Design and context

Instrumental study ²⁷, carried out in four departments of Peru (Ancash, Ayacucho, Puno and Cuzco) according to the variety of Quechua. We selected adults from the departments of Puno and Cuzco, located in southwestern Peru in order to evaluate the Cuzco-Collao variety. Puno has 538,127 (57.0%) and Cuzco has 709,892 (74.7%) Quechua speakers ²⁸; both departments are characterized by commercial activities, livestock farming and tourism due to their richness in historical and archaeological sites. To evaluate the variety of Quechua Chanca, we selected inhabitants of the department of Ayacucho, which has 389,045 (81.2%) Quechua speakers, mostly dedicated to commerce, livestock farming and agriculture ²⁸. In order to evaluate the variety of Central Quechua, we selected inhabitants of the department of Ancash, which has 289,172 (34.0%) Quechua speakers, mostly engaged in commerce, agriculture and tourism ²⁸.

Cultural adaptation phase

Translation

We used the forward translation and back translation methods during the translation process ²⁷. The original version of the PHQ-9 (English) was translated directly into each variety of Quechua (Central, Cuzco-Collao, Chanca) by two independent translators, who are native Quechua speakers of each variety and with advanced knowledge of English. Once the translation for each variety was completed, the two translators and two native Quechua-speaking researchers met to discuss the differences in the translations. Once the discrepancies were resolved and the translations unified, we proceeded with the back translation (Quechua to English) of the three Quechua variants of the PHQ-9. This process was carried out by two translators who had English as their native language and advanced knowledge of Quechua for each variant. Once the back translations were completed, the translators for each variant met with two study researchers to verify the back translation along with the first translation, fine-tuning details and giving their approval to the final version for the three languages.

Cultural adaptation to Quechua contexts

By applying the Delphi method, we created a cultural adaptation form, which can be found at: https://doi.org/10.5281/zenodo.8312191. This form included some open questions for uncommon words in Quechua such as “depression” or “without hope”. We also sought to consult on the change of “reading newspaper” to “listening to radio”, this change involves the PHQ-9 response categories and the relationship between the PHQ-9 in Quechua with the DSM-V for the diagnosis of major depression. The Central Quechua version of the PHQ-9 was reviewed by two Quechua-speaking psychologists, who had at least three years of experience in the care of Quechua-speaking patients with depression in the provinces of Ancash; one of them had a master’s degree and the other a bachelor’s degree in psychology. The Cuzco-Collao Quechua version was evaluated by five Quechua-speaking psychologists, all with professional degrees and at least one year of experience in the care of Quechua-speaking patients with depression in Puno or Cuzco. The Chanca Quechua version was evaluated by two psychologists with professional degrees and at least one year of experience in the care of Quechua-speaking patients with depression in the provinces of Ayacucho. The interaction between each expert and the research team took place in two (Central) and four rounds (Cuzco-Collao and Chanca) of mailings. In addition, we assessed the improvements (based on the recommendations of the experts) using indicators of relevance, representativeness, clarity and cultural equivalence on a scale of 0 to 3 (where 3 was the best rating). Once all the experts rated the improvements in all indicators with a score of 3, we held a virtual meeting with the research team to evaluate the suggestions and reach a consensus.

Subsequently, a focus group (via Zoom) was organized for each variety of Quechua. The moderator for each variety was a Quechua-speaking psychologist proficient in qualitative methods. The meetings lasted approximately 60 minutes. First, we asked participants to answer the PHQ-9 in Quechua through an online version shared via Google Forms. Then, the moderator invited the participants to provide their opinion on the clarity and comprehension of the items in a language that was common and simple for Quechua speakers. Five Quechua speakers participated (three women and two men) in the focus group for the Cuzco-Collao variety; four people participated (two women and two men) for the Quechua Chanca variety; and four adults participated (three men and one woman) in the focus group for the Central Quechua variety. We considered, for the three varieties of Quechua, that the participants be bilingual (speak Quechua and Spanish) and over 18 years of age (Figure 1). The focus group participants provided a favorable opinion regarding the clarity and comprehension of the PHQ-9 in the three variants of Quechua, confirming that this adaptation is clear and in accordance with their cultural context.

Figure 1 Flow chart of the translation and cultural adaptation of the PHQ-9. 

Psychometric analysis phase

Participants

The final version of the instrument, obtained after the previous phase, was applied to a sample of 970 adults selected by non-probabilistic convenience sampling, ensuring that for each variety of Quechua we had no less than 200 participants. This number is even higher than the recommended standard for factor analyses (20 participants per item and considering that there are 9 items, the minimum size would be 180 participants) and reported as adequate in simulation studies ²⁹. It also complies with the minimum size calculated for factor analysis (n=218, see: https://doi.org/10.5281/zenodo.8312191). Six surveyors participated in the data collection, they presented the study to the participants, and once informed consent was obtained, they requested the participants to complete the survey via Google Forms. The collection was carried out in those towns and cities with the greatest Quechua presence. Men and women, over 18 years of age, living in urban and rural settings, located in three regions (according to Quechua variant) were included. We included inhabitants of the departments of Puno and Cusco (n=525) to evaluate the Cuzco-Collao variant, inhabitants of Ayacucho (n=226) to evaluate the Chanca variant, and inhabitants of the department of Ancash (n=219) to analyze the Central Quechua variant. All participants were bilingual Quechua speakers (Quechua and Spanish), with sufficient academic background to read Quechua (at least incomplete primary school). Those who spoke a variety of Quechua other than the three studied were excluded.

Instrument

The PHQ-9 consists of 9 items corresponding to DSM-IV depressive symptoms ⁶. The response options evoke the frequency of occurrence of such symptoms in the last two weeks, considering the following Likert-type scale: 0 = not at all, 1 = several days (1-6 days), 2 = most days (7-11 days), 3 = almost every day (12 days or more). The Spanish version of the PHQ-9 has demonstrated validity (e.g., goodness of fit as a unidimensional measure: CFI [comparative fit index] = 0.936; RMSEA [root mean square error of approximation] = 0.089; SRMR [standardized root mean square residual] = 0.039), as well as adequate reliability (α = ω = 0.87) in the Spanish-speaking Peruvian population ¹.

Covariates

They were used to characterize the population, as well as to study the measurement invariance of the unidimensional model according to age (in years), sex (female, male), education (incomplete primary, complete primary, incomplete secondary, complete secondary, incomplete higher, complete higher), marital status (single, married/cohabiting, divorced/separated, widowed) and place of residence (rural/urban).

Procedure

For data collection, one interviewer for the Central Quechua variety, two interviewers for the Chanca Quechua variety, and three interviewers for the Cuzco-Collao Quechua variety were trained in the use and application of the instrument. All interviewers were third- or fourth-year psychology students. The interviewers identified WhatsApp groups of parents from educational institutions, Christian churches and groups of peasant community associations to whom they presented the survey in Google Forms format. Data collection began in February and ended in June 2022.

Statistical analysis

Relative and absolute frequencies were used during descriptive analysis ³⁰. Subsequently, we conducted a Confirmatory Factor Analysis (CFA) of the unidimensional model (its unidimensionality was verified in Peru in a representative sample of 30,449 individuals: https://doi.org/10.5281/zenodo.8312191) using a WLSMV estimator (weighted robust squares with adjusted mean and variance), as we did previously ¹. In addition, the WLSMV is an unbiased estimator for items with ordinal responses and non-normal distribution ³¹. We report the standardized betas of the model and standard goodness-of-fit measures: the chi-square (X²) for the model versus baseline, considering values <3 as acceptable; the CFI, which is adequate when >0.90; the TLI (Tucker-Lewis index), which is acceptable with values >0.90. Likewise, the SRMR and RMSEA, were considered adequate with values ≤ 0.08 ³².

We assessed measurement invariance across groups defined by Quechua variants, sex and location using a multigroup CFA. The change in CFI (ΔCFI) and RMSEA (ΔRMSEA) was used as the main criterion to compare models with more constraints versus models with fewer constraints. Models first assumed configurational invariance (i.e., similar factor structure across groups) as the base model, moving up to metric invariance (i.e., similar factor loadings and factor structure across groups), strong invariance (i.e., similar thresholds, factor loadings, and factor structure across groups), and strict invariance (i.e., similar residual variances of items, thresholds, factor loadings, and factor structure across groups). Between each model, we examined whether ΔCFI < 0.01 or ΔRMSEA < 0.01 in order to establish whether the more restricted model was appropriate ³⁰.

Alternatively, MIMIC (Multiple Indicators and Multiple Causes) models were adjusted for the evaluation of measurement invariance according to age, marital status and educational level (variables for which the multigroup CFA was not feasible). We assessed the invariance of indicator intercepts and mean differences of latent dimensions, all across groups according to these covariates. We preferred to use ∆CFI rather than ΔX² because the former is not affected by sample size or model complexity ³². For multigroup CFA, we performed a sensitivity analysis using the method recommended by Yoon and Lai ³³ for dissimilar sample sizes. In fact, each covariate was evaluated separately, comparing two types of models for each of them: 1) a saturated version where the covariate explains all the observed items, but not the latent dimensions, and 2) a version of the invariant intercept model where the covariate explains all the latent dimensions, but not the items. Similarly, the above fit indices are reported and interpreted.

Finally, reliability was determined using Cronbach’s Alpha ³⁴⁾ and McDonald’s Omega ³⁵⁾ coefficients. To ensure reproducibility, the main codes of the analysis can be found at: https://github.com/JCBAZO/R-PHQ9-Quechua. All analyses were performed in R Studio version 4.0.4, with the packages “lavaan” ³⁶, “lavaan.survey” ³⁷ “semTools” ³⁸, “semPlot” ³⁹⁾ and “Psych” ⁴⁰.

Ethical Considerations

This study was evaluated and approved by the Ethics Committee of the Faculty of Health Sciences of the Universidad Peruana Unión with report number 2022-CE-FCS-UPeU-059. The instrument was applied as a self-report in virtual format (Google Forms), designed to present the informed consent first, so only those who agreed to participate in the study accessed the survey. Authorization was requested from the copyright owner of the original instrument (Pfizer) via e-mail, who authorized its use and adaptation.

Cultural adaptation phase

The experts and the research team interacted until achieving the highest rating of appropriateness for each item with respect to its cultural adaptation (3 on a range of 0 to 3), highlighting its relevance, representativeness, clarity and cultural equivalence. Additionally, they provided valuable suggestions which are mentioned below.

Some recommendations contributed to improve the adaptation of the word “depression”, which was initially translated as “llaqui”. Experts suggested to add “sinchi”, which gives a greater intensity, leaving “sinchi llakisqa” (a lot sadness) as the best understanding of depression (item 2). Regarding the expression “without hope”, the experts recommended implementing the question as “manañan q’anchay karqanchu kausayñiykipi / yanqallañan kausaranki” (item 2). For contextual reasons, the daily activity “reading the newspaper” was changed to “listening to the radio”, because the latter is the most used media in the three regions (item 7).

The response options (Likert-type) of the translated PHQ-9 also required special attention on the part of the judges and focus group participants. In particular, the category “almost every day” presented problems of clarity in its translation. The team’s final recommendation was to use the expressions “Mana hayk’aqpas”, “Wakin p’unchawkunalla”, “Ashka p’unchawkuna”, “Yaqa llapa p’unchawkuna”, representing the Spanish equivalent of “Never”, “Some days”, “Several days”, “Almost every day”.

Psychometric analysis phase

Characterization data

Of the 970 bilingual participants, 560 (57.7%) were female, 580 (59.8%) were between 18 and 30 years old, 621 (64%) reported being currently at or having completed college, 577 (59.5%) were single, and 614 (63.3%) lived in an urban area (Table 1).

Validity of internal structure

The three final instruments (one for each variety of Quechua) were validated independently. The single-factor model reported adequate goodness-of-fit values for the three varieties of Quechua (Central, Chanca and Cuzco-Collao), as well as for the total sample (CFI= 0.990; TLI= 0.987; SRMR=0.048; RMSEA= 0.071) (Table 2). On the other hand, the sensitivity analysis for the multigroup CFA reported results similar to those shown here. The only latent factor of the measurement model (depression) loaded a minimum of λ = 0.57 and a maximum of λ = 0.79 to the PHQ-9 items (Figure 2).

Figure 2 General measurement model of the PHQ-9 with betas and standardized errors (N=970). 

Reliability

The PHQ-9 reported good reliability for all Quechua varieties, with Cronbach’s Alpha values between 0.865 and 0.915, and with Omega values between 0.833 and 0.881 (Table 2).

Measurement invariance

The results of the multigroup CFA confirm the invariance of measurement according to Quechua variants, sex and place of residence (Table 3). The configural model showed good fit to the data (CFI = 0.935) for the three Quechua variants. After confirming configural invariance, metric invariance was proven with ΔCFI = 0.001 and ΔRMSEA=0.007 (both <0.01). Then, the values of ΔCFI=0.018 and ΔRMSEA=0.005 demonstrated strong invariance, with at least the latter being <0.01. Finally, strict invariance was confirmed with ΔCFI=0.009 and ΔRMSEA=0.001. The results and interpretation were similar for the invariance models according to sex and residence (Table 3).

The results of the MIMIC models confirmed the invariance by age, marital status and educational level (Table 4). The CFI and TLI values were >0.98 for all the variables, while the SRMR and RMSEA values were <0.08. Likewise, the absolute values of ΔCFI, ΔTLI and ΔRMSEA were <0.01.