Retrospective Risk Adjustment

Retrospective Risk Adjustment in Health Plan Payment

Retrospective risk adjustment settles what a health plan or provider is finally paid for a coverage period once that period has closed and its diagnosis and cost records are complete. The payer starts with a provisional amount, then recomputes risk scores or shares part of the realised costs, and pays or recovers the difference. The design matters in health insurance markets financed by capitation or community-rated premiums, because the final settlement decides how far a plan that enrols sick people is compensated and how much of any cost saving the plan keeps. Two current United States examples are the final reconciliation of Medicare Advantage risk scores and the Affordable Care Act (ACA) transfers between insurers in the individual and small group markets.

Two meanings of retrospective in risk adjustment

The word retrospective is used for two different design choices. Ellis uses it as another name for concurrent prediction from same-year diagnoses, a sense covered under concurrent risk adjustment. In the second sense it means settlement after the period, which a concurrent model needs and which Medicare Advantage also applies to its prospective risk adjustment model.

A narrower usage comes from United States plan operations. The HHS Office of Inspector General describes chart reviews as retrospective reviews of medical record documentation by Medicare Advantage organisations, used to add diagnoses not previously submitted or to delete diagnoses submitted in error. Such reviews feed the settlement data without changing the payment formula.

Three ways payments are settled after the period

Payers correct payments after the period in three broad ways, often combined. The trigger for the correction is either new information about expected cost or the cost actually incurred. Only the second kind ties revenue directly to spending.

Recalculating risk scores with complete data

In Medicare Advantage, federal regulation (42 CFR 422.310) bases each payment year's risk factors on data for services in a 12-month period before that year. After the payment year is completed, CMS recalculates the risk factors to see whether payment adjustments are necessary, and a reconciliation process accepts data until a final deadline announced by CMS, no earlier than 31 January of the following year. After that deadline an organisation can submit data to correct overpayments but cannot submit diagnoses for additional payment. The model stays prospective, and only the data used to score it are completed late.

Concurrent transfers settled after the benefit year

The HHS-HCC model behind ACA risk adjustment is concurrent. Kautter and colleagues explain the choice by the absence of prior-year diagnoses at the 2014 launch and by people moving in and out of these markets. Because the scores need the whole benefit year's diagnoses, transfers are computed afterwards, and 45 CFR 153.310 requires issuers to be told their risk adjustment payments or charges by 30 June of the year following the benefit year. The CMS summary report for the 2025 benefit year, released on 30 June 2026, describes funds moving towards plans with higher-than-average risk enrolees, with the absolute value of state transfers about 9.0% of total premiums nationally, excluding the high-cost risk pool.

Risk sharing on actual costs

Van Kleef and colleagues separate risk equalisation, which redistributes funds among insurers using indicators of expected cost such as age and health, from risk sharing, which redistributes funds based on actual cost. Outlier pooling, proportional sharing and high-cost pools all belong to the second group. The ACA scheme includes one such element. For the 2025 benefit year its high-cost risk pool reimbursed issuers for 60% of an enrolee's paid claims above USD 1 million, funded by a charge set as a share of premium on all covered plans. A stop-loss contract in private reinsurance follows the same logic.

Payment formula and the cost control trade-off

A retrospective cost element can be written as a weighted mix of a risk-adjusted amount and the cost the plan actually incurred. Geruso and McGuire use this form to compare payment systems on fit, the share of cost variation that payments explain, and power, the share of costs at the margin borne by the plan.

$$P_i = \alpha \hat{C}_i + (1 - \alpha) C_i$$

where $P_i$ is the payment for member $i$, $\hat{C}_i$ is the risk-adjusted payment from the model, $C_i$ is the member's realised cost and $\alpha$ is the weight on the risk-adjusted part, between 0 and 1.

Power equals $\alpha$ only when the risk-adjusted part does not move with the plan's own costs, as in the pure mixed system of Geruso and McGuire, which pays the population mean. They show that fit for this payment is given by the following expression.

$$\text{Fit} = 1 - \alpha^2 \left(1 - R^2_{RA}\right)$$

where $R^2_{RA}$ is the share of cost variance explained by the risk adjustment model on its own. In their example, a model with $R^2_{RA} = 0.10$ and $\alpha = 0.5$ gives a fit of 1 − 0.25 × 0.90 = 0.775. Moving weight towards actual costs raises fit and lowers power. A payment equal to realised cost has a fit of one and a power of zero, which is cost-based reimbursement.

An outlier or high-cost scheme applies the retrospective weight only above a threshold.

$$S_i = s \times \max\left(0,\; C_i - T\right)$$

where $S_i$ is the risk-sharing payment for member $i$, $s$ is the share of excess cost reimbursed and $T$ is the threshold. For a member above the threshold the plan bears one minus $s$ of each extra unit of cost, so power on that margin is also one minus $s$, while members below the threshold are unaffected.

Van Kleef and colleagues state the price of this protection directly, since ex post cost-based payments reduce insurers' incentives for cost containment and prevention. Van Barneveld and colleagues show how the best extent of risk sharing may depend on the weights a regulator gives to selection and to efficiency. Geruso and McGuire add that risk adjustment built on diagnoses from claims also has less than full power, especially when the diagnoses come from the same period.

Worked example with a year-end settlement for one plan

All numbers in this example are illustrative. A plan has 1,000 members and receives a base rate of GBP 3,000 per member per year, scaled by its average risk score. The payer also reimburses 60% of each member's annual cost above GBP 100,000.

1. Provisional payment. Scores calculated before the year from incomplete data average 0.98, so the plan receives 1,000 × 3,000 × 0.98 = GBP 2,940,000 during the year.

2. Score reconciliation. After the year, diagnoses submitted before the final deadline raise the average score to 1.02. The final risk-adjusted payment is 1,000 × 3,000 × 1.02 = GBP 3,060,000, so the reconciliation pays 3,060,000 − 2,940,000 = GBP 120,000.

3. High-cost sharing. Two members cost GBP 180,000 and GBP 130,000. Their excess costs are 180,000 − 100,000 = 80,000 and 130,000 − 100,000 = 30,000, a total of GBP 110,000, and the sharing payment is 0.6 × 110,000 = GBP 66,000.

4. Final position. Total revenue is 3,060,000 + 66,000 = GBP 3,126,000. With actual costs of GBP 3,150,000, the result is 3,126,000 − 3,150,000 = −24,000, a loss of GBP 24,000.

5. Power at the margin. Had care management cut the GBP 180,000 member's cost by GBP 10,000, the sharing payment would fall by 0.6 × 10,000 = GBP 6,000, so the plan keeps 10,000 − 6,000 = GBP 4,000, a power of 4,000 / 10,000 = 0.4 on that margin. Below the threshold the saving is kept in full unless a scored diagnosis is lost.

6. Full reconciliation to cost. If the payer instead settled each member at observed cost, as a rule of "payment adjustment equals observed minus predicted cost" would do, revenue would equal the GBP 3,150,000 of costs whatever the plan did, and power would be zero.

Step 2 corrects the estimate of expected cost, while step 3 moves part of the realised cost back to the payer. In a scheme funded by charges on all plans, as the ACA high-cost pool is, the GBP 66,000 comes from charges on all plans, this one included, and not from new money.

Use in research and budget analysis

Retrospective adjustment also appears outside payment, when analysts compare provider or care-model costs after the fact using same-period diagnoses from claims data. Ellis notes that a concurrent framework is used for quality measurement and for normalising performance measures, a form of case-mix adjustment. In an evaluation, diagnoses recorded after an intervention starts include its complications, so adjusting for them can remove part of the effect being measured, and baseline risk is better taken from a period before the intervention. A model of the budget impact of a scheme with retrospective elements needs the final settled amounts, not the provisional payments.

Boundaries with neighbouring payment ideas

Several terms describe parts of the same payment chain, and each answers a different question about timing or cost. The table places them side by side, and the risk adjustment page covers the general method.

Risk pooling, the spreading of individual losses across a group, is the broader idea, and retrospective risk sharing applies it between plans. A diagnosis-based risk score is the input that a reconciliation recalculates, not the settlement itself.

Misreadings and limits

Most problems with retrospective schemes come from reading greater accuracy as greater fairness. Others come from forgetting how and when the data arrive.

1. Equating observed minus predicted cost with risk adjustment. That rule pays realised cost and removes the incentive to control spending. A risk-adjusted payment compensates predicted cost.
2. Assuming settlement after the year means a concurrent model. Medicare Advantage settles retrospectively on a prospective model.
3. Treating higher fit as a better system. Geruso and McGuire show that risk adjustment aimed at discouraging cream skimming weakens cost control incentives in practice.
4. Overlooking coding. A reconciliation window pays for any diagnosis added before the deadline, whether it completes the record or reflects more intensive coding. The Office of Inspector General found that over 99% of Medicare Advantage chart reviews in its study added diagnoses, and estimated USD 6.7 billion in 2017 risk-adjusted payments from diagnoses reported only on chart reviews.
5. Reading a transfer as a saving. The American Academy of Actuaries notes that risk adjustment is typically budget neutral, so its net effect is a movement of money among participating organisations, not a change in total payments.

Sources

• American Academy of Actuaries. Risk Assessment and Risk Adjustment. Issue brief, May 2010. https://www.actuary.org/wp-content/uploads/2017/11/Risk_Adjustment_Issue_Brief_Final_5-26-10.pdf
• Centers for Medicare & Medicaid Services, CCIIO. Summary Report on Individual and Small Group Market Risk Adjustment Transfers for the 2025 Benefit Year. 30 June 2026. https://www.cms.gov/cciio/programs-and-initiatives/premium-stabilization-programs/downloads/ra-report-by2025.pdf
• Code of Federal Regulations. 42 CFR 422.310(g), Risk adjustment data. Current text read 28 October 2026. https://www.ecfr.gov/current/title-42/chapter-IV/subchapter-B/part-422/subpart-G/section-422.310
• Code of Federal Regulations. 45 CFR 153.310(e), Risk adjustment administration. Current text read 28 October 2026. https://www.ecfr.gov/current/title-45/subtitle-A/subchapter-B/part-153/subpart-D/section-153.310
• Ellis RP. Risk adjustment in health care markets: concepts and applications. In: Lu M, Jonsson E, eds. Paying for Health Care: New Ideas for a Changing Society. Wiley-VCH; 2008. Author manuscript, February 2007. https://people.bu.edu/ellisrp/EllisPapers/2007_Ellis_Riskadjustment25.pdf
• Ellis RP. Risk adjustment. In: The New Palgrave Dictionary of Economics, online edition. 2011. Author manuscript, 19 October 2011. https://blogs.bu.edu/ellisrp/files/2012/01/Ellis_Palgrave-Dictionary-of-Economics_Risk-Adjustment_20111019.pdf
• Geruso M, McGuire TG. Tradeoffs in the design of health plan payment systems: fit, power and balance. Journal of Health Economics. 2016;47:1-19. https://doi.org/10.1016/j.jhealeco.2016.01.007
• Kautter J, Pope GC, Ingber M, et al. The HHS-HCC risk adjustment model for individual and small group markets under the Affordable Care Act. Medicare & Medicaid Research Review. 2014;4(3). https://doi.org/10.5600/mmrr.004.03.a03
• Office of Inspector General, HHS. Billions in Estimated Medicare Advantage Payments From Chart Reviews Raise Concerns. OEI-03-17-00470, December 2019. https://oig.hhs.gov/oei/reports/oei-03-17-00470.pdf
• van Barneveld EM, Lamers LM, van Vliet RCJA, van de Ven WPMM. Risk sharing as a supplement to imperfect capitation: a tradeoff between selection and efficiency. Journal of Health Economics. 2001;20(2):147-168. https://doi.org/10.1016/S0167-6296(00)00077-1
• van Kleef RC, Reuser M, Stam PJA, van de Ven WPMM. A framework for ex-ante evaluation of the potential effects of risk equalization and risk sharing in health insurance markets with regulated competition. Health Economics Review. 2024;14:57. https://doi.org/10.1186/s13561-024-00540-4