MACHINE LEARNING MODELS FOR DESIGNATION OF SUBJECTS FOR TREATMENT AND/OR EVALUATION

DETAILED ACTION Notice of Pre-AIA or AIA Status The present application, filed on or after March 16, 2013, is being examined under the first inventor to file provisions of the AIA . Applicant Argument Response Subject Matter eligibility: The examiner addressed all applicant arguments regarding subject matter eligibility on pages 16-21 filed on 08-12-26 and found them unpersuasive for the reasons below: Applicant argues amended claims 1, 26, and 45 now recite the shared-model architecture disclosed in paragraphs [0082] and [0091], including single stored model instances, per-terminal inference-time weights, and avoidance of per-terminal datasets and retraining, thereby providing the technological improvements recognized in Desjardins. The Examiner respectfully disagrees. Although the amendments now recite the asserted arrangement, under the broadest reasonable interpretation, the different terminal outputs arise because the intermediate component applies different global list weights to the same model outputs. Paragraph [0082] confirms that different combined priority lists are obtained by adjustment of the global list weights, while the trained models and their parameters remain unchanged; paragraph [0091] likewise describes combining priority-list information, not changing model operation. Desjardins credited reduced storage and complexity because the claimed training protected prior knowledge while the model learned new tasks, thereby improving how the machine learning model itself operates. Here, fewer training runs and stored models result from omitting user-specific models and differently weighting prioritization data, not from an improved training technique, inference mechanism, model structure, or storage architecture. The claims therefore improve implementation of the abstract prioritization itself, not computer or machine-learning technology under MPEP §§ 2106.04(d)(1) and 2106.05(a). The § 101 rejection is maintained. Applicant argues amended claims 1, 26, and 45 now expressly recite the shared-model arrangement disclosed in paragraphs [0082] and [0091], thereby satisfying both requirements of MPEP § 2106.05(a). The Examiner respectfully disagrees. Because the claims reflect the disclosed arrangement, but that arrangement improves the production of customized prioritization results, not computer or machine-learning functionality. Applicant argues the avoidance of per-terminal datasets, training runs, and model storage is the same category of technological improvement recognized in Desjardins. The Examiner respectfully disagrees. Desjardins did not treat reduced storage or complexity as independently sufficient. The credited benefits resulted from changing how the model operated during continual learning: parameter values were adjusted so the model could learn a new task while protecting knowledge of a previous task and avoiding catastrophic forgetting. Here, the trained models and their parameters are not updated to acquire or preserve knowledge. Different rankings result from applying different global list weights to their outputs during inference. The avoided storage and training follow from not creating separate user-specific models, not from improved learning, inference, or model architecture. The Desjardins reasoning therefore does not establish integration here. Applicant argues generating N terminal-specific outputs from one trained model is a structurally superior computer architecture comparable to the self-referential table in Enfish. The Examiner respectfully disagrees. Enfish claimed a specific logical table structure and configuring algorithm that changed how the database stored and retrieved information. Claims 1, 26, and 45 do not recite a corresponding change to model structure, memory organization, storage layout, or execution architecture. They obtain N rankings by supplying N sets of numerical weights to the same unchanged models. The reduction in training and storage therefore results from omitting N user-specific training pipelines and substituting inference-time weighting of prioritization data. That improves the efficiency of producing customized rankings, not the operation of the computer itself. MPEP §§ 2106.04(d)(1) and 2106.05(a). Applicant argues the claims satisfy MPEP § 2106.05(f) because they recite how differentiated rankings are produced: the same models are reused, weights are applied by the intermediate component, and inference occurs without retraining. The Examiner respectfully disagrees. The recited how consists of receiving numerical weights, applying those weights to priority-list scores, ranking patients, and transmitting the results. Those operations are the mathematical prioritization identified in Prong One. The server, models, intermediate component, network, and terminals store, execute, communicate, and display that process, but the claims do not specify a technological modification to those components. Describing the mathematical process with greater implementation detail does not convert it into a technological solution when the resulting advance remains differentiated patient rankings. The additional elements therefore use the computer system as the tool for applying the exception. MPEP § 2106.05(f). Applicant argues the claimed conservation of model-storage and training resources is comparable to Example 40’s reduction of network traffic and monitoring overhead. The Examiner respectfully disagrees. Example 40 recited collecting resource-intensive NetFlow data only after detecting an abnormal network condition. That claimed control directly changed network monitoring and avoided excess traffic volume on the network and hindrance of network performance. Claims 1, 26, and 45 do not detect a computer condition or control training or storage in response to a technological condition. They avoid user-specific datasets and models because they instead produce different patient rankings by changing numerical prioritization weights. The claimed benefit therefore arises from selecting a different analytical method for producing rankings, not from improving network, storage, or processing operation. Example 40 does not support integration. Applicant argues the shared-model deployment resembles Example 47’s network-intrusion system and Example 48’s speech-separation process because each uses a trained model in a specifically claimed manner. The Examiner respectfully disagrees. Example 47 applied the model output to a technological process by identifying source addresses, dropping malicious packets, and blocking future network traffic. Example 48 converted clustered representations into separate speech waveforms and generated a new audio signal excluding an unwanted source. Claims 1, 26, and 45 instead terminate with informational patient scores and rankings. They do not control a medical device, perform treatment, change an electronic signal, or remediate a computer condition. Using a trained model in a specified manner is not alone sufficient; the claimed application must produce the asserted technological improvement. These claims improve the prioritization result itself. Applicant argues storing and reusing one model instance is an affirmative machine operation that concretely achieves reduced storage, system complexity, and training burden. The Examiner respectfully disagrees. The single-instance limitation is given full weight, but under its broadest reasonable interpretation it requires one stored instance to be reused for the terminal-specific inference calculations. It does not recite a changed memory structure, shared-memory protocol, model-serving technique, processor-scheduling operation, or storage-management mechanism. Claim 46 adds concurrent execution but likewise does not specify how concurrency is technically achieved. The recited single instance reduces the number of models stored because it does not create separate user-specific models; it does not improve how storage, training, or inference operates. Accordingly, the asserted machine operation does not integrate the mathematical prioritization into a technological application. Applicant argues the claims do not manage personal behavior or interactions because computing a back-end patient ranking neither causes the ranked subjects to act nor requires interaction between them. The Examiner respectfully disagrees. Under the broadest reasonable interpretation, claims 1, 26, and 45 are not limited to an internal record-review queue. They expressly rank subjects for prioritized medical evaluation or treatment by resource-limited interventions. The resulting lists determine the sequence in which providers evaluate or treat patients, thereby organizing provider conduct and provider-patient interactions according to the computed priorities. MPEP § 2106.04(a)(2)(II) explains that the subgroup includes following rules or instructions and may encompass the activity of one person; direct interaction among every affected person is unnecessary. The claimed healthcare-prioritization limitations therefore describe managing personal behavior or relationships or interactions between people. Applicant argues the mathematical-concept grouping is the stronger Prong One basis and, because Desjardins integrated a mathematical concept through a machine-learning improvement, the same result follows here. The Examiner respectfully disagrees. Removing the organizing-human-activity characterization would not end the eligibility analysis because the claims independently recite multiplying scores by global list weights and ranking the resulting numerical values, which constitute mathematical calculations under MPEP § 2106.04(a)(2)(I). Desjardins does not make every computer-implemented mathematical calculation involving machine learning eligible. Its claims changed how a model learned successive tasks while protecting previously acquired knowledge. Here, unchanged models produce different patient rankings because different numerical weights are applied during inference. As explained in Prong Two, that changes the abstract prioritization inputs and results, not the operation of the model or another technology. Applicant argues Example 42 confirms eligibility because the amended claims likewise recite a specific technical manner of delivering a healthcare-information function. The Examiner respectfully disagrees. Example 42 solved a technological incompatibility among separately maintained medical-record systems by converting non-standardized updates into a standardized format, storing the converted information, and providing real-time network access regardless of the originating platform. Claims 1, 26, and 45 recite no corresponding data-format conversion, interoperability mechanism, or change in network accessibility. The server instead receives different numerical weights, applies them to priority-list scores, and returns different patient rankings. Storing and reusing the same model identifies the computational tool used, but the claimed result remains customized prioritization information. Example 42 therefore does not establish that these claims integrate the exception into a practical application. Applicant argues the manual-replication analysis does not reach the amended claims because humans cannot store or reuse trained machine-learning models or perform concurrent inference from one stored parameter set. The Examiner respectfully disagrees. Prong One does not classify the server, stored models, or concurrent inference as mental acts. It identifies the underlying acts of reviewing patient information, assigning priority weights, multiplying scores, comparing results, and ordering patients as mathematical evaluations and clinical prioritization capable of human performance. The server and model limitations are additional elements evaluated separately under Prong Two. Automating the abstract process for multiple terminals does not remove the mathematical and organizational concepts recited by the claims. Applicant argues that paragraphs [0098]–[0101] and [0107]–[0109] describe only computing hardware and program delivery and therefore do not establish that the complete shared-model arrangement is well-understood, routine, and conventional. The Examiner does not rely on MPEP § 2106.05(d) to establish that the complete ordered combination is well-understood, routine, and conventional. The additional elements nevertheless fail to provide significantly more under MPEP §§ 2106.05(a), (f), and (h), because they use the server, models, intermediate component, terminals, and network to execute and distribute the mathematical patient-prioritization process without improving the operation of those components or applying the result to another technological process. Applicant argues that claim 46 is eligible because concurrent generation of different priority lists from one stored parameter set cannot be performed mentally and is not a method of organizing human activity. The Examiner respectfully disagrees. Claim 46 inherits the mathematical weighting, ranking, and patient-prioritization limitations of claim 1. The concurrent server limitation is an additional element; it does not determine whether the underlying operations constitute an abstract idea. At Step 2B, the relevant question is whether concurrency supplies significantly more than the exception. Under the broadest reasonable interpretation, claim 46 requires the server to perform the same inference concurrently for multiple terminal-specific weights using one stored parameter set, but it does not recite a concurrency-control method, processor-scheduling technique, shared-memory arrangement, or changed inference mechanism. Concurrent execution therefore increases the simultaneous performance of the exception without improving computer or model operation. MPEP §§ 2106.05(a) and (f). Applicant argues that independent claims 26 and 45 are eligible on the same shared-model and inference-time-parameterization basis asserted for claim 1. The Examiner respectfully disagrees. Claims 26 and 45 recite the same material additional elements: a server and processor, machine-learning models, an intermediate component, client terminals, a network, GUIs, and single-instance model storage and reuse. Their different formulations concern feeding, weighting, computing, and ranking patient information, which remain part of the judicial exception. As explained for claim 1, different terminal outputs arise from applying different numerical weights to unchanged models. Neither claim adds a changed model structure, training technique, inference mechanism, storage architecture, or technological application of the rankings. The same Step 2A and Step 2B determinations therefore apply. Applicant argues that the dependent claims are allowable by virtue of their dependency. The Examiner respectfully disagrees. Dependency carries forward every limitation of claim 1, including the identified judicial exceptions and additional elements evaluation. Dependency alone therefore does not overcome the § 101 rejection. 35 U.S.C 103: The examiner addressed all applicant arguments regarding 35 U.S.C 103 on pages 16-21 filed on 08-12-26 and found them persuasive for the reasons below: Subject Matter Free from Prior Art The closest teachings are these: Nida, US20200279641A1, describes specific predictive models, an optional machine learning model trained to combine and sort their outputs, and separate clinical queues for different facilities (0043–0047). It does not identify one stored instance of each model reused for every client, or client supplied list weights applied to the specific model outputs before input to the combining model. Banis, US20200210867A1, describes stored models and requests that may identify the same model (and/or same version of the model) (0069–0071, 0077). Model identity does not establish a single stored instance of each of Nida’s specific models and its combined model. Micaelian, US6714929B1, describes client adjustment of weights that changes ranked search results (FIGS. 2 and 8a; see also the discussion of weighted scoring). It does not describe specific clinical machine learning models, a combined prioritization machine learning model, or weighted specific priority lists fed into that model during inference. Together, these references provide a plausible reason to investigate applying client selected weights to Nida’s clinical queues: Micaelian shows that changing weights changes a ranking, while Banis shows model use across requests. The cited passages do not, however, establish the particular implementation recited in claims 1 and 26: one stored instance of each specific model and of the combined model serves all client terminals; each client’s different global list weights are applied by intermediate component code to produce weighted specific priority lists; and those lists are supplied to the same combined model during inference to yield different outputs because of the weights, without creating client specific training datasets or training client specific models. Claim 45 recites the related result through its weighted scores, without the same intermediate component wording. Accordingly, these three references, on the passages identified, do not establish a complete § 102 or reasoned § 103 case for that shared-model inference arrangement. Claim Rejections - 35 USC § 101 35 U.S.C. 101 reads as follows: Whoever invents or discovers any new and useful process, machine, manufacture, or composition of matter, or any new and useful improvement thereof, may obtain a patent therefor, subject to the conditions and requirements of this title. Claims 1-5, 8-10, 12-16, 20, 25, 26, and 45-46 are rejected under 35 U.S.C. § 101 because the claimed subject matter is directed to a judicial exception (an abstract idea) without reciting elements that integrate the exception into a practical application or provide an inventive concept amounting to significantly more than the exception itself. Step 1 The claims are directed to statutory subject matter, encompassing the following statutory category: Machine (Claims 1-5, 8-10, 12-16, 20, 25, 26, and 45-46): The language reciting A system... comprising: at least one processor executing a code for... (Claims 1, 26, and 45) describes a concrete thing consisting of parts, aligning with the definition of a machine in MPEP § 2106.03. Having confirmed the claims are directed to statutory subject matter, the analysis proceeds to Step 2A. Step 2A, Prong One: Judicial Exception Analysis Prong One asks whether the claim recites a judicial exception. The following representative claim 1 recites an abstract idea non-bold and the additional elements in bold: Independent Claim 1 Analysis [1] A system for training a plurality of machine learning models and for training a combined prioritization machine learning model for dynamic prioritization of target subjects for at least one of evaluation, and treatment, comprising: at least one processor of a server in communication with a plurality of client terminals over a network, executing a code for: [2] accessing electronic medical records (EMR) of a set of a plurality of subjects;for at least one of each of a plurality of specific medical interventions that are resource limited, and for each respective target clinical outcome: accessing a respective specific priority list of a respective sub- set of the plurality of subjects scheduled in a prioritized sequence for at least one of treatment and evaluation for at least one of the respective target clinical outcome and by the respective specific medical intervention; [3] creating a respective specific training dataset that includes data extracted from the EMRs of at least the respective sub-set of the plurality of subjects labelled with the specific priority list; [4] training a respective specific machine learning model on the respective specific training dataset for generating an outcome of a respective target specific priority list of a sub-set of target subjects for prioritized at least one of evaluation and treatment for the at least one of respective [5] target clinical outcome and by the respective specific medical intervention, in response to an input of data extracted from EMR of at least the sub-set of target subjects; [6] accessing code of an intermediate component for computing a plurality of weighted specific priority lists by assigning a respective preselected global list weight to each respective target specific priority list outcome of respective specific machine learning models, creating a main training dataset that includes the plurality of weighted specific priority lists obtained as outcomes of the plurality of specific machine learning models assigned respective preselected global list weights, labelled with a respective combined priority list of the set of the plurality of subjects for at least one of treatment and evaluation; [7] training a combined prioritization machine learning model on the main training dataset for generating an outcome of a target combined priority list of target subjects for prioritized at least one of evaluation and treatment in response to an input of the plurality of weighted specific priority lists; and [8] providing the plurality of specific machine learning models, the code of the intermediate component, and the combined prioritization machine learning model, [9] wherein the plurality of specific machine learning models and the combined prioritization machine learning model are implemented as one or more or combination of the following architectures: neural network, logistic regression, decision tree, and boosting, [10] wherein the plurality of specific machine learning models and the combined prioritization machine learning model generate a respective numerical score for each respective subject, and the specific priority lists and the combined priority list are created by ranking subjects according to respective numerical scores, [11] wherein the combined prioritization machine learning model computes,for each respective subject of the plurality of target subjects, a respective weighted score by multiplying each respective numerical score assigned to the respective subject on each respective priority list by the respective global list weight assigned to the respective list; And [12] a plurality of client terminals in communication with the server over the network, each client terminal including at least one processor executing a code for: [13] presenting on each respective display of each respective client terminal of the of the plurality of client terminals, graphical user interface (GUI) configured for dynamically adjusting the respective preselected global list weights indicating the magnitude of prioritization of the respective target specific priority list to respective values, [14] wherein different GUIs dynamically adjust the preselected global list weights to different values, wherein each respective preselected global list weight indicates a magnitude of prioritization of the respective target specific priority list for the respective client terminal, wherein the same respective preselected global list weight is applied to all scores of all subjects of the same respective target specific priority list for the respective client terminal, wherein each respective global weight of each respective priority list is adjustable to a different value by a different client terminal; [15] transmitting the respective dynamically adjusted preselected global list weights of the respective client terminal to the server, wherein the server performs inference using the combined prioritization machine learning model using the different values of the preselected global list weights received from the plurality of client terminals;and [17] receiving from the server and presenting within the GUI on each respective display, for each respective subject, at least one of: [18] the respective numerical score computed by the combined prioritization machine learning model, and a ranking of the respective subject on the combined priority list, computed according to the respective dynamically adjusted preselected global list weights, wherein the server sends different numerical scores and/or rankings to corresponding client terminals according to the dynamically adjusted preselected global list weights. [19] wherein the server stores a single instance of each of the plurality of specific machine learning models and a single instance of the combined prioritization machine learning model, and reuses said single instances to perform the inference for each of the plurality of client terminals. [20] wherein the server generates the different numerical scores and/or rankings for the plurality of client terminals using the same plurality of specific machine learning models and the same combined prioritization machine learning model for all of the plurality of client terminals, by performing the inference with the different values of the preselected global list weights received from the plurality of client terminals, such that different combined priority lists are obtained for different client terminals of the plurality of client terminals, rather than creating different training datasets for each of the plurality of client terminals and rather than training different machine learning models for each of the plurality of client terminals; and [21] wherein the intermediate component executes code to apply the respective dynamically adjusted preselected global list weights to the respective target specific priority list outcomes during the inference performed by the server to compute the plurality of weighted specific priority lists that are input to the combined prioritization machine learning model. Claims 26 and 45 mirror claim 1, with claim 26 detailing the operational sequence as a method and claim 45 as executable instructions on a non-transitory computer-readable medium. Since both describe the same operations without additional technological elements or differing physical limits, they are grouped with claim 1 for the judicial exception evaluation. The non-bold language of limitations [2] through [11], [13] through [18], and [20] through [21] recites judicial exceptions under three statutory categories. Limitations [2], [3], [4], [6], and [7]: Under the broadest reasonable interpretation, the non-bold language describes accessing patient electronic medical records, extracting patient data to create training datasets, and generating specific and combined priority lists of human subjects scheduled in prioritized order for medical evaluation or treatment under resource-limited interventions. This describes certain methods of organizing human activity under MPEP 2106.04(a)(2)(II), specifically managing personal behavior or relationships or interactions between people, because the claimed process schedules human patients for clinical care based on healthcare resource availability. This language also describes mental processes under MPEP 2106.04(a)(2)(III) because gathering, evaluating, and grouping patient records into clinical priority tiers are data-evaluation steps capable of being performed in the human mind. Limitations [5], [6], [11], [13], [14], [15], and [21]: Under the broadest reasonable interpretation, the non-bold language describes assigning preselected global list weights indicating priority magnitude, applying uniform weights across scores of a list, multiplying each numerical score assigned to a subject by a respective global list weight to compute a weighted score, transmitting adjusted weights, and applying weights to compute weighted input lists during inference. This describes mathematical concepts under MPEP 2106.04(a)(2)(I) because limitation [11] recites an explicit mathematical calculation and formula multiplying numerical scores by list weights, and limitations [5], [6], [13], [14], [15], and [21] recite mathematical relationships governing numerical weighting, scalar multiplication, and parameter scaling. Limitation [9]: Under the broadest reasonable interpretation, the non-bold language describes implementing models as neural networks, logistic regression, decision trees, and boosting. This describes mathematical concepts under MPEP 2106.04(a)(2)(I) because logistic regression formulas, decision tree logical branches, boosting iterations, and neural network activation functions are mathematical algorithms and statistical computational formulations. . limitations [10], [18], and [20]: In the broadest reasonable interpretation, the non-bold language refers to calculating numerical scores, comparing these scores to rank subjects on priority lists, transmitting various rankings based on adjusted weights to client terminals, and generating different priority lists across client terminals using different weights. This does not involve creating different datasets or training separate models. This aligns with mathematical concepts in MPEP 2106.04(a)(2)(I), which involves calculating numerical scores and sorting numbers. Additionally, it covers specific methods of organizing human activities under MPEP 2106.04(a)(2)(II) and mental processes under MPEP 2106.04(a)(2)(III), since coordinating clinical triage across hospital units and managing patient priority rankings from a single standard are considered administrative and clinical judgment practices.. Limitations [8] and [17]: Under the broadest reasonable interpretation, the non-bold language describes providing model code and intermediate weighting instructions, and receiving and presenting numerical scores and rankings for each subject. This describes mental processes under MPEP 2106.04(a)(2)(III) because providing decision-making rules and receiving or observing patient priority rankings are informational steps directed to human mental observation and clinical assessment. Under the broadest reasonable interpretation, the abstract prioritization process can be performed by humans without computer assistance. In a clinical setting, a triage committee or clinical team can review medical records, assess interventions, and compile patient lists. They can follow established decision logic or guidelines, assign weights based on departmental priorities, and calculate scores by multiplying with these weights. Subsequently, they can rank patients by urgency and distribute personalized priority lists to various hospital units. By maintaining a single set of triage guidelines and applying department-specific weighting sheets as needed, they can efficiently generate customized rankings without creating separate evaluation criteria—mirroring typical human triage practices.. Under MPEP 2106.04(II)(B), because the mathematical calculations directly execute the clinical triage ordering and mental comparisons, the overlapping mathematical concepts, mental processes, and methods of organizing human activity across limitations [2] through [11], [13] through [18], and [20] through [21] are considered together as an integrated judicial exception directed to mathematical patient prioritization for medical resource allocation. Dependent Claims Analysis: Dependent claims 2 - 5, 8 - 10, 12 - 16, 20, 25, and 46 retain the judicial exception of claim 1: Dependent claims 2 - 5, 9, 10, 12, 16, 20, and 25 recite judicial exceptions without reciting any additional elements. Under the broadest reasonable interpretation, claims 2 - 5, 9, and 25 recite mathematical concepts under MPEP 2106.04(a)(2)(I) and mental processes under MPEP 2106.04(a)(2)(III) by defining weights that comprise a numerical value (claim 2), are selected according to a geographical location (claim 3), defined according to a set of rules (claim 4), fixed within a defined range of values (claim 5), aggregated by multiplying each respective numerical score... by the respective weight (claim 9), or dynamically adjustable by a user via a user interface (claim 25). These recite mathematical formulas, numerical ranges, and mental evaluation rules. Claims 10, 12, 16, and 20 recite certain methods of organizing human activity under MPEP 2106.04(a)(2)(II), specifically managing personal behavior or relationships or interactions between people, by denoting a maximal recommended time interval for interventions (claim 10), denoting a recommended interval for an effective allocation of resources (claim 12), sequentially at least one of treating and evaluating the plurality of target subjects (claim 16), and generating instructions for at least one of treating and evaluation target subjects (claim 20). Coordinating patient scheduling and clinical care instructions directly governs human medical behavior and provider-patient interactions. Because no additional technological elements are recited, these claims do not overcome Prong One. Dependent claims 8, 13, 14, and 15 recite additional elements, including processor execution code, machine learning models, and training dataset structures, but do not overcome Prong One because they continue to recite and directly execute judicial exceptions. Under the broadest reasonable interpretation, claim 8 recites aggregation code that when executed by a processor aggregates... respective numerical scores... into an aggregated score and ranks subjects, which is an explicit mathematical calculation under MPEP 2106.04(a)(2)(I). Claims 13 and 14 recite training datasets that include a schedule of availability of resources for performing each respective corresponding medical intervention (claim 13) and an indication of risk of at least one of a respective clinical outcome and/or a diagnosis to prioritize subjects (claim 14). These recite certain methods of organizing human activity under MPEP 2106.04(a)(2)(II), specifically managing personal behavior or relationships or interactions between people, by scheduling clinical resources and organizing medical care based on patient risk, as well as mental processes under MPEP 2106.04(a)(2)(III) for diagnostic evaluation. Claim 15 recites generic code for receiving a set of target EMRs and feeding data into models to obtain priority lists, which merely routes patient information to execute the abstract triage process. These generic elements do not remove the recited judicial exceptions. Dependent claim 46 recites additional server execution elements, including performing inference using a single model instance concurrently for multiple client terminals from a single stored parameter set, but does not overcome Prong One because it still recites judicial exceptions. Under the broadest reasonable interpretation, claim 46 recites that the server performs the inference using the single instance of the combined prioritization machine learning model concurrently for the plurality of client terminals, generating the different combined priority lists for the different client terminals from a single stored set of trained model parameters. This limitation evaluates the mathematical formulas and model weighting calculations of limitations [19] through [21] across multiple client weighting parameters simultaneously, which constitutes a mathematical calculation under MPEP 2106.04(a)(2)(I). Furthermore, distributing customized patient priority rankings concurrently to multiple hospital departments constitutes a method of organizing human activity under MPEP 2106.04(a)(2)(II), specifically managing personal behavior or relationships or interactions between people, by coordinating clinical triage operations across medical facilities. Executing mathematical scoring and patient triage concurrently on generic server hardware does not remove the judicial exceptions from the claim. Because claims 1 - 5, 8 - 10, 12-16, 20, 25, 26, 45, and 46 recite at least one judicial exception, the eligibility analysis proceeds to Step 2A, Prong Two. Step 2A, Prong Two Prong Two evaluates whether the additional elements, individually and as an ordered combination, integrate the abstract process of patient-prioritization process identified in Prong One into a practical application. The additional elements do not improve computer technology, apply the exception through a particular machine, transform an article, or otherwise meaningfully limit the exception. Independent Claims 1, 26, and 45 Claims 1, 26, and 45 recite substantially the same additional elements and computer arrangement. Claim 1 includes model training, claim 26 principally recites the inference sequence, and claim 45 principally recites weighted-score computation. Those differences concern the mathematical and patient-prioritization operations identified as the judicial exception in Prong One and do not provide a different technological application. Additional Elements The additional elements are a server and processor executing code, machine-learning models, an intermediate component, a network, client terminals and processors, displays and GUIs, and the server storing and reusing single instances of the trained models for multiple client terminals. Individual Additional-Element Evaluation The additional elements do not overcome prong two for the following reasons: Under the broadest reasonable interpretation, the server, processors, network, and client terminals access and communicate EMRs, numerical weights, scores, and rankings; execute the claimed calculations; and return the resulting information. The claims do not specify a processor design, communication protocol, network arrangement, or server configuration that changes how those components operate. The components therefore provide computer tools and a network environment for carrying out the judicial exception, rather than a particular machine that meaningfully limits the exception. MPEP §§ 2106.05(b), (f), and (h). The machine-learning models and intermediate component implement neural networks, logistic regression, decision trees, boosting, and weighted-score calculations. The claims identify alternative model architectures and assign them the functions of generating, weighting, and ranking patient scores. They do not recite a changed model structure, training operation, inference mechanism, or data structure. These elements therefore execute the mathematical calculations and patient prioritization identified in Prong One without improving how the model or computer operates. MPEP §§ 2106.04(d)(1), 2106.05(a), and 2106.05(f). The displays and GUIs permit users to adjust global list weights and view the resulting scores or rankings. The claims do not change how information is displayed, how the interface processes input, or how the user navigates the interface. The GUI supplies numerical values to the weighting calculation and presents the informational result. It therefore facilitates use of the judicial exception without applying it to an improved interface or another technological process. MPEP §§ 2106.05(a) and (f). The claims’ use of single-instance storage and reuse does not improve the operation of the computer or the machine-learning models. Paragraph [0082] indicates that the claimed approach avoids collecting additional data, creating user-specific training datasets, and training separate models. These avoided activities may reduce the amount of data or processing that the system would otherwise use, but the claims do not recite any improvement to the way the processor performs computations, the way the models are trained or execute inference, or the way memory or storage is managed. Instead, the claims obtain different terminal rankings by applying different numerical global-list weights to outputs generated by otherwise unchanged models. Thus, the alleged efficiency results from performing fewer instances of the claimed training and data-generation activities, rather than from an improved computer operation, model architecture, training technique, inference mechanism, or storage technique. Paragraph [0091] similarly describes combining priority-list outputs but does not identify a technological improvement resulting from that combination. Accordingly, the single-instance storage, reuse, and combination of outputs merely reduce duplicated activity while continuing to use the mathematical prioritization process to generate informational rankings; they therefore do not integrate the judicial exception into a practical application. See MPEP §§ 2106.04(d)(1) and 2106.05(a). Ordered Combination As an ordered combination, the client terminals select numerical weights, the network communicates those weights, the server applies them through stored models, and the GUIs display different patient rankings. This arrangement changes the prioritization data supplied to unchanged models but does not change how the computer, model, network, or another technology operates. The combination therefore implements the judicial exception with networked computer components without integrating it into a practical application. MPEP §§ 2106.04(d)(1), 2106.05(a), (f), and (h). Dependent Claims Claims 2–5, 9, 10, 12, 16, 20, and 25 add only further mathematical rules, clinical-prioritization criteria, or medical-management results identified as part of the judicial exception in Prong One. Because these claims introduce no additional element beyond that exception, they require no separate Prong Two evaluation and do not alter the determination for claim 1. Claim 8 adds aggregation code executed by a processor to combine numerical scores and rank subjects. The code and processor perform the claimed mathematical aggregation but do not improve processor operation or another technology. MPEP §§ 2106.05(a) and (f). Claims 13 and 14 add training datasets supplied to machine-learning models, including resource-availability schedules and clinical-risk information. These limitations change the informational content used to train the models but do not recite a changed dataset structure, storage technique, or model-training operation. They therefore refine the patient-prioritization information without improving technology. MPEP §§ 2106.05(a), (f), and (h). Claim 15 adds code for receiving EMRs, feeding data into the models, and obtaining priority-list outputs. This code routes information through the claimed mathematical process but does not alter how the server, models, or data-processing components operate. MPEP § 2106.05(f). Claim 46 adds concurrent inference using one model instance and generation of multiple priority lists from one stored set of trained parameters. Under the broadest reasonable interpretation, concurrency identifies when the same inference calculations are performed for different terminal weights, but the claim does not recite a concurrency-control mechanism, shared-memory arrangement, processor scheduling technique, or changed inference operation. The added elements therefore increase simultaneous execution of the prioritization calculations without improving the underlying computer or model technology. MPEP §§ 2106.04(d)(1) and 2106.05(a). Claims 1–5, 8–10, 12–16, 20, 25, 26, 45, and 46 do not integrate the judicial exception into a practical application. The analysis therefore proceeds to Step 2B. Step 2B Step 2B evaluates whether the same additional elements, individually and as an ordered combination, provide significantly more than the mathematical patient-prioritization process. They do not supply an inventive concept because they implement that process with computer components without improving their operation or applying the exception to a distinct technological process. Independent Claims 1, 26, and 45 Additional Elements The additional elements do not overcome prong two for the following reasons: The additional elements remain a server and processor executing code, machine-learning models, an intermediate component, a network, client terminals and processors, displays and GUIs, and the server storing and reusing single instances of the trained models for multiple client terminals. Individual Additional-Element Evaluation The server, processors, network, and client terminals execute, communicate, and return the data used in the mathematical prioritization. The claims define these components by the information they process and the result they produce, without specifying a technological modification to their structure or operation. Their use therefore amounts to instructions to perform the exception through a computer and within a networked healthcare environment. MPEP §§ 2106.05(f) and (h). The machine-learning models and intermediate component generate scores, apply global list weights, and rank patients. These are the mathematical and triage operations identified in Prong One. The recited model labels do not add a changed training technique, inference mechanism, model structure, or computer capability beyond performing those operations. MPEP §§ 2106.04(d)(1), 2106.05(a), and (f). The GUIs and displays receive numerical weights and present the computed rankings. They do not add an interface operation that changes computer functionality or produces a technological result separate from the exception. MPEP §§ 2106.05(a) and (f). The single-instance storage and reuse limits the number of user-specific models created and stored, but the claims obtain different results by changing numerical global list weights supplied to unchanged models. The element does not alter model learning, parameter retention, inference processing, memory operation, or storage architecture. It therefore does not provide a computer-functionality improvement beyond implementing the mathematical prioritization with fewer separately trained models. MPEP §§ 2106.04(d)(1) and 2106.05(a). Ordered Combination Viewed as a whole, the claimed arrangement receives EMRs and terminal-specific weights, executes the same models, mathematically weights and ranks patient scores, and communicates different rankings to the respective terminals. The interaction among the components follows the information flow required to perform and distribute the judicial exception. Reusing one model instance changes the number of models employed, but not how the claimed computer or model operates. The ordered combination therefore does not provide significantly more than the judicial exception. Dependent Claims The dependent claims do not overcome step 2B for the following reasons: Claims 2–5, 9, 10, 12, 16, 20, and 25 add only further limitations that remain part of the judicial exception identified in Prong One. Because these claims do not add a separate technological element, they provide no additional feature to evaluate for an improvement in computer functionality or another technological field. They therefore cannot supply the significantly more required under Step 2B. MPEP §§ 2106.04(d)(1) and 2106.05. Claim 8 adds processor-executed aggregation code. That code performs the mathematical aggregation and ranking already identified as the judicial exception; it does not change the processor, improve the execution of the underlying calculations, or provide another technological function. Thus, the processor merely performs the exception in computer-readable form, which does not add an inventive concept or amount to significantly more. MPEP §§ 2106.05(a) and (f). Claims 13 and 14 add training datasets containing resource schedules, risks, and diagnoses. These limitations change the subject matter or information evaluated by the models, but they do not change how the models are trained or operate, how the computer processes the information, or how the data is stored. The added data therefore supplies information for the claimed prioritization rather than a technological improvement, and it does not provide significantly more than the exception. MPEP §§ 2106.05(a), (f), and (h). Claim 15 adds code for receiving, feeding, and obtaining EMR data and model outputs. Those operations perform ordinary data receipt, movement, and retrieval so that the computer components can execute the claimed prioritization. The claim does not specify an unconventional data-transfer technique, a changed computer architecture, or another technological result produced by those operations. Accordingly, the additional code is directed to gathering and communicating information in support of the exception, rather than adding an inventive concept. MPEP §§ 2106.05(a), (f), and (h). Claim 46 adds concurrent inference from a single model instance and a single stored parameter set. Although the claim specifies that multiple inferences occur concurrently, it does not recite any improvement to processor scheduling, shared-memory access, parameter storage, concurrency control, or inference execution. The limitation therefore changes the timing or number of instances in which the same mathematical prioritization is performed, but not the technological manner in which the computer or model operates. Considered individually and in combination with the other limitations, the claim merely applies the judicial exception using generic computer components and consequently does not provide significantly more. MPEP §§ 2106.04(d)(1), 2106.05(a), and 2106.05(f). Accordingly, claims 1–5, 8–10, 12–16, 20, 25, 26, 45, and 46, considered individually and according to their dependencies, do not recite additional elements that amount to significantly more than the judicial exception. Conclusion Applicant's amendment necessitated the new ground(s) of rejection presented in this Office action. Accordingly, THIS ACTION IS MADE FINAL. See MPEP § 706.07(a). Applicant is reminded of the extension of time policy as set forth in 37 CFR 1.136(a). A shortened statutory period for reply to this final action is set to expire THREE MONTHS from the mailing date of this action. In the event a first reply is filed within TWO MONTHS of the mailing date of this final action and the advisory action is not mailed until after the end of the THREE-MONTH shortened statutory period, then the shortened statutory period will expire on the date the advisory action is mailed, and any nonprovisional extension fee (37 CFR 1.17(a)) pursuant to 37 CFR 1.136(a) will be calculated from the mailing date of the advisory action. In no event, however, will the statutory period for reply expire later than SIX MONTHS from the mailing date of this final action. Any inquiry concerning this communication or earlier communications from the examiner should be directed to JOSHUA DAMIAN RUIZ whose telephone number is (571)272-0409. The examiner can normally be reached 0800-1800. Examiner interviews are available via telephone, in-person, and video conferencing using a USPTO supplied web-based collaboration tool. To schedule an interview, applicant is encouraged to use the USPTO Automated Interview Request (AIR) at http://www.uspto.gov/interviewpractice. If attempts to reach the examiner by telephone are unsuccessful, the examiner’s supervisor, Shahid Merchant can be reached at (571) 270-1360. The fax phone number for the organization where this application or proceeding is assigned is 571-273-8300. Information regarding the status of published or unpublished applications may be obtained from Patent Center. Unpublished application information in Patent Center is available to registered users. To file and manage patent submissions in Patent Center, visit: https://patentcenter.uspto.gov. Visit https://www.uspto.gov/patents/apply/patent-center for more information about Patent Center and https://www.uspto.gov/patents/docx for information about filing in DOCX format. For additional questions, contact the Electronic Business Center (EBC) at 866-217-9197 (toll-free). If you would like assistance from a USPTO Customer Service Representative, call 800-786-9199 (IN USA OR CANADA) or 571-272-1000. /J.D.R./Examiner, Art Unit 3684 /Shahid Merchant/Supervisory Patent Examiner, Art Unit 3684