Industries / Health / Health insurers and PBMs
Plans and enrollment
Designing the cover people can buy, setting who may join and at what price, and moving money between insurers so that one that takes sicker members is not punished for it.
7
Milestones achieved, from 1996 to 2024.
1
To watch for: announced, not yet confirmed outside.
2
Open deliverables. Nobody has shipped them.
The road
AchievedWatch forOpen
Achieved
1996
Everyone must buy cover
Switzerland's Federal Health Insurance Act (KVG) took effect on 1 January 1996. Basic health insurance is compulsory and is sold by about 50 non-profit insurers, which must accept every person required to insure in their area and treat everyone alike whatever their health; premiums do not depend on income and differ by insurer, place of residence and the form of insurance chosen, by the Federal Office of Public Health's account. Under the old law, premiums graded by age at joining, age limits and permitted exclusions meant that mostly young, healthy people could switch insurers, and costly members piled up at some funds. An emergency federal decree brought in risk equalization, payments from insurers with healthier members to those with sicker ones, from 1 January 1993, and the KVG made it ordinary law; it used only age and sex at first, and added stays in hospital or a nursing home from 2012.
Lessonletting people switch insurers freely only works with a way to move money toward the insurers that take sicker members.
Credit
Swiss Federal Assembly; Gemeinsame Einrichtung KVG, the joint body that runs the equalization
2000
A price for joining late
Australia's Lifetime Health Cover started on 1 July 2000. A person who first takes out private hospital cover after the 1 July following their 31st birthday pays 2% more for each year they are over 30, up to 70%, and the extra stops after 10 years of continuous cover, by the government's own guide. Together with a 30% premium rebate from the same period, it raised the share of Australians with private cover from just over 30% in 1998 to just under 50% by the end of 2000, mostly among the richest fifth of the population, by the count of Walker, Percival, Thurecht and Pearse (2005).
LessonRyan found in 2024 that by 2021 the insured pool was as tilted toward older people as before the policy, because the large group aged 45 to 59 who joined in 2000 and 2001 kept their cover as they aged.
Credit
Australian Government; evaluations by Walker and colleagues and by J. B. Ryan
2006
One compulsory private policy
The Netherlands' Health Insurance Act (Zorgverzekeringswet) took effect on 1 January 2006. Everyone who lives or works in the country must buy the standard policy from a private insurer; insurers must accept every applicant and charge everyone the same premium for a policy, whatever their age or health; and a risk-equalization contribution moves money between insurers. Members may change insurer each year: 7.0% did for 2025 and 6.4% for 2026, by the count of the Dutch Healthcare Authority (NZa), which links the drop to a small premium rise.
Lessonthe equalization model keeps being retuned; the NZa wrote in May 2025 that premium gaps had narrowed, partly because of a statistical method added to the model called constrained regression, and that insurers' pull toward young, healthy members seemed to have weakened.
government.nlwetten.overheid.nlzoek.officielebekendmakingen.nlzoek.officielebekendmakingen.nl
Credit
Dutch government and parliament; NZa
2009
Funds paid by diagnosis
Germany's statutory sickness funds have shared costs through risk structure equalization (Risikostrukturausgleich) since 1994, at first using only age, sex and reduced earning capacity. From 1 January 2009 a central health fund has paid each sickness fund an amount per member that also reflects the member's recorded illnesses, the morbidity-based equalization (Morbi-RSA); until 2020 it counted 50 to 80 selected diseases, about a quarter of possible diagnosis codes. A law of 2020 made it count all diseases from 2021, added a pool for members with very high costs, and set out to make the system harder to manipulate.
Lessononce payments followed recorded illness, protecting the system from manipulation became a stated goal of reform.
Credit
German Bundestag; the Federal Office for Social Security (BAS), which runs it and says it paid out about €350 billion to the funds in 2025
2014
Health no longer sets premiums
For US cover starting on or after 1 January 2014, the Affordable Care Act lets insurers in the individual and small-group markets vary premiums only by age (at most 3 to 1 for adults), tobacco use (at most 1.5 to 1), family size and geography, by CMS's summary. A risk adjustment program moves money from insurers with healthier members to those with sicker ones: for 2024, 592 insurers took part and transfers equalled about 8.9% of total premiums, 11.0% in the individual market and 3.9% in the small-group market, by CMS's report.
Credit
US Congress; CMS, which runs the program
2022
Court voids price tables
Chile's private health insurers, the ISAPREs, set the price of each member's plan using "factor tables", multipliers that change the price for each person covered. The Constitutional Court struck down the legal basis of the tables then in use in 2010, as the Supreme Court has since recalled, and in late 2022 the Supreme Court ordered the ISAPREs to apply the single table set by the regulator, in force since 1 April 2020, and to return what they had charged above it. A later law, No. 21.674, set up repayment plans of up to 13 years. By 30 June 2026 the Superintendency of Health put the total owed at about 1.2 trillion Chilean pesos, of which 21.2% had been paid into members' accounts.
Lessona pricing rule left in use for a decade after its legal basis fell can come back as a debt that takes insurers more than a decade to repay.
Credit
Supreme Court of Chile; Superintendencia de Salud
2024
Risk model rebuilt for coding
Medicare Advantage plans, private plans that run Medicare benefits, cover more than half of eligible beneficiaries since 2023, and 55% (35.2 million people) in 2026, by KFF's count. They are paid more for members with more recorded diagnoses. From 2024 to 2026 CMS phased in a new risk model, Version 28 (V28), meant to shrink the gap between how fully these plans and traditional Medicare record diagnoses. MedPAC, the commission that advises Congress on Medicare, estimated that coding made risk scores about 14% higher in 2024 than for similar people in traditional Medicare, projected about 10% for 2026, and said V28 "is having the intended effect". It still put 2026 payments to the plans at 14% above traditional Medicare, or $76 billion, of which it attributes $57 billion to favorable selection: plans enrolling people whose risk scores overstate their costs.
Lessonpaying by diagnosis rewards recording diagnoses, and once coding was reined in, selection became the larger cost.
Credit
CMS; MedPAC for the measurement
Watch for
Ten million more uninsured
The Congressional Budget Office (CBO) estimates that the budget reconciliation law signed on 4 July 2025 will raise the number of people without health insurance in 2034 by 10 million: 7.5 million from changes to Medicaid, 2.1 million from changes to the ACA Marketplaces and 0.4 million from other provisions and interactions. Combined with the scheduled end of the enhanced Marketplace premium tax credits, CBO's estimates show a rise of more than 14 million.
Achieved whenthe Census Bureau's or CDC's national surveys record a rise in the uninsured of that order as the provisions take effect.
Claimed by
CBO, as compiled by KFF, August 2025
Open
01
One patient, four risk formulas
An open calculator that takes an invented member (age, sex, region, diagnoses, prescriptions and hospital stays) and works out what four risk-equalization formulas would pay that member's insurer for a year: Germany's Morbi-RSA, the Dutch equalization under the Health Insurance Act, Switzerland's equalization, and the US HHS-HCC model for Marketplace plans. Run it on 200 invented members, from healthy 25-year-olds to people with several chronic illnesses, and publish where the formulas agree and where they differ, as a share of each country's average payment.
The briefCredit · unclaimed02
Selection spirals, backtested
An open simulation of an insurance market in which invented people decide each year whether to buy cover and which policy, and insurers set next year's prices from this year's claims, under rules such as community rating, late-joining penalties and risk equalization. Calibrate it only on public aggregate data and test it against three recorded histories: Australian hospital cover around the 2000 Lifetime Health Cover change, the US individual market from 2014 to 2018, and the Dutch shift to restricted "budget" policies from 2021 to 2025.
The briefCredit · unclaimed
Open 01 · the brief
One patient, four risk formulas
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
An open calculator that takes an invented member (age, sex, region, diagnoses, prescriptions and hospital stays) and works out what four risk-equalization formulas would pay that member's insurer for a year: Germany's Morbi-RSA, the Dutch equalization under the Health Insurance Act, Switzerland's equalization, and the US HHS-HCC model for Marketplace plans. Run it on 200 invented members, from healthy 25-year-olds to people with several chronic illnesses, and publish where the formulas agree and where they differ, as a share of each country's average payment.
Why it is still open
Each regulator publishes its own formula and pays insurers with it, but comparisons stop at insurers or whole countries. Douven and colleagues compared five countries in 2022 using insurer-level data. A search in November 2026 found no tool that runs the same person through several countries' formulas.
Built with what exists
BAS's yearly decisions on the German classification model, risk groups and weights; Gemeinsame Einrichtung KVG's documents on the Swiss equalization, including its drug-based cost groups (PCG); the equalization chapter of the Dutch Health Insurance Act and the yearly Dutch equalization rules; CMS's HHS-HCC software, which CMS now offers in Python as well as SAS; Synthea for invented medical histories; an AI coding agent checked only against hand-worked cases, as AWS checked an emulator its agents wrote (see the Note "Code no person read").
bundesamtsozialesicherung.dekvg.orgwetten.overheid.nlcms.gov
First step
Implement the Swiss formula, which has the fewest inputs (age, sex, region, prior hospital stay and drug cost group), for 20 invented members.
Done looks like
Four formulas implemented for one recent year, each passing at least 30 hand-worked cases built from that regulator's published tables and checked by a second person; 200 invented members with all four payments; a table of the 20 members on whom the formulas differ most; code and data under an open license.
Open 02 · the brief
Selection spirals, backtested
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
An open simulation of an insurance market in which invented people decide each year whether to buy cover and which policy, and insurers set next year's prices from this year's claims, under rules such as community rating, late-joining penalties and risk equalization. Calibrate it only on public aggregate data and test it against three recorded histories: Australian hospital cover around the 2000 Lifetime Health Cover change, the US individual market from 2014 to 2018, and the Dutch shift to restricted "budget" policies from 2021 to 2025.
Why it is still open
Market simulators exist as teaching tools, not as tested forecasts. The Society of Actuaries published Alan Mills's agent-based models in 2013, one of them a model of adverse selection on the US exchanges, with its code, but without a test against what happened. A search in November 2026 found no open model of this kind checked against a real market's recorded enrollment and premiums.
Built with what exists
APRA's quarterly private health insurance statistics; CMS's Marketplace open enrollment files from 2014 and its rate files; the NZa's reports on Dutch policies and switching; Python with Mesa, an open library for agent-based models; AI models to draft code and tests that people review.
First step
Rebuild Australian hospital cover by age group for the years around 2000 from published statistics, and test whether a simple model with a 2% a year loading reproduces the rise from just over 30% to just under 50%.
Done looks like
For each of the three histories, the model, calibrated only on data from before the change, predicts the share insured (or the share on budget policies) and the average premium for the next four years within 10% of the recorded figures, with code, inputs and every run published.