the distribution system 135b to retrieve, apply, and / or discard a substance. Alternatively and / or additionally, a third transaction record can be generated in response to the first clinician 150a interacting with the electronic medical record (EMR) system 135c when applying a substance to a patient. Therefore, the machine learning model 115 can be trained to recognize one or more transaction records indicating that the first clinician 150a is at the beginning of a shift and / or the end of a shift.

[0085] At 204, the safety controller 110 may identify one or more high-risk periods based at least on a shift associated with the clinician. For example, the safety controller 101 may identify one or more high-risk periods based at least on a shift associated with the first clinician 150a, during which the first clinician 150a may engage in hazardous behavior, such as diverting and / or mishandling a substance. Examples of high-risk periods may include shift changes (e.g., n amount of time before the end of a shift and / or n amount of time after the start of a shift), portions of a shift exceeding a threshold amount of time, and / or the like.

[0086] At 206, the security controller 110 may activate a protection workflow during one or more high-risk periods. In some exemplary embodiments, the security controller 110 may activate a protection workflow configured to prevent the first clinician 150a from engaging in hazardous behavior and / or collecting evidence related to the hazardous behavior during the high-risk period. For example, the protection workflow may include: the security controller 110 generates one or more electronic alerts (e.g., emails, push notifications, short message service (SMS) messages, and / or the like) and sends an electronic alert to a client 120 associated with an institutional monitor, the electronic alert identifying the first clinician 150a as possibly engaging in hazardous behavior during the high-risk period. The protection workflow may also include: the security controller 110 configures one or more medical devices (e.g., dispensing cabinets, infusion pumps, waste stations, and / or the like) to restrict and / or deny interaction with certain substances. For example, the security controller 110 may configure the medical device to prevent the first clinician 150a from retrieving and / or administering certain substances without additional authorization (e.g., from a third clinician 150c).

[0087] Alternatively and / or additionally, the protection workflow may also include: the security controller 110 activates one or more monitoring devices (e.g., video cameras, still image cameras, audio recorders, and / or the like) at the medical device whenever the first clinician 150a accesses the medical device (e.g., a dispensing cabinet, an infusion pump, a waste station, and / or the like). The protection workflow may also include: the security controller 110 configures the medical device to isolate the substances accessed by the first clinician 150a.

[0088] Figure 2B A flow chart is depicted that illustrates another example of a process 250 for machine learning-based security control according to some exemplary embodiments. Figure 1 and Figure 2B , process 250 may be performed by the safety control system 100 , for example, by the safety controller 110 .

[0089] At 252, the security controller 110 may identify a first shift associated with the first clinician by applying at least a machine learning model, the machine learning model being trained to identify one or more shifts associated with the first clinician based at least on a series of transaction records associated with the first clinician. In some exemplary embodiments, the security controller 110 may apply the machine learning model 115 to identify one or more shifts associated with the first clinician 150a. As noted, the machine learning model 115 may be trained to identify one or more shifts associated with the first clinician 150a based at least on a series of transaction records associated with the first clinician 150a.

[0090] At 254, the security controller 110 may compare a first shift associated with the first clinician with a second shift associated with the first clinician and / or a third shift associated with the second clinician. For example, the security controller 110 may compare transaction records included in the first shift associated with the first clinician 150a with transaction records included in the second shift associated with the first clinician 150a and / or the third shift associated with the second clinician 150b. As noted, the security controller 110 may detect the occurrence of abnormal behavior in the shift by at least comparing a shift with similar shifts associated with the same clinician and / or different clinicians. Therefore, the first shift associated with the first clinician 150a, the second shift associated with the first clinician 150a, and / or the third shift associated with the second clinician 150b may share at least one common attribute, including, for example, duration, time of day, institution, department, patient, and / or the like.

[0091] At 256, the security controller 110 can detect the occurrence of abnormal behavior during the first shift associated with the first clinician based on at least a comparison between the first shift associated with the first clinician, the second shift associated with the first clinician, and / or the third shift associated with the second clinician (e.g., according to 204). In some exemplary embodiments, the security controller 110 can compare the transaction records included in the first shift associated with the first clinician 150a and the transaction records included in the second shift associated with the first clinician 150a and / or the third shift associated with the second clinician 150b to identify one or more abnormal transaction records. An abnormal transaction can be a transaction that occurs in one shift but not in another shift. An abnormal transaction can be a transaction that occurs at a given time in one shift and at a different time in another shift. Aggregation can be used to identify abnormal transactions. For example, in a typical shift, a clinician may perform a certain type of transaction (e.g., requesting a substance to be dispensed) ten times, but during different shifts, the type of transaction may occur more frequently or less frequently. Exceptions may be designated using a threshold that allows for small differences between transaction attributes (e.g., differences in transaction time or dispense requests) but flags differences that exceed the threshold. The presence of one or more abnormal transaction records in the first shift associated with the first clinician 150a may indicate that the first clinician 150a has engaged in hazardous behavior including, for example, diversion and / or mishandling of materials.

[0092] At 258, the safety controller 110 may activate a protection workflow in response to detecting an occurrence of abnormal behavior during a first shift associated with the first clinician. For example, the safety controller 110 may determine that the first clinician 150a may have engaged in hazardous behavior, such as transferring and / or mishandling materials. Therefore, the safety controller 110 may activate a protection workflow, which is configured to prevent future occurrences of hazardous behavior and / or collect evidence related to the hazardous behavior. For example, the protection workflow may include the safety controller 110 generating an electronic alert, which includes, for example, a wireless alert message, such as a push notification, a short message service (SMS) message, and / or the like. The electronic alert can be sent to a client 120 that can be associated with a monitor, who is associated with the first clinician 150a.

[0093] In some exemplary embodiments, the protection workflow may also include: the security controller 110 configures one or more medical devices (e.g., dispensing cabinets, infusion pumps, waste stations, and / or the like) to limit and / or deny interaction with certain substances, such as prescription medications, high-value medical supplies, and / or the like. For example, the security controller 110 may configure the medical device to prevent the first clinician 150a from retrieving and / or administering certain substances without additional authorization (e.g., from the third clinician 150c).

[0094] Alternatively and / or additionally, the protection workflow may also include: the security controller 110 activates one or more monitoring devices (e.g., video cameras, still image cameras, audio recorders, and / or the like) at the medical device whenever the first clinician 150a accesses the medical device (e.g., a dispensing cabinet, an infusion pump, a waste station, and / or the like). The protection workflow may also include: the security controller 110 configures the medical device to isolate the substances accessed by the first clinician 150a.

[0095] Figure 3 A block diagram is depicted that illustrates a computing system 300 consistent with an implementation of the current subject matter. Figure 1 2 , computing system 300 may be used to implement security controller 110 and / or any components thereof.

[0096] like Figure 3 As shown in , the computing system 300 may include: a processor 310, a memory 320, a storage device 330, and an input / output device 340. The processor 310, the memory 320, the storage device 330, and the input / output device 340 may be interconnected via a system bus 350. The processor 310 is capable of processing instructions for execution within the computing system 300. Such executed instructions may execute, for example, one or more components of the security controller 110. In some exemplary embodiments, the processor 310 may be a single-threaded processor. Alternatively, the processor 310 may be a multi-threaded processor. The processor 310 is capable of processing instructions stored in the memory 320 and / or on the storage device 330 to display graphical information of a user interface provided via the input / output device 340.

[0097] Memory 320 is a computer-readable medium, such as a volatile or non-volatile medium, that stores information within computing system 300. For example, memory 320 may store a data structure representing a configuration object database. Storage device 330 can provide permanent storage for computing system 300. Storage device 330 may be a floppy disk device, a hard disk device, an optical disk device, a magnetic tape device, a solid-state device, and / or any other suitable permanent storage member. Input / output device 340 provides input / output operations for computing system 300. In some exemplary embodiments, input / output device 340 includes a keyboard and / or a pointing device. In various embodiments, input / output device 340 includes a display unit for displaying a graphical user interface.

[0098] According to some exemplary embodiments, the input / output device 340 can provide input / output operations for the network device. For example, the input / output device 340 can include an Ethernet port or other networking port to communicate with one or more wired and / or wireless networks (e.g., a local area network (LAN), a wide area network (WAN), the Internet).

[0099] In some exemplary embodiments, computing system 300 can be used to perform various interactive computer software applications, which can be used for the organization, analysis and / or storage of data in various formats. Alternatively, computing system 300 can be used to perform any type of software application. These applications can be used to perform various functions, such as planning functions (e.g., generation, management, editing of electronic data sheet documents, word processing documents and / or any other objects, etc.), computing functions, communication functions, etc. The application can include various add-in functions, or can be independent computing products and / or functions. After activation in the application, the function can be used to generate a user interface provided via input / output device 340. The user interface can be generated by computing system 300 (e.g., on a computer screen monitor, etc.) and presented to the user.

[0100] One or more aspects or features of the subject matter described herein can be implemented in digital electronic circuits, integrated circuits, specially designed ASICs, field programmable gate arrays (FPGA) computer hardware, firmware, software and / or combinations thereof. These various aspects or features can include embodiments in one or more computer programs that can be executed and / or interpreted on a programmable system, the programmable system including at least one programmable processor that can be dedicated or general, the processor being connected to receive data and instructions from a storage system, at least one input device and at least one output device and to transmit data and instructions to them. A programmable system or computing system can include a client and a server. The client and the server are usually far away from each other and usually interact through a communication network. The relationship between the client and the server is generated by means of a computer program that runs on a respective computer and has a client-server relationship with each other.

[0101] These computer programs, which may also be referred to as programs, software, software applications, applications, components or codes, include machine instructions for programmable processors and may be implemented in high-level procedural and / or object-oriented programming languages ​​and / or in assembly / machine languages. As used herein, the term "machine-readable medium" refers to any computer program product, device and / or device, such as a disk, an optical disk, a memory and a programmable logic device (PLD), for providing machine instructions and / or data to a programmable processor, including a machine-readable medium for receiving machine instructions as machine-readable signals. The term "machine-readable signal" refers to any signal for providing machine instructions and / or data to a programmable processor. A machine-readable medium may, for example, store such machine instructions non-volatilely, such as a non-volatile solid-state memory or a magnetic hard drive or any equivalent storage medium. A machine-readable medium may alternatively or additionally store such machine instructions in a volatile manner, such as a processor cache or other random access memory associated with one or more physical processor cores.

[0102] To provide interaction with a user, one or more aspects or features of the subject matter described herein may be implemented on a computer having a display device, such as a cathode ray tube (CRT) or liquid crystal display (LCD) or light emitting diode (LED) monitor for displaying information to a user, and a keyboard and pointing device, such as a mouse or trackball, whereby the user can provide input to the computer. Other types of devices may also be used to provide interaction with a user. For example, the feedback provided to the user may be any form of sensory feedback, such as visual feedback, auditory feedback, or tactile feedback; and any form may be used to receive input from the user, including acoustic, voice, or tactile input. Other possible input devices include touch screens or other touch-sensitive devices such as single-point or multi-point resistive or capacitive tracking pads, voice recognition hardware and software, optical scanners, optical pointers, digital image capture devices and related interpretation software, and the like.

[0103] In the above description and in the claims, phrases such as "at least one of..." or "one or more of..." may appear, followed by a combined list of elements or features. The term "and / or" may also appear in a list of two or more elements or features. Unless otherwise implied or explicitly contradicted by the context in which it is used, such phrases are intended to mean any one of the elements or features listed individually, or any one of the recorded elements or features in combination with other recorded elements or features. For example, the phrases "at least one of A and B"; "one or more of A and B"; and "A and / or B" each mean "A alone, B alone, or A and B together". A similar interpretation is also intended for lists including three or more items. For example, the phrases "at least one of A, B, and C"; "one or more of A, B, and C"; and "A, B, and / or C" each mean "A alone, B alone, C alone, A and B together, A and C together, B and C together, or A, B and C together". In the above and in the claims, use of the term "based on" is intended to mean "based, at least in part, on," such that non-recited features or elements are also allowed.

[0104] As used herein, the term "determine" or "determines" encompasses a wide variety of actions. For example, "determining" may include calculating, computing, processing, deriving, generating, obtaining, searching (e.g., searching in a table, database, or other data structure), confirming, etc. via a hardware element without user intervention. In addition, "determining" may include receiving (e.g., receiving information), accessing (e.g., accessing data in a memory), etc. via a hardware element without user intervention. "Determining" may include distinguishing, selecting, choosing, establishing, etc. via a hardware element without user intervention.

[0105] As used herein, the term "providing" or "provided with" encompasses a wide variety of actions. For example, "providing" may include storing a value in a location of a storage device for subsequent retrieval, transmitting a value directly to a recipient via at least one wired or wireless communication medium, transmitting or storing a reference to a value, etc. "Providing" may also include encoding, decoding, encrypting, decrypting, verifying, checking, etc. via a hardware element.

[0106] As used herein, the term "message" encompasses a wide variety of formats for conveying (e.g., transmitting or receiving) information. A message may include a machine-readable aggregation of information such as an XML file, a fixed field message, a comma-delimited message, etc. In some embodiments, a message may include a signal for conveying one or more representations of information. Although described in the singular, it should be understood that a message may be combined, transmitted, stored, received, etc. in multiple parts.

[0107] As used herein, the term "selectively" or "selective" can encompass a wide variety of actions. For example, a "selective" process can include determining an option from a plurality of options. A "selective" process can include one or more of the following: dynamically determined inputs, preconfigured inputs, or user-initiated inputs to make the determination. In some embodiments, an n-input switch can be included to provide selective functionality where n is the number of inputs used to make a selection.

[0108] As used herein, the term "corresponding" or "corresponding" encompasses a structural, functional, quantitative and / or qualitative correlation or relationship between two or more objects, data sets, information and / or the like, preferably a correspondence or relationship that can be used to transform one or more of the two or more objects, data sets, information and / or the like to appear to be the same or equivalent. The correspondence can be evaluated using one or more of a threshold, a value range, fuzzy logic, pattern matching, a machine learning evaluation model, or a combination thereof.

[0109] In any embodiment, data can be forwarded to a "remote" device or location, where "remote" means a location or device outside the location or device where the program is executed. For example, the remote location can be another location in the same city (e.g., an office, a laboratory, etc.), another location in a different city, another location in a different state, another location in a different country, etc. Thus, when a project is indicated as being at the "remote" location of another project, this means that the two projects can be in the same room but separate, or at least in different rooms or different buildings, and can be separated by at least one mile, ten miles, or at least one hundred miles. "Transmitting" information refers to transmitting the data representing the information as an electrical signal through a suitable communication channel (e.g., a private network or a public network). "Forwarding" a project refers to any method that allows the project to be transported from one location to the next location, whether physically or otherwise (where possible), and at least in the case of data, includes physically transporting the medium carrying the data or transmitting the data. Examples of transmission media include radio or infrared transmission channels and network connections to another computer or networking device and the Internet, or include email transmissions and information recorded on websites, etc.

[0110] One or more aspects of the described artificial intelligence may be implemented in whole or in part by a model. The model may be implemented as a machine learning model. The learning may be supervised, unsupervised, enhanced, or hybrid learning, whereby a plurality of learning techniques are used to generate the model. The learning may be performed as part of training. Training the model may include obtaining a training data set and adjusting the characteristics of the model to obtain a desired model output. For example, three characteristics may be associated with a desired device state. In such instances, training may include receiving three characteristics as input to the model and adjusting the characteristics of the model so that for each of the three characteristic sets, the output device state matches the desired device state associated with the historical data.

[0111] In some embodiments, training can be dynamic. For example, the system can use a set of events to update the model. Detectable characteristics from the events can be used to adjust the model.

[0112] The model may be an equation, an artificial neural network, a recursive neural network, a convolutional neural network, a decision tree, or other machine-readable artificial intelligence structure. The structural characteristics available for adjustment during training may vary based on the selected model. For example, if a neural network is the selected model, the characteristics may include input elements, network layers, node density, node activation thresholds, weights between nodes, input or output value weights, etc. If the model is implemented as an equation (e.g., regression), the characteristics may include weights of input parameters, thresholds or limits for evaluating output values, or criteria for selecting from a set of equations.

[0113] After model training, retraining can be included to optimize or update the model to reflect additional data or specific operating conditions. The retraining can be based on one or more signals detected by the apparatus described herein or as part of the methods described herein. Upon detection of a specified signal, the system can activate a training process to adjust the model, as described.

[0114] Additional examples of machine learning and modeling features that may be included in the implementations discussed above are described in “A survey of machine learning for big data processing” by Qiu et al., EURASIP Journal on Advances in Signal Processing (2016), which is incorporated herein by reference in its entirety.

[0115] As used herein, "user interface" (also referred to as interactive user interface, graphical user interface or UI) may refer to a network-based interface that includes data fields and / or other control elements for receiving input signals or providing electronic information and / or for providing information to a user in response to any received input signals. Control elements may include dials, buttons, icons, selectable areas or other perceptible marks presented via the UI, and when interacting with the UI (e.g., clicking, touching, selecting, etc.), the UI initiates data exchange for the device presenting the UI. The UI may be implemented in whole or in part using technologies such as hypertext markup language (HTML), FLASHTM, JAVATM, NETTM, web services or rich site summaries (RSS). In some embodiments, the UI may be included in an independent client (e.g., a fat client, a fat client) configured to communicate (e.g., send or receive data) according to one or more aspects of the description. Communication may be to or from a medical device, a diagnostic device, a monitoring device, or a server communicating with them.

[0116] Depending on the desired configuration, the subject matter described herein may be embodied in systems, devices, methods and / or articles. The embodiments described in the foregoing description do not represent all embodiments that conform to the subject matter described herein. Instead, these embodiments are only some examples that conform to aspects related to the described subject matter. Although several variations have been described in detail above, other modifications or additions are also possible. Specifically, in addition to those described herein, additional features and / or variations may also be provided. For example, the embodiments described above may relate to various combinations and subcombinations of the disclosed features and / or combinations and subcombinations of several additional features disclosed above. In addition, the logic flows depicted in the accompanying drawings and / or described herein do not necessarily require the specific order or sequence shown to achieve the desired results. Other embodiments may be within the scope of the appended claims.

Claims

1. A safety control system based on machine learning, comprising: at least one data processor; as well as at least one memory storing instructions that, when executed by at least one data processor, result in operations including: identifying a first shift associated with a first clinician by applying at least a machine learning model, the machine learning model comprising a probabilistic machine learning model trained to identify a probability that the first clinician is on duty or off duty and one or more shifts associated with the first clinician based at least on a series of transaction records associated with the first clinician; determining that the first clinician may have engaged in the compromising behavior based on at least a first shift identified as being associated with the first clinician; as well as A protection workflow is activated in response to determining that the first clinician may be engaging in hazardous behavior, the protection workflow comprising at least one of configuring a waste station, generating an alarm, or activating a monitoring device.

2. The machine learning-based safety control system according to claim 1, further comprising: comparing a first shift associated with the first clinician to a second shift associated with the first clinician and / or a third shift associated with the second clinician; as well as A determination is made based at least on the comparison that the first clinician may be engaging in hazardous behavior.

3. The machine learning-based safety control system according to claim 2, wherein: The first shift is compared to the second shift and / or the third shift by comparing at least a first plurality of transaction records included in the first shift associated with the first clinician and a second plurality of transaction records included in the second shift associated with the first clinician and / or the third shift associated with the second clinician.

4. The machine learning-based safety control system according to claim 3, wherein: In response to detecting one or more anomalous transaction records included in the first plurality of transaction records, it is determined that the first clinician may have engaged in compromised behavior.

5. The machine learning-based safety control system according to claim 2, wherein: The first shift is compared to the second shift and / or the third shift based at least on the first shift associated with the first clinician having one or more attributes in common with the second shift associated with the first clinician and / or the third shift associated with the second clinician.

6. The machine learning-based safety control system according to claim 2, wherein: A first shift associated with the first clinician and a third shift associated with the second clinician are compared based at least on the first clinician and the second clinician having one or more attributes in common.

7. The machine learning-based safety control system according to claim 1, further comprising: identifying, based on at least a first shift associated with the first clinician, one or more high-risk time periods during which the first clinician may engage in compromising behavior; as well as A protection workflow is activated during one or more high-risk periods.

8. The machine learning-based safety control system according to claim 7, wherein: The one or more high risk periods include at least one of a first amount of time after a start of a first shift and a second amount of time before an end of the first shift.

9. The machine learning-based safety control system according to claim 7, wherein: The one or more high risk periods include portions of the first shift that exceed a threshold amount of time.

10. The machine learning-based safety control system according to claim 1, wherein: A first transition from the off-duty state to the on-duty state corresponds to the beginning of a first shift associated with the first clinician, and wherein a second transition from the on-duty state to the off-duty state corresponds to the end of the first shift associated with the first clinician.

11. The machine learning-based safety control system according to claim 1, wherein: Probabilistic machine learning models are trained using reinforcement learning techniques including: Q-learning, Monte Carlo, state-action-reward-state-action, deep Q-networks, deep deterministic policy gradients, asynchronous actor-critic algorithm, trust region policy optimization, and / or proximal policy optimization.

12. The machine learning-based safety control system according to claim 1, wherein: The series of transaction records includes one or more transaction records generated in response to a first clinician interacting with the access control system, and wherein the transaction record indicates scanning of an access token at an entry point to the treatment facility via the access control system.

13. The machine learning-based safety control system according to claim 1, wherein: The series of transaction records includes one or more transaction records generated in response to a first clinician interacting with a dispensing system, and wherein the transaction record indicates accessing a dispensing cabinet associated with the dispensing system to retrieve a substance.

14. The machine learning based safety control system according to claim 1, wherein: The series of transaction records includes one or more transaction records generated in response to a first clinician interacting with the electronic medical record system, and wherein the transaction record indicates administration of a substance to a patient identified in the electronic medical record system.

15. The machine learning based safety control system according to claim 1, wherein: The protection workflow includes generating an alert that identifies a first clinician as potentially engaging in compromising behavior.

16. The machine learning based safety control system according to claim 1, wherein: The protection workflow includes activating one or more monitoring devices in response to the first clinician interacting with the medical device and / or isolating a substance accessed by the first clinician.

17. The machine learning based safety control system according to claim 1, wherein: The protected workflow includes configuring one or more medical devices to prevent a first clinician from retrieving, administering, and / or discarding a substance without authorization by a second clinician.

18. A safety control method based on machine learning, comprising: identifying a first shift associated with a first clinician by applying at least a machine learning model, the machine learning model comprising a probabilistic machine learning model trained to identify a probability that the first clinician is on duty or off duty and one or more shifts associated with the first clinician based at least on a series of transaction records associated with the first clinician; determining that the first clinician may have engaged in the compromising behavior based on at least a first shift identified as being associated with the first clinician; as well as A protection workflow is activated in response to determining that the first clinician may be engaging in hazardous behavior, the protection workflow comprising at least one of configuring a waste station, generating an alarm, or activating a monitoring device.

19. The safety control method based on machine learning according to claim 18, further comprising: comparing a first shift associated with the first clinician to a second shift associated with the first clinician and / or a third shift associated with the second clinician; as well as A determination is made based at least on the comparison that the first clinician may be engaging in hazardous behavior.

20. The machine learning-based safety control method according to claim 19, wherein: The first shift is compared to the second shift and / or the third shift by comparing at least a first plurality of transaction records included in the first shift associated with the first clinician and a second plurality of transaction records included in the second shift associated with the first clinician and / or the third shift associated with the second clinician.

21. The safety control method based on machine learning according to claim 20, wherein: In response to detecting one or more anomalous transaction records included in the first plurality of transaction records, it is determined that the first clinician may have engaged in compromised behavior.

22. The machine learning-based safety control method according to claim 19, wherein: The first shift is compared to the second shift and / or the third shift based at least on the first shift associated with the first clinician having one or more attributes in common with the second shift associated with the first clinician and / or the third shift associated with the second clinician.

23. The safety control method based on machine learning according to claim 19, wherein: A first shift associated with the first clinician and a third shift associated with the second clinician are compared based at least on the first clinician and the second clinician having one or more attributes in common.

24. The machine learning-based security control method according to claim 18, further comprising: identifying, based on at least a first shift associated with the first clinician, one or more high-risk time periods during which the first clinician may engage in compromising behavior; as well as A protection workflow is activated during one or more high risk periods.

25. The machine learning-based security control method according to claim 24, wherein: The one or more high risk periods include at least one of a first amount of time after a start of a first shift and a second amount of time before an end of the first shift.

26. The machine learning-based security control method according to claim 24, wherein: The one or more high risk periods include portions of the first shift that exceed a threshold amount of time.

27. The machine learning-based security control method according to claim 18, wherein: A first transition from the off-duty state to the on-duty state corresponds to the beginning of a first shift associated with the first clinician, and wherein a second transition from the on-duty state to the off-duty state corresponds to the end of the first shift associated with the first clinician.

28. The machine learning-based security control method according to claim 18, wherein: Probabilistic machine learning models are trained using reinforcement learning techniques including: Q-learning, Monte Carlo, state-action-reward-state-action, deep Q-networks, deep deterministic policy gradients, asynchronous actor-critic algorithm, trust region policy optimization, and / or proximal policy optimization.

29. The machine learning-based security control method according to claim 18, wherein: The series of transaction records includes one or more transaction records generated in response to a first clinician interacting with the access control system, and wherein the transaction record indicates scanning of an access token at an entry point to the treatment facility via the access control system.

30. The machine learning-based security control method according to claim 18, wherein: The series of transaction records includes one or more transaction records generated in response to a first clinician interacting with a dispensing system, and wherein the transaction record indicates accessing a dispensing cabinet associated with the dispensing system to retrieve a substance.

31. The machine learning-based security control method according to claim 18, wherein: The series of transaction records includes one or more transaction records generated in response to a first clinician interacting with the electronic medical record system, and wherein the transaction record indicates administration of a substance to a patient identified in the electronic medical record system.

32. The machine learning-based security control method according to claim 18, wherein: The protection workflow includes generating an alert that identifies a first clinician as potentially engaging in compromising behavior.

33. The machine learning-based security control method according to claim 18, wherein: The protection workflow includes activating one or more monitoring devices in response to the first clinician interacting with the medical device and / or isolating a substance accessed by the first clinician.

34. The machine learning-based security control method according to claim 18, wherein: The protected workflow includes configuring one or more medical devices to prevent a first clinician from retrieving, administering, and / or discarding a substance without authorization by a second clinician.

35. A non-transitory computer-readable medium storing instructions that, when executed by at least one data processor, result in operations comprising: identifying a shift associated with the clinician by applying at least a machine learning model, the machine learning model comprising a probabilistic machine learning model trained to identify a probability of the clinician being on duty or off duty and one or more shifts associated with the clinician based at least on a series of transaction records associated with the clinician; Determine that the clinician may have engaged in the compromising behavior based at least on the identified shifts associated with the clinician; and A protection workflow is activated in response to determining that the clinician may be engaging in hazardous behavior, the protection workflow comprising at least one of configuring a waste station, generating an alarm, or activating a monitoring device.

36. A safety control device based on machine learning, comprising: means for identifying a first shift associated with a first clinician by applying at least a machine learning model, the machine learning model comprising a probabilistic machine learning model trained to identify a probability of the first clinician being on duty or off duty and one or more shifts associated with the first clinician based at least on a series of transaction records associated with the first clinician; means for determining that a first clinician may have engaged in a hazardous behavior based at least on an identified first shift associated with the first clinician; as well as Means for activating a protection workflow in response to determining that the first clinician may be engaging in a hazardous behavior, wherein the protection workflow includes at least one of configuring a waste station, generating an alarm, or activating a monitoring device.

37. The machine learning-based security control device according to claim 36, wherein: The machine learning-based security control device is also configured to execute the machine learning-based security control method according to any one of claims 18 to 34.