DETAILED ACTION
The present application, filed on or after March 16, 2013, is being examined under the first inventor to file provisions of the AIA . This action is responsive to pending claims 1-20 filed 8/17/2026.
Claim Rejections - 35 USC § 101
The 35 U.S.C. 101 are withdrawn in view of Applicant’s response. Vectorizing feature vectors via spatial locations on document components is not a process practically performed in the mind, nor is feature selection for a potential mental model training based on vectorized document components. Nor can these steps be said to constitute a mathematical process. Further, the claims are integrated into the practical application of application of document feature extraction in the context of claims processing via machine learning.
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.
Claim(s) 1, 4-5, 14, 17-18 are rejected under 35 U.S.C. 103 as being unpatentable over Anand ("Automating prior authorization decisions using machine learning and health claim data", published 2022) in view of Li (US 20230004604 A1).
For claim 1, Anand discloses: a method of machine learning model training for automated entity update processing (§5.1 ¶1-2 gives overview of a machine learning technique for processing prior authorization (PA) requests, such as to automate approval decisions, see §5.3, hence, automating updated insurance PA processing), the method comprising:
obtaining a training set of multiple historical claims documents, each of the multiple historical claims documents including an associated entity value (§5.2 “Preprocessing and Data Acquisition Staging” discloses training data including historical claim documents associated with accepted and denied decisions);
extracting text portions from the training set of multiple historical claims documents, the extracted text portions including one or more identified text features (§5.1.1: text portions are extracted via NLP including extracting relevant information including medical justifications for surgery);
generating training feature vectors based on the extracted text portions and the one or more identified text features (§5.2 “Training Approach and Model Selection” discloses training with such data, such as via decision trees, random forests, gradient boosting (XGBoost) techniques, hence, training data of §5.2 “Preprocessing” is generated to train machine learning models, hence, structured data set vectors of features are input into these algorithms for training);
supplying the training feature vectors to a treatment identification machine learning model to train the treatment identification machine learning model to generate a prediction output (§5.2.1-5.2.2: supervised machine learning models are supplied the training feature vectors generated above, these ML models identifying treatment in order to generate a predicted outcome, hence, treatment identification ML models), the prediction output indicative of a correct entity value for a claim document (§5.3 disclose accuracy results);
obtaining a target claim document (§5.3 discloses testing on additional claim documents);
extracting text portions and identified text features from the target claim document (§5.1.1: extraction via NLP);
supplying the extracted text portions and the identified text features from the target claim document to the machine learning model to generate a prediction output of the correct entity value for the target claim document (§5.3: generating prediction results); and
automatically processing an entity update associated with the target claim document based on the prediction output of the correct entity value for the target claim document (§5.4: generating automatic PA approvals for high confidence projections).
Anand does not disclose: wherein extracting the text portions includes, for each text portion, assigning a text identifier to the text portion, determining a relative position of the text portion within boundaries of a corresponding one of the multiple historical claims documents, and identifying one or more other text portions located above, below, to a left of, and to a right of the relative position of the text portion;
wherein generating the training feature vectors includes vectorizing, word by word, the identified one or more other text portions located above, below, to the left of, and to the right of the relative position of each text portion, including the determined relative position of each text portion as a feature of the training feature vectors, and applying a feature selection model to iteratively select a specified number of features from the vectorized text portions, the specified number of features being selected from vectorized text portions having at least a minimum number of examples in the training set.
Li discloses: wherein extracting the text portions includes (fig.8, 0165 gives overview of the process for training a knowledge model, including the various data preparation steps in the subsequent figures figs.9-19), for each text portion, assigning a text identifier to the text portion (fig.19, 0188-197 shows various text portions including numeric identifier (0, 1, 2, etc.)), determining a relative position of the text portion within boundaries of a corresponding one of the multiple historical claims documents (fig.19, 0194: relative coordinates), and identifying one or more other text portions located above, below, to a left of, and to a right of the relative position of the text portion (figs.17-19 additional spatial elements are defined, including, for particular centrally located elements, elements in claimed directions);
wherein generating the training feature vectors includes vectorizing (fig.8 gives overview of a process for generating input to knowledge model, e.g., a neural net (see fig. 2, fig.3:338), hence, a vectorization process including generation of row vectors of fig.19), word by word, the identified one or more other text portions (figs.14, 0183 shows OCR-based word-by-word generation of features of text portions, eventually aggregated into the features of figs. 17-19 ) located above, below, to the left of, and to the right of the relative position of each text portion (figs.17-19: spatial elements located to the various sides of centrally located portions are processed), including the determined relative position of each text portion as a feature of the training feature vectors (fig.19:194 shows incorporation of various relative positions incorporate as features), and applying a feature selection model to iteratively select a specified number of features from the vectorized text portions (fig.8: feature engineering, 0201, fig.3:334-336, 0142-143: iteratively applying one or more feature engineering processes to csv vectors to select features), the specified number of features being selected from vectorized text portions having at least a minimum number of examples in the training set (0142-143: selected features having at least one example as they are drawn from the generated csv data vector, e.g., previously identified features).
It would have been obvious before the effective filing date to one of ordinary skill in the art to modify the method of Anand by incorporating the feature vector generation of Li. Both concern the art of information extraction, and the incorporation would have, according to Li, improve document extraction by better leveraging context data to identify documents (0004-5).
For claim 4, Anand modified by Li disclose the method of claim 1, as described above. Anand further discloses: wherein the machine learning model includes a random forest tree-based model (§5.2.1).
For claim 5, Anand modified by Li disclose the method of claim 1, as described above. Anand modified by Li further discloses: wherein extracting the text portions includes, for each text portion:
determining a position of the text portion (Li fig.19, 0189-197: various positions are determined);
identifying one or more other text portions adjacent to the position of the text portion (figs.17-19: additional boxes are determined, including various adjacent boxes); and
determining at least one qualitative indicator associated with each of the identified one or more other text portions (fig.19: group labels, fig.20-21: various features and attributes).
Claim(s) 14, 17-18 recite systems analogous to the above method(s) 1, 4-5 and are hence rejected for the same reasons.
Claim(s) 2, 6, 8-9, 12-13, 15, 19 are rejected under 35 U.S.C. 103 as being unpatentable over Anand ("Automating prior authorization decisions using machine learning and health claim data", published 2022) in view of Li (US 20230004604 A1) in view of Wang (US 20220301072 A1).
For claim 2, Anand modified by Li disclose the method of claim 1, as described above. Anand modified by Li does not disclose: wherein automatically processing the entity update includes automatically storing the prediction output of the correct entity value in a database as an entity field update associated with the target claim document.
Wang discloses: wherein automatically processing the entity update includes automatically storing the prediction output of the correct entity value in a database as an entity field update associated with the target claim document (fig.14:1424, 0186: storing of extracted information prediction output in a database, as an information update to associated claim documents, such as processed via workflow of fig.6, se 0082 to determine claim document policy application).
It would have been obvious before the effective filing date to a person of ordinary skill in the art to modify the method of Anand by incorporating the database technique of Wang. Both concern the art of automatic insurance claim processing, and the incorporation would have, according to Wang, allow implementation of systems and workflows that would more accurate and less expensive than current procedures (0003).
For claim 6, Anand modified by Li disclose the method of claim 1, as described above. Anand does not disclose the limitations of claim 6.
Wang discloses: classifying a page type of the target claim document via a page classification model (fig.9, 0103 gives overview of classifying page images for provision to the information extraction engine 230 (0112), the information extraction engine operation to determine page type and feature, see fig.10:1004 for extracting policy rules, hence, classifying as policy page type, see 0163, medical documents type, see fig.18 0149-150, hence, classifying pages for further processing after image conversion); and
supplying the classified page type to the machine learning model with the extracted text portions and the identified text features from the target claim document (fig.10, fig.18, etc. as above: page type and relevant page text portions and text features (policy, injury, etc.) is obtained for further processing), to generate the prediction output of the correct entity value for the target claim document (fig.6 determines using said data for claim processing, whether conditions are met, see 0089).
It would have been obvious before the effective filing date to a person of ordinary skill in the art to modify the method of Anand by incorporating the database technique of Wang. Both concern the art of automatic insurance claim processing, and the incorporation would have, according to Wang, allow implementation of systems and workflows that would more accurate and less expensive than current procedures (0003).
For claim 8, Anand modified by Li modified by Wang disclose the method of claim 6, as described above. Anand modified by Wang further discloses: wherein the page type includes at least one of a claim form, a pre-authorization document, a claim information document (Wang 0077-84: doctor’s opinions, treatment reports, clinical notes), a claim list, a correspondence document (ibid: doctors opinions and notes constitute correspondence documents), an envoy page, an ID card, an invoice, a payment details document, a medical information document (treatment reports, clinical notes, hospital record, etc.), a shipping label, or a system generated email.
For claim 9, Anand modified by Li disclose the method of claim 1, as described above. Anand modified by Li does not disclose the limitations of claim 9.
Wang discloses supplying the prediction output of the machine learning model to a standardization model to conform a treatment type in the prediction output to a standard treatment type format (Wang fig.18, 0149-154 discloses information extraction of injury types, fig.8, 0098-102 disclosing information extraction of documents, such as treatment reports (0078, 0081), with fig.6 disclosing application of extracted information to policy to generate policy inferences , e.g., whether conditions are met, see 0089, hence, prediction output such as of treatment type prediction, injury, is supplied to a standardization model to determine treatment conformance to policy).
wherein the standard treatment type format includes at least one of a specified treatment claim type (fig.6, 0089: treatment claim type condition met or not met), a specified treatment category, a specified treatment sub-category, and a specified treatment detailed category.
It would have been obvious before the effective filing date to a person of ordinary skill in the art to modify the method of Anand by incorporating the information extraction technique of Wang. Both concern the art of automatic insurance claim processing, and the incorporation would have, according to Wang, allow implementation of systems and workflows that would more accurate and less expensive than current procedures (0003).
For claim 12, Anand modified by Li disclose the method of claim 1, as described above. Anand further discloses: identifying, via a provider information model, at least one extracted text portion (§5.1.1: identifying personally identifiable data (PII) for deletion); and
removing the at least one extracted text portion from the extracted text portions supplied to the machine learning model (ibid).
Anand modified by Li does not disclose: wherein the extracted text portions including a healthcare provider name; wherein the extracted text portion includes the healthcare provider name.
Wang discloses: wherein the extracted text portions including a healthcare provider name; wherein the extracted text portion includes the healthcare provider name (fig.18, 0149-154 discloses entity information extraction that removes various named entities not related to the sickness and treatment information (0149) including proper names and places, hence, removal of names such as healthcare provider such as found in claims data of Anand).
It would have been obvious before the effective filing date to a person of ordinary skill in the art to modify the method of Anand by incorporating the information extraction technique of Wang. Both concern the art of insurance claim processing, and the incorporation would have, according to Wang, allow retrieval or extraction of medical information specific to particular illnesses or injury (0149), such as for further automated claims processing (e.g., fig.3, 0003, fig.6).
For claim 13, Anand modified by Li disclose the method of claim 1, as described above. Anand further discloses: wherein: the target claim document includes a health insurance claim document or a prescription drug coverage claim document for a patient (§5.2 “Preprocessing” contemplates operating on prescription drug treatment data, insurance claims).
Anand modified by Li does not disclose: the prediction output of the machine learning model is a text portion of the health insurance claim document or a prescription drug coverage claim document including a treatment type for the patient.
Wang discloses: the prediction output of the machine learning model is a text portion of the health insurance claim document or a prescription drug coverage claim document including a treatment type for the patient (fig.18, 0149-154 discloses information extraction prediction for medial related terms, the output being an extracted text portion of a insurance claims document related to the injury or condition, hence, combination with Anand’s disclosure of treatment plans yielding application to treatment types based on illness, see Anand §4.2: Data Sources disclosing HER including treatment plans).
It would have been obvious before the effective filing date to a person of ordinary skill in the art to modify the method of Anand by incorporating the information extraction technique of Wang. Both concern the art of insurance claim processing, and the incorporation would have, according to Wang, allow retrieval or extraction of medical information specific to particular illnesses or injury (0149), such as for further automated claims processing (e.g., fig.3, 0003).
Claim 14 recites a system analogous to claim 1 above and is hence rejected for the same reasons. Li further discloses: the system comprising:
memory hardware configured to store computer-executable instructions (fig.26, 2640-2650, 0238-241), and a training set of multiple historical claims documents (Anand §5.2 Preprocessing discloses acquiring of training data, hence, combination with Wang yielding storage of data for training); and
processor hardware configured to execute the computer-executable instructions to perform a process (fig.26:2610, 0236).
Claim(s) 15, 19 recite systems analogous to the above method(s) 2, 6 and are hence rejected for the same reasons.
Claim(s) 3, 16 are rejected under 35 U.S.C. 103 as being unpatentable over Anand ("Automating prior authorization decisions using machine learning and health claim data", published 2022) in view of Li (US 20230004604 A1) in view of Wang (US 20220301072 A1) in view of Blohm (US 20240257260 A1).
For claim 3, Anand disclose the method of claim 1, as described above. Anand does not disclose the method of claim 3.
Wang discloses: wherein automatically processing the entity update includes:
displaying the prediction output of the correct entity value on a user interface (figs.11-12, 0178-182 show GUI for displaying and drilling down on the automatically extracted information and claim processing);
receiving an input from a user in response to displaying the prediction output of the correct entity value on a user interface (ibid: user may provide input based on displayed predicted entity value outputs, including drilling down, action button, including tabs for verifying the inferred results, benefits, etc.).
It would have been obvious before the effective filing date to a person of ordinary skill in the art to modify the method of Anand by incorporating the database technique of Wang. Both concern the art of automatic insurance claim processing, and the incorporation would have, according to Wang, allow implementation of systems and workflows that would more accurate and less expensive than current procedures (0003).
Anand modified by Wang does not disclose: storing the input received from the user in a database as an entity field update associated with the target claim document.
Blohm discloses: storing the input received from the user (0049 discloses manual viewing and adjustment of automatic suggestions, hence, combination with Anand modified by Wang yielding the storing of the user input adjudications in the database as a value associated with the claim documents).
It would have been obvious before the effective filing date to a person of ordinary skill in the art to modify the method of Anand modified by Wang by incorporating adjudication technique of Blohm. Both concern the art of automatic insurance claim processing, and the incorporation would have, according to Blohm, allow an expert user to adjust claim values based (0049).
Claim(s) 16 recite systems analogous to the above method(s) 2, 6 and are hence rejected for the same reasons.
Claim(s) 7, 10-11, 20 are rejected under 35 U.S.C. 103 as being unpatentable over Anand ("Automating prior authorization decisions using machine learning and health claim data", published 2022) in view of Li (US 20230004604 A1) in view of Wang (US 20220301072 A1) in view of Burns ("Classification of current procedural terminology codes from electronic health record data using machine learning", published 2020).
For claim 7, Anand modified by Li modified by Wang disclose the method of claim 6, as described above. Wang further discloses: wherein the page classification model is a machine learning classification model (0043-44, 0047).
Anand modifieds by Wang does not disclose the remaining limitations.
Burns discloses: obtaining a page classification training set, the page classification training set including multiple historical sample documents each tagged with an assigned page classification (p.3: Case Selection: selecting cases with tagged CPT codes, for data preparation, see 4: Data Preparation and Natural Language Processing, to generate training data, hence, generating a supervised machine learning model (see p.4-5) to process treatment information via machine learning by assign unique CPT codes to the treatment type, hence, classifying treatment documentation pages based on type); and
supplying the page classification training set to the machine learning classification model to train the machine learning classification model to generate a classification output indicative of a predicted page type of a claims document (p.4-5: Supervised Training discloses training using the labeled training data via machine learning models to generate an output, such as for billing and adjustment purposes (p.2 Introduction ¶2)).
It would have been obvious before the effective filing date to a person of ordinary skill in the art to modify the method of Anand by incorporating the information extraction technique of Wang. Both concern the art of insurance claim processing, and the incorporation would have, according to Wang, allow retrieval or extraction of medical information specific to particular illnesses or injury (0149), such as for further automated claims processing (e.g., fig.3, 0003, fig.6).
For claim 10, Anand modified by Wang discloses the method of claim 9, as described above. Anand modified by Wang further discloses, wherein the standardization model is a machine learning standardization model (Wang 0102).
Anand modified by Wang does not disclose the remaining limitations of claim 10.
Burns discloses: obtaining a standard treatment type training set, the standard treatment type training set including multiple historical sample documents each tagged with at least one assigned treatment type category (p.3: Case Selection: selecting cases with tagged CPT codes, for data preparation, see 4: Data Preparation and Natural Language Processing, to generate training data); and
supplying the standard treatment type training set to the machine learning standardization model to train the machine learning standardization model to generate a standard format output indicative of a predicted standard treatment type of an identified treatment in a claims document (p.4-5: Supervised Machine Learning Methods: various machine learning models are trained with the training data, see also p.9: Holdout data set performance metrics for training results).
It would have been obvious before the effective filing date to a person of ordinary skill in the art to modify the method of Anand modified by Wang by incorporating CPT NLP technique of Burns. Both concern the art of medical documentation machine learning, and the incorporation would have, according to Burns, allow efficient generation of CPT, such as for billing practices (p.2 Introduction ¶1).
For claim 11, Anand disclose the method of claim 9, as described above. Anand modified by Wang does not disclose the limitations of claim 11.
Burns discloses: assigning a unique code according to the standard treatment type format, wherein the unique code includes at least one of an ICD-10 code, a CPT code, an HCPCS code or a CDT code (p.3: Case Selection: selecting cases with tagged CPT codes, for data preparation, see 4: Data Preparation and Natural Language Processing, to generate training data, hence, generating a supervised machine learning model (see p.4-5) to process treatment information via machine learning by assign unique CPT codes to the treatment type).
It would have been obvious before the effective filing date to a person of ordinary skill in the art to modify the method of Anand modified by Wang by incorporating CPT NLP technique of Burns. Both concern the art of medical documentation machine learning, and the incorporation would have, according to Burns, allow efficient generation of CPT, such as for billing practices (p.2 Introduction ¶1).
Claim(s) 20 recite systems analogous to the above method(s) 7 and are hence rejected for the same reasons.
Response to Arguments
Applicant’s arguments have been fully considered. In the remarks, Applicant argued:
1. The 101 rejections are overcome for the presented reasons.
Examiner agrees and the 101 rejections are withdrawn.
2. The cited art does not disclose the newly added limitations.
Applicant’s arguments have been fully considered but are moot in view of the newly cited art.
3. Anand does not disclose “prediction output indicative of a correct entity value for a claim document”, as Anand’s models train on structured health claim data and output an approval probability, not a correct entity value for a claim document.
Examiner respectfully disagrees. The predictive output indicates adherence to claim accuracy metrics, hence, adherence to correctness of an approval entity value. Further clarification is needed to reach Applicant’s intended scope, which concerns particular field values.
4. Examiner’s reasoning is inferential regarding generating training vectors, and hence is an obviousness rationale not a finding of inherent disclosure.
Examiner respectfully disagrees. The cited machine learning models are fundamentally numeric or mathematical in nature, and input to these models would inherently require transformation of training data to numeric arrays, i.e., feature vectors.
5. Wang does not disclose the newly added limitations.
Applicant’s arguments have been fully considered but are moot in view of the newly cited art.
6. For claim 5, 18, Wang does not disclose the limitations.
Applicant’s arguments have been fully considered but are moot in view of the newly cited art.
7. For claims 6, 19, Wang 103 does not disclose the cited limitations.
Examiner respectfully disagrees. Fig.9, 0103 is provided as a contextual overview of the classification process, with further elements 230, 0112, fig.10:1004, 0163, fig.18, 0149-150 cited as particular to the claim at stake.
8. For claims 7, 20: Burns’s selection of cases with tagged CPT codes is not “multiple historical sample documents each tagged with an assigned page classification”.
Examiner respectfully disagrees. The training data comprises a large set of health records tagged with CPT codes, hence, historical sample documents comprising various sections or pages, the documents tagged or associated with the cited classifications. Further clarification is needed to reach intended scope.
9. For claim 3, the cited art is not clear.
The citation heading for claim 3 has been clarified.
10. For claim 3, Bolom does not disclose the storing limitation.
Examiner respectfully disagrees. Bolom discloses manual feedback based on user, combination with the disclosure of Anand disclosing storing of data associated with the claim documents yielding the storing of said user feedback, as claimed.
11. The motivation to combine is generic boilerplate.
Examiner respectfully disagrees. A POSITA may be motivated to combine the same references for the same reasons.
Conclusion
The prior art made of record and not relied upon is considered pertinent to applicant's disclosure. Maldonado (US 20210287300 A1) discloses a unified GUI for claims processing.
Applicant's amendment necessitated the new ground(s) of rejection presented in this Office action. Accordingly, THIS ACTION IS MADE FINAL. See MPEP § 706.07(a). Applicant is reminded of the extension of time policy as set forth in 37 CFR 1.136(a).
A shortened statutory period for reply to this final action is set to expire THREE MONTHS from the mailing date of this action. In the event a first reply is filed within TWO MONTHS of the mailing date of this final action and the advisory action is not mailed until after the end of the THREE-MONTH shortened statutory period, then the shortened statutory period will expire on the date the advisory action is mailed, and any extension fee pursuant to 37 CFR 1.136(a) will be calculated from the mailing date of the advisory action. In no event, however, will the statutory period for reply expire later than SIX MONTHS from the date of this final action.
Any inquiry concerning this communication or earlier communications from the examiner should be directed to LIANG LI whose telephone number is (303)297-4263. The examiner can normally be reached Mon-Fri 9-12p, 3-11p MT (11-2p, 5-1a ET).
Examiner interviews are available via telephone, in-person, and video conferencing using a USPTO supplied web-based collaboration tool. To schedule an interview, applicant is encouraged to use the USPTO Automated Interview Request (AIR) at http://www.uspto.gov/interviewpractice. The examiner is available for interviews Mon-Fri 6-11a, 2-7p MT (8-1p, 4-9p ET).
If attempts to reach the examiner by telephone are unsuccessful, the examiner’s supervisor Jennifer Welch can be reached on (571)272-7212. The fax phone number for the organization where this application or proceeding is assigned is 571-273-8300.
Information regarding the status of an application may be obtained from Patent Center and the Private Patent Application Information Retrieval (PAIR) system. Status information for published applications may be obtained from Patent Center or Private PAIR. Status information for unpublished applications is available through Patent Center or Private PAIR to authorized users only. Should you have questions about access to Patent Center or the Private PAIR system, contact the Electronic Business Center (EBC) at 866-217-9197 (toll-free).
/LIANG LI/
Primary examiner AU 2143