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 .
Continued Examination Under 37 CFR 1.114
A request for continued examination under 37 CFR 1.114, including the fee set forth in 37 CFR 1.17(e), was filed in this application after allowance or after an Office action under Ex Parte Quayle, 25 USPQ 74, 453 O.G. 213 (Comm'r Pat. 1935). Since this application is eligible for continued examination under 37 CFR 1.114, and the fee set forth in 37 CFR 1.17(e) has been timely paid, prosecution in this application has been reopened pursuant to 37 CFR 1.114. Applicant's submission filed on 05/26/2026 has been entered.
Status of the Claims
The status of the claims as of the response filed 05/26/26, is as follows:
Claims 1, 5–10, 14–19, and 21–29 are pending in the application. Independent claims are 1, 10, and 19.
Amended claims are 1, 6, 10, 15, 19, 22, and 25- 26.
Claims 2–4, 11–13, and 20 were previously canceled. New claims are 27–29.
Claim Objections
Claims 28 and 29 are objected to as being substantially duplicative. Claim 28 recites that the removal of the second target medication from the medication listing is automatic, while claim 29 recites that the same removal is performed without human intervention. Under the broadest reasonable interpretation, in the context of processor-executed operations, automatic removal is not materially different from removal performed without human intervention. Appropriate correction is required.
Response to Arguments
Claim Objections
Applicant’s arguments, see page 24 filed date 5/26/2026, with respect to claims 24 and 26 have been fully considered and is persuasive.
Applicant argues that claims 24 and 26 were objected to as identical in scope, and that amended claim 26 now recites a different scope because it further limits the similarity feedback used to fine-tune the first vector embedding function.
Examiner agrees claims 24 and 26 no longer have identical scope, and the objection to claims 24 and 26 as substantially duplicative is withdrawn.
35 U.S.C 101 Subject Matter
Applicant’s arguments, see page 15-21 filed date 5/26/2026, with respect to Claims 1, 5–10, 14–19, and 21–29 have been fully considered and are not persuasive. The 35 U.S.C 101 Subject Matter
rejections is sustained.
Applicant argues that claim 1, generating a plurality of vector embeddings corresponding respectively to the plurality of standard medication codes and storing, in a second data repository, the plurality of vector embeddings, improves computer function because the embeddings are generated once and reused at query time without re-computation.
The Examiner respectfully disagrees because, under BRI, the cited language only requires generating numerical text representations, storing them, and later accessing them; the claim does not require computing the embeddings only once, preventing re-computation, or changing memory, cache, index, or database operation. MPEP 2111 requires BRI to be reasonable in view of the specification, not the broadest possible interpretation, and MPEP 2106.05(a), as updated after Desjardins, requires that the claim itself include the components or steps that provide the asserted improvement. The record shows vector embeddings are mathematical data because vector embedding functions are mathematical functions that map objects... into vector representations in a multi-dimensional space, and the repository is generic because a data repository 102 is any type of storage unit and/or device (Spec. paras. 61, 22). Thus, the argument is not persuasive, and the § 101 rejection is maintained.
Applicant argues that claim 1, non-standardized natural language identifiers for medications that are not resolvable to standard medication codes... through string matching or lookup-table comparison and applying a second vector embedding function, improves technology because it resolves inputs that conventional lookup or string matching cannot resolve.
The Examiner respectfully disagrees because the claim defines the data problem and applies another embedding function to the free text; it does not claim an improved NLP model, improved embedding architecture, improved tokenization technique, or improved computer operation. The specification confirms that the second embedding function converts free text into numerical vectors for downstream comparison, stating that each unmapped medication code is represented as a numerical vector or vector embedding... used for various downstream tasks such as similarity comparison, clustering, classification, or information retrieval (Spec. para. 88). Desjardins does not make a mathematical model eligible merely because it handles difficult data; the relevant inquiry is whether the claim reflects a technological improvement rather than use of the abstract idea in a field of use, and this limitation remains part of the mathematical mapping rule. Therefore, the argument is not persuasive, and the § 101 rejection is maintained.
Applicant argues that claim 1, computing a similarity measure and based at least on determining that each of the first similarity measure and the second similarity measure exceed a threshold... mapping, improves computer function because threshold-conditioned mapping gates record reconciliation on two similarity measures.
The Examiner respectfully disagrees because the cited limitation is the mathematical/evaluative rule used to decide whether mapping occurs, not an additional technological element that integrates the exception into a practical application. The specification describes the threshold as a significance cutoff for a similarity calculation: the system uses a similarity measure, e.g., cosine similarity, Euclidean distance and a similarity metric exceeding 0.9 meets the threshold (Spec. paras. 91-92). Under MPEP 2106.05(a), the claim must reflect a particular technological solution, not merely the idea of achieving a desired mapping result, and this threshold only controls when the abstract medication-code mapping is accepted. Therefore, the argument is not persuasive, and the § 101 rejection is maintained.
Applicant argues that claim 1, mapping the first target unmapped medication code to the first medication grouping, mapping the second target unmapped medication code to the first medication grouping, and removing the second target medication... as duplicative, improves other technology through a grouping-based deduplication cascade.
The Examiner respectfully disagrees because the cited cascade is medication-record classification and list editing, not a claimed improvement to EHR architecture, interoperability protocol, database structure, or computer performance. The specification states the healthcare benefit, not a computer-functionality change: Deduplication ensures that each patient has a unique and accurate record and deduplicating healthcare data helps optimize resource utilization, reducing storage requirements and improving the overall performance of information systems (Spec. paras. 125-126). Those benefits flow from applying the abstract deduplication result to healthcare records; MPEP 2106.05(h) states that limiting a judicial exception to a technological environment or field of use does not integrate the exception into a practical application. Therefore, the argument is not persuasive, and the § 101 rejection is maintained.
Applicant argues that claim 1, limitations (a)-(d), reflect improvements to the functioning of a computer, at least by improving the computer’s storage, learning, data sets and data structures, relying on the December 4, 2025 memorandum stating that Desjardins credited claims for improving the functioning of the machine learning model itself, citing reduced storage requirements, lowered system complexity, and the prevention of catastrophic forgetting.
Examiner respectfully disagrees because, under proper BRI, claim 1 does not train, adjust, or preserve knowledge of a machine-learning model; it generates embeddings, compares vectors, maps medication text to codes/groupings, and removes a duplicate medication entry. The USPTO guidance supports applicant’s legal premise only where the claim reflects the disclosed technological improvement, because the Desjardins update requires that the claim must be evaluated to ensure that the claim itself reflects the disclosed improvement in technology and that the claim include the components or steps of the invention that provide the improvement described in the specification. Here, the specification describes embeddings as mathematical data, not improved computer architecture: Vector embedding functions are mathematical functions that map objects... into vector representations in a multi-dimensional space (Spec. para. 61). Therefore, the argument is not persuasive, and the § 101 rejection is maintained.
Applicant argues that claim 1, generating a plurality of vector embeddings and storing, in a second data repository, the plurality of vector embeddings, improves data sets and structures by persisting machine-generated vector embeddings for re-use, eliminating re-computation of the standard-code embedding pass at each incoming patient record and reducing the computational and I/O cost of every reconciliation query.
Examiner respectfully disagrees because, under proper BRI, claim 1 requires storing embeddings and later accessing them, but does not require one-time computation, cache persistence, avoidance of recomputation, reduced I/O, or a new data structure. The specification confirms generic storage, stating a data repository 102 is any type of storage unit and/or device (Spec. para. 22), and describes generating standard-code embeddings as applying the same embedding technique to text attributes, not as a changed storage structure or query architecture (Spec. para. 90). Desjardins does not permit importing unclaimed performance effects into Prong Two; the claim itself must reflect the asserted improvement under BRI. Therefore, the argument is not persuasive, and the § 101 rejection is maintained.
Applicant argues that claim 1, removing the second target medication... from the listing of medications for the patient as duplicative, improves storage by removing duplicative target medications from the listing of medications for the patient based on common medication-grouping membership, thereby reducing redundant storage and downstream query work through a specific grouping cascade.
Examiner respectfully disagrees because, under proper BRI, the claim removes an entry from a patient medication listing after abstract mapping and grouping determinations; it does not recite deletion from physical memory, a changed database schema, a new storage index, or a new query engine. The specification states the benefit at the healthcare-data level: Deduplication ensures that each patient has a unique and accurate record and deduplicating healthcare data helps optimize resource utilization, reducing storage requirements and improving the overall performance of information systems (Spec. paras. 125-126). Those benefits flow from cleaner information, but the claimed additional elements do not improve how the computer stores, retrieves, or processes data in the Enfish/Desjardins sense. Therefore, the argument is not persuasive, and the § 101 rejection is maintained.
Applicant argues that claim 1 need not have the specification explicitly set forth the improvement because the specification describes the invention such that the improvement would be apparent to one of ordinary skill in the art, citing MPEP 2106.05(a) and the December 4, 2025 memorandum.
Examiner respectfully disagrees because applicant states the rule incompletely as applied to this record. The guidance does state that the specification need not explicitly set forth the improvement, but it also requires that the specification describe the invention such that the improvement would be apparent to a POSITA and that the claim itself reflect that disclosed improvement. The present specification supports medication-event deduplication using NLP, embeddings, similarity thresholds, and groupings, but the cited passages do not show a POSITA-recognized improvement to computer storage, learning, data sets, or data structures beyond using generic repositories and mathematical embeddings to improve medication reconciliation. Therefore, the argument is not persuasive, and the § 101 rejection is maintained.
Applicant argues that claim 1, non-standardized natural language identifiers for medications that are not resolvable to standard medication codes... through string matching or lookup-table comparison, affirmatively forecloses the manual-replication scenario relied on in the Office Action because the claim operates on the residue - the inputs that lookup-table and string-matching reconciliation cannot resolve and supplies a specific computer-implemented mechanism, namely a vector embedding function and threshold-gated similarity over persisted standard-code embeddings, to resolve those inputs.
Examiner respectfully disagrees because, under proper BRI, claim 1 excludes exact string matching and lookup-table comparison, but does not exclude human semantic evaluation, judgment, classification, or pen-and-paper comparison of medication meaning; nor does it recite a new computer component, model architecture, data structure, or database operation. The amended claim still requires applying embedding functions, computing similarity measures, determining whether thresholds are exceeded, mapping medication codes to standard codes and groupings, determining common grouping membership, and removing a duplicate medication entry. The specification confirms these operations are numerical comparison and classification, stating that each unmapped medication code is represented as a numerical vector or vector embedding in the high-dimensional space for similarity comparison, clustering, classification, or information retrieval (Spec. para. 88), and that the threshold is a similarity cutoff, such as cosine similarity, Euclidean distance and a similarity metric exceeding 0.9 meets the threshold (Spec. paras. 91-92). MPEP 2106 treats concepts involving observations, evaluations, judgments, and opinions as mental processes even when performed with pen and paper, and Desjardins requires a claimed technological improvement, not merely a computer-applied mathematical mechanism for resolving harder data. Therefore, the argument is not persuasive, and the § 101 rejection is maintained.
Applicant argues that claim 1, generating/storing standard-code embeddings, applying a second vector embedding function to non-standardized natural-language identifiers not resolvable through string matching or lookup-table comparison, threshold-conditioned mapping, and grouping-cascade deduplication, is not a mathematical concept applied with a computer, but a specific logical structure and process that integrates any judicial exception into a practical application, relying on Desjardins and Enfish for the proposition that software improvements may be defined by logical structures and processes.
Examiner respectfully disagrees because, under proper BRI consistent with MPEP 2111, claim 1 recites a logical medication-reconciliation process, but the relied-upon logic is the identified abstract idea itself: converting medication text into numerical vectors, computing similarity values, applying thresholds, assigning medication codes and groupings, and removing a duplicate medication entry. Desjardins supports eligibility where the claim reflects an improvement to computer functionality or another technology, but MPEP 2106.04(d) and 2106.05(a) require the additional elements, beyond the judicial exception, to integrate the exception into a practical application; merely performing mathematical comparison and classification with generic repositories and processors does not do so. The record confirms that the claimed mechanism is mathematical comparison and classification, not an improved computer structure: Vector embedding functions are mathematical functions that map objects... into vector representations in a multi-dimensional space (Spec. para. 61); each unmapped medication code is represented as a numerical vector or vector embedding for similarity comparison, clustering, classification, or information retrieval (Spec. para. 88); and the threshold is a similarity cutoff using measures such as cosine similarity, Euclidean distance and a similarity metric exceeding 0.9 meets the threshold (Spec. paras. 91-92). The specification’s stated benefit is cleaner healthcare information, not a changed processor, model architecture, storage architecture, or data structure, because the asserted reducing storage requirements and improving the overall performance of information systems flows from removing duplicate healthcare records, not from an improvement to how the computer operates (Spec. paras. 125-126). Therefore, the argument is not persuasive, and the § 101 rejection is maintained.
Applicant argues that claim 1, non-standardized natural language identifiers for medications that are not resolvable to standard medication codes... through string matching or lookup-table comparison, and the ordered combination of storing standard-code embeddings, threshold-gated similarity matching, and grouping-cascade deduplication, recites unconventional additional elements because conventional systems, including Agresta, resolve medication free text through RxNorm/RxTerms lookup tables and string normalization.
Examiner respectfully disagrees because, under proper BRI consistent with MPEP 2111, the alleged unconventional features are the identified judicial exceptions themselves, not additional elements beyond the exceptions. Claim 1 uses generic media, processors, repositories, and data sources to perform mathematical vector generation, similarity scoring, threshold comparison, code/group mapping, and duplicate-record removal. MPEP 2106.05(d) evaluates whether the additional elements are more than well-understood, routine, conventional activity, but the asserted difference from Agresta concerns the abstract reconciliation logic, not an unconventional processor, memory arrangement, database schema, API protocol, or computer architecture. The specification confirms generic implementation because a data repository 102 is any type of storage unit and/or device and the system components may be implemented in software and/or hardware (Spec. paras. 21-22), while the claimed embedding and matching operations remain mathematical comparison because Vector embedding functions are mathematical functions that map objects... into vector representations in a multi-dimensional space (Spec. para. 61). The fact that lookup-table or string-matching systems may be conventional does not make the claimed abstract alternative an inventive concept; it only identifies a different medication-reconciliation rule implemented on generic computer components. Therefore, the argument is not persuasive, and the § 101 rejection is maintained.
Applicant argues that claim 1, non-standardized natural language identifiers for medications that are not resolvable to standard medication codes... through string matching or lookup-table comparison, together with persisted standard-code embeddings, threshold-gated similarity matching, and grouping-cascade deduplication, improves computer capabilities and the technological field of automated medication-record reconciliation, similar to Amdocs, because it provides an unconventional technological solution to resolving and deduplicating medication free text that conventional lookup-table systems cannot resolve.
Examiner respectfully disagrees because, under proper BRI consistent with MPEP 2111, claim 1 improves the informational result of medication reconciliation, but does not recite an unconventional additional computer element or an unconventional arrangement of computer components like Amdocs. Step 2B asks whether additional elements beyond the judicial exception amount to significantly more, and MPEP 2106.05(d) requires evaluating whether those additional elements, individually or in combination, are more than well-understood, routine, conventional activity. Here, the alleged unconventional mechanism is the abstract reconciliation logic itself: generating numerical embeddings, comparing similarity values, applying thresholds, mapping codes to groupings, and removing a duplicate entry. The specification confirms that the computer implementation is generic because a data repository 102 is any type of storage unit and/or device and system components may be implemented in software and/or hardware (Spec. paras. 21-22), while the vector mechanism is mathematical because Vector embedding functions are mathematical functions that map objects... into vector representations in a multi-dimensional space (Spec. para. 61). The asserted field improvement is therefore limited to using mathematical comparison and classification in medication-record reconciliation; MPEP 2106.05(h) states that generally linking a judicial exception to a particular technological environment or field of use does not add significantly more. Therefore, the argument is not persuasive, and the § 101 rejection is maintained.
Applicant argues that claim 1, persisting and reusing machine-generated vector embeddings of standard medication codes, applying a vector embedding function to inputs not resolvable through string matching or lookup-table comparison, threshold-conditioned mapping, and grouping-cascade deduplication, recites meaningful limitations beyond a generic healthcare field of use and therefore adds significantly more under Step 2B.
Examiner respectfully disagrees because, under proper BRI consistent with MPEP 2111, the cited limitations are the medication-reconciliation rule itself, not additional elements beyond the judicial exceptions. Claim 1 uses the generic computer elements, namely the non-transitory media, hardware processor, data repositories, and data sources, to execute mathematical vector generation, similarity scoring, threshold comparison, code/group mapping, and duplicate-entry removal. MPEP 2106.05(d) evaluates whether additional elements amount to significantly more, but the asserted concrete mechanisms are the abstract mathematical and evaluative steps already identified in Prong One; they do not recite a particular processor architecture, memory arrangement, database schema, embedding-model architecture, indexing structure, or EHR interoperability protocol. The specification confirms generic implementation because a data repository 102 is any type of storage unit and/or device and the components may be implemented in software and/or hardware (Spec. paras. 21-22), while the claimed embedding mechanism remains mathematical because Vector embedding functions are mathematical functions that map objects... into vector representations in a multi-dimensional space (Spec. para. 61). The medication-record context does not supply the inventive concept because MPEP 2106.05(h) treats limiting an exception to a field of use or technological environment as insufficient. Therefore, the argument is not persuasive, and the § 101 rejection of claim 1 is maintained.
Applicant argues that claims 5-10, 14-19, and 21-26 are patent-eligible based on dependence from, or parallelism with, amended claim 1, and requests withdrawal of the § 101 rejection.
Examiner respectfully disagrees because the eligibility defects in claim 1 are carried into the parallel independent claims and dependent claims unless an additional claim limitation supplies a practical application or inventive concept beyond the abstract medication-code mapping, vector comparison, and deduplication logic. Claims 10 and 19 recite the same substance in method and system form, and the dependent claims merely narrow the same abstract process by reciting medication type, weighted cosine similarity, ranked or thresholded candidate codes, named embedding models, transmission to healthcare systems, treatment recommendation, feedback-based model adjustment, consolidation, API retrieval, or automatic removal. These limitations narrow the data, calculation, output, or healthcare use of the same abstract process, but do not change the generic computer implementation into significantly more. Therefore, the applicant’s request to withdraw the § 101 rejection is not persuasive, and the rejection of claims 1, 5-10, 14-19, and 21-29 is maintained.
35 U.S.C 103
Applicant’s arguments, see page 21-14 filed date 5/26/2026, with respect to Claims 1, 5–10, 14–19, and 21–29 have been fully considered and are not persuasive. The 35 U.S.C 103 rejection is sustained.
Applicant argues that claim 1, accessing medication groupings, mapping the first and second target unmapped medication codes to the first medication grouping, determining common grouping membership, and removing the second target medication as duplicative, are not taught because Hane’s vector embeddings only identify similarity and do not perform the later grouping/removal chain.
Examiner found the argument is not persuasive. The rejection does not rely on Hane alone for the full grouping/removal chain. Hane supplies the medical-code vector framework and similarity comparison; Agresta supplies the medication-reconciliation problem, multi-source medication list, RxNorm/NDC/RxTerms matching environment, and duplicate-medication handling. Under MPEP 2141 and 2143, the proper question is whether the combined teachings would have suggested the claimed sequence with predictable results, not whether Hane alone performs every step. A POSITA would have used Hane’s vector similarity to map difficult medication text to standard medication-code relationships and Agresta’s RxNorm/RxTerms reconciliation workflow to group corresponding medications and remove duplicates from the patient medication list. Therefore, the rejection is maintained. References: Hane, Abstract; col. 1, ll. 29-45; col. 2, ll. 1-25; col. 4, ll. 1-26; col. 15, l. 65-col. 16, l. 8; col. 18, ll. 19-55. Agresta, p. 1, Abstract; p. 2, Introduction and Section II.A; p. 3, Section II.B; p. 5, Section III.B; p. 7, Section IV; p. 9, Section V. and obvious rationale below
Applicant argues that Hane’s consolidation is misplaced because Hane filters alphanumeric medical-code sequences for training data, not natural-language identifiers in a patient-facing medication list, and Hane does not use a medication-grouping ontology.
Examiner found the argument is not persuasive. Hane is relied on for computer-executed duplicate filtering and vector similarity in prescription/drug-code data, not for the entire patient-facing medication-list reconciliation workflow. Agresta supplies the patient medication-list context because it teaches retrieving medications from multiple EHR, PHR, and HIT systems and combining/reconciling them into a medication list that identifies conflicts between the same or different medications. The proposed combination uses each reference for its known function: Hane detects similarity and duplicate prescription/drug-code data; Agresta reconciles duplicate medication information from multiple patient-medication sources. This is a predictable combination under MPEP 2143 and does not require bodily incorporation of Hane’s training-data process into Agresta. Therefore, the rejection is maintained. References: Hane, col. 13, ll. 44-67; col. 14, ll. 19-52; col. 15, ll. 1-24; col. 18, ll. 19-55. Agresta, p. 1, Abstract; p. 2, Introduction and Section II.A; p. 3, Section II.A; p. 7, Section IV.A; p. 9, Section V. and obvious rational below
Applicant argues that Agresta does not cure Hane because Agresta does not teach vector-embedding similarity for semantically normalizing unmapped medication identifiers into standard medication codes.
Examiner found the argument is not persuasive. Agresta does not need to teach the vector-embedding portion because Hane teaches that portion. Hane teaches an embedding vector dictionary in which each multi-dimensional vector corresponds to a medical code and teaches similarity by distance, angle, or cosine distance. Agresta teaches the missing medication-reconciliation context, including RxNorm, RxTerms, RxCUI/TTY relationships, FHIR data sources, and matching medication records from different EHR formats. The combination would have predictably used Hane’s vector similarity as the matching engine inside Agresta’s medication-reconciliation workflow. Therefore, the rejection is maintained. References: Hane, Abstract; col. 2, ll. 1-25; col. 4, ll. 1-26; col. 12, ll. 1-28; col. 15, l. 65-col. 16, l. 8; col. 18, ll. 50-55. Agresta, p. 2, Section II.B; p. 4, Section II.B; p. 6, Section III.D; p. 7, Section IV; p. 9, Section V. and obvious rational below
Applicant argues that even if Hane and Agresta were combinable, the combination still would not teach mapping normalized medication identifiers into medication groupings, determining same-group membership, and removing a duplicate medication.
Examiner found the argument is not persuasive. Agresta teaches that a medication in an EHR may be a brand name while a filled medication may be a generic, and that correct matching uses corresponding product identifiers, including RxNorm and NDC, linked through standardized mapping tables. Those standardized relationships reasonably read on medication groupings under BRI because the claim requires standard-code group membership, not Applicant’s exact grouping-engine implementation. Once Hane maps the free-text medication target to a standard code by vector similarity, Agresta’s RxNorm/NDC/RxTerms relationships provide the known grouping and duplicate-resolution path. Therefore, the rejection is maintained. References: Hane, col. 1, ll. 29-60; col. 2, ll. 1-25; col. 4, ll. 1-26; col. 16, ll. 1-20; col. 18, ll. 19-55. Agresta, p. 3, Section II.B; p. 5, Section III.B; p. 6, Section III.D; p. 7, Section IV.A; p. 9, Section V. and obvious rational below
Applicant argues that the remaining claims are patentable for the same reasons as claim 1.
Examiner found the argument is not persuasive. Applicant does not present separate reasons why the additional limitations of the remaining claims distinguish over the applied references. Because the argument for claim 1 is not persuasive and the remaining claims are argued only through claim 1, the rejection of the remaining claims is maintained. References: Hane, citations applied in the claim-by-claim rejection for the corresponding limitations.
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.
Subject Matter eligibility Rejection 35 U.S.C 101
Claims 1, 5-10, 14-19, 21-29 are rejected under 35 U.S.C. § 101 because the claimed subject matter is directed to a judicial exception (an abstract idea) without reciting additional elements that integrate the exception into a practical application, and without reciting an inventive concept amounting to significantly more than the exception itself.
Step 1: Statutory Categories Analysis
The claims are directed to statutory subject matter, encompassing the following statutory categories:
Process (Claims 10, 14-18): The language reciting "A method comprising: accessing a plurality of standard medication codes... generating a plurality of vector embeddings... and removing the second medication" defines a series of acts or steps for deduplicating medication events, aligning with the definition of a process in MPEP § 2106.03.
Machine (Claim 19): The language reciting "A system comprising: at least one device including a hardware processor; the system being configured to perform operations" describes a concrete thing consisting of functional parts and hardware, aligning with the definition of a machine in MPEP § 2106.03.
Manufacture (Claims 1, 5-9, 21-29): The language reciting "One or more non-transitory computer-readable media comprising instructions which, when executed... cause performance of operations" describes a tangible article of manufacture given a new form and utility through the encoding of computer-executable instructions, aligning with the definition of a manufacture in MPEP § 2106.03.
Having confirmed the claims are directed to statutory subject matter, the analysis proceeds to
Step 2A, Prong One.
Prong one
Step 2A, Prong One requires determining if a claim recites a judicial exception, such as an abstract idea, law of nature, or natural phenomenon. According to MPEP § 2106.04, abstract ideas are categorized into Mathematical Concepts, Certain Methods of Organizing Human Activity, and Mental Processes.
Representative Claim Recites the following non-bold abstract idea parts:
Claim 1.
One or more non-transitory computer-readable media comprising instructions which, when executed by one or more hardware processors, cause performance of operations comprising:
accessing, from a first data repository, a plurality of standard medication codes, each standard medication code being mapped to a corresponding set of attributes, each set of attributes associated with at least one medication;
generating a plurality of vector embeddings corresponding respectively to the plurality of standard medication codes, wherein generating the plurality of vector embeddings comprises:
applying a first vector embedding function to text of a first set of attributes associated with a first standard medication code of the plurality of standard medication codes for a first medication event, to generate a first vector embedding, wherein the first medication event is associated with a first medication, and
applying the first vector embedding function to text of a second set of attributes associated with a second standard medication code of the plurality of standard medication codes for a second medication event, to generate a second vector embedding, wherein the second medication event is associated with a second medication;
storing, in a second data repository, the plurality of vector embeddings corresponding respectively to the plurality of standard medication codes:
accessing patient medication data of a patient from one or more data sources to generate a listing of medications for the patient, wherein the patient medication data comprises a first target unmapped medication code corresponding to a first target medication event and a second target unmapped medication code corresponding to a second target medication event, the first target unmapped medication code comprises a first set of medication free text associated with a first target medication and the second target unmapped
medication code comprises a second set of medication free text associated with a second target medication,
wherein the first set of medication free text associated with the first target medication and the second set of medication free text associated with the second target medication each comprise non-standardized natural language identifiers for medications that are not resolvable to standard medication codes in the plurality of standard medication codes through string matching or lookup-table comparison;
applying a second vector embedding function to:(a) the first set of medication free text to generate a first target vector embedding for the first target unmapped medication code,
and (b) the second set of medication free text to generate a second target vector embedding for the second target unmapped medication code;
accessing, from the second data repository, the plurality of vector embeddings corresponding respectively to the plurality of standard medication codes;
computing a similarity measure for each of the first target vector embedding and the second target vector embedding and each of the plurality of vector embeddings to generate a plurality of similarity measures, wherein the plurality of similarity measures comprise:(a) a first similarity measure for the first target vector embedding and the first vector embedding, and
(b) a second similarity measure for the second target vector embedding and the second vector embedding;
based at least on determining that each of the first similarity measure and the second similarity measure exceed a threshold:(a) mapping the first target unmapped medication code associated with the first target medication to the first standard medication code,
and (b) mapping the second target unmapped medication code associated with the second target medication to the second standard medication code;
accessing, from a third data repository, plurality of medication groupings, wherein the first medication associated with the first standard medication code and the second medication
associated with the second standard medication code are mapped to a first medication grouping;
based at least on the mapping of the first standard medication code to the first medication grouping and the mapping of the first target unmapped medication code to the first standard medication code, mapping the first target unmapped medication code to the first medication grouping;
based at least on the mapping of the second standard medication code to the first medication grouping
and the mapping of the second target unmapped medication code to the second standard medication code, mapping the second target unmapped medication code to the first medication grouping;
based at least on the mapping of the first target unmapped medication code and the second target unmapped medication code to the first medication grouping, determining that the first target medication associated with the first target unmapped medication code and the second target medication associated with the second target unmapped medication code belong to the first medication grouping ; and
based on determining that the first target medication and the second target medication belong to the first medication grouping, removing the second target medication associated with the second target unmapped medication code from the listing of medications for the patient as duplicative of the first target medication associated with the first target unmapped medication code to generate a deduplicated listing of medications for the patient.
Claim Abstract Classification Rational
Claims 1, 10, and 19 recite mathematical concepts and mental-process as explain below.
Limitations 2, 6-9, 12, and 15-23 recite the mental-process abstract idea because, under BRI with the computer environment removed, they cover steps that can be practically performed by a human using observation, evaluation, judgment, classification, and pen-and-paper recordkeeping. Limitation 2 functionally means obtaining a reference list of standard medication codes and their attributes, like a person obtaining a drug dictionary, formulary, RxNorm report, or paper table. Limitation 6 functionally means recording or keeping the generated code representations, like writing stored values or labels on paper. Limitations 7-9 functionally mean reviewing patient medication information, making a medication list, recognizing free-text medication entries, and determining that the free text is not resolved by exact lookup or string matching. Limitation 12 functionally means retrieving the previously recorded standard-code representations. Limitations 15-23 functionally mean deciding that similarity results are sufficient, assigning free-text entries to standard codes, consulting medication group information, assigning the entries to the same medication group, determining that both target medications belong to that group, and removing one listed medication as a duplicate. These limitations match the mental-process grouping because MPEP 2106 treats concepts performable in the human mind, including observations, evaluations, judgments, and opinions, as abstract ideas, and expressly recognizes that the use of pen and paper does not remove the limitation from the mental-process grouping.
Limitations 3-5 and 10-14 recite the mathematical-concept abstract idea because, under BRI, they require converting medication text into vector embeddings and computing similarity measures between vectors. Limitation 3 functionally means creating numerical semantic representations for standard medication codes. Limitations 4 and 5 functionally mean applying a vector-embedding function to text attributes for first and second standard medication codes to generate first and second vectors. Limitations 10 and 11 functionally mean applying a vector-embedding function to first and second medication free-text entries to generate first and second target vectors. Limitations 13 and 14 functionally mean calculating numerical similarity values between target vectors and standard-code vectors. These limitations match the mathematical-concept grouping because the specification describes vector embeddings as numerical representations of text and similarity metrics as calculated measures such as cosine similarity or Euclidean distance.
Dependent Claims Analysis
Dependent claims 5-9, 14-18, and 21-29 do not change the Prong One determination because they only narrow the same medication-code mapping, vector-similarity calculation, clinical grouping, and medication-list deduplication abstract ideas recited in independent claims 1, 10, and 19.
Claims 5 and 14 recite identifying a generic medication and a name brand medication. This is a mental process because a person can classify medication entries as generic or brand-name using a formulary, drug dictionary, RxNorm record, or medication chart.
Claims 6 and 15 recite a weighted cosine similarity measure. This is a mathematical concept because it defines a numerical vector-comparison calculation used to determine similarity between medication embeddings.
Claims 7, 8, 16, and 17 recite identifying the n highest similarity measures or a subset meeting a threshold and presenting candidate standard medication codes. These limitations recite mental processes because they require ranking, filtering, and selecting candidate medication-code matches based on comparative values.
Claims 9 and 18 recite using BioWordVec fastText or SAPBERT. These limitations remain mathematical concepts because the named models generate vector representations of medication text for similarity comparison; naming the model narrows the math tool but does not add a technological improvement.
Claims 21 and 23 recite transmitting the deduplicated listing to a clinical decision-support engine or an external EHR system. These limitations add healthcare workflow and record-synchronization activity. They do not change the abstract idea because they merely route the result of the medication-deduplication process to another healthcare information system.
Claim 22 recites generating a treatment recommendation, including an adjustment, substitution, or discontinuation. This is a mental process because it is clinical judgment based on medication information. The claim does not recite administering a treatment or applying the recommendation in a concrete treatment step.
Claims 24 and 26 recite fine-tuning the vector embedding function using similarity feedback or healthcare-provider-confirmed mappings. These limitations recite mathematical model optimization because they adjust the embedding function using feedback data. The healthcare-provider source of the feedback only limits the data source; it does not change the mathematical character of the limitation.
Claim 25 recites consolidating prescription quantities and refill data into a unified medication record. This is a mental process and healthcare record-management activity because a person performing medication reconciliation can combine duplicate medication records into one record.
Claim 27 recites retrieving standard medication codes using one or more APIs. This does not recite an abstract idea by itself, but it is only a generic computer-access mechanism for obtaining the reference medication-code data used in the abstract mapping process. It is treated as an additional element for Prong Two, not as a basis to change the Prong One classification.
Claims 28 and 29 recite automatic removal without human intervention. These limitations do not change the Prong One result because automating an abstract determination and removal step does not change the nature of the underlying abstract idea. The claim still determines duplicate medications by grouping and removes one listed entry.
Because claims 1, 10, and 19, and dependent claims 5-9, 14-18, and 21-29 recite mathematical concepts and mental processes under Step 2A, Prong One, the analysis proceeds to Step 2A, Prong Two.
Step 2A, Prong Two
Under Step 2A, Prong Two, the analysis evaluates whether the claim integrates any identified judicial exception into a practical application, such that the claim as a whole is not merely directed to the exception itself.
The additional elements are the claimed computer implementation features: the one or more non-transitory computer-readable media, one or more hardware processors, first data repository, second data repository, third data repository, one or more data sources, and, for claim 27, the one or more application programming interfaces used to retrieve standard medication codes.
Under BRI, these additional elements do not integrate the mathematical concepts and mental processes into a practical application.
The media and processors provide a generic execution environment for the abstract medication-code mapping, vector comparison, medication grouping, and duplicate-removal logic. The claims do not recite a particular processor architecture, memory arrangement, embedding-model architecture, database schema, API protocol, network protocol, or other computer-functionality improvement. The specification likewise describes the system components broadly, including a data repository as any type of storage unit and/or device and components (par.0022) that may be implemented in software, hardware, distributed applications, or combined machines.
The repositories, data sources, and APIs also do not integrate the exception into a practical application. They merely provide locations or access mechanisms for the medication-code data, patient-medication data, vector-embedding data, and medication-grouping data used by the abstract analysis. The specification describes patient medication data as coming from multiple sources and describes external sources as HIEs and healthcare databases, which confirms that the claim uses ordinary healthcare information repositories as the field of use for the abstract reconciliation process.
The dependent claims do not add a Prong Two integration. Claims 5, 7, 8, 14, 16, 17, 21-23, 25, 28, and 29 further narrow the clinical classification, candidate selection, healthcare workflow, record consolidation, or automation of the abstract result. Claims 6, 9, 15, 18, 24, and 26 further narrow the mathematical similarity measure, embedding model, or feedback-based tuning. These limitations were identified in Prong One as part of the judicial exceptions, and therefore they are not relied upon as additional elements that integrate the exceptions into a practical application.
Considering only the additional elements individually and in combination, the claims merely apply the identified judicial exceptions using generic computer components in a healthcare-record environment. The claims therefore do not integrate the exceptions into a practical application under Step 2A, Prong Two, and the analysis proceeds to Step 2B.
Step 2B
At Step 2B, the same additional elements are evaluated to determine whether they amount to significantly more than the judicial exceptions, individually or as an ordered combination. The additional elements do not provide an inventive concept because they perform generic computer functions of storing, retrieving, accessing, and executing instructions for the abstract medication-reconciliation analysis.
The one or more non-transitory computer-readable media and one or more hardware processors do not add significantly more. They are recited at a high level of generality and are used only to execute the abstract operations identified in Prong One. The claims do not require any non-generic hardware configuration or any improvement to processor operation, memory operation, NLP processing, or computer performance.
The first data repository, second data repository, third data repository, and one or more data sources do not add significantly more. They are used to hold or supply the data consumed by the abstract process, including standard medication codes, patient medication data, vector embeddings, and medication groupings. This is ordinary data storage and data access tied to the healthcare-record field of use, not an inventive computer implementation.
The one or more APIs recited in claim 27 also do not add significantly more. Under BRI, the API limitation only specifies a generic mechanism for retrieving standard medication-code information. The claim does not recite a new API structure, a specific interoperability protocol improvement, a security improvement, a data-transfer improvement, or any technical change to how APIs operate.
Viewed as an ordered combination, the additional elements still do not transform the claim into patent-eligible subject matter. The ordered combination is a generic computer environment used to perform the Prong One abstract sequence: obtain medication information, generate and compare vector representations, map medication codes, assign medication groupings, identify duplicate medication entries, and update the medication list. The alleged improvement is to the accuracy or usefulness of medication-record reconciliation, not to the computer components or another technology.
Accordingly, claims 1, 5-10, 14-19, and 21-29 are directed to the judicial exceptions identified in Prong One and do not recite additional elements, individually or in ordered combination, that integrate the exceptions into a practical application or amount to significantly more than the exceptions. The claims are rejected under 35 U.S.C. § 101 as being directed to patent-ineligible subject matter.
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.
Claim(s) 1, 7-10, 16-19 and 21-29 are rejected under 35 U.S.C. 103 as being unpatentable over Hane - US10891352 and further in view of Agresta – PTO-892 U
Hane teaches Claim 1.
One or more non-transitory computer-readable media comprising instructions which, when executed by one or more hardware processors, cause performance of operations comprising: (Hane, Col. 5, ll. 25-45)
accessing, from a first data repository a plurality of standard medication codes, each standard medication code being mapped to a corresponding set of attributes, each set of attributes associated with at least one medication; (Hane, abstract, Col.1, ll.29-40, 63 – Col. 2., ll. 25, Col. 12, ll. 9 –28, Col. 4, ll. 27-41, fig. 4B)
Hane’s embedding vector dictionary is read on the claimed first data repository because it is stored and later accessed as a dictionary of code-linked data.
generating a plurality of vector embeddings corresponding respectively to the plurality of standard medication codes, wherein generating the plurality of vector embeddings comprises: (Hane, abstract)
applying a first vector embedding function to text of a first set of attributes associated with a first standard medication code of the plurality of standard medication codes for a first medication event, to generate a first vector embedding, wherein the first medication event is associated with a first medication, (Hane, abstract “Hane, See at least, The computing entity generates an embedding vector dictionary comprising a plurality of multi-dimensional vectors based on a medical embedding model trained using machine learning and the one or more medical sentences. Each multi-dimensional vector corresponds to a medical code.” , Col. 2, ll1-25, Col. 12, ll. 9 –28, Col 13, ll. 1-23)
Above limitation require, transforming alphanumeric identifiers into numerical vectors by executing a mathematical process on descriptive data points that characterize a clinical encounter involving a drug. Reference Applicant Language, See at least, Standard medication codes, as referred to herein, are alphanumeric identifiers that represent medication events, par. 14. Initially, the system generates vector embeddings for the standard medication codes by applying a vector embedding function to a set of attributes associated with the standard medication codes. Applying a vector embedding function to the set of attributes includes applying the vector embedding function to text of the set of attributes, par. 15
Hane demonstrates the creation of numerical vectors where each vector specifically maps to a medical code, such as a drug or prescription code. The mathematical transformation is performed by a medical embedding model acting as the required vector embedding function which processes medical sentences to output multi-dimensional vectors. Under the Broadest Reasonable Interpretation (BRI), Hane’s medical sentences, which are disclosed as strings of alphanumeric medical codes, constitute the text of a first set of attributes characterizing a medication event. Hane explicitly links these codes to specific events such as a particular patient visit, a prescription, or a claim.
and applying the first vector embedding function to text of a second set of attributes associated with a second standard medication code of the plurality of standard medication codes for a second medication event, to generate a second vector embedding, wherein the second medication event is associated with a second medication; (Hane, abstract, Col.12, ll. 1-28)
Hane et al. teaches generating a multi-dimensional vector per medical code using the same embedding model across many codes, and expressly identifies at least two different medication-related codes tied to different medications
storing, in a second data repository, the plurality of vector embeddings corresponding respectively to the plurality of standard medication codes: (Hane, the embedding vector dictionary may be stored in memory 210, 215; the embedding vector dictionary comprises a set of medical codes and the corresponding, linked, and/or assigned multi-dimensional vectors; the embedding vector dictionary may indicate the multi-dimensional vector corresponding, linked, and/or assigned to each medical code of a plurality of medical codes, col. 15, l. 65-col. 16, l. 8; see also the embedding vector dictionary is generated as described above with respect to FIG. 4A and then stored in memory 210, 215, col. 17, ll. 1-3.)
accessing patient medication data of a patient from one or more sources to generate a listing of medications for the patient, wherein the patient medication data comprises a first target unmapped medication code corresponding to a first target medication event and a second target unmapped medication code corresponding to a second target medication event, the first target unmapped medication code ; (Hane, abstract, Col. 12, ll. 50 – 67, abstract, Col. 15, ll. 34-64, Col.1, ll. 29-45, Col. 13, ll. 44-67, Col. 14, ll. 26-39)
Hane reads on accessed source data includes a patient identifier and prescription/drug codes for a patient visit, then generates a code listing as medical sentences from the accessed codes; Hane further describes two event-level medication codes because a patient may have the prescription/drug code corresponding to the statin and the prescription/drug code corresponding to the antibiotic in the same instance, and also teaches extracting two or more medical codes to form the sentence/listing.
wherein the first set of medication (Hane, col. 11, ll. 40-67; medical sentences consist of one or more medical codes, col. 1, ll. 30-40;)
Hane’s disclosure of prescription or drug codes is read on medication-related standard codes.
applying a second vector embedding function to(Hane, abstract)
accessing, from the second data repository, the plurality of vector embeddings corresponding respectively to the plurality of standard medication codes; (Hane, Col. 16, Col. 17, ll. 1-7)
Hane’s memory 210, 215 is read on the claimed second data repository because Hane stores the embedding vector dictionary in that memory and later accesses it from that memory. Hane’s accessed dictionary is read on the claimed plurality of vector embeddings corresponding respectively to the plurality of standard medication codes because the dictionary contains multi-dimensional vectors corresponding, linked, or assigned to each medical code of a plurality of medical codes, including prescription or drug codes as medication-related codes.
computing a similarity measure for each of the first target vector embedding and second target vector embedding and each of the plurality of vector embeddings to generate a plurality of similarity measures, wherein the plurality of similarity measures comprise: (Hane, abstract)
a first similarity measure for the first target vector embedding and the first vector embedding,(Hane, Col. 2, ll. 1 – 25, Col. 4, ll. 1-20)
Hane et al. utilizes mathematical distance as a metric to evaluate the relationship between different subject records.
and (b) a second similarity measure for the second target vector embedding and the second vector embedding; (Hane, Col. 4, ll. 9-16, Col. 16, 50-67)
based at least on determining that each of the first similarity measure and the second similaritymeasure exceed a threshold: (Hane, Col. 4, ll. 1-26), col. 18, ll. 19-35)
Hane describes a computer performing distance and angle calculations between aggregate vectors to identify similarity between subjects like patients or providers.
(a) mapping the first target unmapped medication code associated with the first target medication to the first standard medication code, and (b) mapping the second target unmapped medication code associated with the second target medication to the second standard medication code; (Hane, Col.2, ll. 15 -50)
Hane describes identifying whether two vectors are similar to find related patients or to link a resubmitted insurance claim to an original claim. This process evaluates the relationship between existing data objects.
accessing, from a third data repository, plurality of medication groupings, (Col.11, ll. 45-60, Col. 12, ll. 45-60, Col. 1, ll. 40-60)
Hane’s disclosure of accessed medical-code data including prescription or drug codes, together with medical sentences and aggregate vectors formed from sets of medical codes, may be read on accessing stored medication-related groupings.
wherein the first medication associated with the first standard medication code and the second medication associated with the second standard medication code are mapped to a first ;(abstract, Col.11, ll. 45-60, Col. 12, ll. 45-60, Col. 1, ll. 30-60, Col. 2, ll. 15-25)
Hane describes medical information including prescription or drug codes, links medical codes to multidimensional vectors, and aggregates vectors corresponding to sets of medical codes or subjects for similarity analysis.
based at least on the mapping of the first standard medication code first target unmapped medication code to the first standard medication code, mapping the first target unmapped medication code ; (Col. 1, ll. 30-60, Col. 4, ll. 5-20, abstract)
Hane describes accessing medical information encoded with medical codes, including prescription or drug codes, generating an embedding vector dictionary that links medical codes to multidimensional vectors, generating aggregate vectors from those code vectors, and analyzing distances or angles between aggregate vectors to determine similarity
based at least on the mapping of the second standard medication code to the first medication grouping and the mapping of the second target unmapped medication code to the second standard medication code, (Col. 16, ll.1-20, Col.17, ll.1-20, abstract)
Hane describes accessing stored medical-code/vector correspondence and aggregating vectors for sets of medical codes that may include prescription or drug codes.
removing the second target medication associated with the second target unmapped medication code from the listing of medications for the patient as duplicative of the first target medication associated with the first target unmapped medication code to generate a deduplicated listing of medications for the patient.(Col.1, ll. 29-50, Col. 13, ll. 15-25, Col. 14, ll. 1-53)
Hane describes filtering patient-specific medical-code data, including prescription/drug-code data, to remove repeated or duplicate code entries and retain only the first instance, which supports generating a non-redundant patient-specific medical-code.
35 U.S.C Obviousness Rational:
Hane teaches the base vector-code framework because Hane accesses medical information comprising prescription or drug codes, generates medical sentences comprising one or more medical codes, generates an embedding vector dictionary in which each multi-dimensional vector corresponds to a medical code, analyzes vectors by distance or angle, and filters patient-specific sets to remove repeated/duplicate prescription/drug codes. However, Hane does not teach free-text medication normalization to a standard-code group, and group-based medication-list deduplication.
Agresta teaches the missing free-text normalization and group-based deduplication because Agresta discloses medication reconciliation that retrieves medications from multiple EHR/HIT sources and combine and reconcile medication into a medication list that identifies potential conflicts between the same and/or different medications (Agresta, pag.9 Section V). Agresta further teaches RxNorm/RxTerms standard-code normalization because RxNorm provides standard normalized names and unique identifiers, assigns an RxCUI to each concept, and uses TTYs to delineate branded and generic drug names (Agresta, p.2, section B.); RxTerms allows matching where prescriptions are recorded with more free-form names that do not match any of the provided RxNorm term type formats (p.4, section b, Agresta). Agresta also teaches the grouping and removal step because the algorithm sends the medication name to RxNorm approximate search, obtains similar RxCUIs, uses RxNorm’s all related API to return concepts related by TTY, adds concept-group names and synonyms to a duplicate set, and removes any Medication Statement with a name in that duplicate set.(Agresta, page 9, section C)
A person of ordinary skill in the art would have combined Hane with Agresta by applying Agresta’s RxNorm/RxTerms free-form medication-name normalization and RxCUI/TTY duplicate-set workflow to Hane’s medical-code vector-similarity system. Hane provides the similarity mechanism because it teaches medical information encoded with prescription or drug codes, medical sentences comprising one or more medical codes, an embedding vector dictionary where each multi-dimensional vector corresponds to a medical code, and similarity based on distance or angle between vectors. Agresta provides the reason and missing medication-reconciliation application because it teaches retrieving medications from multiple EHR/HIT sources to combine and reconcile medication into a medication list, recognizes duplicate medication-list problems, and teaches that RxTerms supports matching prescriptions recorded with more free-form names, RxNorm assigns an RxCUI to each concept and uses TTYs for brand/generic specificity, and RxNav/RxNorm APIs support approximate term search for parsing natural language or free-text entered prescriptions and identifying potential duplicates. The modification would have predictably normalized free-text medication entries into RxNorm/RxCUI standard-code relationships, used RxCUI/TTY-related concepts as the standard-code group, and removed duplicate medication statements from the patient list, because the references apply known medication-code similarity, standard-drug normalization, and medication-reconciliation duplicate removal for the same medication-data accuracy problem.
Hane in combination with Agresta teaches Claim 7.
The one or more non-transitory computer-readable media of Claim 1, wherein the operations further comprise:
identifying n highest similarity measures of the plurality of similarity measures;
Hane explicitly teaches identifying a "configurable number of one or more closest aggregate vectors, Col. 18, ll. 19-35" based on the "smallest distances". One of ordinary skill in the art would understand that in a multi-dimensional vector space, similarity and distance are inversely related; thus, identifying a configurable number of vectors with the "smallest distance" is mathematically identical to identifying "n highest similarity measures"
and presenting standard medication codes, mapped to embedding vectors that correspond to the n highest similarity measures, as candidate standard medication codes for mapping to the first target unmapped medication code.
Hane describes providing an "output identifying the investigation subject... [and] one or more identified similar or different subjects, Col. 18, ll. 44-56" to determine which subjects are "most similar". Under the Broadest Reasonable Interpretation (BRI), providing a list of similar identified subjects for the purpose of investigation is functionally identical to presenting candidates for mapping to an unmapped entry.
Hane in combination with Agresta teaches Claim 8.
The one or more non-transitory computer-readable media of Claim 1, wherein the operations further comprise: identifying a subset of similarity measures, of the plurality of similarity measures, that meet a threshold similarity measure;
Hane explicitly teaches a filtering step where the computing entity identifies "all of the aggregate vectors that satisfy a threshold distance requirement with respect to an investigation aggregate vector, Col. 18, ll. 19-44". This identifies a specific "subset" of vectors from the broader population that qualify based on a defined mathematical boundary.
and presenting standard medication codes, mapped to embedding vectors that correspond to the subset of similarity measures, as candidate standard medication codes for mapping to the first target unmapped medication code. (Hane, Col. 18, 19-53, Col. 2, ll. 43-55, Col. 1. Ll.30-45)
Hane’s disclosure of outputting aggregate vectors to identify similar subjects is functionally identical to the claimed presentation of candidate standard medication codes, as both processes select and display specific data points that satisfy a similarity threshold. Because Hane explicitly teaches that these vectors correspond to medical codes such as "prescription or drug codes," the output of vector-based subjects meeting the threshold constitutes the same functional disclosure as presenting "standard medication codes mapped to embedding vectors" for an unmapped entry.
Hane in combination with Agresta teaches Claim 9.
The one or more non-transitory computer-readable media of Claim 1, applying the second vector embedding function to the first target medication event comprises using one or more of BioWordVec fastText or Self-Alignment Pretraining for Biomedical Entity Representations (SAPBERT). (Hane, See at least, the medical embedding model may be a modified fastText, word2vec, GloVe, or other algorithm, Col. 14, ll. 53-65)
Hane in combination with Agresta teaches, Claim 21.
The one or more non-transitory computer-readable media of claim 1, wherein the operations further comprise: transmitting the deduplicated listing of medications for the patient to a clinical decision support engine for generating medication alerts or treatment recommendations. (Hane, See at least, the computing entity 200 may provide the output as input to another application and/or program operating on the computing entity 200, Col. 18, ll. 44 - 67. the predictive model may be used to predict the occurrence of a clinical event for an investigation subject based on the corresponding investigation aggregate vector. determining the medical event prediction, Col. 19 ll. 61 – Col. 20, ll. 17)
Hane demonstrates the electronic transfer of analyzed medical data, which includes filtered medication lists—to integrated software components designed to evaluate patient outcomes . The system expressly describes providing output data as input to other programs for determining a medical event prediction, such as probabilities related to specific medical codes. Hane’s disclosure of inputting patient-specific vector data into predictive models to identify clinical risks constitutes transmitting medication data to a clinical decision support engine for generating alerts and recommendations, as the technical purpose is to provide automated insight for medical intervention.
Hane in combination with Agresta teaches, Claim 22.
The one or more non-transitory computer-readable media of claim 1, wherein the operations further comprise: generating, based on the first medication grouping and the patient medication data
Hane discloses forming an "embedding vector dictionary" that links medical codes (including "prescription or drug codes") to multi-dimensional vectors, effectively grouping them in a mathematical space.
Hane utilizes "medical sentences" consisting of "one or more medical codes" extracted from a patient identifier's history as the primary data input.
Hane teaches identifying "similar" or "different" clinical events based on the distance between vectors, which serves as the technical foundation for identifying discrepancies .
However does not disclosed “a treatment recommendation comprising a suggested (a) adjustment, (b) substitution, or (c) discontinuation of at least one medication in the listing of medications for the patient.”
Agresta teaches a suggested (a) adjustment, (b) substitution, or (c) discontinuation of the Claim 5, that required a specialized decision-making layer to reconcile medication list conflicts. Agresta provides the missing clinical species by teaching a process to "Make clinical decisions based on the comparison, page 2" and utilizing the "CancelRx, page 2" standard to electronically terminate therapy. Specifically, Agresta’s teaching of a clinical decision to resolve a dosing error is the functional equivalent of the claimed suggested adjustment.
A skilled artisan in the art who read the Hane application would combine Agresta with Hane because both references are directed to the same field of endeavor—medical informatics and clinical data processing—and address the shared problem of resolving discrepancies within fragmented medical records to ensure patient safety.
The combination of Hane + Agresta makes obvious the full limitation "generating... a treatment recommendation comprising a suggested (a) adjustment, (b) substitution, or (c) discontinuation" because a POSITA would combine Hane’s automated vector-similarity detection with Agresta’s clinical rationalization framework using KSR Rationale (Combining prior art elements according to known methods to yield predictable results). A POSITA would implement Agresta’s clinical decision outputs as a routine configuration to make Hane’s "similar/different" alerts actionable within a healthcare workflow, because Hane seeks to identify similar medical practices and Agresta teaches an algorithm that predictably achieves that goal by resolving the discrepancies Hane identifies (Hane, See at least, “identify providers that have similar practices, Col. 4, ll. Ll. 19-26”; Agresta, See at least, “algorithm for medication reconciliation... to identify potential conflicts, page 1”).
The integration of Agresta’s specific clinical reconciliation outputs (such as discontinuation, rationalization, and adding missing medications) into Hane’s machine learning grouping engine resolves the technical ambiguity of "similarity" by providing a specific functional purpose for the distance calculations: the automated detection and correction of medication errors. A PHOSITA would be motivated to integrate these teachings because Hane provides the "how" (the computational embedding method) while Agresta provides the "what" (the clinical reconciliation logic), and the combination yields the predictable result of a high-speed, automated system capable of not just grouping data, but performing the specific clinical task of "Medication Reconciliation" with higher accuracy than manual review. This combination represents the application of a known technique (Agresta's clinical reconciliation categories) to a known device/method (Hane's vector embedding engine) to improve it in the same way (by making the data analysis actionable for patient safety).
A skilled artisan would have a Reasonable Expectation of Success because both systems are built upon the FHIR (Fast Healthcare Interoperability Resources) standard and RxNorm terminology. These protocols are specifically designed to enable interoperability between disparate clinical data sources, allowing for the predictable automation of drug "substitution" or "adjustment" by linking branded names to their generic equivalents (SCD/SBD).
Hane in combination with Agresta teaches ,Claim 23.
The one or more non-transitory computer-readable media of claim 1, wherein the operations further comprise: transmitting the deduplicated listing of medications to an external electronic health record system in . (Hane, Col. 18, ll. 43-67)
Hane teaches transmitting the deduplicated listing of medications of the Claim, transmitting the deduplicated listing of medications to an external electronic health record system in a standardized format for longitudinal patient record synchronization, that required the electronic movement of medication data from a processing engine to a remote destination. Hane describes a computing entity that provides its output (identified similar/deduplicated medical codes) via a network interface to other applications or programs using wired transmission protocols. (Hane, See at least, “the computing entity 200 may provide (e.g., transmit) the output via a network interface 220,... and/or provide the output as input to another application and/or program... Col. 18, 43-67 Such communication may be executed using a wired data transmission protocol, such as... Ethernet” [Col. 8, lines 51-67]). However, Hane does not describe an external electronic health record system, the standardized format (as a clinical data structure), or the specific goal of longitudinal patient record synchronization. Hane’s disclosure of "Ethernet" relates to the hardware transport layer rather than the clinical data formatting required by the claim.
Agresta teaches transmitting the... medications to an external electronic health record system in a standardized format for longitudinal patient record synchronization of Claim 23, that required the use of an interoperable data exchange protocol to ensure a patient’s medical history remains consistent across multiple provider organizations over time. Agresta teaches a FHIR-based extensible software solution for medication reconciliation which can seamlessly include new medication sources to improve the longitudinal sharing of this information across the various health IT platforms (Agresta, Abstract; Section I, page 2.). Agresta specifically utilizes the Fast Healthcare Interoperability Resources (FHIR) standard (a standardized format) to promote secure sharing of healthcare data among multiple health information technology (HIT) systems (including external EHRs) to avoid medication errors such as... duplications.
The combination of Hane + Agresta makes obvious the full limitation obvious, because it represents the combination of prior art elements according to known methods to yield predictable results (MPEP 2143, Rationale A). Hane provides the "what" (highly accurate identification of duplicates via vector similarity metrics) and Agresta provides the "how/where" (the standardized FHIR communication layer for external EHR transmission). A POSITA would integrate Hane’s deduplication output into Agresta’s reconciliation workflow to resolve the "Data Silo" problem explicitly mentioned in both arts, where disparate clinical and pharmacy information systems... often contain duplicate, missing, or inaccurate information (Agresta, Section I page 2; See also Hane, Col 1, ll 1-30). Using Hane’s vector-based "similarity" to identify duplicates for Agresta’s "longitudinal sharing" predictably results in the claimed synchronized, deduplicated listing.
A skilled Artisan in the art who read Hane’s application would combine Agresta with Hane because Hane expressly suggests providing its output as input to "another application and/or program" (Hane, Col 18, ll 47-67) but identifies that its medical codes are non-interpretable, alpha-numeric strings (Hane, Col 1, ll 1-40). The artisan would be motivated to integrate Agresta’s standardized mapping (using RxNorm and FHIR) because Hane’s raw vector data is clinically unusable without being formatted into the standardized format taught by Agresta, which is designed exactly for this type of integration (MPEP 2143, VI). By combining these, the artisan achieves a longitudinal patient record synchronization that is human-readable and clinically actionable within an external EHR. The artisan would have a Reasonable Expectation of Success because Agresta confirms that FHIR is web-based and free for use, and allows... interoperability features (Agresta, Section II.C, page 4), making the transmission of processed data from Hane to an EHR a matter of routine system integration using established APIs.
Hane in combination with Agresta teaches, Claim 24.
The one or more non-transitory computer-readable media of claim 1, wherein generating the plurality of vector embeddings further comprises: fine-tuning the first vector embedding function based on similarity feedback from previously confirmed mappings.
Hane teaches generating the plurality of vector embeddings of Claim 24, that required converting medical codes into high-dimensional numerical representations using a machine learning model. Hane describes a computing entity that generates an embedding vector dictionary comprising a plurality of multi-dimensional vectors based on a medical embedding model trained using machine learning (Hane, abstract, Col. 2, ll. 8-12). Hane further demonstrates using these vectors to identify two or more aggregate vectors that are similar to resolve discrepancies such as resubmitted insurance claims (Hane, abstract, Col. 4, ll. 15-18). However, Hane does not describe fine-tuning the first vector embedding function based on similarity feedback from previously confirmed mappings.
Agresta teaches confirmed mappings of Claim 24, that required verified associations between disparate medication records to serve as a source of truth. Agresta teaches a system where medication reconciliation is the process of comparing a patient's medication orders and making clinical decisions based on the comparison. Agresta specifically notes that the final authority as to which medications are duplicates rests with the user and that the system allows for a shared reconciliation of medications where clinicians validate the accuracy of the records (Agresta, The medications are displayed reconciled in the app, allowing the user to confirm that the reconciled medications are correct (Page 8)). This validation by a clinician constitutes the confirmed mappings (verified data points) missing from Hane’s automated vector analysis.
The combination of Hane + Agresta makes obvious the full limitation [fine-tuning the first vector embedding function based on similarity feedback from previously confirmed mappings] because it involves combining prior art elements according to known methods to yield predictable results (MPEP 2143, Rationale A). Hane teaches a model trained using machine learning (Hane, Col. 12, ll. 1-15) and Agresta in page 5 teaches the generation of a best possible medication list through user-validated clinical decisions. A POSITA would be motivated to use the confirmed mappings from Agresta’s interface as similarity feedback to fine-tune Hane’s embedding function because Hane acknowledges that human medical knowledge is generally required for determining the actual similarity of a set of medical codes (Hane, Col 1, ll. 23-25). Using the "human authority" results from Agresta to refine the mathematical weights of Hane’s model is a common-sense application of "Active Learning" to improve model accuracy over time.
A skilled Artisan in the art who read Hane’s application would combine Agresta with Hane because both references seek to solve the problem of "non-interpretable" and "duplicate" medical data (Hane, Col 1, ll 12-25; Agresta, Section I, page 2). The artisan would recognize that while Hane’s vectors provide a mathematical estimate of similarity, the results can be further trained to reduce errors (Hane, Col 13, 60-67, Col.20, ll. 46-67). Agresta provides the corrective data (the confirmed mappings) necessary to perform this refinement. Integrating this feedback loop would resolve the "communication gaps" mentioned in Agresta by ensuring the processing engine (Hane) becomes more aligned with actual clinical reality through a recursive learning process. The artisan would have a Reasonable Expectation of Success because fine-tuning a model using verified labels (supervised learning) is a foundational technique in the art of machine learning.
Hane in combination with Agresta teaches, Claim 25.
The one or more non-transitory computer-readable media of claim 1, wherein removing the second medication from the listing of medications for the patient comprises: consolidating prescription quantities and refill data associated with the first medication and the second medication into a unified medication record. (Hane, Col. 14, ll. 15-35)
Hane teaches removing the second medication of the Claim 25, removing the second medication from the listing of medications for the patient comprises: consolidating prescription quantities and refill data associated with the first and second medications into a unified medication record, that required the functional exclusion of a redundant medical entry from a patient list to maintain record integrity. Hane identifies similar drug vectors and is configured to exclude subsequent occurrences of those codes to ensure the data set is clean for downstream model application. (Hane, Col. 14, ll. 15-35)
However Hane, does not describe consolidating prescription quantities and refill data associated with the first and second medications into a unified medication record.
Agresta teaches consolidating prescription quantities and refill data of the Claim 25, missing in Hane that required the clinical aggregation of numerical counts and refill instances into a single accurate health profile. Under MPEP 2111, this requires capturing cumulative properties of a drug record because failing to merge these attributes misrepresents a patient’s total medication supply. Agresta teaches a system designed to retrieve and reconcile medication lists by merging information specifically to avoid mistakes in clinical treatment. (Agresta, See at least, reconciliation is done to avoid medication errors such as omissions, duplications, dosing errors, or drug interactions and combine and reconcile medication into a medication list. Section I, page 2 and Abstract, page 1.)
The examiner applies MPEP 2143 Rationale A involving the combination of prior art elements according to known methods to yield predictable results. This rationale is utilized because the consolidation of numeric attributes is a predictable variation of the reconciliation logic taught by Agresta to ensure no clinical data is lost during the deduplication trigger identified by Hane.
The combination of Hane + Agresta applications make obvious claim 25 because a POSITA would recognize that deleting a record in Hane results in the loss of critical fulfillment data that Agresta expressly preserves to prevent patient harm. A POSITA would implement consolidating as a routine configuration to make the reconciliation technique of Agresta operate within the deduplication engine of Hane, because Hane seeks to identify similar medical codes and Agresta teaches a medication reconciliation technique that predictably achieves the goal of an accurate medication list to avoid clinical discrepancies on the same patient input.
Hane, See at least, identify two or more aggregate vectors that are similar abstract, Col. 2, lines 17-20. Agresta, See at least, reconciliation is done to avoid medication errors such as... dosing errors and accurate in order to maximize therapeutic impact. Section I, page 2 and Abstract.
A skilled Artisan in the art who read Hane application, would combine Agresta with Hane, because Hane admits that medical codes are non-interpretable, alpha-numeric strings and provides an explicit invitation for clinical improvement by stating that human medical knowledge is generally required for determining the actual similar of a set of medical codes. This admission creates a technical gap regarding clinical interpretation that Agresta’s standardized reconciliation framework is specifically designed to fill. (Hane, See at least, medical codes are non-interpretable, alpha-numeric strings and human medical knowledge is generally required. Col. 1, lines 1-30)
The integration of consolidating prescription quantities resolves the problem of information loss in Hane where discarding records creates a dosing error by misrepresenting the total medication volume available to the patient. Because the total medication volume available to the patient is an attribute of the medication list, failing to consolidate these quantities misrepresents the record. Integrating Agresta’s reconciliation ensures that the unified medication record preserves the refill data identified by Hane as similar, thereby preventing the life-threatening patient safety events that occur when a clinical history is incomplete or misrepresented. (Agresta, See at least, it is extremely important that medication lists are accurate in order to... prevent potentially life-threatening patient safety events and avoid medication errors such as... dosing errors. Abstract and Section I, page 2.)
A skilled artisan would have a Reasonable Expectation of Success because merging clinical attributes via FHIR RESTful APIs and RxNorm mapping tables is a standardized and routine interoperability function.
Hane in combination with Agresta teaches, Claim 26.
The one or more non-transitory computer-readable media of claim 1, wherein generating the plurality of vector embeddings further comprises: fine-tuning the first vector embedding function based on similarity feedback from previously confirmed mappings. wherein the similarity feedback comprises healthcare provider-confirmed mappings from a medication reconciliation workflow.
Hane teaches generating the plurality of vector embeddings using a trainable medical-code embedding model, as shown by the medical embedding model may be trained using machine learning and the one or more medical sentences may be used as a training data set to train the medical embedding model using machine learning. Hane further teaches that the embedding dictionary maps medical codes to vectors whose vector-space relationship expresses relatedness, because the distance or angle between two multi-dimensional vectors within the multi-dimensional space indicates how closely and/or strongly related the two corresponding medical codes are, and Hane identifies cosine distance as an example distance measure. Hane, col. 14, ll. 53-60; col. 15, ll. 18-24; col. 16, ll. 53-59.
However, Hane does not teach fine-tuning the first vector embedding function based on similarity feedback from previously confirmed mappings.
Agresta teaches the missing provider-confirmed medication-reconciliation feedback source, as shown by a medication service providing a single place to manage transactions add, modify, cancel, comment, validate, and reconcile and ensuring that the right prescriber validates and updates the right information on a patient’s medication list. Agresta also teaches the matching context because correctly matching them via corresponding product identifiers RxNorm, NDC linked through standardized mapping tables is identified as a task, and Agresta’s FHIR data sources evaluate de-duplication, semantic matching, and allow end users to choose the best medication list. Agresta, p. 3, Section II.A.
Agresta further supports algorithm updating because it teaches adaptive multi-use algorithms for medication reconciliation for medications from different EHRs and teaches a FHIR solution that facilitates easily updating the medication reconciliation algorithm. It also states that medication reconciliation requires an adaptive algorithm able to read and match medications from multiple HIT systems with varying structure. Agresta, p. 5, Section III.B.
A POSITA would have combined Hane with Agresta by using Agresta’s provider-validated reconciliation outcomes as feedback examples to update Hane’s machine-learned medical-code embedding function. The reason is that Hane already learns vector relationships among medical codes for similarity analysis, while Agresta identifies the medication-reconciliation problem requiring semantic matching, provider validation, and algorithm updating; using confirmed reconciliation outcomes to refine Hane’s embedding model would predictably improve later medication matching. Hane, col. 14, ll. 53-60; col. 15, ll. 18-24; Agresta, p. 3, Section II.A; p. 5, Section III.B.
Hane in combination with Agresta teaches, Claim 27.
The one or more non-transitory computer-readable media of claim 1, wherein accessing the plurality of standard medication codes comprises retrieving the plurality of standard medication codes using one or more application programming interfaces.
Hane teaches accessing the plurality of standard medication codes to the extent Hane accesses medical information encoded with medical codes, including prescription or drug codes, and generates an embedding vector dictionary in which each multi-dimensional vector corresponds to a medical code. Hane does not expressly teach retrieving those codes using APIs. Hane, Abstract; col. 1, ll. 14-20; col. 2, ll. 31-44.
Agresta teaches the missing API-based retrieval mechanism because its medication-reconciliation system uses the FHIR RESTful API and other data standards to acquire medication data from multiple electronic and human sources, and further teaches that the MedRec FHIR API aggregates and reconciles FHIR medication resources using the NDC, RxNorm, and RxTerms APIs. Agresta therefore teaches retrieving medication information and medication-code mappings through APIs for the same medication-reconciliation environment. Agresta, p. 2, Introduction; p. 7, Section IV.
A POSITA would have combined Hane with Agresta by retrieving Hane’s prescription/drug medical-code data through Agresta’s FHIR/RxNorm/NDC/RxTerms API architecture. Hane needs medical-code data to generate vector embeddings, and Agresta teaches known healthcare APIs for acquiring and reconciling medication data from multiple HIT sources. The modification would predictably provide Hane’s embedding process with standardized medication-code data using known API retrieval techniques.
Hane in combination with Agresta teaches Claim 28.
The one or more non-transitory computer-readable media of claim 1, wherein removing the second target medication associated with the second target unmapped medication code from the listing of medications for the patient as duplicative of the first target medication associated with the first target unmapped medication code to generate a deduplicated listing of medications for the patient is automatic.
Hane teaches computer-performed duplicate filtering for patient-associated prescription/drug-code data, as shown by the set of medical sentences corresponding to a patient identifier may be filtered to remove any repeated/duplicate prescription/drug codes and only the first instance of prescription/drug code corresponding to the statin may be included in the set of medical sentences corresponding to the patient/patient identifier. This supports the automatic aspect of duplicate filtering in prescription/drug-code data, but Hane applies that filtering to medical sentences used for embedding training, not to removal of a duplicate target medication from a patient medication list. Hane, col. 14, ll. 19-52.
Agresta teaches the missing patient-medication-list reconciliation context because its mHealth application can retrieve medications from multiple electronic health records, personal health records, and other health information technology systems and combine and reconcile medication into a medication list that identifies potential conflicts between the same and/or different medications. Agresta further identifies the problem as medication lists containing duplicate, missing, or inaccurate information and teaches evaluating de-duplication, semantic matching, and selection of the best medication list. Agresta, Abstract; p. 1-3, Introduction and Section II.A.
A POSITA would have combined Hane with Agresta by applying Hane’s computer-performed duplicate prescription/drug-code filtering to Agresta’s medication-reconciliation workflow for de-duplicating medication information from multiple HIT sources. Hane supplies the automated duplicate-filtering technique for prescription/drug-code data, while Agresta supplies the patient medication-list problem and de-duplication objective. The modification would have predictably generated a deduplicated patient medication list by automatically removing the second medication entry after determining that it duplicates the first medication entry. Hane, col. 14, ll. 19-52, and Agresta, Abstract; p. 1-3, Introduction and Section II.A.
Claim 29.
Hane in combination with Agresta teaches, The one or more non-transitory computer-readable media of claim 1, wherein removing the second target medication associated with the second target unmapped medication code from the listing of medications for the patient as duplicative of the first target medication associated with the first target unmapped medication code to generate a deduplicated listing of medications for the patient is performed without human intervention.
Hane teaches machine-executed duplicate filtering of patient-associated prescription/drug-code data, as shown by the set of medical sentences corresponding to a patient identifier may be filtered to remove repeated/duplicate prescription/drug codes and only the first instance of prescription/drug code corresponding to the statin may be included in the set of medical sentences corresponding to the patient/patient identifier. This supports duplicate filtering without human intervention, but only for preparing medical sentences used in embedding training, not for removing a duplicate medication from a patient medication list. Hane, col. 14, ll. 28-52.
Agresta teaches the missing patient-medication-list reconciliation context because its mHealth application can retrieve medications from multiple electronic health records, personal health records, and other HIT systems and combine and reconcile medication into a medication list that identifies potential conflicts between the same and/or different medications. Agresta also identifies the problem as medication lists from disparate sources containing duplicate, missing, or inaccurate information and teaches evaluating de-duplication, semantic matching, and selection of the best medication list. Agresta, Abstract; p. 1-3, Introduction and Section II.A.
A POSITA would have combined Hane with Agresta by using Hane’s machine-executed duplicate prescription/drug-code filtering in Agresta’s medication-reconciliation workflow to de-duplicate medication information from multiple HIT sources. Hane supplies the machine-executed duplicate-filtering technique, while Agresta supplies the patient medication-list problem and de-duplication objective. The modification would predictably remove the second medication entry without human intervention after the system determines that the second medication duplicates the first medication. This rationale is supported by Hane, col. 14, ll. 28-52, and Agresta, Abstract; p. 1-3, Introduction and Section II.A.
Note: Claims 10 and 16-19 are rejected with the above analysis for being very similar to claims 1, 7-9.
Claim(s) 5 and 14 is/are rejected under 35 U.S.C. 103 as being unpatentable over Hane - US10891352 and further in view of Agresta – PTO-892-U and further view of US11556579- Bhatia
Hane in combination with Agresta teaches, Claim 5.
The one or more non-transitory computer-readable media of Claim 1, wherein the first medication comprises a
Hane teaches first and second medications of the Claim 5, "wherein the first medication comprises a generic medication and the second medication comprise a name brand medication," that required a digital storage medium to process distinct medication identifiers within a multi-dimensional space to identify similar clinical entities. Hane provides the functional infrastructure for this limitation by processing "drug codes" as key attributes in generating vector embeddings for medical events (Hane, See at least, Col. 4, ll. 27 -41“Medical information is often encoded using medical codes such as... prescription or drug codes.”). However, Hane does not describe the first medication comprises a generic medication and the second medication comprise a name brand medication. Hane processes medical codes as generic alphanumeric data attributes without expressly categorizing the drug hierarchy into branded vs. generic types.
Bhatia teaches a generic medication and a name brand medication of the Claim 5 that required a specialized classification system to link unstructured drug text to standardized medical categories. Bhatia provides the specific "species" classification missing from Hane’s "genus" of drug codes by using an ontology that expressly distinguishes between branded and generic drug identities (Bhatia, See at least, “the standardized ontology is a standardized medical ontology for generic and branded medication names.” [Claim 18], Col. 3, ll. 21-40).
The combination of Hane + Bhatia applications make obvious the full limitation "wherein the first medication comprises a generic medication and the second medication comprise a name brand medication" because a POSITA would combine Hane’s vector-based similarity analysis with Bhatia’s ontology-based drug labeling using KSR Rationale 1 (Combining prior art elements according to known methods to yield predictable results). If Hane seeks to “identify providers that have similar practices, Col. 4, ll. 20-30” and Bhatia teaches a “standardized medical ontology, Col. 3, ll. 21-40” for drug types, a POSITA would implement Bhatia's categorization as a routine data-labeling configuration to make Hane’s similarity engine operate with higher clinical specificity. Specifically, tagging medications as generic or brand-name before vectorization predictably ensures that the model recognizes different identifiers for the same active ingredient (Hane, See at least, “identify two or more aggregate vectors that are similar, abstract”; Bhatia, See at least, “standardized medication name for medication text, claim 17”).
Note: Claim 14 is rejected with the same analysis above for being very similar to claim 5.
Claim(s) 6 and 15 are rejected under 35 U.S.C. 103 as being unpatentable over Hane - US10891352 and further in view of Agresta – PTO-892-U and further view of US20140310013A1- Ram.
Hane in combination with Agresta teaches Claim 6.
The one or more non-transitory computer-readable media of Claim 1, wherein the first similarity measure comprises a weighted cosine similarity measure for the first target vector embedding and the first vector embedding. Hane, See at least, analyze at least a portion of the plurality of aggregate vectors to identify two or more aggregate vectors, claim 12… that are similar or different based on a distance or angle between the two or more aggregate vectors in the multi-dimensional space, Col. 1-15. the distance within the multi-dimensional space may be Euclidean distance, cosine distance, and/or other distance measure. In an example embodiment the distance between the multi-dimensional space is an angle or value indicative of an angle (e.g., cosine distance), Col.16, ll. 41-60.)
and wherein the second similarity measure comprises . (Hane, Col.4, ll.1-26, Col. 17, ll. 45-67, col. 18, ll. 12-18)Hane uses cosine distance or angle between vector pairs as a similarity metric. However, Hane does not expressly teach first weighted cosine similarity measure and second weighted cosine similarity measure.
Ram teaches the missing weighted cosine similarity measure, as shown by the system measures a similarity between feature vectors of each challenge and individual users based on a number of criteria, including but not limited to: cosine measure, weighted cosine measure, and Pearson correlation. Ram, para. 0084. Ram further grounds the weighted vector context by teaching that the user feature vector may include text information comprising a number of groups of weighted words, and that the challenge feature vector includes text information before similarity is measured. Ram, paras. 0082-0084. Functionally, Ram’s weighted cosine measure is the same kind of operation required by Claim 6: a similarity measure computed between two feature/vector representations, with weighting included in the cosine-based similarity criterion.
A person of ordinary skill in the art would have combined Hane with Ram before the effective filing date by using Ram’s weighted cosine measure as Hane’s cosine-based vector-similarity metric. The combination would have been reasonable because Hane already compares medical-code vectors by distance or angle, including cosine distance, to determine similarity, and Ram expressly identifies weighted cosine measure as a similarity criterion for comparing feature vectors. The modification would have predictably produced the claimed first and second weighted cosine similarity measures by applying the same weighted cosine comparison to each recited vector pair.
Note: Claim 15 is rejected with the same analysis above for being very similar to claim 6.
Conclusion
Any inquiry concerning this communication or earlier communications from the examiner should be directed to JOSHUA DAMIAN RUIZ whose telephone number is (571)272-0409. The examiner can normally be reached 0800-1800.
Examiner interviews are available via telephone, in-person, and video conferencing using a USPTO supplied web-based collaboration tool. To schedule an interview, applicant is encouraged to use the USPTO Automated Interview Request (AIR) at http://www.uspto.gov/interviewpractice.
If attempts to reach the examiner by telephone are unsuccessful, the examiner’s supervisor, Shahid Merchant can be reached at (571) 270-1360. The fax phone number for the organization where this application or proceeding is assigned is 571-273-8300.
Information regarding the status of published or unpublished applications may be obtained from Patent Center. Unpublished application information in Patent Center is available to registered users. To file and manage patent submissions in Patent Center, visit: https://patentcenter.uspto.gov. Visit https://www.uspto.gov/patents/apply/patent-center for more information about Patent Center and https://www.uspto.gov/patents/docx for information about filing in DOCX format. For additional questions, contact the Electronic Business Center (EBC) at 866-217-9197 (toll-free). If you would like assistance from a USPTO Customer Service Representative, call 800-786-9199 (IN USA OR CANADA) or 571-272-1000.
/JOSHUA DAMIAN RUIZ/Examiner, Art Unit 3684
/Shahid Merchant/Supervisory Patent Examiner, Art Unit 3684
Prosecution Insights