Introduction

Recently, treatment of HIV infection has shown remarkable progress. In the 1990s, a combination of drugs exhibiting a high capacity to suppress viral loads was discovered. Reverse transcriptase inhibitors along with protease inhibitors comprise what is now called highly active antiretroviral therapy (HAART). Can reduce the viral count to undetectable levels, thus improving the functioning of the immune system, preventing transmissibility, and reducing morbidity and mortality ^1-3. However, achieving these effects requires a treatment adherence level of close to 90%-95% ⁴; therefore, optimal level of treatment adherence is a required to achieve undetectable viral load, which was one of UNAIDS'S 90-90-90 goal ⁵.

A meta-analysis reveal that approximately 60%-70% of patients undergoing treatment in different countries ⁶ show this level of treatment compliance, particularly those in Latin America and the Caribbean ⁷. Owing to the problems that may arise from poor treatment adherence, such adherence is one of the main challenges associated with the treatment of HIV infection today. Notably, inadequate adherence can lead to the development of strains that are resistant to the most effective treatment available to date ¹. Furthermore, risk behaviors (e.g., sexual intercourse without a condom) expose uninfected individuals to these strains and those with HIV to re-infection, thus limiting treatment options and worsening the situation ^1,3.

Different techniques have been proposed to address the lack of long-term treatment adherence, including the Information, Motivation, and Behavioral (IMB) skills model ⁸. This concept differs from others as it was designed specifically for HIV-infected individuals. The IMB model allows the identification of key intervening factors that can help evaluate and understand the problems associated with HAART adherence as well as determine the target variables to design interventional strategies.

The IMB model uses a simple and dynamic scheme that links different factors that influence HAART adherence. It specifically states that if a person with HIV has enough information about items, including regime and side effects, and the necessary motivation to undergo successful treatment (favorable attitudes towards adherence and perception of social support), then they will put specific behavioral skills into practice to achieve treatment adherence, such as through acquiring the drugs and including the regimen in their daily routine. These skills will result in treatment adherence behavior ⁸.

The Information factor here refers to aspects including the dose, how and when to take the medication, side effects, degree of treatment adherence required, and interactions with other drugs ⁹. Different studies have highlighted the impact of this factor on adherence to antiretroviral treatment ^10-15.

The individual's attitude towards treatment adherence and perceived social support are included in the Motivation factor. In this sense, Wasti et al. state that not informing others about their diagnosis increases the probability of non-adherence to treatment ¹⁵. Several studies support this hypothesis for antiretroviral treatment as well as for treatment in general ^{10,13,14,16,17}.

Other studies consider Information and Motivation as independent factors that play important roles in preventive behaviors that improve the health status ¹⁸, with Information being mandatory for treatment adherence ⁹.

Behavioral skills in the IMB model include the individual's specific abilities to be adherent and the self-efficacy for all actions related to treatment adherence ⁹. Several studies refer to the relationship between specific behavioral skills, perceived self-efficacy, and adherence to HAART ^10,14,19,20.

To assess the barriers to treatment adherence, an instrument was developed within the framework of the IMB model called the "Life Windows - Information, Motivation, Behavioral Skills-Antiretroviral Treatment Adherence Questionnaire" (LW-IMB-AAQ) ²¹. This model has been used in different countries including India ²², Romania ²³, South Africa ²⁴, and Ethiopia ²⁵. The use of the Motivation subscale has been reported in Argentina as well ²⁶. However, many of these studies do not report the adaptation and validation of the instrument for local use. The original study underestimates the cultural differences that can affect not only the validity of the instrument but also that of the theoretical model; therefore, psychometric investigations of the model have been conducted in populations other than that of the original study ²⁷.

One of the LW-IMB-AAQ adaptation studies was conducted by Peng et al. for application in Shanghai ⁴. The authors proposed a structure with three sections that corresponded to the three subscales of the original instrument based on the confirmatory factor analysis but with two subscales in each of these sections. Items 1 and 4 of the motivation subscale were eliminated for the model to yield acceptable fit indices. The reliability values (a) were .84 for Information, .81 for Motivation, and .91 for Behavioral Skills.

The psychometric properties of the LW-IMB-AAQ were assessed in another study ²⁸. Exploratory factorial analysis was conducted with a varimax rotation where the original factorial structure was replicated. Further, validity in terms of other variables whose oblique factors were related to measures of treatment adherence was evaluated. Although it is important to acknowledge this effort, it should be noted that there are some limitations to the procedures. Confirmatory factor analysis is currently recommended when starting from a delimited structure ²⁹. Furthermore, varimax rotation assumes that the factors are orthogonal, a condition that is not met in the LW-IMB-AAQ. Therefore, in this case, the use of oblique rotations is suggested as they are more in line with the representation of the constructs studied in psychology ³⁰.

A literature review revealed that no instruments for measuring antiretroviral treatment adherence and/or factors related to it that have been developed, adapted, and/or validated in Argentina. A questionnaire to evaluate the possible drawbacks in adherence with the appropriate properties and technical requirements is needed to address the challenges posed by the current HIV epidemic ^14,13,31. Therefore, this study aimed to validate the instrument for the application in the Argentinian population.

Materials and methods

he sample population comprised 190 people with HIV selected using the nonrandom accidental type who were undergoing antiretroviral treatment at the Córdoba province, Argentina. Ten people were excluded from the study because of missing data, thus exceed-ing the criteria suggested for the data imputation process ³². The final sample population included 180 people. The mean age of the participants was 40.61 (SD = 12.032) years and 82.8% were men. Regarding ongoing medication, 81% stated that they complied with the schedules but 60% stated that they sometimes forgot.

We included patients aged >18 years who were undergoing antiretroviral treatment and provided their informed consent in compliance with the ethical aspects listed in the Declaration of Helsinki ³³. The project was endorsed by the Research Ethics Committee of the College of Psychologists of the Province of Córdoba (File No. 3678).

Two main instruments were used: a self-reported adherence questionnaire and the instrument to be adapted. Data collection was conducted using a questionnaire on sociode-mographic variables and others topics related to the positive diagnosis of HIV as a control.

The questionnaire on sociodemographic and other variables was developed ad hoc and it included questions about gender, age, occupation, date of HIV diagnosis, the person(s) the patient lives with, level of education attained, how they found out about the diagnosis, how many people they told about their diagnosis, and their relationship with those individuals.

The Simplified Medication Adherence Questionnaire (SMAQ) ³⁴ was used to asses treatment adherence. The structure of this test is unidimensional and it comprised six items, five with a dichotomous response option and one graded option. Questions were about compliance or noncompliance to drug intake according to the schedule determined for the patient during the past 3 months, the last weekend, and the last week. Forgetfulness, carelessness with the times of the shots, and differential behavior due to discomfort are also explored. This instrument was previously used in a study on an Argentinian population with HIV whose results reported that its validity and reliability were adequate ³⁵.

The LW-IMB-AAQ test comprises 33 items grouped into three subscales: Information, Motivation, and Behavioral skills. The first includes 9 items, the second includes 10 items, and the third includes 14. The psychometric properties are reported together with the publication of the original version of the instrument. The Information subscale had a low index (a = .59) of internal consistency, which was expected because the items were intended to consider aspects of the treatment that were not interrelated. Reliability indices (a) of .70 and .75 were found for the Motivation subscale. Finally, the Behavioral Skills subscale presented reliability indices of .90 in all except the first item ²¹. A Spanish version was provided by Amico (personal communication, November 16, 2016).

The methodological design of this study meets the criteria established by Montero and León for instrumental studies ³⁶. Participants were invited to participate and informed about the attendance policy for patients regarding the medical appointments of the institution and private offices. Questionnaires were provided after they expressed their agreement and signed the informed consent form. The data were then systematized for the corresponding analyses and tests.

A confirmatory factor analysis was conducted using the MPLUS v.7 program, for the evidence of validity of the internal structure through the weighted least squares means and variance adjusted as a parameter estimation method ^36,38. It is more precise when assessing ordinal variables with distributions that do not approach normality and with small samples ^39-41; factor loads greater than .40 were considered acceptable ⁴². The adjustment of the measurement model was assessed following the recommendations of Hu and Bentler, Steiger and DiStefano, Liu, Jiang and Shi where CFI values > .95, RMSEA < .07, TLI > .95, WRMR < .90, are considered as indicators of acceptable fit ^43-45. Transverse to this entire process, the modification indices were also analyzed where values >.20 of the Expected Unstandardized Parameter Change (EPC) indicated specifications that had to be performed in the evaluated model ^46,47.

The structural hypothesis initially suggested by the original authors of the instrument was analyzed. According to this model (M1), the 33 items can be grouped into three oblique factors (M1). New models (M2 and M3) were proposed based on the previous one, with the purpose of improving the fit of the measurement model.

The reliability of the scores was then estimated using the omega coefficient (i) and the correction for correlated errors (to'), according to the recommendations of Gadermann et al., and Greco et al. ^48-52 for interpretation.

We finally sought to provide evidence of validity regarding the other variables. Specifically, a concurrent validity analysis was conducted with the aim of analyzing the relationship between treatment adherence (total SMAQ score) and the three subscales of the LW-IMB-AAQ using linear regression analysis. Statistically significant relationships can be expected based on previous studies ⁵³.

Results

Descriptive analysis of the data was conducted first. As shown in Table 1, 27% of the items have asymmetry and kurtosis values >±2, suggesting that the data does not approximate a normal distribution ^42,54.

Further, the fit of the measurement model was assessed. Unacceptable fit indices were obtained for the original model (M1), which were items with factor loads of <.40 and EPC values that suggested adding specifications between the items of the three oblique factors. Considering this, a review was performed based on the content of the items where potentially overlapping reagents were identified. Accordingly, items 3 and 5 (Information), 2, 4, 5, 6, 8, and 9 (Motivation), and 1, 3, and 4 (Behavioral Skills) were removed from this process.

Although the fit indices improve for M2, they do not optimally meet the criteria mentioned in the literature. Furthermore, there are still EPC values >.20 that are related to the correlations between items that can be considered to be under parameterization. As a result, items 2 and 13 (Behavioral Skills) were removed as well.

Finally, for M3, a model of three oblique factors with acceptable fit indices was observed, with factor loads >.40 and without specifications to add or remove from the proposed model (EPC < .20) (Table 2).

Third, the internal consistency was analyzed considering the different models. The reliability of the M1 scores were as follows: to_F1 = .765/to'_F1 = .757; ω_F2 = .781/ω'_F2 = .702 and ω_F3 = .889/ω' _F3 = .886; for model M2: ω_F1 = .832/ω' _F1 = .767; ω_F2 = .759 and ω_F3 = .917/ ω'_F3 = .873; and for model M3: ω_F1 = .833; ω_F2 = .759 and ω_F3 = .888. In all cases, it can be inferred that the values are acceptable.

As a last step, evidence on the validity in terms of other variables was collected. As shown in Table 3, there is a statistically significant relationship between the scores of the Information and Behavior subdimensions of the LW-IMB-AAQ with treatment adherence. The evaluated model presents a value of R ² = .134, which is considered mid-level ⁵³. The Motivation subdimension did not provide statistically significant values.

Discussion

The objective of this study was to evaluate the psychometric properties of the LW-IMB-AAQ in a sample population including individuals diagnosed with HIV in Córdoba, Argentina. This study was needed in the absence of other studies, specifically those on the internal structure validity, relationship with other variables, and analysis of the reliability of the scores. All this was performed as per the recommendations of the Standards for Educational and Psychological Testing ⁵⁵. The final version of the instrument included 20 items in the Information, Motivation, and Behavioral Skills subscales, with significant associations with treatment adherence and acceptable reliability results for application in the Argentinian population.

The results indicated that the original factorial structure is not replicated in terms of the evidence of validity of the internal structure. There were specifications in M1 that suggested adding relationship parameters between items, allowing us to infer a possible factorial complexity because of the poorly differentiated factorial loads of some items and because there were high interfactorial correlations between the Motivation and Behavioral Skills subscales ^43,56. This evidence is supported by the high correlations between the same subscales by Santillán-Torres-Torrija et al. ²⁸. For this reason, in view of the previous issues and the representation of the construct through the theoretically proposed components, we decided to remove items with redundant content and low factor loading so we could obtain a robust version of three oblique factors that continues to represent what was proposed by Fisher et al. ⁹.

The working hypothesis was that the subscales measured through the SMAQ are important predictors of treatment adherence for the evidence of validity regarding other variables of the final version of the LW-IMB-AAQ. The results of the linear regression analysis indicated that the factors of Information and Behavioral Skills are the most important predictors of the variance associated with taking medications. However, when analyzing the propor-tion of variance predicted by each factor, this is considered high for all subscales. This result is consistent with that reported by Fisher et al., who suggest that information is a necessary condition and the most important factor for engaging in behaviors involved in treatment adherence ⁹. Another reason for this is that there are people who have the information but low levels of motivation and vice versa ¹⁸. This result is consistent with those obtained by Botempi et al., Miller et al., Jones et al., Varela-Arévalo et al., Wasti et al., Simoni et al., Horvath et al., Ladero-Martín et al., and Fisher et al. ^{9-12,14-16,19,20}.

The evaluated models can be considered acceptable according to the standards of Gadermann et al. and Greco et al. in terms of score reliability ^51,52. However, there is a difference between the values of the omega coefficient and the correction for correlated errors as it has been reported that the under-parameterization of correlated errors in statistical models causes reliability to be a biased estimator ⁵⁷. One of the causes is the presence of redundant items, as seen in the case of the LW-IMB-AAQ, and can lead false positives and false negatives that have a possible direct effect on decision making ^58-60. These values suggest that the LW-IMB-AAQ can be applied for group description purposes and not for individual decision-making considering that values around .90 are needed in the three subscales.

In short, the findings of this study coincide with that of the original study in terms of factors as well as internal structure ²¹. The factorial structure confirms the presence of three main factors, coinciding with the Information, Motivation, and Behavioral Skills subscales, which was also reported by Santillán-Torres-Torrija et al. ²⁸. After removing two items, Peng et al. ⁴ identified three sections that coincide with the abovementioned factors but with two subscales each. Moreover, the analysis of the internal consistency of the subscales also yielded acceptable values both in this study and in the research by Peng et al. ⁴. However, in the original study and in the one conducted by Santillán-Torres-Torrija et al. the Information subscale showed a a of .70 ^21,28.

There are certain limitations to this study. The sample mainly comprised patients from the same institution, and the majority included patients who adhered to treatment; this could influence the results of the study. Therefore, future studies should include sample populations including participants from both public and private health institutions. Likewise, we recommend increasing the number of cases to obtain greater variability and to conduct more complex analyses, such as structural equations.

This is an initial study on LW-IMB-AAQ; therefore, we recommend evaluating the factor structure through confirmatory analysis of measurement invariance to corroborate its equivalence in different populations, thus contributing to the generalization of its use. This is especially considering that the removal of the items in the final structure of this version of the instrument should be confirmed with other samples for consolidation ⁶¹.

In conclusion, this research constitutes a significant theoretical and practical contribution to the approximation of adherence to antiretroviral treatment, because information was obtained on the best representation of the constitutive definition of the construct among the Argentinian population, and because this study can be a precedent for the LW-IMB-AAQ to be most widely supported, and to be used by health professionals as evidence for the design of specific intervention strategies.