Why Behavioral Health Referrals Still Run on Fax in 2026 and What It Takes to Close the Loop

The short answer

Behavioral health referral coordination still runs on fax for four structural reasons, none of which are about reluctance to modernize:

1. Federal funding excluded the sector. The 2009 HITECH Act directed roughly $35 billion in EHR incentive payments to hospitals and eligible professionals, but substance use and mental health treatment facilities were largely left out. The sector began digitizing roughly a decade and a half behind everyone else, without subsidy.
2. Consent law makes automated sharing legally complicated. 42 CFR Part 2 imposes stricter requirements on substance use disorder records than HIPAA alone. Most EHRs and health information exchanges still cannot segment data at the attribute level, so the compliant workaround is frequently to keep sensitive records out of automated exchange entirely.
3. Behavioral health providers largely aren’t on the exchanges. In ONC’s analysis of the 2024 National Substance Use and Mental Health Services Survey, just 19% of facilities reported participating in a health information exchange, and 67% said they were unfamiliar with HIEs or unaware of availability in their service area.
4. Referral sources are irreducibly heterogeneous. A behavioral health intake queue receives referrals from emergency departments, primary care, courts, jails, schools, child welfare agencies, employee assistance programs, crisis lines, and self-referring families. Fax is the only channel every one of those senders can use today.

Fax persists because it is the lowest common denominator satisfying all four constraints simultaneously. That is also why “just get rid of fax” has failed as a strategy for fifteen years, and why the productive question is different:

What has to happen to a faxed referral after it arrives?

Key takeaways

• Adoption is no longer the gap; exchange is. More than two-thirds of substance use and mental health treatment facilities now maintain records in an EHR only. The problem moved from “do they have a system” to “can their system talk to anyone.”
• Consent is a workflow problem, not only a legal one. The 2024 Part 2 revisions permit a single consent covering treatment, payment, and operations — a meaningful simplification most organizations have not yet operationalized.
• Referral leakage is severe. Industry analyses put the share of referrals that never complete at 25–40%, with some estimates higher. In behavioral health, where readiness to engage is time-limited, a delayed referral is frequently equivalent to a lost patient.
• The failures are administrative, not clinical. Referrals stall at intake, at the chase for missing information, and at patient contact — three stages that are entirely document and communication handling.
• Ownership gaps between your referral partners are enormous. State-run facilities reported EHR adoption at 38% versus 97% for federal facilities. You are not dealing with one interoperability environment; you are dealing with several at once.
• The fix is not eliminating the channel. It is making the inbound channel structured, tracked, consent-aware, and closed-loop.

Part 1: What the data actually says about behavioral health health IT

For years the sector’s story was low EHR adoption. That story is now out of date, and repeating it obscures the real problem.

ONC’s data brief drawing on SAMHSA’s 2024 National Substance Use and Mental Health Services Survey found more than two-thirds of substance use and mental health treatment facilities maintain patient records in an EHR only, rather than on paper. Adoption largely happened.

What didn’t happen was exchange capability.

Capability	Share of facilities
Participate in a health information exchange	19%
Unfamiliar with HIEs or unaware of local availability	67%
Integrate external information electronically (EHR-only facilities)	48%
Integrate external information electronically (EHR + paper facilities)	36%
Use EHR for patient messaging	45%
Use EHR for patient access to records	44%

The ownership spread

Facility ownership	EHR adoption
Federal	97%
Private for-profit	68%
State-run	38%

ONC characterized the state-facility gap as significant, pointing to cost, data fragmentation, and workforce challenges as likely contributors.

If you run a behavioral health organization receiving referrals across that landscape, your intake process must accommodate the weakest sender, not the average one. That single constraint explains more about persistent fax volume than any cultural explanation.

ONC’s own read on the findings

ONC identified two distinct gaps: shortfalls in what implemented EHRs can actually do, and shortfalls in how facilities implemented and use them. That distinction is the whole ballgame. Buying a better system doesn’t help if the workflow around it never changed — a point the implementation research reinforces later in this article.

The demand context

SAMHSA’s 2024 NSDUH data indicates roughly a third of U.S. adults — on the order of 86.6 million people — experienced a mental health or substance use condition. Research consistently finds that most people who could benefit from SUD treatment don’t receive it. Meanwhile, KFF analysis has found that nearly half the U.S. population lives in an area with a behavioral health workforce shortage, and HRSA projections point to continuing shortfalls among addiction counselors, mental health counselors, marriage and family therapists, psychologists, and adult and child psychiatrists.

Put those facts together and the stakes of referral coordination sharpen considerably. When clinical capacity is scarce, every referral that fails for administrative reasons wastes a slot that someone else needed. Referral leakage in behavioral health isn’t only lost revenue; it’s misallocated scarce capacity.

Most articles wave at 42 CFR Part 2 and move on. It deserves precision, because the details determine exactly what you can and cannot automate.

What Part 2 covers

Part 2 applies to federally assisted programs that both hold themselves out as providing, and do provide, substance use disorder diagnosis, treatment, or referral for treatment. Note that referral for treatment sits inside the scope. The referral document itself can be protected information.

Why it’s stricter than HIPAA

HIPAA permits disclosure for treatment, payment, and operations without specific patient authorization. Part 2 historically required patient consent for each disclosure to each recipient. That difference is the entire reason a behavioral health record cannot simply flow through the same pipes as a cardiology record.

What changed in 2024

The revised rule permits a single patient consent covering treatment, payment, and health care operations, rather than requiring fresh consent for every disclosure to every entity. For programs that had been managing consent per-transaction, this is a substantial reduction in administrative burden. The revisions also brought Part 2 records under HIPAA’s enforcement framework for certain purposes, including breach notification obligations — so the simplification came packaged with elevated enforcement exposure.

Why that didn’t end the fax era

Two reasons, one technical and one operational.

Technical. Compliant automated exchange requires data segmentation — the ability to tag and control SUD-specific content at the data level rather than blocking or releasing whole documents. Data Segmentation for Privacy (DS4P) standards exist, and FHIR supports security labeling. But most deployed EHRs and HIEs still don’t implement segmentation in practice. Programs without it face a binary choice: overshare or block.

Operational. The failure mode of manual redaction runs in both directions, and both directions are bad. Overshare and you have a reportable disclosure with HIPAA penalties now attached. Undershare and the receiving clinician makes decisions without knowing about an active SUD or a medication interaction — a documented complaint from primary care physicians who couldn’t access relevant records during emergencies.

Faced with that, keeping sensitive referrals on a channel where a human reviews every page before it moves is a defensible risk decision. It is also slow, unmeasurable, and entirely dependent on staffing levels.

Part 3: Anatomy of a behavioral health referral

Here is the actual path, with failure points marked.

The referral source landscape

Referral source	Typical sender infrastructure	Common channel	Data quality
Hospital ED / inpatient psych	Enterprise EHR	Fax, occasionally direct message	Moderate — often lacks insurance detail
Primary care / FQHC	Certified EHR	Fax, referral order	Good clinically, weak on consent
Crisis line / 988 follow-up	Call center system	Fax, phone, secure email	Urgent, minimal documentation
Courts and probation	Case management or paper	Fax, mail	Court order attached; clinical detail sparse
Jails and corrections	Varies widely	Fax	Time-critical at release; frequently incomplete
Schools and child welfare	Non-clinical systems	Fax, email	Consent complexity high
Employee assistance programs	Vendor platform	Fax, portal	Insurance detail present, clinical detail thin
Self / family	None	Phone, web form	No records; highest engagement urgency

No single electronic standard reaches all eight. That is the operational reality any referral automation strategy has to accommodate.

The seven stages, and where each breaks

1. The referral is generated. A social worker, PCP, court liaison, or counselor decides a patient needs behavioral health services.

Breaks here: the sender doesn’t know your intake criteria, current availability, or which program fits. Referrals arrive mismatched to level of care, which means a clinician has to triage before anything else can happen.

2. It transmits. Fax, in most cases, for the reasons above.

Breaks here: transmission failures nobody discovers. Multi-page packets that split or arrive out of order. Documents landing on a shared line with no owner.

3. It arrives at intake. Someone opens it, reads it, determines what it is, identifies the patient, and decides which program it belongs to.

Breaks here: the largest single latency in the chain, and almost nobody measures it. Documo’s 2025 Stuck in the Fax Lane survey of more than 100 healthcare administrators, HIM specialists, and IT professionals found 52% of inbound faxes still require manual intervention and 44% are time-sensitive, with 88% of practitioners reporting that fax-related delays negatively affect patient care and only 29% describing their workflows as fully automated.

4. Missing information gets chased. Insurance details, consent forms, prior treatment history, medication lists, court documentation.

Breaks here: the chase is unbounded. Staff call the referring office, wait for a callback, re-fax a consent form, wait again. Meanwhile the engagement clock runs.

5. Benefits and authorization are verified. Behavioral health carries some of the highest initial denial rates of any specialty, with high-acuity services routinely exceeding 15%.

Breaks here: authorization delay pushes the first available appointment past the window in which the patient is still willing to attend.

6. The patient is contacted and scheduled.

Breaks here: by the time contact happens, phone numbers are stale or circumstances have changed. In SUD care especially, readiness to engage is measured in hours and days, not weeks.

7. The loop closes — or doesn’t. The referring provider learns whether the patient was seen.

Breaks here: most often. Without closed-loop confirmation, the referring organization cannot document care coordination for quality measures, and nobody in the chain knows the patient fell through.

How much leaks

Industry analyses of referral completion put the share that never complete at 25–40%, drawing on athenahealth network data and Health Affairs research; some closed-loop referral vendors cite figures as high as 65% depending on definition and setting.

Whatever the precise number for your organization, the operative finding is that failures cluster at stages 3, 4, and 6 — all document and communication handling, none of them clinical.

Part 4: Why the clinical research points at the same bottleneck

The literature on referral effectiveness converges with the operational picture in a way worth taking seriously.

Interventions built on assertive linkage produce measurably better connection rates. One quasi-experimental analysis of a motivational engagement intervention at care transitions found participants had roughly twice the odds of connecting to step-down SUD treatment, and better odds of connecting within 10 days. Research on recovery management checkups in primary care settings has reported very high rates of patients agreeing to treatment and attending intake when repeated, assertive follow-up is built into the model.

What those interventions share is speed and human contact at the moment of readiness. Every hour consumed by manual document handling at intake is an hour subtracted from the window in which a warm handoff is still possible. Automation of the document layer isn’t a substitute for clinical engagement work — it is what creates room for it.

The cautionary finding

There is a result here that cuts against easy technology optimism, and it belongs in any honest business case.

A study of SUD treatment data integration at a safety net health system found that after integrating program data into the EHR, the number of patients trackable as initiating treatment rose 250% — from 562 to 1,411 — while measured referral linkage declined from 74% to 48%. Ninety-day retention rose from 45% to 74%. The authors attributed the linkage decline largely to a broadened denominator and changed outcome definitions. Addiction therapists appreciated having information in one place but did not report large time savings shortly after integration.

Three lessons:

1. Better visibility often makes your numbers look worse before they look better. Prepare leadership for this or the project gets killed in month four.
2. Define your metrics before you change the system, or you cannot tell improvement from measurement artifact.
3. Technology alone doesn’t produce the savings. The authors specifically noted that greater preparatory workflow analysis would likely have reduced end-user burden.

Part 5: What closing the loop actually requires

Five capabilities, in order of implementation difficulty.

1. Structured capture at the door

Every inbound referral gets classified on arrival — referral versus records versus consent form versus authorization correspondence — with the patient identified and the document associated to a record. This is where intelligent document processing earns its keep: extraction of referral source, patient identifiers, requested level of care, insurance, referring provider, and referral date from an unstructured fax.

Why it matters most: everything downstream depends on knowing what arrived and for whom. An unclassified document cannot be prioritized, tracked, or escalated.

2. Missing-information detection at intake rather than at review

The highest-leverage automation in behavioral health referral coordination is not routing — it is flagging, within minutes of arrival, that a referral lacks the consent form, insurance detail, or clinical documentation required to proceed.

That converts a multi-day chase into a same-day callback while the referring office still has the chart open and remembers the patient. The difference in completion rates between a callback on day zero and a callback on day four is not marginal.

3. Consent-aware routing

Sensitive content is identified and handled according to the applicable consent, with consent status attached to the record rather than living in a coordinator’s memory or a separate binder.

This is precisely where the 2024 Part 2 single-consent provision becomes operationally valuable: one properly executed TPO consent, tracked in the workflow and queryable at the moment of disclosure, replaces a per-disclosure scramble. Organizations that updated their consent forms but not their consent workflow captured only half the available benefit.

4. Timestamped tracking with an owner

Every referral has a current status, a named owner, and an elapsed clock visible to leadership. This is the capability that makes referral conversion measurable at all — and the absence of which explains why most organizations cannot answer basic questions about their own conversion rate.

5. Automated loop closure

Confirmation back to the referring source when the patient is scheduled and when the patient is seen — through whatever channel that sender can receive, including fax.

This is the point where “get rid of fax” thinking causes real damage. If your court liaison partner can only receive fax, closing the loop by fax is the correct engineering decision. Outbound fax that fires automatically from a workflow event is not legacy technology; it’s meeting a partner where they are.

What stays human

The pattern in healthcare automation adoption consistently favors machine handling of identification, extraction, and routing while keeping humans on clinical decisions and on first contact with vulnerable patients.

Automate: classification, extraction, patient matching, missing-information detection, consent status tracking, deadline calculation, status visibility, loop closure notifications, reporting.

Keep human: level-of-care determination, clinical triage of acuity and risk, first outreach in high-acuity or crisis cases, and any judgment where being wrong has clinical consequence.

Be explicit about that boundary internally. Staff resistance to referral automation is usually a fear about the wrong boundary being drawn — that software will decide who gets treated. Naming the boundary early converts opposition into participation.

Part 6: Compliance checklist for automating referral intake

Before any automation touches a Part 2 record, confirm:

• Vendor status and BAA. Any platform handling Part 2 data needs an appropriate business associate agreement, and you should confirm how it handles re-disclosure notice requirements.
• Segmentation capability. Can the system tag SUD-specific content at the data level, or does it handle only whole documents? Whole-document handling is workable but constrains what can be routed automatically.
• Consent tracking and queryability. Where does consent status live, and can it be checked at the moment of disclosure rather than retrieved from a separate file?
• Audit trail. Every view, transmission, and export of Part 2 content must be logged. This is both a compliance requirement and your defense in a dispute.
• Break-glass handling. Emergency access patterns need to exist, be documented, and be auditable.
• State overlay. For multi-state operations, confirm which state’s consent standard governs each program and whether your workflow can enforce different rules per site.
• Breach notification alignment. Post-2024, Part 2 breaches carry HIPAA notification obligations. Your incident response plan should reflect that.
• Minor-specific requirements. Consent rules for adolescent behavioral health vary by state and are frequently missed in system design.

Nothing on that list is a reason to delay. All of it is a reason to involve compliance counsel in the design phase rather than the deployment phase.

Part 7: A 90-day implementation roadmap

Days 1–30 — Measure what you have.

Inventory every inbound channel and fax number. Sample 100 recent referrals and record five timestamps each: sender date, arrival, first human touch, first patient contact attempt, and first appointment. Classify by referral source. Calculate your incomplete-referral rate by source. Most organizations discover their worst-performing source is one nobody was tracking.

Days 31–60 — Consolidate and assign.

Route every fax number into a single capture point. Assign a named owner and monitoring cadence to every channel, including the ones currently landing in individual mailboxes. Build the ownership matrix: referral type × owning role × backup × escalation trigger. Update consent forms to the 2024 single-TPO structure if you haven’t, and — more importantly — decide where consent status will be recorded so it can be checked in workflow.

Days 61–90 — Automate the highest-volume source.

Pick one referral source, usually hospital ED or primary care, and implement classification, extraction, patient matching, missing-information flagging, and loop closure for that source only. Measure against your day 1–30 baseline. Expand once results hold.

Resist automating everything at once. Organizations that do generally cannot tell which change produced which result, and cannot defend the investment at renewal.

Build, buy, or hire?

Hire if your arrival-to-decision latency is already low and your incomplete-referral rate is under 15% — your constraint is clinical capacity, and coordinators won’t fix that.

Build only with genuine engineering capacity and a stable referral source mix. Classification and extraction models require ongoing maintenance as referral forms change, and behavioral health forms change constantly.

Buy if your channels are fragmented, your exception rate is high, or you cannot currently produce the metrics in Part 8. Prioritize vendors that handle both inbound and outbound fax, because loop closure to low-tech partners is where many referral platforms quietly fail.

The clarifying question for a budget conversation: are we competing with software, or with another FTE? In behavioral health, where coordinator turnover is high and hiring is slow, the answer is often the second — and that changes both the ROI model and the procurement path.

Part 8: The metrics that prove it worked

If you cannot currently produce these numbers, that itself is the finding:

1. Referral arrival to intake decision — median and 90th percentile.
2. Arrival to first patient contact attempt — the metric most directly tied to engagement, and the one most affected by document automation.
3. Referral conversion rate — referrals received that become completed first appointments.
4. Incomplete-referral rate by source — share arriving without information required to proceed. Your best lever for upstream conversations with referring partners.
5. Loop closure rate — share of referrals where outcome was communicated back to the sender.
6. Time to third-next-available appointment — the standard access measure, and the one that connects intake performance to patient access outcomes.
7. Cross-site variance — the spread between best and worst performing site on metrics 1 and 3.

That last one tends to be the most uncomfortable and the most actionable. Wide variance between sites running identical workflows is almost always a staffing or process ownership issue, and it stays invisible until intake is instrumented centrally.

Benchmarks to aim at

There is no authoritative national benchmark for behavioral health referral conversion, which is itself a meaningful gap. Reasonable internal targets, based on what coordination-focused models have demonstrated:

• Arrival to intake decision: same business day
• Arrival to first contact attempt: within 24 hours
• Incomplete-referral rate: under 15% from any single high-volume source
• Loop closure: above 90%

Part 9: Objections worth taking seriously

“Our referral partners will never leave fax.” Correct, and the strategy doesn’t require them to. The goal is structuring what arrives, not changing what senders do.

“Part 2 means we can’t automate.” Part 2 constrains what content moves where under which consent. It does not prohibit classifying a document, identifying a patient, or flagging that a consent form is missing. Most of the automation value sits in operations that don’t require disclosure at all.

“We tried a referral platform and it didn’t help.” The most common failure mode is a platform that manages referrals your staff enter manually, which relocates the bottleneck rather than removing it. Capture has to happen at the channel, not after a human has already read the document.

“Our volume isn’t big enough to justify it.” Possibly true. Run Part 7’s day 1–30 measurement first. If arrival-to-contact is already under 24 hours, spend the money on clinicians instead.

“Won’t this depersonalize intake?” The opposite argument is stronger and better supported: the clinical literature says engagement depends on fast human contact at the moment of readiness. Automating document handling is what makes that contact possible sooner.

Frequently Asked Questions