Context and Consistency Matter - Does The Boston Globe Care?

Many of you have likely seen The Boston Globe’s report on Massachusetts Skilled Nursing Facilities (SNFs). The newspaper follows the typical narrative about an operator from out-of-state buying up pristine providers, cutting costs, reducing staff, and moving money from one pocket to the other. Quality declines, physical and psychological care needs go unattended, all punctuated with official terminology and data-driven analysis.

To be clear, I'm an advocate for the provider community, but I am no shill. I do not lobby for policies that benefit subsets of providers. I do not make excuses for bad operators. I am not a clinician, but I am highly sensitive to resident and family experiences described in the Globe, wherever they occur. I have no basis to challenge the accounts detailed in the story. There are processes for such matters that I would not presume to address. Then there's the data - official terms that hold gravitas because they are sourced from publicly available data released by CMS. That’s where I come in. Not because I was hired to (I was not) or am being paid to do so (I am not). I’m here because the Globe repeatedly treats reported healthcare data as though the numbers validate the narrative simply by existing. This is all too common; it’s also misleading and weaponized. My goal is not to stop the use of SNF data, it is to rationalize it.

Skilled Nursing data is a fragmented collection of inconsistent and distorted datasets that define a SNF’s Data Profile. Much of it is sourced from Medicare and Medicaid cost reports, Payroll Based Journal (PBJ) staffing, MDS assessments, claims, surveys, short- and long-term measures, state filings, and a growing assortment of supplemental reporting systems that describe different aspects of the same provider.

Many of these documents were not designed for analytics. They serve different purposes; rules evolved around specific environments and were retrofitted for the purposes of calibrating nominal quality. If only they aligned; they span various reporting periods, incentives differ, and oversight is inconsistent at best, theoretical at worst, but most often perfunctory. There are countless fields that could be picked out from one domain, compared to the same measure from another, and we receive a call to de-escalate the argument. “You’re both right and you’re both wrong.” The fact that two fields appear to address the same subject does not mean they’re measuring the same thing, and it certainly doesn’t mean they can be compared without first understanding how each figure was calculated.

The Globe's treatment of staffing at RegalCare at Taunton shows what can go wrong. Unfortunately, it doesn’t provide insight into how often it does indeed go wrong.

RegalCare took over on Sept. 1, 2022. The table below itemizes routine costs for respective cost-reporting periods. Maybe I’m missing something - these are allowable costs per day. The Globe mentioned severe cost cutting. I’m not seeing it. Cost per patient day was $350.39 in the eight months before the sale, $373.72 in the four months right after it, $385.34 in 2023, and $361.12 in 2024. Even where cost eased slightly in the most recent year, The Boston Globe leaves out a monumentally relevant factor. The facility maintained 100 certified beds in 2019, then dropped to 92 when RegalCare took over. They are down to 89 beds today. Billable days fell from 29,567 in 2021 - the Globe’s own pre-RegalCare baseline year - to 26,455 in 2024, roughly 10.5 percent. That comparison undersells it: 2021 was still a pandemic year, with occupancy suppressed industrywide. Measured against 2019, the last full pre-pandemic year, days are down nearly 19 percent, from 32,589 to 26,455. Either way, a shrinking census is a bit relevant, don’t you think?

The Alleged Staffing Collapse

The Globe’s central staffing claim compares 2021, the year before RegalCare took over, to 2025: RN hours down more than half, LPN hours down more than a third. Here’s the problem - the 2021 figure comes from Longmeadow’s Massachusetts state cost report, filed with CHIA. The 2025 figure comes from Taunton’s federal Medicare cost report, filed with CMS and published through HCRIS. Those aren’t two years of the same dataset. They’re two different regulatory filings, and we already know they don’t reconcile even for the identical facility in the identical year: Longmeadow’s own 2021 CHIA filing reports 179,376 combined RN, LPN, and CNA hours against 29,567 patient days, while the federal HCRIS series for that same year shows only 113,509 total direct care hours for the same building, same days. HCRIS runs 37 percent below CHIA before anyone compares a single year to another.

Apply that same 37 percent gap to the discipline-level numbers the Globe is working from - 23,561 CHIA-reported RN hours and 40,002 CHIA-reported LPN hours for 2021 - and the corrected, apples-to-apples starting point comes out to roughly 14,900 RN hours and 25,300 LPN hours. Compare those to Taunton’s actual 2025 HCRIS filing - 10,547 RN hours, 24,002 LPN hours - and the RN decline shrinks from “more than half” to about 29 percent. The LPN decline all but disappears: about 5 percent, not “more than a third.” I’m estimating that correction proportionally, not pulling it from a worksheet, so treat it as an illustration rather than a certified number. But it illustrates something real: most, if not all, of the Globe’s headline “collapse” is the two-filing gap we can already document, not a change in who was staffing the building.

And look - there’s another one. Checked year by year, the federal series on its own, 2017 through 2025, shows exactly one year that breaks the pattern, and it isn’t 2021. More on that in a moment. First, here’s a third way the numbers don’t hold up, this time inside a single federal filing.

Taunton's 2024 Medicare cost report shows 181,398 employee nursing hours, consisting of 24,443 RN hours, 41,625 LPN hours, and 115,330 CNA hours. Nursing salary expense was approximately $3.13 million. The 2025 cost report shows 90,350 employee nursing hours, including 10,547 RN hours, 24,002 LPN hours, and 55,801 CNA hours. Base nursing salary expense remained approximately $3.08 million.

Read the hours alone and the conclusion looks obvious: staffing appears to have fallen almost 50 percent. The problem appears when your gaze shifts to the wage column.

The 2024 filing reports average hourly wages of $24.27 for RNs, $23.08 for LPNs, and $13.68 for CNAs. The aggregate nursing wage works out to $17.26 an hour. Using the 2025 salaries and hours produces base hourly wages of approximately $49.48 for RNs, $46.67 for LPNs, and $25.74 for CNAs.

Anybody who has looked at a Massachusetts nursing home P&L in the last five years should stop right there. A 2024 RN wage of $24.27 in Massachusetts does not pass the "sniff" test. That's a terrible pun, but things get more interesting.

Divide the 181,398 nursing hours reported in 2024 by two and you get 90,699. The facility reported 90,350 hours in 2025 - a difference of four-tenths of one percent. I do not know exactly how the error occurred, and I am not going to manufacture an explanation without documentation, but I do know the 2024 nursing hours cannot be accepted at face value. They do not reconcile with the reported salary expense, they do not reconcile with the following year's filing, and they do not reconcile with another federal staffing dataset.

That should have stopped anyone relying on the 2024 hours until somebody figured out what happened. No one did. Why bother when the source is a CMS document?

For the record, the Globe’s own 19-percent figure is a decline in facility operating spending, not patient census - a different number that happens to land on the same digits. Census fell too, just not by the same amount: about 10.5 percent measured against the Globe’s 2021 baseline, nearly 19 percent against 2019.

Nine Years, One Outlier

One clarification before I go further, because it matters: the wage-mismatch finding above comes from Taunton’s federal Medicare cost report, filed with CMS and published through HCRIS. Longmeadow of Taunton also files a separate Massachusetts state cost report with CHIA, under the facility’s prior operator, and that filing tells a different story about 2021 - one worth checking before leaning on it. Those are two separate filings, built for two separate regulators, under two separate sets of rules. The cleanest way to settle which picture is right is to pull the full federal series and look at every year at once.

Longmeadow’s 2021 CHIA filing reports 179,376 combined RN, LPN, and CNA hours against 29,567 patient days - 6.07 HPRD. Read on its own, that number looks like a pandemic-inflated baseline. The federal HCRIS series below tells a different story.

Direct Care HPRD sits in a tight band across eight of the nine years shown - 3.28 to 4.93, including 3.84 in 2021 itself. The one year that breaks the pattern is 2024, at 6.86, immediately followed by a drop back to 3.54 in 2025. That’s the same 2024 filing the wage reconciliation already flagged as internally inconsistent. Two independent checks - the wage math and the nine-year trend - now point at the same single year. Whatever the CHIA number is picking up, it isn’t a pandemic-era staffing surge, and it’s better to run the two filings against each other now than to build an argument on the one that turns out to be the outlier.

Census matters, significantly. The facility carried 100 licensed beds and 29,567 patient days in 2021; it carries 89 beds and 25,509 days in 2025 - beds down 11, days down about 14 percent. A raw-hours comparison will look worse than an HPRD comparison for that reason alone, independent of anything that happened in 2024.

Payer mix, on the other hand, turns out not to explain much. Medicare’s share of patient days was 16.9 percent in 2021 and 17.5 percent in 2025 - essentially flat. The absolute Medicare-day count fell, from 5,010 to 4,465, but that tracks the building getting smaller, not a shift away from higher-acuity residents. I’d rather drop that explanation than force it.

PBJ Should Have Been the Next Stop

Payroll-Based Journal exists specifically to measure nursing home staffing (employees and contract labor) and feeds CMS Care Compare and the Five-Star staffing methodology. If a cost report appears to show a 50 percent reduction in nursing hours, PBJ is an obvious place to look next. Surprise, surprise: PBJ shows nothing resembling a 50 percent collapse in Taunton's total nursing staffing.

At that point the issue changes. We have a cost-report field producing an extraordinary decline, salary data inside the same filing suggesting the hours are wrong, and a second federal Data Domain that does not confirm the alleged collapse. That should trigger reconciliation before publication, litigation, policymaking, or anything else that attaches meaning to the number.

Instead, the number survives because it comes from a government filing and can be converted into a clean percentage and a razor-sharp sound bite. There's lazy and there's ignorant - pick your poison.

This is one of the central problems with Skilled Nursing data. A number acquires credibility based on the source, even when the surrounding data are telling you that something is wrong. The kicker is that it all comes from CMS.

How Did This Get Through?

Medicare cost reports are filed with Administrative Contractors. They are processed, accepted, and eventually made available in the HCRIS database. From there, they are used and abused by researchers, policymakers, consultants, investors, attorneys, reporters, and anyone else with a computer and enough interest to download the file.

How does a reported RN wage of $24.27 in Massachusetts make it through that process? How does a nursing-hours figure that appears to be almost exactly double what it should be avoid scrutiny? How does a facility report nearly identical nursing salary expense in consecutive years while reported nursing hours fall by half without anybody asking a question? How does a major-market newspaper let the obvious slip through?

I have written about the atrophy of cost-report accuracy for years. Medicare moved away from facility-specific cost-based reimbursement long ago, but the cost report never disappeared. What changed was the immediate economic consequences. A field that has little effect on settlement attracts less attention than it did when I started preparing these reports in the 1990s.

The irony is that cost reports have become more important as public research tools at roughly the same time that institutional expertise around them has weakened.

I may be the last professional still active nationally who prepared SNF Medicare cost reports when they actually drove facility-specific payment. Back then, you learned quickly that preparing the report was only part of the job. You had to understand how the pieces fit together. I called the Medicare cost report the "Reimbursement Rosetta Stone," and without a working understanding of the decades-old principles, it's difficult to appreciate how woefully inaccurate industry data has become. We lost a generation of cost-report professionals when PPS kicked in, and with it, the discipline essential for grounding the SNF economy. It's all connected, and if you're going to accuse a facility of an unprecedented statistical outlier without a second thought, how can anything reported be taken at face value?

Before reading this paragraph, note that it's AI: HHS OIG has documented weaknesses in Medicare Administrative Contractor cost-report oversight. Its 2025 review found that MACs did not consistently meet cost-report oversight requirements, with every one of the 12 jurisdictions missing at least one desk-review or audit-quality requirement in at least one year reviewed. That does not establish what happened with Taunton's filing, but it does make the broader concern about cost-report accuracy difficult to dismiss.

OK, I'm back. The previous text sugarcoats the problem. My team and I have gone deeper into every SNF Data Domain than any commercial effort could justify. The cost-report canard is that the document is unsuitable for rate setting - but in the theatre of the absurd, it gave the Globe a kickass sound bite.

Meanwhile, CMS protects itself with its own HCRIS disclaimers. The files reflect what providers reported, and users are responsible for the conclusions they draw. I understand the disclaimer, but the practical problem remains: the data carries the appearance of federal validation even though the individual fields may not have been validated in a manner appropriate for the uses now being made of them. In other words: file it and forget it - while academics, attorneys, and analysts produce research that informs policy. It's intolerable. CMS doesn’t weigh in to say, “the Cost Report has no clothes!”

The Plaintiffs' Bar Has Figured This Out

Plaintiffs' attorneys have become increasingly sophisticated in their use of public nursing home data. There is nothing inherently wrong with that. Public information should be used when it can identify poor performance, resident risk, or questionable practices. The problem appears when extraction is confused with analysis.

A federal form number looks impressive in an exhibit. A CMS source looks authoritative. They distill down to figures and terms a jury can understand. A percentage change without a relative anchor remains accurate - but that doesn't make it meaningful. If Mr. Jones could take four steps last week but only two steps today, that's a 50 percent decline. It sounds worse than a 30 percent decline - even though a 30 percent decline that took someone from 100 steps to 70 represents a far larger absolute change.

If the figure supports the theory of the case, there is not much incentive to keep looking until another dataset ruins the narrative.

Expert witnesses can compound the problem. Plenty of people know how a Medicare cost report is prepared. Far fewer understand how cost-report staffing should be reconciled with PBJ, MDS, case mix, census, agency utilization, payer mix, and Five-Star. Those are very different levels of expertise, and the distinction matters once somebody begins assigning operational meaning to the numbers.

Taunton is a perfect example. The 2024 and 2025 cost reports can be used to calculate an apparent nursing-hours decline approaching 50 percent. The arithmetic is easy. Once the salary line is brought into the analysis, the conclusion becomes difficult to defend. Bring in PBJ and the problem grows.

The proper conclusion at that stage is that we have a data-quality problem that requires investigation before anyone claims to know how much staffing actually changed. That is what an expert should recognize.

Real-life Examples

The most expensive term CMS ever attached to SNF policy may be “Expected Hours” - picture an attorney dramatically announcing that “CMS expected (emphasis added) this provider to deliver XX hours of direct care a day.” I have stories that would be comical if they didn’t drive up the cost of insurance to staggering heights, but I must share the deposition that sticks with me. Exasperated after I wouldn’t concede the point, the plaintiff’s attorney finally asked why I was dragging it out. I told him resident profiles differ and the data isn’t designed to reconcile resource use like that. I asked him a question back: “How many attorneys would a firm need to handle 1,000 parking violations a week?” Four or five, he guessed. “How many to handle a single class-action lawsuit against a major corporation?” At least 40, he said, plus full support staff. What he was arguing about this SNF amounted to the ABA mandating that every firm employ exactly 3.48 full-time attorneys per-100 active cases, regardless of what those cases entailed. Would he call the parking-violation firm understaffed? The comparison isn’t much of a stretch.

Case Mix Adds Another Layer

Massachusetts, like many states, transitioned its Medicaid acuity methodology to an MDS-based PDPM nursing framework, creating longitudinal comparison problems across different case-mix systems. Once the change occurs, the same facility can appear different simply because the measurement methodology changed.

This issue reaches well beyond Massachusetts. States that moved from legacy acuity systems into PDPM created a break in comparability that analysts need to recognize. Resident acuity may have changed, but the method used to describe it changed too, and those two effects should not be casually blended together.

PDPM also influences staffing expectations. A facility that improves MDS documentation can increase its reported CMI even when the underlying resident population has not materially changed. Better capture of depression, functional status, respiratory services, and other clinical characteristics increases recorded acuity and reimbursement, which in turn raises the staffing level CMS expects. This explains why case-mix states historically reported higher acuity than the others. CMS now suggests the phenomenon is "creep" - and some of it is - but in our experience, much of the need was always there, it’s just now being captured.  Go figure.

The Globe is comparing performance across a period that spans the state's move from MMQ to PDPM. The same caution applies anywhere a state changed the case-mix methodology underneath the data. Without accounting for the transition, methodology change can easily be mistaken for operational change.

MedPAC Should Know Better

MedPAC relies heavily on Medicare cost reports for its SNF margin work, which is one reason I have been frustrated by how little attention it has paid to the deterioration of cost-report reliability.

I rely on the cost report for our analytics and advisory work. They remain enormously valuable. The problem is the precision attached to conclusions derived from fields that may be inconsistently reported, weakly reviewed, or interpreted outside the context in which they were created. We are forced to scrub them like an unsolicited news circular in a blue plastic bag tossed on your driveway.  By the time it’s picked up, the document is cost report comparable. It’s gross.

MedPAC illustrated the sensitivity of its own analysis when it revised the methodology used to calculate SNF Medicare margins. Its calculated 2023 Medicare margin moved from 17.6 percent under the old method to roughly 22 percent under the updated method. Incorporating contract labor moved the result one way, while changing the nursing case-mix methodology under PDPM moved it substantially in the other direction.

The underlying providers did not change. The year did not change. The Medicare program did not change. The analytical treatment changed, and the answer moved materially. That should have generated a serious discussion about source-data integrity, comparability, and the extent to which cost-report outputs should be presented as precise measures of provider economics. Instead, we continue debating SNF margins to one decimal place while spending far too little time discussing what sits underneath the decimal.

This only scratches the surface. The Globe also never addressed two more variables that routinely distort year-over-year staffing comparisons: change of ownership (CHOW) and turnover. A CHOW can reset how a facility's cost-report periods align, introduce a short initial or stub reporting period, and change the entity preparing the filing altogether - any of which can produce hour and salary figures that look like an operational collapse when they're really an artifact of the transaction.

Turnover complicates the picture further. High RN and LPN turnover, especially when agency staff are substituted for employees during a hiring gap, can make employee-hours-only data - which is all a cost report captures - collapse even while total staffing, employee plus contract labor, holds steady or barely moves. PBJ captures both categories; the cost report does not. Neither CHOW nor turnover explains what happened at Taunton on its own, but a serious analysis rules them in or out before reaching for the word "collapse." For what it’s worth, Taunton’s current turnover runs higher than turnover at several of RegalCare’s other Massachusetts facilities, many of which report turnover below the state average. That spread argues against treating one facility’s staffing story as representative of the whole chain.

The Rent Line Isn’t What It Looks Like

One more item worth checking before anyone repeats it: the claim that rent at Taunton “roughly quadrupled to $522,000.” Taunton’s 2025 federal cost report shows exactly that figure - $522,000 in rent paid to a related real estate entity. It also shows CMS throwing that number out entirely.

Worksheet A-8-1 of the filing applies a rule that has governed Medicare cost reporting for decades: rent paid to a related organization isn’t an allowable cost at face value. What’s allowable is the related party’s actual cost of owning the building. So, the $522,000 gets backed out, and in its place the filing adds $383,296 in mortgage interest and $181,818 in depreciation - $565,114, more than the rent figure it replaces. The same worksheet also catches a related-party management fee of $225,681 and disallows $148,013 of it, down to an allowable $77,668. Net effect across all of it: allowable costs actually fall by $104,762 relative to what was originally booked.

So, the $522,000 number is real, but it was never going to be recognized as rent, quadrupled or otherwise - it's a related-party building cost that Medicare's own related-organization rule immediately converts into interest and depreciation, at a higher combined number. And the worksheet built to catch related-party overcharges caught one, on the management fee line, independent of anything the Globe reported. Citing the reported rent figure without opening the worksheet that sits three tabs over in the same filing isn't a data-entry error. It's not reading the document.

The Globe also cites $1.6 million paid to related companies last year - about 16 percent of operating expense - and the state’s health commissioner called that figure “concerning.” Worth sitting with for a second. Related-party transactions are common throughout the nursing home industry. A large share of SNFs lease their building from an affiliated real estate entity and contract management or therapy services from one, too. Federal rules don’t prohibit that. They require it to be disclosed, on exactly the schedule Taunton used: Worksheet A-8-1.

The $1.6 million isn’t a number RegalCare buried - it’s a number the filing is built to surface, sitting in the one place a regulator, a reporter, or a competitor would know to look for it.

And the same worksheet shows the disclosure requirement doing its job. It didn’t wave $1.6 million through. It cut the claimed management fee from $225,681 to an allowable $77,668 - a $148,013 disallowance. It backed the $522,000 rent out entirely and replaced it with the landlord’s actual ownership cost. Net effect: allowable costs came out $104,762 lower than what was originally booked. If the worry is that a related-party structure lets an owner quietly move money around unchecked, this filing is evidence of the opposite - the check is built into the form, and in this case, it caught something.

Why Data Domains Matter

The Taunton cost report happens to offer an unusually obvious example because the inconsistency nearly jumps off the page. Most data problems are less generous.

Cost reports, PBJ, MDS, claims, survey findings, Five-Star ratings, and ownership disclosures each describe a different part of the facility, and none of them can be interpreted responsibly in isolation. The analytical value comes from understanding how those domains interact, where they should reconcile, and why they sometimes do not.

If nursing expenses decline, I want to know what happened to nursing hours, wage rates, census, agency utilization, and PBJ. If staffing changes, I want to know which disciplines moved, how resident days changed, and whether case mix changed at the same time. If CMI rises, I want to understand whether the residents changed, documentation improved, reimbursement methodology changed, or several things happened together. A Five-Star decline requires the same discipline, because the overall rating tells me very little until I understand which component moved and what changed underneath it.

This is what I mean by cross-contextualization. We do not need more data nearly as badly as we need to understand the data we already have.

RegalCare happens to be the subject today. Another provider will be next. Some of the Globe's conclusions may survive a proper analysis; others may not survive at all. I have no idea, because the data can't be used to assess the quality of care.

I've been involved in enough depositions to gauge the knowledge base on the other side. One cost-report "expert," when asked for the worksheet reference, responded: "Worksheet S-3 Part V – ummm, well, I'm not quite sure if the 'V' is a Roman numeral or not, so it may be Part 5." That's a true story. I explained during my deposition that, considering the previous worksheet is Part IV, he'll figure it out eventually.

I've sat across from experts like that in depositions more times than I can count, and it never takes long to get to the worksheet they can't name. I almost feel bad by the end.

We don't handle cases like that much anymore. The schedule is disruptive, and I consider it blood money. It always comes down to whether I can tell staffing was adequate just by looking at the data. My answer is simple: "I have no idea if staffing was appropriate. The data can't be used like that. That's why I don't understand why you cherry-picked numbers across different Data Domains to make your case."

Numbers Are Not Data

Healthcare has confused access to data with expertise. Anyone can download HCRIS, pull PBJ, scrape Care Compare, and calculate a percentage. The difficult part begins after the spreadsheet is open, when somebody has to determine what the field actually means, why it was collected, what period it represents, whether the methodology changed, whether the number is internally plausible, and which other Data Domain should confirm or challenge it. Those questions determine whether the number is useful.

The Taunton filing offers an unusually clean example of what happens when they are skipped. A federal cost report contains an obvious error. The error survives the Medicare contractor process, reaches the public dataset, and can then be used to support a highly consequential claim about resident care.

There is no shortage of data in Skilled Nursing.

There is a shortage of people willing or able to figure out what it means. Numbers are not data. Context matters, and in Skilled Nursing, cross-contextualization is how you find the facts.

One final observation before I go. The New York Rangers and Boston Bruins both opened their seasons last night at TD Garden. Sadly for me, the Rangers lost 3 – 0. Yet when I picked the Globes sports section, the headline read: Rangers rookie makes three pointer in the 14th inning, dropping Boston to three games under .500.

The numbers sound precise enough. Someone please explain to the Globe that cross-contextualization doesn’t work in sports. SNFs are different. That’s why we’re here.

Marc Zimmet is the CEO of Zimmet Healthcare Services Group.