Psychometric properties of a culturally adapted Spanish version of the Attitudes Toward Genomics and Precision Medicine instrument

Abstract

Introduction:

There is evidence of growing racial and ethnic disparities in genomic healthcare and precision medicine. Validated survey instruments and measures are required to understand the needs of diverse populations to appropriately tailor person-centered approaches and end disparities in genomic healthcare and precision medicine.

Methods:

We aimed to examine the psychometric properties of a culturally adapted Spanish version of the Attitudes Toward Genomics and Precision Medicine (AGPM). First, we culturally adapted the AGPM. We then conducted a web-based evaluation of the Spanish AGPM in a cohort of 486 individuals identifying as Hispanic to establish the Spanish version’s reliability, factor structure, and measurement invariance relative to the English version. We also compared AGPM responses between Spanish- and English-speaking Hispanic individuals.

Results:

The Spanish version of the AGPM demonstrates robust internal consistency with Cronbach alpha ranging from 0.84–0.98 across domains. All AGPM items significantly loaded on their respective factor (p < 0.001). Configural, metric, strict, and residual invariance models all met absolute and relative fit criteria. Significant differences were observed between Spanish and English-speaking participants in some AGPM subscales.

Conclusions:

The Spanish version of the AGPM demonstrates sound psychometric properties and may be useful for informing culturally empowered approaches to genomic healthcare and precision medicine for people identifying as Hispanic.

Introduction

Genomics has important population health implications in relation to prenatal, newborn, and adult screening as well as genetic testing [1]. However, a 2018 report from the National Academies of Sciences, Engineering, and Medicine identified mounting racial/ethnic disparities in genomic healthcare/precision medicine [2]. In light of mounting disparities in genomic healthcare, some have advocated for prioritizing health equity in genomics [1] - particularly among the Hispanic population which constitutes 19% of the population in the United States [3, 1]. The Clinical Sequencing Evidence-Generating Research consortium (CSER) note structural factors (e.g., geography, insurance coverage) are key drivers affecting access - yet investigators underscore the critical role of psychosocial factors (e.g., attitudes, beliefs, literacy/numeracy) as critical barriers to uptake [4]. Indeed, prior research on Hispanic individuals has identified a range of barriers to genomic healthcare/precision medicine, underscoring the importance of considering cultural norms, and the need for culturally and linguistically tailored approaches [5–8].

An important starting point to foment health equity in genomic healthcare/precision medicine, is to have validated instruments that are linguistically and culturally relevant [9]. These validated instruments are essential to understand the attitudes, beliefs, norms, and preferences of Hispanic individuals, which will play an important role in the development of culturally empowered, person-centered interventions to promote their access to genomic healthcare/precision medicine.

Understanding if genomic healthcare/precision medicine services are perceived as beneficial and consistent with one’s health beliefs and values is essential to support high-quality decisions (i.e., informed and aligned with preferences) [10]. Given that 68% of Hispanics in the U.S. speak Spanish [11], having validated, culturally adapted instruments in Spanish represent an essential first step of a coherent public health strategy that could also include community engagement, coalition building, enhancing genomic literacy/numeracy, and diversifying the genomic healthcare workforce [1].

A number of studies have examined beliefs and attitudes towards genomic healthcare/precision medicine - yet relatively few efforts have examined survey structure and psychometric properties, particularly among Hispanic populations. Further, the focus and constructs of validated instruments vary widely. For example, the Genomic Orientation Scale examines optimism/pessimism about the future of genomic medicine [12]. In contrast, the Genetic Utility Scale evaluates perceived informational and emotional utility of genetic screening in adult and pediatric populations respectively [13, 14]. The Attitudes Toward Genomics and Precision Medicine (AGPM) measures perceived benefits, concerns, and overall support of genomics and precision medicine [15]. The AGPM is unique as it is written at an 8th grade level and embeds educational information so that the attitudes assessed are informed by basic information about genomics and precision medicine. The instrument measures concerns spanning a range of constructs (i.e., privacy, embryo, nature, social justice) and was designed to assesses the latent factors explaining overall attitudes toward genomics and precision medicine. As such, the AGPM is applicable to a wide range of populations and can inform interventions to surmount barriers to genomic healthcare and precision medicine. We posit that a validated, culturally adapted Spanish version of the AGPM for Hispanic populations could help inform culturally empowered approaches to improve access to genomic healthcare and precision medicine among Hispanic adults.

Having validated Spanish language instruments is important because more than 30% of the Hispanic population in the United States was born in a Latin American country–where Spanish is the predominant language [6, 16]. Previous research has shown heterogeneity of the Hispanic population, including different levels of acculturation [17], language (e.g., Spanish and English) [18, 19], as well as values and beliefs [20] that can all impact instrument psychometric properties [21, 22]. In light of the diversity of the Hispanic population, establishing equivalence between English and Spanish language versions enable meaningful comparisons between Hispanic English and Spanish speakers. Understanding similarities and differences within populations can inform development of tailored interventions to engage communities and increase access and uptake of genomic healthcare and precision medicine. Indeed, validating culturally adapted instruments is a key component of a public health strategy for addressing disparities in genomic healthcare and precision medicine [1].

We undertook a rigorous process following best practices to culturally adapt a Spanish language version of the AGPM. Briefly, best practices involve ten sequential steps to produce high quality translation (i. preparation, ii. forward translation, iii. reconciliation, iv. back translation, v. back translation review, vi. harmonization, vii. cognitive debriefing, viii. review of cognitive debriefing and finalization, ix. proof-reading, x. producing a final report). The multi-step process helps ensure fidelity of constructs between the translated and original instruments. Subsequently, examining the psychometric properties of the culturally adapted instrument can support validity of the translated version. We aimed to develop a culturally adapted AGPM and examine the psychometric properties and invariance of the Spanish language version. Since the primary purpose of this examination is to facilitate comparisons across different groups of Hispanic adults, we also compare the scores of the English- and Spanish-speaking groups.

Methods

Permission to translate the Attitudes towards Genomics and Precision Medicine (AGPM) scale was obtained from the Bioethics Research Center (Washington University, St. Louis, MO). The study was reviewed and approved by the local Institutional Review Board (#24.210.01e) and was registered on ClinicalTrials.gov (NCT06386861).

Attitudes towards Genomics and Precision Medicine (AGPM)

The original AGPM [15] consists of 37 items and five subscales, each targeting a certain issue related to precision medicine: social justice concerns (n=4) e.g., “Genetic tests could cause people to be treated unfairly”; privacy concerns (n=8) e.g., “I have concerns about how my information will be kept private”; nature concerns (n=7) e.g., “I think gene editing is wrong because it is like playing God”; embryo concerns (n=9) e.g., “It bothers me that embryonic stem cell research destroys embryos”; and perceived benefits (n=9) e.g., “I am curious to know about my own genes”. Later, the developers of the AGPM added a six-item subscale representing overall support of genetic medicine (e.g., “I generally support the use of genetic testing”) [15]. Per the user manual, the scale is not included in the final score of the AGPM. Items are organized in blocks focusing on different topics (e.g., lifestyle monitoring, information storage). Since many people are unfamiliar with terms like precision medicine or gene editing, participants are presented with a brief explanation about the topic before each block. All items are rated using a 7-point Likert-type scale (1 = strongly disagree to 7 = strongly agree) and coded such that higher scores represent higher levels on the construct (i.e., more perceived benefits or more concerns).

Translation and cultural adaptation of the AGPM

We employed best practices of the Professional Society for Health Economics and Outcomes Research (ISPOR) and employed the ten sequential steps to translate and culturally adapt the AGPM [23].

Independent forward translation and reconciliation (steps 1–3)

First (step 1), we obtained permission from the AGPM developers to respect copyright and reviewed instrument concepts. Next (step 2), two native Spanish speakers independently performed forward translations from English to the target language (Spanish). Reconciliation (step 3) involved a native Spanish speaker (M.P.L.) comparing the forward translations to create a single forward translation.

Back translation, independent review, and harmonization (steps 4–6)

An experienced translator performed back translation (step 4) from Spanish to the source language (English). Back translation (step 5) involved investigators (A.A.D., M.P.L.) independently comparing the English back translation to the original source instrument (English AGPM). Harmonization (step 6) was performed in a meeting of investigators (A.A.D., M.P.L., I.R.M) to review any translation discrepancies to ensure conceptual equivalence.

Cognitive debriefing, refinement, proofreading, and finalization (steps 7–10)

A Spanish-speaking investigator (M.P.L.) conducted cognitive debriefing (step 7) using the Spanish AGPM with two native Spanish speakers using the “think aloud” method to assess comprehensibility, cultural relevance, and cognitive equivalence. A review of cognitive debriefing and finalization (step 8) was conducted (A.A.D., M.P.L., I.R.M) to incorporate findings from the cognitive debriefing process and make final refinements to improve the performance of the Spanish AGPM. Proofreading (step 9) was conducted by a Spanish-speaking study investigator (M.P.L.). The final step, the finalization report (step 10) is reported in this manuscript.

Psychometric validation of the Spanish AGPM

According to acceptable validity standards in survey research [24], instrument validation is a procedure where the instrument developer collects evidence supporting its use for certain purposes. Such evidence involves the steps described above. For example, ensuring items align with the target construct, testing associations with similar measures (performed by the original developers), and testing whether respondents understand and use the instrument as intended (e.g., “think aloud” exercise). Another component of validation involves collecting data to support evidence of the instrument’s internal structure including reliability, factor structure, and equivalence across groups of interest (in this case, English- and Spanish-speakers). These are the steps taken at this stage of the validation process.

Participants were recruited in April 2024 from Amazon Mechanical Turk (MTurk) [25, 26]. Briefly, MTurk is a large (250,000+ members), secure, web-based crowdsourcing platform that has been widely used for online social and behavioral sciences research [25, 27, 28]. Studies indicate that MTurk produces data/results that are comparable to traditional data collection methods [29, 30]. We targeted adults (18+ years old) who identified as either Hispanic and whose preferred language was either Spanish or English. All participants provided opt-in electronic consent prior to participation in the online survey. Participants selected their preferred language (Spanish or English) for completing the online survey (Qualtrics™). To mitigate fatigue effects, we started the survey with the AGPM. We implemented several validation items throughout the survey, aligning with best practices when working with MTurk [26]. Participants reported their age at the beginning and end of the survey. Only participants with matching responses were included in the analysis. In the English version, participants were asked to endorse a certain response option indicating careful reading of the question (e.g., “select option C”). The Spanish version included a similar validation question as well as a second question asking them to complete a well-known proverb (“Agua que no has de beber… déjala correr”). This validation question was included as a way to ensure participants completing the Spanish version were indeed Spanish speakers familiar with the language. Participants provided demographic information including age, sex, self-reported race (multiple entries possible), highest educational attainment, annual income, whether they live in the U.S., and self-reported health status ( 5-point Likert-type scale: 1 = poor to 5 = excellent).

Data cleaning

Inclusion criteria for analysis included: a) self-identifying as Hispanic; b) answering ≥ 75% of AGPM items (i.e., at least 28 items); and c) correct/matching validation questions. Two groups were created based on the language chosen to complete the survey (i.e., English or Spanish). Additional data cleaning measures included exclusion of participants whose responses clearly indicated the use of translation programs (e.g., open-ended responses started with the word “translate”), automatically generated responses (i.e., identical responses to open-ended response items), or duplicate entries (e.g., multiple [12 of 14] duplicate responses to socio-demographic questions). In cases of suspected duplication we retained the entry with the longest time to completion as a proxy for the participant’s attention.

Statistical analysis

Participant characteristics are reported using descriptive statistics. Internal consistency was determined by calculating Cronbach α. We performed the invariance analysis using the lavaan R package [31, 32]. We applied confirmatory factor analysis (CFA) to test if the AGPM is invariant across languages. Briefly, CFA assumes the data are multivariate normal [33]. Since our data did not meet this assumption (Mardia’s skewness = 43,994, p < .001, Mardia’s kurtosis = 206.37, p < .001), we used a robust maximum likelihood estimator which is less sensitive to violations of the multivariate normality assumption [34].

The invariance tests were performed following the steps proposed by Bowen and Masa [35]. First, we tested whether the hypothesized model fits well in each group (i.e., English and Spanish) separately. Based on the original structure of the AGPM, we tested the fit of a second-order factor model. In second-order factor models, each item is loaded on their respective scale (e.g., “Genetic tests could cause people to be treated unfairly” is loaded on the social justice concerns scale) and the scales all load on a single factor representing general attitudes (i.e., a combination of social justice concerns, privacy concerns, etc.). After establishing that this model fits well to the data in the English- and Spanish-speaking samples separately (see fit statistics below), we tested the fit of a series of nested models posing stricter constraints in terms of English- and Spanish-speaking group equivalence. These models’ fit was examined both independently and relative to the previous, less constrained model. If any model did not fit the data, we determined that the instrument did not meet this level of invariance.

Invariance testing

The first level of invariance is called configural invariance, and it refers to the similarity of the factor structure between the groups, that is, that in both the English- and Spanish-speaking survey groups the items load on the expected scales. When testing for configural invariance, we performed a CFA where the factor structure is set to be equal in both groups. The next level is metric invariance, where the factor loadings are set to be equal across the samples. This means that the degree to which each item “belongs” to their respective scale is similar across the English- and Spanish-speaking groups. This is followed by scalar or strict invariance which constrains the intercepts (as well as the loadings) to be equal. Strict invariance is the lowest level which allows for valid group comparisons. Since the intercepts are equal, strict invariance can be interpreted as same manifest scores representing the same level of the latent variable in both survey groups, English- and Spanish-speaking. Therefore, any score differences reflect true differences in the construct. The highest level of invariance is called residual invariance, where in addition to the item loadings and intercepts, the residuals are also set to be equal across the English- and Spanish-speaking groups. Residual invariance can be interpreted as having equal reliabilities across groups, which is not strictly necessary for comparing them (provided that the scales are sufficiently reliable in both language response groups).

The aforementioned models were evaluated using several fit statistics as proposed by [36, 37]: root mean square error of approximation and standardized root mean squared residual (RMSEA and SRMR, respectively; for both, ≤ .08 indicating an acceptable fit), and comparative fit index (CFI) and Tucker-Lewis index (TLI; for both, ≥ .90 indicating an acceptable fit). Some relative fit criteria we applied include a χ² test (to compare each model with its previous, less restrictive version), change in CFI ≥ −.01, and change in RMSEA ≤ .015. Generally, these fit statistics represent the differences between the data expected based on a model and the observed data, with various adjustments for sample size degrees of freedom [38].

Group comparisons

After establishing the AGPM’s measurement invariance, we looked at the factor loadings of the items for the full sample (i.e., English and Spanish together). We then employed Student’s t-tests to compare sub-scale scores between the English and Spanish samples. A Bonferroni correction was used to control for multiple comparisons (α = 0.05/6 scales = 0.008). We also used Cohen’s d as a measure of effect size [39], with d < 0.50 interpreted as a small effect, 0.50 ≤ d ≤ 0.80 interpreted as a medium effect, and d ≥ 0.80 interpreted as a large effect.

Results

In total, data was obtained from 851 participants. After data cleaning steps, 486 (57%) participants were included for analysis. All included participants self-identified as Hispanic. The majority of participants completed the survey in English (n = 295, 61%) rather than in Spanish (n = 191, 39%). Characteristics of the sample are reported in Table 1.

Table 1.

Demographic characteristics of the sample.

Variable1	Total (n = 486)	English (n = 295)	Spanish (n = 191)
Mean age (SD)	31.74 (5.58)	32.06 (6.22)	31.24 (4.40)
Live in the U.S. (%)	94.24	94.67	93.62
Sex (%)
Female	25.77	26.79	24.34
Self-identified race (%)
American Indian / Alaska Native	12.06	8.26	16.42
Asian	8.12	7.82	8.46
Black / African American	9.28	1.74	17.91
Native Hawaiian / Pacific Islander	9.74	10.00	9.45
White	60.79	72.17	47.76
Mixed race	4.19	2.71	3.29
Education (%)
Less than high school	2.23	0.38	4.81
High school	6.90	4.58	10.16
Some college	2.67	1.53	4.28
Associate’s degree	4.01	1.14	8.02
Bachelor’s degree	57.91	74.04	35.29
Master’s degree	25.39	17.56	36.36
Doctoral degree	0.89	0.76	1.07
Annual income (USD)
$0 – 23,000	1.10	1.13	1.06
$23,001 – 45,000	21.81	19.62	24.87
$45,001 – 75,000	42.07	43.40	40.21
$75,001 – 112,000	28.85	29.43	28.04
> $112,000	6.17	6.42	5.82
Health status (mean , SD)	3.78 (0.98)	3.88 (0.83)	3.63 (1.14)

¹Sample sizes represent the full dataset. Some variables had fewer responses due to missingness. Percentages are based on available data. Participants could endorse multiple racial categories.

Internal consistency

Comparing the amount of shared variance/covariance of the AGPM subscales (by language and overall) revealed good internal consistency (all Cronbach α ≥ .80; see Table 2).

Table 2.

Reliability (Cronbach’s α) of the Spanish AGPM domains

Subscale	Total	English	Spanish
Total scale	.98	.97	.98
Overall support	.92	.93	.89
Perceived benefits	.92	.91	.94
Privacy concerns	.91	.90	.93
Embryo concerns	.92	.91	.93
Nature concerns	.90	.89	.92
Social justice concerns	.84	.80	.89

Invariance testing

In both the English- and Spanish-speaking survey response groups, all of the items significantly loaded on their respective second-order factor, and the models’ fit statistics were acceptable (Table 3). Examining configural, metric, strict, and residual invariance models revealed that all four models met our absolute and relative fit criteria (Table 4). However, the residual model fit was significantly worse compared to the strict model based on the χ² test. These findings indicate that the English and Spanish AGPM versions are at least strictly invariant, suggesting that they are comparable. Standardized factor loadings for all items across subscales were very strong (all > 0.65) (Supplemental Material).

Table 3.

Fit statistics for the second-order factor model (English and Spanish samples).

Measure	Configural	Metric	Strict	Residual
Δχ2 (df)	-	15.34 (36)	37.01 (31)	143.67 (37)
p-value	-	0.999	0.211	< 0.001
ΔRMSEA	-	−0.002	0.00	0.002
ΔCFI	-	0.002	0.00	−0.007
RMSEA	0.054	0.052	0.052	0.054
CFI	0.937	0.939	0.939	0.932
TLI	0.932	0.937	0.938	0.933
SRMR	0.052	0.058	0.058	0.060

df, degrees of freedom; RMSEA, root mean square error of approximation; CFI, comparative fit index; TLI, Tucker-Lewis index; SRMR, standardized root mean squared residual.

Table 4.

Fit statistics for the measurement invariance models

Measure	English	Spanish
χ2 (df)	1983.35 (624)	1640.505 (624)
p-value	< 0.001	< 0.001
RMSEA	0.057	0.048
CFI	0.920	0.956
TLI	0.915	0.953
SRMR	0.058	0.047

df, degrees of freedom; RMSEA, root mean square error of approximation; CFI, comparative fit index; TLI, Tucker-Lewis index; SRMR, standardized root mean squared residual.

English- and Spanish-speaking group comparisons

Having established invariance between the two groups, we compared AGPM scores of English- and Spanish-speaking respondents identifying as Hispanic (Table 5). We observed significant differences (p < 0.001) between the groups in ‘perceived benefits’ (d = −0.41), ‘privacy concerns’ (d = −0.40), ‘embryo concerns’ (d = −1.93), and ‘nature concerns’ (d = −1.07). Compared to the English-speaking sample, the Spanish-speaking sample scored higher in all significant domains. The effects for ‘perceived benefits’ and ‘privacy concerns’ were small while the effects for ‘embryo concerns’ and ‘nature concerns’ were large.

Table 5.

Discriminant validity correlations English-speaking vs. Spanish-speaking

AGPM	English Mean (SD)	Spanish Mean (SD)	t(df)	p-value	Cohen’s d
Total	4.21 (0.35)	4.23 (0.34)	−0.64 (418.59)	0.522	−0.06
Overall support	5.32 (1.18)	5.25 (1.11)	0.66 (424.29)	0.511	0.06
Perceived benefits	4.84 (0.93)	5.22 (0.96)	−4.35 (396.95)	< 0.001	−0.41
Privacy concerns	4.70 (0.92)	5.07 (0.94)	−4.25 (400.14)	< 0.001	−0.40
Embryo concerns	2.95 (0.66)	4.12 (0.50)	−21.99 (471.48)	< 0.001	−1.93
Nature concerns	3.84 (0.79)	4.64 (0.68)	−11.94 (446.62)	< 0.001	−1.07
Social justice concerns	5.33 (1.14)	5.43 (1.27)	−0.85 (375.16)	0.398	−0.08

Discussion

Advances in genomics and precision medicine have enabled evidence-based approaches to reduce morbidity and mortality for millions of people – namely what the CDC terms ‘Tier 1’ conditions (i.e., hereditary breast and ovarian cancer, Lynch syndrome, and familial hypercholesterolemia) [40]. However, advances have not benefitted all, leading to ethnic/racial disparities in genomic healthcare and precision medicine [1]. A public health agenda is needed to prevent the further widening of health disparities in genomics and precision medicine [1]. Having a deeper understanding of human factors relating to genomic healthcare and precision medicine is essential for ending such disparities. Herein we report the psychometric properties of a culturally adapted Spanish version of the AGPM.

The culturally adapted Spanish AGPM exhibits robust internal consistency across all five subscales (α > 0.80). Standardized factor loadings were very strong (all > 0.65) across subscales and individual items. Invariance testing showed the English and Spanish AGPM versions were comparable. While all participants identified as Hispanic, AGPM scores differed between cohorts completing the Spanish and English versions. Specifically, the group completing the Spanish version had significantly higher scores for perceived benefits, privacy concerns, embryo concerns, and nature concerns. These findings are consistent with those from a recent study on the largest sample of Hispanic/Latinos in the United States: the Study of Latinos (SOL) [41]. The study included 5,769 Hispanic adults and identified significant differences within subgroups of the Hispanic population. The SOL study also identified low levels of awareness and utilization of genetic testing in the overall cohort. Only 6.5% of participants reported being offered a genetic test, and a mere 3% reported having undergone genetic testing [41]. Such findings are very concerning given the numerous advantages that genomic healthcare and precision medicine can offer [41].

In the present study, the group completing the Spanish AGPM had higher perceived benefits compare to the group completing the English AGPM. However, those completing the Spanish AGPM had greater privacy, nature and embryo concerns – with the embryo concerns exhibiting the largest effect size. It is possible that the observed differences in stem from a lack of information available in Spanish or in a language that resonates with them. Further, observations point to the need for targeted interventions to elicit and respond to concerns observed in the group completing the Spanish AGPM. Even when information is available, it is likely that it is not provided by race or ethnic concordant healthcare providers, with whom Hispanic clients may feel a greater level of trust [42, 43]. These findings are an important consideration for developing tailored approaches. Our results support that individuals identifying as Hispanic are not a monolithic group. A more nuanced view of an individual’s attitudes and beliefs are needed to inform individualized approaches to genomic healthcare and precision medicine. Moreover, contextualizing the individual may help deconstruct genetic essentialism and social constructs (i.e., race, ethnicity) [44].

The present study has a number of limitations. First, the sample may be biased. For example, we observed that the majority of participants identified their sex as male and were relatively highly educated. While we used MTurk to recruit a sizeable cohort and employed validation checks and a strict data cleaning protocol, our process yielded a relatively small sample size in the Spanish-speaking group, and it is still plausible that some responses may not be trustworthy. Last, the cross-sectional nature of the study does not afford insights into how stable or malleable AGPM scores are.

Genomics and precision medicine are rapidly evolving fields. As such, future work could examine how the AGPM performs with new and emerging discoveries in the field. Additional work examining the AGPM in a larger cohort of people identifying as female and Hispanic would support a better understanding of attitudes and perspective. Longitudinal data on the AGPM in response to tailored psychoeducational interventions (i.e., pre/post) would be useful in building an evidence base for more person-centered approaches to genomic healthcare/precision medicine that support high-quality decisions (i.e., informed and aligned with values and preferences).

To advance health equity in genomic healthcare and precision medicine, there is a need for psychometrically sound, culturally empowered instruments to understand the values, beliefs, and concerns of diverse people. The English and Spanish versions of the AGPM demonstrate robust internal consistency and factor loading among Hispanic adults. The Spanish version of the AGPM can inform the development of culturally empowered approaches to propel genomic healthcare/precision medicine for people identifying as Hispanic.

Data Availability Statement

The data that support the findings of this study are not publicly available due to copyright restrictions but are available from the Bioethics Research Center at Washington University in St. Louis, Missouri (https://bioethicsresearch.org/) upon reasonable request.