Healthcare was the fastest-growing buyout sector last year; McKinsey put the increase in buyout deal value at 51%. For deals above $500 million, the increase was 173%. Bain counts a record $191 billion in global healthcare private equity deal value, while provider and related services rose 57% to about $62 billion, and deal volume stayed flat.

Flat volume with higher value means larger checks on the same operational base.

The sources of that return have also changed. For example, debt fell to 37% of entry multiples in 2025 (vs 44% in 2016).  An analysis by StepStone on buyout deals entered between 2010 and 2022 revealed that leverage and multiple expansion produced 59% of returns, with the remaining 41% from revenue growth and margin expansion.

Revenue growth and margin expansion are what a sponsor can act on directly.

Entry multiples and the debt markets are set outside the deal. Debt now funds less of the entry price than it did in 2016, so a larger share of the return must come from the operating side. In healthcare services, that means the revenue cycle – and an error in how you measure- now costs more than it did previously.

With that in mind, I decided to find the primary source behind four claims that appear in nearly every healthcare revenue cycle discussion. My findings uncovered something interesting: two of the data points have no source at all, one traces to a nearly decade-old vendor report, and the last point traces to a single consulting firm’s project work, described in a trade journal by the managing director who leads that practice.

The four claims are as follows:

1. $200 per missed appointment
2. $150 billion a year in missed appointments
3. 55 to 65% of referrals leak out of network
4. Half of denied claims are never reworked, according to MGMA

When you line up the claims, they start to look like a single patient visit instead of four disparate data points:

• Patient books an appointment and doesn’t show ($200 no-show)
• Patient gets a referral to a specialist (55 to 65% leakage)
• Clinician does the work, and someone writes it down (charge capture)
• Payer decides whether to pay (denial)

Next, consider what each number measures: a handoff between two systems that the organization bought separately, years apart, from different companies. The no-show number needs the scheduling system and the general ledger; the leakage number needs the referral order and the claim; charge capture needs the clinical note and the charge master; denial number needs the claim and the remittance.

No single system owns both sides of those four handoffs. As a result, no one measures them.

The operator needs a number to get a project funded, but they can’t compute it because the data sits on both sides of a handoff. As a next best option, they go looking for a published number to reference.

Now, let’s unpack the published data and where it originates from.

Widely Accepted Claim #1: $200 leakage per missed appointment, resulting $150 billion leaked annually

These circulate as two findings, but they are derived from one number.

Healthcare Finance News reported both findings in November 2016 from a vendor report by SCI Solutions:

• Each 60-minute open or no-show slot costs a physician ~$200

• National annual cost is $150 billion

• The second number is the first number multiplied across the system

That same report put no-show rates between 5% and 30%. To be clear, a practice at 5% and a practice at 30% do not have the same problem. That range is the part an operator can act on, and it disappeared from the data story. What survived is a single cost number, now in its tenth year.

The number has also changed hands.

SCI Solutions sells scheduling software and MGMA is a professional association for medical group leaders. At least one widely circulated industry roundup now credits the $200 to MGMA rather than the vendor that produced it.

That swap is what makes the number viewed as credible.

Widely Accepted Claim #2: 55 to 65% of referrals leak out of network

Despite my attempts, I found no attribution for this number.

The closest peer-reviewed work measures referral completion, which is a different metric entirely. Christopher Forrest and colleagues published the results in the Annals of Family Medicine in 2007. They tracked 776 referred patients from 133 physicians across 81 practices and 30 states.  In the study, they asked one question: did the patient see the specialist within three months?

Physicians reported 79.2%.

Patients reported 83.0% for the same referrals.

Both figures come from questionnaires, so neither is derived from a claim or a record. Agreement between the two accounts was only fair, at a kappa of 0.34. The people closest to the referral disagreed about whether it actually happened.

The reasons for non-completion settle the leakage question. Patients who did not go cited a lack of time, or a belief that the problem had resolved itself.

An important outcome of this work is that those patients did notgo to a competitor; they simply did not go at all. In other words, the study measures a scheduling and follow-up failure, and the 55 to 65% number claims a network failure. The two data points have different fixes.

Size a network problem with completion data, and you’ll fund the wrong project.

The study did find one operational lever, however: completion rates rose when the physician or the staff booked the specialty appointment, rather than leaving the patient to call. Medicaid patients completed less often and met health plan denials more often.

Payer mix moved the result: network ownership was never actually part of the study.

Widely Accepted Claim #3: 65% of denied claims are never reworked

This claim does have attribution, but it changes depending on who is citing it.

Billing and vendor publications credit the same 65% to MGMA, to Change Healthcare, and to the American Medical Association. One number, three different authorities, and no original study behind any of them.

A real measurement does exist. Vabson, Hicks, and Chernew published it in Health Affairs in June 2025.They analyzed 270 million Medicare Advantage claim submissions from 2019, using Inovalon data that covers about 30% of the MA market. Medicare Advantage plans denied 17% of initial claim submissions. Providers appealed, and 57% of all denials were eventually overturned.

Next comes the number that does belong in a model; after every reversal, denials still cut ~7% from provider Medicare Advantage revenue. The authors weighted that figure by dollars instead of by claim count. The authors measured only the direct effect on revenue, and they note the indirect effect may be bigger.

Two costs sit outside the 7%: one is the labor of chasing a denial, the other is the care a patient never received because the denial stood.

Widely Accepted Claim #4: Charge capture leakage of 1 to 3% of net revenue

The Healthcare Financial Management Association actually published the 1%. The article was written by a managing director at a financial management and consulting firm who leads its revenue transformation practice. The basis was his own firm’s revenue transformation engagements, and his words were “as much as 1 percent”, which sets a ceiling.

In circulation, the ceiling became the floor, and vendors who sell charge capture software added the “3%” on top.

Every data distortion impacts the shape of the problem

After careful review, it’s easy to see how these slight distortions of data become accepted industry norms; a range collapses into a point; a ceiling inverts into a floor; one firm’s engagement experience becomes a constant for the entire industry. A study of whether patients saw a specialist becomes a claim about network ownership.

Every one of these numbers got bigger, and big numbers help get projects funded. The problem? You size the opportunity with a number that no one can trace, spend against it, and the expected return misses.

What this costs in diligence

An unsourced number costs nothing at entry. No one stress-tests the size of an opportunity you are arguing for. The number does its job; the project gets approved, and it goes into the plan.

The test comes later, and McKinsey data shows why later is expensive.

Among deals exited since 2019, the final holding year produced 6% of ending EBITDA margin; 2018  produced 4%, and every year before that produced ~1%. Operating improvement is crammed into the end of the hold, which is also when a buyer starts checking the math.

Holds are getting longer. In 2025, 52% of buyout-backed portfolio companies carried four or more years of ownership, up from 43% a year earlier. A longer hold means more parties re-underwrite the same plan; a new deal team; a lender; a buyer’s quality of earnings provider does it last, and which is when the number fails.

One untraceable number is enough to put the whole plan into question.

A diligence team traces one operating number to a blog post. The question then moves from the number to the company. The buyer stops asking whether that number is right and starts questioning whether the company can measure anything at all. Every remaining number then gets re-derived at your expense, on the buyer’s assumptions, with the buyer’s discount on the gap.

Finally, someone is checking references

Until this year, there was no public data on denial rates, which is why data that no one could source survived for so long.

In March, a CMS rule forced payers to publish their numbers. Health policy research nonprofit, Kaiser Family Foundation, analyzed the first full year (2025) on August 13th, 2026.

Not surprisingly, the published rates are nowhere near each other.

In Medicare Advantage, Elevance denied 5% of standard requests. UnitedHealth denied 17%. This means that a buyer can now look up payers at each clinic they own and estimate denial exposure before they ask you for it. If your business case still runs on 65%, they will find the gap.

The right way to run an engagement

At Teragonia, our healthcare team does not put an industry benchmark into a model. Instead, the first deliverable in a revenue cycle engagement is the client’s own baseline, built from their records and raw data as the foundation.

That is a deliberate position, and it sits between the two places operators usually go for help:

• Consulting services will benchmark your denial rate against a peer set, and hand you a recommendation

• AI point solutions will work one queue extremely well, and see nothing on either side of it

Neither can trace a dollar back to the event that produced it, because neither owns the connection between the systems where that dollar was created.

That connection is the most important work, and that work takes three stages.

The first stage reconciles the systems.

Scheduling, authorization, clinical documentation, charge and remittance each hold one part of a single encounter. In a group built from a dozen acquisitions, those five records sit in four or five systems that no one designed to agree.

Most of the work is making one dollar trace back to the event that produced it. This is the pass that puts both sides of a handoff in the same place for the first time.

The second stage computes the rates at the level where people make decisions.

Those rates are denial rate by payer and by site, overturn rate by denial reason and net revenue per completed visit by service line. These are the local equivalents of the four borrowed numbers, and the client already pays for the systems that hold them.

No new data had to be bought.

The third stage marks what cannot be measured yet.

Clients argue with this stage, and yet it is the stage that survives examination. The surgical system, the rounding application, and the imaging system often do not connect to billing. Then, nothing reconciles what the clinician did against what the biller billed. No benchmark closes that gap.

Write “we cannot see this yet, and here is the cost to fix it” into the memo instead. A buyer’s analyst will trace a confident percentage to a 2016 vendor report. That sentence survives the analyst.

The output is not just a better number, but a business that conveys it can produce its own numbers on demand, from records it already owns.

When a buyer traces one figure in your deck, they find out whether the company can produce any figure at all from its own records. A single failed trace tells them how to read the rest of it.

The Traceability Standard: Three Questions to Ask

• Where did this data come from? Not what it is, but where it came from. If the answer is a blog post that cites another blog post, treat the number as unknown.
• What is our own version of the data? Nearly every borrowed benchmark has a local equivalent you can compute.  The cost of an empty slot is net revenue per visit for that service line. The denial rate is in the remits.
• Which direction does the error run? If the number in the deck is larger than the number you can defend, ask who gains from the difference.

Each of the four claims measures a handoff that no system owns both sides of. Until March, nothing could contradict them. A buyer can now pull denial rates for every payer you contract with, and the gap between their figure and yours comes out of the price.