ARTICLE

Third-Party Genetic Interpretation Tools:

A Mixed-Methods Study
of Consumer Motivation and Behavior
Sarah C. Nelson,1,2,* Deborah J. Bowen,3 and Stephanie M. Fullerton3

In an effort to meet ethical obligations and/or participant expectations, researchers may consider offering ‘‘raw’’ or uninterpreted genetic
data for result return. It is therefore important to understand the motivations, behaviors, and perspectives of individuals who might
choose to access raw data before such return becomes routine. In the direct-to-consumer (DTC) context, where raw data are often
made available to customers, the use of third-party interpretation tools has raised concerns about genotype accuracy, data privacy, reli-
ability of interpretation, and consumption of limited health care resources. However, relatively little is known about why individuals
access raw data or what they do with the information received from third-party interpretation. Accordingly, we conducted a survey
on raw data access and third-party tool usage among 1,137 DTC customers recruited through social media. Most survey respondents
(89%) reported downloading their raw data. Among downloaders, 94% used at least one tool, most commonly Promethease (63%) or
GEDmatch (84%). More than half (56%) used both health-related and non-health-related tools and differed significantly from those
who used only one tool type in terms of demographics, participation in research, DTC tests ordered, and testing motivations. Explor-
atory interviews were conducted with 10 respondents and illustrated how social networking, initial lack of interesting findings, and
general curiosity contributed to use of multiple tool types. These results suggest that even when initially motivated by ancestry and
genealogy, consumers frequently also pursue health information in a largely unregulated and expanding suite of third-party tools,
raising both challenges and opportunities for the professional genetics community.

Introduction However, the implications of making raw genetic data

There is a widening consensus that researchers should,
wherever possible, offer individual results to research par-
ticipants, both to enhance autonomy, agency, and reci-
procity of participants and to increase engagement with
and contribution to both traditional and citizen science
research initiatives.1–3 While the return of results has
been the subject of scientific, ethical, and legal debate for
some time,4–6 the discourse has focused mostly on inter-
preted results rather than ‘‘raw’’ or uninterpreted genetic
data. Here, raw data refer to genotype calls from either
array genotyping or sequencing but may also encompass
upstream formats such as the sequence alignment data
used to make genotype calls (e.g., BAM files).1 Empirical ev-
idence suggests that members of the general public would
want their raw genetic data if they were to participate in a
genome sequencing study,7 and some investigators have
begun offering such data to research participants.1 Indeed,
organizers of the National Institutes of Health ‘‘All of Us’’
research cohort have stated plans to make raw genetic
data available to participants,8,9 and the National Acade-
mies of Sciences, Engineering, and Medicine (NASEM)
have issued recommendations that are supportive of re-
turning ‘‘individual research results’’ (both raw data and
interpreted results) where feasible.10 Therefore, under-
standing how individuals make use of their raw genetic
data is important for genetic researchers considering the
return of such information to participants.
available to research participants remain largely unknown.
One avenue to explore potential outcomes is to examine
current uses of raw genetic data from direct-to-consumer
(DTC) genetic testing, which to date has been the most
common route of access. Broadly, there are two types of ac-
tivities DTC customers can undertake with their raw data.
The first is to contribute to collective research efforts by
sharing genotypic and phenotypic data on platforms
such as Open Humans and openSNP.11,12 The potential
benefits of such sharing include increasing the size of data-
sets available to traditional researchers and creating novel
opportunities for participant-driven or ‘‘citizen science’’
research initiatives.13,14 The second type of activity is the
pursuit of individually tailored information via third-
party interpretation (TPI) tools such as Promethease or
GEDmatch. As we found in a prior study,15 TPI tools are
quite heterogeneous in terms of types of information re-
turned (e.g., health, ancestry, genealogy), methods for
generating that information, and cost to users. Further
complicating this picture is that some TPI tools also pro-
vide opportunities to contribute to collective research ef-
forts (e.g., openSNP and DNA.Land).
Recently, TPI tools have received growing attention in
both academic and lay media discourse, where concerns
have been raised about genotype accuracy, data privacy
and security, reliability of health-related information, po-
tential for false positives or false negatives, and down-
stream consumption of limited health care resources. For

1
Institute for Public Health Genetics, University of Washington, Seattle, WA 98195, USA; 2Genetic Analysis Center, Department of Biostatistics, University
of Washington, Seattle, WA 98195, USA; 3Department of Bioethics and Humanities, University of Washington School of Medicine, Seattle, WA 98195, USA
*Correspondence: sarahcn@uw.edu
https://doi.org/10.1016/j.ajhg.2019.05.014.
Ó 2019 American Society of Human Genetics.

122 The American Journal of Human Genetics 105, 122–131, July 3, 2019
example, media reports have recounted individuals
receiving distressing results from TPI, which upon clinical
confirmatory testing turned out to be false positives.16–19
These individuals experienced emotional and financial
hardship in unnecessary follow-up, often tracing back to
errors in the raw DTC genotypes. Accounts in academic
literature have illustrated outcomes of bringing TPI results
to genetic counselors (GCs),20 genetics specialty clinics,21
or otherwise leading to follow-up clinical sequencing.22
In addition to health-related concerns, novel uses by law
enforcement of online genealogy databases such as
GEDmatch are raising questions about data privacy and
consent in third-party services.23,24
However, few studies have reported on the perspectives
of the tool users themselves. One survey of DTC customers
found a high volume of raw data download and TPI tool
use (67%, or 321/478 respondents), where the majority
of tool users (81%) were satisfied with the information
received.25 However, that study did not explore the rela-
tionships between use of specific tools and reactions,
follow-up actions, and DTC testing motivations. Some
tool developers have also surveyed their own users,12,26
but this provides limited insight into use of raw data and
TPI tools more broadly and does not measure the degree
to which users access multiple tools or make distinctions
among the tools that they use.
Here we contribute new information about consumers’
use of raw genetic data and TPI tools from the results of a
survey of more than 1,100 DTC customers and follow-up
interviews with a subset of survey respondents. Our aim
was to better understand users’ motivations and behaviors
from initial DTC testing through to downstream use of
specific TPI tools and resulting follow-up actions, with
the broader goal of understanding how raw data return
in non-DTC contexts may unfold. This primarily descrip-
tive study contributes novel insights into how individuals
leverage their raw genetic data in multiple ways, often
concurrently, including to learn about health risks,
ancestry, and genealogy.
Subjects and Methods
Participant Recruitment
We recruited survey respondents during October and November
2017 with staggered postings to various social media venues: six
genomics-related sub-Reddits, Twitter, and several Facebook
groups (four genealogy groups and the DNA.Land page). Early in
the 2-month period and once per venue, we posted a brief study
description and link to the survey, after seeking permission from
group moderators/administrators. Additionally, the survey was
sent via newsletter to openSNP27 users and was posted on the first
author’s academic website and the Institute for Translational
Health Sciences (ITHS) Participant Portal. Recruitment messages
stated eligibility criteria as being 18 years or older and having
taken at least one DTC genetic test—i.e., raw data download and
TPI tool use were neither mentioned nor required, though some
of the groups targeted were oriented toward TPI tools (e.g., Prom-
The American Journal of Human Genetics 105, 122–131, July 3, 2019 123
ethease sub-Reddit or DNA.Land Facebook page). No incentives
were offered for participating. The survey was closed 7 weeks after
it first opened and after posting final reminders on each recruit-
ment venue.
Survey Design and Implementation
The survey questionnaire was designed to cover three main topics:
DTC testing, raw data download, and TPI tool use. Respondents
were asked which DTC test(s) they had ordered, when, what moti-
vated them to order the test(s), and whether they had downloaded
their raw data. Non-downloaders were asked about reasons for not
downloading while downloaders proceeded to a series of questions
about TPI tools. Respondents who had used multiple tools were
prompted to select one on which to base their responses. All
respondents filled out a demographics section and were invited to
provide contact information for voluntary follow-up interviewing.
We developed survey questions primarily based on our prior
study of TPI tools and developers.15 However, some items were
based on existing instruments: motivations for DTC testing were
adapted from the PGen (Impact of Personal Genomics) Study base-
line survey,28 and questions on how the respondent learned about
tools were adapted from Wang et al.25 A draft instrument was eval-
uated via cognitive interviewing29 with six individuals recruited
from a combination of personal and professional networks. The
self-administered, online questionnaire was implemented in
REDCap.30 While the majority of survey questions were fixed
response, some open text comment boxes were also included.
Follow-up Interviews
We purposively sampled interview participants from among those
survey respondents who volunteered for follow-up contact and
reported using a specific combination of tools that spanned
health, ancestry, and genealogy (Promethease, DNA.Land, and
GEDmatch, respectively)—or ‘‘crossover’’ use, described further
below. Initially, 179 respondents met these criteria. Because we
wanted to explore crossover in both directions (i.e., from health
to non-health and vice versa), we stratified this sample by ‘‘DTC
test(s) ordered’’ as a proxy for initial motivations for tool use
among potential interviewees, and, within each bin, randomly
sampled participant IDs to contact.
We conducted semi-structured interviews via phone or Skype
and had audio recordings transcribed on Rev (see Web Resources).
The interview guide comprised six questions focused on gaining a
deeper understanding of respondents’ timing and motivations to
download data and use multiple types of tools. For example, did
respondents start by using GEDmatch with an interest in geneal-
ogy and go on to additionally use Promethease and if so, why?
Or, conversely, did initial interest in health and use of Prome-
thease eventually lead to use of genealogy and ancestry-focused
tools as well and if so, why? We conducted a total of ten inter-
views, after which point we had observed multiple examples of
crossover in each direction (i.e., participants initially using health
tools before using non-health tools, and vice versa) and halted
recruitment for this exploratory follow-up phase of the study. In-
terviews averaged 36 min (SD ¼ 14 min); an interview with one
deaf respondent was carried out via email.
Data Analysis
Quantitative survey data were analyzed via univariate descriptives,
bivariate, and multivariate analyses (i.e., logistic regression) using
all available, non-missing data. The total number of potential
responses changes between survey sections due to branching
logic; sporadic missing answers also affect the count of available
responses for any given survey item. Therefore, throughout we
report both percentages and counts. All quantitative analyses
were carried out in R statistical and graphing software.31
Throughout, DTC testing motivations were dichotomized into
‘‘very important’’ versus ‘‘somewhat important’’ or ‘‘not at all
important,’’ following precedent in analyses of the PGen survey
from which those survey items were adapted.32,33 To understand
factors influencing use of specific third-party tools, we performed
a series of logistic regression analyses using tool use (yes/no) as an
outcome and the following covariates: each of the eight dichoto-
mized DTC testing motivations, survey recruitment venue, and
DTC test(s) ordered. The survey instrument, results dataset, and
R analysis code are available on openICPSR (see Web Resources).
We thematically analyzed qualitative survey data from open text
boxes in Atlas.ti v8. Interview analysis was also conducted in
Atlas.ti v8 and focused on understanding how and why partici-
pants came to use tools across the multiple domains of health,
ancestry, and genealogy.
This study was approved by the University of Washington (UW)
Institutional Review Board as minimal risk human subjects
research, protocol #50238.
Results
A total of 1,137 eligible respondents took the survey (see
Table 1), including 268 respondents (24%) who did not
progress to the end of the survey. The most common
recruitment venue was Facebook (624/1,137 or 55%), fol-
lowed by Reddit (357/1,137 or 31%), Twitter (71/1,137 or
6%), and the openSNP newsletter (62/1,137 or 5%). Fewer
than 20 respondents were recruited from each of the re-
maining venues. Below we report patterns of DTC testing
and TPI tool use, including respondents’ impressions of
information received and follow-up actions taken. We
then analyze users by categories of TPI tools, focusing on
the common phenomenon (56%, or 458/819 tool users)
of using both health-related and non health-related tools.
Patterns of DTC Testing
DTC Tests Taken
Respondents reported ordering a range of DTC tests, with
36% (413/1,137) ordering multiple tests: 21% (236/
1,137) ordered two tests, 12% (135/1,137) ordered three,
and 4% (42/1,137) of respondents ordered four or more.
While asked explicitly about 23andMe, AncestryDNA,
and FamilyTreeDNA, respondents noted additional tests
via an open text box, most commonly MyHeritage (n ¼
37), Living DNA (n ¼ 32), National Genographic (n ¼
19), and Genes for Good (n ¼ 18). The majority of DTC
tests were ordered after 2015 (see Figure S1), potentially re-
flecting a rise in DTC test popularity and/or a recency effect
in that those who recently ordered tests were more likely to
be active in our online recruitment venues.
DTC Testing Motivations
The most common motivations for pursuing DTC testing
were general curiosity about genetic make-up (661/972 or
124 The American Journal of Human Genetics 105, 122–131, July 3, 2019
68% rated as very important) and curiosity about ancestry
(645/977 or 66% rated as very important; see Table 2). Less
common motivations were limited information about
family health history (201/972 or 21% rated as very impor-
tant) and other family members pursuing testing (104/969
or 11% rated as very important). In free text responses,
20% of respondents (225/1,137) noted additional motiva-
tions, which included pharmacogenomics; being adopted;
breaking through ‘‘brick walls,’’ or dead ends, in genealogy
research by matching with new relatives;34 and profes-
sional interests (e.g., teaches genetics or is a GC). Notably,
81% of respondents (787/969) rated desire to have their
raw genetic data file as either a very important or some-
what important motivation to pursue DTC testing.
The prevalence of raw data access as a motivation was
borne out in the rate of data download: 26% (252/974)
of respondents reported they had not or were unsure
whether they had downloaded their raw data from at least
one of the DTC tests they had taken (note respondents
were asked about download separately for each DTC test
taken). Of these 252, 148 had downloaded data from
another one of the DTC tests they had taken, leaving 104
(11% of 974) who had not downloaded any of their
available raw data files. Demographically, non-down-
loaders did not differ from downloaders; however, in pur-
sing DTC testing, downloaders were more motivated to
find relatives (b ¼ 0.839, p ¼ 0.0007) and obtain raw
data (b ¼ 1.49, p ¼ 9.62e08) compared to non-down-
loaders (see also Figure S2).
Third-Party Tool Use
Impressions and Follow-up Actions
A total of 820 respondents who downloaded raw data also
reported using at least one tool and formed the basis
for subsequent analyses. Most used multiple tools, with
76% (623/820) using two or more (median number of
tools ¼ 3, max number ¼ 11). Thirteen tools were specified
in fixed-response survey questions, though respondents
could note additional tools via free text. The most
commonly used tools were GEDmatch (n ¼ 688 respon-
dents), Promethease (n ¼ 515), and DNA.Land (n ¼ 450;
see Table S1). Additional tools most frequently noted in
the open text box were WeGene (n ¼ 39), FamilyTreeDNA
(n ¼ 35), and MyHeritage (n ¼ 34).
Respondents’ impressions of the information they
received varied by tool (see Figure 1). Promethease users
overwhelmingly agreed they received information about
health (171/186, or 92%) and about risk of specific disease
(180/186, or 97%). Notably, some respondents indicated
receiving information outside a tool’s scope: 16% of
GEDmatch users (83/505) agreed they received some
health information while 37% of Promethease users (69/
188) agreed they received results related to ancestry or
genealogy. (It is possible some respondents who used mul-
tiple tools did not limit responses to the selected tool as
directed in survey instructions.) Across all tools, respon-
dents largely agreed using the tool increased their
Table 1. Survey Respondent Characteristics
All Tool Non-health Health Only Test
Variable Overall Users Only Tools Crossover Tools Statistica p Valuea

Number of respondents 1,137 820b 263 458 98

Mean age (SD; range) 46.4 (15; 46.7 (15; 51.8 (14; 45.5 (15; 39.4(12; 27.2 3.87e12
18–>89) 18–84) 18–83) 18–84) 20–73)

Gender

Women (%) 67.4 67.1 69.8 68.7 53.3 8.99 0.0111

Race (%)

Asian 1.8 1.6 0.9 1.9 2.2 N/A 0.0650

Black or African American 1.7 1.6 0.9 2.1 1.1

Hawaiian or Pacific Islander 0.1 0.1 0.0 0.2 0.0

White 81.6 80.6 76.2 81.7 86.7

Other 3.7 4.0 6.8 2.8 2.2

Prefer no answer 2.2 2.4 4.7 1.4 1.1

Multiplec 8.9 9.8 10.6 10.0 6.7

Hispanic/Latino (%) 6.7 6.6 8.1 6.0 5.6 N/A 0.677

Lives in US (%) 75.9 74.9 74.5 74.2 78.9 1.45 0.485

Max education (%)

Less than high school 1.0 1.1 2.1 0.5 1.1 12.3 0.139

High school graduate or GED 26.1 26.8 27.0 28.1 20.0

College degree 41.0 41.3 39.9 42.2 41.1

Master’s degree 23.1 22.8 21.5 23.2 24.4

Doctorate/terminal degree 8.9 8.1 9.4 6.0 13.3

Occupation (%)

Business, financial, management, sales 14.2 14.2 15.0 15.1 7.8 31.0 0.0556

Computer, engineering, math 16.9 17.1 16.7 15.5 25.6

Life, physical, and social science 9.2 8.2 6.8 7.2 15.6

Legal 2.5 2.6 0.9 3.5 3.3

Education, training, library 14.3 13.6 13.2 15.3 6.7

Arts, design, entertainment, sports, media 4.7 4.6 3.4 5.1 5.6

Healthcare practitioner 8.9 8.9 10.3 7.7 11.1

Office, administrative support 7.7 7.5 6.4 8.8 4.4

Construction, maintenance, natural 1.9 1.9 2.6 1.6 1.1

resources

Production and transportation 1.5 1.7 2.1 1.6 1.1

Other 18.2 19.7 22.6 18.6 17.8

Works in genetic research/medicine (%) 5.1 3.8 3.0 2.3 13.3 24.8 4.11e06

Participant in genetic research (%) 14.9 16.2 11.5 19.0 15.6 7.68 0.0215

Survey respondent characteristics, overall and grouped by type(s) of tools used. For categorical variables, values are given as within-group percentage, excluding
NA/missing values from the denominator. Statistical tests of difference are reported for comparison between groups of tool users: users of non-health only tools,
crossover users (used both health and non-health tools), and users of health-only tools. SD indicates standard deviation.
a
Comparing three groups of tool users: non-health only tool users, crossover tool users, and health-only tool users. For categorical values where all cell counts >5,
the test statistic and p value are from a chi-square test. For categorical values with low cell counts (i.e., race and Hispanic/Latino), the p value is from a Fisher exact
test, and the test statistic is N/A. ANOVA was used to compare continuous variables; the test statistic given is the F value.
b
Of the 820 respondents who used at least one tool, one respondent reported using only ‘‘various R packages’’ and could therefore not be assigned a tool user
group (non-health only; crossover; health only).
c
Respondents who checked more than one box for self-identified race are counted under ‘‘Multiple.’’ Note all American Indian/Alaska Native respondents checked
more than one box and are therefore all counted under ‘‘Multiple’’ here.

The American Journal of Human Genetics 105, 122–131, July 3, 2019 125
Table 2. DTC Tests Ordered and DTC Testing Motivations
Variable Overall All Tool Users Non-health Only Tools Crossover Health Only Tools Test Statistica p Valuea

Number of respondents 1,137 820 263 458 98

DTC Tests Ordered (%)

23andMe 62.4 61.7 40.7 68.1 87.8 85.3 <2.2e16

AncestryDNA 53.9 59.8 73.0 62.0 14.3 104.5 <2.2e16

FamilyTreeDNA 26.8 30.2 37.3 32.1 3.1 41.2 1.14e09

b
Rating of DTC Testing Motivations (% Very Important)

General curiosity 68.0 67.6 55.5 71.4 82.7 30.7 2.1e07

Ancestry 66.0 66.3 69.6 69.8 41.2 31.1 1.8e07

Find relatives 46.9 49.9 62.4 50.4 14.3 66.1 4.4e15

Risk for specific diseases 31.4 30.9 16.1 34.2 54.1 53.9 1.97e12

Limited family health history 20.7 20.7 21.4 21.0 17.7 0.621 0.733

Other family members 10.7 10.7 10.8 11.0 8.3 0.656 0.720
are using

Participate in research 32.3 31.6 26.3 33.8 34.7 4.79 0.0911

Raw genetic data file 50.8 53.1 44.4 56.4 60.4 11.9 0.00263

Comparisons of DTC tests ordered and DTC testing motivations, overall and grouped by type(s) of tools used.
a
Comparing non-health only tool users, crossover tool users, and health-only tool users. Test statistics and p values are from chi-square tests.
b
Responses to DTC testing motivations were dichotomized into ‘‘very important’’ versus ‘‘somewhat important’’ or ‘‘not at all important.’’

understanding of genetics in general (593/778 or 76%
agreed) and of how DTC companies interpret genetic
data (522/775 or 67% agreed). The majority also felt satis-
fied with the information received (678/770 or 88%),
though some reported feeling confused (266/767 or 35%)
or upset (48/758 or 6%). These reactions were relatively
consistent across tools (see Figure S3).
The most common follow-up actions were sharing the
results with a family member (664/780 or 85% shared) or
non-family member friend or loved one (552/778 or 71%
shared). Other common follow-up actions included pursu-
ing additional analysis via a different tool (430/776 or
55%), pursuing more genetic testing (253/776 or 33%),
or participating in a genetic research study (275/775 or
35%). Few respondents made changes to either health in-
surance (7/770 or 0.9% changed) or other types of insur-
ance (i.e., life or long-term care; 8/771 or 1% changed).
Fifteen percent of respondents (116/775) indicated sharing
results with a health care provider (HCP), most commonly
a general practitioner (92/116, or 79% of those who
shared). As might be expected based on the differing types
of information offered by each tool, the reported rate of
HCP sharing differed significantly by tool: 8.1% (39/484)
of GEDmatch users shared while 31% (56/183) of Prome-
thease users shared (Chi-square test statistic ¼ 53.4, p ¼
2.69e13). Notably, across all respondents who shared
with a HCP, 13 (11% of 116) reported sharing results
with a medical geneticist or GC. A total of 52 respondents
(45% of 116) wrote in an additional type of HCP with
whom they shared results, including practitioners from
non-genetics specialties (e.g., cardiologist, gastroenterolo-
gist, ophthalmologist), psychiatry/psychology, and alter-
native medicine (e.g., naturopath, acupuncturist).
Use of Specific Third-Party Tools
Next, we sought further understanding of what factors
influenced an individual to use a given tool or combina-
tion of tools, in particular when respondents used multiple
tools across the domains of health, ancestry, and geneal-
ogy. We performed a series of logistic regression analyses
of tool use, separately for the 7 tools with at least 40 users:
GEDmatch, Promethease, DNA.Land, openSNP, Genetic-
Genie, Interpretome, and Livewello (see Table S2). Desire
to learn about ancestry was significantly and positively
associated with GEDmatch (b ¼ 0.957, p ¼ 3.76e05)
and DNA.Land (b ¼ 0.447, p ¼ 0.00775). Desire to learn
about personal risk for specific diseases was positively asso-
ciated with use of Promethease (b ¼ 0.604, p ¼ 0.00146),
Genetic Genie (b ¼ 1.23, p ¼ 0.000286), and Livewello
(b ¼ 1.30, p ¼ 0.000766). Desire to obtain raw genetic
data was positively associated with use of DNA.Land (b ¼
0.517, p ¼ 0.00176) and Interpretome (b ¼ 0.854, p ¼
0.0166). Survey recruitment venue was strongly associated
with tool use, as GEDmatch and DNA.Land users were
more likely to have been recruited via Facebook while
Promethease users were more likely to have been recruited
from Reddit or the openSNP newsletter. DTC test(s) or-
dered were also linked to tool use (see also Figure 2), with
the most significant associations between 23andMe and
Promethease (b ¼ 0.791, p ¼ 3.e06) and FamilyTreeDNA
and GEDmatch (b ¼ 1.53, p ¼ 0.000199).

126 The American Journal of Human Genetics 105, 122–131, July 3, 2019

Categories of Third-Party Tool Use
The large proportion of respondents using multiple third-
party tools, including tools spanning the disparate cate-
gories of health, ancestry, and genealogy, led us to next
examine combinations of tools used. We grouped respon-
dents into those using only health-related tools (n ¼ 98
respondents); only non-health tools (n ¼ 263); and those
using both types (n ¼ 458), which we refer to as the ‘‘cross-
over group’’ (see Table S1 for tool characterization). We pre-
sent differences in DTC testing motivations, DTC test(s) or-
dered, and demographics between the three groups
(health-only tool users, non-health only, and crossover
group) in Tables 1 and 2. There was a linear trend in age,
where non-health only tool users were oldest (mean
age ¼ 51.8), health only tool users were youngest (mean
age ¼ 39.4), and the crossover group was in between
(mean age ¼ 45.5). The health-only tool group had a signif-
icantly lower proportion of women (53% versus >68% in
the other two groups) and higher proportion of respon-
dents working in genetic research or medicine (13%
versus %3% in the other two groups). The crossover group
had a slightly higher proportion of respondents partici-
pating in (non-DTC) genetic research (19%) compared to
the health only (15.6%) or non-health only (11.5%) user
groups (p ¼ 0.0215). DTC test(s) ordered were significantly
different between the three groups, with the proportion of
23andMe customers increasing from the non-health group
(40.7%) to crossover group (68.1%) to health-only group
(87.8%) and a reverse pattern for both AncestryDNA and
FamilyTreeDNA (proportion decreasing from non-health
to crossover to health-only; see Table 2). Notable associa-
tions between DTC testing motivations and user group
included increasing importance of general curiosity about
genetics from non-health to crossover to health-only.
Interest in having raw data was highest among the
The American Journal of Human Genetics 105, 122–131, July 3, 2019 127
Figure 1. Perceived Information Re-
ceived from Tools
Agreement with statements about infor-
mation received from TPI tools, separately
by tool, over 820 respondents who re-
ported using at least one tool. Counts
plotted to the right of x ¼ 0 are for agree-
ment; counts plotted to the left of x ¼
0 are for disagreement.
health-only tool group and lowest in
the non-health only, suggesting that
those who exclusively used health-
related TPI tools were more highly
motivated by raw data access when
initially pursuing DTC testing.
Interviews with Crossover Tool Users
In follow-up interviews with respon-
dents in the crossover group, we
observed tool crossover in both direc-
tions: three initially used health-
related tools before trying non-health
tools, and four initially used non-health tools before trying
health-related tools, illustrated by quotes in Box 1. In
contrast to those who began with an interest in one
domain then moved to another, for three interviewees
who were either adopted or had an adopted parent, inter-
est in health and genealogy were inextricably bound
together in their search to learn simultaneously about their
biological family and about potential implications for their
own health risks.
The primary phenomena responsible for tool crossover
were social networking, general curiosity, and initial lack
of interesting findings. Facebook and Reddit were the
primary social networks in which participants learned
about multiple tools, including those seemingly outside
the scope of the given Facebook group or sub-Reddit.
Once participants learned about additional tools, they
tried them often out of curiosity or general hunger to learn
more, in particular to go beyond the information provided
in DTC company reports. An initial lack of interesting find-
ings in one domain also pushed some participants to seek
out tools in another area.
Discussion
In this sample of DTC customers, we found high rates of
raw data download and usage of TPI tools. Given some of
our recruitment venues, this volume of TPI tool use may
be expected; however, the scale and scope were notable.
Specifically, respondents reported using on average three
different TPI tools, often spanning a range of health,
ancestry, and genealogy. Users of different tool categories
(health, non-health, and both) differed in their demo-
graphics, motivations for DTC testing, and DTC tests
taken. In follow-up interviews with a subset of crossover
users (i.e., those using both health and non-health tools),

we observed individuals often migrated to using tools
outside their original scope of interest due to peer-to-peer
sharing on social networks, general curiosity, and/or initial
lack of interesting results. Below we discuss implications of
these findings for the professional genetics community,
including both researchers and providers.
Our findings suggest that in the research context, return-
ing raw data to participants could present both a benefit
and a liability to researchers. Raw data return may benefit
the research community by increasing rates of participa-
tion in genetic research and engagement with and enthu-
siasm for genetics more broadly.12 In our survey, 35% of
tool users reported that using third-party tools led them
to participate in a genetic research study, while 76% agreed
that it increased their understanding of genetics in general.
The potential liability for the researcher community is that
third-party tools vary widely in the quality, scope, and
complexity of information returned,15,35 in addition to
having variable data security and privacy practices.36,37 Ad-
dressing potential legal liability is outside the scope of this
paper, but we suggest that researchers may bear some
ethical responsibility if participants are harmed by
downstream third-party tool usage in the way some DTC
customers have been harmed.16–19 Notably, in our results,
only 6% of respondents reported feeling upset by informa-
tion received from third-party tools while 35% felt
confused. While not reported by a majority of our respon-
dents, these negative outcomes merit further study and
consideration on how to avoid and/or mitigate.
As with DTC testing, one concern with TPI tool usage is
downstream overutilization of scarce health care resources,
or ‘‘raiding of the medical commons.’’38 Indeed, of the few
accounts of TPI tool usage to date, several have focused on
interactions with the health care system.20–22 These prior
studies rightly illustrate the potentially alarming outcomes
128 The American Journal of Human Genetics 105, 122–131, July 3, 2019
Figure 2. Tools Used Based on DTC
Test(s) Taken
Results are plotted for 820 respondents
who reported using at least one TPI tool.
Darker blue shading on bars indicates tools
that offer health-related information (see
Table S1).
of patients’ misunderstandings of TPI
reports; however, they do not indicate
how often these scenarios result from
‘‘garden variety’’ TPI tool usage. In the
current study, we observed a relatively
low reported rate of sharing TPI re-
sults with HCPs (15%). This rate was
lower than the 30% (of 321 surveyed)
previously reported by Wang et al.,25
which may be due to our participants
responding based on a specific tool
rather than across all tools used.
Indeed, when limiting to respondents
who answered based on Promethease, our rate of HCP
sharing was comparable (31%). Likely of concern to ge-
netics professionals is that among respondents who shared
with HCPs, the majority (79% of 116) did so with general
practitioners or non-genetics specialists, rather than with
medical geneticists or GCs. This will present challenges
for primary care physicians, especially given that TPI re-
ports are typically longer and harder to digest compared
to DTC company reports.15,20,35
Regardless of whether they brought TPI reports to pro-
viders, respondents were frequently engaging with health
information via TPI (i.e., 68% of tool users used health-
related tools; see also Table S1). This was true even for re-
spondents who were initially intent on finding relatives
and receiving genetic ancestry percentages. Due to the
flow of information on social media platforms such as
Facebook and Reddit, respondents who started off using
tools in one domain often switched or ‘‘crossed over’’ to us-
ing tools in another. This has implications for those whose
TPI reports eventually do prompt them to interact with the
health care system. For example, those initially interested
in genealogy who later use health tools may be more likely
to overestimate the reliability and comprehensiveness of
the health information. There is relatively little uncer-
tainty in identifying close relatives from genotyping array
information; as one interviewee said about genetic geneal-
ogy, ‘‘DNA doesn’t lie.’’ This is in stark contrast to disease
prediction based on similar data, which is far more
probabilistic and uncertain. Furthermore, the raw data
file is incomplete given that it is often based on array gen-
otyping rather than genome sequencing, and may even be
incorrect.22 The opposite effect is also possible: those
involved in both paper trail and genetic genealogy may
realize that genetics alone does not provide complete in-
formation when researching family history and so may
 Box 1. Interview Quotes (Participant ID in Parentheses)

Initial Use of Health-Related Tools:

I was interested in finding out exactly what my DNA meant, not just what 23andMe wanted to tell me, but maybe other information I could
glean from it.I think I used Promethease first.from what I could tell, it was the most information...[On additionally using GEDmatch] I may
have just Googled ‘‘things to do with raw DNA.’’ I just wanted something different from Promethease. Specifically, I wanted to see if I could find
matches other than what 23andMe had found for me. (574)
I got quite frustrated how boring I am in terms of my ancestry [from DTC reports], there’s hardly any variation...I think that’s why I decided to
concentrate more on the health side of things rather than the rather dull ancestry part, because I was hoping to find something quite exciting
and something I didn’t know about before, and the health side provided that — but the ancestry didn’t.[On additionally using GEDmatch and
DNA.Land] On a lot of these forum posts, they tend to list the third party websites along with their cost. GEDmatch and DNA.Land come up
frequently in these lists on the posts. (58)
Initial Use of Non-health Related Tools:
I was primarily focusing on genealogy at that point [when first using TPI tools].I may have found Promethease mentioned online or in an
article I read.I just said, ‘‘Oh, that’s cool.’’ So I did that.It’s just general interest, I just wanted to see if there was anything to the health
thing, and I don’t carry anything specific.I didn’t think I did, but it’s good to know. But the genealogy thing is what I’ve been focusing on
mostly, because the health is all like, ‘‘Ok, I know. I could get heart disease.’’ I don’t have any ‘‘bad genes,’’ as they say — the high correlation
ones. I don’t really have any of those. (798)
I wasn’t doing it [using TPI tools] for health at all and just when I saw the Promethease and I thought, ‘‘Well for five dollars, you know, we’ll see
what it says.’’ (559)
When I did download [raw data] for the first time, it was specifically to upload to GEDmatch.After that, I explored what else I could do and
found out that I could get more matches by doing free uploads to other sites. I uploaded to DNA.Land, Geni, MyHeritage and FamilyTreeDNA.
Through membership of a genealogy Facebook group, I found out about Promethease and did that out of curiosity. (664)
Inextricably Linked (e.g., for Adoptees):
I’m adopted, and I know a limited amount about my biological mother. I wanted to see if I could fish around and find out what was going on the
other side, find anybody that I was connected to. I was doing all kinds of testing.partly I was trying to take control of whatever health
information I could get, but then also fishing around for family connections. I was doing some of each. (577)
Partially it was ancestry, partially it was finding my bio family, partially it was finding out what [family history] boxes can I check now when
I go to the doctor’s office. It’s a little bit of everything which all comes together for a big ball of happy.It’s not any one single reason. (226)

be better equipped to understand limitations of health-
related genetic information. Indeed, some individuals
who primarily pursue ancestry testing over health-related
testing may do so because they perceive the latter to lack
accuracy and utility. Providers who better understand the
course patients have taken to leverage their raw data via
TPI may be better equipped to calibrate and manage pa-
tient expectations and understanding.
Another important finding is that approximately 40% of
tool users agreed they had received health information,
including about disease risk. This contrasts with our prior
study of tool developers in which they characterized tools’
direct linking to scientific publications or variant annota-
tion databases as merely ‘‘bridging to the literature’’ and
hence stopping short of actual interpretation.15 However,
our survey data suggest users regard TPI reports as
providing personally relevant health information. At the
same time, developers’ claims that ‘‘bridging’’ may increase
understanding of genetic risk15,35 were supported by our
survey results: the majority of respondents agreed that us-
ing TPI tools increased their understanding of genetics in
general (76%) and in particular how DTC companies inter-
pret genetic data (67%).
To our knowledge, this is the largest study to date of raw
data and TPI tool usage. However, our recruitment of a con-
venience sample via social media limits the ability to
generalize findings to DTC customers more broadly, as it
is difficult to know how our survey respondents compare
to the larger population of DTC customers. It is reasonable
to infer, however, that our respondents are likely highly
motivated individuals who were active in these online
forums and thus may overestimate the degree of data
download and tool usage. The relatively small number of
non-downloaders may not have provided enough power
to detect differences with downloaders. However, we
contend that limited generalizability is mitigated in part
by our collection of qualitative data through open text sur-
vey responses and follow-up interviews, which generated
deeper understanding of users’ motivations and experi-
ences. Convenience sampling also allowed a rapid collec-
tion of a large number of respondents already engaged in
the topics of interest, which seems appropriate for gaining
preliminary insight into a relatively understudied area.
Furthermore, by recruiting from social media venues
where individuals were likely to be engaging with raw
data and third-party interpretation, we were able to
observe more of the phenomena of interest. The length
of the survey may have contributed to the 24% non-
completion rate and therefore potential survey item
response bias, though compared to prior surveys,25 we
collected more extensive and granular information about
specific tool usage. Some incongruities between informa-
tion reportedly received and the actual offerings of tools
suggests that respondents may not have limited responses

The American Journal of Human Genetics 105, 122–131, July 3, 2019 129
to the selected tool as directed; however, these incongru-
ities were not widespread.
We have focused on the consumer genomics context as
it is currently the most common way to access raw data.
However, this work can help explore how broadening
routes of access may unfold. We have discussed return of
raw data from genetic research above, but another poten-
tial route is through clinical sequencing. Since 2014, the
HIPAA direct access right has allowed individuals to access
the contents of their designated record sets,39 which for
clinical sequencing laboratories would likely include unin-
terpreted sequence data.40 Laboratories are not required to
provide additional explanation or interpretation, which
may lead recipients to seek out TPI. Indeed, many TPI tools
accept the Variant Call Format (VCF) file type common to
genome and exome sequencing.15 Future research should
evaluate how individuals’ interactions with their raw
data potentially differ across the contexts of DTC testing,
clinical sequencing, and return of results from research.
In summary, moving forward individuals will have
increasing routes to access their raw genetic data and
leverage it in an expanding menu of largely unregulated
TPI services. These activities raise a set of concerns related
to but distinct from DTC genetic testing and thus merit
further investigation to more fully understand potential
harms and benefits. Rather than taking sides in a potential
ensuing ‘‘culture war’’ about raw data,41 the professional
genetics community has an opportunity to proactively
engage with users, understand the complexity of their mo-
tivations for pursing third-party analysis, and ultimately
educate them about potential limitations.
Supplemental Data
Supplemental Data can be found online at https://doi.org/10.
1016/j.ajhg.2019.05.014.
Acknowledgments
We thank our participants as well as the administrators of the Face-
book groups and openSNP newsletter who aided in recruitment.
Interview transcription was supported by funds from the UW
Institute for Public Health Genetics. This work was partially sup-
ported by the National Human Genome Research Institute
(NHGRI) and the National Cancer Institute (NCI) CSER Con-
sortium, U01 HG006507 and U24 HG007307 (Jarvik, PI). This
research used statistical consulting resources provided by the Cen-
ter for Statistics and the Social Sciences at UW. REDCap and the
Participant Portal at ITHS are supported by the National Center
for Advancing Translational Sciences of the National Institutes
of Health under Award Number UL1 TR002319.
Declaration of Interests
The authors declare no competing interests.
Received: March 12, 2019
Accepted: May 20, 2019
Published: June 13, 2019
130 The American Journal of Human Genetics 105, 122–131, July 3, 2019
Web Resources
Service used to transcribe audio recordings from participant inter-
views, https://www.rev.com
Survey instrument, quantitative survey dataset, and R analysis code,
http://doi.org/10.3886/E105721V3
References
1. Thorogood, A., Bobe, J., Prainsack, B., Middleton, A., Scott, E.,
Nelson, S., Corpas, M., Bonhomme, N., Rodriguez, L.L.,
Murtagh, M., Kleiderman, E.; and Participant Values Task
Team of the Global Alliance for Genomics and Health
(2018). APPLaUD: access for patients and participants to
individual level uninterpreted genomic data. Hum. Genomics
12, 7.
2. Nelson, S. (2016). Geneticists should offer data to participants.
Nature 539, 7–7.
3. Lunshof, J.E., Church, G.M., and Prainsack, B. (2014). Infor-
mation access. Raw personal data: providing access. Science
343, 373–374.
4. Beskow, L.M., and Burke, W. (2010). Offering individual ge-
netic research results: context matters. Sci. Transl. Med. 2,
38cm20.
5. Bredenoord, A.L., Kroes, H.Y., Cuppen, E., Parker, M., and van
Delden, J.J.M. (2011). Disclosure of individual genetic data to
research participants: the debate reconsidered. Trends Genet.
27, 41–47.
6. Wright, C.F., Middleton, A., Barrett, J.C., Firth, H.V.,
FitzPatrick, D.R., Hurles, M.E., and Parker, M. (2017). Return-
ing genome sequences to research participants: Policy and
practice. Wellcome Open Res 2, 15.
7. Middleton, A., Wright, C.F., Morley, K.I., Bragin, E., Firth, H.V.,
Hurles, M.E., Parker, M.; and DDD study (2015). Potential
research participants support the return of raw sequence
data. J. Med. Genet. 52, 571–574.
8. Karow, J. (2018). All of Us Program Plans to Return Disease Var-
iants, PGx Results, Primary Genomic Data (GenomeWeb).
9. National Institutes of Health (2018). Informational Webinar
on the ‘‘All of Us Genetic’’ Counseling Resource Funding
Announcement, https://allofus.nih.gov/sites/default/files/
genetic_counseling_resource_webinar.pdf.
10. NASEM (2018). Returning Individual Research Results to Par-
ticipants: Guidance for a New Research Paradigm (Washing-
ton, D.C.: National Academies Press).
11. Haeusermann, T., Fadda, M., Blasimme, A., Tzovaras, B.G., and
Vayena, E. (2018). Genes wide open: Data sharing and the
social gradient of genomic privacy. AJOB Empir. Bioeth. 9,
207–221.
12. Haeusermann, T., Greshake, B., Blasimme, A., Irdam, D., Ri-
chards, M., and Vayena, E. (2017). Open sharing of genomic
data: Who does it and why? PLoS ONE 12, e0177158.
13. Ball, M.P., Bobe, J.R., Chou, M.F., Clegg, T., Estep, P.W., Lun-
shof, J.E., Vandewege, W., Zaranek, A., and Church, G.M.
(2014). Harvard Personal Genome Project: lessons from partic-
ipatory public research. Genome Med. 6, 10.
14. Swan, M. (2012). Crowdsourced health research studies: an
important emerging complement to clinical trials in the pub-
lic health research ecosystem. J. Med. Internet Res. 14, e46.
15. Nelson, S.C., and Fullerton, S.M. (2018). ‘‘Bridge to the Litera-
ture’’? Third-Party Genetic Interpretation Tools and the Views
of Tool Developers. J. Genet. Couns. 27, 770–781.
16. Kolata, G. (2018). The Online Gene Test Finds a Dangerous
Mutation. It May Well Be Wrong. New York Times, July 3,
2018. D1. https://www.nytimes.com/2018/07/02/health/
gene-testing-disease-nyt.html.
17. Almendrala, A. (2018). Home Genetic Tests May Be Riddled
With Errors, And Companies Aren’t Keeping Track. Huffingt.
Post, April 3, 2018. https://www.huffpost.com/entry/home-
genetic-test-false-positives_n_5ac27188e4b04646b6451c42.
18. Hercher, L. (2018). 23andMe Said He Would Lose His Mind.
Ancestry Said the Opposite. Which Was Right? New York
Times, September 16, 2018. SR7. https://www.nytimes.com/
2018/09/15/opinion/sunday/23andme-ancestry-alzheimers-
genetic-testing.html.
19. Matloff, E. (2019). I Had Lynch Syndrome For 30 Hours.
Forbes, February 12, 2019. https://www.forbes.com/sites/
ellenmatloff/2019/02/12/i-had-lynch-syndrome-for-30-hours-
2/#66ed644a2567.
20. Allen, C.G., Gabriel, J., Flynn, M., Cunningham, T.N., and
Wang, C. (2018). The impact of raw DNA availability and cor-
responding online interpretation services: A mixed-methods
study. Transl. Behav. Med. 8, 105–112.
21. Moscarello, T., Murray, B., Reuter, C.M., and Demo, E. (2019).
Direct-to-consumer raw genetic data and third-party interpre-
tation services: more burden than bargain? Genet. Med. 21,
539–541.
22. Tandy-Connor, S., Guiltinan, J., Krempely, K., LaDuca, H.,
Reineke, P., Gutierrez, S., Gray, P., and Tippin Davis, B.
(2018). False-positive results released by direct-to-consumer
genetic tests highlight the importance of clinical confirma-
tion testing for appropriate patient care. Genet. Med. 20,
1515–1521.
23. Erlich, Y., Shor, T., Pe’er, I., and Carmi, S. (2018). Identity
inference of genomic data using long-range familial searches.
Science 362, 690–694.
24. Ram, N., Guerrini, C.J., and McGuire, A.L. (2018). Genealogy
databases and the future of criminal investigation. Science
360, 1078–1079.
25. Wang, C., Cahill, T.J., Parlato, A., Wertz, B., Zhong, Q., Cun-
ningham, T.N., and Cummings, J.J. (2018). Consumer use
and response to online third-party raw DNA interpretation
services. Mol. Genet. Genomic Med. 6, 35–43.
26. Yuan, J., Gordon, A., Speyer, D., Aufrichtig, R., Zielinski, D.,
Pickrell, J., and Erlich, Y. (2018). DNA.Land is a framework
to collect genomes and phenomes in the era of abundant ge-
netic information. Nat. Genet. 50, 160–165.
27. Greshake, B., Bayer, P.E., Rausch, H., and Reda, J. (2014).
openSNP–a crowdsourced web resource for personal geno-
mics. PLoS ONE 9, e89204.
The American Journal of Human Genetics 105, 122–131, July 3, 2019 131
28. Carere, D.A., Couper, M.P., Crawford, S.D., Kalia, S.S., Duggan,
J.R., Moreno, T.A., Mountain, J.L., Roberts, J.S., Green, R.C.;
and PGen Study Group (2014). Design, methods, and partici-
pant characteristics of the Impact of Personal Genomics
(PGen) Study, a prospective cohort study of direct-to-
consumer personal genomic testing customers. Genome
Med. 6, 96.
29. Krosnick, J.A., and Presser, S. (2010). Question and Question-
naire Design. In Handbook of Survey Research, P.V. Marsden
and J.D. Wright, eds. (Emerald), pp. 263–313.
30. Harris, P.A., Taylor, R., Thielke, R., Payne, J., Gonzalez, N., and
Conde, J.G. (2009). Research electronic data capture
(REDCap)–a metadata-driven methodology and workflow pro-
cess for providing translational research informatics support.
J. Biomed. Inform. 42, 377–381.
31. R Core Team (2013). R: A language and environment for statis-
tical computing (R Foundation for Statistical Computing).
32. Baptista, N.M., Christensen, K.D., Carere, D.A., Broadley, S.A.,
Roberts, J.S., and Green, R.C. (2016). Adopting genetics: moti-
vations and outcomes of personal genomic testing in adult
adoptees. Genet. Med. 18, 924–932.
33. Koeller, D.R., Uhlmann, W.R., Carere, D.A., Green, R.C., Rob-
erts, J.S.; and PGen Study Group (2017). Utilization of Genetic
Counseling after Direct-to-Consumer Genetic Testing: Find-
ings from the Impact of Personal Genomics (PGen) Study.
J. Genet. Couns. 26, 1270–1279.
34. Kirkpatrick, B.E., and Rashkin, M.D. (2017). Ancestry Testing
and the Practice of Genetic Counseling. J. Genet. Couns. 26,
6–20.
35. Badalato, L., Kalokairinou, L., and Borry, P. (2017). Third party
interpretation of raw genetic data: an ethical exploration. Eur.
J. Hum. Genet. 25, 1189–1194.
36. Ney, P.M., Ceze, L., and Kohno, T. (2018). Computer Security
Risks of Distant Relative Matching in Consumer Genetic Data-
bases. ArXiv, 1810.02895.
37. Hazel, J.W., and Slobogin, C. (2018). Who Knows What, and
When?: A Survey of the Privacy Policies Proffered by U.S.
Direct-to-Consumer Genetic Testing Companies. Cornell J.
Law Public Policy 28, 35–66.
38. McGuire, A.L., and Burke, W. (2008). An unwelcome side ef-
fect of direct-to-consumer personal genome testing: raiding
the medical commons. JAMA 300, 2669–2671.
39. U.S. DHHS (2014). CLIA Program and HIPAA Privacy Rule;
Patients’ Access to Test Reports. 42 CFR x 493, 45 CFR x 164.
40. U.S. DHHS (2016). Individuals’ right under HIPAA to access
their health information 45 CFR x 164.524.
41. Evans, J.P., and Green, R.C. (2009). Direct to consumer genetic
testing: Avoiding a culture war. Genet. Med. 11, 568–569.