OCR VA Claims File Pages: Searching a Veteran's C-File

To OCR a VA claims file, keep the copy VA sent you untouched, split a working copy into parts your tool accepts, and convert every part to searchable PDF. Then build an index that gives each document in the file a date, a type and a page range, so a search hit leads to a document you can name.

A veteran's C-file can span fifty years and several kinds of paper: induction physicals typed on a manual typewriter, carbon copies, handwritten sick call entries, and last year's laser-printed decision letter. Each converts differently, and the index is what turns the result from a searchable pile into a file.

Rating letters and the separation form are the two searches to run first. Look through the converted C-file for decisions that discuss a service-connected disability and for the document that states discharge status. Those pages are the core of a veteran benefits claim. Confirm diagnostic codes, effective dates, and the character of discharge on the image before you cite them.

What is a VA claims file and how does a representative get a copy?

A VA claims file, or C-file, is the record the Veterans Benefits Administration keeps of a veteran's benefit claims: applications, service records, medical evidence, examination reports, decisions and correspondence. A copy comes through a Privacy Act request, and accredited representatives may also be approved to read a client's records in VA's own systems.

The VA pages cited in this post use the term without defining it. VA's page on Privacy Act requests refers to "VBA Privacy Act records, such as compensation and pension records" and tells requesters that including "the individual's Social Security number or C-File number may aid in the identification of records." What VA gathers into the file is described in its duty-to-assist regulation. 38 CFR § 3.159(c)(2) lists federal records VA will seek, which "include but are not limited to military records, including service medical records; medical and other records from VA medical facilities; records from non-VA facilities providing examination or treatment at VA expense; and records from other Federal agencies, such as the Social Security Administration." Paragraph (c)(1) adds private providers, employers and state and local agencies. Each of those sources is a kind of page you will meet.

Requesting a copy

The right of access is in 38 CFR § 1.577(a): "any individual upon request may gain access to his or her record" and may "have a copy made of all or any portion thereof". Paragraph (f) of that section says that when VA benefit records "are requested by the individual to whom the record pertains, the duplication fee for one complete set of such records will be waived."

VA's privacy page gives three ways to make the request: complete VA Form 20-10206 and send it to the Centralized Support Division, submit the same form online, or send a signed written request. The page says the request "must be submitted by the individual" with a signed request and identifying information. How a representative's power of attorney fits with that wording is something to confirm with VA for your own accreditation and client.

Reading the file in VA's systems

Separately, 38 CFR § 1.600(b) says VA "will provide access to specific VA IT systems, the Veterans Benefit Management System (VBMS) and the Caseflow products Queue and eFolder Express", to approved attorneys, agents and service organization representatives, for claimants who appointed them, and on a read-only basis. The conditions run to several paragraphs and are worth reading in full. This post assumes you hold a copy of the file as one or more PDFs, whichever way it reached you.

Whichever route you use, write down the date of the copy. A claims file grows with every submission and decision, and your index describes the file as it stood on one day.

Why does a VA claims file arrive as thousands of unsorted pages?

A VA claims file runs to thousands of pages because it collects decades of records from a long list of custodians, and VA's older paper material was scanned into its electronic system. Whether your copy arrives sorted, labelled or as one continuous PDF is not stated on the VA pages cited here, so examine your copy before planning the work.

VA's own description of its claims system explains where the page images come from. The privacy impact assessment for VBMS calls it "an electronic, paperless environment in which to maintain, review, and make rating decisions for veterans' claims" and says VCIP, its claims intake program, "is used to scan-in paper documents that exist for veterans that were used during the legacy paper claim filing process." The same document says VBMS provides an interface for "third-party scanning vendors" to "upload scanned images of documents sent as part of the claims processing workflow."

A scanned image of a document is a picture of a page. It may or may not have had text added afterwards, and the assessment does not address that. So the first hour with a new file is a survey:

1. Count. Note the number of PDF files, the page count of each, and the file sizes.
2. Look for structure. Open the bookmarks panel. If VA's copy has bookmarks or separate files per document, your index is half built. If it has neither, you build all of it.
3. Test for text. On ten pages spread through the file, old and new, try to select a line. Paste what you select into a blank document. Record for each page whether there was no text, good text, or garbled text. The post on why you cannot copy text from a PDF has more ways to tell whether a PDF has real text.
4. Check the order. Read the dates on every fiftieth page. If they rise steadily, the file is chronological. If they jump between decades, it is in some other order, such as the order of receipt.
5. Sample for duplicates. Pick one distinctive page, such as the separation document, and search or page through for copies. A page submitted with three different claims can be in the file three times.

Write the findings at the top of your index. They determine whether you convert the whole file or only the parts with no usable text, and they tell the next person on the matter what they are looking at.

One practical consequence of size: the copy may exceed the page or file limits of whatever tool you use. The post on batch OCR for discovery explains how a batch should be split before upload, and the same arithmetic applies to one very long PDF.

Service treatment records and decades-old military forms

Service treatment records are hard pages to convert: typewritten forms, carbon copies, rubber stamps and handwritten clinic entries, some of them copies of copies. OCR in general does best on the typed forms, worse on faint copies and worst on handwriting. Test a sample of your own pages, and read the entries that matter from the image.

The National Archives describes what these records are. Its page on veterans' medical and health records says health records "cover the outpatient, dental and mental health treatment that former members received while in military service" and "include induction and separation physical examinations, as well as routine medical care (doctor/dental visits, lab tests, etc.) when the patient was not admitted to a hospital." The page adds that clinical (hospital inpatient) records were generated separately and are generally not filed with the health records. The same page explains why these records are in VA's hands: "In 1992, the Army began retiring most of its former members' health records to the Department of Veterans Affairs (VA)", with the other services following.

What each kind of page does in conversion

The right-hand column describes OCR in general, not any one tool, and your own pages are the test.

The character-level reasons typewritten pages misread, including strikeovers and the typist's habit of using a lowercase L for the numeral one, are covered in the post on OCR for typewritten documents. For a claims file the point to carry over is that a date such as l2 Jun 68 may be in the text with a letter where a digit should be, and a search for "12 Jun 68" will miss it.

A routine for the entries that matter

A single line in a service record may be the entry a claim depends on, so convert first and then read by hand:

1. Convert the service records so that typed forms, stamps and dates become searchable where they can be.
2. Page through the images of the whole service-records portion once, with the veteran's claimed conditions beside you. Converted text does not replace this pass.
3. For each entry that bears on a claimed condition, transcribe it into your notes exactly, with the page number, and mark any word you cannot read as illegible.
4. Have a second person check each transcription against the image.

Schema Law's homepage says recognition is strongest on clean print and weaker on handwriting, faint faxes and stamps over text, and that nothing marks a passage as uncertain. On pages like these, assume errors and look.

When older records are thin

A file for an older veteran may hold few service records, or records that look rebuilt from other sources. Before concluding that pages were lost in scanning, read the National Archives' account of the 1973 fire at the National Personnel Records Center, which "destroyed approximately 16-18 million Official Military Personnel Files". It estimates the loss at 80% for Army personnel discharged November 1, 1912 to January 1, 1960, and 75% for Air Force personnel discharged September 25, 1947 to January 1, 1964 with names alphabetically after Hubbard, James E. Whether a particular veteran's records were affected is a question for the Archives. Note the gap in your index and what you did to check it.

Rating decisions, examination reports and agency letters

Rating decisions, examination reports and VA's letters are the modern typed documents in a claims file, the clean print on which OCR is strongest. A brief cites them by date and percentage, so convert them early, confirm the dates and percentages against the image, and use their fixed structure to move through them.

Decision notices

The statute fixes what a decision notice has to contain, and that list is a map of the document. 38 U.S.C. § 5104(b) requires each notice to include:

• "Identification of the issues adjudicated."
• "A summary of the evidence considered by the Secretary."
• "A summary of the applicable laws and regulations."
• "Identification of findings favorable to the claimant."
• "In the case of a denial, identification of elements not satisfied leading to the denial."
• "An explanation of how to obtain or access evidence used in making the decision."
• "If applicable, identification of the criteria that must be satisfied to grant service connection or the next higher level of compensation."

The current wording of subsection (b) dates from a 2017 amendment, so a decision issued before that amendment took effect was written under the earlier text. For each decision, record in your index the date of the notice, each issue, the outcome, and the effective date and percentage where one is stated. Check every date and number against the page. A 30 read as a 50 is the kind of error that typed pages still produce.

The evidence summary in a decision is useful twice over. It lists what the adjudicator looked at, with dates, and each item on that list is a document you should be able to find in the file. Tick them off against your index. Anything listed that you cannot find goes on a follow-up list.

Examination reports

VA's page on the VA claim exam says of the exam, "The provider will write up an exam report and send it to us." It also says the veteran can request a copy with VA Form 20-10206. Reports may be completed on a Disability Benefits Questionnaire. VA's public DBQ page says the questionnaires "help collect necessary medical information to process your disability claims."

A questionnaire is a form: printed questions, then boxes, selections and typed remarks. The printed questions are a dependable search target, and they repeat across every exam of that type in the file. The selections are not dependable in converted text. A marked box may appear as a letter, a symbol or nothing. For every answer you rely on, open the page image and record what is marked there. Typed narrative remarks and medical opinions at the end of a report read like any typed paragraph.

Letters

Development letters, notices of scheduled exams and acknowledgment letters add pages without adding medical evidence. Convert them with everything else, index them by date and type, and move on. Their value is in dates: when VA asked for something, and when it said it received something.

How do you split a converted VA claims file into dated documents?

Split a converted VA claims file by finding where each document begins, giving it a date, a type and a page range in an index, and only then cutting the PDF into one file per document. Build the index against the page numbers of the copy VA sent, so every entry can be traced back to that copy.

Step 1: fix the reference numbering

Before anything else, decide how you will refer to pages. If VA's copy is one PDF, use its page positions, 1 through the last page. If it is several PDFs, use file name plus page. Never renumber after you start. If you add your own page stamps, add them to a working copy and record that they are yours.

Step 2: find the boundaries

Go through the converted file from the front and mark each page where a new document starts. Signals of a first page:

• A letterhead, form number or form title at the top
• A date line and an address block
• "Page 1 of" in a footer, or page numbering that resets
• A change of typeface, paper size or scan quality from the page before
• A fax header with a new transmission date

Converted text speeds this up. Search for the form titles and phrases that open documents in this file, such as the heading of a decision notice or the title line of an application form, and each hit is a candidate boundary. Then confirm by eye. Attachments stay with the document they came with: a private medical record sent in under a cover statement is part of that submission, and the index should say so.

Step 3: date each document

A page in a claims file can carry several dates: the date written, the date signed, a received stamp, a fax line, and a scan date. Choose one rule and apply it throughout. A workable rule is to use the document's own date for the Document date column and to record a received stamp separately when one is legible. For treatment records covering a period, record the first and last dates of treatment.

When a document has no date, leave the date blank and describe its position: "undated, follows the June 2009 exam report". Do not guess a date into a column people will sort by.

Step 4: build the index

Keep the list of types short and fixed: application, statement, service personnel record, service treatment record, VA treatment record, private treatment record, examination report, decision notice, VA letter, appeal document, other. A short list makes the index sortable. Free-text types cannot be sorted or counted.

Step 5: mark duplicates, do not delete them

When the same document appears again, index it and put the start page of the first copy in the Duplicate of column. Leave the pages in place. A later copy may carry a different received stamp or a handwritten note, and the fact that a document was submitted twice can matter.

Step 6: cut and name

Once the index is complete, split the working PDF at the indexed page ranges. Name each file so that it sorts by date and still points back to the source: 2009-06-18 exam report knee p1482-1489.pdf. Undated documents take a prefix that sorts them together, such as 0000-00-00. Keep one unsplit searchable copy as well, because a search across the entire record is still the fastest first look.

For a fictional veteran, Raymond Ostrow, a 3,900-page file might index to 610 documents, of which 140 are duplicates. Sorted by date, the 470 distinct documents read as a history: enlistment, treatment in service, separation, a first claim, a denial, new evidence, an exam, a grant. Sorted by type, the eleven decision notices sit together and can be read in an afternoon.

The index is a spreadsheet you build by hand. Schema Law reads the pages into text and tables and exports them. It does not identify where one document ends and the next begins or assign dates, so the boundary and date columns come from a person reading the file. Its export formats are listed under formats on the homepage.

Which searches find a diagnosis or an in-service event in a claims file?

The searches that find a diagnosis or an in-service event are the condition's names in every form the records use, the symptom words around it, and the places, units and dates of the event. Run them across the converted file, log each hit by page, then read the handwritten service records by eye.

The regulation on service connection says what the evidence has to show, and that frames the searching. 38 CFR § 3.303(a) says service connection "basically" means the facts establish "that a particular injury or disease resulting in disability was incurred coincident with service in the Armed Forces, or if preexisting such service, was aggravated therein", and that each condition is considered "on the basis of the places, types and circumstances of his service as shown by service records". So there are three things to find: the current condition, the event or onset in service, and whatever ties them together.

Searching for a diagnosis

• The formal name and the plain one. "lumbosacral strain" and "low back"; "tinnitus" and "ringing"; "PTSD" and "posttraumatic" with and without a hyphen or space.
• Abbreviations. Clinic notes shorten: "LBP", "DJD", "HTN", "s/p". List the abbreviations you see in the first few records and search each.
• Diagnostic codes. If a decision assigns a numbered diagnostic code to a condition, search that number across the decisions.
• Misspellings and misreads. Search a distinctive fragment, such as "tinnit" or "lumbo", to catch variants and words broken by recognition errors.

Searching for an in-service event

• Places and units. Base names, ship names, unit designations, country names, in the spellings of the period.
• Dates. The month and year in several formats: "Jun 68", "June 1968", "6/68", "680612". Then the same with a lowercase L in place of a 1, for typewritten pages.
• Event words. "fell", "accident", "blast", "struck", "treated", "sick call", "profile", "light duty", "hospital".
• Separation exam language. Read the separation examination and any medical history completed with it, and search the condition's plain name on the pages near the separation date.

Searching for the link

• Opinion phrases. "at least as likely as not", "less likely than not", "related to", "due to", "caused by", "aggravated", "nexus". Add any other phrasing you see an examiner in this file use to state a conclusion.
• Continuity. The condition's name between separation and the first claim, year by year. Gaps and their length are part of the record either way.
• Favorable findings. Search decisions for "favorable", since 38 U.S.C. § 5104(b) requires the notice to identify findings favorable to the claimant.

Keeping a hit log

Log every hit that matters: term, page, document from the index, date, the sentence quoted exactly, verified against image or not. A hit list sorted by document date shows the condition's history in order.

Search has a floor. A word misread in conversion will not be found, and handwritten entries may not be in the text at all. A search that returns nothing means nothing was recognized, and the page-by-page read of the service records is still required.

Where the same veteran also has a Social Security disability claim, the two files may hold some of the same medical evidence, arranged differently. The post on OCR for a Social Security disability file covers the exhibit sections of that file. A claims file holds health records and identifiers for the veteran and may hold them for dependents, so check any outside service against your duties before uploading. Schema Law's commitments are under security on the homepage.