Randomized 180-day readmission62.3% vs 61.7%Intervention versus usual care
Adjusted difference+0.82 points95% CI −5.97 to +7.61; P=.81
Memory hook: The intervention group improved. The control group improved too. The large apparent improvement was not caused by the program intervention.
Read that precisely: the dramatic before/after decline was not demonstrated to be the program’s added effect. The randomized comparison found no significant readmission benefit; it did not prove a zero effect.
Why this is such a useful population-health study
The 2020 randomized trial enrolled 800 medically and socially complex patients; 782 were linked to discharge data. Participants received substantial contact: an average of 28.1 encounters. The comparison group exposed the misleading before/after story.
What it does not prove
One program in one setting is not a verdict on every intervention. Modest benefit remained compatible with the interval. The result does not establish effects on every patient outcome, or show that hospice or palliative care fails.
Meeting-ready line: “Show what happened to comparable patients without the program before crediting the whole decline to the intervention.”
Patients may move toward their own longer-run expected spending, not necessarily toward the population average. Some remain very sick and expensive.
The trap: selecting people at a recent peak, then crediting every subsequent decline to your intervention.
The remedy: use an appropriate comparison group, multiple baseline periods, consistent follow-up and a clear treatment start. None eliminates all bias automatically.
As the Camden lesson shows, “after treatment” does not by itself mean “because of treatment.”
Every year has high spenders. They are not always the same people.
Picture 10,000 people lined up by annual spending.
A small group at the expensive end accounts for a large share of the bills. Repeat the lineup next year: the picture can look similar, even though many people have changed places.
Medicare FFS, 2023 data5% → 45%The costliest 5% accounted for 45% of annual spending. MedPAC, July 2026.
VA + Medicare persistence study32% → 7%32% stayed high-cost in the first follow-up year; 7% stayed high-cost through all four follow-up years.
Rank change is not recovery. Death, coverage loss, intermittent high costs and changing needs can all change the group. The 7% is continuous persistence, not everyone who was high-cost in year four.
Better targeting question: Who is likely to have a costly, preventable event next, and which intervention can actually help?
Persistent high cost is not automatically avoidable cost. Necessary treatment can remain expensive despite excellent care.
Do not join two separate cohorts into one story: it can be a costly trap when strategically planning.
A one-year persistence estimate starting in 2011 and another starting in 2012 describe two starting groups. You cannot chain them into a two-year result for the original patients. A linked patient-level follow-up is needed.
Ignoring MSSP truncation can catastrophically overstate a program’s projected ACO return.
Start here · Published real-world example · Iris / Aledade, 2022
A large medical-cost difference became a much smaller ACO-level figure.
Headline medical-cost difference$994 PMPM$11,928 annualized per person
→
Reported after MSSP financial stop-loss$292 PMPM$3,504 annualized per person
Only 29.4% remained. The reported amount was 70.6% lower.
The authors reported an ACO-level adjustment, not an actual CMS payment. Annual figures above are derived ×12 illustrations; the report followed participants for an average of 14 months.
Read the original report and its limits
The 2022 author report describes a comparison group, 335 treated patients, and a financial stop-loss adjustment. It does not provide enough detail to independently reconstruct that adjustment or establish randomized allocation. Authors disclosed employment and financial interests in Aledade. Do not apply 29.4% as a universal factor or subtract costs already included in a total-cost result.
CMS limits extreme annualized beneficiary spending included in its financial calculations. Preventing dollars above that limit can lower real medical bills without lowering the ACO’s counted amount by the same number.
Now make the mechanism visible · Hypothetical, not the published study
They budgeted around $150,000 of medical savings. Only $35,000 survived the cap.
Imagine one beneficiary would cost $400,000 without the program, but $250,000 with it. Use an illustrative $285,000 CMS spending cap.
What the medical bills suggest$150,000$400,000 − $250,000Medical spending avoided
→
What the capped comparison recognizes$35,000$285,000 − $250,000Spending reduction after truncation
The missing $115,000 was above the cap. It can be real medical spending avoided, but it was not fully counted in this ACO spending comparison to begin with.
Same full-year beneficiary, one enrollment type, nonnegative included expenditures, common cap. The $285,000 cap is invented for teaching. The $35,000 is still not the shared-savings check.
For PY2026, use Methodology v14, §§3.1.1–3.1.5. Select the year/enrollment-type threshold, annualize included beneficiary spending, apply the threshold, completion and person-time weighting, then aggregate. Do not cap service categories or a whole cohort total separately. The program’s actual payment still depends on the entire ACO’s benchmark and reconciliation.
The same care can change two budgets in different ways.
Delegated MA · contract dependent
Delegated / capitated revenue − costs inside the delegated risk pool − program operating cost = delegated surplus
MSSP · fee-for-service ACO
Updated benchmark − included, adjusted A/B expenditures = gross benchmark difference → full CMS reconciliation and sharing
Part D is the pharmacy trap: MA-PD includes prescription-drug coverage, but the physician group’s contract decides whether those costs affect its pool. Standard MSSP performance spending is Parts A+B, not Part D.
The plan’s financial responsibilities and a physician group’s delegated contract are not the same thing. Inspect the actual contract before transporting an MA result.
Difference to check
MA / MA-PD
MSSP
Part D prescriptions
MA-PD plans include drug coverage with Part D financing. Whether pharmacy costs affect a provider’s risk pool depends on its contract.
Part D prescription spending is not directly included in the A/B benchmark and performance spending. Medication use can still change A/B utilization.
Part B-covered drugs
Some outpatient drugs are medical benefits, not Part D prescriptions. Check the delegated medical-cost definition.
Included Part B drug claims remain in A/B spending, even though Part D prescriptions do not. “Part D excluded” does not mean “all drug spending excluded.”
Hospice
Original Medicare generally pays the hospice benefit for an MA enrollee. Confirm which remaining expenses and payments enter the delegated contract.
Included hospice claim payments count in A/B expenditures. Hospital reductions and hospice increases must be considered together.
Payment and population
Plan enrollment and negotiated payment/risk arrangements.
ACO assignment is not enrollment in a separate insurance plan. FFS payments continue; shared savings are a separate calculation.
Extreme claims / risk adjustment
Contractual stop-loss, carve-outs, coding and payment rules can differ.
Use the applicable CMS methodology, including beneficiary-level expenditure truncation. Do not import an MA adjustment factor.
Model all included reductions minus all included increases. Hospital ↓ and SNF ↓ can coexist with hospice ↑, home health ↑ and Part B care management ↑.
Seeing Part D data in CCLF7 does not mean those pharmacy payments belong in the MSSP A/B spending total.
Another difference: program FFS collections can reduce an organization’s required subsidy, while Medicare-paid claims can add to measured A/B expenditures. Keep those ledgers separate.
Meeting-ready conclusion: “Use the studies to identify a promising approach. Use your population, intervention and comparator to estimate the likely local effect.”
Appropriate eligibility, informed choice and patient goals come first. Hospice enrollment is not a savings quota.
Before asking “Why did it fall?” ask “Are these the same measurement?”
Same numbers, different scales
$262.75 PMPM = $3,153 PMPY = $3,153,000 per 1,000 person-years.
0.72 admissions per person-year = 720 per 1,000 person-years.
Same label, different populations
Assigned beneficiaries, enrolled participants, engaged patients, survivors and completers are not interchangeable denominators.
SOURCE · POPULATION · PERIOD · UNIT · DENOMINATOR
Watch for: immature recent claims compared with final claims; a monthly result compared with a cumulative result; different risk or service definitions; deaths or coverage loss treated as improvement; coding or benefit changes.
CCLF, an analytics platform and a CMS expenditure report can differ legitimately. Reconcile the methods and explain the remaining difference instead of assuming exact totals must match.
Assigned ≠ assignable. ACO financial assignment is also different from the statistical phrase “intervention-attributable effect.” Say “intervention-caused savings in assigned beneficiaries” when that is what you mean.
Individual opportunities can mature. The team can still get better.
1 · Initial catch-up
Close the neglected gaps.
A newly supported panel may have access, medication, transition or serious-illness needs that have not been addressed.
2 · Maintenance & renewal
Protect gains. Find new risk.
New discharges, illnesses and crises create new opportunities. Yesterday’s stabilized patients may need less intensive support.
3 · Operating improvement
Make the work more reliable.
Better targeting, access, follow-up, referral habits, staffing and billing can improve delivery.
Planning framework, not a proven causal savings decomposition or a year-by-year forecast.
The levers to consider
Acute utilization, post-acute pathways, serious-illness care, site of service, chronic-care support, clinician learning, team infrastructure and population composition.
Match intensity to need: broad reliable primary-care processes can serve everyone; expensive interventions need a well-defined, impactable target.
CMS is a map, not a guarantee. A reimbursed service can improve care without necessarily lowering total spending in every population. Payment availability does not substitute for a local business case.
Do not confuse selection with improved care. A changing roster or patient mix can improve an average even if the care process itself is unchanged.
Do not silently turn a historical finding into a current dollar promise.
Keep the published result
Show the original dollar amount, price year, data period, population and study design. Publication year and dollar year may differ.
Then add a separate illustration
Identify the chosen index, source, start value, end value and target month/year.
Illustrative updated dollars = original dollars × target index / original index
Medical CPI measures a particular price basket; it is not a Medicare payment schedule. Inflation adjustment does not update clinical effectiveness, care patterns, coding, risk adjustment or MSSP policy.
Do not label an inflation-adjusted study estimate “expected current MSSP savings.”
No current-dollar conversion is supplied here because a defensible calculation requires verification of the original price year and the selected index.
