A systematic review of the psychometric properties of the cross-cultural translations and adaptations of the Multidimensional Perceived Social Support Scale (MSPSS)

Background Social support (SS) has been identified as an essential buffer to stressful life events. Consequently, there has been a surge in the evaluation of SS as a wellbeing indicator. The Multidimensional Perceived Social Support Scale (MSPSS) has evolved as one of the most extensively translated and validated social support outcome measures. Due to linguistic and cultural differences, there is need to test the psychometrics of the adapted versions. However, there is a paucity of systematic evidence of the psychometrics of adapted and translated versions of the MSPSS across settings. Objectives To understand the psychometric properties of the MSPSS for non-English speaking populations by conducting a systematic review of studies that examine the psychometric properties of non-English versions of the MSPSS. Methods We searched Africa-Wide Information, CINAHL, Medline and PsycINFO, for articles published in English on the translation and or validation of the MSPSS. Methodological quality and quality of psychometric properties of the retrieved translations were assessed using the COSMIN checklist and a validated quality assessment criterion, respectively. The two assessments were combined to produce the best level of evidence per language/translation. Results Seventy articles evaluating the MSPSS in 22 languages were retrieved. Most translations [16/22] were not rigorously translated (only solitary backward-forward translations were performed, reconciliation was poorly described, or were not pretested). There was poor evidence for structural validity, as confirmatory factor analysis was performed in only nine studies. Internal consistency was reported in all studies. Most attained a Cronbach’s alpha of at least 0.70 against a backdrop of fair methodological quality. There was poor evidence for construct validity. Conclusion There is limited evidence supporting the psychometric robustness of the translated versions of the MSPSS, and given the variability, the individual psychometrics of a translation must be considered prior to use. Responsiveness, measurement error and cut-off values should also be assessed to increase the clinical utility and psychometric robustness of the translated versions of the MSPSS. Trial registration PROSPERO - CRD42016052394. Electronic supplementary material The online version of this article (10.1186/s12955-018-0912-0) contains supplementary material, which is available to authorized users.

Various outcome measures have been developed to measure SS [3,11]. Originally created to measure SS in American adolescents, the Multidimensional Scale of Perceived Social Support (MSPSS) has evolved as one of the most extensively used SS outcome measure [3, 6-8, 11, 12, 18]. The MSPSS has 12 items that measure the perceived adequacy of the available amount of SS [15,21,22] (See Additional file 1). It measures the amount of SS an individual receives from three sources i.e. friends, family and significant other/special person. The amount of SS is rated on a seven-point Likert scale; with responses ranging from very strongly disagree (=1) to very strongly agree (=7). The cumulative/total scores ranges from 12 to 84. As no item response theory calibration has been applied to the tool, the scores are interpreted as, the higher the score, the greater the amount of available SS [21]. The original version of the MSPSS yielded a three-factor structure, high internal consistency (α = 0. 88), stability (yielded α = 0.85 after 3 months from first administration) and moderate construct validity as the SS scores were negatively correlated to anxiety (r = − 0.18; p < 0.01) and depression scores (r = − 24; p < 0.01) [22].
The drive towards evidence-based practise (EBP) and patient-centred care has led to an increase in the crosscultural adaptation and translation of patient-reported outcome measures (PROMs) [7,[23][24][25]. To this end, the MSPSS has undergone extensive translation and adaption across linguistic and socio-economic contexts and settings i.e. from low- [7,13,26] and middle [11,18,[27][28][29] to high-income countries [3,6,9,17,18,[30][31][32][33][34]. However, evidence of the psychometric properties of the translated and adapted versions of the MSPSS is fragmented, but deeply important to both clinical practice and research. For instance, due to conceptual differences, some authors have collapsed the scoring system (response options) to three [30,35], four [12], five [13,26,36], and six [31] levels against the original seven-point Likert scoring system. However, the category "reordering" was based on "qualitative" analysis yet in those circumstances, item response theory techniques such as Rasch analysis are a prerequisite for ensuring the interpretability of response categories for translated tools [37,38]. Secondly, due to linguistic differences, some of the terminology of the original MSPSS have been changed, for example, the term "someone special" has been replaced by the term "husband" for some translations [5]. Thirdly, cultural differences are likely to influence perception of social support [10-12, 17, 18] thus potentially influencing the structural validity of the translated versions of the MSPSS. For example, single factor structures were reproduced in the validation of the MSPSS in Asian countries such as Turkey [39][40][41][42], Thailand [28,43,44] and Pakistan [45][46][47][48][49], which are considered as "collectivistic" societies [50]. This implies that respondents could not differentiate between support provided by family, friends and significant others as postulated by the developers. The MSPSS was originally validated in the US which is considered an "individualistic" society [51]. Given the wide variation in MSPSS translations and cultural adaptations, a systematic evaluation of the adapted and translated versions of the MSPSS will help bring an understanding of the quality of the existing tools, and gaps in knowledge and adaptation [7,23,24,52]. Further, a recent literature review asserts the psychometric robustness of the MSPSS across various settings and study populations [53]. However, the methodologies of the cited studies were not critically appraised. This is a limitation as the reported psychometrics are dependent on the quality of the methodologies applied in accordance with the COnsensus-based Standards for the selection of health status Measurement Instruments (COSMIN) criterion [54][55][56][57] . For instance, issues such as the sampling, handling of missing responses, the model used for analysis, bias in research reporting amongst other factors affects both the internal and external validity of the psychometric evaluation studies [55,56,58]. More so, other psychometric properties such as floor and ceiling effects, critical/cut off values and responsiveness were not reported and this weakens the level of evidence of the purported psychometric robustness of the MSPSS [53]. Therefore, the aims of the present review are to: (I) systematically identify the translated and adapted versions of the MSPSS; (II) evaluate the methodologies applied in the adaptation and translation process; and (III) appraise the psychometric properties of the translated and adapted versions.

Protocol and registration
This review was conducted in accordance with the Preferred Reporting Items of Systematic Reviews and Meta-Analyses Protocol (PRISMA) guidelines (Additional file 2). The protocol was registered with and published on the PROSPERO database (Ref-CRD42016052394) [59].

Eligibility criteria
Studies were included if one of the aim(s) was to: translate and culturally adapt the MSPSS; evaluate the psychometrical properties of translated version(s) of the MSPSS or if they measured SS using translated versions of the MSPSS. Additionally, only studies applying a quantitative study designs and reporting on the psychometrics of the translated, 12-item version of the MSPSS were included. Due to limitation in resources for translation, only full text articles published in English were included. Studies based in more than one country that applied the MSPSS as an outcome measure, commentaries, review articles and studies which utilized several linguistic versions of the MSPSS were excluded. Additionally, studies which utilized both the original version of the MSPSS and the translated version were also excluded as the aim of the review was to appraise the psychometric properties of the translated versions of the MSPSS.

Information sources
We searched Africa-Wide information, CINAHL, PubMed, Psych INFO, and Scopus for peer-reviewed articles. Google Scholar was also searched to identify grey literature such as dissertations. We also contacted the developer of the MSPSS to identify the translations which we might have missed through database searches. We did not impose a time limit to publication dates to gather as many articles as possible. In cases where the abstract was available online and where it was unclear as to whether the translated version of the MSPSS was applied, the authors were contacted for clarification. Reference lists of the identified articles were manually searched for possible articles for inclusion for the attainment of literature saturation. Literature searches were conducted between November 2016 and February 2017. Table 1 below is the search strategy applied in retrieving articles on EBSCO-host search engine:

Outlined in
As an illustration, articles on the translation and adaptation of the French version of the MSPSS were retrieved as follows: {multidimensional scale of perceived social support OR MSPSS OR mspss} AND {translation OR translated OR translat*} AND {france OR France OR French OR french OR francias}.

Study selection
A previously described study selection process was utilized [60,61]. One author (JD) ran the search strategy across all databases. Two independent reviewers (MC and LC) then selected the eligible titles and abstracts for further investigation using a predefined search strategy. Reviewers resolved disagreements about inclusion through discussion, and a, third reviewer (JD) was brought in if the two reviewers could not agree. Another author (MC) manually searched the reference lists of identified articles to screen full texts for inclusion.

Data collection process
The principal investigator (JD) developed the data collection sheet. The tool/checklist was validated by three co-investigators (MC, TM & JD) with prior experience with psychometrics and psychometrics systematic reviews. The tool was then piloted on ten randomly selected studies of definite rating. Three reviewers (MC, JD & TM) independently applied the COSMIN checklist to rate the methodological quality of the ten studies. The inter-rater reliability was 0.8 as measured by the Kappa coefficient. Most of the disparities emanated from the rating of the structural validity and differences were discussed upon further reference to the COSMIN user manual. Afterwards, two reviewers (MC & TM) independently extracted data from the retrieved studies and their inter-rater reliability was 0.9. The two data collection sheets were reconciled into one data set through discussions between the principal author (JD) and two reviewers (MC & TM).

Data items
The extracted information included the research setting and design, study sample, demographic and or clinical characteristics of the participants, target translation language and secondary outcome measures e.g. depression. The methodological quality of the translation process and evidence for reliability and validity of the questionnaires was also documented.

Risk of bias in individual studies
The methodological quality of the retrieved articles was assessed using the COnsensus-based Standards for the selection of health status Measurement Instruments (COSMIN) checklist [54]. It consists of eight boxes which evaluate the methodological quality of the following psychometrics: internal consistency (box A), reliability (box B), measurement error (box C), content validity (box D), structural validity (box E), hypotheses testing (box F), cross-cultural validity (box G), and criterion validity (box H) [54,55]. Methodological quality is rated on a four-point Likert scale i.e. poor, fair, good and excellent. In assessing a domain, the lowest rating of an item is assigned as the overall quality of the domain under investigation. For example, in assessing structural validity, if an inappropriate rotation method is employed i.e. if orthogonal rotation instead of oblique rotation is used to interpret factors on the MSPSS, the evidence will be rated as poor even if all the other sub-domains are rated as of excellent quality [54][55][56][57].

Best evidence synthesis
To come up with best evidence synthesis for each psychometric property, we used the Cochrane Collaboration Back Review Group criterion [62] as outlined in Table 2 to synthesise results from the COSMIN methodological assessment [54][55][56][57] (see Table 3) and Terwee et al.

Study selection
Description of adaptations For seven of the translations, the response options were reduced from the original seven -point Likert scale to a five (n = 4), three (n = 1), four (n = 1) and six-point (n = 1) scale. Some of the original terms on  * and ** denotes findings from the USA and Spain respectively the MSPSS were modified/changed in four of the studies i. e. the term special person/significant other was changed to "husband" or "spouse". The MSPSS was self-administered in most studies [54.5%, n = 12]. Depression, general psychological well-being, social networks and anxiety were the most commonly measured secondary outcome measures Fig. 1 and Table 6.

Results of individual studies Arabic
Two variants of Arabic translations were retrieved [27,35].

Arabic generic version
The Arabic generic version was described in one study [27]. The methodology for this cross-cultural validation study was poor, as scanty details were provided for the adaptation process. The evidence for structural validity was poor as only exploratory factor analysis (EFA) was performed. There was limited evidence for internal consistency (IC) as the handling of missing responses was not reported. There was unknown evidence for construct validity as no specific hypotheses were formulated.

Arabic version for women (MSPSS-AW)
Two studies on the MSPSS-AW were available [35,63]. The methodology for the cross-cultural validation was poor. There was no description of; translators' expertise, whether the translations were done independently, the number of forward and backward translations performed and the reconciliation process. This version was not reviewed by a committee and was not pretested. There was moderate evidence for IC, the handling of missing responses was not reported. There was fair evidence for structural validity, although confirmatory factor analysis (CFA was performed, the number of missing responses Arabic women -USA Poor [5] Fair [5] Fair [5] Fair [5,47] Fair [5] Arabic Generic -Lebanon Poor [27] Poor [27] Fair [27] Fair [27] Fair [27] Chichewa-Malawi Fair [26] Excellent [26] Excellent [26] Fair [26,36]] Chinese (Simplified) -Malaysia Poor [12] Poor [12] Fair [12] Poor [12] Poor [12] Poor [12] Chinese (Traditional) -Hong Kong, China was not documented. There was limited evidence of construct validity as no specific hypotheses were formulated and the psychometrics of the comparator instruments were not adequately described.

Chichewa and Chiyao
Two studies were available [26,36]. The methodology for the cross-cultural validation was fair as only one forward and one backward translations were performed. Further, the expertise of the translators, pre-test sample and the reconciliation of the forward translation were poorly described. There was strong evidence for structural validity and IC. Both EFA and CFA were applied for structural validity evaluation. There was limited evidence for construct validity, no specific hypotheses were formulated.

Chinese
Two versions in simplified and traditional Chinese versions were retrieved [12,17].

Traditional Chinese
Twenty-one studies, applying the traditional Chinese version of the MSPSS, were available [17, . The methodology for the cross-cultural validation was poor. Solitary forward and backward translations were performed; the translators' expertise were not outlined and it was not clear whether the translations were done independently and if the tool was pretested in the target population. Two studies tested structural validity of the MSPSS-C in different populations [17,64]. There was poor evidence for structural validity as only EFA was performed. There was conflicting evidence for IC. The methodological quality of one of the studies was questionable [17] with the second study yielding a Cronbach's alpha< 0.70 despite fair methodological quality [65]. There was limited evidence for reliability as the test conditions and the stability of the re-test sample were not clearly outlined. There was moderate evidence for construct validity, no specific hypotheses were formulated. The evidence for criterion validity was unknown as the psychometrics of the "purported" gold standard measure was questionable.

Creole
Only one study was available [6]. The tool was poorly translated as; there were no multiple translations, the tool was not pre-tested and factorial analysis was not done. The evidence for IC was indeterminate as the subscales unique ICs were not computed and handling of missing responses was not documented. There was indeterminate evidence for reliability as the conditions for the administrations were not clearly stated and the evidence for no systematic change in the outcomes was not provided.

French
Only one study was available [9]. The cross-cultural translation and adaptation process was poor. The tool was not pre-tested and only solitary forwardbackward translations were performed. There was limited evidence for structural validity, IC, reliability and construct validity. The methodologies applied were of fair quality, the handling of missing response was not reported, the conditions for the test-retest were not clearly outlined and no specific hypotheses were formulated respectively.

Hausa
Three studies were available [7,14,85]. The methodology for the cross-cultural validation was good. A solitary backward translation was done and reconciliation process was poorly described. There was strong evidence for structural validity and IC. There was limited evidence for test-retest reliability and construct validity. It was not clear if administrations were independent, if patients were stable in between administrations and the MSPSS was re-administered after a week against the recommended 2 weeks [57,86]. Further, the handling of missing responses was not reported and no specific hypotheses were formulated.

Korean
One study was available [87]. The methodology for the cross-cultural validation was poor. There was a scanty description of the expertise of the translators and whether the translations were done independently. It was not clear if the tool was pretested in the target population and solitary forward and backward translations were performed. There was poor evidence for structural validity as only EFA was performed. There was strong evidence for IC as the methodology was of excellent quality. There was limited evidence for construct validity, no specific hypotheses were formulated.

Luganda
One study was available [13]. The methodology for the cross-cultural validation was fair. A solitary forward translation was performed; the characteristics of the pretest sample were not clearly described and details of the reconciliation of the original and forward translation

Malay
Three studies were available [11,88,89]. The methodology for the cross-cultural validation was poor. There was a scanty description of; the expertise of the translators, whether the translations were done independently, the reconciliation process, and the tool was not pretested in the target population. There was poor evidence for structural validity as only EFA was performed. There was indeterminate evidence for IC, the handling of missing responses was not documented. There was unknown evidence for construct validity; no specific hypotheses were formulated with poor/no description of the psychometrics of the comparator instruments. There was no report of test-retest reliability; the stability of the respondents was not clearly outlined; the tool was readministered after a week and there was a disparity in administration conditions as the items were reshuffled for the retest. There was unknown evidence for criterion validity, the psychometrics for the purported "gold standard" outcome measure was not provided.

Persian
Three studies were available [18,90,91]. The methodology for the cross-cultural validation was poor. The expertise of translators, handling of missing responses and  reconciliation process was poorly described. Further, solitary forward and backward translations were performed and the tool was not pre-tested. There was poor evidence for structural validity as only EFA was performed. There was limited evidence for IC and reliability as the methodologies were of fair quality. Only 71 participants were recruited for test re-test reliability and the conditions and stability for the re-test sample were not clearly stared. There was unknown evidence for construct validity as no specific hypotheses were formulated and no the psychometrics of the comparator instruments were not provided.

Polish
Four studies were available [92][93][94][95]. The methodology for the cross-cultural validation was good. The reconciliation of the translations was poorly described and the tool was not reviewed by a committee. There was strong evidence for IC and structural validity. There was moderate, negative evidence for construct validity as no specific hypotheses were formulated.

Portuguese
Three studies were available [31,96,97]. The methodology for the cross-cultural validation was fair. The expertise of the translators was not stated, if was not clear if translations were done independently, only solitary forward and backward translations were done and the tool was not reviewed by a committee. There was excellent evidence for both structural validity and IC. There was unknown evidence for test-retest reliability; a sub-optimal sample (n = 52) was utilized, the stability of the participants and the conditions for the re-test were not stated. There was limited evidence for construct validity as no specific hypotheses were formulated.

Spanish
Five studies were available [34,[98][99][100][101]. The methodology for the cross-cultural validation was poor. The expertise of the translators was not stated; only solitary forward and backward translations were done and the tool was not pre-tested. There was conflicting evidence for structural validity as the cited studies were of both poor and fair quality. For instance, for one of the studies, EFA contrary to the CFA reported was done and authors performed varimax (orthogonal) rotation [101]. There was limited evidence for IC as the handling of missing responses was not recorded. There was moderate evidence for construct validity as no specific hypotheses were formulated.

Swedish
Two studies were available [3,102]. The methodology for the cross-cultural validation was poor. A solitary backward translation was performed; the handling of missing responses was not reported and the credentials of the translators were not clearly described. Evidence for structural validity was poor, only EFA was performed and an inappropriate rotation method (orthogonal varimax) was utilized. There was moderate evidence for IC, the handling of missing responses was not described.

Depression -EPDS
EFA exploratory factor analysis, CFA confirmatory factor analysis Evidence for reliability was limited as a sub-optimal sample size (n = 44) was used for the retest and the conditions of the re-test administration were not clearly described. There was limited evidence for construct validity as no specific hypotheses were formulated.

Tamil
Only one study was available [103]. The methodology for the cross-cultural validation poor. There was scanty description of: the expertise of the translators was, if forward translations were done independently, the reconciliation process, whether the tool was not pre-tested in the target population and the profile of the pre-test sample. There was limited evidence for IC, a sub-optimal sample size (N = 94) was recruited and handling of missing responses was not reported. There was unknown evidence for structural validity, construct validity and criterion validity. Only EFA was performed, no specific hypotheses were formulated and the psychometrics of the purported "gold standard" were not provided.

Thai
Three studies were available [28,43,44]. The methodology for the cross-cultural validation was poor. The forward translators did not work independently; only solitary forward and backward translations were performed and scanty details were provided for the reconciliation process and the pre-test sample profile. There was moderate evidence for structural validity and IC. Both EFA and CFA were performed, however, the percentage of missing responses was not stated. There was unknown evidence for construct validity as the no specific hypothesis were formulated. There was limited evidence for test-retest reliability, a suboptimal sample (N = 72) was utilized, the conditions and stability of patients for the re-test were not clearly outlined.

Turkish
Two versions of the Turkish translations were available i.e. the original Turkish version [39,41] and the revised Turkish version [29,104].

Original Turkish version
Four studies were available [39][40][41][42]. The methodology for the cross-cultural validation was poor. A solitary forward translation was performed; it is not clear if the forward translators worked independently and the tool was not pre-tested. There was poor evidence for structural validity as CFA was not performed. There was moderate evidence for IC, the handling of missing responses was not described. There was unknown evidence for construct validity as no specific hypotheses were formulated and one of the studies was of poor methodological quality [41].

Revised Turkish version
Two studies were available [29,104]. There was moderate evidence for structural validity as the handling of missing values was not described. Evidence was; conflicting for IC and unknown for reliability and construct validity. A sub-optimal sample was utilized; the test conditions and stability of the participants was neither described and no specific hypotheses were formulated.

Urdu
Five studies were available [45][46][47][48][49]. The methodology for the cross-cultural validation was poor. The following were not stated; the expertise of the translators, if translations were done independently, the number of forward and backward translations. Further, the tool was not reviewed by a committee and was not pretested in the target population. There was limited evidence for structural validity [47,49]. One of the studies was of poor quality, only EFA was performed [49]. For the second study, although CFA was performed, the handling of missing responses was not documented [47]. There was indeterminate evidence for IC as the methodology was of poor quality. There was moderate evidence for construct validity, no specific hypotheses were formulated.

Settings
The MSPSS has been translated across a range of settings and populations.

Translation quality
Trans-cultural adaptation, translation and validation aim to succinctly capture the meaning of a latent construct in another population. As such, a rigorous translation process is essential [25,52]. None of the studies included in this review were translated using robust methodologies, with 16 of the 22 studies being of poor methodological quality in accordance with the COSMIN criteria [56]. The lack of quality of the translations affects the generalizability and comparability of the study findings. For example, if the MSPSS is applied in a large multinational trial, there is risk of misleading results if one of the translations was poorly conducted. The findings could have negative implications on policy formulation, over−/under estimation of an intervention effect size amongst other. The lack of detailed descriptions of both language and construct expertise of the translators, whether the translations were done independently and reconciliation of the translations compromised the methodological rigor of most of the retrieved studies. Furthermore, the absence of a panel of experts review process for content and face validation, as was the case in 13 of the 21 included studies, could have jeopardised the ability to produce a culturally acceptable translation [25,52]. Ideally, the panel should consist of experts with diverse professional backgrounds to ensure the attainment of semantic, idiomatic, and conceptual equivalence [25]. For instance, given the differences in cultures, the interpretation of the term "special person" can vary from setting to setting. It is argued that respondents from collectivistic cultures may not distinguish between family and a significant other as sources of SS [47,49]. For example, in Turkey, when the term "special person" was changed to "husband" following a panel of experts' review, the resultant/revised translation yielded a three-factor structure as opposed to the earlier two-factor structure [29,104]. This illustrates that a more rigorous reconciliation and adaptation can yield a more reliable factor structure. Lastly, pretesting/cognitive debriefing of the translated and adapted tool is essential before the tool can be applied to a larger population [25,52,56]. This should be done in the target population as translation is an integrated and iterative process and requires input from "experts" and the "target users" of the PROM [52]. Unfortunately, only five of the 23 translations described this process in detail, including description of sample selection, hence this could also be a source of methodological limitation for the retrieved studies.

Structural validity
Structural/factorial validity is defined as the extent to which scores on an outcome measure adequately reflect the dimensions/structure of the construct to be measured [41]. Factorial validity can be envisaged as the 'backbone' for the statistical evidence of the validity or lack thereof of a translated tool. Ideally for translated outcome measures, both EFA and CFA should be performed to test factorial validity [3,31]. EFA is a technique used to explore/discover the number of factors a tool possesses [105][106][107]. The original MSPSS has a three-factor structure, it is essential to test if this is the same for the translated versions as SS is a multidimensional, subjective construct which is dependent on sociocultural contextual factors [1, 3, 10-12, 17, 18, 108]. To this end, it is acceptable to obtain a one-or two-factor structured translation if the translation method is adequately robust. However, EFA alone is inadequate, as was the case in most of the retrieved translations; therefore, CFA ought to also have been performed [57,86]. CFA is an advanced structural equation modelling statistical technique which combines the concepts of EFA, correlation and multiple regression [109,110]. It provides evidence as to whether the translated versions replicated the original three-factor structure as postulated by the developers of the MSPSS. Unfortunately, a minority of the studies [9/23] performed CFA with only four studies [4/23] performing both EFA and CFA which is a major shortcoming for the level of evidence for structural validity. Further, in some instances, some authors/studies refer to EFA as CFA [28,43,44,103] and this again yields inaccurate conclusions. In instances were only EFA was performed, some authors utilized an inappropriate rotational method i.e. orthogonal instead of oblique rotation [3,13,102]. Orthogonal rotation is used when the factors are hypothesized to be unrelated [105,107,111], which is not the case for the MSPSS as the domains are stipulated to be correlated [15,21,22]. For studies which performed CFA, only three adequately described the goodness of fitness (GOF) indices. These are important as they provide concrete evidence to the degree to which the data/translation fits into the original factor model [109,110,112]. Furthermore, given that the MSPSS can yield one-, two-or three-factors, all the three models should be tested using CFA before a decision on the degree of fit can be made. None of the studies which performed both EFA and CFA included this analysis, hence this could be envisaged as a potential source of reporting bias. Replication of the original factorial structure is not necessarily a benchmark for an accurate translation process [25]. For instance, authors may be tempted not to report the results of a two-factor model if the degree of fit is much better than for a three-factor model. Provision of multiple GOF indices for all the three models should be a "standard" reporting practise as it provides the potential readership with all the essential information for them to critique the methodological quality and subsequent conclusions in keeping with the evidence supplied [110].

Reliability
Most of the translated tools displayed adequate evidence for IC as most attained a Cronbach's alpha of at least 0. 70. However, given the limitations in the structural validity testing and lack of rigour in the translation process, the results for IC may need to be interpreted with caution. This is because a tool can be reliable, yet not valid [25,113]. To illustrate this, if only EFA is performed, the factorial validity will be poor, however, the tool can still yield a high alpha statistic. In that instance the reliability findings can be deemed as "misleading" [114]. The validity of the alpha scores is also dependent on the homogeneity/unidimensionality of a test and this can be established through factor analysis [56]. Therefore, if CFA (the preferred unidimensionality test for translated tools) is not performed, the IC for that test will/may not be valid [56,114,115]. Additionally, the longitudinal validity (test-retest) also gives further evidence of the stability of an outcome measure over time [56]. Only four studies reported the stability of the translated versions which is another potential limitation. Given the potential limitation in relying solely on the IC as an indicator of reliability, other indices such as the alternative forms, split-half and test-retest reliability are recommended for concrete evidence of reliability of outcome measures [113,115]. More so, it is argued that despite its wide usage and popularity, the Cronbach alpha is least desirable index for reliability estimation [115].

Construct validity
Construct validity refers to the extent to which scores on an instrument relate to other measures in a manner that is consistent with theoretically derived hypotheses concerning the concepts that are being measured [54]. Depression, anxiety, self-esteem and general mental well-being were the most commonly reported outcomes against which SS scores were compared. To prevent report bias, the developers of the COSMIN checklist recommend that authors should formulate specific hypothesis before data collection [54,57,86]. None of the studies specified the expected magnitude of correlations with only three studies formulating specific hypotheses. Further, there is need for authors to describe in detail the comparator instruments as well as demonstrating their reliability and validity in the study population as failure to do so affects the both internal and external validity [86]. For example, some of the translations [eight out of twenty-three] did not report the psychometrics of the comparator instruments. In other instances, the authors refer to the psychometrics of the comparator from another population which again is questionable [45][46][47][48][49]. Failure to demonstrate the psychometric robustness of the comparator instruments would thus affect the construct validity of the translated versions of the MSPSS.

Criterion validity
Criterion validity is defined as the extent to which scores on an outcome measure perform against an established gold standard [58]. Given that SS is a latent variable [3,11], it is difficult to establish a gold standard against which the MSPSS can be assessed against. Nevertheless, for the three studies which evaluated criterion validity, the psychometrics of the purported gold standard were either questionable or were not well described. Therefore, there was poor evidence for criterion validity.

Limitations
The use of the COSMIN checklist for the evaluation of the methodological quality may have been a potential limitation. This is because the checklist came into effect in 2011 and some of the translations had been performed prior to its' publication. The stringent nature of the checklist has also been reported in almost similar systematic reviews [24,116]. For example, in the assessment of IC and factorial validity, if the handling of missing responses is not reported, the domain(s) are rated as fair quality despite the rest of the ratings being of excellent quality. Inconsistencies within the COSMIN checklist may also be viewed as a potential source of limitation. For example, in evaluating the structural validity of translated tools, if CFA is not performed, item 6 for the structural validity/Box E is rated as good and the same is rated as poor for item 14 under Box G/cross-cultural validity. As the COSMIN guidelines are currently under review, it is hoped the revised guidelines will further harmonize the terminology utilized in the methodological and further increase the checklist validity in rating methodological quality of the translation and adaptation of PROMs. Additionally, we could not evaluate fifteen language versions of the MSPSS which were published in other than English language and this could have introduced language bias for the present review.

Conclusions
We identified 22 translated versions of the MSPSS. The psychometric properties which were most often reported included internal consistency, test-retest reliability, structural validity and construct validity. Many of the tools did not follow a rigorous translation process and there was poor evidence for structural validity. The advent of EBP and increased usage of PROMs requires quality translations to ensure reliable and valid outcome measures. The retrieved MSPSS translations therefore need to be utilized with precautions. There is also need to assess other psychometric properties such as responsiveness, measurement error and establishment of cutoff values to increase the clinical utility and psychometric robustness of the translated versions of the MSPSS. We also recommend the development of a standardized protocol for the translation and adaptation of the MSPSS. Future translation studies should utilize the backward-forward translation method with special emphasis on the use of multiple translators, reconciliation of translations, panel of expert assessment and both EFA and CFA should be performed for factorial analysis.

Acknowledgements
The assistance by Mrs. Morgan from the University of Cape Town library in formulating the search strategy is greatly appreciated.

Disclosure statement
The views expressed in this publication are those of the author(s) and not necessarily those of AAS, NEPAD Agency, Wellcome Trust, or the UK government.

Funding
The systematic review is part of the primary authors' PhD work at the University of Cape Town. The work is being funded by The African Mental Health Research Initiative (AMARI). AMARI is a consortium of four African universities whose overall goal is to build excellence in leadership, training and science amongst African scholars in mental, neurological and substance use (MNS) research in Ethiopia, Malawi, South Africa and Zimbabwe. AMARI, at the University of Zimbabwe College of Health Sciences (UZCHS) secured funding from the Wellcome Trust through the Developing Excellence in Leadership and Science (DELTAS) Africa initiative. The DELTAS Africa Initiative is an independent funding scheme of the African Academy of Sciences (AAS)'s Alliance for Accelerating Excellence in Science in Africa (AESA) and supported by the New Partnership for Africa's Development Planning and Coordinating Agency (NEPAD Agency) with funding from the Wellcome Trust [DEL-15-01] and the UK government.

Availability of data and materials
The data will not be shared since it forms part of ongoing research.
Authors' contributions JMD-conceptualized the systematic review, developed the search strategy, ran database searches and drafted the manuscript. JJ -conceptualized the systematic review and edited the manuscript. MC, LC TMdata extraction and editing of the manuscript. HJ -editing of the manuscript. All authors read and approved the final version of the manuscript.

Ethics approval and consent to participate
This study was conducted as systematic review, so no ethical approval was sought and there was no need for informed consents.