Page number 34
of stroke and geriatric patients as well as for patients with various neurological diagnosis (14-16, 19, 20). Those studies have indicated that the SADL has good reliability, whereas the findings regarding the scale’s validity were less con- sistent (15, 19-22). The latter spe- cifically concerns whether each item on the SADL assesses the same underlying construct and the functioning of the intervals of the scoring scale (14, 15). Hence, there is still a need to investiga- te further the internal construct validity of the SADL scale for the group of stroke patients. Rasch measurement model analysis is the recommended method of evaluating the internal validity and reliability of the item scale for criterion-referenced outcome measures used in reha- bilitation (23). Criterion-referen- ced outcome measures evaluate the patient’s performance against pre-specified criteria describing different levels of efficient or independent performance (24). The resulting raw scores are ordinal and do not exhibit basic features of measurement, such as unidimensionality, hierarchical order and equal interval scaling. The use of Rasch measurement model analysis transforms ordinal data into interval equal level data, enabling the use of parametric statistical analyses. In addition, the internal scale validity and reliability of the outcome measure can be established, for example, whether the construct measured by the test items (for example independence in ADLs) remains stable over the range of person abilities in the population of inte- rest (25). Thus far, only one study has used Rasch analysis to inves- tigate the construct validity of the SADL (14). That study, inclu- Characteristics Participants (n (%)) 200 (100) Gender (n (%)) Men 136 (68) Women 64 (32) Age in years (Mean (±SD)) 58 (12) Men 56 (12.6) Women 59 (11.7) Diagnosis (n (%)) Stroke 188 (94) Others* 12 (6) T.S.O. (Median (min-max)) 33 (6-3195) *Brain tumour, encephalitis, other neurological diagnosis, n= Number of participants, SD=Standard Deviation; min= minimum, max= maximum, T.S.O. = Time since onset given in days Table I. Characteristics of included persons with stroke and other acquired brain injuries. ding a heterogeneous sample of geriatric patients, indicated that several items did not measure the same construct as the other items (14). However, methodological weaknesses, including missing descriptions of statistical analyses and the resulting estimates, make it difficult to draw firm conclusi- ons from the study. Furthermore, the results for a geriatric patient group may not apply to the po- pulation of patients with stroke, for whom the SADL was originally developed. The aim of our study was th- erefore to examine the construct validity of the SADL for patients with stroke by exploring its inter- nal scale validity and aspects of its reliability (internal consistency, Rasch analysis based item- and person reliability coefficients and person separation ratios). Methods STUDY DESIGN AND SAMPLE We used convenience sampling in our cross-sectional study. The inclusion criteria were being a patient of at least 18 years old with stroke or similar motor and cognitive impairments following an acquired brain injury. The SADL was administered to 200 patients admitted to the stroke unit at Sunnaas Rehabilitation Hospital in Norway from 2012 to 2017 (see Table I). The stroke unit is a secondary care unit with pati- ents that are referred for complex rehabilitation from primary hospitals in the South-East Health Region of Norway. The average age and length of stay for pati- ents admitted to the stroke unit at Sunnaas Rehabilitation Hospital in 2017 was 54,8 years and 46.5 days (26). The number of persons included in the study was chosen according to recommendations from Linacre (27) in order to obtain precise, robust measure- ments. THE SUNNAAS ADL INDEX (SADL) The SADL was developed in 1985 by Norwegian occupational the- 34 Ergoterapeuten 5–2023
Page number 35
rapists at Sunnaas Rehabilitation Hospital (10). The assessment tool was intended to provide an overview over the areas of ADL’s relevant to consider the patient’s independence. It contains twelve ADL items: 1 eating 2 continence 3 indoor mobility 4 toilet management 5 transfer 6 dressing and undressing 7 grooming 8 cooking 9 bathing and/or showering 10 housework 11 outdoor mobility 12 communication. Each item has four ordinal scoring categories ranging from 0 to 3 points, for a maximum sum score of 36 points. Scores from 0 to 1 indicate total or partial dependen- ce on assistance, whereas scores of 2 to 3 indicate independence with or without adaptation or assistive devices. The items are structured into three hierarchical groups based on frequency of activity performance and degree of assistance needed. Items 1 to 4 (eating, continence, indoor mobi- lity and toilet management) refer to activities performed several times daily that cannot be plan- ned to occur at a specific time. By contrast, items 5 to 8 (transfer, dressing and undressing, groo- ming and cooking) refer to activi- ties also performed several times daily, but that can be arranged to occur at a specific time. Items 9 to 11 (bathing and/or showering, housework and outdoor mobility) refer to activities that can be plan- ned, but occur only once or twice a week. Last, item 12, communica- tion, is not included in any group due to its frequency. The SADL should be administered and sco- red according to specific criteria outlined in the manual (10). DATA COLLECTION The SADL was administered as part of regular clinical evaluations by occupational therapists or oc- cupational therapy students who were trained in using the SADL. The assessment was completed as an interview with the patient and/or relatives and supplemen- ted by clinical observations when considered to be necessary. ETHICS The study was approved as a Quality Improvement Project by the Data Protection Officer at Oslo University Hospital, Norway. The approved aim of the study was to investigate the measu- rement properties of the SADL, not to investigate characteristics of the individuals in the group or to produce new knowledge about people or disease. Thus, the study was not required to be evaluated by the Regional Committee for Medical and Health Research Ethics in Norway, and consent was not obtained from the participants (28). Data was de-identified with the link key stored separately. Thereafter, the data was handled and analysed pseudonymously. DATA ANALYSIS The Rasch measurement model was used to evaluate the SADL’s internal scale validity and aspe- cts of item and person reliability using Winsteps version 3.71.0.1. Six areas were investigated in an iterative analytical process: (i) functioning of the rating scale (ii) unidimensionality (iii) targeting of item difficulty to person ability (iv) item and person reliability (v) item invariance (vi) hierarchical structure The rating scale model and the functioning of the rating scale First, rating scale functioning was investigated to determine which derivation of the Rasch polyto- mous model to use for further analysis (29). For the rating scale to function well, the threshold values should increase by >1.4 logits between each category. A minimum of ten responses for each category is recommended for the scale to function as ex pected (30). Unidimensionality Unidimensionality was investi- gated by principal components analysis (PCA) and item and person goodness-of-fit statistics. The PCA may indicate existence of a secondary dimension if the eigenvalue in the first contrast amounts to more than two (31). We also investigated potential multidimensionality by exploring if there was clustering between groups of items. This is more important than whether the loa- dings exceed certain values (31). Local independence of the items was also explored. Since local dependency may inflate reliability indices, analyses were repeated with testlets (correlated items pooled together) if the standardi- zed residual correlation between two items exceeded 0.3 (32). Goodness-of-fit statistics indi- cate how well the items fit the un- derlying construct and how close to the expected value the persons perform (30). Since misfitting infit statistics pose a greater threat to test validity than misfitting outfit statistics, criterion for an accep- table infit mean square (MnSq) Ergoterapeuten 5–2023 35
