Psychometric Properties and Hidden Dimensionality of the Sinhala Version of End Stage Renal Disease Adherence Questionnaire (SINESRD-AQ) Among Patients Receiving Haemodialysis in a Selected Hospital in Sri Lanka

Adherence to a complex therapeutic regimen is crucial to maintain the quality of life of patients receiving haemodialysis. End Stage Renal Disease Adherence Questionnaire (ESRD-AQ) is a valid and most widely used instrument to assess all dimensions of treatment adherence among patients receiving in-center haemodialysis. Though the precise evaluation is fundamental, availability of valid and reliable instruments to evaluate treatment adherence is limited in Sri Lanka. Therefore, the purpose of this study was to evaluate the psychometric properties of Sinhala translated version of ESRD-AQ.

Renal replacement therapy including long term dialysis or kidney transplantation is needed for patients with ESRD for survival [3]. Though the Kidney transplantation is the best of choice in management of patients with ESRD, resource constraints and shortage of kidney donors has made haemodialysis as most e cient and practical treatment modality in Sri Lanka [3]. The statistics of National Hospital of Sri Lanka demonstrate that the number of dialysis sessions carried out has dramatically increased from approximately 3000 in 2003 to more than 11600 in the year 2013 with approximately thousand haemodialysis sessions per month carried out in the hospital [4].
The increase of patients with ESRD necessitates management on dialysis for better outcomes, thus the adherence to prescribed treatment regimen is essential [5]. Patients receiving haemodialysis are required to adhere to a complex treatment regimen consisting of regular haemodialysis attendance, prescribed medications, recommended diet and uid restrictions [5]. Although the health care team fosters adherence, the ability to comply with treatment regimen depends on the understanding and motivation of the patients with ESRD. Nevertheless, recently published literature revealed that approximately 50% of individuals with ESRD undergoing haemodialysis did not adhere to their prescribed treatment regimen [6]. In fact, nonadherence has been associated with increased morbidity and mortality among patients resulting in high cost for the health care system and substantial burden on their family [7].
Wide variations in the reported non-adherence rates have been observed due to lack of reliable measurement tools that address all the facets of treatment adherence behavior of the patients on maintenance haemodialysis. Adherence to haemodialysis treatment regimen is measured by a variety of methods including clinical measures such as biological markers (interdialytic weight gain) and biochemical markers (serum creatinine, blood urea nitrogen, serum potassium and phosphate levels etc.). In general, biological and biochemical markers can be viewed as objective measures while a valid and reliable tool to assess subjective phenomenon was in dearth in literature [8]. Few self-reported adherence scales have been developed and tested for use with patients with ESRD. Assessment of treatment adherence among patients receiving haemodialysis favors the use of multidimensional instrument to evaluate all the dimensions of adherence to the treatment regimen.
End Stage Renal Disease Adherence Questionnaire (ESRD-AQ) is one of the widely accepted multidimensional instruments designed by Kim et al. in Los Angeles, California to assess self-reported treatment adherence behavior in patients receiving haemodialysis [8]. ESRD-AQ has been culturally adapted and validated into different languages and used to assess treatment adherence among patients receiving haemodialysis in Portugal, United states, Spain, Brazil, Egypt and Palestine [3,7,8,9,10,11]. In each language version, ESRD-AQ demonstrated acceptable validity and reliability. The content validity of the ESRD-AQ was assessed by calculating the Content Validity Index (CVI) for each item based on the ratings of an expert panel according to the degree of relevance of each item. Construct validity has been assessed by employing a known group analysis comparing adheres group with a group of non-adheres. Reliability of the instrument was evaluated with test-retest reliability using intraclass correlation coe cient (ICC) in a random sample. The original study conducted by Kim et al., demonstrated good Item level Content Validities for the forty-six items of the questionnaire ranged between 0.86-1.00, which resulted in average of I-CVI of 0.99. ICCs were ranged from 0.83-1.00. The mean scores from the directly measuring adherence behaviors on four different areas (haemodialysis attendance, prescribed medications, recommended diet and uid restrictions) of treatment adherence indicated that the ESRD-AQ clearly distinguished adheres and non-adheres [8]. The validity and reliability indices (CVI = 0.97 -0.99, Average CVI= 0.98 and ICC= 0.82 to 1.00) of the Spanish version (SESRD-AQ) were also within an acceptable range [9]. The Portuguese version of ESRD-AQ (PESRD-AQ) showed strong test-retest reliability across all the items and I-CVI ranging from 0.94-1.00 with an average I-CVI of 0.98 [7]. The Brazilian version showed I-CVI for all the items greater than 0.8 and average I-CVI of 0.96. It also showed high reliability with ICC of 0.98 and 0.91 for direct adherence behavior subscale and attitude/perception subscale respectively. The study further evaluated kappa statistic coe cient for each item of the two subscales of ESRD-AQ which were greater than 0.8 except two items presenting 0.45 and 0.79. Internal consistency (Cronbach's alpha) was 0.57 considered not satisfactory. Acceptable level of Cronbach's alpha is not necessarily important for the instrument as the six questions measuring adherence evaluate different aspects of treatment regimen and no homogeneity between items [10].
With the growing CKD problem and increasing number of patients receiving haemodialysis in Sri Lanka, a valid and reliable instrument is required to determine the degree of adherence to the haemodialysis treatment regimen among Sinhala speaking patients. It enables to address any de ciencies identi ed with treatment adherence among patients to prevent the occurrence of adverse events. Among the previous studies conducted in Sri Lanka, none have used ESRD-AQ to assess self-reported adherence to haemodialysis treatment regimen and no other Sinhala version of instruments measuring adherence are available for use among Sinhala speaking patients. Thus, availability of an instrument from Sinhala language is immensely worth to measure self-reported adherence in patients receiving haemodialysis in Sri Lanka. Therefore, the aim of the study was to evaluate psychometric properties of Sinhala translated version of ESRD-AQ for use among patients receiving haemodialysis in Sri Lanka.

Study design and setting
The study was a validation study which consisted of translation, cultural adaptation and evaluation of psychometric properties (face, content, consensual and construct validity) of the instrument (Figure 1). A descriptive cross-sectional study was conducted among Sinhala speaking patients receiving haemodialysis in Teaching Hospital, Kurunegala, Sri Lanka to evaluate construct validity of the instrument. Teaching Hospital, Kurunegala which is located nearly 100 km away from the capital city of Sri Lanka, has one of the main government sector haemodialysis units. It provides haemodialysis treatment to approximately fortyfty patients daily during morning, afternoon and evening shifts.
The procedure of translation and cultural adaptation of original English version of ESRD-AQ into Sinhala

Language
The translation process was conducted according to the recommended guidelines provided by Beaton et al. on cross-cultural adaptation of self-reported measures [12]. The translation process included ve stepsforward translation, synthesis of common version, back translation, expert committee consolidation and pretesting. The initial translation of ESRD-AQ from the original English language to the Sinhalese language was done by two independent professional bilingual translators (language expert and subject expert).
Both translations were compared, and discrepancies were resolved with the addition of an unbiased bilingual translator who was not involved in the forward translation process and synthesized a common Sinhala version of the instrument. The forward translated common Sinhala version of the instrument was independently back translated to the original English language by two other independent professional bilingual translators (language expert and subject expert). Both translations were compared, and discrepancies were resolved with the addition of an unbiased bilingual translator who was not involved in the back-translation process and produced a common English version of the instrument. A committee consisted of experts who were familiar with the constructs of interest, questionnaire designing and subject experts in Index for the whole scale (S-CVI) [13], Item-level Content Validity Ratio (I-CVR) [14] and Modi ed Kappa statistic coe cient (k) [15] for each item were calculated.
The resulting version was administered to a purposively selected group of patients (n=10) receiving haemodialysis to evaluate face validity. The patients were asked to comment on each item of the instrument to determine any di culty in understanding the words of the items or answer the items. In fact, their feedback was used to improve comprehensibility of the items of the instrument. Based on comments of the participants, no further modi cations were indicated in the instrument. The nal SINESRD-AQ was produced, and proof reading was carried out to eliminate the spellings, grammatical and formatting errors.

Sample and sample size calculation
The study included all the patients above the age of 18 years, diagnosed with ESRD, receiving haemodialysis at least for six months and able to speak Sinhala language from all ethnic groups attending haemodialysis at Teaching hospital, Kurunegala, Sri Lanka. Patients who were critically ill were excluded from the study. The sample size was calculated considering the rule of thumb 5-10 subjects per each item [16]. Considering 14 items in two subscales measuring adherence in ESRD-AQ, calculated sample size for the current study was 140 (1:10). Considering nonresponse rate of 5%, the nal calculated sample size for the study was 150. Participants who ful lled selection criteria were recruited for the validation study until the sample size was achieved.

Data collection procedure
Data collection was carried out from November 2019 to April 2020. Patients who were willing to participate in the study were recruited after obtaining written informed consent. Subsequently, data were collected using interviewer administered SINESRD-AQ and using a self-designed questionnaire for socio-demographic characteristics regularly by the principal investigator without disturbing the haemodialysis treatment they received. An independent sample of ten patients receiving hemodialysis completed the questionnaire twice with a retest interval of one week in order to assess test-retest reliability of the instrument. Construct validity of the SINESRD-AQ was assessed by Con rmatory Factor Analysis (CFA) using LISREL 11.0 software for windows.
Prior to perform factor analysis, the data set was assessed for suitability, quality, missing data and for violations of the assumptions of analytical techniques of CFA. No missing data were observed. The sample data assumed univariate marginal distribution. Both Shapiro-Wilk test and Kolmogorov-Smirnov test were involved in assessing univariate normality of data as the sample size is greater than 50 [17]. The p value was 0.000 (p<0.05) for all the variables in the data set which rejects the null hypothesis for the normality for the sample at a 5% level of signi cance assuming non-normality of data. Considering total adherence score and perception score yielded from two subscales of SINESRD-AQ, standardized values (z scores) were computed to identify univariate outliers. Any Z score above +3.29 or below -3.29 is considered as an outlier case [18].
Five extreme outliers (case 55,81,86,88 and 101) were detected and removed prior to performing CFA. The raw data set was checked for errors in data entry. Since the absence of errors, no adjustments were made to the data set prior to perform CFA. The sampling adequacy was assessed using the Kaiser-Meyer-Olkin (KMO) test. A value 0.663 indicating that the sample is barely adequate to perform factor analysis. Bartlett's test of sphericity was less than 0.001 indicating an adequate amount of collinearity between items [19].
Robust unweighted Least Squares (RULS) estimation method in LISREL 11.0 which was especially designed for ordinal data and adjusted to non-normality was used to estimate model parameters as it yields better results in lower sample sizes and estimates parameters with less standard error compared to the other estimation methods [20]. Index] were used to evaluate the goodness of t of the CFA model of two subscales of ESRD-AQ and to examine the adequacy of the model t to the data. RMSEA index less than 0.06 and SRMR less than 0.09 [21,22] were considered as an acceptable model t. For the GFI, AGFI, CFI and NNFI [22], values greater than 0.95 were considered as good t to the data. While no threshold levels have been recommended for

Results
Socio-demographic characteristics of the participants The mean age of the study participants was 54.08±10.78 (±SD years) (Age range 18-74 years). Majority were males (72.7%), married (96.0%), Sinhalese (98.0%) and not employed (86.0%). More than 70% (71.3%) of patients were diagnosed to have chronic kidney disease for less than ve years and majority (64.7%) were receiving hemodialysis for less than one year (Table 1).   [19]. Eight Outliers (case 5,9,11,20,32,87,92 and 101) were detected in the data set after performing initial CATPCA analysis by looking at objects scores that roughly exceeding the range of -3.5 -+ 3.5 [27]. Then the analysis was re-run after removing those outliers with a Ten preset number of components. In addition to CATPCA, O'Connor parallel analysis (2000) was performed to determine the exact number of components to be extracted for further analysis [28].
Considering the eigenvalues obtained from both CATPCA and O'Connor parallel analysis, ve components were extracted with eigenvalues greater than 1 (Figure 2). The Cattell's (1966) scree plot also revealed a clear break in the slope after the 5th factor suggesting that ve components to be retained for further analysis ( Figure 3).
The analysis was re-run with ve preset number of dimensions/components. Values of Variance Accounted For (VAF) each item was obtained by CATPCA analysis to determine the number of items to be retained in nal analysis considering the communality of each item. Average value of 0.5-0.6 communality is acceptable for sample sizes between 100-200 [29]. Therefore, ten items with communality less than 0.5 were removed and nal analysis was performed with twenty-two items. After performing nal CATPCA, ve components with eigenvalue greater than 1 were extracted and approximately 74% of total variance was explained by those ve components. Eigenvalues for extracted components ranged between 6.455 -1.741 (Table 2). All twenty-two items had communality greater than 0.5 and factor loadings greater than 0.6 ( Table 3).

Reliability analysis -Internal consistency
Internal consistency reliability of new dimensionality was assessed with Cronbach's Alpha and composite reliability. Cronbach's Alpha for three components were greater than 0.6 and two were demonstrated 0.45.
Composite reliability (CR) values were greater than 0.7 for all the components.  .929

Convergent and discriminant validity of new dimensionality of SINESRD-AQ
The calculated AVE for each component was greater than 0.5 and CR was greater than 0.7. According to Fornell and Larcker criterion, the square root of AVEs of components are higher than the inter construct correlations (Table 4) and Heterotrait-monotrait (HTMT) ratio of correlations were less than 0.85 (Table 5).   According to the Hu and Bentler's two index presentation strategy (1999), it is required to meet thresholds of combination of SRMR<0.09 with another model t index including either NNFI (TLI)>0. Alpha coe cient for scales with similar length and variance. Highly correlated items or increasing the number of items of a scale will also produce high Alpha coe cient [33]. In the present study, new dimensionality has no similar length in the scales neither item of each scale are not assessing the similar content of adherence, i.e. barriers to follow uid restriction recommendations -consists items relevant to frequency of following uid restriction recommendations, the level of di culty and type of di culty in following uid restriction recommendations. Therefore, the internal consistency reliability might not be an accurate method to assess reliability of this tool, whereas more valid reliability measures such as test-retest reliability can be assessed to determine the reproducibility and ability to provide consistent scores overtime with the scale [32].
The construct validity of new dimensionality was assessed by convergent and discriminant validity. The current study has a few limitations. The authors have not performed con rmatory factor analysis to con rm the construct validity of new dimensionality of SINESRD-AQ which gives more insight and con dence to assess treatment adherence behaviors among patients receiving haemodialysis. Test-retest reliability which is a more powerful measure to assess reproducibility of the instrument was also not evaluated on new dimensionality of SINESRD-AQ in the current study. It is recommended to conduct future studies with a larger sample size to examine validity and reliability of new dimensionality of ESRD-AQ using different settings and to check whether those ndings are consistent with the results of the current study.

Conclusions
The current study concludes that the SINESRD-AQ is a scienti cally and statistically valid and reliable instrument to assess treatment adherence behavior among Sinhala speaking patients receiving haemodialysis in Sri Lanka. Availability of data and materials The raw data utilized to support the ndings of the current study will be available from the corresponding author upon reasonable request.
Ethics approval and consent to participate Written and verbal informed consent was obtained from all the patients prior to commencement of data collection of the study.