Actuary-grade data trust for mental health parity
The 2024 MHPAEA final rule requires health plans to prove parity through data, not assertions. Most plans lack the data quality infrastructure to pass an actuary-grade comparative analysis. Here is what parity compliance data trust actually requires across network adequacy, NQTL design, and claims adjudication.
The 2024 MHPAEA final rule turned mental health parity from a design standard into a data standard. Health plans can no longer assert that their nonquantitative treatment limitations apply comparably to medical/surgical and behavioral health benefits. They must prove it with outcomes data. That proof requires data infrastructure most plans do not have.
The gap is not regulatory awareness. Compliance teams know what the rule says. The gap is data quality. When a plan cannot trust its own claims data, provider directories, or utilization records to the level an actuary would certify, every parity analysis built on that data is suspect.
What the 2024 final rule changed about data requirements
The Mental Health Parity and Addiction Equity Act of 2008 established that financial requirements and treatment limitations for mental health and substance use disorder benefits cannot be more restrictive than those for medical/surgical benefits. For years, plans demonstrated compliance through plan design documentation.
The 2024 final rule from CMS, DOL, and HHS (2024) changed that. Plans must now conduct comparative analyses for every NQTL applied to MH/SUD benefits and collect outcomes data demonstrating that NQTLs do not restrict access to behavioral health services more than comparable medical/surgical services. The rule explicitly requires that plans examine "the relevant data evaluated" and "any other data relevant to the factors" in their NQTL designs.
This is not a paperwork exercise. It is a data provenance exercise. An NQTL comparative analysis that cites data of unknown origin, unknown recency, or unknown completeness will not survive a DOL audit.
The five data domains where parity compliance breaks
| | Value (%) | |---|---| | BH reimbursement gap | 20 | | BH out-of-network use (vs med/surg) | 200 | | Provider directory error rate | 50 |
Parity analysis touches at least five distinct data domains. Each has its own trust problems.
Claims adjudication data
Comparative analysis requires apples-to-apples comparison of denial rates, prior authorization approval rates, and reimbursement levels between MH/SUD and med/surg claims. According to CMS analysis of MHPAEA enforcement actions (2023), the most common deficiency in plan submissions was incomplete or inconsistent claims data that prevented meaningful comparison.
The problem starts with coding. ICD-10 codes for behavioral health conditions are notoriously inconsistent across providers. A study published in Psychiatric Services (2019) found that diagnostic coding agreement between behavioral health providers was below 70% for common conditions like adjustment disorders and anxiety spectrum diagnoses. When the denominator of a parity analysis is built on codes that providers apply inconsistently, the ratio is unreliable before the analysis begins.
Network adequacy data
The 2024 rule requires plans to demonstrate that their provider networks for MH/SUD are comparable in access to their med/surg networks. This means comparing metrics like appointment wait times, geographic distance, and provider-to-member ratios.
Provider directory accuracy is the foundational problem. The CMS Medicare Advantage and Part D final rule (2023) found that roughly half of provider directory entries contained at least one inaccuracy. For behavioral health providers, the error rate is likely higher because they change practice locations, accept or drop insurance panels, and adjust availability more frequently than most medical/surgical specialists. A network adequacy comparison built on directory data with a 50% error rate is not a comparison. It is noise.
Prior authorization and utilization management data
NQTL analysis must compare how prior authorization is applied across MH/SUD and med/surg benefits. The Department of Labor's 2023 MHPAEA report to Congress (2023) found that many plans could not produce consistent prior authorization data segmented by benefit classification. The data existed in different systems, used different category definitions, and applied different time windows.
When a plan uses one vendor for med/surg utilization management and another for behavioral health UM, the data structures rarely align. Prior auth denial reasons are coded differently. Appeal timelines are tracked differently. The result is two datasets that cannot be compared without significant reconciliation, and most plans have not invested in that reconciliation.
Reimbursement rate data
Plans must demonstrate that reimbursement rates for MH/SUD providers are comparable to rates for med/surg providers of similar training and licensure. A report from the Bowman Family Foundation and Milliman (2019) found that behavioral health providers were reimbursed at rates 20% lower than comparable medical providers, with out-of-network utilization for behavioral health services running at rates two to five times higher than for medical services.
The data trust problem here is not just the rate itself but the comparability framework. What constitutes a "comparable" med/surg provider for a licensed clinical social worker? The rule requires plans to define and defend that comparison using data, but most plans lack a structured crosswalk between behavioral health and med/surg provider taxonomies.
Outcomes data
The 2024 rule's most significant addition is the outcomes data requirement. Plans must collect and evaluate data on the impact of NQTLs on access to MH/SUD benefits, including metrics like utilization rates, out-of-network usage, and claims denials, and compare those outcomes to med/surg equivalents.
This is where the data trust requirement reaches its peak. Outcomes data must be longitudinal, consistently defined, and attributable to specific NQTLs. According to the American Psychiatric Association's analysis of the final rule (2024), most plans have never built the data infrastructure to track outcomes at the NQTL level. They track aggregate utilization. They do not track how a specific prior authorization requirement for intensive outpatient program admission affects downstream emergency department utilization for the same population.
Why actuarial standards demand higher data trust than compliance minimums
A parity comparative analysis is, at its core, an actuarial exercise. The Actuarial Standards Board's ASOP No. 23 (2016, revised) sets the professional standard for data quality in actuarial work. It requires actuaries to assess whether data is "appropriate, reasonable, and sufficient" for the intended analysis and to disclose any limitations.
ASOP No. 23 is not optional guidance. It is a binding standard for any actuary who signs a parity analysis. If the underlying data has unknown provenance, inconsistent definitions, or unquantified error rates, the actuary cannot certify the analysis without substantial caveats. Those caveats, in turn, undermine the plan's compliance position.
The practical effect: parity compliance data must meet a higher trust threshold than most operational health plan data. A claims file that is "good enough" for payment processing may fail actuary-grade scrutiny because payment processing tolerates a level of coding inconsistency that a comparative ratio analysis does not.
The consent and 42 CFR Part 2 collision
Substance use disorder data carries special federal protections under 42 CFR Part 2 (2024, amended). These protections restrict the redisclosure of SUD treatment records even within a health plan's own analytics environment.
This creates a direct conflict with parity analysis requirements. A plan must compare SUD treatment access and utilization to med/surg equivalents, but it may lack consent to use individual SUD records for that comparison. The SAMHSA final rule aligning Part 2 with HIPAA (2024) narrowed some gaps, but plans still face scenarios where their SUD data is either incomplete because patients did not consent to data sharing, or siloed in systems that cannot feed a comparative analysis pipeline.
Without a consent architecture that tracks patient-level permissions across use cases, plans cannot build a defensible SUD parity dataset. The consent dimension alone can reduce the usable denominator for SUD parity analysis by 30% or more, depending on the plan's consent capture processes.
What actuary-grade data trust looks like in practice
| | Value (%) | |---|---| | Provenance | 25 | | Consent | 20 | | Recency | 15 | | Quality | 10 | | Concordance | 10 | | Validation | 10 | | Breadth | 5 | | Stability | 5 |
A parity analysis that would satisfy both DOL enforcement and actuarial professional standards needs data scored across several dimensions before the analysis begins.
Provenance. Every data element in the comparison must trace to a source system, a capture method, and a timestamp. Claims that were manually re-adjudicated, directory entries that were bulk-updated without provider verification, and utilization records that were backfilled from secondary sources all carry different provenance profiles. A trust score that weights provenance at 25% of the total score reflects the reality that unattributed data cannot support a regulatory comparison.
Recency. Parity analyses have reporting windows. Data that is stale relative to the analysis window produces misleading ratios. Provider directory data that was last verified 18 months ago cannot support a current-year network adequacy comparison. Claims data with a 90-day lag means the most recent quarter of the analysis window is incomplete.
Concordance. The same provider, the same procedure, and the same benefit classification must be coded consistently across the MH/SUD and med/surg datasets being compared. When a plan uses CPT codes for med/surg services and HCPCS codes for behavioral health services, concordance scoring identifies the mapping gaps before they corrupt a ratio.
Completeness. Missing fields are not neutral in a parity analysis. A denial record without a denial reason code cannot be classified. A prior authorization record without a turnaround time cannot support a timeliness comparison. Data completeness scoring quantifies what percentage of records in each dataset carry every field required for the analysis.
The enforcement trajectory
DOL enforcement is accelerating. The DOL's FY 2023 report (2023) documented that of the comparative analyses submitted by plans, not a single initial submission fully satisfied the requirements. Every plan required additional data or analysis.
State attorneys general are opening a second enforcement front. The National Association of Insurance Commissioners' model bulletin on MHPAEA (2023) encouraged state regulators to request NQTL comparative analyses and supporting data. Several states, including New York and California, now require annual parity compliance reports with data appendices.
The regulatory environment is moving toward continuous compliance, not periodic attestation. Plans that cannot produce trusted data on demand will face repeated remediation cycles. Each cycle costs time, money, and reputational risk.
How this connects to broader behavioral health data infrastructure
Parity compliance does not exist in isolation. The same data quality failures that break parity analyses also break behavioral health integration, care management targeting, and value-based payment models. A plan that cannot trust its behavioral health claims data for parity purposes cannot trust it for risk adjustment, network design, or quality reporting either.
The investment in parity compliance data quality is an investment in behavioral health data infrastructure broadly. Scoring every record before it enters an analysis pipeline, enforcing minimum trust thresholds for inclusion in regulatory submissions, and maintaining consent-aware data governance are not parity-specific requirements. They are prerequisites for any analytical use of behavioral health data.
The DTI Engine scores every record 0 to 100 across eight dimensions before your AI model sees it. If your team is building parity comparative analyses, NQTL documentation, or behavioral health network adequacy reporting and needs to know which records meet actuary-grade trust thresholds, talk to the SuperTruth commercial team. Schedule a conversation or call (215) 918-4140.
Further reading:
Sources
Jason Alan Snyder
Co-founder of SuperTruth and Artists & Robots, and an inventor on the Data Trust Index patents. Twenty-plus years building technology inside Interpublic Group. He writes here nearly every day on data trust, provenance, and what AI should be allowed to act on, and publishes essays on his Substack.
About SuperTruth · LinkedIn · Substack · jasonalansnyder.com
See it in practice
DTI scores the record, not the patient.
8 dimensions. 0 to 100. Travels with every record permanently.