Co-located BHI fails without shared data trust
Co-located behavioral health integration fails when PHQ-9 scores, substance use disorder records, and psychiatric diagnoses live in separate EHR instances with no shared identifier. Fewer than half of integrated primary care sites achieve reliable data exchange between behavioral health and medical records, and the downstream cost hits quality measurement, reimbursement, and patient safety.
Co-located behavioral health programs place a therapist or psychiatric consultant inside a primary care clinic. The assumption is that physical proximity solves coordination. It does not. When the behavioral health clinician documents in a separate EHR module, uses different coding conventions, or lacks access to the patient's medical problem list, co-location becomes co-existence. The data stays siloed even when the clinicians share a hallway.
The result is a behavioral health integration model that looks integrated on paper but fragments at the record level. That fragmentation breaks quality reporting, blocks accurate risk adjustment, and leaves AI models trained on these records with a false picture of the patient.
The scale of the integration gap
The Collaborative Care Model (CoCM), the most evidence-backed form of behavioral health integration, requires a shared care plan, a patient registry, and systematic follow-up using validated measures like the PHQ-9 for depression or the GAD-7 for anxiety. A 2020 Psychiatric Services study found that only 43% of primary care practices with co-located behavioral health services reported having a shared electronic care plan accessible to both primary care and behavioral health providers.
That means 57% of sites calling themselves integrated are operating with parallel documentation. Two clinicians, one patient, two records.
SAMHSA's 2023 National Survey on Drug Use and Health estimated that 57.8 million adults in the United States had a mental illness in 2023. Among those, roughly 55% received no mental health treatment. Co-located models are supposed to close that gap. But when the data infrastructure underneath them is broken, the models cannot prove they are working.
Where behavioral health data breaks in co-located settings
| | Value (%) | |---|---| | Sites without shared care plan | 57 | | BH clinicians unable to send data electronically | 64 | | Providers over-restricting Part 2 data sharing | 60 | | MIPS BH measures below completeness threshold | 40 | | CoCM practices with 48+ hour documentation lag | 31 |
The failure points are specific and measurable. Each one degrades the trust score of any record that passes through a co-located behavioral health site.
Separate EHR instances
Many co-located programs use a behavioral health EHR that does not write to the same instance as the primary care record. A 2022 Office of the National Coordinator for Health IT data brief reported that only 36% of behavioral health clinicians could electronically send patient health information to clinicians outside their organization. Inside the same clinic, the problem persists when behavioral health documentation lives in a module that the primary care physician cannot query.
PHQ-9 and GAD-7 capture without structured fields
Screening instruments are the backbone of measurement-based care. But practices frequently record scores in free-text notes rather than structured, coded fields. When a PHQ-9 score of 14 appears in a clinical note but not in a discrete data element, population health dashboards miss it. Quality measures that depend on screening completion rates report falsely low numbers. A 2021 JAMA Network Open study examining depression screening in primary care found that structured PHQ-9 data capture rates varied from 12% to 89% across participating sites, with the lowest rates at practices that had recently added co-located behavioral health services.
42 CFR Part 2 consent fragmentation
Substance use disorder records carry additional federal protections under 42 CFR Part 2. Even within a co-located setting, these records cannot flow to the primary care physician without specific written patient consent. The 2024 final rule updates from SAMHSA aligned Part 2 more closely with HIPAA, but implementation remains inconsistent. A 2023 report from the Government Accountability Office found that confusion about Part 2 requirements led 60% of surveyed providers to over-restrict data sharing, withholding behavioral health information even when consent was in place.
Billing code misalignment
BHI services billed under CoCM CPT codes (99492, 99493, 99494) require documented minutes of care management time, systematic psychiatric case review, and measurement-based treatment adjustments. When behavioral health clinicians bill under standard therapy codes (90834, 90837) instead of CoCM codes, the claim data tells a story of therapy delivered alongside primary care rather than integrated care. That distinction matters for value-based contracts that tie reimbursement to integration metrics. CMS's 2024 Physician Fee Schedule final rule maintained separate payment structures for CoCM and standard behavioral health, making billing code accuracy a direct determinant of revenue.
What co-located care data trust actually requires
Four conditions must hold for behavioral health integration data to be trustworthy enough for quality reporting, AI model training, or value-based payment.
Unified patient identity across clinical domains
The primary care record and the behavioral health record must resolve to the same patient identifier. Master patient index failures in co-located settings create duplicate records at rates between 8% and 12%, according to a 2021 AHIMA analysis. When a patient has one MPI entry in primary care and another in behavioral health, longitudinal outcome tracking becomes impossible.
Structured screening instrument capture
PHQ-9, GAD-7, AUDIT-C, and Columbia Suicide Severity Rating Scale scores must be stored as discrete coded data elements, not embedded in clinical notes. Structured capture enables automated registry reporting, population-level trending, and downstream analytics. The LOINC panel for PHQ-9 (code 44249-1) provides the standard encoding, but adoption remains inconsistent across behavioral health EHR products.
Consent-aware data routing
Substance use disorder data, psychotherapy notes, and HIV-related behavioral health records each carry distinct consent requirements. A co-located data architecture must enforce consent segmentation at the field level, not the document level. Blocking an entire behavioral health encounter because one data element requires restricted consent means the primary care physician loses access to the depression diagnosis, the medication list, and the care plan along with the restricted element.
Temporal alignment of encounters
CoCM billing and quality reporting require documentation that a psychiatric consultant reviewed the case within a defined period, that the patient's PHQ-9 was re-administered at a specified interval, and that care plan adjustments happened in response to measurement changes. When timestamps across systems drift, or when behavioral health encounters are documented days after they occur, the temporal chain breaks. A 2022 Health Affairs study on CoCM implementation found that 31% of practices had documentation lag of more than 48 hours between the behavioral health encounter and its entry into the shared record, making real-time care coordination impossible.
How data quality failures hit BHI quality measures
CMS and NCQA both track behavioral health integration through specific quality measures. When the underlying data is untrustworthy, these measures produce misleading results.
The HEDIS measure for Follow-Up After Hospitalization for Mental Illness (FUH) requires evidence of an outpatient visit within 7 or 30 days of psychiatric discharge. If the co-located behavioral health visit is documented in a system that does not feed into the plan's claims or encounter data pipeline, the visit looks like it never happened. The patient is flagged as lost to follow-up. The practice's quality score drops. The plan's star rating is affected.
CMS's Merit-based Incentive Payment System (MIPS) includes measures for depression screening and follow-up (CMS measure 134) and screening for clinical depression with a follow-up plan. A 2023 analysis by the American Medical Association found that 40% of practices reporting MIPS behavioral health measures had data completeness below the threshold needed for reliable performance comparison.
The AI model risk
AI models trained on co-located behavioral health data inherit every gap described above. A predictive model for depression relapse that trains on records where PHQ-9 scores are missing from 50% of behavioral health encounters will learn that depression monitoring does not happen regularly. It will underestimate the severity of patients who were actually being monitored but whose data was captured in free text.
Risk adjustment models that cannot access substance use disorder diagnoses because of over-restricted Part 2 consent will systematically under-score the clinical complexity of patients with co-occurring conditions. The financial and clinical consequences compound: lower risk scores mean lower capitated payments, which mean fewer resources allocated to the highest-need patients.
For any organization building or buying AI tools that touch behavioral health data, the question is not whether the model architecture is sound. The question is whether the training data carries verified provenance, current consent, and structured quality scores. Without those properties, the model's output is unauditable.
What trust-scored data changes
| | Value (%) | |---|---| | Provenance | 25 | | Consent | 20 | | Recency | 15 | | Quality | 10 | | Concordance | 10 | | Validation | 10 | | Breadth | 5 | | Stability | 5 |
Scoring behavioral health records across dimensions like provenance, consent, recency, and quality creates visibility into exactly where integration breaks. A record with a provenance score of 90 but a consent score of 30 tells you the clinical data is well-sourced but the consent governance is failing. A record with high quality but low recency tells you the screening data is structured but stale.
This is not abstract. When a health system can see that 40% of its co-located behavioral health records score below a consent threshold, it knows where to invest: consent workflow redesign, Part 2 training for front-desk staff, or EHR configuration changes that enable field-level consent segmentation.
When a health plan can see that encounter data from co-located sites has a recency lag of 72 hours on average, it knows its real-time care coordination dashboards are showing yesterday's picture of today's patients.
The point of scoring is not to grade clinicians. It is to make the data infrastructure visible so that the people responsible for integration can fix the pipes before the data reaches a model, a quality report, or a payment calculation.
The consent problem is the hardest problem
Behavioral health data sits at the intersection of the most restrictive consent regimes in healthcare. Psychotherapy notes under HIPAA. Substance use disorder records under 42 CFR Part 2. State-level mental health confidentiality statutes that vary across all 50 states. HIV-related behavioral health data with its own disclosure rules.
Co-located models make this harder, not easier. When a patient sees a primary care physician and a behavioral health clinician in the same visit, the consent architecture must handle the possibility that one data element from that visit is shareable, another requires specific authorization, and a third is governed by state law that differs from federal standards.
Most EHR systems handle this with document-level consent flags. The entire behavioral health note is either shared or blocked. That binary approach destroys integration. The clinician who needs to see the depression diagnosis and medication list cannot, because the note also contains a substance use disorder reference that requires Part 2 consent.
Field-level consent enforcement is the technical answer. But it requires consent architectures that most health IT systems do not have. The gap between what co-located care needs and what consent infrastructure delivers is the single largest barrier to trustworthy behavioral health integration data.
For a deeper treatment of this problem, see Mental health data: the most sensitive consent domain in healthcare AI and Substance use disorder data and the special status challenge for AI systems.
What comes next
The expansion of CoCM billing codes, the 2024 Part 2 alignment with HIPAA, and CMS's increasing emphasis on behavioral health quality measures all point in the same direction: co-located behavioral health integration will be measured, reported, and reimbursed based on data that most sites cannot currently produce at the quality level required.
Sites that invest in structured screening capture, unified patient identity, consent-aware data routing, and temporal alignment now will be able to report accurately, bill correctly, and contribute training data that AI models can actually trust. Sites that do not will find themselves unable to demonstrate the value of integration, even when their clinicians are doing excellent work.
The clinical model is not the bottleneck. The data model is.
The DTI Engine scores every record 0 to 100 across eight dimensions before your AI model sees it. For co-located behavioral health programs, that means scoring consent completeness, screening data structure, encounter recency, and cross-system concordance at the point of ingestion, not after a quality audit finds the gaps. If your team is building integrated care analytics or reporting BHI quality measures, 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.