We developed and applied the following inclusion criteria to identify and select eligible studies: published in English; focused on the collection or capture of information on the occurrence of GD in life insurance whether direct (through patients' or participants' self-report, such as interviews or surveys) or indirect (through professionals such as doctors, genetic counselors and insurers); focused on a primary or follow-up study (multiple publications on the same study were grouped together and treated as a single study). Narrative and systematic reviews were included only if they also presented additional empirical data on the occurrence of GD in the context of life insurance (that is, other than the data contained in the studies reviewed therein).

Eligible studies were excluded if: they did not focus in whole or part on the occurrence of GD in the context of life insurance; they offered insufficient evidence; or they contained serious methodological flaws. Editorial letters with no primary data, comments, opinions, abstracts and unpublished studies were excluded. Studies capturing the fear of GD rather than actual experiences were also excluded.

Searches were conducted from March until May 2012 on PubMed, Google Scholar, Social Science Research Network and Hein Online, using the Boolean operators 'OR' and 'AND' with various permutations of the following keywords: 'genetic discrimination', 'study', 'life insurance', 'survey', 'data', 'empirical evidence' and 'genetic test'.

First, we conducted keyword searches that produced a list including 534 search results. Using the terms 'genetic discrimination' and 'life insurance' combined ('AND'), we obtained a list of 29 publications from the database PubMed. We executed a search on Google Scholar including all the following keywords combined ('AND'): 'genetic discrimination', 'study', 'life insurance', 'survey', 'data' and 'empirical evidence'. This produced 155 search results. Our search on Social Science Research Network using the terms 'genetic discrimination' yielded 100 results. Our last search was conducted on Hein Online using the following combination of terms '("genetic discrimination" AND "life insurance") AND "genetic test" OR "data" OR "empirical evidence" OR "survey"', and generated 250 results.

Second, applying our selection criteria, we reviewed the titles and available abstracts of each of the 534 search results identified through our keyword searches, and retained 29 publications.

Third, we performed a systematic hand search in the references list and bibliography of each of the 29 articles to identify additional relevant publications and assess cross-referencing. This led to 16 more eligible studies. Hence, we uncovered a total of 45 eligible studies relevant to GD in the context of life insurance.

Thereafter, eligible studies were each independently assessed for relevance by two researchers (YJ, INF). Applying the selection criteria above, we retained studies documenting experiences of discrimination through both direct and indirect evidence on the occurrence of GD. Based on these criteria, 12 studies were excluded. This final list was also compared with references and resources provided by a recent review dealing with GD in numerous contexts including life insurance 9. The 33 studies retained represent a systematic overview of the empirical evidence of GD in the field of life insurance. For an overview of the search strategy, see Figure 1.

The vastly different nature of the available studies suggested that formal statistical analysis and comparison of discrimination cases would be inappropriate. Instead, we analyzed the data through a social science comparative approach that incorporates both quantitative descriptive analysis and qualitative content analysis. Key elements of the selected studies were coded independently according to their relevance to pre-selected themes. We extracted information on study scope (country and context), genetic conditions (whether the study focused on a single or multiple conditions), definition of GD (how GD was construed by the researchers), validation (whether a formal validation process was disclosed by the authors), conclusions (how results were qualified: whether evidence of GD was found, whether policies or laws were required), and number of citations (how often the publication was cited by peers). Two trained researchers (YJ, INF) independently evaluated the results and discrepancies were resolved by consensus. Results obtained on individual themes were qualitatively analyzed and, where appropriate, converted to statistical data (values were rounded according to convention; this process was based on the information presented in Table 1121314151617181920212223242526272829303132333435363738394041424344).

Together, the 33 studies represent two decades' worth of research published from 1991 to 2012 (see Table 1). A small peak in the number of studies (eight studies) can be identified between 2006 and 2007, which can be linked to the charged political climate in the US and renewed academic interest that prefaced the adoption of GINA. These studies generally gathered evidence through direct sources (that is, individuals commenting on their own experience with insurance companies), but some obtained their data through secondary sources such as insurers, insurance associations, health professionals or patient groups.

The majority of studies reviewed (73%) aimed at providing evidence of GD in a variety of fields (for example, employment, immigration, adoption and access to healthcare), including that of life insurance. A minority (27%) focused exclusively on the context of life insurance (see Table 1). The majority of the available evidence (58%) comes from studies involving North American population groups (see Figure 2 and Table 2).

Surprisingly, given the rather strong European policy response mentioned above, only six studies (18%) provided empirical data on the situation prevailing in continental Europe, and five of these studies were very specific in nature, addressing the context of familial hypercholesterolemia and hypertrophic cardiomyopathy in the Netherlands, hereditary breast and colorectal cancer in Norway, and Huntington's disease in Germany 24384044.

The substantial number of studies carried out in Canada (seven studies), where no specific laws have been adopted to limit the use of genetic information by life insurers, and the absence of studies in a highly legislated European context, could suggest that the number of GD studies carried out in a given country, as well as the number of GD cases reported, does not necessarily have a strong correlative impact on policymaking. A notable exception to this trend could be Australia, whose Disability Act was amended following the publication of major studies on GD and an extensive report from the Australian Law Reform Commission 45.

The mitigated overall results are not easily interpretable and the task is exacerbated by the serious methodological challenges faced in some of these studies. One of the constraints concerns the range of genetic diseases investigated in the literature. Reviewing the scope of the 33 studies, it is apparent that they only uncovered evidence on a very limited number of highly penetrant, familial, adult-onset, relatively well-known genetic conditions.

The majority of the evidence is based on the following five conditions: Huntington's disease, hereditary breast and ovarian cancer, hemochromatosis, familial hypercholesterolemia and hereditary colorectal cancer (see Figure 3). Of the 33 studies reviewed, 19 (58%) specifically focus on one of these five conditions (see Figure 4). Moreover, evidence on genetic discrimination in the context of Huntington's disease is provided in over 14 of the 33 studies, and on hereditary breast and ovarian cancer in more than 10 (see Figure 3).

The high number of studies focusing on a single condition (see Figure 4) makes it particularly difficult to generalize from the results of a systematic comparison of the literature to reach a broad, robust conclusion on GD applicable to the whole research or clinical genetic context. Moreover, research on GD in the fields of personalized medicine and/or pharmacogenomics, infectious diseases and genome-wide association studies remains absent in the literature, thereby resulting in a complete lack of data on GD in the context of emerging 'omics' research. This is particularly concerning given that the amount of genomic information in the typical individual's medical record is likely to increase tremendously in the next few years as whole-genome sequencing costs are reduced and personalized medicine becomes more common in clinical settings 46.

The authors of the 33 studies all struggled with the meaning of GD (examples of GD definitions are provided in Table 3). Several studies refrained from using the term 'genetic discrimination' in their questionnaire so as to avoid biasing responses. However, the paradoxical consequence of this methodological approach was over-reporting due to the tendency of participants to declare any negative outcome they faced while applying for life insurance as an instance of discrimination 47. When specifically included in survey questionnaires or data analysis strategies, the definition of GD varied widely (see Table 3), greatly reducing the possibility of meaningful comparison. Indeed, the challenge of defining GD led a study author to conclude that '[t]he notion of finding wholly objective and overt evidence, as opposed to subjective and implicit accounts of discrimination may [n]ot be entirely realistic' 42. Studies choosing to adopt a broad definition, or no definition at all, tended to report the highest incidence of GD cases 36. Studies using a legal definition of GD obtained lower results 3. But, because laws on this topic vary significantly across jurisdictions, these studies are difficult to compare or integrate with one another outside of their national context.

To obtain more robust and comparable results, some studies have used the criteria of 'irrational discrimination' - discrimination that is not based on scientifically validated and actuarially relevant genetic information - as a selection criterion to assess the practice of insurers. However, because negative decisions by life insurers against some of the most genetically at-risk individuals who might have pressing need to obtain life insurance (for example, an asymptomatic patient having tested positive for a monogenic dominant serious condition) would not necessarily constitute irrational GD, use of this criterion could arguably be perceived as unethical. The partly subjective nature of the underwriting process (illustrated by the high variability between the guidelines and questionnaires used by different insurance companies) and our limited knowledge of the genomics of complex diseases further limit the use of the rationality criterion to determine objectively the prevalence of GD in insurance.

The context of Huntington's disease can be used to illustrate the impact of definitional choices on the results of GD studies. Because this disease is a relatively well-known autosomal dominant genetic condition, obtaining a positive test result has serious implications for the future health of an asymptomatic individual. This explains the life insurers' interest in being able to use test results or family history information, regarding this particular disease, for underwriting. This is in turn reflected in the results of GD studies. Studies investigating GD in the context of Huntington's disease and using a broad definition of GD or no definition at all would likely identify a significant number of GD cases (exclusion, higher premiums or conditional acceptance). However, studies using a rationality or legality criteria would generally report a much lower number of discrimination incidents.

Around half of the studies reviewed (48%) found that, although GD had some empirical basis, its incidence was rare and it was not a significant source of insurance denials 414161719202122273139404244. A second category, comprising a considerable number of studies (42%), concluded that the existence of GD in life insurance was documented by the evidence they provided and that the situation gave grounds for serious concern. Within this category, Huntington's disease came up often 36373843. Early US studies in this category often advocated the adoption of laws and policies to prohibit access to genetic information by life insurers 17 or the development of a more generous public insurance system that would provide a minimum amount of life insurance to all applicants 16. Finally, a minority of studies (9%) found no empirical evidence to support the existence of GD in the life insurance context.

It should be mentioned that some, but not all, of the countries covered by these studies, have already adopted laws prohibiting insurers' access to genetic information (for example, the Netherlands, Norway, and Germany) (see Table 1). Authors of the 9% of studies finding no empirical evidence of GD were often of the opinion that the GD problem was more linked to media hype and fear of discrimination than to GD itself. These studies consequently pointed out the importance of educating the public and reassuring the concerns of patients and research participants about GD 2240.

Among the 19 studies dealing with a single genetic condition (see Table 2), a significant number of studies (47%) concluded that there was sufficient evidence to raise serious concerns about GD. Half of these studies concerned Huntington's disease 36373843. A second important category (42%) found that, while GD existed, it was of rare occurrence 1931394044. Finally, a minority of studies (11%) concluded there was no evidence of GD 2528.

To highlight the broad trends, it is possible to further group the 33 reviewed studies into two categories: a majority of studies (58%) that believes that GD in the context of life insurance is a negligible issue that does not warrant the substantial societal debate and policy concern generated to date, while a substantial minority (42%) concludes that GD exists and has impacted access to life insurance negatively.

Validating the study results (that is, avoiding biases and concealments as well as ensuring that the data reflect cases of 'real' rather than 'perceived' discrimination) was another significant hurdle. For the purpose of our research, validation was considered to be any additional independent step(s) or method(s) taken by the researcher to confirm the accuracy of reported discrimination events. A majority of the reviewed studies (76%) could not be considered as validated (see Table 1 and 2); in this case it seems more accurate to talk about studies assessing the 'perceived' level of GD rather than objective manifestations of it. Testifying to the importance of the validation process, Wertz found that, '[W]hen asked to give details of their refusals, almost all [participants] described situations that are characteristic of general insurance practice. They were apparently objecting to what they perceived as unfair insurance practices in general, rather than practices specific to genetics.' 20.

Studies of patients were not the only ones in which investigators noted the importance of verifying findings. Otlowski et al. observed that insurers surveyed on the topic of GD were likely to under-report unfavorable underwriting decisions 3.

To attempt to reduce the biases associated with non-validated results, several verification techniques have been used over the years. They include follow-up phone calls or in-person interviews to elicit additional information about reported cases of discrimination; review of the participant's medical file (to verify if an unfavorable decision could be due to a pre-existing condition); audit of the documentation or correspondence relating to any discrimination complaint; corroboration of discrimination reports by independent sources (ombudsman or similar administrative instances); case law; and so on. These validation techniques were used, alone or in combination, with various degrees of rigor by researchers and these choices significantly impacted the results of the reviewed studies.

Other study limitations that impacted the results and their compatibility with one another included the sample size and type of people surveyed. Some studies would include individuals already affected by a genetic disease but asymptomatic 16, whereas others would include healthy carriers 30. Some would include asymptomatic untested individuals with a family history of disease 37, and some would include information on patients that was obtained from indirect sources (family members, genetic counselors or members of a disease support group) 21. The lack of large-scale studies of well-characterized individuals also made it difficult to extrapolate from the results to objectively estimate the prevalence of GD in the life insurance sector.

Treloar and colleagues have written that, 'conceptualizing, investigating and verifying individual's experience of genetic discrimination constitute a challenging endeavour' 47. Discrimination surely can take many subtle forms. For example, rather than charging a higher premium or excluding an applicant, an insurance company could decide to process an application more slowly or ignore phone calls and emails in the hope of discouraging pursuit of the application process. In this case, the applicant might not even be aware that she or he has been discriminated against.

Given these serious challenges, it should come as no surprise that the five most influential studies on GD within academia, as measured by Web of Science and Google Scholar citation rates (see Table 4), all suffer from important limitations. For example, the most cited article on GD in life insurance, a precursor 1992 pilot study by Billings et al., used a broad definition of GD and reported 29 responses describing 41 separate incidents of possible discrimination (32 in the field of insurance) 13. The study undertook an extensive advertising campaign (1,119 letters mailed to genetic professionals, an advertisement in the American Journal of Human Genetics, and similar advertisements published in several patient organization newsletters) to elicit this relatively small number of potential discrimination cases from an under-defined population that included symptomatic respondents. The authors of the study acknowledged the limitations of their work stating that it is not meant to demonstrate the prevalence or the full range of discriminatory practices. Nevertheless, their conclusion that unfair and discriminatory use of genetic data existed and that new laws and sanctions should be considered does not seem to accord with the limited data and exploratory methodology provided in the study. Learning from early experiences and challenges in this field, more recent studies tend to draw more cautious or qualified conclusions and to recognize their own substantial methodological limitations 32.