Structural validity of the Dutch version of the disability of arm, shoulder and hand questionnaire (DASH-DLV) in adult patients with hand and wrist injuries

Hand and wrist injuries are commonly seen in adults [1‐4]. About 20% of all visits to the emergency departments are due to hand and wrist injuries [5, 6]. Considering the ageing of the population, the incidence for these injuries is going to grow [7, 8].

The prevalence of chronic pain following distal radius fractures is reported to be as high as 30%. Of these patients, 11% report moderate to very severe pain 1 year after the initial injury [9, 10]. Longterm disability largely affects elderly patients, of whom 46–95% report some degree of disability 1 year following the initial accident, and 7–16% even report moderate to very severe disability [9, 10]. Aforementioned complaints may result in patients’ inability to perform daily activities.

The International Classification of Functioning, Disability and Health, the ICF, provides a standard language and framework for the description of functioning and disability [11]. In the ICF, functioning problems are classified in three areas: Impairments, Activity limitations and Participation restrictions. The broad concept of disability can refer to any or all areas of functioning in the ICF. Patient reported outcomes (PROs) are one of the most common techniques to assess the different facets of functioning. These outcomes are reported by patients and not defined by an observer [12]. They may be used in clinical decisionmaking, as well as in health care policies and reimbursement decisions [13, 14]. To ensure a PRO can be used in clinical practice for these abovementioned functions, they have to be validated. [14]

Recently, recommendations for a core set of domains for standardized reporting in distal radius fracturs have been published. [15] Pain and function were considered as primary domains.

In every day practice, mostly traditional outcome measures are used to determine results of treatment. For hand and wrist injuries these include physical examining, range of motion, grip strength and radiographic imaging. These examinations mainly reflect aspects of disability in bodily functions. However, the traditional outcome measures are “clinician based” and do not correlate well with aspects that patients find important, such as activity limitations [16]. Therefore, PROs are increasingly used to evaluate the result of treatment and rehabilitation, also in patients with hand and wrist injuries.

The American Academy of Orthopedic Surgeons, the Council of Musculoskeletal Specialty Societies and the Institute for Work and Health developed a questionnaire which reflects the impact of injury on function of a variety of upper extremity musculoskeletal disorders or injuries and developed the Disabilities of the Arm, Shoulder and Hand, questionnaire (the DASH) [17]. The DASH is a 30-item, self-report questionnaire to measure physical function and symptoms in people with musculoskeletal disorders of the upper limb [17]. The questionnaire consists of 3 subscales: a physical subscale, a symptoms subscale and the psychosocial subscale. The DASH has been translated and adapted into several languages [18‐32].

In literature exploratory factor analyses (EFA) have been conducted by several authors in different languages to examine the underlying factors of the DASH questionnaire [22, 23, 33]. EFA is a data-driven method without making specifications about the number of and relationships between the latent factors. This approach is used as an exploratory technique. In contrast, confirmatory factor analysis (CFA) requires strong empirical or conceptual grounds to guide the specification and evaluation of the structure of the model in advance [34]. To date, only two studies reported on CFA of the DASH, which were performed on the Italian and American version of the DASH [35, 36].

In this study, the structural validity of the existing translated Dutch version of the DASH (DASH-DLV) was investigated in a patient population with hand and wrist injuries [37]. Particularly, a CFA was conducted, followed by an assessment of internal consistency. Because Veehof et al. already translated the DASH into a Dutch version, we chose not to translate the DASH again [33].

As described previously, a cohort of 466 patients who sustained isolated hand or wrist injury were eligible, of which 370 (79.4%) patients (188 males and 182 females) participated in the current study, with a mean age of 43.6 (SD = 14.2) years [38]. The majority of the hand and wrist injuries (82%) were treated conservatively. A large proportion of the injuries were fractures, mainly of distal radius (130/334) (Table 1). The follow-up time ranged from 1 to 25 months. The DASH-DLV questionnaire was completely filled in by 329 of the responders (88.9%). Sixteen patients (4.3%) had a missing response on the item “sexual activities”. The rest of the items were missing in less than 3%. Total scores could be calculated according to the DASH manual for all patients [40].

The 3 CFA models with corresponding fit statistics are presented in Table 2, the standardized factor loadings are presented in Table 3. Although associated with a significant χ² goodness-of-fit (584.83, df = 405, p < 0.001) adequate levels of absolute (RMSEA and SRMR) and incremental fit indexes (CFI and TLI) were calculated for Fig. 1. All factor loadings for this model were higher than 0.70 (Table 3).

Figure 2 also yielded a significant χ² goodness-of-fit value (498.12, df = 402, p = 0.001), but satisfactory absolute and incremental fit indexes. In Fig. 2, all items loaded high on one of the three correlated subscale factors Physical, Symptoms and Psychosocial. The factor loadings ranged from 0.75 to 0.95. Only 5 and 4 items loaded on subscale factors Symptoms and Psychosocial, respectively. The covariance between the correlated factors was positive and statistically significant (Physical versus Symptoms = 0.89, Physical versus Psychosocial = 0.94, Symptoms versus Psychosocial = 0.92, all p-values< 0.001).

The bifactor Fig. 3 was associated with good levels of model fit indexes: χ² value of 419.96 (df = 375, p = 0.054), RMSEA = 0.018, SRMR = 0.041, CFI = 0.998, TLI = 0.998.

However, in Fig. 3, many items loaded high (ranging from 0.69 to 0.93) on the general factor, but low on the subscale factors (Table 3). As an example, the correlated model (Fig. 2) suggests that item 8 ‘Garden or do yard work’ was a strong indicator of the Physical subscale (i.e. a factor loading of 0.93). In contrast, Fig. 3 (the bifactor model) indicated that item 8 was a weak indicator (i.e. a factor loading of 0.06).

The ECV is 0.92 in Fig. 3. The factor strength indexes are also presented in Table 3. The coefficient ω_H was high for the general factor (0.96), but ω_S was low for the group factors (Phyical, Symptoms and Psychosocial; which were 0.01, 0.26 and 0.11 respectively). These results indicate that a large portion of the total variance is explained by the general factor, and only a very small portion of the total variance is explained by subscale factors. Regarding internal consistency, Crohnbach’s α of the single and the 3-correlated factor models (Figs. 1 and 2) were high, ranging from 0.88 to 0.97. These findings suggest that the DASH-DLV measures a single factor model and that it is not beneficial to report subscale scores.

The various CFA models were used to clarify how the items of the DASH-DLV relate to each other, and to explore if there were any subscale scores that should be used when scoring the questionnaire. This study suggests that the DASH-DLV reflects a unidimensional trait, and thus reporting subscale scores in the Dutch translation of the DASH is of very limited value and should be avoided.

The Upper Extremity Collaboration Group used principle component analysis to determine the dimensionality of the DASH. Although a two-factor model explained more variance and the scree plot suggested two factors, a one-factor model is recommended given its simplicity [40].

While principal component analysis aims to explain all variance in the data set, making it most appropriately applied as a data reduction technique, EFA is used to only explain the common variance of all items, discovering a set of yet unknown latent variables based on the data. In contrast, confirmatory factor analysis makes it possible to test whether the data fit a prehypothesized factor structure based on empirical data or theory, making this technique more appropriate to confirm the factor structure (i.e. dimensionality) of a questionnaire. The choice for a particular method of factor analysis is crucial, because the different techniques have different assumptions about the data and answer different research questions [58].

In this study, we used CFA since our reseach question was to confirm the factor structure of the DASH-DLV. To our knowledge, only two studies have conducted CFA to examine the DASH questionnaire [35, 36]. Franchignoni et al. investigated the factor structure of the Italian version of the DASH [35]. After an exploratory approach, the 3-factor structure showed adequate fit, nonetheless with some misfitting items. A 1-factor model of the DASH was not confirmed as indicated by poor fit statistics.

In the American version, Lehman et al. also tested a 3-factor model after excluding item 20 and 21 because of their unacceptably low factor loadings [36]. Although the TLI and SRMR values indicated good fit, the CFI and RMSEA do not. In addition, they found high interfactor correlations (> 0.83).

All models in our study yielded adequate fit to the data (Table 2). Both Fig. 1 (one-factor) and Fig. 2 (3-correlated factors) showed high and statistically significant factor loadings. However, the subscales Symptoms and Psychosocial of Fig. 2 included only 3 and 2 items, respectively, potentially compromising the coverage of the construct’s theoretical domain. All items in the bifactor model (Fig. 3) were associated with high factor loadings on the general factor, but low on the group factors. Bifactor analysis allows researchers to empirically examine the appropriateness of using subscales. To date, research in assessing the structural validity of DASH has not included bifactor models.

Several important findings support that the DASH-DLV is sufficiently unidimensional. First, the covariance between the 3 correlated factors in Fig. 2 were all positive and significant, indicating unidimensionality. Second, the factor loadings of the general factor in the bifactor model (Fig. 3) are very similar to the loadings in the single factor model (Fig. 1). Furthermore, the factor loadings are high and statistically significant on the general factor, but substantially lower on the group factors. This suggests that the subscale factor contribution ‘over and above’ the general factor is very limited. [46] Third, the general factor of Fig. 3 accounted for more than 90% (ECV = 0.92) of the common variance, indicating a high degree of unidimensionally. Finally, although the coefficient omega total values estimated in the bifactor model showed very good reliability for the general and subscale factors, the values of omega hierarchical of the general factor differed significantly from the omega subscale of the subscale factors. Omega hierarchical (ω_H) coefficient gives the proportion of variance in scale scores accounted for by a general factor, whereas the omega subscale represents the reliability estimate of the subscales, accounting for the effects of the reliability due to the general factor in bifactor models [51, 59]. The coefficient omega hierarchical therefore provides useful information on whether scores for subscale factors can be interpreted with confidence, or that only the general factor score should be used. In this study, ω_S was very low for the subscale factors (ranging 0.01–0.26), but ω_H was high (0.96) for the general factor. This indicates that the subscale factors account for only 1 to 26%, while the general factor accounts for 96% of the variance. This implies that reporting subscale scores in the DASH-DLV is of extremely limited value.

This study has some limitations. The patients who were included mainly experienced distal radius fractures, and were mostly treated non-surgically. This distribution of patients may limit the generalizability of the results. For this study, we only included trauma cases and no elective cases. This may have caused a selection bias towards elderly females. In addition, an existing Dutch translation of the DASH questionnaire was used without employing a translation and culturally adaptation process. However, this Dutch version is widely used and supported by the Institute for Work & Health [37]. Despite these limitations, the response rate was sufficiently high and an adequate sample size was included. There was only a small number of missing values, from which total scores for all patients could still be calculated according to the DASH manual. [40] Finally, future studies should assess validity in more detail, and other measurement properties of the DASH, such as test-retest reliability and responsiveness, should be evaluated.

In conclusion, this study suggests that the DASH-DLV reflects a unidimensional trait, and thus reporting subscale scores in the Dutch translation of the DASH is of very limited value and should be avoided. Further studies should assess the validity of the DASH-DLV in more detail, as well as other measurement properties, such as test-retest, reliability, measurement error and responsiveness, to ensure reliable interpretation of this patient reported outcome measure in clinical practice.