Systems and methods for optimal operating room scheduling under uncertainty

Abstract

A method optimizes operating room scheduling under uncertainty by: receiving a set of prospective activities to be scheduled, said prospective activities relating to operating room surgeries that consume healthcare resources for a duration of time; generating a set of mathematical distributions associated with a collection of prospective activity types, where a prospective activity type can be defined in terms of the surgery code, the surgeon ID, and other attributes, wherein each mathematical distribution quantifies the uncertainty with regard to the time duration for the retrospective prospective activity type and is constructed using stored historical data relating to the operating room surgeries; and constructing at least one schedule for the prospective activities that optimizes the utilization of healthcare resources within a defined set of constraints using the mathematical distributions such that the at least one schedule incorporates the quantified uncertainty. The method can adapt to new historical data and operational constraints.

Description

SYSTEMS AND METHODS FOR OPTIMAL OPERATING ROOM SCHEDULINGUNDER UNCERTAINTYCROSS-REFERENCE TO RELATED APPLICATION(S)

[0001] This application claims the benefit of US Provisional Application No. 63 / 555,456, filed on February 20, 2024.BACKGROUND

[0002] Operating rooms (OR) contribute to more than 30% of total expenses and 40% of total revenues in hospitals. A well-planned operating room schedule controls the patient flow in and out of the system. A major challenge faced by many hospitals in implementing effective operating room schedules is the uncertainty surrounding the number and types of surgeries that need to be scheduled each day or week (e.g., input to the system), the durations of these surgeries, and the post-surgical lengths of stays (LOS). If not properly accounted for, these uncertainties can result in inefficient resource utilization, leading to surgery cancellations that can cause significant disruptions to the delivery of care, as well as to reduced revenue affecting hospital operations.

[0003] Currently, surgery duration or length of stay values relied upon for scheduling are estimated based on surgeons’ experience or using commercial surgical scheduling systems, which calculate moving averages of previous cases for the same surgery code. At the same time, current OR scheduling systems are generally reactive, scheduling cases one-at-a-time, as they arrive. These systems have very' little room for optimization as they are not able to correct for scheduling mistakes of the past or plan for future schedules as a whole.BRIEF SUMMARY

[0004] Systems and methods for optimal operating room (OR) scheduling under uncertainty are described. The scheduling framework utilizes mathematically defined distributions of durations for surgical and post-surgical activities, as opposed to using point estimates (e.g., single values). The mathematical distributions are constructed in a data driven manner that leverages historical data captured within electronic health records (EHR) and internal hospital systems.

[0005] In some aspects, a method optimizes operating room scheduling under uncertainty by: receiving a set of prospective activities to be scheduled, the prospective activities relating to operating room surgeries that consume healthcare resources for a duration of time;generating a set of mathematical distributions associated with a collection of prospective activity types, where a prospective activity type can be defined in terms of the surgery code, the surgeon ID, and other attributes, wherein each mathematical distribution quantifies the uncertainty with regard to the time duration for the retrospective prospective activity type and is constructed using stored historical data relating to the operating room surgeries; and constructing at least one schedule for the prospective activities that optimizes the healthcare resources within a defined set of constraints using the mathematical distributions such that the at least one schedule incorporates the quantified uncertainty. Appreciably, one or more schedules are able to be generated, where all the generated schedules quantity7the uncertainty7. Multiple schedules may, for example, tune the tradeoff between the different risk metrics and objectives in different ways.

[0006] In some aspects, a system optimizes operating room scheduling under uncertainty including: a past history data store containing information relating to operating room surgeries that consume health care resources for a duration of time; a processor with access to the past history data store, the processor executing instructions to: receive a set of prospective activities to be scheduled, said prospective activities relating to the operating room surgeries; generate a set of mathematical distributions using stored historical data relating to the operating room surgeries, wherein each of the set of mathematical distributions quantifies uncertainty with regard to time for the respective prospective activity based on the historical data; and construct at least one schedule for the prospective activities that optimizes the healthcare resources within a defined set of constraints using the mathematical distributions such that the at least one schedule incorporates the quantified uncertainty'.

[0007] In some aspects, a computer-readable storage medium optimizes operating room scheduling under uncertainty7, by receiving a set of prospective activities to be scheduled, said prospective activities relating to operating room surgeries that consume healthcare resources for a duration of time; generating a set of mathematical distributions using stored historical data relating to the operating room surgeries, wherein each of the set of mathematical distributions quantifies uncertainty7with regard to time for the respective prospective activity based on the historical data; and constructing at least one schedule for the prospective activities that optimizes the healthcare resources within a defined set of constraints using the mathematical distributions such that the at least one schedule incorporates the quantified uncertainty.

[0008] This Summary is provided to introduce a selection of concepts in a simplified form that are further described below in the Detailed Description. This Summary is not intended toidentify key features or essential features of the claimed subject matter, nor is it intended to be used to limit the scope of the claimed subject matter.BRIEF DESCRIPTION OF THE DRAWINGS

[0009] FIG. 1A shows a flow chart of an optimization scheme for optimizing operating room scheduling under uncertainty.

[0010] FIG. IB shows a schematic diagram of the type of historical surgery data contained in data store, which is used to generate mathematical distributions.

[0011] FIG. 1C illustrates surgery' time distributions for a procedure for different surgeons.

[0012] FIG. ID illustrates length of stay distributions for the same surgery across different surgeons.

[0013] FIG. 1 E shows a chart of typical scheduling for patient surgeries within an operating room environment.

[0014] FIG. IF shows a chart of typical patient discharging after surgery recovery.

[0015] FIG. 2 shows a schematic diagram of an OR scheduling system including modeler and schedule optimizer.

[0016] FIG. 3A illustrates a computer implemented method for optimizing operating room scheduling under uncertainty.

[0017] FIG. 3B illustrates a use case of FIG. 3A, where risk metrics are calculated and utilized.

[0018] FIG. 3C illustrates an embodiment where mathematical distributions reflecting uncertainty are based at least in part upon surgeon and procedure pairing.

[0019] FIG. 4A shows one workflow utilized when constructing schedules that optimize health care resources.

[0020] FIG. 4B illustrates a mathematical formula for expressing a workflow able to mitigate / revolve the operating room scheduling problem.

[0021] FIG. 4C provides an example of the elements in the tuple T of the workflow formula of FIG. 4B.DETAILED DESCRIPTION

[0022] Systems and methods for optimal operating room (OR) scheduling under uncertainty are described. A predictive model uses historical data to determine mathematical distributions, as opposed to point-estimations, to quantify uncertainty' in time dependent activities likesurgery duration and length of stay. Analyzing the historical records avoids estimation bias and creates schedules with less real-world variability than otherwise possible. In certain embodiments, modeled distributions for surgery' duration and length of stay are used to calculate risk metric values, which are utilized in processing scheduling tradeoffs within the predictive model. An example of the utility' of risk metrics is the ability to establish risk-averse predictions for surgery duration so that scheduling is more effective and unexpected overtime is reduced. The scheduling framework is also adaptive, meaning that as new healthcare data is acquired over time the predictive models and the distributions that are produced are updated to reflect any changes captured within the new data. The optimized scheduling practices reliant on the distributions streamline operating room management, while increasing patient satisfaction and improving hospital resource efficiency.

[0023] Certain embodiments described provide an operating room (OR) scheduling framework, system, or method that leverages past electronic health records (EHR) and other historical data as well as current operational / resource constraints. The scheduling framework is able to respect and adapt to new operational constraints and requirements. The scheduling framework improves the efficiency and predictability of both surgical and post-surgical operations. In some embodiments, the scheduling framework includes an ability to update underlying models and generated mathematical distributions to capture changes in the historical data. The operating room scheduling methodology does not necessarily replace human schedulers but instead is able to seamlessly work with human schedulers, surgeons, and patients to improve scheduling efficiency, lower costs, and improve stakeholder experiences.

[0024] The disclosed scheduling optimizer can use fully interpretable predictive models, that are the full mathematical distributions of surgery' durations and length of stay. In some embodiments (see FIG. 1C, ID, and 3C) modeled time elements are associated with surgeonsurgery pairs, which have been found to vary from surgeon to surgeon. Using mathematical distributions to model surgery durations and length of stay avoids biases of common predictive models and allows a richer, more data-driven approach. In certain embodiments (See FIG. 3B), risk metrics are calculated, specific objectives and operational / resource constraints (e.g., operating room scheduling optimization, operating room overtime, patient throughput, surgeon and staff preferences, patient preferences, step-down bed utilization, etc.) are weighed, and a controlled tradeoff between the various obj ectives is used when constructing the optimized schedules. Controlling tradeoffs are possible only because of the mathematical distributions being able to capture and quantify uncertainty That is. defining risk measures is possible because of the mathematical distributions. Controlling tradeoffs is possible because thescheduling framework is a mathematical optimization model that explicitly captures objectives (including those that measure risks) and operational constraints.

[0025] To elaborate, the mathematical distributions of surgery duration and / or length of stay can be used to calculate important risk metrics (e.g. the Conditional Value at Risk (CVaR), Value at Risk (VaR), and Mean- Variance) that can capture different levels of performance, ranging from risk-neutral (e.g., expected value) to worst-case (e.g.. very conservative). In an example, surgery duration distributions for surgeon-surgery pairs allow risk-averse surgery duration thresholds to be calculated, which can avoid large operating room overtimes. Similarly, using the length of stay distributions for surgeon-surgery' pairs or based on other historical data, the variability of daily discharges can be controlled conditioned on schedules optimized by the predictive model.

[0026] In certain embodiments (see FIG. 4A, 4B, and 4C), a workflow or workflow formula leverages the scheduling optimizations in a manner permitting clinicians, schedulers, and patients to interact with the scheduling system to handle every' eventuality. In the proposed workflow / formula. scheduling rounds take place periodically (e.g., every few days, every week, etc.) and prospective activities (e.g.. cases) are scheduled in batches / groups during every scheduling round. The frequency of scheduling rounds depends on striking a balance between having sufficiently large batches of prospective activities to control the uncertainty' inherent in the completion of these tasks and minimizing delays between the time cases are created and surgeries are scheduled. Minimizing delays can be due to clinical reasons (e.g.. urgent cases that need to be scheduled immediately), patient satisfaction / experience considerations (e.g., minimize the wait time until patients are informed of their surgery day), or combinations of factors. During every' scheduling round, the prospective activities can be “guaranteed” to be scheduled within a prespecified time window (e.g., within two weeks, three weeks, etc.). The length of this window is decided so that all cases during each round can be scheduled within the decided time frame. This guaranteed performance is another contribution of the proposed mathematical optimization model.

[0027] FIG. 1A shows a flow chart of an optimization scheme for optimizing operating room scheduling under uncertainty. A set of prospective activities 110 to be scheduled is received by modeler 120. Each prospective activity 1 10 relates to operating room surgery and can be an activity' that requires a duration of time during which a level of healthcare resources is consumed. A modeler 120, or similar computing component, is a predictive modeler that uses historical surgery’ data of data store 122 to produce a mathematical distribution 124 quantifying uncertainty related to the time duration of relevant activities. A schedule optimizer128, or similar computing component, uses the mathematical distributions 124 as well as objectives and operational / resource constraints and requirements. Since the model is adaptive, objectives and operational constraints and requirements can be updated at every round to capture changes, the schedule optimizer 128 constructs one or more schedules 130 that optimizes the healthcare resources. Healthcare resources and their constraints can be defined within data store 126.

[0028] Operational constraints affect both surgery scheduling and post-operational treatment. The post-operational treatment requires a length of stay in a recovery room, hospital room, or hospital bed during which staff monitors the post operation recovery7process. Recovery rooms, operating rooms, and respective staff, nurses, and doctors are all constrained healthcare resources. Further, patients have lives and commitments, so schedules 130 can be constrained by patient blackout dates. Surgeons, anesthesiologists, recovery physicians, nurses, and staff all have work schedules that need to be considered when scheduling, which are also treated as constraints. For example, it is common for surgeons to be available for operating room activities for certain days of the week and not others. Data store 126 can include data on all relevant resources and their availability and constraints.

[0029] FIG. IB shows a schematic diagram of the type of historical surgery data contained in data store 122, which is used to generate mathematical distributions 124. As shown, a set of historical records are maintained (records 0 to 4 shown), where each record includes a patient identifier, a primary procedure identifier, a surgery room identifier, a physician (surgeon) identifier, minutes spent in a care facility outside an operating room, a length of stay in days, a creation day for a prospective surgery7, a discharge day, and the like. This information is not comprehensive and is a simplistic example of a portion of maintained records able to be used by modeler 120 to create mathematical distributions 124 that incorporate or quantify uncertainty in prospective activities, which affects schedules 130.

[0030] Information data store 122 can be maintained in a variety of other computing systems, such as electronic health records (EHR) for patients, hospital billing systems, employee payment systems, room reservation systems, hospital inventory systems, and the like. In various embodiments, the historical data of data store 122 can be anonymized, such as by using identifiers for patients and physicians. In some embodiments, data scrubbing activities can be sufficient to implement an online (e.g., Web based) version of a software tool for prospective modeling (e g., generating mathematical distributions 124) and / or for optimizing schedules 130 using the distributions. Along with the distributions, data elements are also used to generate operational / resource constraints and requirements in the optimization model. Forexample, quantities of specific equipment, a number of nurses working on a given day, surgeon block schedules, and the like. Often important data elements are not included in electronic health record data but exist locally within a hospital inventor or other system. Use cases are contemplated where the data relied upon does not include Health Insurance Portability and Accountability Act (HIPAA) concerns. In other use cases, the schedules 130 created herein can be presented to a manual human scheduling agent that interacts with a hospital’s scheduling computing system. Thus, the schedules 130 can be advisory in nature in certain embodiments. In some embodiments, components (e.g., modeler 120 and schedule optimizer 128) and features expressed herein can be implemented as plug-ins or functional components integrated within existing healthcare software products to provide integrated scheduling optimizations.

[0031] The variation in surgery times and length of stay across surgeons for the same surgery can vary significantly. FIG. 1C illustrates surgery time distributions for a procedure for different surgeons. In the chart, the left-hand side is labeled in minutes for the duration of the surgery procedure. The bottom of the chart shows eight different surgeons by ID number. As shown, the time in minutes per the distributions span from approximately three hundred minutes to approximately five hundred and sixty minutes. For a given surgeon, the distribution in minutes shows an uncertainty' of approximately fifty minutes for most surgeons. Because historical data in data store 122 is used to construct accurate mathematical distributions 124, based on the data used to construct the graph of FIG. 1C, surgeon identity and procedure type is a significant factor in determining an accurate mathematical distribution 124 when modeling (using modeler 120) a prospective activity 1 10 (e g., the procedure having the noted Procedure ID).

[0032] FIG. ID illustrates length of stay distributions for the same surgery' across different surgeons. In the chart, the left-hand side is labeled days of length of stay within a post-operation recovery room. The bottom of the graph denotes the same eight surgeon identifiers from FIG. 1C. As show n, the surgeon with a specific ID has a significantly' greater uncertainty or variance for length of stay post procedure as a different surgeon. Thus, the historical data in data store 122 will create different mathematical distributions 124 for the same procedure based on the assigned surgeon. As new procedures are performed, the underlying historical data is updated and the distributions 124 can change accordingly.

[0033] Significantly, when relying on point-estimates and surgeon provided estimates, the uncertainty quantified in the charts of FIG. 1C and ID cannot be properly captured. Situations exist where an assigned surgeon has not performed a specific type of procedure to sufficiently capture a distribution using a surgeon / procedure pair. In such a scenario, a machine learningalgorithm, used by the modeler 120 to determine the sought mathematical distributions can rely on transfer learning techniques. In general, transfer learning can be utilized to improve the accuracy of the predictive modeling and the generated mathematical distributions 124. Transfer learning is a machine learning technique in which knowledge gained through one task or dataset is used to improve model performance on another related task and / or different dataset. In other words, transfer learning uses what has been learned in one setting to improve generalization in another setting.

[0034] In certain embodiments, modeler 120 estimates statistical distributions for surgeonsurgery' t pe pairings that do not have sufficient data within the historical data store / repository. Transfer learning based on different procedures the surgeon performs in relation to performance of other surgeons for the different procedures as well as performance of different surgeon performance for the procedure of note are leveraged to estimate a surgeon’s distribution for the procedure of note. In other words, a determination can be made that sufficient historical data is lacking to construct an accurate mathematical distribution for a procedure of note based on the assigned physician performing that procedure in the past. Mathematical distributions can be constructed for the procedure at least in part by using a combination of surgery duration of similar surgeons performing the same type of prospective surgery'. The mathematical distributions constructed in the manner can use a weighted mean of distributions. In one implementation, a Wasserstein barycenter distance algorithm can be used.

[0035] Generally, use of the mathematical distributions 124 for scheduling reduces variance otherwise experienced. The variance is reduced because uncertainty quantified within the distributions is taken into account in a mathematical optimization framework. The operating room scheduling problem, resolved at least in part by using the distributions 124 when scheduling is graphically expressed in charts of FIG. IE and IF.

[0036] FIG. IE shows a chart of typical scheduling for patient surgeries within an operating room environment. As shown, three different operating rooms, each a healthcare resource, exist. In week 1, day 1, operating room 2 is utilized in a manner resulting in overtime, while operating room 1 was underutilized. Overtime pay will often occur when tightly scheduled surgeries run over a time estimate that conventional scheduling utilizes. Overtime pay can also result in surgery cancelations. Implementing the mathematical distributions 124 for scheduling minimizes the variance in usage rates between operating rooms and minimizes overtime while maintaining or increasing patient throughput.

[0037] FIG. IF shows a chart of typical patient discharging after surgery recovery. Because patient discharges occur responsive to and after surgery completion (plus recovery time)operating room scheduling affects daily patient discharges. When variance is reduced in operating room scheduling, variance in patient discharges is also reduced. Further, rises or falls in patient discharging (e.g., high variability) places increased pressure on recovery staff and adds variability to the number of recovery rooms (e.g., a healthcare resource) being utilized. Use of mathematical distributions 124 to quantify uncertainty in the length of stay enables the variability of discharges to be minimized, which equates to more efficient use of recovery rooms.

[0038] FIG. 2 shows a schematic diagram of an OR scheduling system including modeler 120 and schedule optimizer 128. The OR scheduling system 200 is a computer system having at least one processor 210, circuitry 214, and a set of executable instructions 212, which include software. The processor 210 is part of a set of one or more computer processors, which may be local to a computing device, distributed across network 230, or contained in the cloud. The circuitry 214 includes motherboards, network interface cards, memory, and the like. The instructions 212 can include executables written in a computer language, which cause the computer system and / or processor 210 to perform a series of defined steps. In certain embodiments, the instructions 212 can include a machine learning algorithm, generative Al programming, a trainable neural network, and / or an optimization algorithm, which is utilized at least in part to generate the mathematical distributions 124 and schedules.

[0039] The operating room scheduling system 200 can be hosted by a server having access to a network 230, which includes one or more intranets and the internet. Hospital systems can be complex and can utilize numerous tailored software applications, including an Electronic Health Record (EHR) application 240, a patient billing application 242, a patient scheduling application 244, a staff scheduling application 246, a room scheduling application, mobile health applications, recovery tracking and communication applications, additional software for medical records and data collection, and the like.

[0040] Patient scheduling applications 244 like CERNER and EPIC, may calculate moving averages of previous cases for the same surgery code, but are presently limited to providing point estimates of surgery and LOS duration. These types of programs can be extended and improved via use of the mathematical distributions 124 to ensure uncertainty is captured and taken into account. Other known systems for operating room scheduling are reactive, scheduling cases one-at-a-time as they arrive. Such systems are challenging to optimize and generally cannot correct for scheduling mistakes in the past. System 200, which can be integrated to improve existing scheduling applications, overcomes these challenges using a scheduling interval or workflow formula as elaborated upon in FIG. 4A, 4B, and 4C.

[0041] Historical data and health care resource data are indicated as being stored in data store 122 and 126. Data stores 122 and 126 can utilize a variety of different structured formats for storing the data and different cataloging and indexing schemes for data cross referencing / searching. The operating room scheduling system 200 is able to utilize a set of application interfaces 220 to interact with various applications 240, 242, 244, and 246 to effectively access the needed historical data and to discern relevant health resources as detailed herein. In some embodiments, personally identifiable information can be obscured or anonymized by data obscurer 222. For example, data obscurer 222 can permit compliance with Health Insurance Portability and Accountability' Act (HIPAA) regulations, local policies, and other medical record requirements. Security 224 module can restrict access to records, can perform encryption, and generally ensure security policies are properly maintained with regard to the historical and health care resource data.

[0042] In some embodiments, various user interface 226 components can be implemented to permit user to machine interactions. For example, the user interfaces 226 may enable patients to enter a set of available dates for a surgery or other prospective activity. Other coordination efforts, such as surgeon availability and / or confirmation can be performed via the user interface 226. The user interfaces 226 can include intranet user interfaces restricted to hospital staff use, interactive Web pages, mobile application interfaces, extended reality' interfaces, and the like.

[0043] In certain embodiments, the modeler 120 and schedule optimizer 128 can include third party software and may be able to read and write spreadsheet, database, or other structured files (e.g., character separated values or CVS files). The PYTHON OPTIMAL TRANSPORTION (POT) or similar library routines can be helpful in generating risk values and mathematical distributions 124, as can the GUROBI SOLVER and other mathematical optimization / analytic platforms. In some embodiments, the mathematical distributions 124 and scheduling functions detailed herein can be integrated with various operating room scheduling platforms such as SURGISTREAM, MAX-OR, CASECTRL, HEALTHSTREAM, and the like.

[0044] FIG. 3A illustrates a computer implemented method for optimizing operating room scheduling under uncertainty. In step 302, a set of prospective activities to be scheduled are received. These activities can relate to an operating room, scheduling operating room surgeries, reserving, and / or scheduling post-operative resources, and the like. In step 304, for each activity, a set of mathematical distributions representing uncertainty are generated usinghistorical data. In step 306, a schedule is constructed for the set of prospective activities that optimizes the health care resources needed for these activities.

[0045] FIG. 3B illustrates a use case of FIG. 3A, where risk metrics are calculated and utilized. Steps 302 and 304 are the same as detailed in FIG. 3 A. In step 310, a set of risk metrics are calculated from the set of mathematical distributions. In step 312, various objectives are defined for optimizing the health care resources. That is, often tradeoffs are made to optimize for divergent or conflicting factors and the objectives inform the method of a specific manner in which a balance between the tradeoffs is to occur. Step 314 controls the tradeoff between the various objectives using the risk metrics when constructing a schedule that optimizes health care resources.

[0046] As shown with reference to FIG. 1C and ID, statistically significant differences can exist in the amount of time needed for different surgeons to perform a similar surgery. FIG. 3C illustrates an embodiment where mathematical distributions 124 reflecting uncertainty are based at least in part upon surgeon and procedure pairing. When these parings are used, after step 302. step 320 executes. In step 320 a set of surgeons available and able to perform the surgical procedure are determined. The surgical procedure is one of the prospective activities being modeled and scheduled as discussed. In some situations, time constraints for performing the surgical procedure from the time a need is identified can exist. Further, the scheduling and optimizing system can be biased or weighted to ensure a patient is able to have a preferred surgeon perform their operation. As shown by step 322, a surgeon from the set of surgeons is assigned in accordance with availability and established preferences. The process may be iterative, as prior to a surgery' an initially assigned surgeon can be re-tasked and another surgeon assigned.

[0047] In step 324, historical data for the amount of time the assigned surgeon is likely to take for the procedure for which that surgeon was assigned. In this manner, the mathematical distributions 124 can be data-driven, at least in part, as opposed to being solely reliant on surgeon provided estimations, which are often inaccurate. In some embodiments, a particular surgeon can be shown to provide especially accurate or inaccurate estimates and a weight or adjustment applied to the surgeon estimate can be adjusted accordingly. An accurate distribution based on a surgeon / procedure pairing may require a threshold or minimal set of historical data in order to be relied upon. When this historical information is sufficient, a mathematical distribution is constructed, and the method proceeds from step 324 to step 328.

[0048] When the amount of information is insufficient, transfer learning techniques can be applied. As shown in step 326, a combination of similar surgeons and / or surgeries can beobtained from the historical data and used for constructing the mathematical distribution 124. In step 326, the mathematical distributions (possibility adjusted for risk factors, objectives, and tradeoffs as noted in FIG. 3B) are used to construct a schedule that optimizes the health care resources. To be precise, the mathematical distributions are not adjusted for risk factors but are instead leamt from historical data. The mathematical distributions capture the uncertainty' in the case length and the length of stay that are inputs to the mathematical optimization problem that determines schedules. The mathematical optimization problem controls the tradeoff between risk factors and enforces operational / resource constraints among others.

[0049] FIG. 4A shows one workflow utilized when constructing schedules that optimize health care resources. As shown, a scheduling interval can be established for gathering a sufficient set of prospective activities, including planned operating room surgeries, to construct an optimizable schedule. In other words, sets of prospective activities are run as a batch. Critical cases can be escalated and manually scheduled and / or used to expedite the scheduling prior to a default scheduling interval. Each prospective activity can include patient blackout dates, where a patent is unable to have a surgery performed or is unable to be in a recovery’ room. Similarly, a surgeon can blackout dates consistent yvith his / her availability-. The process is iterative, and patients can be manually scheduled (subject to capacity limits) as shown. The input received prior to constructing a schedule for a scheduling interval includes a set of new requests for patient surgeries. Sometimes, recovery rooms are at a premium, due to past surgeries having extensive length of stays. The scheduler (e.g.. schedule optimizer 128) considers availability of recovery rooms when scheduling. For each schedule run, anticipated surgery' time in an operating room and anticipated recovery' time post-surgery is modeled as a mathematical distribution 124 by modeler 120. As shown, the output of a scheduling run is a set of assigned surgeries by time and operating room as yvell as a set of post-operative resources (e.g., recovery rooms) anticipated. An automated caller, or human agent, can confirm the scheduled dates with patients and surgeons, and the schedule can be updated, as necessary. Appreciably, use of the workflow of FIG. 4A and the mathematical optimization model that relies on mathematical distributions 124 to quantify uncertainty- reduces variability in both patient scheduling and discharge dates, significance of which was elaborated upon in reference to FIG. IE and IF.

[0050] FIG. 4B illustrates a mathematical description of the operating room scheduling problem resulted / mitigated herein. FIG. 4C provides an example of a scheduling problem flow consistent with the mathematical description and conventions of FIG. 4B. FIG. 4C also shows the type of data, including operational / resource constraints and requirements used herein. Incertain embodiments, the mathematical description of FIG. 4B and / or derivatives thereof is used by the schedule optimizer 128. The schedule optimizer is not limited in this regard and other approaches exist in the art and can be used in other contemplated use cases.

[0051] Appreciably (see FIG. IE and IF), a major challenge faced by many hospitals in implementing effective OR schedules are the uncertainty surrounding the number and types of surgeries that need to be scheduled each day or week (e.g., input to the system). These uncertainties are not considered by existing scheduling systems leading to inefficient utilization of resources and problems. Scheduling includes a set of related activities involving surgeries in an operating room itself, operating room post anesthesia care units (PACU), intensive care units (ICU), and step-down beds, and the like. The problems can lead to surgery cancellations, staff burnout, and patient dissatisfaction that together cause significant disruptions to care delivery and reduced revenue for the health system.

[0052] The schedule optimizer 128 leverages past Electronic Health Record (EHR) data and other historic data to improve the efficiency of OR utilization and the predictability of postoperative patient flow. Specifically, modeler 120 uses predictive models of mathematical distributions 124 (compared to only point estimates) of surgery durations and length of stay (LOS). The mathematical distributions 124 quantify uncertainty and can be calculated in a data- driven manner using historical data. In some embodiments, the mathematical distributions (see FIG. 3C) are based in part upon unique combinations or pairings of surgeon and surgery procedure, which is referred to as a surgeon-surgery pair.

[0053] Per FIG. 4B, the Operating Room (OR) scheduling problem as a sequential decisionmaking problem, defined by the tuple:T = {Nt,Wt, Pt.Ct,Dt,Xt,R} --Variable t is a time index, also called a scheduling round, which captures the time when the schedule is updated, (e.g., once a week, twice a week, etc.) Nt represents the set of new cases recorded at time step t. Wt represents a given time horizon starting at time t during which the new Nt cases need to be scheduled. Pt is the prior OR schedule containing cases that were scheduled before time step t but have not been performed yet. Ct is the set of constraints that the new' schedule at time step t must satisfy, (e.g., surgeon availability, patient blackout dates, etc.) Dt denotes the dataset that contains information on all cases that have been performed and discharged before time step t. (e.g., case duration, length of stay, etc.) Xt = {XOt ,Xlt } is the updated OR schedule at time step t that includes the prior schedule Pt and the additionalschedule that assigns the new cases in Nt to specific days and ORs within the time horizon Wt while respecting the constraints in Ct. The partial schedule XOt represents the cases whose scheduled surgery days are before the next time step t + 1 and Xlt represents the remaining cases whose scheduled surgery days are after the next time step t + 1 and will constitute the prior schedule for the next time step, (i.e., Pt+1 = Xlt) R is the objective (reward) function that measures the performance of the OR schedule Xt. (i.e., R(Xt) e R and the higher the reward, the better the OR schedule is.)

[0054] The optimal schedule Xt at each time step is defined as follows:X(Pt,Ct,Wt) denotes the set of all schedules that respect the prior schedule Pt, satisfy the constraints in Ct. and finish all new cases within the window Wt. The new cases Nt and data Dt constitute inputs to the OR scheduling problem during scheduling round t. FIG. 4C provides example of the elements in the tuple T.

[0055] Finding the optimal schedule Xt that optimizes health care resources requires the ability to evaluate a schedule at time step t. Since the outcome of a schedule Xt cannot be observed at the decision time step t, except for perhaps some counting statistics (e.g., the number of daily scheduled cases and weekly scheduled cases for each surgeon), a predictive model P(Xt) (see modeler 120) is trained from the dataset Dt and captures the randomness in the outcome of the schedule Xt. (e.g., the distribution of surgery duration and LOS).

[0056] Embodiments of the described decision management system and LLM Business Knowledge Base can be embodied as a computing system implemented as a single system but may also be implemented across multiple systems or sub-systems co-located or distributed relative to each other. Computing systems generally include one or more processors that transform or manipulate data according to the instructions of software, one or more storage devices on which the software and data are stored, and communication systems for wired and / or wireless communication across devices and systems.

[0057] It should be understood that as used herein, in no case do the terms “storage media,” “computer-readable storage media” or “computer-readable storage medium” consist of transitory carrier waves or propagating signals. Instead, “storage” media refers to non- transitory media.

[0058] Although the subject matter has been described in language specific to structural features and / or acts, it is to be understood that the subject matter defined in the appended claimsis not necessarily limited to the specific features or acts described above. Rather, the specific features and acts described above are disclosed as examples of implementing the claims and other equivalent features and acts are intended to be within the scope of the claims.

Claims

CLAIMSWhat is claimed is:

1. A method for optimizing operating room scheduling under uncertainty comprising: receiving a set of prospective activities to be scheduled, said prospective activities relating to operating room surgeries that consume healthcare resources for a duration of time; generating a set of mathematical distributions associated with a collection of prospective activity types, wherein each mathematical distribution of the set of mathematical distributions quantifies uncertainty with regard to a time for a corresponding one of the prospective activity types and is constructed using stored historical data relating to the operating room surgeries; and constructing at least one schedule for prospective activities corresponding to the prospective activity types that optimizes utilization of healthcare resources within a defined set of constraints using the mathematical distributions such that the at least one schedule incorporates the quantified uncertainty.

2. The method of claim 1, wherein one of the prospective activities is for an operation to be performed on a patient by surgical staff in an operating room, wherein one of the mathematical distributions is for a time needed in the operating room for the operation to be performed on the patient, wherein the at least one schedule includes an operating room schedule constrained by availability of at least two of the surgical staff, a set of patients in need of surgeries, equipment available, and operating room availability.

3. The method of claim 2, wherein one of the prospective activities is for post- operational treatment for the patient after the operation, wherein post-operational treatment requires a length of stay in a recovery room by the patient, wherein one of the mathematical distributions is for the length of stay needed in the recovery’ room for the patient, wherein the at least one schedule includes a recovery room schedule constrained by availability of recovery staff and a set of recovery rooms available.

4. The method of claim 1. further comprising: calculating a set of risk metrics from the mathematical distributions; andcontrolling a tradeoff between various objectives based on values of the risk metrics when constructing the at least one schedule that optimizes health care resources.

5. The method of claim 4, wherein the set of risk metrics comprise a computation for at least one of Conditional Value at Risk (CVaR), Value at Risk (VaR), and Mean- Variance.

6. The method of claim 1, further comprising: controlling a tradeoff between different objectives when constructing the at least one schedule that optimizes health care resources.

7. The method of claim 6, wherein the different objectives comprise: operating room scheduling optimization, operating room overtime minimization, operating room underutilization minimization, and step-down bed utilization.

8. The method of claim 1. wherein the mathematical distributions are created by a predictive model that utilizes a learning algorithm trained using the historic data to capture the uncertainty within the mathematical distributions.

9. The method of claim 8. further comprising: constructing the mathematical distribution for one surgery of the prospective surgeries using the predictive model and past surgeries of a same type of the prospective surgery' by a same physician that is to perform the one surgery.

10. The method of claim 9, further comprising: determining sufficient historical data to construct an accurate mathematical distribution for the one surgery based on the same physician performing the same ty pe is insufficient; and constructing the mathematical distribution for the one surgery’ at least in part by using a combination of surgery duration of similar surgeons performing the same type of prospective surgery.

11. The method of claim 10, comprising: constructing the mathematical distribution using the combination and using a weighted mean of distributions.

12. The method of claim 1. further comprising: establishing a time window to have a prospective surgery and relative prospective activities be scheduled, wherein the time window is one of the defined constraints, whereby the time window reduces delays between a time a case for the prospective surgery occurs and a time at which the prospective surgery takes place per the at least one schedule.

13. The method of claim 1, wherein the defined constraints comprise patient provided blackout dates, and schedules of surgeons.

14. A system for optimizing operating room scheduling under uncertainty comprising: a data store containing information relating to operating room surgeries that consume health care resources for a duration of time; and a processor with access to the data store, the processor executing instructions to: receive a set of prospective activities to be scheduled, said prospective activities relating to the operating room surgeries; generate a set of mathematical distributions associated with a collection of prospective activity types, wherein each mathematical distribution of the set of mathematical distributions quantifies uncertainty with regard to a time for a corresponding one of the prospective activity types and is constructed using stored historical data relating to the operating room surgeries; and construct at least one schedule for prospective activities corresponding to the prospective activity’ types that optimizes the utilization of healthcare resources within a defined set of constraints using the mathematical distributions such that the at least one schedule incorporates the quantified uncertainty.

15. The system of claim 14, wherein one of the prospective activities is for an operation to be performed on a patient by surgical staff in an operating room, wherein one of the mathematical distributions is for a time needed in the operating room for the operation to be performed on the patient, wherein the at least one schedule includes an operating room schedule constrained by availability of the surgical staff, a set of patients in need of surgeries, and operating room availability.

16. The system of claim 15, wherein one of the prospective activities is for post- operational treatment for the patient after the operation, wherein post-operational treatment requires a length of stay in a recovery room by the patient, wherein one of the mathematical distributions is for the length of stay needed in the recovery room for the patient, wherein the at least one schedule includes a recovery room schedule constrained by availability of recovery staff and a set of recovery rooms available.

17. The system of claim 14, said processor further executing the instructions to: calculate a set of risk metrics from the mathematical distributions; and control a tradeoff between various objectives based on values of the risk metrics when constructing the at least one schedule that optimizes the health care resources.

18. The system of claim 14, said processor further executing the instructions to: controlling a tradeoff between different objectives when constructing the at least one schedule that optimizes the utilization of health care resources, wherein the different objectives comprise: operating room scheduling optimization, operating room overtime, operating room underutilization minimization, and step-down bed utilization.

19. A computer-readable storage medium for optimizing operating room scheduling under uncertainty, comprising instructions stored thereon, that when executed on a processor, perform the steps of: receiving a set of prospective activities to be scheduled, said prospective activities relating to operating room surgeries that consume healthcare resources for a duration of time; generating a set of mathematical distributions with time being a variable of each of the mathematical distributions using stored historical data relating to the operating room surgeries, wherein each of the set of mathematical distributions quantifies uncertainty with regard to time for the respective prospective activity based on the historical data; and constructing at least one schedule for the prospective activities that optimizes the utilization of healthcare resources within a defined set of constraints using the mathematical distributions such that the at least one schedule incorporates the quantified uncertainty.

20. The computer-readable storage medium of claim 19. that when the instructions are executed on the processor, further perform the steps of: calculating a set of risk metrics from the mathematical distributions; and controlling a tradeoff between various objectives based on values of the risk metrics when constructing the at least one schedule that optimizes health care resources.

Images