Showing evidence of risk selection requires showing evidence of “actions with the goal and/or the effect that the cross-subsidies as intended by the regulator are not fully achieved”. Because it is hard to show evidence of the
goal of certain actions, we will first concentrate on showing evidence of the
effects of actions. In doing so, we first restrict ourselves to showing evidence that the cross-subsidies as intended by the regulator are not fully achieved. We discuss two methods of estimating risk selection (“
Residual expenses” and “
Overrepresentation of over- or undercompensated groups”). First, we discuss residual expenses as an estimate of risk selection (“
Residual expenses”). Second, the level of overrepresentation of over- or undercompensated groups per insurer or health plan is discussed as an estimate of risk selection (“
Overrepresentation of over- or undercompensated groups”). Third, we provide a list of signals of selection that can be measured and that, in particular in combination, can show evidence of risk selection (“
Signals of risk selection”).
Residual expenses
Showing evidence that “the cross-subsidies as intended by the regulator are not fully achieved” requires that it is known what these intended cross-subsidies are. These intended cross-subsidies
2 can be derived from (1) the risk equalization payments per insured and (2) the restrictions on the premium rates. For example, in the case of community ratings per health plan an identical risk distribution across the health plans is implicitly assumed to yield the cross-subsidies as intended by the regulator. In most European countries the equalization payment per individual equals the risk-adjusted
predicted expenses for that individual minus
p % of the overall average expenses per person,
3 with, e.g.,
p = 0 in Israel,
p = 50 in the Netherlands, and
p = 100 in Switzerland.
4 In addition, the insurer may charge the insured a community-rated premium reflecting the insurer’s efficiency.
5 This implies that in most European countries the cross-subsidies as intended by the regulator are such that the ‘residual expenses (i.e., actual expenses minus risk-adjusted predicted expenses) in the case of perfect risk equalization’ for each insured
in expectation are zero, assuming average efficiency.
“Cross-subsidies such that the residual expenses on each insured
in expectation are zero” imply that ex-ante the statistically
expected/
predicted residual expenses are zero for each insured. Because the unpredictable variation in individual residual expenses is large, ex-post there will always be a large variation in the actual residual expenses per
individual insured, even with perfect risk equalization. Therefore, showing evidence of risk selection cannot be done on the basis of one individual insured who ex-post (by accident) has extreme positive or negative residual expenses. Showing evidence of risk selection requires a sufficiently large number of individuals and can only be done with a certain level of statistical significance.
6
One could be inclined to measure risk selection by calculating for each insurer the average residual expenses of its insured.
7 The intuitive idea to do so is that if an insurer has an overrepresentation of overcompensated insured, this insurer will have, after risk equalization, lower-than-average residual expenses per insured and vice versa. The conclusion could then be that, if these average residual expenses are different from zero for at least one insurer, with a certain level of statistical significance, there is risk selection because at least one insurer is over- or undercompensated and thus the cross-subsidies as intended by the regulator are not fully achieved.
8 However, in most cases this conclusion is incorrect and the measure of risk selection is biased, as will be argued below.
Biased estimates of selection because of differences in insurers’ efficiency
Because the insurers’ average residual expenses are influenced by both selection and the insurers’ efficiency, these average residual expenses are biased estimates of risk selection if they are not adjusted for the differences in insurers’ efficiency. For example, negative average residual expenses per insurer can be the consequence of (1) being more efficient than average (and no risk selection), and/or (2) risk selection (and having average efficiency). Therefore, it is important to take care that the measure of risk selection is not biased by the insurer’s efficiency.
9
Insurers’ efficiency has two components: (1) efficiency at the insurer level, i.e., the insurer provides healthcare efficiently or has selectively contracted efficient providers; and (2) efficiency at the insured level, i.e., the insured have a preference for efficiency.
10 These two components can go together, but not necessarily. For example, an insurer with average efficiency but with an effective marketing campaign in creating a reputation of ‘delivering (or contracting) efficient and appropriate and no unnecessary care’ may attract many insured who prefer to make use of healthcare services in an efficient way and who avoid unnecessary care. Consequently, this insurer will have lower than average expenses within the risk groups used for the risk equalization. Although this insurer has a selective risk composition of insured, there is no risk selection because this situation could also occur in the case of perfect risk equalization,
11 when the cross-subsidies as intended by the regulator are fully achieved.
Underestimation because positive and negative selection effects cancel out
Even if the average residual expenses per insurer would be adjusted for the differences in efficiency among insurers, they may underestimate the true risk selection. One reason is that several forms of risk selection, both positive and negative, may occur simultaneously; therefore, positive and negative selection effects may cancel out. For example, an insurer may have an overrepresentation of selected groups of undercompensated insured (e.g., due to offering the best care for the chronically ill) as well as an overrepresentation of selected groups of overcompensated insured (e.g., due to selective advertising).
Underestimation because of selection within the insurers’ portfolio
A second reason why the insurers’ average residual expenses may be an underestimation of the true risk selection is related to the level of measurement. The risk equalization is mostly done at the level of risk-bearing insurers, while each insurer is often allowed to offer several health plans with different premium rates. Because risk selection may (often) take place at the health plan level, ideally, the average residual expenses should be measured at the health plan level and not at the insurer level. However, often the regulator and researchers only have access to expenditures data at the insurer level. If an insurer has one health plan with positive risk selection and another health plan with negative risk selection, the positive and negative selection effects may (partly) cancel out at the insurer level. In that case, the average residual expenses at the insurer level underestimate the true risk selection. In reality there may be serious market segmentation
within the insurer’s portfolio’, if the undercompensated insured choose a health plan with a high premium and the overcompensated insured choose a health plan with a low premium.
12
Underestimation because selection actions may be ‘unsuccessful’
A third reason why the insurers’ average residual expenses may be an underestimation of the true risk selection is because “actions with the goal” may not be successful and therefore not reflected in the insurers’ average residual expenses. For example, in the extreme, if all insurers are equally successful in risk selection and have an identical risk composition of their insured, the average residual expenses are zero for each insurer, apart from differences in efficiency. Nevertheless, there may be substantial risk selection, with all of its negative effects (e.g., distorting the quality of care).
Signals of risk selection
If the above-mentioned measures of risk selection indicate the existence of risk selection, this is sufficient for knowing that risk selection is indeed present. However, the reverse is not true. If the measures do not indicate that there is risk selection, this does not mean that risk selection is absent. Yet there still might be (substantial) selection (see “
Residual expenses”). Therefore, as a third method we provide a list of signals of selection that can be measured and that, in particular in combination, can show evidence of risk selection.
13 We do not pretend to present a complete list of all possible signals. We primarily focus on type-1 actions because these may have the most worrisome consequences (e.g., distorting the quality level of care). Measurement of these signals of risk selection is also useful in the case that the above-mentioned measures of risk selection indicate that risk selection is present, but it is still unknown which forms of risk selection this entails, and with which effects. It is important to note that the extent to which certain observed actions can (or cannot) be characterized as risk selection crucially depends on the quality of the risk equalization system. For example, actions with the effect of having only young insured can be considered as risk selection only if age is
not included as a risk adjuster in the risk equalization formula. Because in most countries the model for calculating the risk equalization payment is continuously improved, the measurement of (signals of) risk selection is a dynamic process. Actions that are currently considered as risk selection may no longer be risk selection after improving the risk equalization. Therefore, for the right interpretation of observed actions, it is necessary to know which selected groups of insured are over- or undercompensated, and to what extent (see, e.g., Table
2).
Health plan differentiation via contracting and delivering care
In different countries, insurers have different tools for contracting, organizing, managing, and delivering healthcare.
14 The use of each of these tools may result in health plan differentiation and market segmentation, as different (risk-)groups of insured prefer different health plans. Therefore, dependent on the quality of the risk equalization, the application of each of these tools can be considered as risk selection. For making a list of signals of such risk selection it is important to know (1) which selected groups of insured are over- or undercompensated, and (2) what degrees of freedom the insurers have in differentiating their health plans.
In several countries, insurers are free to negotiate with the providers of healthcare about the quality and price of healthcare, including the providers’ financial incentives (e.g., pay for performance, or risk sharing). Insurers and providers may agree on protocols for medical treatments and the level of efficiency of the care, i.e., the price-quality ratio, e.g., of implants, pharmaceuticals, medical devices, and diagnostic tests. Another degree of freedom is that insurers can selectively contract with preferred providers only, and can decide on the level of reimbursement in case an insured is treated by non-contracted providers. It is not difficult to imagine how these tools can be applied with the goal or the effect of market segmentation such that the over- and undercompensated people end up in different health plans with different premium rates.
The regulator can measure the following signals of risk selection. First, the regulator can monitor the quality of the contracted providers, the level of reimbursement of selected pharmaceuticals, the level of reimbursement of care received from selected non-contracted providers, and the rules for necessary pre-authorization.
A second option is that the regulator holds interviews with representative organized groups of undercompensated patients that negotiate with the insurers about price, quality, and the insurer’s purchasing strategy. The regulator could ask them questions such as: Which health plans do (not) allow you to have much influence on the quality of care, on the selected preferred providers, and on the composition of the supplementary insurances? Do you have the feeling of (not) being welcome with certain health plans?
A third option is that the regulator holds interviews with providers of care who specialize in treating chronic conditions of undercompensated patients. Questions could include: Is it difficult for you to get a contract with insurers? Which health plans in particular present a challenge? These answers can be compared with the answers by other providers.
Finally, the regulator can create opportunities for whistle-blowing, e.g., by employees of insurers who have ethical problems with their insurer’s policy and their own duties.
Health plan differentiation via service level
Health plan differentiation can also take place by differentiating the service levels of health plans, such as: having all contacts with the insured only via internet and email, rather than having an office building; the speed and quality of answering emails and phone calls; and advice to patients when the insurer acts as an intermediary for patients asking for guidance about the best providers or about waiting times.
Differentiation of service level can take place at a personal level. For example, based on administrative data such as costs and utilization from prior years, insurers may be able to qualify an insured as an over- or undercompensated risk type. If the insurer expects that a certain individual is overcompensated, the insurer may offer him/her short response times and excellent mediation when care is needed, and the opposite when the insurer expects that the individual is undercompensated.
The regulator can monitor these tools for risk selection by means of interviews with selected groups of insured or via ‘mystery’-insured: on the one hand, very healthy (overcompensated) persons, and on the other hand, unhealthy (undercompensated) patients with several chronic conditions.
Selective marketing, also by insurance agents
There are many ways that insurers can selectively market their health plans. In addition, many people do not buy their health plan directly from the insurer but via an insurance agent, i.e., a person or organization that advises and assists consumers regarding insurance products. Insurers often provide insurance agents with a bonus fee for each (new) applicant. Whereas insurers have to respect open enrollment, this generally does not apply to agents. Insurance agents can easily distinguish between over- and undercompensated individuals (e.g., just by observing and asking questions about health status) and use this information when channeling applicants to health plans.
The regulator can monitor this tool for risk selection by analyzing the marketing activities of all insurers and their insurance agents. In which media do they advertise? What is their marketing strategy? Who is the target group? What is the insurer’s image? Are over- and undercompensated people equally attracted by the marketing campaign? Do selected groups of consumers receive special (financial) benefits if they purchase a health plan, e.g., free supplementary insurance or rebates on other (insurance) products?
Selective enrollment and disenrollment
To measure signals of selective enrollment and disenrollment the regulator could submit so-called ‘mystery’-applications to insurers and insurance agents, and let ‘mystery’-insured ask for more information by letter, email, phone, and internet: this would seek to compare experiences from overcompensated (very healthy) persons and undercompensated persons (unhealthy patients with specific chronic conditions).
Another option is to hold interviews with insured consumers who switched insurers or health plans and ask them: Why did you switch? Were you not satisfied with the quality level of care that was delivered or contracted by your previous insurer or health plan? Did you have the feeling of not being welcome with your previous insurer? Did you have the feeling of being kicked out?
Supplementary insurances and other tie-in products
Supplementary insurances can also be an effective tool for risk selection. This holds true in particular if (1) health plans and supplementary insurances are (seemingly) sold as one product, and (2) no special regulation applies to supplementary insurances. The latter implies that insurers are free to require new applicants for supplementary insurances to fill out a health questionnaire, to reject applicants, and/or to charge risk-rated premiums for supplementary insurances. This is the case in, e.g., the Netherlands and Switzerland [
23]. The outcomes of a health questionnaire for supplementary insurance may help insurers to distinguish between applicants who are expected to be over- and undercompensated for regulated basic health insurance. By rejecting high-risk individuals for supplementary insurances (or by charging them excessive premiums for supplementary insurances), an insurer will be unattractive for these individuals.
In addition, insurers may give special financial benefits to the overcompensated insured if they purchase a health plan, e.g., rebates on other insurance products such as car insurance, fire insurance, or travel insurance. As soon as these insured switch to another health insurer, they no longer receive the rebates on these other products. The Dutch government facilitated such market segmentation and risk selection by allowing that health plans may provide up to 10 % in rebates to members of a ‘group’. This stimulated the forming of selected risk-groups. About two-thirds of the Dutch population have purchased their health plan via a ‘group’. Such groups can be organized by any legal entity (e.g., employers, shops, sports clubs, patient organizations, and private initiatives). Whereas insurers have to respect open enrollment, groups are free to reject applicants. For example, anyone can organize a group of overcompensated individuals and negotiate with insurers about (financial) benefits for the group. There are many examples of risk selection in the Netherlands via groups [
10].
15
The regulator can monitor these forms of risk selection by closely monitoring the market and the insurers’ behavior with respect to supplementary insurances and other (insurance) products, ‘groups’, and via ‘mystery’-applications for supplementary insurances and health plans. Do only selected (overcompensated) groups of consumers receive special financial benefits if they purchase a health plan? Finally, in the case of strong signals of risk selection, the regulator can measure whether an insurer, a health plan, or a group has an overrepresentation of over- or undercompensated groups (see “
Overrepresentation of over- or undercompensated groups”) and, if the regulator is authorized to do so, the regulator can ask for the reports of meetings of relevant employees working for the insurer.
Priorities
The measurement of all possible signals of risk selection is very costly. Therefore, the regulator should set priorities. The regulator should make a good estimate of the likelihood of different forms of risk selection and the seriousness of its consequences, both in the short and in the long run. For example, a form of risk selection that only results in one group of consumers paying 100 euro per year more than another group, could be considered lower than the social loss resulting from a form of risk selection that distorts the quality of care and thereby reduces or eliminates the access to good quality care for the underpriced high-risk patients. By multiplying the estimated probability of each form of risk selection with its estimated social loss, the regulator may give priority to potential signals of forms of risk selection with the highest expected social loss.
Effective supervision can also prevent undesired forms of risk selection. In any case, the regulator must have a permanent update of which selected groups of insured are over-and undercompensated by the current risk equalization, and to what extent.
Showing the absence of risk selection is impossible
Showing the absence of risk selection requires showing the absence of “actions with the intention and/or the effect that the cross-subsidies as intended by the regulator are not fully achieved”.
If risk equalization is
perfect, risk selection is absent. However, it is impossible to show that the risk equalization is perfect. Perfect risk equalization exists if and only if there exists no single group of over- or undercompensated insured. Because in principle the number of subgroups is unlimited, it is practically impossible to show that there exists no single group of over- or undercompensated insured.
16
If risk equalization is imperfect, it is also impossible to show the absence of risk selection. In principle, the number of actions that can be qualified as risk selection is unlimited. It is impossible to show the absence of all these actions. Showing that all insurers or all health plans have an equal risk portfolio of insured is also no proof of the absence of risk selection, because all insurers or all health plans could be equally successful in risk selection. It could also mean that with one or more insurers there is both positive risk selection (e.g., an underrepresentation of chronically ill insured) and negative risk selection (e.g., an overrepresentation of low-educated low-income people), and that these selection effects cancel out. Finally, not rejecting the null-hypothesis “that a selected group of insured is not over- or undercompensated” with a certain level of statistical significance is not a proof that “the selected group of insured is not over- or undercompensated”. Possibly, this group is over- or undercompensated, but the size of the group is too small to come to statistically reliable conclusions, e.g., in the case of rare diseases.
The conclusion is that although the evidence of risk selection can be shown with a specified level of statistical significance, it is impossible to show the absence of risk selection.