Taking actuarial fairness seriously: what is required for the ethical use of genetics in insurance?

Abstract

Whether life insurers should be able to consider genetic information during underwriting is a long-standing debate often characterized by strong opinions on both sides. Insurers push for full access to applicants’ genetic information, and consumer advocates often call for a ban on insurer use of the information. Both sides employ concepts of fairness and discrimination in supporting their position. This article considers the concept of actuarial fairness, where individuals are expected to pay for the risks they bring to an insurance pool. Currently, law and policy adopting this standard most often take a deferential approach, allowing insurers to utilize genetic information with wide latitude. This article takes seriously a middle-ground approach, broadly labeled as actuarial utility. Building from prior literature examining this issue, this article proposes a framework US policy can adopt to assist in the assessment of the actuarial utility of genetic information with a particular focus on emerging genetic technologies.

I. INTRODUCTION

A 2013 case study described an Australian man with Lynch Syndrome, a genetic predisposition for colorectal cancer, who applied for life insurance but was denied full coverage.¹ Undeterred, he gathered epidemiological evidence showing that, with regular colonoscopies, his risk of dying from colon cancer was minimal. He then submitted this evidence to a life insurance company, noting that he was also sending the information to the Australian Human Rights Commission. Soon after, the man was offered full coverage by that same company.² The man’s course of action and overall experience is unusual for several reasons. First, at the outset, life insurers generally do not ask for genetic test information in initial applications. Instead, they may learn of genetic predispositions through follow-ups from family history inquiries or medical records. Second, given the opaque nature of much insurance underwriting, applicants may be unaware of whether and how their genetic information is being utilized in the decision process for determining eligibility or setting premiums. Third, evidence indicates that insurers are only considering a small number of genetic conditions, such as hereditary cancers and neurological conditions. However, as genomic science advances and more individuals receive genetic testing, there are mounting questions about how and when insurers will and should utilize larger amounts and types of genetic information.

The Australian case example highlights several of the complexities and unresolved issues inherent to proposals that would allow insurers to use genetic information in their underwriting. Do insurers have sufficient actuarial evidence to support some of the underwriting decisions that they are making in the complex and rapidly evolving world of genomic information? What does the law mean when it holds that insurance premiums must be actuarially fair?

Actuarial fairness describes the principle that insurance policyholders should pay premiums that reflect the likelihood they will file an insurance claim. While this initially appears relatively straightforward, there are contrasting views about whether genetic information should be used to predict the likelihood that sub-groups within an insurance pool will file a claim. Health insurers are legally barred from considering genetic information in many countries, including under US federal law, but regulations vary regarding whether other types of insurers are allowed to consider an applicant’s genetic risk. Insurers maintain that access to genetic test results is vital to ensure a fair and functioning insurance market and to realize the goals of actuarial fairness.³ Consumers and advocates, on the other hand, tend to hold the view that actuarial decisions based on genetic information are discriminatory, since individuals cannot control their genetic make-up⁴ and argue that use by insurers should be barred to encourage genetic testing and promote health without the concern of a negative impact on premium rates or a denial of coverage.⁵ In the US, the two sides of this debate often take extreme policy positions, either arguing that insurers should have complete regulatory freedom to use genetic information as they deem appropriate or that there should be a complete ban on their ability to do so.

In this article, we argue that this polarized debate results from differing understandings of ‘fairness’ on each side, as well as a prioritization of competing goals and interests. Currently, across the vast majority of US states, non-health insurers are given wide deference to collect and consider genetic information in underwriting. Although they are ostensibly both practically and legally constrained by the need to have actuarial justification for their decisions, there is little check on insurers to understand how they are defining this concept, especially when it comes to quickly advancing genomic technologies.

We consider a middle ground by proposing a framework that involves a rational discrimination approach⁶ to using genetic information in actuarial decision-making. While the authors of the paper disagree about whether US policy should ultimately prioritize such a rational discrimination approach or a more stringent ban on insurer use of genetic information, we agree that the current state of affairs gives too much latitude to insurers and does not do enough to balance between insurers’ concerns about market sustainability and consumers’ concerns about social fairness. Thus, our framework represents a transparent articulation of the concepts that would constitute actuarial fairness, which could be used to implement a necessary set of baseline protections if genetic information is to be used by insurers. In this article, we set the debate by outlining both the theoretical underpinnings of actuarial and social fairness as well as laying out the competing goals and interests amongst stakeholder positions. The tensions between these competing interests highlight the potential merits of a middle-ground approach. We outline a framework for the US that builds upon existing literature and focuses on the concept of actuarial utility, which is made up of several central characteristics: predictive value, disease characteristics, and strength of evidence. We illustrate the utility of our framework by using practical examples that highlight the value of transparent industry standards for the use of genetic information in insurance eligibility and pricing decisions. We conclude with some considerations about the challenges that may arise from the implementation of our framework and suggest some procedural protections that may be necessary to mitigate those concerns.

II. SETTING THE DEBATE

II.A. Theoretical Underpinnings of Actuarial Versus Social Fairness

In the debate over insurer use of genetic information, insurers and consumers both use the terms of fairness and discrimination, but they often have different conceptions of these terms, which lead them to arrive at different conclusions.⁷ Fairness and discrimination are morally-laden terms that are difficult to define.⁸ On a simplistic interpretation, fairness means treating likes alike. That is, those in the same situation should be treated the same. Whereas discrimination entails treating people differently. This can sometimes be warranted, such as in medical triage, when emergency patients are prioritized over patients with minor injuries. However, discrimination can be unfair when it is unjustified or based on prejudiced notions.

Insurers tend to utilize the language of ‘actuarial fairness’. Generally speaking, a premium is actuarially fair if and only if a rise (or fall) in premium is concomitant with a rise (or fall) in an individual’s expected payout (ie a life insurance payment or a disability insurance claim). Here the sum of expected payout is equal to the probability of triggering an insured event multiplied by the amount paid out by the insurer for that event. This conception of actuarial fairness tracks with two theoretical visions of what it means to be actuarially fair. On the first interpretation, actuarial fairness is concerned solely with the distribution of risk and premium between the insured and insurer. This is exemplified by the traditional interpretation of Kenneth Arrow’s account of actuarial fairness, where the fair premium is equal to the risk that is covered by the insurer.⁹ On the second target of comparison, fairness is concerned with the distribution of premiums across insureds within the pool. This means that insureds with the same risk ought to be paying the same premium, otherwise those at lower risk may be seen as ‘subsidizing’ those at higher risk.¹⁰

Alternatively, when consumers speak about fairness, they do not typically refer to actuarial fairness. Instead, they often evoke themes of social fairness where premiums and underwriting are unfair if they systematically prevent individuals from becoming insured. While actuarial fairness governs what counts as fair between members of an insurance pool strictly based on risk and expected returns, social fairness draws on other accounts of fairness that are based on exclusion and disadvantage of members of social groups.¹¹ Some have argued that this is simply another way of interpreting actuarial fairness—by comparing those within insurance pool and those outside the insurance pool.¹² Those who lie outside of the insurance pool are uninsured, often either because insurers do not want to accept their risks or they cannot afford their premiums. According to some, it is unfair when some go uninsured by virtue of their inability to afford the premiums that insurers demand.¹³

Thus, an analysis of fairness could be concerned about the distribution between three different groups: (i) insurers and insureds; (ii) insureds and other insureds in the same pool; or (iii) insureds and uninsured.¹⁴ Policies may seem fair when analyzed under the perspective of one comparison group, but may be unfair when considering another perspective. Such a tension came to a head in the early-90s in the US, when health insurers refused to cover HIV-positive individuals at affordable rates. Under the perspective of fairness across the insurance pool, insurers argued that it would be unfair to not price premiums in an actuarially-fair manner as it would lead to cross-subsidization. However, under the perspective of fairness between members within the pool and outside of the pool, it would be unfair to potentially limit access to healthcare for this marginalized group.¹⁵

II.B. Competing Policy Visions and Goals

Regulatory approaches focused on insurance underwriting tend to either completely ban insurer use of genetic information or give deference to insurers to use results in their decisions as they see fit. For a health insurance example, the 2010 US Affordable Care Act disallows insurers from considering pre-existing conditions, including genetic information, as the basis for changing premium prices.¹⁶ Furthermore, the Genetic Information Nondiscrimination Act (GINA) had already federally banned the use of genetic information in 2008.¹⁷ Thus, in health insurance, companies are completely barred from considering genetic information. However, such regulations do not extend to life, disability, or long-term care insurance. Florida is the only state that has banned the use of genetic information in underwriting for all three insurances unless the risk information is accompanied by an already manifested, medically-diagnosed condition.¹⁸ For most non-health insurances, state regulation of use of genetic information is minimal and highly deferential.¹⁹

Insurers and consumers often advocate for these policy extremes because they are prioritizing different goals and interests at stake. Insurers are focused on preventing cross-subsidization and protecting market sustainability. Consumers, on the other hand, are often interested in combatting actuarially unfair genetic discrimination, easing fear of genetic discrimination that could result from actuarially fair genetic discrimination, protecting insurance access for individuals, and maintaining public acceptance of the system. Scholars similarly have come to different recommended policy options depending on how these competing value-laden goals are prioritized.²⁰ Notably, the variation across US state regulation of insurer use of genetic information closely reflects these competing priorities. In an empirical analysis of insurance anti-discrimination laws across US states, Avraham, Logue, and Schwarcz identified three factors that explain regulatory variation: (a) the predictive capacity of the characteristic in question, (b) the degree of the adverse selection problem created if the characteristic is restricted, and (c) the degree to which discrimination on that basis is considered socially suspect.²¹ This provides a useful descriptive framework for understanding existing law but does not provide guidance on the specific scientific and actuarial evidence needed to evaluate predictive capacity for genetic conditions.

While it is clear that there are important interests on both sides of the debate, it is less clear how either policy extreme would actually impact the insurance market. In particular, the evidence that the worst-case outcomes (eg insurance market death spiral, widespread genetic discrimination) will come to bear is mixed at best.

1. Interests/Goals Prioritized by Insurers

i. Preventing cross-subsidization

As mentioned above, insurers argue that they need to consider genetic information, amongst other risk information, to avoid cross-subsidization across their insurance pool. In other words, it is actuarially fair that individuals pay premiums commensurate with their levels of risk, regardless of the reason for their levels of risk. On this argument, lower-risk consumers should not be ‘cross-subsidizing’ higher risk individuals.²² For example, some believe it would be unfair to charge smokers and non-smokers the same amount for life insurance given their differing risks of death. If they were charged the same premiums, non-smokers could be seen as subsidizing the risk of smokers.²³ Similarly, insurers argue that failing to consider an applicant’s genetic risk would allow for cross-subsidization from those with low genetic risk to those with higher genetic risk.²⁴

ii. Ensuring market sustainability

Insurers are also interested in prioritizing sustainability of the insurance industry.²⁵ They argue that it may not be financially sustainable for consumers to be aware of their genetic results, while insurers remain blind to this information, a situation known as information asymmetry or adverse selection.²⁶ This is because there is some evidence that individuals at higher risk of several genetic conditions are more likely than average to purchase insurance,²⁷ although the strength of this evidence is debated.²⁸

For example, one 2000 study assessed the insurance purchasing habits of 282 women with a family history of breast or ovarian cancer, some of which had undergone BRCA1 testing under a research protocol.²⁹ The study found no significant differences in the probability of having life insurance or the amount of life insurance purchased across those who received genetic testing and those who did not. In contrast, a 2003 study of 636 women who had received risk assessment for breast and ovarian cancer, including 109 who had tested positive for a BRCA1/2 pathogenic variant, were asked questions about insurance purchasing behavior.³⁰ The study found 27 individuals who increased life insurance coverage after undergoing genetic counseling and/or genetic testing. They concluded that those who increased coverage were more likely to have a positive BRCA1/2 test. Although this study is often cited to show risks of adverse selection and insurance purchasing behavior change, in total, only 27 study participants (4 per cent) increased their life insurance policy, making any large-scale conclusions about widespread adverse selection inapt from this study.³¹ Similar mixed evidence has been found in other conditions, such as Alzheimer’s disease (AD)³² and Huntington’s disease (HD).³³ Additionally, most genetic conditions are unlike AD and HD in terms of severity, stigma, and predictive nature of genetic markers, calling into question the generalizability of any findings related to insurance purchasing behavior.³⁴

In the absence of clear data of insurancepurchasing behavior, economic modeling attempts to predict impacts on insurance, but these are similarly mixed. Under traditional economic models, restrictions on risk classification cause welfare-reducing adverse selection. But the degree of welfare loss and inefficiency from adverse selection is unclear and any bans on the use of genetic information should include systematic monitoring of the welfare loss from adverse selection.³⁵ Even if adverse selection is detected amongst individuals with genetic risks, it is not clear that such an outcome would always be undesirable. For instance, insurer access to genetic test results can reduce social welfare by deterring individuals from obtaining clinically beneficial genetic tests.³⁶ From a public policy perspective, the relevant objective need not be the minimization of selection-driven market distortion. Instead, ‘loss coverage’, understood as the expected share of the population’s losses that are actually compensated by insurance, is a reasonable policy goal since compensation of losses is a central social purpose of insurance.³⁷

What is particularly worrying about adverse selection is that proscriptive legislation will lead to the dreaded ‘death spiral’, where insurers must continuously raise premiums because of their inability to accurately assess risk, leading to a collapse of the insurance market.³⁸ For instance, Monheit et al. argue that the health insurance reforms enacted in New Jersey in the early 1990s caused an insurance death spiral.³⁹ As the state enforced guaranteed issue (that is, insurers cannot deny insurance to any individual regardless of risk levels) and community-rating (setting the same premium for everyone in a designated area), premiums rose dramatically while enrollments dropped, eventually leading to scaled back coverage. With community-rating, individuals were not paying their actuarially-fair premium and with guaranteed issue, no one could be denied insurance. Thus, high-risk individuals signed up for insurance, causing insurance companies to have to raise premium prices across the board to cope with high costs. At the same time, low-risk individuals dropped out of insurance pools as the premiums rose above their expected utility from insurance. Without proper mechanisms, voluntary insurance programs that are not actuarially-fair are likely to lead to adverse selection and could result in unsustainability of the insurance pool.⁴⁰

While more recent evidence of death spirals are limited, The Grattan Institute’s 2021 report posits that the Australian health insurance industry is in a critical state because rising premiums have forced a ‘downgrading’ phenomenon where more members choose lower-tier policies with more exclusions. Older Australians receive significantly higher benefits, yet pay the same base premiums under the community rating system. As the average age (as well as risk) of the pool increases, the cross-subsidy from the young might encourage further dropouts.⁴¹

But empirical evidence of adverse selection leading to death spirals is not uncontentious. While New Jersey’s experience in the 90s shows some evidence of a death spiral, other studies in New York, Pennsylvania, and Connecticut failed to significantly show a dramatic fall in insured individuals or a spike in premiums because of community-rated plans.⁴² Thomas Buchmueller and John DiNardo even suggest that ‘the notion that community-rating leads to adverse selection death spirals appears to have passed into the conventional wisdom’, and that ‘there has been little systematic attempt to document the magnitude of such an effect’.⁴³ Furthermore, it is too early to tell if the Australian case will result in a death spiral, since Australian lawmakers have policy tools at their disposal to encourage participation in the private insurance.⁴⁴ Similarly, although there were fears of a death spiral in the American health insurance system when the individual mandate was dismantled, these fears have not been realized.⁴⁵

Despite contested evidence, the death spiral still looms as a perceived threat for the financial survivability of insurance firms. However, in the case of regulating insurer use of genetic information specifically, there has been little to no evidence that policies barring insurer use have drastically affected insurance premiums.⁴⁶ The overarching conclusion from the literature suggests that insurance death spirals are a conditional phenomenon. While insurance markets are theoretically structurally vulnerable to adverse selection, the eventuality of a death spiral is dictated by the strength of compensating policy instruments. Current evidence suggests that rising risk as a result of aging populations may undermine risk-pooling. But in the case of genetic information, there is little evidence suggesting that adverse selection will occur and lead to death spirals.

If adverse selection and a death spiral materialize, the insurance industry may eventually collapse as coverage becomes increasingly unaffordable, harming all actors. Even if a death spiral is not fully realized, insurers worry that a ban on the use of genetic information may avoid a ‘genetic underclass’ of individuals unable to afford high premiums set from their genetic information while inadvertently creating a ‘financial underclass’ affecting all consumers due to the overall increase of premiums caused by adverse selection.⁴⁷

2. Interests/Goals Prioritized by Consumers

i. Combatting actuarially unfair genetic discrimination

Consumers and their advocates have generally been against the use of underwriting based on genetic information in large part because they are worried about genetic discrimination and the denial of insurance. This worry is particularly acute when the use of genetic information would be both actuarially ‘and’ socially unfair. It is not in an insurance company’s best interest to charge actuarially unfair rates, since this should lead to customers leaving for competitors that more accurately price risk. However, there is anecdotal evidence that actuarially unfair use of genetic information has occurred in the life insurance industry, most notably with single-gene conditions.⁴⁸ Concerningly, there is also anecdotal evidence that insurers have misinterpreted genetic tests and subsequently erroneously denied consumers insurance.⁴⁹ In one particularly egregious case, a consumer with family history of HD was incorrectly denied coverage based on flawed reasoning that the individual may later develop HD—ignoring that the individual did not have the necessary single-gene mutation.⁵⁰ Genetic tests can be misinterpreted⁵¹ or erroneously overemphasized when predicting future disease.⁵² A genetic test may indicate increased risk that has been subsequently mitigated through preventative measures. An example of this is when a person may opt for a mastectomy after learning of a mutation in the BRCA gene, which leads to ‘lower’ risk of breast cancer as compared to the general population.⁵³ Mechanisms for recognizing the nuances and incompleteness of genetic information may be necessary to prevent insurers from using information in an actuarially unfair way. This could include requirements that insurers are transparent about when and how they are using genetic information to make decisions or an appeals process for adjudicating cases of alleged unfair use of genetic information.

ii. Combatting fear of genetic discrimination

There has been documented public fear of both actuarially unfair⁵⁴ ‘and’ actuarially fair⁵⁵ genetic discrimination.⁵⁶ It is important to note, however, that although there are some evidence and reports of insurance denials based on genetic information,⁵⁷ there is no robust evidence of widespread discrimination. This is likely because, as discussed above, at this time insurers may not widely be using genetic information.

Nevertheless, public fear of genetic discrimination continues to lead to tangible medical and research consequences. Clinically indicated genetic testing may provide valuable data that can inform individual clinical decisions (eg selecting between treatment options, implementing preventative behavioral changes, or determining optimal screening frequencies).⁵⁸ At a societal level, the correct utilization of genetic information can minimize the population-level burden of disease.⁵⁹ However, there is evidence that individuals are sometimes hesitant to undergo medically indicated testing due to concern about subsequent genetic discrimination.⁶⁰ In the research context, potential participants may also be reluctant to participate in studies that involve genetic testing,⁶¹ which can hinder the understanding of diseases and creation of new therapeutics. This widespread fear was the primary reason behind the passage of GINA.⁶² Yet, because of gaps in the law, such as the lack of application to life, disability, or long-term care insurance, there continues to be acute public concern about the potential for discriminatory use of genetic information by insurers.⁶³

Specific concern about genetic discrimination is likely informed by a history of stigmatization against individuals based on genetic traits.⁶⁴ Additionally, genetic information can reveal highly sensitive information such as predisposition to severe, incurable, degenerative diseases. The information revealed through genetic testing impacts not only the individual but may also have implications for genetically related family members. Genetic information is sometimes described as being unique because genetic codes are immutable characteristics within a person.⁶⁵ Unlike other characteristics such as behavior and environmental exposures, genetic codes are static.⁶⁶ Additionally, due to the expansive amount of information in genetic data and the exponential pace of research in understanding this information, genetic tests have been argued as being uniquely deserving of privacy and protection.⁶⁷

The need for a robust genetic information framework is also partly driven by a concern for equity, as the fear of insurance discrimination is a well-documented barrier to genetic testing, and this fear falls disproportionately on communities that already face the greatest barriers to insurance access. Research consistently shows that racial and ethnic minorities express greater concern about insurance discrimination arising from genetic testing.^68,69 African American women were significantly less likely to undergo BRCA1/2 testing than white women with comparable family histories of breast and ovarian cancer.⁷⁰ African American and Hispanic populations also express stronger concerns about abuses of genetic testing than white populations, with medical mistrust being a primary predictor of these concerns.^71,72 Thus, providing a robust and transparent framework that outlines the use of genetic information in insurance can alleviate much of these fears. Any evaluation of whether to permit insurer use of genetic information must account not only for aggregate welfare effects, but also the distributive dimension of the costs of the policy, particularly when the populations most likely to be deterred from testing are those who could benefit most from early detection and intervention.

iii. Protecting access for individuals

If the use of genetic information is permitted, there is a need to protect against exclusion of those with ‘high-risk’ genetic results.⁷³ Denial of insurance is an understandable concern for the public. However, it is unclear what consumers may think about fluctuations in insurance premiums based on more nuanced genetic information, such as polygenic scores or epigenetic markers. Would they ever trust insurers to adjust premiums based on genetic information in a ‘fair way’ and could insurers be worthy of this trust?⁷⁴

iv. Public acceptability

If life insurance companies take the controversial step of considering genetic testing information, it must be done in a politically and socially acceptable manner.⁷⁵ Traits that could be informative for risk classification have historically been excluded when deemed objectionable.⁷⁶ For example, characteristics such as gender, race, religion, or history of being a victim of intimate-partner violence are seen as inappropriate to include, no matter how statistically accurate they are in predicting risks.⁷⁷ Exclusion of these characteristics is sometimes enforced by law and other times by social norms.⁷⁸

Related to public acceptability is perceived fairness of the policy. Some believe that using genetics to determine premiums is unfair because individuals have no choice in their genetic composition.⁷⁹ Characteristics such as where one lives and whether they smoke are seen as more modifiable and subsequently more acceptable. However, there is increasing understanding that ‘behavioral’ risk factors are often related to socio-economic status and environmental influences that are difficult or even implausible to control.⁸⁰ Public opinion survey data will be useful for identifying the types of genetic information that are most socially objectionable. Stakeholders (eg regulators, advocates, and insurance companies) will have to balance the actuarial utility of information against the way that using it will be perceived by the public.⁸¹

II.C. Potential Impacts of Policy Choices

As we have shown, insurers and consumers come to the debate about insurer use of genetic information with discrete conceptualizations of fairness and discrimination and, thus, prioritize different goals and interests. Yet, the most dire predictions on both sides of the debate—the unraveling of insurance markets and widespread genetic discrimination—are currently theoretical concerns. This is not to minimize the potential harmful impacts of small insurance market effects or denials of insurance for individuals at high risk. However, neither risk currently appears to be widespread.

Indeed, current insurance practice reinforces this conclusion. For example, although insurers consistently warn about the potential destabilization of insurance markets if their ability to collect genetic information were circumscribed, there is little evidence that insurers are currently underwriting based on genetic information on a widespread basis.⁸² Many life insurers do not ask questions about genetic information on their applications,⁸³ instead relying on family medical history and, potentially, an applicant’s medical records to identify risk. This practice is likely because insurers currently seem to consider only a small number of genetic markers in underwriting.⁸⁴

The lack of systematic collection of genetic information indicates that some applicants with genetic predispositions that increase risk are being underwritten without taking that risk into account. For now, this might be accurate enough for insurers. Much current genetic information does not help predict risk beyond already collected family medical history and current health status,⁸⁵ and there are plenty of other sources of risk-relevant data that the insurers can consider.⁸⁶ Indeed, collecting genetic information could be inefficient for insurers because they could receive ‘noisy’ responses, such as applicants indicating they have had an ancestry genetic test or carrier screening during pregnancy. These test results, while genetic, do not contain useful information for predicting health and mortality risk.

This indicates that, despite claims that lack of access to genetic testing information will cause catastrophic results to insurance markets, these effects, if ever realized, are unlikely to be immediate. Insurers are likely more concerned with the future ability to predict risk from common genetic information as opposed to our current ability which often is limited to relatively rare monogenic conditions.⁸⁷ Additionally, they may be avoiding widespread use of genetic information today in order to avoid future proscriptive legislation.⁸⁸ However, while the choice to balance the costs associated with collecting genetic information with the marginal benefit of the information gained may make current economic sense for the insurance industry, it fails to meet the true goals of actuarial fairness. Actuarial fairness, to be true to its underlying goals, demands full collection and use of risk-related information, no matter the cost.⁸⁹ Despite this inconsistency between theory and practice, the insurance industry generally continues to lobby against bans on use of genetic information,⁹⁰ likely to preserve the ability to underwrite based on genetics in the future. However, this also leaves many individuals fearful of discrimination that isn’t currently occurring on a widespread basis, but could grow given legal structures allowing deference to the industry.

We are thus poised at a moment when current genetic test results appear to have few consequential impacts for insurance underwriting and therefore are not causing vast genetic discrimination. As such, now is an opportune time to explore in depth what an actuarial fairness policy should look like moving forward.

III. A MIDDLE GROUND

While many argue that there seems to be a distinction between social and actuarial fairness,⁹¹ there is also a case to be made about the ways in which the concepts overlap. Both take the idea that different treatments across people require justification. Despite often taking maximalist positions about the appropriateness of using genetic information, proponents of actuarial fairness would agree that different premiums require actuarial justification. Such a view is echoed by Anne-Marie Cotter’s account that discrimination is wrong because it treats people unequally ‘without rational justification’.⁹² Furthermore, both notions of social fairness and actuarial fairness rest on a consequentialist notion of maximizing overall value. After all, two justifications for actuarial fairness are to stabilize insurance markets (a social good) and to provide individuals an avenue to convert uncertain risk into certain losses. Whereas social fairness aims, in part, to improve the welfare of those that are worse off in society.

Given the challenges of either fully permitting or entirely prohibiting the use of genetic information, it is worth considering whether policy alternatives exist that could meaningfully address concerns on both sides. Internationally, various policies have been employed to regulate insurer use of genetic information.⁹³ Several countries have completely barred life insurers from using genetic information. For example, Canada’s Genetic Non-discrimination Act makes it a criminal act to use genetic information in contract formation, including insurance contracts, although the comprehensiveness of this bill in practice as it relates to insurance has been questioned.⁹⁴ Recently, Australia’s government has passed a law to bar life insurers from using genetic information.⁹⁵

Two approaches used across the globe seek some middle ground between insurers and consumers. First, under the fair limits approach, insurers are only allowed to use genetic information for policies above a set monetary threshold.⁹⁶ For example, Singapore has adopted a fair limits approach on the use of genetic information. Insurers are prevented from using genetic information for insurance products valued below SGD$2 million (approximately USD$1,575,000), which far exceeds the estimates of Singapore’s Life Insurance Association of the average protection needed per individual at SGD$813,892 (approximately USD$640,000).⁹⁷ Such limits aim to contain adverse selection, as individuals with known genetic risks would be unable to purchase high-value insurance plans without incurring high premiums, while still ensuring adequate access to insurance that provides sufficient coverage. Singapore’s approach also remains highly sensitive to locally salient risks to strengthen protections for participants in an ongoing national genetic testing program.⁹⁸

Second, under a rational discrimination approach, insurers can use genetic information that has been independently reviewed for scientific validity, clinical significance, and actuarial relevance.⁹⁹ Although both policies are potential options, this article focuses on what a rational discrimination approach would look like in the US context.

The rational discrimination approach provides a possible mechanism for balancing interests across stakeholders; however, the potential success of this approach greatly depends on whether there is rigorous review of what kinds of information can be used.¹⁰⁰ In the US, seven states have laws that require actuarial justification before a life insurer can use genetic information.¹⁰¹ In theory, this would allow state insurance departments to assess whether information utilized by insurers has sufficient justification. In practice, however, these laws have been deferential to insurers.¹⁰² In the sections that follow, this article lays out a potential framework for a US-based rational discrimination approach. The framework builds upon existing actuarial justification models and explores how a US-based framework could apply to several key emerging technologies.

IV. A US-BASED RATIONAL DISCRIMINATION FRAMEWORK

A rational discrimination approach has the potential to balance many of the competing interests of insurers and the public, occupying a middle ground between the extremes of uninhibited use or no use of genetic information. Building on existing frameworks, this section provides further elaboration, sketching a structure for the US, with a particular focus on scientific validity, clinical significance, and actuarial relevance of genetic information.¹⁰³ Given the competing interests discussed in the previous section, the goal is to build a methodology that simultaneously combats unfair actuarial discrimination based on genetics, addresses public fear of allowing discrimination, promotes access for individuals, and ensures market sustainability.

IV.A. Existing Frameworks

Several groups and countries over the years have developed or proposed frameworks to assess whether and when a life insurance company should be able to consider genetic information. Under these models, an independent third party would be responsible for assessing specific genetic tests to determine if they could be incorporated into life insurance underwriting. For example, in the early 2000s, two legal organizations, the Australian Law Reform Commission and the US Uniform Law Commission, undertook separate efforts to set guidelines for when insurers could use genetic information.¹⁰⁴ These efforts were similar in many ways, including in their standards and in the fact that neither proposal was implemented. Under the Australian proposal, genetic information needs to meet thresholds of actuarial relevance and scientific reliability before use.¹⁰⁵ The Uniform Law Commission focused less on actuarial evidence, instead relying on scientific standards of clinical validity, clinical utility, and analytical validity, which mirror the Australian category of scientific reliability.¹⁰⁶ The Uniform Law Commission also included another standard called the ethical, legal, and social implications (ELSI) category, where regulators would need to assess ELSI issues regarding life insurer use of any particular piece of genetic information.

A similar structure had been implemented in the UK in 2001. This policy provides an example that combines the fair limits and rational discrimination approaches with a robust independent review process. The UK adopted an incredibly robust review process, requiring insurers to submit extensive evidence of the scientific and actuarial justification for using specific genetic tests before their use is approved.¹⁰⁷ In the nearly two decades that have elapsed since the policy began, only HD for life insurance policies above £500,000 (approximately USD$675,000) has been approved.¹⁰⁸ Despite these high standards of actuarial justification, there is no evidence of insurance market instability in the UK.

However, there have been important policy updates in more recent years. In recognition that the Code itself lacked a transparent, structured basis for evaluating additional conditions beyond HD, the Association of British Insurers commissioned the Cambridge Centre for Health Services Research (CCHSR) in 2021 to develop a supplementary framework. Importantly, the fact that the UK itself recognized the need for a structured evaluative framework underscores the very gap our framework aims to fill.¹⁰⁹

Within actuarial practice, the primary US standard for risk classification is derived from the Actuarial Standard of Practice No. 12 (ASOP No. 12), Risk Classification, issued by the Actuarial Standards Board. ASOP No. 12 which requires actuaries to select risk characteristics that are related to expected outcomes, consider the interdependence of risk characteristics, account for adverse selection potential, and assess the reasonableness of results while complying with applicable law. While ASOP No. 12 provides the professional standard that any US actuary working on genetics-based risk classification must satisfy, it does not address the specific scientific and clinical considerations needed to evaluate genetic information that our proposed framework below is designed to fill.¹¹⁰

In summary, existing frameworks providing analysis on the use of genetic information in insurance share several key gaps. First, none provide detailed criteria for evaluating the specific scientific and clinical evidence that should underpin actuarial use of genetic information. Second, most existing frameworks do not clearly distinguish between information that is merely actuarially relevant versus a heightened threshold of information that has, what we call, genuine actuarial utility. Third, existing frameworks tend to operate within a single disciplinary domain without integrating clinical genomics, actuarial science, and ethical considerations into a unified evaluative structure. It is these gaps that the framework we propose below seeks to fill.

IV.B. A US-Based Framework

If the US were to adopt a rational discrimination model as a middle-ground approach to regulating insurer use of genetic information, we propose utilizing a framework similar to the UK Code on Genetic Testing and Insurance.¹¹¹ Here we lay out several specifics of a potential framework that are particularly relevant given continued advances in genomic technologies. The first is the general concept of actuarial utility, which serves as a scaffolding on which additional considerations (ie predictive value of genetic information; disease characteristics/actionability; and scientific validity/strength of evidence) can be specified (Figure 1). Our framework is an early attempt at articulating a rigorous way to analyze when genetic information might justifiably be used in actuarial decisions. As such, though we do try to explore how these considerations will generally operate, we do not try to develop formal thresholds to precisely differentiate between cases. Furthermore, the concepts can sometimes be in tension with each other, complicating the analysis of a given case. Our goal will be to highlight these possible tensions, but resolving them will often require making normative judgments about how to balance their relative weight in the context of specific cases and cultural/political settings.

Figure 1.

Framework of actuarial utility.

1. Actuarial Utility

This framework allows for the use of genetic information under some circumstances to keep insurance costs down and to maintain the viability of the insurance industry. But this presumes actuarial utility, which mirrors the UK Moratorium standard of actuarial relevance, requiring assessment of the likelihood that a given piece of information would play a meaningful role in setting actuarial decisions. Thus, as a threshold analytic step, criteria are needed to determine when genetic information is sufficiently important to justify discrimination.

Note that we have intentionally used the term actuarial utility, as distinguished from actuarial relevance. Information can be relevant to actuarial decisions, but unless it plays a meaningful role in influencing those decisions, discrimination is not justified. Thus, this is a higher bar than the current US standard of actuarial justification, which has been determined largely through deference to the insurance industry. Furthermore, actuarial utility also requires that adding genetic information to modeling provides some marginal advantage over the status quo. For example, information about unhealthy behaviors (eg smoking, poor nutrition, low physical activity) can be substantially more predictive of cardiovascular disease than genetic information, meaning that the latter has relatively reduced actuarial utility.¹¹² This similar standard was incorporated into the UK review model.¹¹³ Finally, actuarial utility will vary across different types of insurance. For example, genetic information related to patient longevity will be more useful to life insurance decisions. Information connected to a patient’s disease course would be important for estimating the cost of future care needs, which would be more useful to health insurers. Long-term care insurers would likely be interested in information related to both longevity and cost of care.

It is useful to compare our concept of actuarial utility with the criteria used in specific frameworks. The CCHSR framework discussed above asks four questions: test utility for characterizing risk, disease impact on length and quality of life, test uptake, and behavioral responses. Our framework subsumes the first two of these questions within its analysis of predictive value and disease characteristics, but goes further in several respects. First, by explicitly requiring that genetic information provide marginal predictive value beyond what is already available from non-genetic sources (such as family history or existing clinical data), our framework imposes a higher standard than mere test utility. Second, our framework does not consider test uptake as a criterion for permissibility; while uptake is relevant to the magnitude of any market impact, it should not determine whether the use of a given genetic test is actuarially justified. Third, compared to ASOP No. 12 which does not demand causal relationship between risk characteristics and outcomes, our framework implicitly demands a higher evidentiary standard for genetic information, requiring not merely correlation but evidence of predictive value, disease actionability, and scientific validity. This higher bar reflects the distinctive ethical considerations associated with genetic information, including its immutability and implications for family members.

2. Predictive Value of Genetic Information

A number of factors will be important to consider when assessing whether genetic information has actuarial utility. The UK Moratorium requires genetic tests to have clinical relevance—we agree. However, there are two specific considerations that are especially relevant within this theme. The first consideration is tied to the idea of penetrance, or the probability that a person with a genetic variant that has been associated with a disease will develop any features of that disease. Some variants are completely penetrant, meaning that if you have the variant, you will eventually develop all features of the disease. But penetrance is most often low or incomplete, in which case patients with a given pathogenic variant are unlikely to develop the disease or are unlikely to develop all disease symptoms (although can still be at higher risk than the general population).¹¹⁴ As a general principle, when a given genotype is more predictive of actual disease manifestation, it has more actuarial utility. Other factors, such as variable expressivity similarly impact the predictive value of genetic information. Predictive value builds on, and extends, the idea of ‘actuarial accuracy’: precisely estimating the probabilities and losses associated with insured events.¹¹⁵ But actuarial estimates can be precise yet practically useless, such as in the case of low-penetrance genetic variants or genetic data that adds nothing beyond family history. Relatedly, while genetic variants can sometimes independently cause disease, it is often the case that external factors (eg health-related behaviors, environmental exposures) also play a large role in disease manifestation. Some diseases are completely driven by genetics, some are completely driven by external factors, and most are driven by both.¹¹⁶ When the genetic contribution to disease is relatively stronger, knowledge about a person’s genotype will be more directly useful for actuarial decision-making.

Given that the predictive value of genetic information (and the associated actuarial utility) exists on a broad continuum, it could be helpful to articulate a threshold of predictive value, below which genetic information should not be used for actuarial purposes. Such a threshold would enhance social fairness, helping to reassure an understandably anxious public that allowable discrimination is limited to cases where the information has substantial actuarial utility. Of note, this will not reassure those who are concerned about allowable discrimination in cases of actuarial utility. A minimum threshold would also have an ancillary benefit of mitigating (although not completely eliminating) the risk that patients would forgo testing because of a fear that their results would be used to discriminate against them in an unfair way.

It is also worth discussing the social fairness concerns raised by actuarial use of genetic information with very high predictive value. Unlike cases where behaviors and lifestyle choices (and the corresponding individual responsibility) can play a role in development of a disease, high-penetrance genetic disease are largely outside of a person’s control. This does not necessarily mean that actuarial use of high penetrance information should be prohibited, but rather that it is important to think proactively about ways to protect high-risk individuals. We suggest that high penetrance genetic information should not be used as a justification for denying coverage; even with the certainty (or high likelihood) of developing a genetic condition, there is always an upper bound price that can be determined, given fixed coverage amounts in life insurance. Such a premium may be very unattractive to consumers with certain genetic conditions, given their higher likelihood of triggering an insurable event. Nevertheless, a price that insurers are willing to accept, even if higher due to genetic risk or pure uncertainty, allows for further policy options that could be employed to ensure that individuals are not effectively priced out of coverage (eg subsidies, capping the allowable difference in premiums between two groups, etc.).¹¹⁷

3. Disease Characteristics and Actionability

The next relevant consideration under the broad concept of clinical relevance has to do with the relationship between genetic information, expected disease manifestation, and the ability to impact the course of a disease. Knowing the clinical prognosis associated with a genetically-mediated condition can help insurers to estimate future health costs and life expectancy, which are useful for actuarial decision-making. Actuarial use of genetic information will be more likely to be justified in cases where the genetic information meaningfully contributes to understanding of disease course. Naturally, this can cut in either direction. Genetic information predicting that a patient’s disease will likely lead to early death would be actuarially important for setting a higher premium for a life insurance policy.

Even if genetic information can be used to help predict the course of a disease, that knowledge should be discounted by the availability of preventive steps or effective treatments. Thus, the actuarial utility of some genetic information will be mediated by how individuals change in response to the genetic information that they receive. While genetic predisposition to a disease is often understandably viewed negatively, it can be seen as a positive when framed as an opportunity to take actions that prevent or mitigate the impact of that disease. The actuarially relevant question is not whether someone has a genetic predisposition for a disease, but rather how knowledge about a genetic predisposition will actually impact an individual’s health. Of course, this then raises a question about how to grapple with the actuarial implications of people who choose not to take advantage of prophylactic or therapeutic options, but that is an issue reaching beyond genetics, into broader questions about the fairness of imposing costs on people who make suboptimal health choices.¹¹⁸ This is especially important to consider when systemic barriers, such as lack of insurance coverage, can make it difficult for individuals to engage in optimal health choices.¹¹⁹

Finally, it is relevant to consider the severity or health burden of a given condition.¹²⁰ Generally, when predictive information is associated with higher severity conditions (eg a disease characterized by substantial neurological deterioration or a very aggressive form of cancer), the importance of predictive genetic information should increase. Note, however, when estimating the marginal risk or severity associated with a genetic variant, careful attention must be paid to the choice of baseline comparator. Some studies report relative risks using population-level controls rather than specifically comparing carriers to non-carriers, and this has important implications for how such risks are interpreted.¹²¹

It is also important to distinguish between different conceptions of severity. A disease can be severe in the mortality sense, meaning that it typically causes death well before average life expectancy. A disease can also be severe in the morbidity sense, meaning that it causes substantial disability or suffering. Some severe diseases cause both morbidity and mortality, but others can be associated with one or the other. These two conceptions of severity will have different actuarial utility across types of insurance.

4. Scientific Validity and Strength of Evidence

As a final consideration, it is vital to evaluate the strength of the evidence base being cited to support the clinical importance of genetic risk information.¹²² This mirrors the UK Moratorium’s requirement of technical relevance. Stronger evidence provides more justification for actuarial use and bolsters public acceptance. In particular, there will be a stronger case for actuarial use of genetic information when the existing data can provide a plausible theory about biological causality; mere correlation is generally a weaker form of support.¹²³ Reasonable people can disagree about how much evidence is sufficient, but it will be important to establish widely applied industry evidentiary standards to reassure the public that any actuarial use of genetic information is being implemented consistently and with sufficient justificatory guardrails.

Establishing a strong evidence base to justify actuarial use of genetic information will sometimes be straightforward, but a number of challenges exist that can make this process more difficult. First, strength of evidence can differ across ancestry groups. Underserved groups are generally less studied, partially because they are less likely to participate in research¹²⁴ and are not widely represented in biobanks.¹²⁵ Since understudying underserved communities compounds their disadvantage, the research enterprise is actively engaged in efforts to recruit more diverse research participants.¹²⁶ But until those efforts have matured, insurers would have to be cautious about recognizing the limits of non-diverse data for actuarial decision-making.

A second challenge relates to the fact that prevention and treatment evidence often lag behind research establishing the etiology and natural history of a disease.¹²⁷ This makes sense given that a basic understanding of the disease is typically required before potential interventions can be identified and examined. For purposes of our framework, however, the worry is that insurers will prematurely use genetic risk information to increase premiums, without waiting for a full understanding about how genetics can be used to help inform beneficial strategies for preventing or treating a disease.

A final challenge is defining the population of interest that is being examined. For example, in the early days of life insurance underwriting, insurers only used a handful of characteristics to differentiate their insurance pool, such as age and gender. Over time, insurers began to identify additional sub-groups to place into their own risk classes, such as separating smokers from non-smokers. In the context of genetic information, there are many possible ways to create subgroups to consider the epidemiological and actuarial value of test results. For example, with a predisposition to breast cancer, possible subpopulations of analysis include those with a family history of breast cancer, those with a pathogenic variant in BRCA1/2 genes, and those with genetic predispositions who have taken preventive measures to mitigate their risks. In general, insurers aim to create risk classes that are as homogeneous as possible. In the context of genetic information, insurers should clearly define their population of analysis and regularly reevaluate whether technology advances and medical discoveries necessitate splitting risk classes into smaller, more specific sub-groups.

V. THE FRAMEWORK IN PRACTICE

In the previous section, we argued that not all genetic information is equally useful, and that actuarial utility differs by insurance type and type of genetic information. The time is ripe to examine what a rational discrimination approach in the US would entail across differing types of information given rapid growth in genomic technologies. There are currently over 175,000 genetic tests on the market,¹²⁸ and 10 new tests being introduced daily.¹²⁹ Thus, new developments in genetic technologies may shift the current landscape, necessitating a reconceptualization of actuarial fairness as a potential policy mechanism that can balance competing concerns.

Indeed, the UK government recently released a call for evidence regarding whether the UK Code on Genetic Testing and Insurance needs to be updated given the changing genomic landscape.¹³⁰ Many of the questions asked of the UK Code focus on the distinction between predictive and diagnostic genetic test because the UK framework rests on this divide. Yet, this call shows how emerging technologies can evolve quickly, thus requiring a means of assessing when they begin to provide information of sufficient relevance and importance for insurers.¹³¹

Our proposed framework offers general considerations to help distinguish whether genetic information provides a marginal advantage to insurers beyond what they routinely collect. We include emerging technologies given speculation that insurers will find these types of genetic tests useful in the future.¹³² While an exhaustive application of our framework to all these types of genetic information in all insurance markets is infeasible within the scope of this article, in the following section, we apply our framework to three kinds of genetic information that life insurers might consider, to demonstrate the ability of our framework to distinguish between genetic information that is useful for risk stratification purposes from that which is not.

V.A. Scenario 1

Pathogenic expansions of a trinucleotide repeat in the HTT gene are the cause of HD, an autosomal dominant neurodegenerative disorder characterized by progressive motor dysfunction, psychiatric disturbances, and cognitive decline.¹³³ The mean age of onset is 35 to 44 years, and the median survival time is 15 to 18 years after onset, though earlier onset is observed with longer repeat expansions. Individuals with ≥40 repeats almost invariably develop the disease, with penetrance approaching 100 per cent by late adulthood. Symptoms worsen over 15–20 years, ultimately leading to severe disability and premature death, often from complications such as aspiration pneumonia, falls, or cardiovascular issues.¹³⁴ There is currently no cure or disease-modifying therapy, and management is focused on supportive care. Given its hereditary nature, predictive genetic testing is available for at-risk individuals with a family history, though uptake is variable due to concerns about the psychological implications of test results. Studies have shown that individuals with HD have life expectancies that are approximately 13 years shorter than the population average.¹³⁵

V.B. Application of Framework to Scenario 1

The existence of robust and replicable population data demonstrating shorter life expectancies in individuals with HD is sufficient to establish the actuarial utility of information about HTT repeat expansions to a life insurer. The predictive value of information about HTT repeat expansions is also high given their proximity to 100 per cent penetrance. The disease is not actionable, meaning that there is no effective treatment or prevention for individuals with HTT repeat expansions, and the strength of evidence associated with this condition is strong, in that several large registry studies have replicated similar findings about truncated life expectancies in affected individuals. According to our framework, HTT test results are actuarially useful and may be used by insurers for risk stratification purposes.

V.C. Scenario 2

Pathogenic variants in MYH7 are a cause of hypertrophic cardiomyopathy (HCM) and dilated cardiomyopathy (DCM), two conditions that cause changes to the thickness of the heart muscle resulting in increased lifetime risks of abnormal heart functioning and sudden cardiac death. In DCM, the cardiac muscle wall grows progressively thinner, leading to an abnormal enlargement of the left ventricle. DCM usually initially manifests in adults in their 40s to 60s, although it may present at any age.¹³⁶ In HCM, the left ventricular wall grows progressively thicker, making it harder for the heart to pump blood effectively. The clinical presentation of HCM is highly variable, encompassing a spectrum from asymptomatic left ventricular hypertrophy to arrhythmias—including atrial fibrillation and life-threatening ventricular arrhythmias—and, in some cases, progressive heart failure.¹³⁷ Population-level studies show that carriers of pathogenic variants in MYH7 have higher mortality risks than non-carriers.¹³⁸ However, epidemiologic data also show that MYH7 exhibits incomplete and age-dependent penetrance, complicating the predictive value of information about this gene. Not all MYH7 pathogenic variant carriers will develop disease, and lifetime disease risks vary by variant. Furthermore, most population studies of MYH7 carriers are in European-ancestry cohorts, limiting the generalizability of these data to more diverse populations.

V.D. Application of Framework to Scenario 2

From a life insurance perspective, information about the MYH7 pathogenic variant is potentially useful because it provides insight into which population subgroups have current and future mortality risks above and beyond what would be discernible based on information about other characteristics and exposures. However, the penetrance and, therefore, predictive value of the MYH7 variant is incomplete, as current evidence suggests that it is >50 per cent penetrant over the course of a lifetime. There is evidence that MYH7 pathogenic variants can increase a person’s risk of dying prematurely, and while there are prevention and screening options for these conditions, evidence for their effectiveness is derived from relatively small observational studies and registry data rather than randomized clinical trials. In addition, monitoring adherence to prevention and screening may be more cumbersome for an insurer than it is worth. The population studies that exist have been conducted primarily in cohorts of European origin, limiting the relevance of this evidence to more diverse insurance pools. Furthermore, the interventions used to mitigate the risks associated with pathogenic variants in MYH7 have not been proven to be effective in randomized controlled trials. Thus, while it may be tempting for a life insurer to use MYH7 test results to risk stratify insurance pools, according to our framework, this is not necessarily warranted given current evidence.

V.E. Scenario 3

Polygenic risk scores (PRS) are now available for use to calculate disease risk estimates by evaluating the cumulative impact of multiple disease-associated common genetic variants in an individual.¹³⁹ The risk estimates provided by PRS are not currently recommended for clinical use because they perform poorly in ancestral populations that are not well-represented in genomic reference databases.¹⁴⁰ In oncology, some of the most promising early studies of PRS show that these scores have the most predictive power in high-risk women and when combined with other data points.¹⁴¹ PRSs provide associational data and do not demonstrate firm evidence of disease causality, and their positive predictive value is not well-established by population evidence at this time.

V.F. Application of Framework to Scenario 3

If a PRS indicating a two-fold increased cancer risk compared to a member of the average population for that cancer were made available to a life insurer, an analysis using our framework would proceed as follows: as the positive predictive value of the PRS is not yet well-established in the literature, especially across all populations, the actuarial utility of this PRS score would not be possible to determine with confidence. While patients with a two-fold increased lifetime risk of developing cancer may benefit from heightened screening at earlier ages than members of the average population, there is no compelling evidence that this screening translates to longer life expectancies. As PRSs do not provide mechanistic information about the cause of cancer, it is unlikely that PRSs would provide actionable information about targeted treatments. The evidence base supporting PRSs is relatively young, and longitudinal studies of their predictive value and actionability are limited. Furthermore, taking into account PRS for all obscures the differences in evidentiary weight and predictive value across populations, raising equity concerns. Thus, according to our framework, PRS for cancer risk has limited value for the purpose of risk stratification.

V.G. Additional Notes on the Framework

We recognize that applying our framework to these three examples results in the same conclusions reflected in the UK’s Code on Genetic Testing and Insurance at the time of writing. The UK Code singles out HD as the sole genetic condition that insurers may find it useful to know about for risk stratification purposes. This parallel is not surprising. The use of HD as an exemplar in genetics is often critiqued because the hereditary nature of HD, including its dominant inheritance and penetrance, is rare amongst conditions. Yet these are the features that make HD clearly fit an actuarial utility model. The fact that most genetic conditions are not like HD and instead fall into the murky middle ground of actuarial utility represented by MYH7 highlights the importance of a clear framework.

Thus, while the headline ‘results’ of using our framework are not novel at this time, we believe it still makes an important contribution. By offering transparent criteria to guide the use of genetic information in insurance pricing, we have provided an explicit rationale for or against using genetic information that seeks to balance the distinct interests of consumers and insurers, and which can be applied to a rapidly changing evidence base over time.

At least three additional insights arising from these scenarios are instructive. First, even in the same gene, different variants may confer differing degrees of predictive value based on what is known about their specific impacts on gene function and the extent of population evidence about them. Second, the evidence base for genomic medicine is typically most relevant to populations with European ancestry, as the cohorts that have been studied mostly come from this group. Since most insurance markets consist of admixed and ethnically diverse sub-populations, the predictive value of a genetic variant may not be as strong as is suggested by the population studies that have been conducted. As genetic information cannot be randomized, the quality of the evidence base about a genetic variant may take a long time to reach the standard that other data points are held to in order to be considered useful for risk stratification, meaning that genes which were described earlier are more likely to have a robust evidence base supporting or invalidating their use in actuarial decision-making. The requirement that there be strong supporting evidence positive predictive value in order for genetic information to be used in actuarial decision-making may, in some instances, supersede all other considerations, as the actuarial utility, predictive value, and actionability of the information all depend on the existence of a robust evidence base.

A third insight illustrated by these examples is that some genetic conditions have evidence-based prevention and treatment strategies which may, in themselves, provide an actuarially useful basis for further risk stratifying insurance pools. For example, women with pathogenic BRCA2 variants who have had risk-reducing surgeries will have lower lifetime risks of developing breast and ovarian cancers than those who have not, though their estimated lifetime risk of developing pancreatic cancer would be the same. Within the group of high-risk prospective life insurance beneficiaries, population evidence for successful treatment and prevention strategies may provide a basis for further risk stratification within a group that has been deemed high risk by dint of genetic information alone.

The application of our framework to these scenarios suggests that insurance companies may benefit from having transparent and consistent industry standards for what constitutes a robust evidence base for genetic information in the context of their specific insurance market.

These examples demonstrate how, building on previous models, our framework can help distinguish actuarially useful genetic information from merely actuarially relevant genetic information by taking into account the information’s predictive value, strength of the supporting evidence, and how much utility it offers beyond other available data. In addition, the examples illustrate how the same genetic information might hold varying relevance across insurance markets. Currently, our framework leads to the conclusion that most genetic data (including PRS and epigenetic information) lack the predictive accuracy or evidence needed to meet the bar for actuarial utility and inform underwriting or pricing decisions. However, as the evidence base strengthens, these types of genetic information may become more useful for actuarial purposes. A strength of our framework is that it allows for the possibility that the influence of genetic information on actuarial decisions may increase over time as genomic science evolves.

An additional strength of our framework is that it grapples with the fact that the debate about actuarial use of genetic information has thus far been characterized by stakeholders’ extreme policy positions. Our framework is designed to explore the middle ground, serving the interests of both insurers and consumers. Permitting insurance underwriting for genetic results with actuarial utility helps protect insurance market stability and may prevent insurers from leaving the market over concerns regarding information asymmetry.

Our framework also protects against unjustified genetic discrimination by establishing robust standards for usage of genetic information. While our framework protects against unjust discrimination, it does not protect against actuarially fair genetic discrimination. If it became necessary to mitigate the impact of actuarially fair genetic information other policy options could be employed, such as a prohibition on refusing to issue a policy because of genetic information, subsidies to offset increased premiums for genetic risk, and limits on the premiums that people with genetic risk will be charged. Relatedly, it will be important to create a two-tiered appeals system. First, there should be a mechanism for regulators to critically assess the proposed use of a given piece of genetic information and to intervene when there is weak or insufficient evidence of its actuarial utility. Second, a mechanism could provide the means for individual consumers to appeal when it appears that the use of genetic information was unfair in their particular situation, such as in the opening case of an Australian man with Lynch Syndrome. However, with or without these policy options, consumers may continue to experience fear of genetic discrimination. It may be impossible to fully mitigate this fear through a middle-ground approach; however, the inclusion of transparent, procedural protections can ensure that consumers can learn what genetic information is used by insurers as well as appeal if necessary.

Specifically, it will be important to require transparency into the evidence used by insurers to justify the actuarial relevance of a given piece of genetic information. Currently, insurers do not share detailed information about their actuarial decision-making processes, making it challenging to compare the use of our framework to current industry standards. At present, few, if any, insurers routinely ask for genetic information as a part of the policy application process. From this, we can infer that most genetic information does not currently meet insurers’ threshold for actuarial relevance. While it is currently ‘possible’ for insurers to obtain genetic information about prospective policy holders, a lack of transparency about industry standards generates uncertainty about the extent to which insurers ‘are’ obtaining and currently using this information, or how they plan to use such information in the future. If there were to be widespread incorporation of genetic information into actuarial decision-making, it would be necessary for insurers to be transparent about why they believe the information is actuarially relevant, and how it is being utilized. Transparency would allow insurers to balance their potential future interest in using genetic information for actuarial purposes with consumers’ interests in avoiding unfair discrimination by providing clear guidelines to help consumers evaluate the likelihood that their genetic information will ever be used to inform policy eligibility or pricing in the future.

To implement the framework, considerations of administrative complexity and cost will be relevant at the level of policy design. However, we view these considerations as analytically distinct from the ethical question this framework is designed to address. The aim of this analysis is to articulate normative standards for when the use of genetic information in actuarial decision-making is ethically justifiable, not to prescribe a specific regulatory mechanism or cost structure for implementation. Conflating ethical adequacy with anticipated expense risks allowing contingent features of current institutional arrangements to determine the scope of moral evaluation.

VI. CONCLUSION

Under the status quo, there is substantial uncertainty about the current and future actuarial uses of genetic information. Insurers would certainly like to be able to use genetic information to make actuarial decisions, but the state of the science is such that most genetic information is not yet sufficiently actuarially useful, and consumers are actively worried about discriminatory use. In this article, we consider an actuarial fairness framework that would set requirements that insurers would have to meet in order to incorporate genetic information into their insurance underwriting. We set out the outline of a framework for the US as a starting point—specific thresholds and determinations will have to be developed over time. One novel contribution of our framework is that it provides heightened transparency about insurers’ potential uses of genetic information without committing insurers to a blanket prohibition on using any genetic information at all. For some, this might not provide enough protection for consumers; however, it is a necessary prerequisite for our current regulatory system that rests on concepts of actuarial fairness unmoored to any established standards or transparency.

Though insurers are not currently making widespread use of genetic information to inform their actuarial decisions, that reality becomes increasingly likely as our clinical understanding of the relationship between genetic variants and human health continues to rapidly evolve. There are good reasons for insurers to have access to any information that can accurately help to measure risk, but consumers and advocates are understandably have fairness concerns about how genetic information would be used by insurers. Our article builds on existing thinking to present a compromise approach that allows for using genetic information in actuarial decisions, but within justificatory guardrails. While actuarial utility is not a new concept, we think that incorporating it into insurance practices in a meaningful and transparent way is a prerequisite for allowing use of genetic information for actuarial decisions.