Smart EHR MCP Server

## TypeScript Schema for Policy Index JSON ```typescript /** * Defines the overall structure for the JSON output representing a single medical policy document. * The goal is to index treatments, indications, and their coverage status for easy searching, * without embedding the full complexity of the policy logic. */ interface PolicyIndex { /** * Contains metadata about the specific policy document being indexed. * Source: Typically found in the header or footer of the policy document. */ policyMetadata: PolicyMetadata; /** * An array defining all distinct treatments, procedures, or technologies * discussed in the policy. Each treatment should have a unique ID within this document. * Source: Identify all specific medical services, devices, or techniques mentioned. * Look in the title, introduction, policy criteria, investigational sections, coding section, and regulatory status. */ treatments: TreatmentDefinition[]; /** * An array defining all distinct indications (reasons for treatment), contraindications * (reasons *not* to treat), or specific policy rule contexts (like adjunctive use) * discussed in the policy. Each should have a unique ID within this document. * Source: Identify all diseases, conditions, symptoms, patient characteristics, or specific usage scenarios * mentioned in relation to the treatments. Look in the introduction, policy criteria sections (medical necessity, * investigational, contraindications), and potentially the evidence review for context. */ indications: IndicationDefinition[]; /** * An array summarizing the coverage determinations for specific combinations * of treatments and indications/contraindications/rules defined above. * This is the core linkage showing what the policy concludes for each pair. * Source: Synthesize information from the "Policy Coverage Criteria", "Indications Considered Medical Necessity", * "Indications Considered Investigational", and "Contraindications" sections. */ coverageRules: CoverageRule[]; } /** * Metadata about the source policy document. */ interface PolicyMetadata { /** * The official title of the medical policy. * Source: Document header. * Example: "Transcranial Magnetic Stimulation as a Treatment of Depression and Other Psychiatric/Neurologic Disorders" */ policyTitle: string; /** * The unique identifier or number assigned to this policy by the issuer. * Source: Document header. * Example: "2.01.526" */ policyNumber: string; /** * The date when this version of the policy becomes effective. * Format: YYYY-MM-DD. * Source: Document header or effective date section. * Example: "2024-11-01" */ effectiveDate: string; /** * The date when this policy was last revised or updated. * Format: YYYY-MM-DD. * Source: Document header or revision history. * Example: "2024-10-21" */ lastRevisedDate: string; /** * (Optional) The policy number this policy replaces, if specified. * Source: Document header or revision history. * Example: "2.01.50" */ replacesPolicy?: string; /** * (Optional) Reference to a related BCBSA (Blue Cross Blue Shield Association) policy, if specified. * Source: Document header. * Example: "2.01.50" */ bcbsaRefPolicy?: string; /** * The original filename of the source document (e.g., PDF). Helps link back to the source. * Example: "premera_bc_2_01_526_tms.pdf" */ sourceFilename: string; /** * Version number for this JSON schema structure itself, in case the schema evolves. * Example: "2.0" */ schemaVersion: string; } /** * Defines a specific treatment, procedure, or technology mentioned in the policy. */ interface TreatmentDefinition { /** * A unique, concise identifier for this treatment *within this policy file*. * Convention: Use a prefix like "tms_" followed by a descriptive term (e.g., "tms_standard", "tms_deep", "tms_any"). Must be unique within the 'treatments' array. */ id: string; /** * The primary, standardized name for the treatment as used in the policy. * Example: "Standard/Conventional Repetitive TMS", "Deep TMS", "Any TMS Type" */ name: string; /** * A comprehensive list of keywords, synonyms, acronyms, brand names, codes, * and related terms associated with this treatment. Crucial for searchability. * Be thorough: include terms from the policy text, common clinical usage, and the coding section. * Source: Policy text (title, intro, criteria, investigational list, regulatory status, coding), common knowledge. * Example: ["TMS", "rTMS", "Transcranial Magnetic Stimulation", "Standard TMS", "SNOMED:123", "NeuroStar"] */ keywords: string[]; /** * An array of CPT, HCPCS, or other specified codes explicitly associated with this *specific* treatment * in the policy's coding section or text. Only include if directly linked. * Source: Coding section of the policy. * Example: ["CPT:90867", "CPT:90868", "CPT:90869"] */ codes?: string[]; } /** * Defines the type of an indication entry. Helps differentiate between clinical conditions, * reasons *not* to perform a procedure, and specific usage rules. * Use only these exact values. */ type IndicationType = /** The indication represents a clinical disease, disorder, symptom, or patient group. */ | 'Condition' /** The indication represents a contraindication – a reason the treatment should NOT be performed. */ | 'Contraindication' /** The indication represents a specific policy rule or context (e.g., adjunctive use, use with other tech). */ | 'PolicyRule'; /** * Defines a specific indication (clinical condition, contraindication, or policy rule context). */ interface IndicationDefinition { /** * A unique, concise identifier for this indication *within this policy file*. * Convention: Use a prefix like "ind_", "contra_", or "rule_" followed by a descriptive term (e.g., "ind_mdd", "contra_implant", "rule_adjunctive"). Must be unique within the 'indications' array. */ id: string; /** * Categorizes the indication using the predefined IndicationType values. * This helps interpret the coverage rules correctly (e.g., a 'Contraindication' rule means NMN). */ type: IndicationType; /** * The primary, standardized name for the indication, contraindication, or rule. * Example: "Major Depressive Disorder (Unipolar)", "Contraindication: Metallic Implant", "Adjunctive Use (Boosting other treatments)" */ name: string; /** * A comprehensive list of keywords, synonyms, acronyms, related terms, or specific examples * associated with this indication/contraindication/rule. Crucial for searchability. * Be thorough: include terms from policy text, common clinical usage, related conditions/situations. * Source: Policy text (criteria sections for necessity, investigational, contraindications), common knowledge. * Example: ["MDD", "Major Depressive Disorder", "Unipolar Depression", "TRD"] * For Contraindications: ["Contraindication", "Not Medically Necessary", "Metal", "Implant", "Ferromagnetic"] */ keywords: string[]; } /** * Defines the coverage status determination for a specific treatment/indication pair. * Use only these exact values. */ type CoverageStatus = /** The treatment is considered medically necessary for the indication, provided specific criteria (detailed in the full policy) are met. */ | 'Medically Necessary' /** The treatment is considered NOT medically necessary for the indication. This often applies to contraindications or explicitly excluded scenarios. */ | 'Not Medically Necessary' /** The treatment is considered investigational (experimental, unproven) for the indication. */ | 'Investigational'; /** * Represents a specific coverage rule linking one treatment to one indication/contraindication/rule, * summarizing the policy's determination for that combination. */ interface CoverageRule { /** * The unique ID of the treatment this rule applies to. Must match an 'id' from the 'treatments' array. * Use generic IDs like "tms_any" when a rule applies broadly to multiple or all defined treatments. */ treatmentId: string; /** * The unique ID of the indication, contraindication, or rule this rule applies to. Must match an 'id' from the 'indications' array. * Use generic IDs like "ind_any" when a rule applies broadly to multiple or all defined indications. */ indicationId: string; /** * The summary coverage determination for this specific treatment-indication combination, * using one of the predefined CoverageStatus values. * Source: Synthesize from the main policy statements (Medically Necessary, Investigational, Not Medically Necessary/Contraindicated). */ coverageStatus: CoverageStatus; /** * (Optional) A brief, concise note providing essential context or clarification for this specific rule, * *without* detailing the full criteria. Focus on key exceptions, scope limitations, or important nuances. * Example: "Criteria apply: Age >= 18, moderate/severe, specific medication failures required.", "Excludes accelerated/SNT/SAINT protocols.", "Unless specific clearance documented." */ notes?: string; } ``` --- ## LLM Prompting Strategy and Guide for Data Abstraction **Core Prompt Structure:** ```text You are an expert medical policy analyst tasked with extracting structured information from medical policy documents into a specific JSON format. Your goal is to create an index summarizing coverage decisions for treatments and indications, focusing on findability via keywords, without embedding complex clinical criteria. **Input:** The full text content of a single medical policy document will be provided. **Output:** Generate a single JSON object that strictly conforms to the following TypeScript schema: <PASTE_TYPESCRIPT_SCHEMA_HERE> **Instructions:** 1. **Parse Metadata:** Extract the information required for the `policyMetadata` object from the document's header, footer, or revision history sections. Ensure dates are in YYYY-MM-DD format. Include the original filename and the specified schema version ("2.0"). 2. **Identify and Define Treatments:** * Carefully read the policy (title, introduction, criteria, investigational list, coding, regulatory status) to identify *all distinct* treatments, procedures, devices, or technologies mentioned. * For each distinct treatment, create an object in the `treatments` array. * Assign a unique `id` using the convention `tms_[descriptive_term]` (e.g., `tms_standard`, `tms_deep`, `tms_saint`). Use `tms_any` for rules applying broadly. * Assign the most appropriate standardized `name`. * Populate the `keywords` array *comprehensively*. Include: * The primary name and acronyms (TMS, rTMS, dTMS, TBS). * Full names (Transcranial Magnetic Stimulation). * Synonyms or variations mentioned (Standard, Conventional, Repetitive). * Specific parameters if they define a type (Accelerated, >= 3/day, Neuronavigation, H-Coil). * Brand names mentioned (NeuroStar, Brainsway, Cerena). * Relevant CPT/HCPCS/other codes mentioned anywhere in the policy (even if also listed in `codes`). * Related concepts (Image-guided, qEEG guided). * If specific CPT/HCPCS/other codes are explicitly linked *only* to this treatment type in the coding section, add them to the optional `codes` array. 3. **Identify and Define Indications/Contraindications/Rules:** * Read the policy (introduction, medical necessity criteria, investigational criteria, contraindications, other rules sections) to identify *all distinct* clinical conditions, reasons *not* to treat (contraindications), or specific policy contexts/rules (e.g., adjunctive use, use with other therapies). * For each distinct item, create an object in the `indications` array. * Assign a unique `id` using conventions: `ind_[condition_term]` (e.g., `ind_mdd`), `contra_[reason_term]` (e.g., `contra_implant`), `rule_[context_term]` (e.g., `rule_adjunctive`). Use `ind_any` for rules applying broadly to indications. * Assign the appropriate `type`: 'Condition', 'Contraindication', or 'PolicyRule'. * Assign the most appropriate standardized `name`. * Populate the `keywords` array *comprehensively*. Include: * The primary name and acronyms (MDD, OCD, PTSD, TBI, ICP). * Synonyms (Unipolar Depression, Manic-Depressive Illness). * Related concepts or severity levels mentioned (Treatment Resistant, TRD, Moderate, Severe). * Specific examples listed under broader categories (e.g., list specific metallic implants under the 'Metallic Implant' contraindication keywords). * For Contraindications, *always* include "Contraindication" and "Not Medically Necessary" as keywords. * For Policy Rules, include terms describing the context (Adjunctive, Combination, Concurrent Use, Bridge Therapy). 4. **Determine and Record Coverage Rules:** * For *each relevant combination* of a Treatment and an Indication/Contraindication/Rule explicitly or implicitly addressed in the policy: * Create an object in the `coverageRules` array. * Set `treatmentId` to the ID of the relevant treatment (use `tms_any` if the rule applies to all/multiple TMS types discussed). * Set `indicationId` to the ID of the relevant indication/contraindication/rule (use `ind_any` if the rule applies regardless of specific indication). * Determine the overall `coverageStatus` based *only* on the policy's explicit statements (Medically Necessary, Investigational, Not Medically Necessary). * "Medically Necessary": Found in sections defining coverage criteria. * "Investigational": Found in sections listing investigational uses. * "Not Medically Necessary": Found in contraindication sections or explicit statements of non-coverage (e.g., use with ECT). * If necessary, add a *brief* `notes` field to capture critical context mentioned alongside the determination (e.g., age limits, specific exclusions like "Excludes psychotic features", prerequisite conditions like "Unless specific clearance documented"). **Do NOT copy detailed clinical criteria here.** Keep notes concise (max 1-2 short sentences). 5. **Consistency and Completeness:** * Ensure all `treatmentId` and `indicationId` values used in `coverageRules` correspond exactly to IDs defined in the `treatments` and `indications` arrays. * Be thorough in identifying all distinct treatments and indications/rules mentioned. * Prioritize comprehensive keywords for robust searching. * Strictly adhere to the allowed values for `coverageStatus` and `indicationType`. * If the policy mentions a treatment or indication but provides no clear coverage determination for a specific combination, do *not* create a rule for it unless there's a default rule (like "all other indications are investigational"). **Example Snippet Translation:** * **Policy Text:** "Theta burst stimulation is considered investigational for the treatment of Obsessive Compulsive Disorder." * **Should Result In (assuming IDs `tms_tbs` and `ind_ocd` exist):** ```json { "treatmentId": "tms_tbs", "indicationId": "ind_ocd", "coverageStatus": "Investigational" } ``` * **Policy Text:** "Contraindications: ... Seizure disorder or a history of a seizure disorder, unless stable and well-controlled on medication..." * **Should Result In (assuming IDs `tms_any` and `contra_seizure` exist):** ```json { "treatmentId": "tms_any", // Applies to all TMS types "indicationId": "contra_seizure", "coverageStatus": "Not Medically Necessary", "notes": "Unless stable/well-controlled or specific exceptions apply." } ``` Generate the JSON output based on these instructions and the provided policy text.