Workers' comp apportionment: the rules, the records and the paragraph that holds up

A worker with a bad back rarely has only 1 reason for it. There is the lift at work that ended with an ER visit. There may be a claim 7 years earlier against a different employer, an MRI from a weekend injury, a decade of disc degeneration, a primary care chart full of back pain complaints. Apportionment asks how much of the permanent disability that exists now belongs to the current industrial injury, and how much to everything else.

The California Supreme Court put the purpose plainly in 2007:

"Employers must compensate injured workers only for that portion of their permanent disability attributable to a current industrial injury, not for that portion attributable to previous injuries or to nonindustrial factors."

3 limits frame everything else in this guide.

1. It applies to permanent disability. Apportionment splits the permanent disability rating. It does not split the employer's duty to provide medical treatment for the industrial injury, which is why chapter 3 spends time on Hikida.
2. It is about causation of the disability, not causation of the injury. The lift caused the injury. The question is what is causing the disability measured at the evaluation: the disc herniation from the lift, the degeneration that was already there, a prior injury to the same spine, or a mix.
3. It is measured at the time of the evaluation. Escobedo requires the physician to explain how and why each factor is causing permanent disability "at the time of the evaluation."

Senate Bill 899 rewrote California apportionment in 2004. Section 4664 took effect on April 19, 2004, and section 4663 was replaced the same day. Most of what follows has been litigated since, but the text is short enough to read in full, and every evaluator should. The core sentence is 4663(c):

"A physician shall make an apportionment determination by finding what approximate percentage of the permanent disability was caused by the direct result of injury arising out of and occurring in the course of employment and what approximate percentage of the permanent disability was caused by other factors both before and subsequent to the industrial injury, including prior industrial injuries."

Read the last clause twice. "Other factors" includes prior industrial injuries. So a prior work injury can be dealt with under 4663 as a causation factor, and, if it produced an award, under 4664 as well. The 2 sections do different jobs.

The rest of 4663 and 4664, in plain words

4663(b)

A physician who reports on permanent disability must address the causation of the permanent disability in that report.

4663(c), second half

If the physician cannot make an apportionment determination, the report must state the specific reasons why, and the physician must then consult other physicians or refer the worker to one who can make the final determination.

4663(d)

A worker claiming an industrial injury must, on request, disclose all previous permanent disabilities or physical impairments. The disclosure is a starting list, not evidence. Check it against the records.

4664(b)

A prior permanent disability award is conclusively presumed to still exist at the time of a later industrial injury. The statute calls this "a presumption affecting the burden of proof."

4664(c)(1) regions

Hearing; vision; mental and behavioral disorders; the spine; the upper extremities including the shoulders; the lower extremities including the hip joints; and the head, face, cardiovascular system, respiratory system and all other systems.

The regions list decides more 4664 fights than people expect. A prior award for a shoulder does not presume anything about a lumbar spine, because they sit in different regions. A prior neck award and a current low back claim are both "the spine." The evaluator should say in plain words which region each prior award covered, with the page cite, before anyone starts arguing overlap.

The statutes say what to decide. The cases say what a decision has to look like before a judge can rely on it. 6 decisions do most of the work, and every evaluator who writes on apportionment in California should be able to name them.

Escobedo v. Marshalls (2005): the how-and-why test

The WCAB decided Escobedo v. Marshalls en banc in 2005, and en banc decisions bind every WCAB panel and workers' compensation judge. It held that a report which makes an apportionment determination still cannot be relied on unless it is substantial evidence, and it said exactly what that means:

"A medical opinion must be framed in terms of reasonable medical probability, it must not be speculative, it must be based on pertinent facts and on an adequate examination and history, and it must set forth reasoning in support of its conclusions."

Then it gave the example that became the test. A physician who says 50% of a back disability comes from the industrial injury must explain how and why the disability is causally related to the injury and how and why the injury is responsible for about 50%. A physician who blames 50% on degenerative disc disease must explain the nature of the disease, how and why it is causing permanent disability at the time of the evaluation, and how and why it accounts for about 50%. The opinion must also, in the Board's words, "describe in detail the exact nature of the apportionable disability, and set forth the basis for the opinion."

Notice what that demands from the record. "The nature of the degenerative disc disease" is a finding on an imaging report with a date. "At the time of the evaluation" means comparing old studies to new ones. An evaluator without the prior imaging can describe degeneration in general terms, and general terms are what get a report rejected.

Kopping v. WCAB (2006): proving a prior award

"First, the employer must prove the existence of the prior permanent disability award. Then, having established by this proof that the permanent disability on which that award was based still exists, the employer must prove the extent of the overlap, if any, between the prior disability and the current disability."

WCAB panels still deny 4664 credit where the defense never put the prior award into evidence. It is the cheapest apportionment argument to lose: the document exists, someone just did not order it.

Benson v. WCAB (2009): separate injuries, separate shares

Where there are successive injuries, the evaluator apportions to each. Benson is also the source of the rule that a physician who weighs every factor and concludes the percentages cannot be approximated to a reasonable medical probability has still made the determination 4663(c) requires. That is a narrow door, and it only opens for an evaluator who shows the work.

Hikida (2017) and Justice (2020): disability from treatment

In Hikida v. WCAB (2017) 12 Cal.App.5th 1249, a worker developed complex regional pain syndrome after carpal tunnel surgery for an industrial condition. The Court of Appeal held the resulting disability was not subject to apportionment, because medical treatment for an industrial injury is not apportioned and neither is new disability that the treatment causes. In County of Santa Clara v. WCAB (Justice) (2020) 49 Cal.App.5th 605, the court limited that: Hikida applies where industrial treatment produces a new, more disabling condition, not every time surgery happens. In a panel section of Vigil (2024), outside its en banc holding, the Board applied Justice to a worker whose hip replacements went well: Hikida did not bar apportionment, but the defense still lost because the QME's opinion was conclusory.

Nunes (2023): no end-run through vocational evidence

In Nunes v. State of California, Dept. of Motor Vehicles (2023) 88 Cal.Comp.Cases 741, an en banc decision, the WCAB held that vocational evidence must address apportionment and may not substitute "vocational apportionment" for valid medical apportionment. A vocational expert's opinion does not replace the physician's apportionment analysis.

People who have not read many apportionment reports picture 1 number. In practice there are usually 3 or more causes and at least 2 reports that disagree. The visual below uses the hypothetical file from chapter 5: a lumbar disability with a current industrial injury, a prior industrial injury that ended in an award, and non-industrial degeneration documented on an MRI.

A few cautions about reading a bar like this.

• The bar shows causation shares, not a rating formula. The evaluator states approximate percentages of the permanent disability under 4663(c). A disability rating specialist or the parties then apply those shares to the rating. How a 4664 prior-award credit is applied on top, and how it interacts with a 4663 share assigned to the same prior injury, is a rating and legal question that the parties argue and the judge decides. Keep the medical report to the medical facts: what the prior award covered, which region, and whether and how the prior disability overlaps the current one.
• Each segment needs its own how and why. The 25% for the prior injury needs a reason. So does the 20% for degeneration. So does the 55%, because under Escobedo the applicant carries the burden on the industrial share.
• The 10% in the first report was not wrong because it was small. It was weak because it rested on a history in which the worker denied prior back trouble, and the record contradicted the history. A report built on an inadequate history fails Escobedo no matter what number it lands on.

What each record contributes

The worker's "no prior back problems" is not necessarily a lie. People forget a claim from 9 years ago, or do not count a closed claim as a "problem." The record, not an accusation, corrects the history.

An apportionment opinion is only as good as the record set it was built on, and in California the path that records take to a medical-legal evaluator is regulated. Section 4062.3 sets it out.

Step 4 is the one evaluators under-use. The 3 lists in 4062.3(d) are a statutory requirement, and they are also the best defense of an apportionment paragraph: a reader can see at a glance that the prior award was received, reviewed and relied on. Build them from the actual file, with page ranges, not from the cover letter's description of what was supposedly enclosed.

What to pull, by source, in priority order

APPORTIONMENT RECORDS LIST
Claim: [CLAIM NO. / CASE NO.]      Worker: [NAME]      DOI: [DATE]
Body parts claimed: [e.g. LUMBAR SPINE, RIGHT SHOULDER]
Look-back period: [e.g. 10 YEARS BEFORE DOI] to present

A. PRIOR CLAIMS AND AWARDS (same and related body regions)
[ ] Prior claim numbers, employers, dates of injury
[ ] Awards, stipulations with request for award, compromise and release
[ ] Rating strings and the body part / region each award covered
[ ] Prior P&S / MMI reports, prior QME and AME reports, depositions
[ ] Prior work restrictions and return-to-work status at claim closure

B. PRIOR IMAGING (region claimed)
[ ] Radiology reports: X-ray, MRI, CT, with dates and facility
[ ] Images on disc or portal, where comparison will be disputed
[ ] Any report referenced in notes but not yet produced: [LIST]

C. PRE-INJURY TREATMENT
[ ] Primary care: problem list, visit notes for the region, referrals
[ ] Chiropractic, physical therapy, acupuncture: visit logs, discharge notes
[ ] Pain management, injections, specialist consults
[ ] Urgent care and ER visits for the region

D. PHARMACY
[ ] Fill history from [PHARMACY / PBM] for [DATE RANGE]
[ ] Analgesics, NSAIDs, muscle relaxants, neuropathic agents

E. NON-INDUSTRIAL CONDITIONS
[ ] Documented conditions the evaluator may weigh: [e.g. DIABETES,
    INFLAMMATORY ARTHRITIS] with the records that diagnose them
[ ] Non-work injuries before or after DOI (MVA, falls, sports)

F. CURRENT CLAIM
[ ] All treating records, diagnostics, operative reports
[ ] Work status slips and restriction history
[ ] P&S / MMI report and any later treatment

G. DISCLOSURE
[ ] Worker's 4663(d) disclosure of prior disabilities, if requested
[ ] Gaps: providers or dates named in records but not produced

Prepared by: [NAME]   Date: [DATE]   Served on: [PARTY] [DATE]

The evaluator's record review for apportionment is a sequence of questions, and each answer tells you what to look for next. This is the path we would follow on a file like the one in chapter 5. It is a method for organizing the review, not a substitute for the physician's judgment on any branch.

The difference between a report the judge relies on and one sent back for development is usually visible in 1 section. Below is the hypothetical QME's apportionment section before and after the prior records arrived. The first version is not unusual. That is the problem.

The original fails on at least 4 of the grounds the WCAB lists: inadequate history, no reasoning, "training and experience" in place of explanation, and an opinion about a factor (age) without describing the disease at all. A thin opinion also costs time. In a 2025 panel decision, a judge found a QME's 30% industrial, 70% non-industrial split self-contradictory and not substantial evidence; the Board granted reconsideration because the record might need further development. That means more reports, more months, and the same records someone could have gathered the first time.

Anatomy of a paragraph that holds

The "why 25%" row is the weakest part of most real reports, and it is the part the case law is least forgiving about. There is no formula. There is only a stated reason that a reader can test: which restrictions overlap, which findings are new, what the imaging comparison shows. If the reason is "roughly half the current limitations were already present in 2017," say which limitations and cite the pages.

APPORTIONMENT: [BODY PART]
Legal basis: Labor Code 4663 and 4664; Escobedo v. Marshalls (2005)
70 Cal.Comp.Cases 604.

Records considered for apportionment:
Received: [LIST WITH PAGE RANGES]
Reviewed: [LIST]
Relied on: [LIST]
Referenced in the records but not received: [LIST, WITH THE PAGE
THAT REFERENCES EACH]

Current permanent disability: [SUMMARY OF IMPAIRMENT AND
RESTRICTIONS AT THIS EVALUATION, DATE]

FACTOR [N]: [NAME, e.g. INDUSTRIAL INJURY OF MM/DD/YYYY /
PRIOR INJURY OF MM/DD/YYYY / DEGENERATIVE DISC DISEASE L4-S1]
Evidence: [FINDING], [SOURCE, DATE], p. [X]; [FINDING], p. [Y]
Prior award (if any): [%] PD, [BODY PART AS WRITTEN], [DATE],
p. [Z]. Region under Labor Code 4664(c)(1): [REGION].
Nature of the condition: [WHAT IT IS, LEVELS, SEVERITY, AS SHOWN
ON (STUDY) OF (DATE), p. (X)]
How it causes disability now: [MECHANISM LINKING THIS FACTOR TO
THE CURRENT IMPAIRMENT OR RESTRICTIONS]
Why [N]%: [REASONING FOR THE SHARE: WHICH FINDINGS OR RESTRICTIONS
ARE ATTRIBUTABLE TO THIS FACTOR AND WHICH ARE NOT]
Treatment note: [WHETHER ANY DISABILITY RESULTS FROM TREATMENT OF
THE INDUSTRIAL INJURY, AND WHY]

SUMMARY
Industrial injury of [DATE]: [N]%
[FACTOR]: [N]%
[FACTOR]: [N]%
Total: 100%
These opinions are stated within reasonable medical probability.

If a determination cannot be made: the specific reasons are
[MISSING RECORDS / OUTSIDE SPECIALTY], and I [CONSULTED /
RECOMMEND REFERRAL TO] [SPECIALTY] per Labor Code 4663(c).

The California Supreme Court's standard, which the WCAB applies to every medical report, is short:

"Medical reports and opinions are not substantial evidence if they are known to be erroneous, or if they are based on facts no longer germane, on inadequate medical histories and examinations, or on incorrect legal theories. Medical opinion also fails to support the Board's findings if it is based on surmise, speculation, conjecture, or guess."

Here is how that plays out in apportionment, with the fix for each. Most of the fixes are record problems, not writing problems.

Our opinion, after reading a lot of these: the single most effective habit is reconciling the history with the records in a dedicated paragraph. It forces the evaluator to read the prior records rather than list them, it defuses the "inadequate history" attack in advance, and it gives the judge a reason to trust the rest of the section.

2 variables change across state lines: which edition of the AMA Guides produces the impairment number, and what the state lets you subtract from it. California's answers are unusual on both. Do not carry its vocabulary, its cases or its edition into another state's claim.

Which book you are rating in

California rates under the AMA Guides 5th edition. Section 4660 incorporates it for injuries before 1 January 2013, and section 4660.1(b) keeps it for injuries on or after that date, with whole person impairment multiplied by 1.4. The legislature has not moved California to the 6th edition. The AMA now maintains the 6th edition as a digital product revised by year, so "6th edition" in another state can mean the 2008 printing or a later version, depending on how that state's rule is worded.

Florida is the sharpest contrast with California. In California, Hikida rests on the long-standing rule that medical treatment for an industrial injury is not apportioned. Florida's 440.15(5)(b) says medical benefits "shall be paid apportioning out the percentage of the need for such care attributable to the preexisting condition." An adjuster who moves between the 2 states and carries a habit across will get it wrong in 1 of them.

Texas is the live one to watch. The division's notice names no effective date for the move to the 6th edition, but if it lands, Texas evaluators will rate the same injuries in a different book, and every rating will need to name which one. The IME physician field manual covers edition questions across states.

Apportionment is a good fit for AI in 1 narrow sense and a bad fit in another. The good fit: the work is finding and ordering evidence across hundreds of pages from several claims and providers, and a large language model (LLM) reading a record set does that fast. The bad fit: the output is a medical opinion under a legal standard, signed by a physician, and no software should write it.

What the law already says about who summarizes the records

California wrote a rule for this before anyone had heard of generative AI. Labor Code 4628(a) says no one other than the signing physician may participate in the nonclerical preparation of a medical-legal report, and lists "reviewing and summarizing prior medical records" as part of that preparation. Section 4628(c) then allows an exception:

"If the initial outline of a patient's history or excerpting of prior medical records is not done by the physician, the physician shall review the excerpts and the entire outline and shall make additional inquiries and examinations as are necessary and appropriate to identify and determine the relevant medical issues."

The statute speaks of persons, and we know of no California rule that addresses AI summaries directly. But the practical reading is easy to defend: treat an AI medical chronology the way 4628(c) treats a staff outline. Read all of it, check the citations against the pages, and make your own inquiries. Section 4628(e) makes a noncompliant report inadmissible, so this is not a place to cut corners. Whether and how to disclose software use in the report under 4628(b) is a question to settle with counsel or the DWC before you need the answer.

Where AI helps on an apportionment file

• Cross-claim chronology. Merging a prior claim file and a current one into 1 dated sequence by body part is exactly the task that takes hours by hand and that software does well, as long as every entry carries a page-level citation back to its source.
• Referenced-but-missing records. A note that says "MRI shows DDD" with no MRI report in the file is a pattern clinical NLP can catch reliably. Chapter 5's gap was on page 391.
• History versus record. Separating what the worker reported from what a clinician found, so the evaluator sees contradictions early.
• Pharmacy and work status timelines. Long, repetitive printouts turned into date ranges with counts.
• Search and questions. "Every mention of lumbar pain before 05/10/2023, with pages" is a better query for retrieval-augmented generation (RAG) than for a tired reviewer at 11 pm.

Where it fails, and what catches it

We feel strongest about 1 point. A tool that volunteers apportionment percentages is a liability to the evaluator who uses it. The percentage has to come out of the physician's reasoning, and a number suggested by software before that reasoning exists is exactly the kind of conclusion without analysis that Escobedo rejects. Agentic AI that chains tasks across a file is useful for assembly. It should stop short of the opinion.

On the courtroom side: federal courts have sanctioned lawyers under Rule 11 for filing briefs with fabricated, AI-generated case citations, most famously in Mata v. Avianca, Inc. (S.D.N.Y. 2023). The WCAB is a different forum, but the lesson carries: verify every citation before you sign or file, whether it points to a case or to page 261 of a claim file.

Vendor checklist for legal AI tools on a claim file

• Signs a business associate agreement (BAA)claim files carry PHI; HIPAA compliant AI starts with the BAA
• SOC 2 report availableask for it, read the exceptions
• No training on your fileswritten into the contract
• A citation on every lineto the page, not the document
• Flags records referenced but not producedthe chapter 5 gap
• Separates patient-reported history from clinician findingsthe inadequate-history attack starts here
• Flags low-confidence OCR and handwritinginstead of guessing silently
• Handles 2 or more claims in 1 fileand keeps each fact's claim of origin
• Keeps an audit trail of AI usewho ran what, on which file, when
• Does not assign apportionment percentageshuman-in-the-loop means the physician decides

0 of 10 checked

For a longer treatment of accuracy and admissibility, see whether AI is accurate enough for court and HIPAA compliant AI medical record review. Adjusters and TPA examiners using AI for workers' comp files will find the claims side, including AI claims processing, in the workers' comp adjuster manual and the TPA manual.

Medrecords AI is medical record review software. You upload the records you have; it drafts a cited chronology and flags, and every line links to its source page. On an apportionment file, that means the record basis for the opinion, organized across the prior and the current claim. It does not decide apportionment, suggest percentages or write the medical opinion. You review, you revise, you sign.

For raters, the impairment rating worksheet pulls measured findings, the MMI evidence and the apportionment record (prior injuries, prior ratings, pre-existing findings) into cited rows. The edition, the table, the apportionment call and the signature stay with the physician. Faxed and handwritten pages go through OCR with per-page confidence flags. Security is SOC 2 and HIPAA with a BAA; see security and HIPAA. Plans are on the pricing page.

What it does not do: retrieve records from providers or search the state case system for prior awards, rate impairment, decide overlap, or tell anyone what percentage to write. Flags are signals, not verdicts. A missing-record flag tells you where to look; it does not prove the record exists.

What is the difference between Labor Code 4663 and 4664?

Section 4663 requires the physician to apportion today's permanent disability by cause, including non-industrial factors and prior industrial injuries. Section 4664 limits the employer's liability to the industrial share and conclusively presumes that a prior permanent disability award still exists. Under Kopping, the employer must still prove the prior award and the extent of overlap.

What did Escobedo v. Marshalls decide?

The 2005 en banc WCAB decision held that apportionment is to the causation of permanent disability, that it can include pathology and asymptomatic prior conditions, and that the opinion must be substantial medical evidence. The physician must explain how and why each factor causes disability at the time of the evaluation and how and why it accounts for the percentage assigned.

Can a workers' comp disability be apportioned to degenerative disease?

In California, yes, if substantial medical evidence shows the degenerative condition is causing part of the permanent disability now. A finding on imaging is not enough by itself; the physician must describe the disease and explain the mechanism. Other states use different rules, so check the statute where the claim is filed.

Who has the burden of proof on apportionment?

Under Escobedo, the applicant proves the percentage of permanent disability caused by the industrial injury, and the defendant proves the percentage caused by other factors. For prior-award credit under 4664, the defendant proves the award and the overlap.

What if the evaluator cannot determine apportionment?

Section 4663(c) requires the report to state the specific reasons, and the physician then consults other physicians or refers the worker to one who can decide. Under Benson, a physician who weighs every factor and concludes the percentages cannot be approximated to a reasonable medical probability has made the determination, if the reasons are stated.

Can AI decide apportionment percentages?

It should not. Apportionment is a medical opinion under a legal standard that requires reasoning a judge can test. AI medical record review software can assemble the prior and current records into a cited chronology and flag gaps; the physician writes the percentages and the how and why.

Is it HIPAA compliant to upload a workers' comp claim file to an AI tool?

It can be, if the vendor signs a business associate agreement, secures the data (a SOC 2 report is the usual evidence), and does not train its models on your files. Consumer chatbots without a BAA are the wrong place for a claim file. Your own obligations depend on your role and state law.

Can ChatGPT summarize prior medical records for a QME report?

A general chatbot can produce a fluent summary, but without page-level citations you cannot check it, and it can hallucinate findings. Under Labor Code 4628(c), a physician who did not excerpt the records personally must review the excerpts and the entire outline. Use a tool that cites every line, and read the source pages.

Does a QME have to disclose using AI software to summarize records?

Section 4628(b) requires disclosure of each person who performed services in connection with the report, other than clerical preparation. It does not mention software, and we know of no California rule that addresses AI summaries directly. Settle your practice with counsel or the DWC, and follow 4628(c) regardless.

Statutes were read on the California Legislature's site in September 2026, and quoted text is verbatim. Escobedo was read in the WCAB's own copy of the en banc decision; Kopping, Brodie, Hegglin, Benson, Hikida, Justice and Nunes are quoted or described as the WCAB quotes and applies them in the decisions linked below. The case example is hypothetical throughout. Product facts come from this site's product pages. Nothing here is legal or medical advice.

• California Labor Code 4663, apportionment based on causation, (a) to (e).
• California Labor Code 4664, liability for the industrial share, prior award presumption, 100% regional cap and the 7 regions. Added by Stats. 2004, ch. 34, effective April 19, 2004.
• Labor Code 4660 (AMA Guides 5th edition, injuries before 2013) and 4660.1 (5th edition with the 1.4 adjustment, injuries from 2013).
• Labor Code 4062.3, records provided to QMEs and AMEs, 20-day service, 10-day objection, and the evaluator's 3 lists.
• Labor Code 4628, who may prepare a medical-legal report, review of excerpts and outlines, inadmissibility and the $1,000 penalty.
• Escobedo v. Marshalls (2005) 70 Cal.Comp.Cases 604 (Appeals Board en banc), holdings and the how-and-why passage.
• Vigil v. County of Kern (2024), WCAB en banc, quoting Brodie, Nunes and Benson; its panel section III applies Hikida and Justice.
• WCAB panel decision, 2025, applying Kopping v. WCAB (2006) 142 Cal.App.4th 1099, the 2-step burden for prior-award credit.
• WCAB panel decision, 2025, quoting Escobedo at 621 and Hegglin v. WCAB (1971) 4 Cal.3d 162, 169.
• WCAB panel decision, 2025, where the judge found a QME's apportionment self-contradictory and the Board granted reconsideration, citing Escobedo, Kopping and Gay v. WCAB (1979) 96 Cal.App.3d 555 on the defense burden.
• 28 Tex. Admin. Code 130.1, AMA Guides 4th edition for Texas ratings, and the Texas DWC notice on transition to the 6th edition, 2025 version.
• Florida Statutes 440.15, (3)(b) rating schedule and (5)(b) apportionment of pre-existing conditions and medical benefits.
• Pennsylvania Bureau of Workers' Compensation, News & Notes, Summer 2019, on Act 111 of 2018 restoring impairment rating evaluations; the 6th edition (second printing, April 2009) and the 35% threshold are set by Act 111's section 306(a.3), 77 P.S. 511.3.
• Mata v. Avianca, Inc., 678 F. Supp. 3d 443 (S.D.N.Y. 2023): Rule 11 sanctions for fabricated AI-generated case citations.

Related guides: workers' comp IME record review, pre-existing conditions and the eggshell plaintiff, how to get medical records for a lawsuit, California QME record review and impairment rating record review.