AI Medical Coding & Michigan No-Fault PIP Utilization Review (MCL 500.3157 & 500.3157a)
Michigan no-fault PIP is now governed by the fee-schedule and utilization-review framework in MCL 500.3157 and MCL 500.3157a. Generative AI now assists with CPT/ICD coding on both the provider and carrier sides. When the AI gets it wrong, the consequences land on the claimant. This guide explains how AI enters the PIP fight and what to do about it.
Where the Fee Schedule Meets AI
Since 2019 PA 21 and 2019 PA 22, Michigan no-fault provider reimbursement has been governed by the fee schedule in MCL 500.3157. For services with a Medicare-payable CPT code, payment is capped at a declining percentage of the Medicare amount (declining from 200% to 190% over the phase-in (per the DIFS-published fee-schedule phase-in)). For services not on the Medicare schedule, providers must use their January 1, 2019 charge-description master, multiplied by a declining DIFS-published percentage (55% in 2021-22, 54% in 2022-23, and 52.5% since July 2023, with annual CPI adjustments thereafter). Code selection is therefore not academic — it directly determines the dollar amount payable.
Generative-AI tools now assist Michigan providers and insurers in selecting the right ICD-10 diagnosis codes and CPT procedure codes from clinical notes. Vendors market AI scribes, automated coding engines, and revenue-cycle platforms that read the EHR, propose codes, and feed them into billing. These tools are also showing up on the carrier side of the no-fault dispute — powering the utilization-review process under MCL 500.3157a.
Utilization Review — MCL 500.3157a
MCL 500.3157a defines “utilization review” as “the initial evaluation by an insurer or the association created under section 3104 of the appropriateness in terms of both the level and quality of treatment, products, services, or accommodations provided under this chapter based on medically accepted standards.” Insurers and the MCCA may use the UR process to:
- Request additional information or records from the provider.
- Deny or reduce charges that, based on UR, are not “medically accepted” or are at a quality or level inconsistent with the standard of care.
- Refer disputes to the Michigan Department of Insurance and Financial Services for the appeal track set out in the statute and the implementing rules.
The UR process is procedurally narrow, but it is exactly the layer where AI-generated coding flags get embedded into denial letters. Plaintiff-side counsel for the provider or the injured claimant should understand what the AI is doing and what records it touched.
How AI Coding Tools Work in PIP Practice
| Workflow stage | What the AI does | What the human MUST verify |
|---|---|---|
| Initial chart capture | Reads dictation / EHR notes and proposes ICD-10 diagnoses | Whether the captured note actually reflects the visit and whether the proposed code matches the documented condition |
| CPT procedure code selection | Suggests the procedure code from documented services | Whether the documented level of service supports the proposed CPT level (E/M leveling especially) |
| Fee-schedule cross-reference | Maps the CPT to the Medicare amount and applies the MCL 500.3157 multiplier | The multiplier is statute-driven; verify against the current DIFS-published phase-in |
| UR flag generation (carrier side) | Flags charges that exceed historical norms, repeat-treatment patterns, or procedure-bundling rules | Whether the flag reflects medical reality or algorithmic bias; medically accepted standard, not algorithm output, controls under MCL 500.3157a |
| Denial letter drafting | Drafts the denial language and the statutory citation | Whether the letter accurately reflects the basis for denial and complies with MCL 500.3142 and 500.3145 |
Discovery Strategy When AI Is in the Loop
Michigan plaintiffs and providers fighting a UR denial should treat the AI-coding pipeline as a discoverable workflow:
- Request the AI vendor name and version. Identify whether the carrier or provider is using an internal model or a third-party platform.
- Request the AI training-data scope to the extent producible without trade-secret destruction. The model’s training corpus shapes its bias toward or against certain claim profiles.
- Request the audit log of human-in-the-loop overrides. A coder or UR nurse who routinely accepts AI suggestions without independent review is a discovery target.
- Request the metadata showing when the code was assigned and by whom. EHR audit trails are typically available under HIPAA and standard practice.
- Preserve everything per Brenner v Kolk, 226 Mich App 149 (1997). Once a UR dispute is reasonably anticipated, the duty to preserve attaches.
When AI Coding Errors Cause Real Harm
An AI-generated CPT or ICD error can result in:
- Underpayment of the provider, which can cascade into the provider dropping the patient or assigning the claim to a collections agency.
- Overbilling that triggers MCL 500.3148(2) — the fee-shifting provision against a claimant or provider whose claim was in some material respect fraudulent or excessive without reasonable foundation. See Moore v Secura Insurance, 482 Mich 507 (2008).
- UR-driven benefits cutoff, with the claimant losing access to treatment.
- Spoliation exposure when the audit trail of AI suggestions is not preserved.
Deadlines — The Same Clocks Still Run
- PIP claim: One year from each medical expense under MCL 500.3145.
- Provider direct-action standing: Limited post-Covenant Med Ctr Inc v State Farm, 500 Mich 191 (2017); providers usually proceed via assignment.
- Attorney-fee shifting: MCL 500.3148(1) on unreasonable delay or refusal; MCL 500.3148(2) on fraudulent or excessive claims.
- Three-year tort SOL: MCL 600.5805(2) for any third-party bodily-injury claim arising from the same accident.
Frequently Asked Questions
Can a carrier deny PIP based on an AI flag alone?
Not legally. MCL 500.3157a requires the UR determination to be based on medically accepted standards. An AI suggestion is evidence at most; the human reviewer’s independent judgment under medically accepted standards is what the statute requires.
Is the AI-coding workflow discoverable?
Generally yes under MCR 2.302(B)(1), subject to proportionality and trade-secret protection. Vendor identity, model version, prompt templates, audit logs, and human-in-the-loop override records are typical targets.
Does HIPAA limit what I can get from the EHR audit trail?
The audit trail of the patient’s own record is generally producible to the patient or their authorized representative under HIPAA’s right-of-access provisions. Production in litigation runs through ordinary subpoena/court order practice, subject to protective orders.
What if the provider used an AI scribe and the note misstates what was said?
The note — AI-generated or not — is the documentation that supports or fails to support the billed code. A misstated note is an evidentiary problem that opens the provider to MCL 500.3148(2) fraud-or-excessive-claim exposure and creates a malpractice angle against the provider for documentation negligence.
Can my lawyer use AI to review the carrier’s denial letters?
Yes, subject to MRPC 1.6 confidentiality (no disclosure of client information to a tool that retains training rights), MRPC 5.3 supervision, and MRPC 1.1 competence. Lawyers remain professionally responsible for the output regardless of who or what generated it.
Does the fee-shifting provision under MCL 500.3148 cover AI-driven denial mistakes?
It can. An unreasonable refusal or delay under MCL 500.3148(1) does not require carrier bad faith — it requires unreasonableness. A carrier whose denial rests on an obvious AI error and whose human reviewer rubber-stamped the output can face fee-shifting exposure under Moore v Secura.
Michigan PIP Claim Denied or Reduced?
Free, confidential review with Attorney Manny Chahal. No fee unless we recover.
Call 1-844-624-2425