Assessments
We recorded risk factors for atherosclerosis, namely diabetes mellitus, hypertension, smoking, renal failure, hypercholesterolemia, history of coronary artery disease or cerebrovascular disease. In patients with IC, we measured ABI at rest and after exercise. In patients with CLI, we measured ABI at rest and TP.
Quality of life was measured using the VascuQol, which is a sum-score based instrument. The questionnaire consists of 25 items on five domains, i.e. Pain (4 items), Activity (8), Emotional (7), Symptoms (4) and Social (2). Each item is rated as a seven point response scale, with a score of one being the worst and a score of seven the best possible. The total average score is the sum of all 25 items scores divided by 25. For each separate domain an average score can be calculated (sum of all items of one domain divided by the number of items of that domain). So, both the overall score as well as the scores per domain range from one to seven [
16]. The VascuQol has shown to be a reliable and valid instrument for assessment of QoL in patients with PAD [
7,
17].
Disability status was evaluated using the ALDS. For the psychometrical details of IRT in relation to the ALDS, see Additional file
1. The current version of the ALDS itembank consists of 77 items, ranging from very easy (e.g., get out of bed into a chair) to relatively difficult (e.g., walk for more than 15 minutes) [see Additional file
2]. Initially, the ALDS was developed within a dichotomous IRT model with two response options 'I can carry out the activity' and 'I cannot carry out the activity' [
9]. However, the dichotomous rating scales were disliked by some respondents as they are perceived as too restrictive. Therefore, the option 'with difficulty' has been added. Currently, each item has three response options, but the response options 'can carry out' and 'can carry out, but with difficulty' are analysed as one response category. In the case that a patient has never performed the activity or answers that he does not know, 'Not applicable' is recorded. The original units of the ALDS scale are (logistic) regression coefficients, expressed in logits. To make the results easier to interpret these scores are linearly transformed into values between 0 and 100. Lower scores represent more disability.
A major strength of an IRT itembank is that researchers, using their clinical judgment, can make their own selections of items from the itembank that are applicable to the population they are investigating. By using a small number of items tailored to the expected ADL level of patients, a detailed clinical picture can be obtained without the need to have all the questions answered by the patient. Even if different sets of items are used for different patient groups, ALDS scores can still be compared because all items are derived from the calibrated itembank. In this way the ALDS can be used to assess patients with a wide range of conditions and levels of functional status.
The methodology [
9], the psychometrics of the ALDS in terms of dealing with missing data [
18], differences between item measurement characteristics of the itembank in relation to age and sex [
19] and the metric properties of ALDS items in mixed types of patient groups [
11‐
14], as well as the statistical power to detect given effect sizes in clinical trials using IRT outcome scales [
20] have been examined in depth.
From the ALDS itembank, two questionnaires were composed in this study: one questionnaire for claudicants (29 items), and one questionnaire for patients with critical limb ischemia (27 items). Twenty-three items were in common, covering the whole range of the ALDS itembank. Besides these common items, the claudication questionnaire encompassed six additional, relatively more difficult activities, whereas in the critical limb ischemia questionnaire four extra, relatively easier activities were offered. Selecting a representative range of items is essential to prevent floor and ceiling effects. For example, presenting a slightly disabled patient only items between an ALDS of 10 to 50, the maximum achieved ALDS will be 50 (ceiling effect), whereas with items ranging from 0 through 100, the 'real score' (for example 80) can be achieved. Since the ALDS is based on the IRT, the score is not influenced by the selected items [
9]. For the complete ALDS item bank and the selected items in this study, see Additional file
2.
Clinimetric evaluation
The clinical measurement properties of the ALDS were evaluated in terms of internal consistency reliability, construct validity and clinical validity.
Internal consistency reliability refers to the statistical coherence of the scale items. One measure of internal consistency is the Cronbach's α coefficient, which is based on the (weighted) average correlation of items within a scale [
21,
22]. Internal consistency is considered to be good if α ≥ 0.80 [
23]. We also calculated item-total correlations which represent the correlation of a single item with the sum of all other items. Correlations ≥ 0.40 were conservatively considered to be sufficient.
Construct validity concerns whether the new scale corresponds with other instruments measuring the same health concept and instruments measuring different aspects of health. We assumed that in order for the ALDS to be valid, the ALDS scores had to show a decreasing pattern of associations, with the highest correlation with the disability related Activity domain of the VascuQol, intermediate correlations with the VascuQol subscales Symptom, Pain, Emotional and Social, and the lowest with the impairments in terms of ABI and TP [
24,
25].
Clinical validity (also known as known-groups validity) refers to the ability of an instrument to discriminate between patient groups with known differences in clinical status. In this study, clinical validity was investigated by comparing the ALDS between patients with IC and patients with CLI, with ALDS scores to be expected higher in patients with IC than in patients with CLI.
The VascuQol was used as benchmark and therefore the analyses focusing the association between functional health and the vascular parameters and the mean score differences between patients with IC and CLI, were also done for the VascuQol and its Activity domain.
Statistical analysis
Patient characteristics and outcome scores were summarized using descriptive statistics. Distribution of the data was tested with a histogram and the Kolmogorov-Smirnov test. In case of discrepancy between both methods, we regarded the data as not normally distributed. ALDS outcome scores were calculated using a dichotomous IRT model, based on previously published item properties [
11] and algorithms implemented in BILOG-MG (version 3.0) and SPSS version 14.0 (SPSS Inc, Chicago, Illinois). In this approach the response options 'can carry out' and 'can carry out, but with difficulty' are analysed as one response category. ALDS items which were rated 'Not applicable' were statistically considered as if they were not presented to that patient [
18].
Cronbach's α was obtained using a specific IRT method that allows for missing item responses. The average item-total correlation was calculated using a biserial correlation. Associations between the ALDS (and VascuQol) and other outcome measures were expressed in Pearsons's or Spearman's correlation coefficients, when appropriate. We labelled the strength of the association: correlation coefficients r = 0.00-0.19 were regarded as very weak, r = 0.20-0.39 as weak, r = 0.40-0.59 as moderate, r = 0.60-0.79 as strong and r = 0.80-1.00 as very strong [
26]. An unpaired
t-test was used to compare ALDS and VascuQol scores between the two patients groups. Difference in mean scores between both diagnosis groups was expressed in Cohen's
d effect size, defined as the difference between the means divided by the pooled standard deviation. An effect size value between 0.50 and 0.80 is considered as a moderate difference, and ≥ 0.80 as substantial [
27].