How to become an AI-native IME physician in 2026

5 job titles describe the same underlying act: a physician who examines someone they are not treating, reads the file, and writes an opinion that a court, a carrier or a state agency will rely on. The titles differ by who appoints you and who pays.

The distinction matters more than the acronyms suggest. In a fee-schedule state your rate is a published number you cannot negotiate, and your economics turn entirely on how many evaluations you complete and how large the record files are. In private IME work you set the price, and your economics turn on what you can defend and who will pay it.

What the opinion has to answer

A California QME report has to reach 7 findings. Other states word them differently and ask for most of the same things.

1. Causation. Is the injury or illness work-related to a reasonable medical probability?
2. Disability. Does it affect the person's ability to work, temporarily or permanently?
3. Return to work. When, and with what restrictions?
4. Impairment rating under the AMA guides, which then drives the disability rating.
5. Apportionment. How much of the impairment belongs to non-industrial causes or an earlier industrial injury?
6. Permanent and stationary date, where one applies.
7. Future medical care, if any is needed.

Since 2013, a California QME may not comment on current or ongoing treatment disputes. That moved to independent medical review. 2 of the former QMEs in RAND's focus groups named this restriction as part of why they left: they could see a treatment need, say so, and watch it count for nothing.

Demand for medical-legal evaluations in California has grown almost every year since 2012, and the supply of evaluators has not tracked it. That gap is the whole opportunity, and it is unevenly distributed across specialties in a way that decides whether this work is worth your time.

Read the ratio, not the volume

Volume tells you how busy a specialty is. The ratio of assignments to evaluators tells you how badly the system needs you.

Physician counts from the Association of American Medical Colleges 2023 data, chiropractor count from the Bureau of Labor Statistics, both as reported by RAND.

2 readings jump out. Orthopedic surgery is saturated in the sense that it has the most evaluators, and still carries 115 assignments per evaluator. Internal medicine has 16,736 physicians in the state and 189 QMEs, which is 1.1% of them. RAND also found subspecialties where the state cannot field a panel at all, oncology among them, because fewer than 5 certified evaluators exist in the subspecialty.

A panel assignment is not a case

This is the number that surprises people who join expecting the volume figures to be income. In a represented case the state issues a panel of 3 evaluators in your specialty; each side strikes one, and one of you gets the exam. RAND linked panels to actual billing and found that of the 2023 panels they could match to a claim, 60.5% produced an evaluation. That figure has climbed steadily from 45.9% in 2016, but it still means roughly 2 in 5 panel appearances never become work.

How much work one evaluator actually gets

Apply the utilisation rate to those medians and the shape of a practice appears. An orthopedic surgeon holding a single specialty sees around 67 evaluations a year at the median. A chiropractor holding one specialty sees around 11. An evaluator carrying multiple specialties including orthopedics sees around 175, which is a real second job.

In California your fee is a published number. You cannot discount it to win work and you cannot raise it to reflect a hard file. The schedule below took effect on 1 April 2021 and was the first change since 2006. As of September 2026 it has no cost-of-living adjustment, and bills to add one have not passed.

8 CCR 9795. Every relative value is multiplied by $16.25.

The modifiers are the pay rise

Modifiers multiply the relative value, so they multiply everything except the per-page record fee.

The psychiatric multiplier is contested by the people who receive it. Psychiatrists and psychologists in RAND's focus groups argued that doubling the fee does not cover an evaluation that runs 3 hours where an orthopedic exam runs 20 minutes, on top of a causation standard that is harder to meet. One active evaluator described the change as an 18 percent pay cut. Take that as a warning about specialty economics, not about the specialty.

Record review is now the bigger half

The 2021 schedule replaced hourly billing with a flat fee plus $3 a page past 200 pages. RAND found that growth in medical-legal spending since then has been driven principally by that per-page fee. Across 198,774 California cases with at least one paid record-review line:

RAND, Figure D.7. Page equivalents are ours: payment divided by $3, plus the 200 included pages.

Read the median line again. On a typical case with any record review at all, the file runs past 500 pages, and the fee schedule pays you $939 on top of the $2,015.00 evaluation to read them. That is the single largest number under your control, and chapter 8 is about what it costs you in hours to earn it.

What the change did to the system

Outside California

Texas pays designated doctors a maximum allowable reimbursement, adjusted each year against the Medicare Economic Index. The 2026 figures:

Texas Department of Insurance, Division of Workers' Compensation, 2026 professional fee schedule. The conversion factor rose 2.7% for 2026.

The gap is deliberate and worth understanding before you relocate a practice. One evaluator in RAND's focus groups holds both a California QME appointment and a Texas designated doctor certification, and travels to California for evaluations because California pays more even after the travel.

Private IME work, where you set the price

Outside a state fee schedule nobody publishes your rate. The closest thing to a market reference is SEAK's 2024 survey of 325+ examiners across 54 specialties.

Note what the hourly figure is doing. Nobody quotes $441 an hour. It is $2,890 divided by 6.5 hours. Cut the hours and that number moves, which is the entire argument of this manual. Testimony prices separately and higher, generally above $500 an hour.

California appoints QMEs in 2-year cycles. Nothing about the process is difficult; it is a sequence of forms, 1 exam and one course, and the whole thing is gated on board certification you already hold.

What you need before you apply

• A California licence in good standing, in one of the licence classes the labor code treats as a physician: MD, DO, psychologist, optometrist, dentist, podiatrist, acupuncturist or chiropractor.
• Board certification in the specialty you want to be listed in. Since 2015 the state enforces board certification for subspecialties at appointment and at reappointment. This is what emptied some subspecialty panels.
• An office the medical unit can certify. You must keep it for at least 180 days from certification. Panels are assigned from the worker's home zip code, so where your offices sit decides which cases reach you.

The exam and the course

The competency exam runs 3 hours and is offered twice a year, with application windows published on the DIR site. Passing it is not the end. Since February 2024 every applicant must also complete a 25-hour course covering the material in 8 CCR 14 before the state will certify them, and chiropractic applicants must have completed it within 3 years of applying. Final certification also requires a report-writing sample and a writing class.

Staying appointed

• 12 hours of continuing education every 24 months. On your first reappointment the state credits the exam itself as 6 of those, so you only need 6 more.
• An annual fee to keep the appointment live.
• 2 recent reports, submitted through the online portal. This is new: 8 CCR 50 was amended in 2024 to require it at reappointment. Your reports are now reviewed by someone other than the attorney who deposes you, which is the first structured quality feedback the system has ever offered.
• Up to 120 unavailability days a year, raised from 90 in 2024, declared on QME Form 109.

What the training does not teach

Former evaluators in RAND's focus groups were consistent on this point: the coursework prepares you to pass, not to work. One described a 1-day course that left them not feeling prepared to do the job alone. Another described continuing education as watching a video. The people who found their footing did it through templates and mentors, not curriculum. Colorado supplies its evaluators with a report template. California does not require one, though the medical unit and specialty societies publish some, and evaluators who mentor consistently hand their mentees a specialty-specific template on day one.

RAND reviewed the medical-legal process in 17 states and found no other state doing exactly what California does. If you work outside California, or you are deciding where to build, the differences are structural and they change the job.

RAND review of workers' compensation policy in 17 states: AZ, CO, FL, GA, HI, IL, MD, MI, NC, NJ, NM, NV, NY, OH, PA, TX, WA.

3 systems worth knowing

2 things stand out for anyone planning where to work. Colorado is the only state in the review that has taken the record packet away from the parties and given it to the agency, which is the single largest source of wasted evaluator time everywhere else. Texas is the only one that will tell you, on a schedule, how your reports compare to everyone else's.

Every structural complaint in RAND's interviews traces back to the same thing: the file. Not its contents, its handling. Records arrive late, arrive duplicated, arrive culled by someone with a stake in the answer, or arrive in a volume no one can read inside the deadline.

The rule that causes it

Under 8 CCR 35 the parties exchange the records and the cover letter with each other at least 20 days before sending them to you, and each side then has 10 days to object. The consequence is that you are not entitled to the records until 10 days after the evaluation. If they have not arrived by then you must write and serve the report anyway, using whatever you have, inside the 30-day deadline under 8 CCR 38.

So the standard sequence is legal, routine, and backwards:

What that does to an evaluator

• You examine without knowing what you are looking for. One evaluator in RAND's focus groups described examining a neck and learning from the cover letter, 10 days later, that the claim concerned a foot. Another with multiple specialties reported not knowing on the day of the exam which specialty they were meant to be applying.
• The writing window collapses. An evaluator described receiving 2,000 pages 10 days after the evaluation, with the report already 90 percent written, leaving 20 days to reconcile the 2.
• Injured workers notice. Represented workers in RAND's interviews described arriving to find the evaluator had no records at all. One unrepresented worker phoned ahead to confirm the records had arrived. Another collected them personally and carried them to the appointment.
• The volume is not theoretical. A medical management company representative told RAND that a packet can run 10,000 to 15,000 pages, and called records arriving 9 days post-evaluation the single issue most in need of fixing.

Culling, and why it costs you money

Since the fee schedule started paying $3 a page past 200, defence attorneys have an incentive to send fewer pages, and evaluators report that they do. One active evaluator described attorneys removing hundreds of pages that other physicians in the file had referenced, so the absence was visible. That evaluator's point was not about the fee. It was that deciding which records are clinically relevant is a medical determination, and a party to the dispute was making it.

The system's own correction is expensive: an evaluator who was not given a record writes a supplemental report later, at $650.00, which is a worse outcome for the payer than sending the pages would have been.

The one thing everyone agrees on

RAND found no participant of any kind, on any side, who objected to removing duplicate records before they reach the evaluator. Defence representatives, employer representatives, active evaluators and former evaluators all endorsed it. One former evaluator described spending significant time removing duplicates personally and pointed out that a secretary could do it.

That is the rarest thing in a contested system: unanimous agreement on a piece of work that nobody is currently paid to do. Chapter 8 is about who does it now.

Medical-legal work is governed by deadlines you did not set and cannot move. Missing one costs you the case, and 1 percent of all replacement panels each year are triggered by a late report. Here is every clock that touches you.

8 CCR 30, 31.3, 33, 34, 35, 35.5, 38 and the labor code sections cited in RAND Table 2.1.

How the scheduling clock actually runs

The scheduling window was 60 to 90 days until 2020, moved to 90 to 120 under a COVID emergency regulation, and was made permanent in 2023. The measured reality is slower than either number.

RAND, Table S.3, linked panel and claims data. n = 67,747 panels in 2023.

The average has improved by 32 days since 2016 and the share reaching an exam within a year has risen from 39.0% to 53.7%. The share reaching an exam within 90 days has not moved. What improved is the long tail, not the front of the queue.

Most writing about AI and medical-legal work is speculation. This part is not. A US state workers' compensation agency has already put an AI system into production on evaluator record packets, told RAND exactly what it does, and told RAND exactly what it cannot do.

What Colorado built

Colorado's division took the record packet away from the parties for DIME cases. The division now receives the records, removes duplicates, organises them, obtains anything missing from the opposing party, and hands the evaluator a finished packet. An AI system does the first pass. A person then checks that the AI did it correctly.

The division was deliberate about scope, and the sequence is the useful part:

1. Chronological order. Put the pages in date sequence.
2. De-duplication. Remove copies of the same document.
3. Bates numbering. Give every page a stable citable number.
4. An index. So the evaluator can find a document without reading to it.
5. Relevance. Explicitly not attempted yet, and described by the division as an entirely different level of complication.

The division described the early returns as positive and the expectations as modest on purpose. They also noted that having the parties give up control of assembling the packet was a significant change in itself.

What it does not do

Colorado was specific about the failures, and these are the limits you should assume for any tool you evaluate:

• Copy-paste is not duplication. The platform removes identical records. It does not catch an office note whose body was copied into every subsequent visit note, which is the most common form of redundancy in a modern chart.
• Faxes and photocopies defeat it. The same document scanned twice at different quality is often not recognised as the same document.
• Relevance is out of reach. A defence representative made the same distinction to RAND: consolidating pages is solvable, deciding which of them matter to the injury is another whole step.

Why this is the right problem

Return to the money. On a California case with any record review at all, the median payment for reading the file is $939, which is about 513 pages. At the 95th percentile it is $10,000, about 3,533 pages. Record review is not a side task in this work; RAND found it is the principal driver of growth in the entire state's medical-legal spending since 2021.

And it is the one part of the job with unanimous agreement that some of it is waste. Not one participant in RAND's interviews, on any side, objected to removing duplicates before the packet reaches the evaluator.

What an AI-native evaluator does differently

This chapter is the working sequence. It assumes California, a panel case and a file of several hundred pages, and it maps onto any fee-schedule state with the names changed.

Before the exam

1. Log the packet the day it arrives. Page count, date range, source facilities, and the date you received it. You will need all 4 for billing and for the limits section of the report.
2. Order, de-duplicate, number, index. This is the Colorado sequence from chapter 8 and it is the same sequence whether a state, a management company or software does it. Record the page count before and after de-duplication: you bill on what you were sent, and you work from what is left.
3. Read the cover letter as a specification. List every question. That list becomes the report's headings, so an unanswered question is visible to you before it is visible to an attorney.
4. Build the chronology with page cites. Every row carries the page number it came from. A row you cannot open is a row you cannot testify to.
5. Write the missing-records list. Anything referenced in the file but not produced. Send it before the exam, not after. Chapter 20 has the letter.

At the exam

1. Confirm the body parts and the specialty in question against the cover letter, out loud, before you begin.
2. Take the history against the chronology, not against memory. Discrepancies between what the worker reports and what the file records are findings, and they belong in the report.
3. Note what you would need to answer any question you cannot answer today.

After the exam

1. Draft against the question list the same day, while the examination is fresh.
2. Expect a second packet. Records may legally arrive up to 10 days after the exam. Reconcile them against the chronology rather than re-reading the file.
3. Verify every cited fact at the page before you sign. This is the step that makes the deposition survivable and it is the step nothing automates.
4. Serve inside 30 days, with the limits section naming what you did not receive.
5. Bill the same week. Delay in payment was the most commonly cited reason evaluators gave RAND for leaving the system. Do not add to it from your own side.

What the arithmetic looks like

The fee schedule is fixed, so the only variable is your hours. Put your own numbers in. We have no published figure for how many hours any tool saves an evaluator, and this calculator does not invent one: the 2 hours fields are yours to set from your own files.

Run the default numbers and the point of the manual appears in 1 line: 48% of that revenue is paid for reading pages, at a rate of $3 each, on a clock you did not set. The evaluation fee is fixed. The reading is not.

This is a category list, not a product list. The test for any tool in this work is the same: does it put the file in order without deciding anything, and can you show your work afterwards.

The 6 things software should do

What to require before a file goes near a vendor

• A business associate agreement. Not a privacy policy, an executed BAA.
• US processing and storage, named, in writing.
• No training on your data, stated as a contractual term rather than a marketing sentence.
• Deletion on request, with a stated period.
• An audit trail, so you can answer the process questions from chapter 16.

The evaluation you should actually run

1. Pick 3 closed files of different sizes, including 1 over 2,000 pages.
2. Run each through the tool and record 4 numbers: pages in, pages out, minutes elapsed, errors you found.
3. Verify 20 random citations against the source pages.
4. Ask the vendor what happens on a scanned fax of a document already in the file. If the answer is confident and absolute, they have not tested it, because Colorado tested it and it fails.
5. Decide on your own numbers, not the vendor's.

Chapter 20 has that trial as a scorecard you can fill in.

The report is the product. The examination is 20 minutes in an orthopedic case and 3 hours in a psychiatric one; the report is what gets read, deposed, adopted or ignored. RAND's earlier work for the division identified what a high-quality medical-legal report contains, and nothing in the 2026 study contradicts it.

The 4 marks of a report that holds

1. A complete review of the prior history and the current issues. Not a summary of the records you happened to receive. A review that says what you had, what you did not have, and what you asked for.
2. A response to every medical-legal question put to you. The cover letter is the specification. An unanswered question is a supplemental report, a deposition, or both.
3. A focus on objective findings. What was measured, by whom, when, and on what page.
4. A well-reasoned permanent disability explanation with references. The rating is the part that decides money, and it is the part opposing counsel will attack line by line.

Impairment rating: stay inside the guides, use all of them

First, know which book you are in. 3 editions of the AMA Guides are in live use across the states, they produce materially different numbers, and the one that governs is fixed by the date of injury rather than by today's date. Chapter 12 has the measured difference and what to write in the report.

Almaraz and Guzman settled how far you may go. You may not go outside the 4 corners of the AMA guides. Inside them, you may use any chapter, table or method that most accurately reflects the impairment in front of you. Say which one you used and why, in the report, before someone asks you in a deposition.

Apportionment is where reports fail

Apportionment asks you to split impairment between the industrial injury and everything else: prior injuries, degenerative change, non-industrial conditions. It is the finding most likely to be the reason a case does not settle, and it is impossible to do honestly without the prior records. Which is the whole argument of chapter 6: an evaluator who never received the 2019 imaging cannot apportion to it, and a supplemental report months later is what that costs.

Write the limits into the report

When records are missing, or the question falls outside your specialty or clinical competence, the regulation requires you to say so in the report. Do it explicitly and in a fixed place, so that the reader and the judge can find it:

• The records you received, with a count and a date range.
• The records referenced in what you received but not provided to you.
• The questions you could not answer, and what you would need to answer them.
• Anything outside your specialty, which may mean a panel in another specialty.

This manual keeps saying "the AMA guides" as though there were one book. There is not. There are editions, states pick between them, and the edition decides how large a number you write down. In April 2026 the Texas Division of Workers' Compensation put a figure on that difference by having 5 faculty-level reviewers re-rate 363 real closed cases under a second edition.

Which book you are rating under

What happened when 363 cases were rated twice

The division drew 375 cases proportionally from the 17,480 impairment-rating claims filed in Texas in 2022, all originally rated 1% or more under the 4th edition, and gave them to 5 reviewers from its Medical Quality Review Panel, 3 physicians and 2 chiropractors, all designated doctor faculty. They re-rated each case under the 6th edition from the original certifying doctor's report.

The bigger the rating, the harder it falls

The averages hide the shape. Sort the same cases by what the 4th edition gave them and the pattern is monotonic almost the whole way up: small ratings mostly survive, large ratings mostly do not.

Texas DWC, 2026 AMA Guides Study. The one group that rose is not in this table: 24 spine cases rated 0% under the 4th edition, of which 12 came back at 1 to 4% under the 6th.

Why this is not an academic point

Ratings sit on top of benefit thresholds. In Texas an impairment rating of 15% or more is one of the conditions for supplemental income benefits. Of the 22 cases in the sample that reached 15% under the 4th edition, 10 fell below it under the 6th. That is not a rounding difference to the person being rated, and it is the reason opposing counsel will read your edition, your chapter and your table before they read your reasoning.

The number about you, not the book

All 5 reviewers rated the same 294 cases. Under the 4th edition those cases averaged 4.74%. Under the 6th, each reviewer's average landed somewhere between 3.06% and 3.7%.

Both halves of that sentence matter. The edition moved the average by about a third; 5 qualified, faculty-level examiners working from identical files moved it by roughly six tenths of a point between them. The book is the larger lever. You are the smaller one, and you are the one a deposition can question.

What to do with this

1. Name the edition in the report. Edition, chapter, table, page. Not "the AMA Guides".
2. Check the edition against the date of injury, not the date you were assigned. Older claims can sit in an older book.
3. Never port a rating across editions. A 4th or 5th edition percentage is not a 6th edition percentage rewritten. If you are asked to convert one, say that you would have to re-rate.
4. Say why you chose the method you chose. In California, Almaraz and Guzman lets you use any part of the 5th edition that most accurately reflects the impairment, provided you stay inside it and explain yourself. Chapter 11 covers that; chapter 16 covers being asked about it.
5. Keep your own distribution. If you rate 40 cases a year, you can plot them. An examiner who cannot say roughly where their own ratings sit cannot answer the question when it is asked under oath.

Until 2024, a California evaluator could write reports for a decade and never receive a structured comment on one. RAND found this to be the most consistent complaint from active evaluators: 8 active and 1 former evaluator said they did not get enough feedback. They do not know how the case resolved, or whether what they wrote helped anyone.

The 3 ways you currently find out you were wrong

1. You get deposed. Expensive, adversarial, and late.
2. You are asked for a supplemental report. A signal that something was missing, delivered as more work.
3. You stop being selected. The quietest and the most damaging, because nothing tells you it is happening.

Evaluators told RAND they would prefer the same information earlier and in a lower-stakes setting. One applicants' attorney made the sharper version of the point: evaluators are often not failing to address mandatory legal questions on purpose, they simply were never taught them, and only learn when an attorney pushes.

What is changing

From 2024, 8 CCR 50 requires you to submit your 2 most recent reports through the online portal at reappointment. That is the first routine, non-adversarial read of a California evaluator's work. 2 other states are further ahead and are worth copying voluntarily.

Both are worth stealing as private practice. You can calculate your own version of the Texas factors from your own files: how many of your reports drew a letter of clarification, how many produced a supplemental request, how many led to a deposition. Nobody is going to send you that report card, so write it yourself every 6 months.

Mentorship, which the system does not provide

3 former evaluators, an active evaluator and a management company representative told RAND that new evaluators need mentors, and that no formal programme exists. RAND's recommendation is to pair new and experienced evaluators voluntarily. One former evaluator's wish was smaller and more revealing: somewhere to compare notes, down to what a billing agent should charge.

Being on a panel is not the same as getting the case. In a represented matter each attorney strikes one of the 3 evaluators, and RAND's replacement data shows exactly why evaluators come off panels.

What each of these means for you

2 structural notes. The strike rate has been essentially flat at 44 to 47 percent since 2012, so it is a property of the system rather than a signal about any year. And replacements are increasingly happening late: by 2024, 23.5% of replacement panels were requested after medical-legal money had already been spent on the case, and nearly a third of that spending was more than a year old.

The scheduling loss is the one to fix

One in 5 replacements happens because nobody could get an appointment inside the window. The window is now 90 to 120 days from the original request. RAND recommended letting evaluators cap the number of panels they accept per month, precisely so that in-demand specialists stop accumulating assignments they cannot schedule. That regulation does not exist yet. Until it does, the only lever is the number of offices you hold and the unavailability days you declare.

Getting paid is a separate discipline from doing the work. RAND asked evaluators why they left the system and the most commonly cited reason was not the fee schedule and not the exams. It was delay in payment.

What you bill in California

Every medical-legal service is billed under an ML code plus a modifier plus, past the included pages, a per-page record review charge. There is no negotiation and no cost estimate. You bill what the schedule says.

8 CCR section 9795, effective 2021-04-01. Relative values times the $16.25 conversion factor.

The 3 billing mistakes that cost you

1. Not counting the pages you were sent. The per-page charge is on what the parties served on you, before you remove duplicates. Count on arrival and record the number. Reconstructing a page count 3 months later after a bill is disputed is a losing exercise.
2. Missing the modifier. The multiplier is on the evaluation, not on the record review. A psychiatric comprehensive evaluation is $4,030, not $2,015. An AME comprehensive is $2,720. Getting this wrong once a month is real money.
3. Billing late. Nothing in the schedule pays you for your own delay, and every week you wait is a week the file gets colder on the other side.

What the second packet does to your bill

Records may arrive up to 10 days after the exam. Those pages are billable too, at the same $3.00, and they are the pages most often left off a bill because they came in after the evaluator mentally closed the file. Bill from the log, not from memory. Chapter 20 has the log.

If you are outside California

Texas sets a maximum allowable reimbursement per designated doctor service rather than a page rate. In the private IME market there is no schedule at all: SEAK's 2024 survey of 325+ examiners across 54 specialties put the average IME fee at $2,890 and the average hourly rate at $441, with a $955 average no-show fee. Read chapter 3 for the full comparison.

A deposition is where the report gets tested. It is also, at $113.75 a quarter hour with a 2 hour minimum, the highest hourly rate in the schedule: $455.00 an hour, or $910.00 in psychiatry.

What gets attacked

• Facts you did not verify. Anything in the report that you took from a summary rather than from the page. This is the single largest exposure created by working faster.
• Records you never saw. If the file was 4,000 pages and you cannot say what you reviewed, the limits section of your report is the only thing standing between you and an unpleasant hour.
• The apportionment reasoning. Percentages without a stated basis do not survive.
• Your method under the AMA guides. Which chapter, which table, why.
• Your process. Increasingly: what software touched this file, and what did you check yourself.

The 5 answers to have ready

1. What you received. Page count, date range, source, date received. From your log.
2. What you reviewed. All of it, or a defined subset with a stated basis. Say which.
3. What you did not receive. The missing-records list, dated, and whether you asked for it.
4. How you found what you cited. Bates number, every time.
5. What you verified personally. Every fact in the report, at the page. That is the only correct answer, and it is only available to you if it is true.

The scheduling rule people forget

A California QME may declare up to 120 days of unavailability a year. Depositions are scheduled around your calendar, not the other way round, and a pattern of unavailability is one of the reasons parties strike an evaluator. Chapter 14 has the numbers.

Medical management companies, sometimes called QME management or IME management companies, sit between the evaluator and the system. RAND's interviews and focus groups produced a consistent picture of what they do, and a consistent set of complaints.

What they do for you

• Schedule the appointments
• Receive and organise the records
• Track the deadlines
• Handle the billing and chase the payment
• Provide office space in some arrangements
• Take a share of your fee

What they cannot do

They may not write the report. The opinion, the causation analysis, the apportionment and the signature are yours, and the entire evidentiary value of the document rests on that being true.

The governance gap

The Division of Workers' Compensation has no jurisdiction over these companies. They are not licensed by the division, not audited by it, and not subject to its discipline. The evaluator is. If the company misses a deadline, the report is late and it is your report that is late.

The mentorship reality

RAND's focus groups identified management companies as a main source of practical training for new evaluators, because no formal mentorship structure exists. That is a real service and also a real dependency: your first understanding of how the work is done comes from a party with a commercial interest in how much of it you do.

What to ask before you sign

1. What share of the fee, on which codes. Including the per-page record review, which is where the volume is.
2. Who counts the pages, and do I see the count. Your bill depends on it.
3. What happens to the deadline clock if you are late getting me the records. Get the answer in writing.
4. Who drafts. The only acceptable answer is that you do.
5. What happens if I leave. Who keeps the calendar, the records and the relationships.
6. Is there an employment or exclusivity clause. An employment contract can interact badly with the QME appointment rules in chapter 4.

The manual has spent 16 chapters on state systems because that is where the public data lives. But the highest fees in this work are in the private market, and everything that makes a state system tolerable is missing there: nobody assigns you a case, nobody publishes your rate, and nobody sends you a file.

The one structural difference

In California the state generates a panel and your name appears on it. In Colorado and Texas the division assigns you. In the private market the case arrives because someone chose to call you, and the reason they chose you is a document they read before they called.

Who actually buys

The last column is the useful one. In panel work you are selected at random and struck for reasons you never learn. Here you are selected on stated criteria, which means you can meet them.

What a buyer reads before they call

1. A CV built for this work. Board certification, clinical volume, the evaluation training you hold, and your medical-legal experience as a separate section rather than buried in the practice history.
2. A testimony list. Case, court or forum, retaining party, year, deposition or trial. Some jurisdictions require you to produce it; every serious buyer wants it; and you will be asked for it under oath eventually. Start it on your first case, not on your fiftieth.
3. A redacted sample report. One complete report with every identifier removed, showing your structure: the questions, the records reviewed, the limits section from chapter 11, the reasoning, the citations.
4. A rate card. Written down, dated, and the same for both sides.
5. A stated turnaround. The number you will actually hit on a file of ordinary size, not your best case.

The rate card

Price every line separately. A single all-in number invites an argument about what it covered, and it hides the record hours that chapter 3 showed are the real cost.

SEAK 2024 Survey of IME Fees and Billing Practices, 325+ examiners across 54 specialties. Lines marked "no published rate" are structure, not benchmarks: we have no source for them and are not inventing one.

SEAK's average across the market was $2,890 for an IME taking 6.5 hours end to end, which is $441 an hour. That last figure is arithmetic, not a quoted rate, and it is the one your own file should be able to beat or explain.

3 things that end a private practice

• Turnaround drift. A carrier with a claim reserve does not care why the report is late. This is the most common reason a referral source quietly stops calling, and like the strike in chapter 14 it arrives with no notice.
• Addenda. Every question you did not answer becomes a second invoice for the payer and a second read for the adjuster. Chapter 11 exists to prevent this.
• A visible lean. Both sides will eventually ask, under oath, what share of your medical-legal work comes from defence and what share from claimants, and what share of your total income this work represents. Track both numbers from your own billing so the answer is a figure rather than an estimate. Chapter 20 has the log.

Conflicts, before you accept

Run the check before the file arrives, not after you have read it. Have you treated this person? Examined them before, for anyone? Do you have a relationship with the employer, the carrier or either firm? Have you already given an opinion in this matter? A conflict found on page 400 costs you the fee and the referral source.

2 paths run through this chapter. Follow the one that matches where you are.

If you are not yet appointed in California

If you are already appointed

The checklist

• Receivables table built for the last 12 months
• Page count logged on arrival for every file
• Modifier applied correctly on every evaluation code
• Order, de-duplicate, number, index before any reading
• Chronology rows all carry page cites
• Missing-records list sent before the exam
• Limits section in the report template
• Every cited fact verified at the page before signing
• Reports served inside 30 days
• Bills sent the same week
• Tool tested on 3 of your own closed files
• BAA executed before any file was uploaded
• Standing answers written for the 5 deposition questions
• Last 10 reports self-scored against the 4 quality marks
• Guides edition confirmed for every state you rate in
• Templates copied and adapted to your own practice
• Rate card written down before the first private enquiry

Download this guide as a worksheet (CSV), with Done and Notes columns.

0 of 17 ticked. Nothing is stored.

Every chapter above tells you to keep something. This chapter is the something. 6 blocks, plain text, meant to be copied into whatever you already write in. Nothing here is stored or sent.

RECORD INTAKE LOG - one row per packet, not per case

Case / claim number
Date packet received
Delivery route (portal, courier, email, disc)
Party who served it
Pages served (count before anything is removed)   <- you bill on this
Pages after de-duplication                        <- you work from this
Date range of the records, earliest to latest
Source facilities and providers listed
Bates range applied
Cover letter received?  Y / N   Date
Questions asked (count)
Records referenced in the file but not produced (count)
Date missing-records request sent
Second packet received?  Y / N   Date   Pages
Exam date
Report served (date)      Days from exam
Bill sent (date)          Amount        Code(s) + modifier
Paid (date)               Days outstanding

MISSING RECORDS REQUEST - send before the examination

To:      [both counsel / the claims administrator, as the rules of your
         jurisdiction require. Serve all parties. Do not contact one side alone.]
Re:      [claimant name] / [claim number] / [date of injury]
Date:    [date]

I have received [N] pages of records in this matter, served on [date],
covering [earliest date] to [latest date].

The following records are referenced within the material provided but were
not themselves produced to me:

  1. [document, provider, date] - referenced at Bates [number]
  2. [document, provider, date] - referenced at Bates [number]
  3. [document, provider, date] - referenced at Bates [number]

These records bear on [causation / apportionment / the impairment rating /
the specific question], and I am requesting them so that my opinion accounts
for them.

If they are not provided, I will proceed on the material I have and will
state in my report which records were requested, on what date, and that
they were not received.

[name], [credentials]
[evaluator number, where one applies]

LIMITS OF THIS REVIEW - a fixed section in every report

RECORDS RECEIVED
  [N] pages, served [date] by [party], covering [date] to [date].
  Sources: [facility, facility, facility].
  Bates [range].
  [If a second packet arrived: a further N pages served on DATE, after the
   examination of DATE, reviewed and incorporated.]

RECORDS REQUESTED AND NOT RECEIVED
  [document, provider, date] - requested [date], not received.
  [document, provider, date] - requested [date], not received.
  [Or: None. All records referenced in the material provided were produced.]

QUESTIONS I AM NOT ABLE TO ANSWER
  [question as put in the cover letter]
  Reason: [the record needed / outside my specialty / requires testing not
  performed].
  What would be required: [the specific record, study or consultation].

MATTERS OUTSIDE MY SPECIALTY
  [issue], which falls within [specialty]. I express no opinion on it.

HOW THIS REPORT WAS PREPARED
  The records were placed in date order, de-duplicated, Bates numbered and
  indexed before review [by me / by NAME / using SOFTWARE]. Every fact cited
  in this report was verified by me against the source page identified in
  the citation. The opinions, the causation analysis and the apportionment
  are my own.

STANDING DEPOSITION ANSWERS - write yours once, then use them

Q. What records did you receive?
A. "I received [N] pages, served on [date] by [party], covering [date range],
   Bates [range]. I logged that on receipt and I have the log with me."

Q. Did you review all of them?
A. "Yes." - or, if not: "I reviewed [defined subset] because [stated basis].
   I said so at page [X] of my report." Never "I reviewed the relevant
   portions" without saying which and why.

Q. What did you not receive?
A. "The records listed at page [X] of my report. I requested them on [date]
   from [both parties]. They were not produced. I noted that in the report
   before I signed it."

Q. Where does that fact appear in the record?
A. "Bates [number]." Every time. If you cannot give a number, you cannot
   give the fact.

Q. Did you use software on this file, and what did you verify yourself?
A. "[Product] placed the records in date order, removed duplicates, applied
   Bates numbers and produced an index and a chronology. It is covered by a
   business associate agreement. It did not form any opinion in this case.
   I read the source pages myself and I verified every fact I cite in the
   report against the page cited. The opinions, the impairment rating and
   the apportionment are mine."

Q. What is your split between defence and claimant work?
A. "[X]% and [Y]% over the last [period], from my own billing records."

Q. What share of your income is medical-legal work?
A. "[Approximately Z]%." Know it before you are asked.

COVER LETTER QUESTION TRACKER - the report's table of contents

Copy every question from the cover letter, verbatim, into this table before
you read a single page. Each one becomes a heading in the report. An empty
"answered at" cell is a supplemental report you have not written yet.

  #  Question, as written by the party        Answered at   Basis / limit
  1  [verbatim]                               p.            
  2  [verbatim]                               p.            
  3  [verbatim]                               p.            
  4  [verbatim]                               p.            
  5  [verbatim]                               p.            

Standard California findings, if the cover letter does not list them:
  Causation to a reasonable medical probability
  Temporary disability
  Return to work and restrictions
  Impairment rating, with edition, chapter and table named
  Apportionment, with the basis stated
  Permanent and stationary date
  Future medical care

TOOL TRIAL SCORECARD - 3 of your own closed files, one over 2,000 pages

For each file, record only these numbers. Do not accept the vendor's.

  File                                  A          B          C
  Pages in (as served)
  Pages out (after de-duplication)
  Duplicates removed
  Minutes elapsed, upload to output
  Citations checked (aim for 20)
  Citations that opened the right page
  Errors you found, described
  Records referenced but not produced,
    flagged by the tool
  Records referenced but not produced,
    that you found and it missed
  Handwritten pages read correctly
  Your own hours on this file, before
  Your own hours on this file, after

Two questions for the vendor, asked after the trial, not before:
  1. What happens to a scanned fax of a document already in the file?
     A confident absolute answer means it has not been tested.
  2. Is there a signed business associate agreement, is processing in the
     United States, and is our data contractually excluded from training?

Everything above this chapter stands on public sources you can check. This chapter is the publisher's, and you should read it as an advertisement. Medrecords AI makes software for the record column in chapter 8: it sorts, deduplicates, reads handwriting, and returns a chronology in which every line cites its source page, under a signed business associate agreement, priced per deduplicated page. It does not form opinions, apportion, or choose a method under the AMA guides. Chapter 8 explains why that line exists.

What recent users say

Professionals who moved live case files onto the platform in August and September 2026, quoted verbatim from their email and attributed by role. One of them reported a defect in the same thread, noted under the quote, because that is what a real test looks like.

Questions readers ask

Is there a real study of IME and QME physicians, or is this a marketing document?

There is a real study and this manual is not it. In May 2026 the RAND Corporation published "Qualified Medical Evaluators and the Medical-Legal Process in California Workers Compensation" (RR-A4655-1), commissioned by the California Department of Industrial Relations. RAND analysed 1,752,557 panel assignments from 2012 to 2024 covering 5,034 evaluators, reviewed the rules in 17 states, ran 27 stakeholder interviews and held 6 focus groups with 11 active and 10 former QMEs. Every RAND number in this manual is cited to it. We did not survey anyone for this document and we do not pretend we did.

What does an IME or QME physician actually earn?

In California it is set by regulation, not negotiated. A comprehensive medical-legal evaluation pays $2,015, a follow-up $1,316.25, a supplemental report $650, and depositions $113.75 a quarter hour with a 2 hour minimum, times a specialty multiplier from 1.1 to 2.45, plus $3.00 a page for records past the first 200. In the private IME market SEAK 2024 survey of 325+ examiners across 54 specialties reported an average fee of $2,890 and an average hourly rate of $441. Chapter 3 has the full tables.

How long does it take to become a QME in California?

Realistically 3 to 6 months from decision to appointment: apply to the Division of Workers Compensation Medical Unit, pass a 3 hour competency exam, and complete a 25 hour course, which has been required before appointment since February 2024. Reappointment needs 12 hours of QME continuing education every 24 months and 2 sample reports. Chapter 4 has the sequence.

Where does AI actually help, and where is it dangerous?

It helps with the record packet: ordering pages by date, removing duplicates, Bates numbering and indexing. Colorado division of workers compensation runs exactly that pipeline in production on DIME cases, with a human verifying the output, and told RAND that relevance was deliberately left out because it is a different order of difficulty. It is dangerous anywhere it touches judgment: causation, apportionment, choice of AMA guides method, or any fact you cite without opening the page. Chapter 8 has the line and where Colorado found it fails.

How much of the job is reading records?

More than most people outside the work assume. Across 198,774 California cases with a record review charge, the median payment for reading the file was $939 and the mean $2,533, with the 95th percentile near $10,000. At $3.00 a page that median is roughly 513 pages and the 95th percentile roughly 3,533. RAND identified record review as the principal driver of growth in California medical-legal spending since the 2021 fee schedule.

Why do evaluators get replaced or struck?

In 2024 the reasons for replacing a California panel evaluator were: struck by a party 46%, no appointment inside the required window 19%, evaluator unavailable 7%, judge ordered 5%, Romero exception 4%, evaluator now inactive 4%, other 3%, wrong specialty 2%, report filed late 1%. 2 of the top 3 are scheduling, which is inside your control. Chapter 14 breaks down what you can change.

Which edition of the AMA Guides should I be rating under?

Whichever one your jurisdiction adopted, fixed by the date of injury rather than by today's date. California uses the 5th edition under Labor Code 4660.1 for injuries on or after 1 January 2005 and has not adopted the 6th. Texas uses the 4th. Editions are not interchangeable: when the Texas Division of Workers Compensation had 5 faculty reviewers re-rate 363 closed 2022 cases under the 6th edition in 2026, the average impairment rating fell from 5.4% to 3.8%, a 30% drop, with 59% of cases rated lower and 30% higher. Name the edition, chapter and table in the report, and never convert a rating from one edition to another. Chapter 12 has the full breakdown.

How do I get private IME work, as opposed to state panel work?

Nobody assigns it to you, which is the whole difference. Carriers, third-party administrators, disability insurers, defence and plaintiff firms, IME brokers and self-insured employers all buy it, and they screen on documents you can prepare in advance: a CV with medical-legal experience as its own section, a testimony list, one redacted sample report, a written rate card and a turnaround time you actually hit. SEAK 2024 is the only published fee reference we found, at an average of $2,890 an IME and a $955 no-show fee. Chapter 18 sets out the buyers and the rate card structure, and marks the lines where no published rate exists.

Does this apply outside California?

The structure does, the numbers do not. RAND reviewed 17 states: some use a state-assigned panel like California, some use a designated doctor list like Texas, some let the parties agree. Texas publishes maximum allowable reimbursements per service rather than a page rate. Colorado assembles the record packet centrally. Chapter 5 maps the differences.

Terms used in this manual. Clinical terms are the reader's own.

QME

Qualified Medical Evaluator. A physician certified by the California Division of Workers Compensation to perform medical-legal evaluations, assigned to a case from a randomly generated 3-name panel.

AME

Agreed Medical Evaluator. A physician the parties agree on instead of using a panel. Fees carry a 1.35 multiplier in California.

DIME

Division Independent Medical Examination, the Colorado equivalent of a state-assigned evaluation.

Designated doctor

The Texas equivalent, assigned by the Division of Workers Compensation and paid per service under a published maximum allowable reimbursement.

IME

Independent medical examination. Used loosely for any evaluation by a physician who is not treating the patient, and specifically for privately retained examinations outside a state panel system.

ML code

The billing code for a medical-legal service in California, set by 8 CCR section 9795. ML201 is a comprehensive evaluation, ML202 a follow-up, ML203 a supplemental report, ML204 a deposition, ML205 a missed appointment.

ML-PRR

The per-page record review charge, $3.00 a page past the first 200 on ML201 and ML202 and past the first 50 on ML203.

Panel

A list of 3 QMEs of a requested specialty generated by the state. Each party may strike 1; the remaining evaluator takes the case.

Strike

A party removing 1 name from a panel without stating a reason. The most common reason a California evaluator loses a case.

MMI

Maximum medical improvement. The point at which a condition has stabilised, which triggers the impairment rating.

Impairment rating

A percentage of whole person impairment calculated under the AMA Guides. Distinct from disability, which is a legal determination.

Apportionment

Dividing the cause of a disability between the industrial injury and other factors. In California it must be to causation and it must be explained.

Almaraz/Guzman

California case law permitting an evaluator to depart from a strict AMA Guides rating when the strict rating does not accurately reflect impairment, provided the departure stays within the 4 corners of the Guides and is explained.

Romero

A California exception allowing a replacement panel in defined circumstances. Accounted for 4% of 2024 panel replacements.

Bates number

A unique stamp on every page of a produced record. A cited fact without one cannot be checked at deposition.

Supplemental report

A further report answering questions raised after the original evaluation, billed under ML203.

This is a field manual, not a study. We surveyed nobody. Every number in it comes from a public source, named at the point it is used, and the sources are these 6.

What we did not do

• We did not interview evaluators for this document. RAND did, and we cite RAND.
• We did not estimate any figure that a source did not state.
• We did not measure how many hours any software saves an evaluator. The calculator in chapter 9 asks you for that number because we do not have it.
• We did not paraphrase a RAND finding into something stronger than RAND wrote.
• We did not carry the Texas edition comparison across to California. That study compared the 4th edition to the 6th; California rates under the 5th, and we say so in chapter 12 rather than doing the arithmetic nobody has published.
• We did not benchmark any private IME rate the SEAK survey does not cover. Chapter 18 marks those lines "no published rate" instead of filling them in.

RAND Corporation, the Texas Department of Insurance, the California Department of Industrial Relations, SEAK and the California Workers Compensation Institute are not affiliated with Medrecords AI, have not reviewed this document, and do not endorse it. Their work is public and you should read it. Chapter 21 is the only part of this manual we are selling anything in, and the notes marked "from the publisher" elsewhere are advertisements; they are labelled so you can skip them.