Improving risk adjustment without diagnostic information – Evidence from reforms in Switzerland
This study demonstrates that Switzerland's risk adjustment scheme can be significantly improved by incorporating morbidity information derived from drug prescriptions, as evidenced by the transition from a cost-based indicator to pharmaceutical cost groups, which enhanced predictive accuracy and reduced compensation errors despite the absence of diagnostic data.
Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer
In the world of health insurance, a fundamental tension exists between fairness and efficiency. Insurance companies collect money from many people to pay for the medical care of a few who get sick. If an insurer could easily tell who is likely to get very sick and very expensive to treat, they might try to avoid signing those people up, leaving them to struggle with higher costs or no coverage at all. To stop this, many countries use a system called risk adjustment. Think of it as a financial balancing act where the government collects money from insurers who have mostly healthy customers and gives it to those who have sicker, more expensive customers. This ensures that every insurer has a fair chance to compete, regardless of who they happen to sign up. The challenge lies in predicting who will be expensive. Most countries use detailed medical records to make these predictions, but in Switzerland, strict privacy laws prevent insurers from seeing these specific diagnoses. Without knowing exactly what a person is sick with, insurers have had to guess based on how much they spent on medicine, a method that is often too blunt to be truly fair.
A team of researchers from Zurich and Lucerne decided to examine how Switzerland has tried to solve this puzzle without using medical diagnoses. They looked at a major shift in how the country calculates these payments, moving from a simple rule based on total drug costs to a more sophisticated system that groups medicines by how they are used. The researchers analyzed data from nearly seven million adults over several years to see which method worked better at predicting future costs and, more importantly, which method stopped insurers from trying to game the system. Their work reveals that while the new method is much better at understanding who needs help, it has also created a few new, subtle ways for insurers to pick and choose their customers, suggesting that the search for a perfect system is ongoing.
For decades, the Swiss system relied on basic information like age and sex, and later added a simple rule: if a person spent more than 5,000 Swiss francs on drugs in a year, they were flagged as high-risk. This was a cost-based indicator. It was easy to use, but it had a flaw. Because the payment jumped up once a person crossed that spending line, insurers had a strange incentive to make sure their customers crossed it. They might encourage people to buy more expensive brand-name drugs instead of cheaper generic ones just to push the total cost over the threshold, ensuring the insurer got a larger payment. This was inefficient and did not necessarily reflect the true health needs of the patient.
To fix this, Switzerland introduced a new system in 2020 based on pharmaceutical cost groups. Instead of looking at the total price tag, this system looks at the types of medicines people take and how many doses they consume. It groups specific medicines together that are known to treat chronic, long-term conditions. For example, taking a certain number of doses of a specific heart medication might place a person in a "heart disease" group, regardless of whether that medication cost a lot or a little. This is a utilization-based indicator. The researchers found that this new approach was a significant improvement. It predicted future spending much more accurately than the old cost-based rule. When they compared the two methods, the new system reduced the number of people who were either paid too much or too little by a large margin. It also stopped the strange incentive to push people over a spending limit, because the payment was now based on the type of medicine, not the total price.
However, the story does not end with a perfect solution. The researchers discovered that while the new system was better overall, it was not flawless. When they looked closely at the specific groups of medicines, they found that some groups were still very unpredictable. For instance, people grouped under "kidney diseases" showed a wide range of actual costs; some were very expensive, while others were relatively cheap, even though they were all in the same payment group. This variation creates a new, quieter opportunity for risk selection. An insurer might look at a person with kidney disease and, based on other clues, guess that this specific person will be cheaper to treat than the average person in that group. If they are right, the insurer makes a profit. The study showed that while the new system reduced the ability to pick and choose customers based on broad categories, it did not eliminate the ability to do so within those categories.
The impact of these changes was visible in the insurance market itself. Before the reforms, insurance premiums varied widely across the country, with some insurers charging much more than others because they were afraid of signing up sick people. After the new system was introduced, the prices across different insurers began to converge. Insurers who had mostly healthy customers raised their prices more than those with sicker customers, suggesting that the fear of being stuck with high costs was fading. The variation in prices across different regions dropped significantly, indicating that the market was becoming more stable and fair.
The researchers concluded that even without access to detailed medical diagnoses, a health insurance system can be made much fairer by using smart, utilization-based indicators. The shift from a simple spending threshold to a system based on drug groups was a clear step forward, improving the accuracy of payments and reducing the temptation for insurers to manipulate costs. Yet, the study also serves as a reminder that no system is perfect. The remaining gaps in prediction mean that insurers still have some room to try and select the healthiest customers within the same risk groups. The Swiss experience offers a valuable lesson for other countries: you can build a robust safety net without invasive medical data, but the work of refining that net never truly stops.
Drowning in papers in your field?
Get daily digests of the most novel papers matching your research keywords — with technical summaries, in your language.