Potential corporate uses of polygenic indexes: Starting a conversation about the associated ethics and policy issues

Summary

Some commercial firms currently sell polygenic indexes (PGIs) to individual consumers, despite their relatively low predictive power. It might be tempting to assume that because the predictive power of many PGIs is so modest, other sorts of firms—such as those selling insurance and financial services—will not be interested in using PGIs for their own purposes. We argue to the contrary. We build this argument in two ways. First, we offer a very simple model, rooted in economic theory, of a profit-maximizing firm that can gain information about a single consumer’s genome. We use the model to show that, depending on the specific economic environment, a firm would be willing to pay for statistically noisy PGIs, even if they allow for only a small reduction in uncertainty. Second, we describe two plausible scenarios in which these different kinds of firms could conceivably use PGIs to maximize profits. Finally, we briefly discuss some of the associated ethics and policy issues. They deserve more attention, which is unlikely to be given until it is first recognized that firms whose services affect a large swath of the public will indeed have incentives to use PGIs.

Despite the relative noisiness of polygenic indexes (PGIs), some commercial firms will use them. We offer a simple model, rooted in economic theory, to show how acquiring PGIs could be profitable and describe two scenarios where firms could use them. Finally, we discuss some associated ethics and policy issues.

Introduction

Genetic data are proliferating and becoming increasingly informative about human traits and outcomes. Until recently, the use of genetic data to predict outcomes was largely limited to identifying individuals with differences in single genes that cause rare medical conditions (e.g., Huntington disease, Tay-Sachs disease). However, the vast majority of human phenotypes are influenced by many genes with small individual effects. In such cases, polygenic indexes (PGIs), also called polygenic risk scores, can summarize the weighted correlations between many single-molecule differences in specific locations throughout the genome (known as single-nucleotide polymorphisms or SNPs) and non-Mendelian or “complex phenotypes.” PGIs have been constructed not only for some medical conditions (e.g., heritable breast cancer, diabetes),^1,2 but also for behavioral and social outcomes (e.g., risk-taking preferences, educational attainment).^3,4

An existing literature describes the possible benefits and risks of using PGIs,⁵ but it has almost exclusively focused on medical and research settings. In clinical applications, PGIs have the potential to improve healthcare by revealing individual medical risk and guiding precision medicine. Such benefits must be weighed against the risks of classification errors, misinterpretation of genetic results by individuals or care teams, and possible stigma or distress for individuals with certain genotypes. In the research domain, PGIs could be used to improve the cost and efficiency of clinical trials and trials of policies or other non-pharmaceutical interventions or to better understand heterogeneity in the impact of environmental factors on an outcome of interest.^6,7 However, researchers and the public can easily misunderstand or mischaracterize statistical relationships between PGIs and complex social outcomes. For example, correlations between genes and educational attainment may capture some of the causal influence of genes on an outcome but will also reflect confounding factors, such as parents’ income or behaviors. Drawing any sort of conclusion requires clear statistical arguments, appropriate data, and careful interpretation. Far more concerning, some actors, to advance racist agendas, have made scientifically untenable claims about differences between populations based on genetics research on social outcomes within populations.^8,9

Less attention has been given to the use of genetic information by firms—profit-maximizing entities that produce goods and services. Most of what has been discussed has focused on cases involving single genes and applications in healthcare or health insurance. For example, there has been some concern related to the disclosure of APOE genotype, which is predictive of Alzheimer disease (AD), to potential insurers who may avoid selling insurance to individuals with a higher risk of developing AD to maximize their profits. Research has demonstrated that people who learn that they have an APOE genotype placing them at increased risk of disease are more than five times more likely to alter their long-term care insurance compared with those who do not know their genotype, triggering concerns of adverse selection in this market.^10,11 Since the emergence of PGIs, there has been some scholarship related to firms that process biological specimens or variant call files and sell people’s PGI reports for the customer’s own purposes. For instance, several articles address the ethics of selling PGI reports—predicting both medical and non-medical phenotypes—either directly online to adults (whether as actionable information or “infotainment”) or to in vitro fertilization (IVF) patients via IVF clinics to be used to inform embryo transfer decisions.^{12,13,14,15,16,17,18} Whatever one thinks of selling such PGI reports, marketing PGIs to potential consumers places these firms’ behavior in the open where it can invite (and indeed has received) scrutiny.

Virtually unstudied are the incentives that firms face in acquiring and using PGIs for purposes other than selling this information to consumers. For instance, a financial firm may wish to use genetic information to create specific products. These applications could be ethically neutral and even welfare-improving if genetic information is used to precisely target products or recommendations that consumers find beneficial. However, risks can arise if firms identify opportunities to increase profits at the expense of consumers’ welfare. For example, firms could identify older individuals with heightened genetic risk for dementia and purposely market complex, over-priced, and low-quality financial instruments to this vulnerable population. While it is likely that many of the associated ethical, legal, and social implications will overlap with those already identified in the literature in other contexts,⁵ the vast scope of phenotypes that PGIs can predict, their increasing predictive power, and the complexities of their interpretation make even well-known quandaries about the appropriate use of genomic information more complex, consequential, and urgent.

We believe that a comprehensive conversation about the potential risks and benefits of firms using PGIs has not occurred, in large part because the idea that firms will use PGIs has not yet been taken seriously. Conventional wisdom in this area instead maintains that PGIs—especially for social and behavioral phenotypes—are simply not informative enough to be profitably used by firms. For example, PGIs for educational attainment have been characterized as “useless for making individual predictions.”¹⁹ While some of the researchers who created these PGIs have themselves plausibly described them as “poor” and “not useful” individual predictors (see Supplemental Data 1, FAQs 1.8 and 2.4 in Okbay et al.²⁰). Whether the predictive power of a PGI is good or useful is ultimately a relative judgment and depends on context. That a PGI often makes inaccurate individual predictions does not necessarily imply that it is useless for profit-maximizing firms operating under uncertainty. After all, a PGI for educational attainment has been shown to be associated with a wide variety of outcomes, including earnings, savings, financial decisions, and wealth.^21,22 Individuals with lower PGIs for educational attainment tend to make poor financial decisions,^23,24 which can be associated with investing in low-quality financial products or being victims of predatory lending practices. One of the distinguishing features of firms is that their objective of profit-maximization is quantitative and quite clearly defined. Tools from economic and operations research can therefore be used to help inform conversations about whether the predictive power of PGIs is large enough to be used in firm-level applications. Depending on the commercial application and the specific PGI involved, even very noisy polygenic measures can be valuable for firms. We are unaware of existing scholarship that assesses such a range of reasonable risks and benefits.

In this perspective, we make the case that PGIs may indeed be informative enough to incentivize firms to use them, which in turn raises ethical and social questions that will necessitate broader and more careful attention (Box 1). We make our case for the relevance of PGIs to real-world actions by firms in two ways. First, we offer a very simple model, rooted in economic theory, of a profit-maximizing firm that can gain information about a single consumer’s genome. We use the model to show that a firm would be willing to pay more for genetic measures that provide more precise information. Crucially, however, depending on the specific economic environment faced by the firm, it would be willing to pay for noisy measures, even if they allow for only a small reduction in uncertainty. We provide a specific calculation characterizing how much firms are willing to pay for a customer’s PGI in the model. Under reasonable parameter assumptions, firms would be willing to pay costs that are comparable to the current costs of genotyping. Second, we describe two plausible scenarios in which different kinds of firms (insurance and financial services) could conceivably use PGIs to maximize profits in ways that pose difficult ethical and policy challenges. Although the sorts of ethical and social questions raised by some types of firms using monogenetic clinical information are well rehearsed, our contribution is to make the case that a wide range of firms will likely be interested in PGIs covering a wide range of phenotypes.

Box 1. Key points.

• (1)Polygenic indexes (PGIs) can provide important information about people pertaining to a vast array of clinical, behavioral, and social phenotypes.
• (2)Some believe that PGIs will be of little value to firms (e.g., those selling insurance and financial services) because they are noisy.
• (3)Using economic modeling, we show that firms nevertheless might be willing to pay for a potential consumer’s PGIs, since this could reduce the uncertainty under which they operate.
• (4)Some uses of PGIs by firms could be in consumers’ interests, while others would not.
• (5)Determining whether particular uses of PGIs by firms are ethically acceptable or unethical necessitates careful analyses in specific cases.
• (6)Laws to prevent unethical uses of PGIs are almost entirely lacking in large parts of the world, and are not panaceas even in jurisdictions like the US where relevant legal action (such as the Genetic Information Nondiscrimination Act of 2008) has been substantial.
• (7)Although there are many uncertainties in exactly how firms might use PGIs, and with what effects, it is essential to start a comprehensive discussion now about the ethical and legal implications across a range of possible scenarios.

A stylized model of a firm’s willingness to pay for genetic information

We assume there is a single firm that enters into a contract with a consumer indexed by i to supply q units of a good to be sold at the prevailing market price of p. The firm chooses q to maximize profits, which consist of the revenues generated by the contract net of its cost. Given a choice of q, the firm earns revenue of p × q. However, the cost of producing this level of output or providing the service depends both on q and on an unobserved consumer characteristic θ_i that is randomly distributed in the population. We further assume that the cost of supplying q units to customer i is C_i = (1/2)θ_iq². In our baseline case, the firm is uncertain about the consumer characteristic but knows the distribution of ln θ_i in the population. Thus, the firm must choose q without knowing the customer’s specific value of ln θ_i. In this kind of environment, firms will maximize profits by choosing larger contracts when the price is higher and when the mean and variance of ln θ_i in the population are lower. A lower mean implies the consumer generates lower costs on average while a lower variance means that the firm faces less uncertainty about the consumer characteristic ln θ_i, both of which increase how much the firm is willing to agree to supply in the contract. For full details on the model, including specific equations and assumptions along with more precise definitions of terms, such as “random,” see supplemental note.

Although the model presented here is heavily stylized, it captures some important tradeoffs present in markets where genetic information could be of interest to firms. For example, in consumer credit markets, we could think of q as the size of a loan made to a consumer, and θ_i could reflect a propensity for late payments or default. In the case of health insurance, q might represent units of coverage with θ_i representing the individual’s unobserved medical needs. In both cases, firms commit to providing a basket of services without having perfect knowledge about traits of the individual that will substantively affect the costs of providing those services.

We next ask how much a firm would be willing to pay to access a consumer’s genetic information in the simplified world of our model. First, focusing on ln θ_i as our phenotype of interest, we suppose that there exists a PGI that predicts ln θ_i. We assume that ln θ_i can be written as a linear function of a PGI and an additive error term. Depending on the setting, this PGI could account for more or less of the variation in ln θ_i.

Imagine that, before choosing q, the firm has the ability to observe the value of an individual’s PGI and then choose q based on this information. This information is valuable, as the firm would optimally supply more (less) q to individuals with lower (higher) values of θ_i. As we show in the supplemental note, this choice is expected to increase profits by a multiplicative factor that depends on the fraction of the variance in ln θ_i that the PGI can account for statistically, i.e., how much the PGI can tell us about θ_i.

Figure 1 plots the firm’s willingness to pay for a customer’s PGI, expressed as a fraction of the profit per consumer the firm could expect without access to the PGI. The only factors that matter for this calculation are the overall variance of ln θ_i and the fraction of that variance accounted for by variation in the PGI. Movement along the x axis corresponds to different levels of predictive power for the PGI, while the separate lines repeat this exercise for different total levels of variation in ln θ_i. These calculations suggest two results. First, unsurprisingly, as the explanatory power of the PGI rises, the firm’s willingness to pay for learning it rises. Second, for the same explanatory power of the PGI, the firm is willing to pay more for access to genetic data if the overall variance of ln θ_i is higher. Again, this should be intuitive. If there is little variation in θ_i across individuals (e.g., if all individuals have the same risk of late payments or developing a costly illness), access to the PGI will not be worth much to the firm. However, in an environment with more uncertainty about individual traits, the firm is willing to pay more for information that resolves some of that uncertainty.

Figure 1.

Firm willingness to pay for genetic information

The x axis shows the incremental R² of a PGI, i.e., the fraction of the total variance of the phenotype ln θ_i that can be explained by the PGI and ranges from 0 to 0.15, where 1 would mean that the PGI is fully informative about ln θ_i. The y axis shows firm willingness to pay (as a fraction of total profit) for the PGI. Plotted lines show the relationship between the incremental R² of a PGI and firm willingness to pay for the PGI for different assumptions on the variance of ln θ_i. If ln θ_i has a low variance (i.e., if all individuals have ln θ_i of similar magnitudes), firms would not be willing to pay a substantial amount for a PGI, even one that is informative since there is little of value to learn. This is illustrated with the flatter lines. In contrast, if ln θ_i has a high variance (i.e., if individuals have ln θ_i of very different magnitudes), firms would be willing to pay a substantial amount for a PGI, increasingly so as it provides more information as illustrated by the steeper lines.

To put the results of Figure 1 into quantitative perspective, consider the market for individual health insurance policies. The Kaiser Family Foundation estimates that the average profit per member per year for such policies was $779 over the period 2016–2018 (https://www.kff.org/report-section/financial-performance-of-medicare-advantage-individual-and-group-health-insurance-markets-issue-brief). For simplicity, think of an insurance contract that lasts ten years, generating an expected undiscounted profit of $7,790. Suppose there exists a PGI that could predict 1% of the variation in the trait ln θ_i linked to medical expenses. Figure 1 suggests that, for a scenario where the variance of ln θ_i is 1, the firm would be willing to pay 1% of total expected profit, or about $78 in our simple example. Note that this already exceeds the cost of sequencing a DNA sample, which can cost around $50. While this will be inflated by other costs associated with getting usable data, the cost of genotyping has been falling in recent years and could continue to decline.⁷ That is, even when a PGI can explain a tiny fraction of the variance of a trait, firms may still be willing to pay to access that genetic information. In short, PGIs are noisy but still provide enough information to make obtaining them profitable under reasonable assumptions about firm behavior in the very real context of health insurance.

We also note that a firm’s decision to seek genetic information may depend not only on the costs associated with genotyping individuals but also on other costs that may be related to social norms and relevant laws and policies. Many firms will discount the profitability of using PGIs by any reputational cost they anticipate this might entail; calculating that cost will depend, in part, on how society comes to view firms who use various PGIs in various ways as well as firms’ calculations of the probability that their use of PGIs will become widely known. Alternatively, firms could decide that the most profitable course would be to use PGIs to target customers with appropriately tailored, if less profitable, products in hopes that advertising their use of this approach would be a promising strategy for gaining market share.

A natural question is whether firms would incur costs to access genetic data given the availability of other kinds of predictive information already widely used in product development and marketing. There may be several advantages to combining genomic information with other predictors (e.g., demographic characteristics) compared to using those other predictors alone or measuring phenotypes directly (see Box 5 in Meyer et al.⁷). Adding PGIs to other predictive information often boosts total predictive power. However, a particular phenotype may be unobservable or otherwise unavailable for measurement. Even when direct measurement is possible, it may be expensive. By contrast, firms need only obtain an individual’s genome sequence once in order to have data that can then be used to predict as many phenotypes as there are PGIs, which will themselves become more predictive over time. Direct measurement also primarily speaks only to the moment of measurement, whereas genomic information can predict past and future phenotypic states (say, past or future weight or adult or geriatric diseases that have not yet manifested). For instance, a firm could track an individual’s behavior over time (e.g., late payments) or they could collect and analyze their PGI only once (e.g., for AD and related dementias [ADRD], which is very predictive of future cognitive decline).

Two scenarios

We now discuss two plausible scenarios where firms might seek genetic information to reduce their uncertainty about consumers.

Insurance

Firms sell insurance to consumers who are willing to buy it in case something goes wrong. Several types of insurance, including health, life, and long-term care insurance, are relevant to the multitude of PGIs that predict the risk of developing a potentially costly disease or engaging in a costly behavior. A key characteristic of insurance is that individual consumers are very different from one another in ways that firms do not fully observe, which can affect how firms price insurance, the profits they can expect, and thus, the viability of their business. If an insurer overestimates the costs they are likely to incur by insuring an individual, they may offer less coverage or charge a price that is too high relative to the consumers’ true odds of needing a payout. If so, consumers may decide not to buy the insurance even though, with more information, a mutually beneficial insurance plan is plausible. In such a case, the insurer loses out, but so can consumers who would like to buy insurance that is priced to reflect the actual risks they pose. On the other hand, insurance firms that lack information about potential consumers may offer too much coverage or charge a price that is too low. For example, a person who is very unhealthy may buy expansive life, health, or disability insurance at a low cost if the firm is unaware of this individual characteristic. If so, firms may exit the market altogether or avoid markets where this type of underpricing occurs too frequently.

If firms knew more about their potential consumers, they could tailor their insurance offerings to individuals and the risks they pose. Consider, for example, the case of long-term care insurance. Many individuals would benefit from such an insurance product but not at prevailing prices, which pool risk with people who may be at much greater risk of massive payouts over long periods of time. If more information were available or shared, there is a possibility that insurance contracts would be written that otherwise would not be. For example, a firm might be willing to offer long-term-care insurance to people who may eventually experience physical limitations but would prefer to avoid consumers who are likely to experience cognitive decline or dementia. Individuals could gain access to such contracts by providing genetic information indicating low risk of cognitive decline. Whether or not this type of insurance contract is something that ought to be permitted is a reasonable question, but there is no reason why such an exchange of information should be rejected out of hand as either implausible or obviously unethical, especially if it could benefit consumers with low risk of ADRD who are nevertheless still vulnerable to other consequences of aging. Alternatively, more information about future risk could allow firms or policymakers to introduce preventative measures that lower costs and thus the price of insurance. In this sense, more information, such as that which could be gleaned from PGIs, could lead to better-functioning markets and more options for consumers with unique sets of needs.

Nevertheless, there are manifold risks of either party incorporating genetic information into insurance markets. Individuals who learn that their own genetic propensity for costly outcomes is higher than expected could hide that information and seek more insurance coverage to prepare for likely health events. If this happens too much, insurers may raise prices for all potential consumers or exit markets altogether. Another risk is that insurers might charge prohibitively high premiums to those who pose greater risks or decline to insure them at all. It is important to note that such practices can effectively take place even in situations where individuals do not disclose their genetic information to firms. For example, suppose that relatively healthy or low-risk individuals voluntarily disclose their genetic information to insurance firms, who in turn offer them expanded coverage or lower premia. In this situation, if a large enough proportion of individuals voluntarily disclose, insurers might infer that the people who do not disclose their information have something to hide and raise premia on anyone who chooses not to disclose. In this way, the actions of profit-maximizing firms can generate complicated externalities—the choices that individuals make with their own private information can end up affecting others even when these secondary actors directly reveal nothing new about their own traits.

The ethical issues here are profound, ultimately forcing us to confront questions that go beyond genetics per se and instead relate to the purpose of private insurance and the proper role of government in our lives.^25,26,27,28 For example, if we could use genetic information to determine that some individuals are likely to experience or cause costly outcomes (e.g., AD or, to use an example unrelated to health, property damage), should private insurers be allowed to turn them away? Alternatively, should firms be forced to pay for treatment and replacement, thus raising prices for consumers unlikely to require payouts? If not, should governments cover these costs? Both within and outside of the health context, uncertainty allows the market to price insurance without knowing who specifically will be associated with costly events and when. However, as genetic information about medical conditions and social behaviors increases, and as uncertainty about future outcomes (e.g., health, accidents, and retirement savings) declines, there is a possibility that markets in at least some kinds of insurance will unravel. That is, the rationale for sharing risks across populations disintegrates as firms and consumers gain greater certainty about the risk that each individual brings to the pool. While market unraveling is far from a certainty, and it is not clear if the advent of PGIs and their continued development and increases in predictive power versus other forms of information will be the catalyst for market changes, we posit that the genome-wide association study revolution has led to a rapid and massive increase in what information insurance firms, which operate based on their ability to price risk, can inexpensively learn about the consumers. It would be shortsighted to postpone a serious conversation about the potential market and policy implications. For example, it would be unwise to wait until markets unravel, low-risk individuals pool together at lower prices, and high-risk individuals are left without insurance and thus vulnerable to health or other shocks to begin a discussion of how policy might respond.

Financial services

Financial services firms offer myriad vehicles for consumers to save and borrow money, ranging from basic checking accounts and mortgages to complex investments tailored to individuals’ risk profiles. Firms generally charge fees for these services. Research provides evidence that differences in PGIs for educational attainment are associated with differences in financial decisions and the returns to savings.^21,22 Other PGIs for outcomes such as general cognition, personality traits, and ADRD are also associated with aspects of financial decision making.^23,24 In this scenario, we focus on recent research relating PGIs to cognitive decline and disadvantageous financial decision making. Specifically, it has been shown that individuals who eventually exhibit cognitive decline or dementia begin to make errors in their financial decisions years before being diagnosed with AD.²⁹ Other research shows that individuals with a high PGI for AD along with a low PGI for educational attainment are rarely diagnosed with AD even though they experience cognitive decline and also exhibit many of the economic consequences of it, such as lower income, earlier retirement, and less wealth at retirement.³⁰ The same research also shows that such individuals tend not to take precautionary measures (e.g., preparing a witnessed will) to secure their financial future; indeed, they tend to be less likely to do so versus people with higher educational attainment and lower AD risk scores. In other words, the individuals who are more likely to benefit from precautionary measures may be the least likely to take them.

These findings suggest a crucial and potentially beneficial role for firms and government policy. First, individuals who recognize their financial vulnerability (or perhaps that of a family member) would likely be interested in purchasing products that are financially protective. For example, a firm could develop products that reduce financial decision-making burdens for individuals at high risk of developing ADRD or direct their assets into low-risk investment vehicles. Second, firms may offer services monitoring expenditures that could reflect confusion or errors. For example, a financial firm could offer a service whereby a designated family member is alerted when an individual at increased genetic risk of cognitive decline makes large or irregular financial transactions that may be due to fraud or scams that could deplete savings. With good policy, more genetic information could compel firms or the government to offer products that serve, rather than thwart, people’s needs.

Risks, however, abound. If a firm knows that a consumer has difficulty with complex financial decisions, it could develop products that take advantage of this information. For example, firms could develop overly complex products that may appear to be good investment vehicles to preserve or increase wealth but have hidden risks, fees, or costs that consumers who are likely experiencing the very early stages of dementia might be expected to miss. Indeed, there is ample precedent for deliberately targeting vulnerable consumers by offering them lousy products.³¹ It is not unfathomable that firms could use genetic information to intensify such practices. Another concern is that those who purchase products to help genetically related family members at high genetic risk of cognitive decline may unwittingly be sharing information about themselves, since a financial firm may flag them as also being at high risk. This could affect what kinds of products, financial vehicles, and loan terms they offer to individuals who never shared direct genetic information about themselves but did share their relationship with someone who had. This perspective is meant, in part, to prompt a consideration of what kinds of safeguards against such practices might be warranted.

Ethical and policy issues

Of substantial concern is that the current legal and policy landscape is not adequate to meet all of the challenges that will arise should firms avail themselves of PGIs. While there are some policies regarding how firms may use genetic information in Australia, Asia, and South America, this does not seem to be the case in the Middle East and Africa.³² In addition, critical gaps remain even in North America and Europe where policy-making activity has been substantial. In the US, for instance, the Genetic Information Nondiscrimination Act of 2008 (GINA) prohibits genetic discrimination in employment but omits the 85% of US firms that employ fewer than 15 people (https://www.census.gov/data/tables/2020/econ/susb/2020-susb-annual.html). GINA’s prohibition on genetic discrimination in health insurance similarly does not apply to other kinds of insurance, including long-term care, disability, life, home, and property insurance. Some US states extend GINA’s prohibitions, but these laws vary widely in scope and strength (https://www.genome.gov/about-genomics/policy-issues/Genome-Statute-Legislation-Database). This patchwork of insurance and employment genetic anti-discrimination laws leaves untouched many other decisions firms make, such as which kinds of services to offer to whom and under what conditions.

We also note that firms’ use of PGIs will likely raise important ethical concerns that have not necessarily been resolved by earlier conversations regarding firms’ use of non-genomic predictors or even monogenetic information. While genetic information is in many ways like any other information (e.g., gender, race, smoking status, or education) that firms might use to customize or market products, the genetic information captured in PGIs has features that merit close consideration. For one, PGIs are complicated statistical measures. Their interpretation requires care, and there are many choices that must be made about how to properly use them in statistical analyses (e.g., controlling for population stratification, using within- or across-family variation, determining under what specific conditions and with what data associations can be interpreted as causal). The complexity of these measures—relative to demographic information or even single genetic markers—increases the risks that firms may misinterpret PGI associations when using them to draw conclusions about individuals. Individuals are unlikely to have a good understanding of the potential implications of PGIs and the kinds of interpretational errors that firms may make when they are used to draw conclusions about them. Moreover, individuals may not be adequately informed about potential future uses of their genetic or other personal information. For example, if an individual chooses to share information about their gender, they at least have some idea of what information is being transmitted. However, if they upload their genetic data to a website, they may not realize how much or what type of information they are revealing, especially if they do not realize that genes are associated with not only phenotypes like disease and height but also social and behavioral outcomes. Furthermore, that information will change since new PGIs can be constructed as they are developed using the same genetic information, and genetic information provided today can be far more predictive of behavior and outcomes in just a few years. By providing genetic data today, individuals are thus providing access to whatever information technological advances can glean from it in the future. Finally, as is true in many contexts, if an individual provides genetic information about themselves, they are also, perhaps unwittingly, providing information about their genetic relatives, whom firms may be able to link to them. While all of these risks exist with other, non-genetic types of information due to intergenerational persistence in many behaviors and characteristics, the genetic links across generations are undeniable; indeed, such intergenerational persistence is often partly attributable to genes. An objective of starting this discussion in this perspective is to promote a more widespread understanding of what individuals are giving away when they provide their genetic information to firms. This could perhaps prompt some individuals to be more wary of sharing their information and more eager to see policies put in place that allow individuals to “claw back” or prohibit certain uses of the information they have already given.

Conclusion

Existing genetic antidiscrimination laws were largely enacted to reassure people who were otherwise interested in undergoing genetic testing, not as a response to actual incidents of genetic discrimination, which were and remain rare.³³ That, however, might change. Whereas the vast majority of people have no genetic risk for rare monogenic diseases, a PGI can be calculated for anyone and predicts a wide range of outcomes. As PGIs become more precise, they will become more useful to firms. Although some PGIs will never be strongly predictive of any individual outcomes, as our model illustrates, they need not be extremely precise to be useful to firms seeking to reduce uncertainty and increase their profits. Indeed, existing PGIs may already be of use to firms given increases in statistical power and decreases in the cost of genotyping.

That said, although we make the case that firms may be interested in PGIs, we acknowledge that precisely how firms will use PGIs, and with what ramifications, is uncertain. Consider, for instance, the two areas noted above that have already received some ethical and legal attention: firms that sell PGI reports to IVF patients to inform transfer decisions and firms that sell PGI reports to adult consumers. Whether polygenic embryo selection will impact population demographics or significantly exacerbate inequalities (two concerns raised in the literature) depends, in part, on the accessibility and uptake of this service and on the aims of consumers who use it (e.g., whether to use PGIs to prioritize embryos for transfer or to reject some embryos, forgoing live births). Similarly, whether selling adults their own PGI reports for a variety of traits, including sensitive phenotypes such as intelligence and educational attainment, will cause the harm that some scholars have anticipated depends, in part, on whether individual consumers hope to use these reports for anything more than “infotainment.” So, too, in the cases we discuss here. For instance, as we mention, many firms will discount the profitability of using PGIs by any reputational cost they anticipate this might entail. Calculating that cost will depend, in part, on how society comes to view firms who use various PGIs in various ways—which is uncertain—as well as firms’ calculations of the probability that their use of PGIs will become widely known. Depending on these variables, firms could decide that the most profitable course would be to use PGIs to target customers with appropriately tailored, if less profitable, products in hopes of advertising their use of this approach as a strategy for gaining market share.

In this perspective, we provided two scenarios in which firms might plausibly seek genetic information from individuals. We have not specified how firms might acquire such information, but it is possible to imagine that they might simply request access to such data from consumers (if individuals have already had their genome sequenced for other purposes), buy these data from third-party data brokers, or incentivize consumers to undergo genotyping by offering in exchange access to a better array of products or lower prices. If consumers are unaware of the value of the information they are thus transmitting, they may be willing to do so for very little in return. Regardless of how firms obtain genetic information, in each scenario, we considered how the same genetic information might be beneficial and harmful for firms and individuals. These observations are conjectural at present but just barely; the scenarios we suggest are more likely to be realized as the predictive power of PGIs continues to grow. As this power increases so will the need to assess the social benefits and harms associated with firms purchasing PGIs. We hope that together with deep stakeholder engagement, our argument for the plausibility of firms’ use of PGIs along with descriptions of just a few of the ways in which firms might use them will help facilitate further discussion aimed at evaluating the overall social benefits and harms.