Detailed Action
Status of Claims
This action is in reply to applicant response filed on January 26, 2026.
Claims 1, 4-5, 7-10, 12-13, 15, 18, and 20 have been amended.
Claims 2 and 16 have been cancelled.
Claims 21-22 have been added.
Claims 1, 3-15, and 17-22 are currently pending and have been examined.
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, 3-15, and 17-22 are rejected under 35 U.S.C. 101 because the claimed invention is directed to an abstract idea without significantly more.
Step 1: Determining that a claim falls within one of the four enumerated categories of patentable subject matter recited in 35 U.S.C. 101 (i.e., process, machine, manufacture, or composition of matter). (MPEP 2106.03)
Claims 15 and 17-22 describe tangible system components, thus falling within one of the four statutory classes; i.e., machine or manufacture.
Claims 1 and 3-14 recite a series of steps, thus falling within one of the four statutory classes; i.e., a process.
Step 2A, Prong 1: Whether the claims recite a judicial exception
Representative claim 1 recites:
A computer-implemented method comprising:
Applying a multi-stage risk prediction framework that executes a plurality of stage-specific machine learning models, each associated with a corresponding stage and including stage-dependent feature data;
Using that framework to monitor a prediction value of risk associated with an element as the element progresses through stages for proactive risk mitigation;
During a first stage:
receiving a first set of feature data associated with the element; and
applying a first stage machine learning model to the first set of feature data to generate the prediction value;
Upon progression to a second stage:
receiving a second set of feature data different from the first set; and
applying a second stage machine learning model to the first and second sets of feature data to generate an updated prediction value; and
Prior to progression to a third stage:
initiating performance of a mitigation action associated with the second stage based on the updated prediction value indicating the element is a high-risk element.
The claims recite an abstract idea.
The PEG groupings implicated are:
Mental processes
Certain methods of organizing human activity
Mental process grouping
The claim recites evaluating information and classifying risk from that information. That is the sort of activity that, at a high level, can be characterized as observation, evaluation, and judgment.
The following clauses are the most problematic: “monitor a prediction value of risk associated with an element as the element progresses through the plurality of stages”; “receiving … a first set of feature data”; “applying … a first stage machine learning model … to generate the prediction value”; “receiving … a second set of feature data”; “applying … a second stage machine learning model … to the first set of feature data and the second set of feature data to generate an updated prediction value”; “based on the updated prediction value indicating the element is a high-risk element”.
Why these recite a mental process
Under the 2019 PEG, a claim recites a mental process when it sets forth acts such as observation, evaluation, judgment, or opinion that can be performed in the human mind or with pen and paper, even if the claim says it is performed on a computer.
Here, stripped to its character as a whole, claim 1 is: obtaining information available at different stages, evaluating that information, updating a risk assessment, and using the updated assessment to trigger a response.
That is fundamentally an information analysis / risk stratification workflow.
The fact that the claim uses the label “machine learning model” does not, by itself, remove the claim from the mental-process grouping. The claim does not recite a specific technical model architecture, a specific training procedure, a specific inferencing pipeline, or any concrete computational mechanism that changes the character of the claim from evaluating data to classify risk.
Certain methods of organizing human activity
The claims also recite managing healthcare-related interventions based on the risk prediction.
Offending clauses “for proactive risk mitigation”, “initiating … performance of a mitigation action associated with the second stage”, In claim 11, “updating a care plan,” “addressing a gap in care,” “generating a follow-up,” “updating a discharge plan,” “contacting the element”.
Why these recite organizing human activity
These limitations are directed to managing care delivery / discharge / follow-up interactions based on a predicted risk level. That is a form of human activity management, akin to coordinating service delivery and intervention workflows. The claim is therefore not merely about computing a value; it is also about using that value to organize healthcare-management actions.
Accordingly, claim 1 recites an abstract idea, at least in the form of a mental process of evaluating staged data to classify risk; and a method of organizing human activity in the form of managing mitigation/care actions based on the classification. Because claims 15 and 20 recite the same functional substance, they likewise recite the same abstract idea.
Step 2A, Prong 2: Whether the claim integrates the exception into a practical application
Why claim 1 does not integrate the exception into a practical application
1. No improvement to computer functionality
The claim does not recite: an improvement in processor operation, a new memory structure, an improved database architecture, a specific inferencing engine, a reduced-latency or reduced-bandwidth mechanism, a new model training or deployment architecture that improves computer performance. Instead, the claim uses: “one or more processors”, stage-specific machine learning models, generalized receiving/applying/updating operations. That is classic generic computer implementation of an abstract analytic concept.
2. No sufficiently concrete improvement to another technology
The claim is framed in a healthcare context, but the actual recitations remain at a high level: receive feature data, score risk, update score, initiate mitigation action.
There is no recited: control of a medical device, modification of a treatment apparatus, sensor-driven actuation, transformation of a physical article, laboratory process, therapy delivery step.
Even the mitigation action is recited generically, and claim 11 confirms that it may be as broad as updating a care plan or contacting the patient. Those are results-oriented intervention directives, not a concrete technological implementation.
3. “Mitigation action” is too high-level to supply integration
The claim does not require a specific treatment protocol or prophylactic step. It does not say, for example, that the system: administers a defined therapy, modifies a defined device parameter, schedules and executes a particular diagnostic protocol using a machine, causes delivery of a particular medical intervention with concrete parameters. Instead, the action is simply initiated based on a high-risk indication. That is still use of an abstract analysis to inform action, not a claimed practical application in the PEG sense.
4. The staged nature of the model is still data-analytic, not technological
The recitation that: different data are available at different stages, later models use prior-stage data plus new-stage data, predictions are updated over time, does not itself integrate the exception. It still amounts to gathering more information and re-evaluating risk. That is an improved analytic scheme, but not necessarily a technological application.
5. Displaying results is insignificant extra-solution activity
Claim 14’s “generating a display” is a textbook example of insignificant extra-solution activity.
6. Field-of-use limitation
The recitation of: resource systems, admission/stay/discharge/post-discharge stages, healthcare-related features, amounts largely to a field-of-use context. Limiting an abstract idea to a healthcare setting does not integrate it into a practical application.
7. No particular machine; no transformation
The claim is not meaningfully tied to a particular machine beyond generic processors, and it does not transform an article into a different state or thing. Transforming input data into a risk score is not the sort of transformation that establishes a practical application.
Accordingly, the claims do not integrate the abstract idea into a practical application.
Step 2B: Whether the additional elements amount to “significantly more” than the judicial exception
The claims do not recite significantly more than the abstract idea.
Additional elements identified: Beyond the abstract idea itself, the claims add, at most: one or more processors, non-transitory computer readable media / memory, generic machine learning models, staged data receipt, periodic updating, initiating mitigation action, optional display generation. These additional elements, individually and as an ordered combination, do not amount to significantly more.
Under Berkheimer, if asserting that claim elements are well-understood, routine, and conventional, the record should have factual support. Here, the specification itself supplies that support.
Intrinsic support for conventional computer components
The specification describes routine computer components in generic terms, including: processor 802, memory 804, bus 808, display 810, input/output device 812, network 830, general-purpose server/client computing environments See, e.g., ¶¶ 85-99. These disclosures read as standard, off-the-shelf computing infrastructure, not specialized hardware.
Intrinsic support for conventional ML tools
The specification also describes conventional model types: neural network, regression, random forest, gradient boosting, general supervised/unsupervised/reinforcement learning
See, e.g., ¶¶ 51, 73, 100-103. Claim 3 itself confirms the generic nature of the model recitation.
Intrinsic support for routine data processing
The specification describes: data cleaning, variable selection, feature derivation, model training, model testing, model scoring, model delivery See ¶¶ 70-74. These are routine analytics pipeline functions.
Regarding depending claims 3-14, 17-19, and 21-22, the claims further narrows the abstract idea or recite additional elements previously rejected in the independent claims.
Accordingly, the claim fails to recite any improvements to another technology or technical field, improvements to the functioning of the computer itself, use of a particular machine, effecting a transformation or reduction of a particular article to a different state or thing, adding unconventional steps that confine the claim to a particular useful application, and/or meaningful limitations beyond generally linking the use of an abstract idea to a particular environment. See 84 Fed. Reg. 55. Viewed individually or as a whole, these additional claim element(s) do not provide meaningful limitation(s) to transform the abstract idea into a patent eligible application of the abstract idea such that the claim(s) amounts to significantly more than the abstract idea itself.
Therefore, Claims 1, 3-15, and 17-22 are, on the present record, ineligible under 35 U.S.C. § 101. The claims recite an abstract idea, namely evaluating stage-based data to predict risk and managing mitigation actions based on that risk. The claims do not integrate that abstract idea into a practical application because they do not recite a specific technological improvement, a particular machine in any meaningful way, a transformation, or a concrete treatment implementation. The additional elements are generic computing and conventional ML/data-processing components; the claims lack an inventive concept.
Claim Rejections - 35 USC § 103
The following is a quotation of 35 U.S.C. 103 which forms the basis for all obviousness rejections set forth in this Office action:
A patent for a claimed invention may not be obtained, notwithstanding that the claimed invention is not identically disclosed as set forth in section 102, if the differences between the claimed invention and the prior art are such that the claimed invention as a whole would have been obvious before the effective filing date of the claimed invention to a person having ordinary skill in the art to which the claimed invention pertains. Patentability shall not be negated by the manner in which the invention was made.
The factual inquiries for establishing a background for determining obviousness under 35 U.S.C. 103 are summarized as follows:
1. Determining the scope and contents of the prior art.
2. Ascertaining the differences between the prior art and the claims at issue.
3. Resolving the level of ordinary skill in the pertinent art.
4. Considering objective evidence present in the application indicating obviousness or nonobviousness.
Claims 1, 3-15, and 17-22 are rejected under 35 U.S.C. 103 as being unpatentable over Ryan et al. (US 2012/0046965 A1, hereafter Ryan) in view of Direct multi-step forecasting - Skforecast Docs – July 14, 2023. Retrieved online from << https://skforecast.org/0.9.1/user_guides/direct-multi-step-forecasting >> (hereinafter Skforecast).
As per claim 1, Ryan discloses a computer-implemented method comprising:
applying, by one or more processors, a multi-stage risk prediction framework that executes a plurality of stage-specific machine learning models, each of the plurality of stage-specific machine learning models associated with a corresponding stage of a plurality of stages and including stage-dependent feature data that is a function of the corresponding stage of the plurality of stages, to monitor a prediction value of risk associated with an element as the element progresses through the plurality of stages for proactive risk mitigation, wherein applying the multi-stage risk prediction framework includes (Ryan describes risk assessment at multiple points (admission, in-patient, discharge, post-discharge), with feature sets and risk models updated at each stage, Figs. 3A, 10, & 13, ¶¶ 38 & 76):
during a first stage of the plurality of stages, receiving, by the one or more processors, a first set of feature data associated with the element that is available at the first stage of the plurality of stages (¶ 38 “In accordance with embodiments of the present invention, when a patient is admitted to a hospital or other clinical facility, the patient's condition may be diagnosed. Based on the patient's condition, a readmission risk prediction model may be selected and used to calculate a readmission risk score that represents the probability of readmission for the patient. The readmission risk may be based on readmission within a predetermined period of time, such as within 7 days after discharge, within 30 days after discharge, within 60 days after discharge, within 90 days after discharge, etc.”. Ryan further discloses that as the patient moves through the care process, new data are collected and used, see ¶¶ 38, 42, 58, 65);
applying, by the one or more processors, a first stage machine learning model, from the plurality of stage-specific machine learning models, associated with the first stage to the first set of feature data to generate the prediction value ([0038] “a readmission risk prediction model may be selected and used to calculate a readmission risk score”), ([0042] “Clinically relevant data for the identified condition is accessed at block 204 for use as training data.”);
upon the element progressing to a second stage of the plurality of stages subsequent to the first stage, receiving, by the one or more processors, a second set of feature data associated with the element that is different from the first set of feature data and available during the second stage of the plurality of stages (Fig. 3A ‘308’ receive patient data while admitted. See also Figure 13); and
prior to the element progressing to a third stage of the plurality of stages subsequent to the second stage, initiating, by the one or more processors, performance of a mitigation action associated with the second stage based on the updated prediction value indicating the element is a high-risk element during the second stage ([0052] “The readmission risk and/or recommended treatments for the patient are presented to a clinician, as shown at block 316.”).
Ryan does not explicitly disclose, but Skforecast teaches applying, by the one or more processors, a second stage machine learning model, from the plurality of stage-specific machine learning models, associated with the second stage to the first set of feature data and the second set of feature data to generate an updated prediction value
PNG
media_image1.png
200
400
media_image1.png
Greyscale
(Skforecast Pg 1, Diagram of direct multi-step forecasting)
Therefore, it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to include the teachings of Skforecast in the system of Ryan, in order to achieve the benefits of such approach; i.e., “achieve better accuracy in certain scenarios, particularly when there are complex patterns and dependencies in the data that are difficult to capture with a single model.” (Skforecast Page 1), and since the claimed invention is merely a combination of old elements, and in the combination each element merely would have performed the same function as it did separately, and one of ordinary skill in the art would have recognized that the results of the combination were predictable.
As per claim 3, the computer-implemented method of claim 1, Ryan further discloses
wherein one or more of the first stage machine learning model or the second stage machine learning model includes one or more of a neural network, regression, random forest, or gradient boosting (¶¶ 24 & 73 teaches linear regression).
As per claim 4, the computer-implemented method of claim 1, Ryan further discloses
wherein the updated prediction value is a first updated prediction value, and the applying, by the one or more processors, the multi-stage risk prediction framework further includes: upon the element progressing to the third stage, receiving, by the one or more processors, a third set of feature data associated with the element that is different from the first and second set of feature data and available during [[a]] the third stage (¶ 56 “recalculating a readmission risk score for discharge planning purposes…the patient's readmission risk score may be used to determine whether to discharge the patient.” ¶ 58 “As noted above, the outpatient activities may include surveillance calls, appointments, as well as a number of other activities.”).
Ryan does not explicitly disclose, but Skforecast teaches applying, by the one or more processors, a third stage machine learning model, from the plurality of stage-specific machine learning models, associated with the third stage to the first set of feature data, the second set of feature data, and the third set of feature data to generate a second updated prediction value (Pg 1, Diagram of direct multi-step forecasting)
Therefore, it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to include the teachings of Skforecast in the system of Ryan, in order to achieve the benefits of such approach; i.e., “achieve better accuracy in certain scenarios, particularly when there are complex patterns and dependencies in the data that are difficult to capture with a single model.” (Skforecast Page 1), and since the claimed invention is merely a combination of old elements, and in the combination each element merely would have performed the same function as it did separately, and one of ordinary skill in the art would have recognized that the results of the combination were predictable.
As per claim 5, the computer-implemented method of claim 4, Ryan further discloses
wherein the applying, by the one or more processors, the multi-stage risk prediction framework further includes: upon the element progressing to a fourth stage of the plurality of stages subsequent to the third stage, receiving, by the one or more processors, a fourth set of feature data associated with the element that is different from the first, second, and third set of feature data and available during the fourth stage (¶ 60 “The process of performing outpatient activities and recalculating readmission risk may be repeated until the patient is readmitted or until it is determined that outpatient activities and readmission risk score monitoring is no longer necessary.”).
Ryan does not explicitly disclose, but Skforecast teaches applying, by the one or more processors, a fourth stage machine learning model, from the plurality of stage-specific machine learning models, associated with the fourth stage to the first set of feature data, the second set of feature data, the third set of feature data, and the fourth set of feature data to generate a third updated prediction value (Pg 1, Diagram of direct multi-step forecasting).
Therefore, it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to include the teachings of Skforecast in the system of Ryan, in order to achieve the benefits of such approach; i.e., “achieve better accuracy in certain scenarios, particularly when there are complex patterns and dependencies in the data that are difficult to capture with a single model.” (Skforecast Page 1), and since the claimed invention is merely a combination of old elements, and in the combination each element merely would have performed the same function as it did separately, and one of ordinary skill in the art would have recognized that the results of the combination were predictable.
As per claim 6, the computer-implemented method of claim 5, Ryan further discloses wherein:
the first stage is a stage of initial utilization of a resource system by the element, the second stage is a stage during a stay of the element in the resource system, the third stage is a stage of discharge of the element from the resource system, and the fourth stage is a stage after discharge of the element from the resource system (Figure 10 ¶¶ 56, 58, 60, & 76 “It should be noted that a readmission risk algorithm and readmission prevention strategy may be applied for patients throughout the care process. This is represented in FIG. 10, which illustrates performance of readmission risk assessment at a number of points in the care process, including pre-admission 1002, admission 1004, while an in-patient 1006, discharge 1008, post-discharge 1010, and re-entry 1012 (if the patient is readmitted). In some embodiments, the same readmission risk algorithm may be employed at the different points in time, while in other embodiments, different readmission risk algorithms may be employed at different points in time depending on factors relevant to each point in time.”).
As per claim 7, the computer-implemented method of claim 1, Ryan further discloses
wherein the first set of feature data includes one or more of element demographics, historical claims, historical prior authorizations, clinical and/or electronic medical records, distance between resource system and element, chronic condition details of the element, risk scores for each resource system, diagnostic code level, historical hospitalization-related features, prescription-related features, lifestyle features, or social determinants of health (¶¶ 38, 65-69) “In accordance with embodiments of the present invention, when a patient is admitted to a hospital or other clinical facility, the patient's condition may be diagnosed. Based on the patient's condition, a readmission risk prediction model may be selected and used to calculate a
readmission risk score that represents the probability of readmission for the patient. The readmission risk may be based on readmission within a predetermined period of time, such as
within 7 days after discharge, within 30 days after discharge, within 60 days after discharge, within 90 days after discharge, etc.”).
As per claim 8, the computer-implemented method of claim 1, Ryan further discloses
wherein the second set of feature data includes one or more of live availability of beds, resource system burnout, partial in hospital treatment, or resource system and facility related features of the element (¶47) “clinically relevant data for the patient.” Fig. 13 e.g., number of days in hospital) see also (¶¶ 38, 65-69).
As per claim 9, the computer-implemented method of claim 4, Ryan further discloses wherein the third set of feature data includes one or more of facility treatment, resource system information, facility information, or a discharge plan (¶ 58 “As noted above, the outpatient activities may include surveillance calls, appointments, as well as a number of other activities.” See also ¶¶ 38, 65-69, 76).
As per claim 10, the computer-implemented method of claim 5, Ryan further discloses
wherein the fourth set of feature data includes one or more of post-discharge follow-up visits, stress tracking, adherence to medication, claims, prior authorizations, resource system and/or facility visits, call comments, vitals of the element, or information related to a smart device of the element ([0058] “The readmission risk score may be calculated, for instance, based on additional information gathered from patient calls and appointments.”) See also ¶¶ 38, 65-69, 76).
As per claim 11, the computer-implemented method of claim 1, Ryan further discloses
wherein the mitigation action includes one or more of updating a care plan for the element in a resource system, addressing a gap in care for the element in the resource system, generating a follow-up for the element, updating a discharge plan for the element, or contacting the element (¶¶ 57 & 58 “Discharge planning may also include planning outpatient activities to be conducted after the patient is discharged. In embodiments, the patient's readmission risk score may be used in planning the outpatient activities for the patient. The outpatient activities may include performing patient monitoring, such as outpatient surveillance calls from a clinician to the patient…”).
As per claim 12, the computer-implemented method of claim 1, Ryan further discloses
wherein the second stage machine learning model is a periodic score model configured to generate the updated prediction value periodically throughout the second stage as additional second sets of feature data are received during the second stage (¶ 88 “The algorithm may be run daily or on some other schedule deemed appropriate by the healthcare facility.”).
As per claim 13, the computer-implemented method of claim 1, Ryan further discloses
wherein the prediction value of risk represents a likelihood of re-utilization of a resource system by the element (¶ 38 “readmission risk score that represents the probability of readmission for the patient.”).
As per claim 14, the computer-implemented method of claim 1, Ryan further discloses wherein the initiating, by the one or more processors, the performance of the mitigation action includes generating a display including one or more of the mitigation action or the updated prediction value (¶ 52 “The readmission risk and/or recommended treatments for the patient are presented to a clinician, as shown at block 316.”).
As per claims 15 and 17-22, these claims recite substantially similar limitations as claims 1 and 3-14 and are rejected under the same art and rationale.
Response To Arguments/Remarks
Applicant's arguments filed 1/26/2026 have been fully considered but they are not persuasive.
Regarding 101:
Applicant argues that the claims integrate any alleged abstract idea into a practical application because the claims recite a multi-stage risk prediction framework using stage-specific machine learning models with stage-dependent feature data, and because the Specification describes improved timeliness, improved accuracy, and improved readmission capture rate relative to single-point-in-time models.
Examiner’s response: The alleged improvement is to the abstract idea itself, not to computer technology or another technical field.
The claims are directed to: receiving first-stage data, applying a first-stage machine learning, model to generate a prediction, receiving second-stage data, applying a second-stage machine learning model using first- and second-stage data to generate an updated prediction, and initiating a mitigation action based on the updated prediction.
As claimed, this is still a scheme for collecting information, analyzing the information to assess risk, updating the assessment as more information becomes available, and acting on the assessment. That is an abstract idea, as set forth in the rejection.
Applicant’s asserted benefits, such as: earlier prediction, more accurate prediction, better capture rate, proactive mitigation, and improved intervention timing, are improvements in the quality of the risk analysis and decision-making, not improvements to the functioning of a computer, machine-learning engine, network, memory, database, or other technology. Stated differently, the claims improve the content of the prediction, not the technology used to generate it.
An improvement in predicting a person’s or element’s future risk based on more information over time is not, without more, a technological improvement under Step 2A, Prong Two.
The recited “multi-stage risk prediction framework” does not amount to a practical application
Applicant contends that the claims recite a particular solution because they require: a plurality of stage-specific models, stage-dependent feature data, a first-stage model applied to first-stage data, and a second-stage model applied to first- and second-stage data.
However, these limitations still define the abstract idea at a higher level of detail; they do not integrate the abstract idea into a practical application. The recited framework is still described in functional terms: what data are received, when the data are received, what model is applied, and what result is generated.
The claims do not recite: a specific machine-learning architecture, a particular model structure, a specific training procedure, a particular feature-vector construction technique, a specific model update mechanism, a specialized inferencing pipeline, a hardware-level implementation, or any technical mechanism that improves the operation of a computer or AI system itself.
Thus, the claimed “framework” remains a data-analysis framework, not a technological implementation that meaningfully limits the exception.
The machine learning limitations are still recited at a high level of generality
Applicant argues that the Office Action improperly characterized the claims at a high level of generality. That argument is not persuasive on this record.
The claims recite: “a plurality of stage-specific machine learning models”; “applying … a first stage machine learning model”; “applying … a second stage machine learning model”
But the claims do not specify: the model topology, the model parameters, the input representation, the weighting or transformation applied to the stage-specific data, the training objective, the feature selection mechanism, the thresholding method, or a concrete technical way in which the models are improved.
Dependent claims 3 and 17 further indicate that the models may be any of several generic model classes, including: neural network, regression, random forest, or gradient boosting.
Those are broad, conventional categories of models, and their inclusion underscores that the claims use machine learning as a tool to perform the abstract analysis rather than reciting a specific technological improvement to machine learning itself.
The Specification’s statements of “technical improvement” are not commensurate with the claim language
Applicant relies on paragraphs such as ¶¶ 42, 43, and 58 of the Specification as evidence of a technical improvement. These statements have been considered but are not dispositive.
While the Specification states that the disclosed approach is “non-conventional,” improves capture rate, and enables earlier intervention, the claims themselves do not recite the technical details that would support a finding of integration into a practical application.
In particular, the claims do not recite: how the framework technically improves model operation, how the system technically handles the transition between stages, how the model is updated in a non-conventional manner, how computing resources are saved, or how the architecture of the computer system is improved.
A statement in the Specification that the invention is a “technical improvement” does not by itself establish eligibility when the claim language is directed to generic data gathering and analysis.
The “mitigation action” does not integrate the abstract idea into a practical application
Claim 1 recites: “initiating … performance of a mitigation action associated with the second stage based on the updated prediction value indicating the element is a high-risk element during the second stage.”. Claim 11 further specifies that the mitigation action may include: updating a care plan, addressing a gap in care, generating a follow-up, updating a discharge plan, or contacting the element. However, these are high-level administrative, planning, or communication actions. They do not recite: a specific treatment step, a particular medical device operation, a concrete transformation of an article, or a technological control action tied to a particular machine. Accordingly, the mitigation-action limitations do not integrate the abstract idea into a practical application.
Applicant’s reliance on Ex parte Desjardins is not persuasive
Applicant cites Ex parte Desjardins and argues that the present claims should not be evaluated at too high a level of generality. That argument is not persuasive here.
First, Applicant has not established that the cited Board decision is binding authority governing the present examination. In any event, the present claims are distinguishable on their own language.
Unlike claims that recite a specific improvement to machine-learning operation itself, the present claims do not recite a concrete technical modification to model structure or function. Instead, the claims recite using different sets of data at different stages to produce updated risk predictions. That is an improvement in the informational basis for the prediction, not a claimed improvement to the machine-learning technology itself.
The present claims therefore remain unlike claims that recite a specific technical manner of improving AI model operation.
Applicant’s Step 2B arguments are not persuasive
Applicant argues that the multi-stage risk prediction framework, particularly the “non-conventional and non-generic arrangement” of the first and second stage models and their stage-dependent feature data, is not well-understood, routine, or conventional, and requests express support under MPEP 2106.07(a) if the rejection is maintained.
This argument has been considered but is not persuasive.
The Office Action is supported by the Specification itself. The rejection is expressly supported by the Specification. The Specification describes the computing environment in generic terms, including: processors, memories, buses, displays, I/O devices, communication interfaces, and networked computing systems. See, e.g., Spec. ¶¶ 85-99.
The Specification also describes the ML tools generically, including: neural networks, regression, random forest, gradient boosting, supervised learning, unsupervised learning, and reinforcement learning. See, e.g., Spec. ¶¶ 51, 73, 100-103.
The Specification further describes standard analytics pipeline operations such as: data cleaning, variable selection, derivation, model training, model testing, model scoring, and model delivery.
See Spec. ¶¶ 70-74. Accordingly, the rejection is supported in writing by the Specification and does not rest on unsupported conclusory assertion.
The claims do not recite significantly more as an ordered combination
Considered as an ordered combination, the claims merely require: receive first-stage data, apply a first-stage model, receive second-stage data, apply a second-stage model using accumulated data, generate an updated risk score, and initiate a high-level mitigation action.
This is an ordered sequence of data intake, analysis, and output action. The sequence does not add significantly more than the abstract idea itself.
The claims still do not recite: a specialized technical architecture, a non-generic hardware configuration, a specific computational improvement, a particular transformation, or a concrete machine control operation. Accordingly, Step 2B is not satisfied.
Conclusion
For the reasons above, Applicant’s arguments do not overcome the rejection.
The claimed “multi-stage risk prediction framework” and “stage-specific machine learning models” do not integrate the abstract idea into a practical application because the claims still merely recite receiving stage-dependent data, applying generic machine learning models to generate and update a risk prediction, and initiating a high-level mitigation action based on the updated prediction. The asserted improvement is to the accuracy and timing of the risk assessment itself, not to the functioning of a computer, machine-learning technology, or other technical field. The claims do not recite a specific model architecture, training technique, feature-processing mechanism, hardware implementation, or technological control operation. Further, the additional elements are well-understood, routine, and conventional, as supported by the Specification’s disclosure of generic processors, memories, networks, standard analytics pipelines, and conventional model types such as neural networks, regression, random forest, and gradient boosting. Any alleged non-conventionality resides in the abstract idea itself, i.e., the concept of updating a risk prediction using different data available at different stages.
The claims remain directed to an abstract idea, namely: evaluating staged information to determine and update a risk prediction, and initiating mitigation activity based on that prediction,
and the claims do not recite additional elements that: integrate the exception into a practical application under Step 2A, Prong Two, or amount to significantly more under Step 2B.
Therefore, the rejection of claims 1, 3-15, and 17-22 under 35 U.S.C. § 101 is maintained.
Regarding 103: Applicant argues that neither Ryan nor Skforecast, alone or in combination, teaches or suggests the following limitation (as now recited in claim 1 and similarly in claims 15 and 20): "applying, by the one or more processors, a second stage machine learning model, from the plurality of stage-specific machine learning models, associated with the second stage to the first set of feature data and the second set of feature data to generate an updated prediction value"
Applicant contends that: Ryan does not disclose applying a “second” stage-specific ML model to both the first and second sets of feature data.
Skforecast, while teaching direct multi-step forecasting with separate models per step, applies each model to the “same” predictor set (e.g., the same lags), not to an “accumulating” feature set as required by the claim.
Therefore, the combination fails to render obvious the claimed limitation.
The examiner respectfully disagrees. Ryan explicitly describes a system that updates a risk score at multiple points in a patient’s journey (admission, in-patient, discharge, post-discharge). (See, e.g., ¶¶54-58, 76, Figs. 3A, 10, 13). At each stage, Ryan teaches that new, additional data become available and are used for updated prediction: ¶ 56 ...recalculating a readmission risk score for discharge planning purposes and/or using a previously obtained readmission risk score or risk score trending in discharge planning." ¶58 ...the patient's readmission risk score is calculated after the patient has been discharged, as shown at block 326. The readmission risk score may be calculated, for instance, based on additional information gathered from patient calls and appointments." Ryan further teaches that the risk model can be re-applied at each stage using both previously available and newly available data.
Further, Skforecast teaches the use of separate models for each “step” (stage), with each model capable of using a different set of predictors (features) as appropriate for that step.
See Skforecast, "To train a Forecaster Direct a different training matrix is created for each model," and the “transformation of a time series into matrices” section. While the basic example uses the same lagged features, the framework is explicitly flexible: This type of model can be obtained with the Forecaster Direct class and can also include one or multiple exogenous variables. “...each model can be trained on different feature sets as needed for each prediction step (stage).” One of ordinary skill would recognize that, especially in a clinical context, it is logical and routine to expand the feature set at each stage as new data become available, and to train/apply the model for each stage with all data available up to that point.
The combination of Ryan and Skforecast provides clear motivation:
Skforecast teaches that using separate models for each stage/step, and including all available features for each, improves accuracy for complex, evolving scenarios ("achieve better accuracy in certain scenarios, particularly when there are complex patterns and dependencies in the data that are difficult to capture with a single model").
It would have been obvious to a person of ordinary skill to configure the system such that, at each stage, the model receives as input all available feature data (both from earlier and current stages) to produce an updated prediction value.
This is a straightforward, predictable extension of the teachings of both references.
Applicant’s argument rests on a narrow reading of Skforecast’s example, but the reference (and the field) contemplates and enables the use of expanding feature sets at each prediction step. The combination of Ryan’s multi-stage, evolving-feature risk prediction with Skforecast’s direct multi-step modeling would have led one of ordinary skill to apply each stage-specific model to all available data (previous and current stage features) to generate an updated prediction value.
Accordingly, the combination of Ryan and Skforecast still renders claim 1 (and claims 15, 20, and all dependents) obvious. Both references, in light of the knowledge of a person of ordinary skill, teach or suggest applying each stage-specific model to all available data (from prior and current stages) at each stage to generate an updated prediction value.
Conclusion
THIS ACTION IS MADE FINAL. 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 FAHD A OBEID whose telephone number is (571)270-3324. The examiner can normally be reached Monday-Friday 8:30am-5:00pm.
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.
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.
/FAHD A OBEID/Supervisory Patent Examiner, Art Unit 3627
Prosecution Insights