Item response theory analysis of the Pain Self-Efficacy Questionnaire

1 Introduction

The Pain Self-Efficacy Questionnaire [1,2] is a 10-item instrument designed to assess the extent to which a person in pain believes s/he is able to accomplish various activities despite their pain. There is evidence for the reliability and validity of the original English-language version, including its ability to predict disability associated with pain, attrition and functional changes after pain management programmes [1]and several translated versions have been used [3,4,5,6,7].

Further evidence for the validity of the PSEQ and its short form may be gleaned from an item response theory (IRT) approach (see [8,9]), in which responses to a set of items are assumed to be probabilistically determined by a latent (unobserved) variable. Whereas classical methods like factor analysis and internal consistency focus on inter-item associations, and therefore are most informative at the questionnaire or sub-scale level, IRT is more informative about properties of individual items. In the case of the PSEQ, an IRT model posits that as the level of self-efficacy (operationalised as some aggregate of all of the items) increases, the response to each item increases along its seven-point response scale. Di Pietro, Catley, McAuley, Parkitny, Maher, Costa, Macedo, Williams and Moseley [10] conducted Rasch analysis, a method mathematically equivalent to a one-parameter IRT model, on the PSEQ using 600 individuals with low back pain. They found that the PSEQ performed adequately against several criteria, including unidimensionality, internal consistency and absence of item bias. A two-parameter model, on the other hand, allows examination of the differences in ability to discriminate between high- and low-scoring individuals.

A further question IRT can address is whether respondents use the PSEQ’s seven-point response scale as intended. Specifically, consistent use of the response scale would be indicated by the observation that as level of self-efficacy increases, the response to each item changes from a lower response category to a higher one, i.e., those very low on self-efficacy (as an aggregate of all items) should be most likely to choose response option 0 (“Not at all confident”) those very high on self-efficacy should be most likely to choose 6 (“Completely confident”) and each of the other five response options should be the most likely for some value of self-efficacy. This relation may not hold if, for example, respondents have difficulty discriminating between consecutive response options, which may occur when there are many such options. Thus, IRT can provide evidence regarding whether respondents use the response categories in the desired manner, and therefore provide a recommendation about whether fewer categories should be used. Di Pietro et al. [10], using Rasch analysis, found evidence for appropriate category ordering in the PSEQ, but presented average curves rather than curves for individual items, finding a disordered item threshold involving the second response category. Whereas the Rasch model assumes that the extent to which items can discriminate between respondents high and low on the level of the latent variable is constant, there are less constrained IRT models that allow discriminative ability to vary, so it would be instructive to examine whether such a model supports the use of seven response categories.

Nicholas, McGuire and Asghari [11] proposed a short form for the PSEQ, consisting of two items: ‘I can do some form of work, despite the pain (“work” includes housework and paid and unpaid work)’ and ‘I can live a normal lifestyle, despite the pain’. These items were selected for the short form on the basis of itemtotal correlations, item-disability correlations (using the modified Roland Morris Disability Questionnaire [12]), responsiveness, contribution to total score variance, and construct validity. IRT can provide additional information regarding location and discrimination parameters that can further inform short form development.

The overarching purpose of the present study is re-analyse Nicholas et al.’s [11] data using a two-parameter IRT model, which supplements validation research already conducted. This study extends the work of Di Pietro et al. [10] by examining the threshold and discrimination parameters for each PSEQ item. We also addressed Di Pietro et al.’s [10] call to examine the properties of the PSEQ for patients in tertiary care and with other pain sites, not just back pain.

2 Materials and methods

3 Results

The mean total PSEQ score was 25.83 (SD = 13.96). The fit of the two-parameter model was significantly better than the fit of a one-parameter model (p < .01), indicating that the discrimination parameters differed between items (i.e., model fit was relatively poor when they were constrained to be equal). The threshold and discrimination parameters for the 10 items are shown in Table 2, as well their means and standard deviations. Item 9 had the highest discrimination parameter, followed by Items 8 and 6. Item 7 had the lowest discrimination parameter, which most likely arises because of the low rate of endorsement for this item (45% of respondents answered “Not at all confident”).

The category probability curves for Items 1 and 7 are shown in Fig.1.All items except item 7 exhibited item response thresholds in the expected order (the curves for Item 1 are included as a representative example of the other nine items). The item information functions (Fig. 2) show that items 6, 8, 9 and 10 were most precise for slightly higher levels of the latent variables than were items 1–5, as indicated by the location of the peaks of the curves in the horizontal direction. This aligns with the observation from Table 2 that the location parameters for Items 1–5 tended to be lower than those for Items 6–10, i.e., Items 1–5 tend to be more precise for lower levels of self-efficacy, and Items 6–10 more precise for higher levels. Item 7 had low precision for all values of the latent variable. The test information function, which is the sum of the item information functions, is shown in Fig. 3.

Fig. 1

Category probability curves for Items 1 and 7 of the PSEQ. The numbers on each graph represent the response options (0 = “Not at all confident”, 6 = “Completely confident”). The horizontal axis represents the value of the pain self-efficacy latent variable, and the vertical axis represents the probability of selecting each of the seven response options. (a) Item 1 (appropriate threshold ordering, representative of all items except Item 7). (b) Item 7

Fig. 2

Item information functions for each PSEQ item, which represents the precision of each item across the range of the latent variable. The horizontal axis represents the value of the pain self-efficacy latent variable. For clarity of presentation, only items with the highest (items 4, 6, 8 and 9) and lowest (items 2 and 7) information are labelled.

Fig. 3

Test information function, which represents the sum of the item information functions. The horizontal axis represents the value of the pain self-efficacy latent variable.

4 Discussion

The results of the IRT analysis conducted here confirm the strong psychometric properties of the PSEQ, adding to the body of evidence supporting its validity and reliability [1,3,10]. The category probability curves provided empirical support for the use of the seven-point response scale, with the exception of Item 7, whose poor performance is consistent with all previous studies examining the validation of the PSEQ [1,10,11]. The large proportion of low responses to this item results in relatively weak relations with the other items.Nonetheless, the item captures clinically important information [1,10], which must be considered alongside statistical considerations in determining its utility. Specifically, the observed low mean item score indicates a large number of respondents are not confident they can cope with their pain without medication. This is important information for clinicians attempting to help their patients to manage their pain without relying on opioids, for example [15]. It is also an important metric for evaluating the outcomes of interventions aimed at enhancing pain selfmanagement. For these reasons we believe the item should not be removed.

It is noteworthy that the mean PSEQ score in this sample (25.83) was much lower than those reported in Di Pietro et al.[10] (range 39.8–44.6). The participants in this study were recruited from a tertiary pain management centre, and their pain was not limited to back pain. Thus, the present results illustrate the psychometric properties of the PSEQ for a sample that is both more impaired (with respect to self-efficacy) and more heterogeneous than in Di Pietro et al. [10], highlighting the wide applicability of the PSEQ.

With respect to the two items included in Nicholas et al.’s [11] short form, Item 9 was the best discriminating item and had greatest precision for higher values of the latent trait. Item 5 was not among the best-discriminating items, but it exhibited no problems. Notably, the precision of Item 9 was greatest for higher levels of self-efficacy, whereas the precision of Item 5 was greatest for lower levels; thus, these two items between capture a range of levels of self-efficacy. Also worth noting is that the four best discriminating items were those used in a 4-item short form proposed recently [16]. In combination with the analyses reported by Nicholas et al. [11], which complement the current analysis, we suggest that no changes are required to the 2 item PSEQ short form.

Item response theory has been applied in a wide variety of research areas. It provides information about the measurement properties of an instrument that complements that of more traditional methods, e.g., factor analysis, internal consistency, and is particularly useful in developing short-form instruments. Thus, its application in pain research has great potential to assist in reducing the burden on patients, particularly when patients are administered a battery of questionnaires.A limitation of this study is that participants were drawn from a single pain clinic. Nonetheless, the sample was large and heterogeneous with respect to origin and site of pain, as well as various demographic characteristics, thus supporting the generalisability of the results. In summary, this study adds to the growing body of evidence supporting the psychometric properties of the PSEQ as a tool for assessing self-efficacy in people with pain.

Highlights

• The Pain Self-Efficacy Questionnaire(PSEQ)has strong psychometric properties.

• Most PSEQ items perform well in item response theory analysis.

• Item 7(coping without medication)performed poorly but has clinical utility.

• The PSEQ is a useful tool for assessing self-efficacy in people with pain.