The Quality of Life (QoL) measurement has been developing as a formal discipline with systematized theoretical bases and specific methodology for more than 30 years (Scientific Advisory Committee of the Medical Outcomes Trust, 2002). QoL has become progressively recognized as an important outcome measure in treatment studies and health service evaluation (Tiffany, Friedman, Greenfield, Hasin, & Jackson, Reference Tiffany, Friedman, Greenfield, Hasin and Jackson2012). The assessment of QoL with the objective of evaluating interventions, adverse effects of treatment and the impact of the disease process is widely used in clinical trials and in observational studies of health (Globe, Reference Globe1999). Notwithstanding, QoL and health status are different constructs, and therefore, instruments designed to assess health status may be inappropriate for assessing QoL. The concept of Health-Related QoL (HRQoL) was created to distinguish outcomes generated by health-based research from earlier sociological enquiries on well-being and life satisfaction in the healthy general population (Smith & Larson, Reference Smith and Larson2003).

In the field of drug and alcohol abuse, HRQoL has been used as an important clinical and research outcome. It was initially used as an outcome measure in methadone maintenance treatment (MMT; Reno & Aiken, Reference Reno and Aiken1993). HRQoL started to be used as an outcome measure also to evaluate “hard” indicators, including criminal behaviors, unemployment, and illicit drug use (Iraurgi, Reference Iraurgi1999). The concept of HRQoL is used in drug abuse research and clinical practice, including the assessment of the impact of therapeutic interventions on patients’ lives (Raisch et al., Reference Raisch, Campbell, Garnand, Jones, Sather, Naik and Ling2012; Wang et al., Reference Wang, Wu, Yen, Yeh, Chung, Chang and Yen2012).

In the field of substance use, the HRQoL has been measured mainly via generic tests. These tests evaluate representative behaviors from daily activities and most frequent symptoms in normal persons. The most commonly used generic tests in drug abuse research, including the Short Form 36-item (SF-36) and 12-item (SF-12) health surveys (Ware, Reference Ware1997; Ware, Kosinski, & Keller, Reference Ware, Kosinski and Keller1996), the World Health Organization Quality of Life (WHOQOL-Bref) assessment instrument (The WHOQOL Group, Reference Orley and Kuyken1994) and the Nottingham Health Profile (NHP; Hunt, McEwen, & McKenna, Reference Hunt, McEwen and McKenna1985), have been widely used in epidemiological research.

Some authors suggest that one of the main advantages of a generic test is its versatility when applied to different populations (Iraurgi, Reference Iraurgi2008). Conversely, these tests may not be sensitive enough to detect clinically relevant changes in patients with specific health disorders. Therefore, it has been advocated that, whenever the main goal is to evaluate the impact of of therapeutic interventions, HRQoL measurement tools should be used.

In recent years, some specific tests have been developed in drug abuse field, including the Injection Drug Users Quality of Life Scale (IDQUOL; Brogly, Mercier, Bruneau, Palepu, & Franco, Reference Brogly, Mercier, Bruneau, Palepu and Franco2003); the Quality of Life Scale for Drug Addicts (QOL-DA;Wan et al., Reference Wan, Fang, Jiang, Shen, Jiang, Tu and Wan2011); and the Health-Related Quality of Life for Drug Abusers (HRQoLDA) test (Lozano, Rojas, & Pérez, Reference Lozano Rojas, Rojas Tejada and Pérez Meléndez2009; Lozano et al., Reference Lozano, Domingo-Salvany, Martínez-Alonso, Brugal, Alonso and de la Fuente2008; Zubaran et al., Reference Zubaran, Sud, Emerson, Zolfaghari, Foresti and Lozano2012). The IDUQOL was conceptualized using the World Health Organization’s (WHO) definition of quality of Life. Initially was developed for injection drug user, but later version have been developed for non-injectors (Morales-Manrique, Valderrama-Zurián, Castellano-Gómez, Aleixandre-Benavent, & Palepu, Reference Morales-Manrique, Valderrama-Zurian, Castellano-Gómez, Aleixandre-Benavent, Palepu and Cocaine Group2007). QOL-DA was designed also using the WHO’s QOL concept. It consists of six domains of physical, psychological, level of independence, social relationships, environment and spirituality beliefs. It was developed for addiction patient, especially for patient with problems related with heroin.

The HRQoLDA test, or TECVASP test as originally known in the Spanish version, is a quality of life instrument developed specifically for the assessment of drug users and is based on the bi-axial concept of addiction of Edwards and Gross (Reference Edwards and Gross1976). According to these authors, the alcohol dependence syndrome was a combination of psychological and physiological process. This was one axis of alcohol problems, while social, legal and other adverse consequences conform another axis. This concept was generalized to other substances abuse, and it was the basis for DSM-IV abuse and dependence construct (APA, 1994). Following this logic, the operational definition of the TECVASP included some of the main symptoms of drug dependence, and physical, psychological and social problems related with drug abuse were included as specific items. The content of operational definition was submitted to experts judgment. After that, psychometric experts checked congruence between item-content. Finally, a pilot study was developed with a drug dependence sample. Understanding the items, time management, and the adequacy of the response format were evaluated (Lozano, Rojas, & Pérez, Reference Lozano, Rojas and Pérez2005). Response categories were a five-likert scale in each item, assessing a 30 days period. This specific instrument was designed to assist clinicians and researchers to determine to what extent and intense drug addiction affects people’s QoL.

Significant psychometric evidence supports the HRQoLDA test as a reliable and useful assessment tool. In preliminary studies, reliability, as determined by Cronbach’s alpha, was 0.798 or above in the assessment of samples of heroin cocaine, crack and alcohol users (Lozano et al., Reference Lozano Rojas, Rojas Tejada, Pérez Meléndez, Apraiz Granados, Sánchez Muñoz and Marín Bedoya2007). A study of the Spanish version of HRQoLDA test applying the Rasch Rating Scale Model demonstrated that the psychometric properties were suitable for measuring the HRQoL in a drug user population (Lozano et al., Reference Lozano Rojas, Rojas Tejada and Pérez Meléndez2009). But, the test has 2 weaknesses: Item 15 did not fit (it was recommended that it was not eliminated because of the importance its content has to the operative definition of the HRQoL) and one category of response utilized in this test shown an inadequate performance.

The Spanish version of the HRQoLDA test also revealed satisfactory validity evidence based on relations to other variables (Known Groups) in that significant differences were observed between heroin users in MMT and non-MMT individuals (Lozano et al., Reference Lozano Rojas, Rojas Tejada, Pérez Meléndez, Apraiz Granados, Sánchez Muñoz and Marín Bedoya2007). In terms of its relations with other variables, HRQoLDA scores have shown significant correlations with severity of dependence, as well as with physical and psychological health (Lozano et al., Reference Lozano, Domingo-Salvany, Martínez-Alonso, Brugal, Alonso and de la Fuente2008). The HRQoLDA test was also useful as a tool in the detection of clinical improvement among patients treated with buprenorphine (e.g. González-Saiz et al., Reference González-Saiz, Lozano Rojas, Martín Esteban, Bilbao Acedos, Ballesta Gómes and Gutiérrez-Ortega2011).

In this study, the authors present the results of the psychometric analysis performed by applying a Polytomous Rasch Model of the adaptation of the original Spanish version to English, as applied to a sample of substance users in Australia. Rasch models estimate both item and person parameters with the same model, offer person-free item parameter estimation and item-free person's measure estimation, provide an optimal scaling of individual differences, etc. (e.g., Reise & Henson, Reference Reise and Henson2003). The application of Item Response Theory (IRT) models (where Rasch models could be included) provides psychometric evidence cannot be obtained through the Classical Test Theory (CTT) analysis. The evidence from each psychometric approach helps to illustrate the methodological quality of an instrument. The COSMIN-COnsensus based Standards for the selection of health Measurement Instruments- checklist (Mokkink et al., Reference Mokkink, Terwee,., Patrick, Alonso, Stratford, Knol and de Vet2012) difference between the psychometric contributions from the CTT and IRT.

The specific aim of this study was to analyze the psychometric properties of the English version of HRQoLDA test applying the Rating Scale Model (RSM), and emphasizing (a) fit between empirical data and theoretical model assumptions; (b) item category probability curves; and (c) precision in terms of information function.

Method

Participants

This study evaluated 121 adults recruited from inpatient and outpatient treatment facilities across the Sydney West Area Health Service (SWAHS) catchment area in western Sydney, Australia. Research sites included Blacktown Hospital, Cumberland Hospital, Nepean Hospital, and the Mount Druitt Centre for Addiction Medicine, all of which are higher education training facilities within Sydney West Area Health Service (SWAHS).

Potential participants presenting for treatment were randomly invited to respond to the questionnaires. The inclusion criteria consisted of fulfillment of the DSM-IV criteria for substance abuse disorders (dependence and abuse) and the ability to understand the aim of the study, as well as the content of the questions in both questionnaires, which entailed a satisfactory command of English. Exclusion criteria comprised of presentations exclusively due to alcohol abuse and/or involuntary admission for inpatient treatment.

This study was granted approval by the SWAHS Human Research Ethics Committee. Prospective participants were provided with a written protocol pertaining to the study and a verbal explanation about the purpose of the study. They were also informed that participation was voluntary, confidential and anonymous. Volunteers were also informed that they could withdraw from this study at any time without any repercussion to their treatments. Research participants were then asked to sign an informed consent form prior to their inclusion in the study.

Instruments

The HRQoLDA Test is a quality-of-life assessment tool specific for drug abuse (Lozano et al., Reference Lozano Rojas, Rojas Tejada and Pérez Meléndez2009). It assesses the physical and psychosocial aspects of life through 20 five-choice items in a Likert-type scale, with choices being designated 1 to 5 points. The choices are scored as follows (see Appendix): items 1–14 (never, 1 to 2 times, 3 to 7 times, 8 to 14 times, 15 times or more), items 15–18 (never, a couple of times, sometimes, many times, all the time) and, items 19–20 (not at all, a little bit, a fair bit, pretty much, absolutely). A structure of one dimension has been checked for this version (Zubaran et al., Reference Zubaran, Sud, Emerson, Zolfaghari, Foresti and Lozano2012). The sum of the 20 scores represents quality of life, such that the higher the score, the better the quality of life.

The Spanish version of the HRQoLDA test, whose acronym is TECVASP test, was translated into English. Recommendations from the International Tests Commission were used in adaptation process, taking into account semantic, idiomatic, experiential, cultural, and conceptual equivalence between the source and the target instruments (e.g., Muniz, Elosua, & Hambleton, Reference Muñiz, Elosua and Hambleton2013). The final English version of the HRQoLDA test was published by Zubaran et al. (Reference Zubaran, Sud, Emerson, Zolfaghari, Foresti and Lozano2012) using Classical Test Theory analysis.

Procedure

Five data collectors underwent a period of training and supervision from the central researcher prior to administering questionnaires independently. The data collectors met regularly to address any queries and ensure each were following the same procedure.

The study participants completed the questionnaires under minimal guidance from the trained examiners, who followed standardized instructional procedures. Interviews took place in a suitable room at one of the research sites mentioned above. Occasionally, specific questions not considered in the initial instruction procedure were answered on a one-to-one basis. Particular care was taken with non-native English speaking participants in order to ensure a satisfactory understanding.

Analysis

The RSM, a Rasch model for polytomous items, was used in the psychometric analysis. The RSM transform person and item raw scores to interval measures which can be located on the same metric. If the data fit the RSM all the items measure one unidimensional construct. In the RSM, probability of response to an item (π_nix) is a logistic function determined by ability of persons (β_n) and item difficulty (δ_i). In the case of the HRQoL construct, the probability that a person will affirm an item is a logistic function of the difference between the person’s level of HRQoL and the level of HRQoL expressed by the item. The RSM is expressed by (e.g., Wright & Masters, Reference Wright and Masters1982)

$${\pi _{nix}} = {{\exp \sum\limits_{j = 1}^x {\left[ {{\beta _n} - \left( {{\delta _i} + {\tau _j}} \right)} \right]} } \over {\sum\limits_{x = 1}^{m + 1} {\exp \sum\limits_{j = 1}^x {\left[ {{\beta _n} - \left( {{\delta _i} + {\tau _j}} \right)} \right]} } }}\quad \quad x = 1,2, \ldots .m + 1$$

where ß_n is the location or value of person n on the scale (ability); δ_i is the location of the item on the scale (difficulty), and τ_j, called the threshold parameter, is the location of threshold m for each response category with respect to item location on the scale. A property of the Rasch models is that item location estimates and person ability estimates are expressed on the same metric and are directly comparable.

Fit analysis.

For the RSM, the response patterns achieved are tested against what is expected (Pallant & Tennant, Reference Pallant and Tennant2007). This is known as fit analysis. Rasch analysis results are interpretable only when data fit the model. A variety of fit statistics determine whether this is the case (Smith, Reference Smith2000). Mean square fit statistics (MnSq) and standardized mean square (Zstd) are two of the most commonly used statistics. Two indexes are reported in these statistics: the INFIT and the OUTFIT indexes. The INFIT index (MnSq and Zstd) is a t standardized information-weighted mean square statistic, which is more sensitive to unexpected behavior affecting responses to items near the person's measure level while the OUTFIT index (MnSq and Zstd) is a t standardized outlier-sensitive mean square fit statistic, more sensitive to unexpected behavior by persons on items far from the person's measure level (Linacre, Reference Linacre2000).

In general, data is fitted to the model when MnSq are between 0.5 and 1.5 logits. Values larger than 1.5 indicate unmodeled noise. Values less than 1.0 indicate overly predictable. For Zstd data is fitted to the model when values are within the range of -2 to +2. In this study, the psychometric software WINSTEPS (vers. 3.64.2) was used (Linacre, Reference Linacre2000).

Results

Items and persons fit

In general, the summary of items and persons show that there is a good fit between the reported data and the model (Table 1). Nevertheless, minimum and maximum values indicate that not all items and persons have acceptable fit values. Item analysis show that 6 of 20 items have MnSq values which go beyond an acceptable range. Specifically, values of items 12, 13, 17, 19 and 20 indicate that they cannot be modeled, and item 5 indicates overly predictable. On the other hand, all items except 19 have good values for Zstd statistics. Persons fit show that 68.6% of patterns of response observed are compatible with model.

Table 1.

Statistics of model and items with five response categories.

Note: MnSq: Mean square fit statistics; Zstd: Standardized mean square.

Items misfit indicates unpredictable responses. In those cases, additional psychometric analyses were pursued.

Thresholds and Category Probability Curve

A Category Probability Curve (CPC) shows the probability of selecting each response category of an item along the HRQoL continuum. The values of threshold parameter values are shown in table 1. For items 1 to 14, the response category 3 has a lower threshold than category 2. In items 19 and 20, the threshold values for category 3 are superior to those for categories 4 and 5. The implication is that categories in these items are disordered.

Graphically, it can be observed that threshold parameters have influenced the CPC in each response format (Figures 1.A, 1.B and 1.C). An item with five response categories produces five regions of probable response in a continuum of values (regions: R1, R2, R3, R4 and R5). In this case, for items 1 to 14, response categories 2 (‘1 to 2 times’) and 4 (‘8 to 14 times’) are not the most likely categories in any segment of the continuum (Figure 1.A). Also, for items 19 and 20 (Figure 1.C.), response categories 3 (‘a fair bit’) and 4 (‘pretty much’) are not the most likely in any segment of the continuum.

Figure 1.

Category Probability Curves for items with five categories.

The five item response categories should generate five distinct peaks where each category is most probable along a range of locations on the latent variable, suggesting empirical observations are consistent with intended category ordering. As it does not happen, the five response categories were reduced (ad doc) to three response categories (recoding 12223). Later, threshold parameters were ordered accordingly. Figures 2.A, 2.B and 2.C can show that CPC has been effective, in that each category defines a region of likely response in the continuum.

Figure 2.

Category Probability Curves for items with three categories.

Results of the recoding procedure have altered the way data fit the model. Table 1 shows that 18 out of 20 items have achieved optimal adjustment values. However, items 12 and 19 have generated inappropriate values of MnSq, but not in ZSTD. Taking into account these results as well as the importance of the content of the operative definition conveyed by these items, the authors have decided to retain these respective items. As a result, the percentage of research participants who presented adjusted parameters of responses increased to 73.7%.

Items and persons map

The items and persons map (figure 3) shows the combined position of persons and items on the HRQoL continuum. On the left are the measurements of the persons, and on the right the items appear in order according to their values (table 2) on the HRQoL construct. Respondents are located to the left, represented by an ‘X’. To the right are positioned all items, including in parentheses the respective value in the scale of each item. Letters ‘M’, ‘S’ and ‘T’ indicate the mean value, with one and two standard deviations respectively.

Figure 3.

Item and person map.

Table 2.

Summary of category structure for items.

Table 3.

Statistics of model and items with three response categories.

Note: MnSq: Mean square fit statistics; Zstd: Standardized mean square.

The items with lower values are items 20 (self-referential expectations), 9 (hallucinations) and 11 (become disoriented). Items in the scale with higher values are item 4 (lack of energy), 2 (sleeping problems) and 5 (felt anxious). This calibration of items generates a more comprehensive and accurate instrument’s structure vis-a-vis the assessment of persons with substance use disorders. For instance, taking into consideration individual ‘A’ with low self-referential expectations, his or her answers in the remaining items would probably indicate a low HRQoL. By contrast, an individual ‘B’ who reports lack of energy, most probably will show a pattern of responses to remaining items indicating a better HRQoL.

The position of persons demonstrates that most of them are located between the item with the highest and the lowest value in the scale. On the other hand, 19.8%, presented HRQoL scores above the item of highest value in the scale.

Test Information Function

The test information function reports the test precision at different levels of the HRQoL construct. Information curves are analogous to reliability of measurement and indicate the precision (reciprocal of the error variance) of a test along the underlying trait continuum (Hays, Morales, & Reise, Reference Hays, Morales and Reise2000). In Figure 4 it can be seen that the scale has its maximum precision in the central values of a continuum of scores.

Figure 4.

Test Information Function.

Rasch models compute a separation index for persons and another for items instead of reliability coefficients. Person separation index (PSI) is used to classify people and represents the number of statistically different performance strata that the instrument can detect in the sample. Low person separation (< 2) indicates that the instrument could be not sensitive enough to distinguish between high and low performers and more items may be needed (Linacre, Reference Linacre2014). The PSI was 2.45, corresponding to a Rasch person reliability of 0.86 per persons. The Rasch reliability (of persons) is comparable to the traditional test reliability.

Item separation is used to prove the item hierarchy. The hierarchy of item is especially important because the proposed items can be compared with the order of item locations estimated from the data (Linacre, Reference Linacre2014). Low item separation (< 3) implies that the person sample is not large enough to confirm the items hierarchy of the instrument (the test has no item with high, medium, low locations; Linacre, Reference Linacre2014). The separation index for items was 5.39, corresponding to an item reliability of 0.97. The mean SE of items was 0.17.