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 .
Status of Claims
Claims 1-4, 6, 7, 9-18 have been amended and are pending; claims 5 and 8 have been canceled.
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 1-4, 6, 7, 9-18 are rejected under 35 U.S.C. 101 because the claimed invention is directed to a judicial exception (i.e., law of nature, natural phenomenon, or an abstract idea) without significantly more.
Under Step 1, claims 1-4, 6, 7, 11-16 are directed to a method, claims 9 and 17 are directed to a system comprising one or more processors, claims 10 and 18 are directed to non-transitory computer readable medium. Therefore claims 1-4, 6, 7, 9-18 are directed to at least one statutory category of invention. (Step 1: yes.)
Under Step 2A, Prong 1, the independent claim 1 recites receiving a question associate with healthcare compliance from a user, identifying a healthcare compliance regulation document associated with the question based on a semantic similarity between the question and the healthcare compliance regulation document; identifying one or more healthcare compliance requirements represented in the healthcare compliance regulation document based on a similarity between terms of the healthcare compliance regulation document and terms of a healthcare compliance vocabulary database; recommending a decision satisfying the one or more healthcare compliance requirements to the user, wherein the compliance recommendation comprises a model trained on training data comprising a plurality of historical compliance decisions; and updating training data comprising the plurality of historical compliance decisions and the recommended decision. Here, the claim set forth or describe a judicial exception that falls under the abstract idea categories of Certain Methods of Organizing Human Activity, specifically managing personal behavior or relationships or interactions between people (including social activities, teaching, and following rules or instructions) (see MPEP § 2106.04(a)(2), subsection II).
Independent claim 11 recites receiving a healthcare compliance regulation document from a health authority website, classifying the healthcare compliance regulation document into a first category, identifying one or more healthcare compliance requirements from the healthcare compliance regulation document, classifying each healthcare compliance requirement into a second category, determining that each healthcare compliance requirement is a new healthcare compliance requirement or the updated healthcare compliance requirement, and updating training data comprising the new healthcare compliance requirement or the updated healthcare compliance requirement. Here, the claim set forth or describe a judicial exception that falls under the abstract idea categories of Certain Methods of Organizing Human Activity, specifically managing personal behavior or relationships or interactions between people (including social activities, teaching, and following rules or instructions) (see MPEP § 2106.04(a)(2), subsection II).
Dependent claims 2-4, 6, 7, 9, 10, and 13-18 serve to further limit or specify the abstract features of the respective independent claims 1 and 11, and therefore are also directed towards the same respective abstract idea.
Accordingly claims 1-4, 6, 7, 9-18 are directed to an abstract idea (Step 2, Prong One: Yes).
This judicial exception is not integrated into a practical application because the additional elements within the claims only amount to mere instructions to implement the judicial exception using a computer (MPEP § 2106.05(f)) and/or amount to no more than mere insignificant pre-/post solution activity (MPEP § 2106.05(g)).
Independent claim 1 recites the following additional elements: “a server configured to process data with trained machine learning models comprising a compliance regulation document classifier, a compliance requirement classifier, and a compliance recommendation engine,” “wherein the compliance regulation document classifier and compliance requirement classifier are each trained on respective training data comprising a respective set of training document and associated labels,” “a healthcare compliance vocabulary database,” “wherein the compliance recommendation engine comprises a trained machine learning model trained on training data stored in a decision precedent database comprising a plurality of historical compliance decisions;” and “training the trained machine learning model of the compliance recommendation engine with updated training data stored in the decision precedent database….”
In view of the specification, a compliance regulation document classifier (Spec, [035]), a compliance requirement classifier (Spec, [038]), and a compliance recommendation engine (Spec, [049]) are machine learning models recited at a high level of generality, without any technical details, and amount to no more than merely applying machine learning to the specific environment. The limitations of training a machine learning model and retraining the machine learning model based on updated training data is also merely applying machine learning techniques to the particular data. The claimed server, database and engines are no more than merely using the computer components as tools to perform the abstract idea.
Dependent claims 2, 3, 6, and 7 do not recite further additional elements. Dependent claim 4 further recites “a robotic system implemented on the server” and “a compliance requirement database” however these additional elements do not integrated into a practical application because in view of the specification (see at least [075]) the robotic system is mapped to the disclosed “robot 108” that is only described at a high level, without any particular technical details, and constitutes merely applying bot software technology to the specific abstract idea, and, similarly, the database is merely being applied as a tool to store the abstract idea data. Dependent claim 9 recites “one or more processors coupled to a memory, the processors and memory configured to perform the method of claim 1” are insufficient to integrate the abstract idea into a practical application as these additional elements amount to no more than mere instructions to apply the abstract idea. Similarly, dependent claim 10 recites “A non-transitory computer readable medium storing instructions for causing the computing system to perform the method of claim 1” is insufficient to integrate the abstract idea into a practical application as these additional elements amount to no more than mere instructions to apply the abstract idea.
Independent claim 11 recites the following additional elements: “a compliance regulation document classifier, … a robot…, wherein the compliance regulation document classifier comprises a trained machine learning model implemented on a server and is trained on training data comprising a set of training documents and associated labels;” “a compliance requirement classifier…wherein the compliance requirement classifier comprises a different trained machine learning model implemented on the server and is trained on different training data comprising a different set of training documents and associated labels;” “a database” and “training the trained machine learning model of the compliance requirement classifier with updated training data stored in the database….” In view of the specification, a compliance regulation document classifier (Spec, [035]) and a compliance requirement classifier (Spec, [038]) are machine learning models recited at a high level of generality, without any technical details, and amount to no more than merely applying machine learning to the specific environment. The limitations of training a machine learning model and retraining the machine learning model based on updated training data is also merely applying machine learning techniques to the particular data. The claimed server and database are no more than merely using the computer components as tools to perform the abstract idea. Claims 12 and 13 do not recite further additional elements than identified in claim 11. Claims 14-16 further recite “a compliance recommendation engine” however in view of the specification (as explained with regard to claim 1) this limitation amounts to no more than mere applying generic machine learning to the abstract idea. Dependent claim 17 recites “one or more processors coupled to a memory, the processors and memory configured to perform the method of claim 11” are insufficient to integrate the abstract idea into a practical application as these additional elements amount to no more than mere instructions to apply the abstract idea. Similarly, dependent claim 18 recites “A non-transitory computer readable medium storing instructions for causing the computing system to perform the method of claim 11” is insufficient to integrate the abstract idea into a practical application as these additional elements amount to no more than mere instructions to apply the abstract idea.
Accordingly, the additional claim elements do not integrate the abstract idea into a practical application (Step 2A, Prong Two: No).
Under Step 2B, the claims do not include additional elements that are sufficient to amount to significantly more than the judicial exception because the additional element individually and taken as a whole in an ordered combination do not amount to more than a mere recitation of instructions to implement the abstract idea on a computer.
When the claims are taken as a whole, as an ordered combination, the combination of limitations in claims 1 and 11 describe machine learning algorithms, described at a high level of technical generality and specifying merely the expected output, on a general server performing the steps of the abstract idea of identifying a healthcare compliance regulation document associated with a received question, identifying one or more healthcare compliance regulations between the question and data stored in a generically described database, recommending a decision outputted by a machine learning algorithm, described at a high level of technical generality and specifying merely the expected output, and updating the model based on updated training data. The Specification supports the conventionality of the generic computer components, particularly at [035], [038], [049], [075], [081]-[082], [086], [098]-[0105]. The additional elements are described at a high level of generality, with insufficient technical details to support a claim of technical improvement. Instead, the limitations of the machine learning models and engines are described by their expected output or function as applied specifically to the abstract idea of healthcare compliance determination. The present claims are similar to Recentive Analytics, Inc. v. Fox Corp., Case No. 2023-2437 (Fed. Cir. Apr. 18, 2025), wherein the steps of training a machine learning model and retraining the model based on updated information is insufficient to show a technological improvement. In Recentive Analytics, Inc. v. Fox Corp. the Federal Circuit held that “The requirements that the machine learning model be ‘iteratively trained’ or dynamically adjusted in the Machine Learning Training patents do not represent a technological improvement … Iterative training using selected training material and dynamic adjustments based on real-time changes are incident to the very nature of machine learning… (‘The way machine learning works is the inputs are defined, the model is trained, and then the algorithm is actually updated and improved over time based on the input’)”. The storing of new or updated healthcare compliance requirements in a database amounts to the storing of information in a memory and a well-understood, routine, and conventional activity, as supported by previous court decisions (see at least storing and retrieving information in memory was held as insignificant extra-solution activity, Versata Dev. Group, Inc. v. SAP Am., Inc., 793 F.3d 1306, 1334, 115 USPQ2d 1681, 1701 (Fed. Cir. 2015); OIP Techs., 788 F.3d at 1363, 115 USPQ2d at 1092-93.)
Accordingly, when viewing the claims as a whole and the claim limitations as an ordered combination, the limitations do not amount to significantly more than the abstract idea (Step 2B: No).
Claim Rejections - 35 USC § 103
The text of those sections of Title 35, U.S. Code not included in this action can be found in a prior Office action.
Claim(s) 11 – 13, 17 and 18 are rejected under 35 U.S.C. 103 as being unpatentable over Pondicherry (US 2020/0111023) in view of Snyder (US 2020/0175110).
(Claim 11) Pondicherry teaches a computer-implemented method, comprising: receiving, by a compliance regulation document classifier, a healthcare compliance regulation document harvested by a robot from a health authority website ([0042] “The web/application server 312 hosts the front end including Web Studio, Java business logic and reporting which are used for the generation of GUIs 180 that enable the data processing system 100 to receive new regulatory documents, the user input related to the new regulatory documents and to provide output via generating the regulatory dashboard 182, updating the action board, etc.” [0019] and [0029]),
wherein the compliance regulation document classifier comprises a trained machine learning model implemented on a server and is trained on training data comprising a set of training documents and associated labels ([0020], [0021] “Each of the plurality of ML models, e.g. the topic extraction model, the entity identification model and the section classification model, are implemented by different ML techniques and are trained on different subsets of data from the regulatory text corpus using different learning techniques. The topic extraction model is based on Latent Semantic Indexing (LSI) and is trained via unsupervised learning on the prior domain-specific documents in the regulatory text corpus.”);
classifying, by the compliance regulation document classifier, the healthcare compliance regulation document into a first category ([0020] “The topic extraction model automatically extracts topics from the received domain-specific document with regulatory text and identifies documents from the regulatory text corpus that are relevant to the received domain-specific document. In an example, the topics can include certain keywords…. For example, if the received domain-specific document pertains to a financial regulation, then the categories can include a government entity, a financial entity, a banking term, etc.”, [0021]);
identifying, by a compliance requirement classifier, one or more healthcare compliance requirements from the healthcare compliance regulation document ([0020] “The relevant documents can be compared with the domain-specific document to track the similarities and differences between prior regulations and the newer regulations…. A section identification model is trained to classify various portions of the domain-specific document into different sections which describe different aspects of newer regulations, such as the requirements put forth in the newer regulations, the rules of the newer regulations, the definitions provided by the newer regulations, etc.”), wherein the compliance requirement classifier comprises a different trained machine learning model implemented on the server and is trained on different training data comprising a different set of training documents and associated labels ([0021] “Each of the plurality of ML models, e.g. the topic extraction model, the entity identification model and the section classification model, are implemented by different ML techniques and are trained on different subsets of data from the regulatory text corpus using different learning techniques.”);
classifying, by the compliance requirement classifier, each healthcare compliance requirement into a second category ([0020] “In an example, the categories can depend on the domain to which the regulatory text in document pertains. For example, if the received domain-specific document pertains to a financial regulation, then the categories can include a government entity, a financial entity, a banking term, etc. A section identification model is trained to classify various portions of the domain-specific document into different sections which describe different aspects of newer regulations, such as the requirements put forth in the newer regulations, the rules of the newer regulations, the definitions provided by the newer regulations, etc.” [0041] “The section classification model 142 is trained on section training data 298 to identify the various sections within the domain-specific document 150. Again, the section identifier 140 is trained to identify different sections for a specific domain. For example, the section training data 298 pertaining to the finance domain or banking sector can include sections labelled as requirements, business rules, definitions, non-core requirements, further reading, cross-references, background, reporting requirements, audit requirements, etc. The section labels/headings may vary for different domains.”);
determining, by the compliance requirement classifier, that each healthcare compliance requirement is a new healthcare compliance requirement or an updated healthcare compliance requirement ([0020] “The relevant documents can be compared with the domain-specific document to track the similarities and differences between prior regulations and the newer regulations.” [0050] “New clauses with respect to a base regulation can also be identified via text comparison techniques between two documents.”-[0051]).
While Pondicherry describes the storing the new compliance requirement in a database or updated compliance requirement in a database ([0049]), but does not explicitly state training a trained machine learning model of the compliance requirement classifier with updated training data stored in the database.
However, Snyder teaches determining a new compliance requirement or an updated compliance requirement ([0048] “the system identifies each regulatory change. Again, a regulatory change may be a change to existing laws, rules, and regulations or the identification of new laws, rules, and regulations. Identifying the regulatory change comprises determining the issuing authority, the rule citation, the standard name of the rule, the region affected by the rule, and the new or changed text of the rule.”), in response to determining, storing, by the compliance requirement classifier, the new healthcare compliance requirement or the updated healthcare compliance requirement in a database ([0063] “an identified webpage that provides a regulatory change may be copied in webpage form and stored in a regulatory change database as a webpage document. In other embodiments, the system may copy at least a portion of the text present in the webpage and store this as the regulatory change documentation.”); and training the trained machine learning model of the compliance requirement classifier with updated training data stored in the database, the updated training data comprising the new healthcare compliance requirement or the updated healthcare compliance requirement ([0082] “the machine learning system may be configured to adjust previous rules or algorithms for assigning the sub-impact values based on deep learning and analysis of historical change management data. In this way, the machine learning system can dynamically update over time as more information is introduced to its knowledge base (e.g., the historical regulatory change database)” See also [0083]-[0085]). It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to combine the teachings of Snyder with the teachings of Pondicherry because as Snyder suggests dynamically updating (i.e. retraining) the machine learning system based on updated regulatory changes allows the system overall to become more accurate and precise over time ([0085]).
(Claim 12) Pondicherry further teaches receiving, at the compliance regulation at the compliance regulation document classifier and from a subject-matter expert (SME), a first validation for classification of the healthcare compliance regulation document into the first category ([0015], [0041], [0046-0047] “It can be appreciated that generally training of the various ML models described herein also involves testing wherein a portion of the training data is used to train the ML models while another portion of the training data or other test data is used to test the ML models. During the testing of the ML model, the output from the ML model being tested can be validated by a SME.”, and [0054]);
in response to the first validation, classifying, by the compliance regulation document classifier, the healthcare compliance regulation document into a third category ([0041] “The section training data 298 can include documents within the regulator text corpus 190 wherein the aforementioned sections are manually labelled by a SME, classifying the various sections in the regulations into the above classes.”);
receiving, at the compliance requirement classifier and from the SME, a second validation for each healthcare compliance requirement into the second category ([0047] “It can be appreciated that generally training of the various ML models described herein also involves testing wherein a portion of the training data is used to train the ML models while another portion of the training data or other test data is used to test the ML models. During the testing of the ML model, the output from the ML model being tested can be validated by a SME.”); and in response to the second validation, classifying, by the compliance requirement classifier, each healthcare compliance requirement into a fourth category ([0046-0047], [0054] “The input from the SME thus obtained is used to train the entity extractor 130 to produce the output shown in the text annotator GUI 1200. Similar GUIs can be used for training the feature selection models and the section identifier 140.”).
(Claim 13) Pondicherry further teaches extracting, by the compliance requirement classifier, at least one healthcare compliance term from the healthcare compliance regulation document ([0020] “The entity identification model is used to analyze the domain-specific document and outputs entities/keywords (nouns) from the received domain-specific document.”); providing, by the compliance requirement classifier, the at least one healthcare compliance term to a subject-matter expert (SME) ([0041] and [0054] “The input from the SME thus obtained is used to train the entity extractor 130 to produce the output shown in the text annotator GUI 1200. Similar GUIs can be used for training the feature selection models and the section identifier 140.”); and adding, by the compliance requirement classifier, the at least one healthcare compliance term to the healthcare compliance vocabulary engine database upon approval of the SME ([0054] and [0030] “The domain-specific data dictionary can include collections of terms, topics, entities, document sections and other words or phrases which can be expected to occur with regulations of that specific domain. The domain-specific data dictionary can include the laws, jurisdictions, governmental bodies, organizations, businesses, products, titles/roles of people implementing regulations within the specific domain etc. The domain-specific data dictionary can be built via unsupervised or supervised training provided to the data processing system 100 in accordance with the examples disclosed herein.”).
(Claim 17) Pondicherry further teaches a computing system comprising: one or more processors coupled to a memory, the processors and memory configured to perform the method of claim 11 (Claim 1 “An Artificial Intelligence (AI)-based regulatory data processing system comprising: one or more processors; and a non-transitory data storage comprising processor-executable instructions that are executed by one or more processors…” and [0019], [0023], [0058-0059]).
(Claim 18) Pondicherry further teaches a non-transitory computer readable medium storing instructions for causing a computing system to perform the method of claim 11 (Claim 1 “An Artificial Intelligence (AI)-based regulatory data processing system comprising: one or more processors; and a non-transitory data storage comprising processor-executable instructions that are executed by one or more processors to” and [0058], [0059]).
Claim Rejections - 35 USC § 103
Claims 14 and 15 are rejected under 35 U.S.C. 103 as being unpatentable over Pondicherry (US 2020/0111023) in view Snyder (US 2020/0175110), and in further in view of Jago et al. “Use of Artificial Intelligence in Regulatory Decision-Making” Journal of Nursing Regulation, Vol 12, Issue 3, October 2021, pages 11-19.
(Claim 14) Pondicherry teaches a server configured to process data with the compliance regulation document classifier, the compliance requirement classifier, and a compliance recommendation engine ([0030]-[0034]). Neither Pondicherry nor Snyder teach receiving a question associated with healthcare compliance from a user and recommending a decision.
However, Jago et al. describes a web-based application that includes a compliance recommendation engine (pg 13, subheading “Risk Prediction” “The tool classifies the complaint as high risk or low risk and provides the probability of the risk prediction. This prediction is achieved by using a “supervised machine leaning” approach, where prominent characteristics of cases that constituted high risk or low risk to the public are leaned from past cases that were processed manually…they allowed us to begin to build a tool that could predict decisions about complaints using mathematical calculations of risk levels and previous judgments by humans with speed and accuracy.”); receiving a question associated with healthcare compliance from a user (page 13, heading “Methods” “This allowed case managers to upload their case files to a web server.” See Figure 1 “2) Choose the data file to predict” page 13, heading “Risk Prediction” describes the question posed to the system as: is the complaint a high risk or low risk);
identifying, by the compliance regulation document classifier the machine learning model, a healthcare compliance regulation document associated with the question and the one or more healthcare compliance requirements corresponding to the healthcare compliance regulation (page 13, subheading “Relevant Standards Matching” “We also aimed to link standards or rules from regulatory codes that were relevant to the new complaint to provide more information to the case managers… The approaches we used relied on semantic text similarity (Reimers & Gurevych, 2019) and textual inference (Williams et al., 2018), which related parts of the complaint to rules. This process identified the three most relevant rules to the case under consideration (Figure 3).”);
recommending, by the compliance recommendation engine, a decision satisfying the one or more healthcare compliance requirements to the user (page 13, heading “Methods” “By clicking a row in the table, users are redirected to a results page (Figures 2 and 3) for the specific case, showing outputs of the system including the predicted risk score, the probability and confidence calculation that this score is correct, the key words used in calculating the risk score, and, for comparison examples of similar cases relevant to the current decision and the regulatory rules pertinent to the case.” It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to combine the teachings of Jago with the teachings of Pondicherry because as Jago suggests there is a need in the healthcare industry to assist and improve the workloads of healthcare practitioners and to assist regulatory staff tasked with processing complaints, particularly in screening complaints for those that are low risk to increase efficiency (see Jago page 11).
(Claim 15) Jago further teaches identifying, by the compliance requirement classifier, at least one precedent of a decision precedent database, wherein the at least one precedent comprises a former decision made by another user associated with the question (page 13, subheading “Similar Case Retrieval” “we used it to return the top three cases to users (with similarity scores and associated risk levels assessed by case managers in the past.”); and recommending, by the compliance recommendation engine, the decision satisfying the one or more healthcare compliance requirements based on the at least one precedent (page 13, heading “Methods” “By clicking a row in the table, users are redirected to a results page (Figures 2 and 3) for the specific case, showing outputs of the system including the predicted risk score, the probability and confidence calculation that this score is correct, the key words used in calculating the risk score, and, for comparison examples of similar cases relevant to the current decision and the regulatory rules pertinent to the case.”).
Claim Rejections - 35 USC § 103
Claim 16 is rejected under 35 U.S.C. 103 as being unpatentable over Pondicherry (US 2020/0111023) in view Snyder (US 2020/0175110), and in further in view of Jago et al. “Use of Artificial Intelligence in Regulatory Decision-Making” Journal of Nursing Regulation, Vol 12, Issue 3, October 2021, pages 11-19, and in further view of Levine, (US 2020/0357000).
Pondicherry does not teach identifying at least one precedent decision. However Levine teaches recommending a decision comprises: identifying, by the compliance recommendation engine, at least one precedent of the decision precedent database, wherein the at least one precedent comprises a former decision made by a former user associated with the question ([0053] “The agency that is performing the review and making the decision on the request. Previous responses 20 from the same agency may be accorded a higher weight than prior responses from other agencies within the same industry. The reviewer at the agency performing the review may also be a factor in the outcome of the review.” [0069]); identifying, by the compliance recommendation engine, at least one insight of an insight database, wherein the at least one insight is provided by another former user ([0053], [0069]); recommending, by the compliance recommendation engine, a plurality of decisions to the user based on the at least one precedent and the at least one insight ([0090]); and receiving, at the compliance recommendation engine, a selection among the plurality of decisions from the user ([0092], [0095]). It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to combine the teachings of Levine with the teachings of Pondicherry, in view of Jago, because as Jago suggests linking past complaints similar to a new complaint question helps case managers cross-check with previous judgements and improve consistency in decision making (Jago, page 13 under heading “Similar Cases Retrieval”), the details of the previous decisions from a database and recommending the decision based on the precedent information, as taught by Levine, allows for easy access to the information ([0127]) and can assist in consistency of decisions for a particular company that may have a history of improper requests ([0070]-[0083]).
Claim Rejections - 35 USC § 103
Claim(s) 1-4, 6, 7, 9 and 10 are rejected under 35 U.S.C. 103 as being unpatentable over Pondicherry (US 2020/0111023) in view of Jago et al. “Use of Artificial Intelligence in Regulatory Decision-Making” Journal of Nursing Regulation, Vol 12, Issue 3, October 2021, pages 11-19, and in further view of Levine, (US 2020/0357000).
(Claim 1) Pondicherry teaches a computer-implemented method, comprising:
receiving, by a server configured to process data with trained machine learning models comprising a compliance regulation document classifier ([0020] “The topic extraction model automatically extracts topics from the received domain-specific document with regulatory text and identifies documents from the regulatory text corpus that are relevant to the received domain-specific document.”), a compliance requirement classifier ([0034] “The section identifier 140 also includes a section feature selection model 146 and a section classification model 142. The section feature selection model 146 extracts linguistic features from textual content for which the section classification model 142 is employed for classifying various portions of the domain-specific document 150 into a plurality of predetermined sections…Other processes enabled by the data processing system 100 include objective interpretation involving identification of requirements, rules definitions and key entities with classification, building expertise in complex regulation through supervised learning and continuously evolving through machine learning. ”), and …machine learning model (Abstract “An Artificial Intelligence (AI)-based regulatory data processing system accesses a regulatory text corpus for training machine learning (ML) models including a topic extraction model, a feature selection model, an entity identification model and a section classification model.”), … wherein the compliance regulation document classifier and the compliance requirement classifier are each trained on respective training data comprising a respective set of training documents and associated labels ([0020], [0021] “Each of the plurality of ML models, e.g. the topic extraction model, the entity identification model and the section classification model, are implemented by different ML techniques and are trained on different subsets of data from the regulatory text corpus using different learning techniques. The topic extraction model is based on Latent Semantic Indexing (LSI) and is trained via unsupervised learning on the prior domain-specific documents in the regulatory text corpus.”); …
identifying, by the compliance requirement classifier, one or more healthcare compliance requirements corresponding to represented in the healthcare compliance regulation document based on a similarity between terms of the healthcare compliance regulation document and terms of a healthcare compliance vocabulary database ([0030] “A model trainer 194 can supply the appropriate training data as detailed herein to train the plurality of ML models on a regulatory text corpus 190 which is designed for improving the models' prediction confidence level. In an example, the regulatory text corpus 190 can pertain to a specific domain wherein a domain-specific data dictionary can also be built within the data processing system 100. The domain-specific data dictionary can include collections of terms, topics, entities, document sections and other words or phrases which can be expected to occur with regulations of that specific domain. The domain-specific data dictionary can include the laws, jurisdictions, governmental bodies, organizations, businesses, products, titles/roles of people implementing regulations within the specific domain etc. The domain-specific data dictionary can be built via unsupervised or supervised training provided to the data processing system 100 in accordance with the examples disclosed herein.” [0034] “the section classification model 142 is trained on labelled training data within the regulatory text corpus 190 to produce a trained classification model for identifying predetermined sections such as but not limited to, requirements, rule and definitions from an input document such as the domain-specific document 150.” [0051] “text similarity can be estimated between the identified relevant documents and the received domain-specific document 150 via techniques such as cosine similarity.”).
Pondicherry does not teach a compliance recommendation engine, receiving a question associated with healthcare compliance from a user, recommending, by the compliance recommendation engine the machine learning model, a decision satisfying the one or more healthcare compliance requirements to the user, wherein the compliance recommendation engine comprises a trained machine learning model trained on training data stored in a decision precedent database comprising a plurality of historical compliance decisions; and training the trained machine learning model of the compliance recommendation engine with updated training data stored in the decision precedent database, the updated training data comprising the plurality of historical compliance decisions and the recommended decision.
However, Jago et al. describes a web-based application that includes a compliance recommendation engine (pg 13, subheading “Risk Prediction” “The tool classifies the complaint as high risk or low risk and provides the probability of the risk prediction. This prediction is achieved by using a “supervised machine leaning” approach, where prominent characteristics of cases that constituted high risk or low risk to the public are leaned from past cases that were processed manually…they allowed us to begin to build a tool that could predict decisions about complaints using mathematical calculations of risk levels and previous judgments by humans with speed and accuracy.”);
receiving a question associated with healthcare compliance from a user (page 13, heading “Methods” “This allowed case managers to upload their case files to a web server.” See Figure 1 “2) Choose the data file to predict” page 13, heading “Risk Prediction” describes the question posed to the system as: is the complaint associated with the submitted case file a high risk or low risk);
identifying, by the compliance regulation document classifier the machine learning model, a healthcare compliance regulation document associated with the question based on a semantic similarity between the question and the healthcare compliance regulation document (page 13, subheading “Relevant Standards Matching” “We also aimed to link standards or rules from regulatory codes that were relevant to the new complaint to provide more information to the case managers… The approaches we used relied on semantic text similarity (Reimers & Gurevych, 2019) and textual inference (Williams et al., 2018), which related parts of the complaint to rules. This process identified the three most relevant rules to the case under consideration (Figure 3).”);
recommending, by the compliance recommendation engine, a decision satisfying the one or more healthcare compliance requirements to the user (page 13, heading “Methods” “By clicking a row in the table, users are redirected to a results page (Figures 2 and 3) for the specific case, showing outputs of the system including the predicted risk score, the probability and confidence calculation that this score is correct, the key words used in calculating the risk score, and, for comparison examples of similar cases relevant to the current decision and the regulatory rules pertinent to the case.”),
wherein the compliance recommendation engine comprises a trained machine learning model trained on training data stored in a decision precedent database comprising a plurality of historical compliance decisions (Page 13, subheading “Risk prediction” “This prediction is achieved by using a “supervised machine learning” approach, where prominent characteristics of cases that constituted high risk or low risk to the public are learned from past cases that were processed manually. Such data are referred to as “training data,” and they allowed use to being to build a tool that could predict decisions about complaint…”). It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to combine the teachings of Jago with the teachings of Pondicherry because as Jago suggests there is a need in the healthcare industry to assist and improve the workloads of healthcare practitioners and to assist regulatory staff tasked with processing complaints, particularly in screening complaints for those that are low risk to increase efficiency (see Jago page 11).
While Jago teaches training the trained machine learning model of the compliance recommendation engine with training data stored in the decision precedent database, the training data comprising the plurality of historical compliance decisions and the recommended decision (see page 13, subheading “Risk prediction” and pages 13-14, subheading “User Feedback” “To improve the system over time, we collected and used feedback from the users throughout the development of the tool to design and fine tune the three features previously described. Case managers provided their own judgment on risk classification of the new case, similarity to other cases, and the relevance of the rules to the given case. This feedback was especially important for identifying features of similar cases and relevant rules because there was insufficient training data to build supervised machine-learning models for them.”), Jago does not explicitly teach retraining the trained machine learning model based on updated data.
However, Levine teaches training the trained machine learning model of the compliance recommendation engine with updated training data stored in the decision precedent database, the updated training data ([0012]-[0013] “The system further includes a prediction model, which has been trained, based on historical data, to output a prediction for a new request for regulatory compliance approval….historical data comprising information extracted from prior requests for regulatory compliance approval from multiple requesting organizations and information extracted from responses to the prior requests generated by at least one regulatory body, training a prediction model to output a prediction for a new request for regulatory compliance approval, the training being based on the historical data….” And [0084] “The prediction components 128, 130, 132, 134, 136 may be updated, as needed and/or periodically, based on updates to the database 106. For example, newly-added historical data extracted from the request 12, and response 26 may be used to retrain/update the prediction model(s).”) It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to combine the teachings of Levine with the teachings of Pondicherry in view of Jago et al. because as Levine suggests the effectiveness of an algorithm can be improved with feedback loops ([0084]).
(Claim 2) Pondicherry does not teach identifying at least one precedent decision. However, Levine teaches identifying, by the compliance recommendation engine, at least one precedent of the decision precedent database, wherein the at least one precedent comprises a former decision made by another user associated with the question ([0049]); and recommending, by the compliance recommendation engine, the decision satisfying the one or more healthcare compliance requirements based on the at least one precedent ([0049], [0060], [0069]). It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to combine the teachings of Levine with the teachings of Pondicherry, in view of Jago, because as Jago suggests linking past complaints similar to a new complaint question helps case managers cross-check with previous judgements and improve consistency in decision making (Jago, page 13 under heading “Similar Cases Retrieval”), the details of the previous decisions from a database and recommending the decision based on the precedent information, as taught by Levine, allows for easy access to the information ([0127]) and can assist in consistency of decisions for a particular company that may have a history of improper requests ([0070]-[0083]).
(Claim 3) Pondicherry does not teach identifying at least one precedent decision. However Levine teaches recommending a decision comprises: identifying, by the compliance recommendation engine, at least one precedent of the decision precedent database, wherein the at least one precedent comprises a former decision made by a former user associated with the question ([0053] “The agency that is performing the review and making the decision on the request. Previous responses 20 from the same agency may be accorded a higher weight than prior responses from other agencies within the same industry. The reviewer at the agency performing the review may also be a factor in the outcome of the review.” [0069]); identifying, by the compliance recommendation engine, at least one insight of an insight database, wherein the at least one insight is provided by another former user ([0053], [0069]); recommending, by the compliance recommendation engine, a plurality of decisions to the user based on the at least one precedent and the at least one insight ([0090]); and receiving, at the compliance recommendation engine, a selection among the plurality of decisions from the user ([0092], [0095]). It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to combine the teachings of Levine with the teachings of Pondicherry, in view of Jago, because as Jago suggests linking past complaints similar to a new complaint question helps case managers cross-check with previous judgements and improve consistency in decision making (Jago, page 13 under heading “Similar Cases Retrieval”), the details of the previous decisions from a database and recommending the decision based on the precedent information, as taught by Levine, allows for easy access to the information ([0127]) and can assist in consistency of decisions for a particular company that may have a history of improper requests ([0070]-[0083]).
(Claim 4) Pondicherry further teaches receiving, by a robotic system implemented on the server, the healthcare compliance regulation document harvested from a health authority website ([0042], [0019], [0029]); classifying, by the compliance regulation document classifier, the healthcare compliance regulation document into a first category ([0020]-[0021]); identifying, by the compliance requirement classifier, the one or more healthcare compliance requirements represented in the healthcare compliance regulation document ([0020]) based on the sematic similarity between the terms of the healthcare compliance regulation document and the terms of the healthcare compliance vocabulary database ([0030], [0034], [0051]); classifying, by the compliance requirement classifier, each healthcare compliance requirement into a second category ([0020], [0041]); determining, by the compliance requirement classifier, that each healthcare compliance requirement is a new healthcare compliance requirement or an updated healthcare compliance requirement ([0020], [0051]); and in response to determination, storing, by the compliance requirement classifier, the new healthcare compliance requirement or the updated healthcare compliance requirement in a compliance requirement database ([0049] storage of the comparison of regulatory documents to regulatory requirements for future use and sending notifications to users, responsible for executing the actions identified).
(Claim 6) Pondicherry further teaches receiving, at the compliance regulation document classifier and from a subject-matter expert (SME), a first validation for classification of the healthcare compliance regulation document into the first category ([0015], [0041], [0046-0047], [0054]); in response to the first validation, classifying, by the compliance regulation document classifier, the healthcare compliance regulation document into a third category ([0041]); receiving, at the compliance requirement classifier and from the SME, a second validation for each healthcare compliance requirement into the second category ([0047]); and in response to the second validation, classifying, by the compliance requirement classifier, each healthcare compliance requirement into a fourth category ([0046], [0054]).
(Claim 7) Pondicherry further teaches extracting, by the compliance requirement classifier, at least one healthcare compliance term from the healthcare compliance regulation document ([0020]); providing, by the compliance requirement classifier, the at least one healthcare compliance term to a subject-matter expert (SME) ([0041]); and adding, by the compliance requirement classifier, the at least one healthcare compliance term to the healthcare compliance vocabulary database upon approval of the SME ([0030]).
(Claim 9) Pondicherry further teaches a computing system comprising: one or more processors coupled to a memory, the processors and memory configured to perform the method of claim 1 (Claim 1, [0019], [0023], [0058-0059]).
(Claim 10) Pondicherry further teaches a non-transitory computer readable medium storing instructions for causing a computing system to perform the method of claim 1 (Claim 1, [0058], [0059]).
Response to Arguments
Applicant's arguments filed 7/01/2025 regarding the applied rejection under 35 USC 101 have been fully considered but they are not persuasive. Applicant contends the claims do not set forth or describe an abstract idea as “the management of personal behavior or relationships or interaction between people.” This argument is not persuasive as the independent claims recite the steps of identifying a healthcare compliance regulation document associated with a question based on a semantic similarity between the question and healthcare compliance regulation documentation; identifying one or more healthcare compliance requirements represented in the healthcare compliance regulation document and terms of a healthcare compliance vocabulary; recommending a decision satisfying the one or more healthcare compliance requirements to the user, the compliance recommendation comprises data comprising a plurality of historical compliance decisions. These steps are an abstract idea, specifically the sub-grouping “managing personal behavior or relationships or interactions between people" which includes social activities, teaching, and following rules or instructions. Similar to the examples in MPEP 2106.04(a)(2)(II) of BSG Tech. LLC v. Buyseasons, Inc., 899 F.3d 1281, 1286, 127 USPQ2d 1688, 1691 (Fed. Cir. 2018) and In re Meyer, 688 F.2d 789, 791-93, 215 USPQ 193, 194-96 (CCPA 1982), the present claims describe managing personal behavior of the activity of recommending a decision for healthcare compliance.
Applicant contents the amended limitations of “training the trained machine learning model of the compliance recommendation engine with updated training data stored in the decision precedent database, the updated training data comprising the plurality of historical compliance decisions and the recommended decision” is an improvement to a computational system to provide more accurate recommendations. The examiner respectfully disagrees. Analogous to the precedential decision Recentive Analytics, Inc. v. Fox Corp. Case No. 2023-2437 (Fed. Cir. Apr. 18, 2025), which found generic requirements that a machine learning model be “iteratively trained” or dynamically adjusted do not represent a technological improvement. “Iterative training using selected training material and dynamic adjustments based on real-time changes are incident to the very nature of machine learning. See, e.g., Opposition Br. 9 (“[U]sing a machine learning technique[] . . . necessarily includes [an] iterative[] training step . . . .” (internal quotation marks and citation omitted)); Transcript at 26:21–24 (“[T]he way machine learning works is the inputs are defined, the model is trained, and then the algorithm is actually updated and improved over time based on the input”).” Id at 12. Similar to Recentive Analytics, the present claims describe generally updating and retraining a trained machine learning model based. The specific data used for the training and retraining (i.e. healthcare compliance requirement) is merely specifying the field of use.
Regarding the previously applied rejection of claims 11-13 and 17-18 as anticipated under 35 USC 102 by Pondicherry, Applicant contends that Pondicherry does not “classifying, by the compliance regulation document classifier, the healthcare compliance regulation document into a first category” but describes classifying sections of a regulatory text document. This argument is not persuasive. Pondicherry describes in [0019] “According to one or more examples described herein, a regulatory data processing system is disclosed for automatically processing information in a complex regulatory document into a form that is usable by a processor-based device for automatic downstream processes.” Per paragraph [0020] Pondicherry describes “The topic extraction model automatically extracts topics from the received domain-specific document with regulatory text and identifies documents from the regulatory text corpus that are relevant to the received domain-specific document….In addition, the entities thus extracted can be classified into one or more of a plurality of categories.” The identification of topics in the document are used to classify the document to one or more of a plurality of categories.
With regard to Applicant’s contentions that Pondicherry does not disclose or render obvious the newly amended limitations of “training the trained machine learning model…”, the Examiner disagrees and refers Applicant to the applied 103 rejection above.
With regard to claims 1-10 previously rejected under 35 USC 103 as unpatentable over Levine in view of Pondicherry, Applicant contends that Pondicherry does not teach “identifying, by the compliance regulation document classifier, a healthcare compliance regulation document associated with the question based on a semantic similarity between the question and the healthcare compliance regulation document.” Specifically, Applicant argues Pondicherry receives a domain-specific regulatory text document and does not identify the document. The Examiner notes that Pondicherry describes in [0020] “The topic extraction model automatically extracts topics from the received domain-specific document with regulatory text and identifies documents from the regulatory text corpus that are relevant to the received domain-specific document. In an example, the topics can include certain keywords. The relevant documents can be compared with the domain-specific document to track the similarities and differences between prior regulations and the newer regulations. The entity identification model is used to analyze the domain-specific document and outputs entities/keywords (nouns) from the received domain-specific document. In addition, the entities thus extracted can be classified into one or more of a plurality of categories. In an example, the categories can depend on the domain to which the regulatory text in document pertains. For example, if the received domain-specific document pertains to a financial regulation, then the categories can include a government entity, a financial entity, a banking term, etc.” Pondicherry describes identifying compliance documents by topic that is determined based on semantic similarity. However, the Examiner further notes that in the currently applied 103 rejection that the prior art Jago et al. teaches the specific newly amended limitation “identifying, by the compliance regulation document classifier, a healthcare compliance regulation document associated with the question based on semantic similarity between the question and the healthcare compliance regulation document” by liking standards or rules from regulatory codes with the submitted complaint being requested if the complaint is high or low risk.
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 CHRISTINE M Tran whose telephone number is (571)272-8103. The examiner can normally be reached Mon-Fri, 9am-5pm.
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, Tariq Hafiz can be reached at 571-272-5350. 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.
CHRISTINE M. TRAN
Supervisory Patent Examiner
Art Unit 3695
/CHRISTINE M Tran/ Supervisory Patent Examiner, Art Unit 3695