This is the first study that has examined the reliability and validity of the German translation of the DEMMI used by physiotherapists in a population of patients admitted to a sub-acute geriatric inpatient rehabilitation hospital. To review the success of cross-cultural validation, it is crucial to compare the results found in this study to the results of other studies on the DEMMI’s psychometrics.
Comparison with other studies
Two studies examined the DEMMI’s psychometrics in Australian older people receiving inpatient rehabilitation [
25,
53]. Mean age (81.8 and 83.4 years, respectively) and proportion of female participants (57.1% and 76%, respectively) was comparable to our sample. However, MMSE (mean 24.0 points) and CCI (mean 1.3) scores were lower in the one study these variables were assessed [
53]. Most participants in our study were assessed in the middle or at the end of their rehabilitation. When this data is compared to discharge values reported in the other studies, DEMMI scores (mean: 41 and 49 points, respectively [
25,
53]) of participants included in this study (mean: 53 points) are even higher, but walking abilities (6MWT: 155 m, gait speed: 33 m/min (approximately 0.54 m/sec) [
53]) are comparable. We thus acknowledge the limited comparability between these samples.
Inter-rater reliability between two trained physiotherapists was excellent (ICC = 0.94), with good agreement in most of the 15 DEMMI items. This result shows high accordance with other reliability studies that found comparable reliability indexes in acute medical patients (r = 0.92) [
24] and in a sub-acute geriatric rehabilitation setting (r = 0.87) [
25].
The absolute reliability (MDC
90 of 9 points) indicates that there must be a change score of at least 9 points for an assessor to be 90% confident that a true change has occurred. This value is similar to the values reported in other trials, where the MDC
90 was between 8 and 10 points in acute and sub-acute geriatric inpatients [
24,
25].
Convergent construct validity was indicated by confirmation of the hypotheses of strong correlations between DEMMI scores, as a measure of mobility, and scores of other mobility related outcome measures. The point correlation between DEMMI scores and gait speed (rho = 0.67) was only slightly lower than expected. These findings are congruent with other trials, where correlations with the 6MWT and gait speed were quite similar in the sub-acute setting [
25] or in older patients with hip and knee ostreoarthritis [
27].
Known-groups validity was evident with respect to dependence in ambulation and walking aid use. The fit of the data for 14 items in the German DEMMI version to a Rasch model confirmed that it is a unidimensional scale. There were, however, some differences in the average location of several items on the logit scale. In the current study the easiest of the 14 analysed items were 3 (lie to sit) and 5 (sit to stand) compared to 1 (bridge) and 7 (stand unsupported) in the comparison study. The most difficult items were 9 (stand on toes) and 10 (tandem stand eyes closed), a result similar to both the Australian [
20] and Dutch [
27] samples.
Strengths and limitations of study
The sample size of the reliability study was as large as calculated
a priori and it is comparable to previous examinations on the DEMMI’s reliability [
20,
25,
27]. The 95% CI was narrower than expected (0.88 to 0.97) and the lower limit of the 95% CI of the ICC (0.88) is higher than the recommended minimum standard of reliability (ICC ≥ 0.70) [
50]. However, inter-rater reliability was only assessed between 2 trained raters with a quite similar level of work experience. One can assume that the inter-rater reliability in a larger sample of raters in clinical practice would be comparable if the same learning procedure is followed. However, reliability studies between more diverse raters are desirable and would provide reliability estimations with higher external validity.
The calculation of the MDC followed the approach described by Stratford et al. [
48]. Therefore, data of stable patients is needed to detect measurement error over time (longitudinal approach) [
47,
48]. The MDC calculation performed in this study includes the ICC between 2 assessors found in this study. As both assessments were performed during a short period of time (30 minutes), the ICC includes the inter-rater variance (between both raters) and the participant’s intra-individual variance (test-retest between both time points). Thus, the MDC of 9 points of the German DEMMI version in the sub-acute geriatric setting might be biased by the short period between both assessments and the inter-rater variance included in the ICC value. Further research should use reliability data of stable patients assessed over a longer period relevant for the inpatient rehabilitation setting (e.g. 2 to 3 weeks) by one single rater to further prove the MDC found in this study.
For the convergent validity analysis, we used only one assessment (POMA) that contained multiple components of mobility. The others (FAC, 6MWT, gait speed) are actually measures of ambulation and gait, and do not rate bed mobility, transfer abilities and higher levels of functional ambulation. The FES-I is only an indirect perception of mobility as it rates fear of falling.
Divergent validity based on DEMMI and MMSE correlations could not be analysed as intended in the study protocol because the included participants did not cover the potential width in MMSE scores. However, the DEMMIs discriminant validity was indicated by a non-significant and low correlation with the CCI and was further proven in another study reporting a weak correlation (0.24) with the MMSE [
20].
The DEMMI was developed in a consecutive acute medical sample [
20]. As we used a cross-sectional design, most sub-acute participants were already an inpatient for some time (13 ± 6 days) and thus presented with a higher mobility level than on admission. That is why we abstained from the calculation of floor and ceiling effects. In the convenient validation sample, assessments were performed by only 3 physiotherapists differing in their level of working experience. This may not entirely represent the real life clinic and its variety of raters and patients, and thus, evidence for the psychometric characteristics of the DEMMI in daily clinical practice is limited. However, in the first study on cross-cultural adaption [
28], the German DEMMI version was performed by a complete clinical section of physiotherapists in 133 consecutive geriatric inpatients, and by doing so no floor- or ceiling effects were observed. Thus, there is evidence for the DEMMI to overcome crucial issues with floor and ceiling effects [
20,
24,
27,
28] that hamper clinical interpretability of most common mobility measurement instruments reported in other studies [
16-
18].
Rasch analysis could only be performed on a 14-item DEMMI as item 4 (sit unsupported) could be performed by all 140 participants. This can be explained by the higher functional abilities of the present convenient sample due to the later recruitment in this study, in contrast to the Australian development sample [
20]. For that reason we kept item 4 (sit unsupported) in the German DEMMI version. Furthermore, this item gives clinically relevant information on the functioning of an elderly patient and it was also the easiest item in the Australian English and the Dutch DEMMI versions [
20,
27]. However, a further study should include a consecutive sample of sub-acute geriatric inpatients who perform the DEMMI immediately after hospital admission.
When it comes to clinical implementation of the DEMMI in a team of health care professions, a short learning phase seems essential to gain such reliable scores between raters. We recommend the approach described in this study, to use the German instruction handbook and to pay attention especially to these items that showed lower agreement in the present study (see Table
2, eg. “lying to sitting”). Cognition can have a significant impact on mobility [
56,
57]. In the current study, patients with low MMSE scores were excluded. However, the high proportion of cognitively impaired patients in geriatric inpatient settings is clinically very important [
58]. These patients were not specially excluded in the general geriatric acute medical DEMMI development sample (mean MMSE: 21.7 ± 7.6, range 0–30) [
20]. Through the successful cross-cultural validation process, the German DEMMI seems to be a valid measure of mobility for patients presenting with various cognitive abilities. However, the DEMMI’s psychometric properties solely in the considerable population of older individuals with cognitive impairment need to be further examined.
The DEMMI is considered to measure a patient’s mobility in various settings and with various disease conditions where mobility functions of older people are an important indicator of independence in the ADLs, quality of life and health status. Thus, more research is needed on the German DEMMI version in acute [
20] and community-dwelling [
59] older people, as well as in nursing home residents.
Mobility can also be affected crucially in several geriatric diseases and syndromes, such as osteoarthritis [
27], Parkinson’s disease [
26], hip-fracture [
53], stroke, frailty, dementia or chronic obstructive pulmonary disease. We recommend further psychometric examination in these conditions, including analysis of responsiveness, interpretability and prognostic validity of the German DEMMI translation.