Measuring personal recovery in people with a psychotic disorder based on CHIME: A comparison of three validated measures

Abstract

Living well in spite of residual symptoms of mental illness is measured with the construct of personal recovery. The CHIME framework might be suitable to evaluate personal recovery measures and guide instrument choice. Three validated measures were evaluated in Dutch patients with a psychotic disorder (N = 52). We compared the Recovery Assessment Scale (RAS), the Mental Health Recovery Measure (MHRM), and the Netherlands Empowerment List (NEL). The measures were assessed on six criteria: content validity (based on CHIME), convergent validity with a social support measure, internal consistency, floor and ceiling effects, item interpretability, and ease of administration. The MHRM scored high on content validity with a balanced distribution of items covering the CHIME framework. The MHRM and the NEL showed moderate convergent validity with social support. In all three measures, internal consistency was moderate and floor and ceiling effects were absent. The NEL scores demonstrated a high degree of item interpretability. Ease of administration was moderate for all three measures. Finally, the CHIME framework demonstrated good utility as a framework in guiding instrument choice and evaluation of personal recovery measures. The MHRM showed the best overall result. However, differences between measures were minimal. Generalization of the results is limited by cultural and linguistic factors in the assessment for the subjective measures (i.e. content validity and item interpretability). The broad and multidimensional construct of personal recovery might lead to ambiguous interpretations. Scientific consensus on a well‐defined personal recovery construct is needed.

Introduction

In recent years, the patient movement raised awareness on the importance of personal recovery (Liberman & Kopelowicz 2005; Roe et al. 2011). Personal recovery is defined by Anthony as a ‘deeply personal, unique process of changing one's attitudes, values, feelings, goals, skills and/or roles. It is a way of living a satisfying, hopeful, and contributing life even with limitations caused by the illness. Recovery involves the development of new meaning and purpose in one's life as one grows beyond the catastrophic effects of mental illness' (Anthony 1993). Several studies advocate for the implementation of personal recovery in nursing care (Jacob et al. 2017; Noiseux & Ricard 2008). The growing amount of research on personal recovery‐focused interventions emphasizes the increasing need for the evaluation of personal recovery measures.

Recently, the CHIME framework for recovery has received increased attention as a way to operationalize recovery (Leamy et al. 2011). This framework is based on a qualitative review of 115 studies. A narrative approach resulted in the five themes of the CHIME framework: Connectedness (support from others and being part of the community), Hope (a positive view on the future and motivation to change), Identity (building a positive identity and overcoming stigma), Meaning (developing meaningful roles and activities, quality of life, and spirituality), and Empowerment (gaining control over life and focus upon strengths; Leamy et al. 2011). The importance of CHIME is widely endorsed in the literature (van Weeghel et al. 2019) which makes the framework suitable for evaluating personal recovery measures.

Self‐report instruments for personal recovery, with straightforward interpretation such as those used in depression (e.g. Inventory of Depressive Symptomatology (IDS)), could serve evaluation of clinical care as well as research. However, personal recovery is characterized by its subjective nature and unique experiences (Wood & Alsawy 2018). Therefore, personal recovery measures contain broadly interpretable items so that a measure can cover a large variety of personal recovery experiences for patients with divergent levels of (cognitive) functioning (Bellack 2006). These aspects make personal recovery measures prone to the bandwidth–fidelity problem. This problem refers to the trade‐off between obtaining a score with high fidelity and narrow bandwidth on a specific trait versus obtaining a score that covers a broad bandwidth with less fidelity (Streiner & Norman 2008). Therefore, developing a generic measure with a broad bandwidth comes at the expense of specificity: mean scores on studies that use a personal recovery measure as the primary outcome can generate multiple interpretations. As an alternative approach, the dimensions of the CHIME framework could serve as single constructs resulting in more specificity compared to the multidimensional construct of personal recovery.

Furthermore, subscales in measures are commonly identified with factor analyses. However, in heterogeneous constructs, factor analyses do not always produce subscales with sufficient content validity as they rely on internal consistency of subscales. In other words, while factor analysis retains items which correlate highly with one another and thus improves internal consistency, it may also cause items to be dropped which have a lower correlation but are important for content validity (Streiner & Norman 2008). Given how CHIME was developed based on a qualitative review of the literature, the dimensions of the CHIME framework may produce subscales based on content validity rather than factor analysis.

In this study, we evaluate the Recovery Assessment Scale (RAS) (Corrigan et al. 1999), the Mental Health Recovery Measure (MHRM) (Young & Bullock 2005), and the Netherlands Empowerment List (NEL) (Boevink et al. 2016). The choice of measures was based on different qualities of the measures. The RAS was included because it is one of the earliest developed measures and is currently the most widely used scale for measuring personal recovery (Salzer & Brusilovskiy 2014). However, the RAS was not validated for the Dutch‐speaking population. Therefore, the MHRM was included as it was the first validated scale for measuring personal recovery in the Dutch mental healthcare population (van Nieuwenhuizen et al. 2014). However, both the RAS and MHRM are developed in Anglo‐Saxon countries and later translated in other languages. Translated measures may have limited generalizability to their targeted population and reliability measures, such as Cronbach's alpha, or floor and ceiling parameters might perform different in their country of development (Streiner & Norman 2008). Validation of the reliability of measures in a specific language is therefore needed. We therefore included the NEL, as it was originally developed in Dutch and validated in the Netherlands.

Personal recovery is a highly subjective construct in which language and cultural elements likely play an important role (Brijnath 2015; Slade 2012). Consequently, we expect the NEL to perform better on measures of applicability as this measure will be evaluated in the same country as it was developed. Based on prior research, we expect all three measures to perform adequately on construct validity, reliability, and applicability (Boevink et al. 2016; Burgess et al. 2011; Cavelti et al. 2012; van Nieuwenhuizen et al. 2014; Salzer & Brusilovskiy 2014; Shanks et al. 2013; Sklar et al. 2013).

Aims

In this study, we hypothesize that a recovery measure shows higher applicability scores when applied in the country of origin, compared to recovery measures from foreign countries. Therefore, we compare the RAS, the MHRM, and the NEL for measuring personal recovery in a Dutch population of patients with a psychotic disorder. Measures will be exploratory assessed on aspects of construct validity (content and convergent validity) and reliability (Cronbach's alpha and floor and ceiling effects) and explanatory on applicability for patients with psychosis (interpretability and ease of administration). Furthermore, we explore the CHIME as framework for analysing content validity.

Methods

Sample

A convenience sample was recruited within Lentis Psychiatric Institute (Groningen, The Netherlands) from October 2015 to February 2017. Patients with a psychotic disorder (schizophrenia spectrum: 295.xx, 297.1, 298,80, 298.90) (DSM IV, American Psychiatric Association 2000)), age ≥18, were eligible for inclusion. In total, 53 patients were recruited for this study. One patient was excluded due to missing data. Therefore, 52 patients were included in the analysis. A description of the sample is presented in Table 1.

TABLE 1.

Description of the sample (N = 52)

Item	Mean (SD)
Age	42.3 (12.69)
Illness duration (years)	15.1 (11.05)
	N (%)
Female/Male	12 (23.1)/ 40 (77.9)
Primary diagnosis
Schizophrenia	34 (65.4)
Psychotic disorder NOS	8 (15.4)
Schizoaffective disorder	7 (13.5)
Other	2 (3.8)
Unknown (missing data)	1 (1.9)
Type of care
Functional Assertive Community Treatment	28 (53.8)
Functional Assertive Community Treatment + sheltered housing	10 (19.2)
Hospital admitted	4 (7.7)
Rehabilitation care	10 (19.2)
Self‐reported psychosis
1	10 (19.2)
2	10 (19.2)
3	8 (15.4)
>3	11 (21.2)
Unknown	13 (25)
Occupation
Paid work	8 (15.1)
Voluntary work	21 (39.6)
Study	3 (5.7)
Other (e.g.: occupational therapy, domestic work)	21 (39.6)
Education †
Primary education	6 (11.6)
Lower secondary education	14 (26.9)
Upper secondary education	24 (46.2)
Bachelor or Master	8 (15.3)

^†International Standard Classification of Education.

Procedures and materials

All included personal recovery measures are self‐report questionnaires and use a Likert scale (range: 1 = strongly disagree to 5 = strongly agree) with higher scores representing more personal recovery.

Recovery Assessment Scale

The RAS consists of 24 items (range 24‐120). Corrigan et al. (1999) conducted a factor analysis resulting in five factors: Personal confidence and hope, Willingness to ask for help, Goal and success orientation, Reliance on others, and No domination by symptoms (Data S1). Good psychometric properties were found in the original English measure, with Cronbach's alpha ranging between α = 0.76 and α = 0.97 (Salzer & Brusilovskiy 2014). Test–retest reliability ranged from r = 0.65 to r = 0.88. Three out of six RCT's found the RAS to be sensitive to change in different psychiatric disorders (Salzer & Brusilovskiy 2014). In four reviews, the RAS received good evaluations for convergent validity and ease of administration (Burgess et al. 2011; Cavelti et al. 2012; Shanks et al. 2013; Sklar et al. 2013). For this study, the RAS was translated to Dutch using forward–backward translation with a native English speaker (Data S2).

Mental Health Recovery Measure

The MHRM was developed based on a recovery model by Young and Ensing (2005). It consists of 30 items with a scoring range between 30 and 150. In two reviews, the MHRM received good evaluations for internal consistency and ease of administration. Convergent validity was measured with constructs of empowerment, resilience, and community living. The correlations varied from 0.57 to 0.75 (Cavelti et al. 2012; Sklar et al. 2013). The Dutch version of the MHRM was used for this study. Exploratory factor analyses conducted on the Dutch MHRM reduced the original seven subscales to three (van Nieuwenhuizen et al. 2014): Self‐empowerment, Learning and new potentials, and Spirituality (Data S1). Cronbach's alphas for the three factors in the Dutch MHRM ranged from α = 0.86 to α  = 0.94 (van Nieuwenhuizen et al. 2014).

Netherlands Empowerment List

The NEL was developed in the Netherlands and based on the results of a conceptual study of empowerment. In this study, concept mapping was used with participation of 56 patients with severe mental illness (Boevink et al. 2017). The 40‐item measure has a scoring range from 40 to 200. The NEL consists of six subscales: Social support, Professional help, Connectedness, Confidence and purpose, Self‐management, and Caring community (Data S1). Good psychometric properties were found with a Cronbach alpha of α = 0.94 and a test–retest reliability of 0.79 (ICC). The NEL showed sensitivity to change in a two‐year randomized controlled trial that evaluated a personal recovery intervention for people with severe mental illness (Boevink et al. 2016).

Social Support List 12 – Interactions

Social support is recognized as a construct closely related to personal recovery (Salzmann‐Erikson 2013; Wood & Alsawy 2018). Social support is embedded in the CHIME framework through the Connectedness dimension. Moreover, support from peers, friends, or family benefits all aspects of the framework to a certain extent (as noted in subdimensions of CHIME such as Hope inspiring relationships or Meaningful life and social roles). The importance of social support is also reflected in studies of personal recovery‐focused interventions, which aim to strengthen the social network (Castelein et al. 2015; Vogel et al. 2019). Therefore, correlation of a personal recovery measure with a social support measure could serve to strengthen theoretical support for the construct validity.

To assess convergent validity with social support, we used the Social Support List 12‐Interactions (SSL12‐I) (van Eijk et al. 1994). The SSL12‐I was validated for use in the geriatric population (Kempen & van Eijk 1995). The 12‐item measure (range: 12–48) contains three subscales: Everyday support, Esteem support, and Support in problem situations. The SSL12‐I showed satisfactory psychometric properties with all subscales demonstrating an internal reliability coefficient of r ≥ 0.70. The mean interitem correlations were r = 0.29 for the complete measure and r ≥ 0.37 for the subscales.

Administration

Each personal recovery measure was completed along with an item on ‘ease of administration' using the same 5‐point Likert scale as the questionnaires. Furthermore, participants provided information about demographic variables, such as highest level of completed education and diagnosis.

Ethical considerations

This study is completed in accordance with the declaration of Helsinki (World Medical Association (WMA) 2013), and all participants provided written informed consent. Participation was on voluntary basis with no compensation.

Analyses

Measures were analysed on six criteria: content validity, convergent validity, internal consistency, floor and ceiling effects, interpretability, and ease of administration. Each criterion was scored on a scale from 0 to 2 (0 = inadequate performance, 1 = moderate performance, and 2 = adequate performance). All criteria and their scoring are further specified below.

Power analysis

The number of participants needed for this study was based on the formula that is described in Streiner and Norman (2008). Based on earlier research, we expected to estimate values of Cronbach's alpha close to α = 0.90 in the current study (Boevink et al. 2016; van Nieuwenhuizen et al. 2014; Salzer & Brusilovskiy 2014). With α = 0.90 (95% CI = 0.85 and 0.95) and >25 items per scale, the power analysis resulted in 50 persons.

Content validity

We operationalized content validity by matching each item of each questionnaire to one of the five dimensions of the CHIME framework. The individual items were mapped to the CHIME framework by three criteria. First, an item was considered valid when it reflected at least one of the five dimensions of the framework. Second, items that did not reflect any of the CHIME dimensions were considered superfluous. Content validity was based on the premise that all aspects of CHIME were considered equally important. A measure was therefore considered an adequate representation of personal recovery when all five dimensions of the CHIME framework were represented, and the items of the questionnaire were evenly distributed across all dimensions (Haynes et al. 1995). The mean number of valid items and their deviation from the mean (variance (σ)) across the different dimensions of CHIME was used as a measure for the total fit of the questionnaire to the framework. If the percentage of redundant items was ≤10% with a total variance of σ ≤ 30, content validity was considered adequate (i.e. two points). Furthermore, content validity was considered moderate with >10% and ≤15% redundant items and σ > 30 and σ ≤ 60 and inadequate with >15% redundant items and σ > 60. Content validity was assessed by three authors (JSV, JB, and RJL). The inter‐rater agreement was analysed with ICC estimates (two‐way mixed) and their 95% CI based on the mean and absolute agreement, resulting in a fair, good, or excellent agreement (Orwin 1994). A fourth author (SC) could be consulted in case discrepancies could not be solved in a consensus meeting. Having allocated each item to one of the CHIME domains resulted in an alternative subscale division for each instrument, with each of the CHIME domains serving as a subscale. Significant correlations between these new subscales and acceptable values of Cronbach's alpha (α = 0.70) would strengthen the validity of using CHIME as a basis for content validity analysis.

Convergent validity

Convergent validity for social support was operationalized by calculating a Pearson product–moment correlation coefficient between the recovery measures and the related construct ‘social support', as measured with the SSL12‐I. Correlations were considered small if r = 0.10, medium if r = 0.30, or high if r = 0.50 (Cohen 1998). Convergent validity was considered adequate if r ≥ 0.50 with P < 0.05, moderate if r ≥ 0.30, and <0.50 with P < 0.05 or inadequate if P > 0.05. Additional analyses were performed on subscales that primarily focused on connectedness (i.e. Reliance on others (RAS) and Social support (NEL)). Furthermore, Pearson correlations between the three measures were calculated to evaluate whether the same construct was measured between the measures.

Internal consistency

Cronbach's alpha was used to analyse internal consistency on total measures and on original subscales as well as the alternative CHIME subscales that were created in this study. Values of α < 0.70 reflect poor internal consistency and values of α > 0.90 reflect redundancy of items (i.e. items have too much similarity) (Streiner & Norman 2008). Measures with ≥14 items produce acceptable alphas (i.e. more items result in a higher alpha; Cortina 1993). Therefore, we also examined the mean interitem correlation of the total measures. Total measures and subscales with an alpha below α< .70 were further evaluated on item level by calculating alpha if the respective item was deleted. A mean interitem correlation between r = 0.15 and r = 0.20 was considered adequate (Streiner 2003). Internal consistency was considered adequate if all of the following criteria were met: α ≥ 0.70 and ≤0.90 with a mean interitem correlation between r = 0.15 and r = 0.20 on total and subscales. Internal consistency was considered moderate if one of these criteria were met and inadequate if none of these criteria were met.

Floor and ceiling effects

Floor and ceiling effects occur when >15% of the respondents attain the lowest or the highest possible score, respectively (Terwee et al., 2007). This is often the result in items which measure the extreme ends of the scales, that is which measure performance at the highest or lowest ends of the spectrum. This results in a negative impact on content validity, as participants at either extreme end cannot be distinguished from one another. The absence of floor and ceiling effects was considered adequate if ≤ 15% of the sample had either the highest or the lowest possible score, respectively (moderate >15% and ≤17.5%, inadequate >17.5%).

Item interpretability

All items of the measures were screened on interpretability. Each item was independently assessed by three authors (JSV, JB, and RJL). Item interpretability was assessed on seven aspects: ambiguous wording; double‐barrelled questioning; jargon; negative wording; lengthiness; complex sentence construction; abstract terms; and face validity (Streiner & Norman 2008). An item was considered adequate if no weak properties applied, moderate if only one weak property applied, and inadequate if ≥ 2 weak properties applied. Agreement on scores was then reached in a consensus meeting. This criterion was considered adequate if ≤ 25% of the items on a measure had weak properties, moderate if > 25% and ≤50% of the items had weak properties, and inadequate if > 50% of the items had weak properties.

Ease of administration

The ease of administration was measured with a single Likert scale item at the end of each measure (‘this questionnaire was easy to complete') ranging from 1 = strongly disagree to 5 = strongly agree. The result is presented with descriptive statistics (means and SD's). A Friedman test was conducted to evaluate significant differences between the three measures. Ease of administration was considered adequate with a mean score of ≥4, moderate with a mean score of ≥3 and <4, and inadequate with a mean score of <3. Patients with higher recovery scores might find it easier to complete the measure. To assess an association between ease of administration and personal recovery, a Pearson correlation was conducted on ease of administration with all three personal recovery measures.

All statistical analyses were performed using SPSS Statistics version 22 (IBM Corp 2013). In case of missing values, pairwise deletion was applied.

Results

Content validity

A good agreement was found between the raters (ICC = 0.783, 95% CI: 0.667–0.857). Table 2 shows the outcomes of the content validity analysis. The MHRM showed the most optimal item distribution across CHIME (σ = 25.6). The RAS scored high on Empowerment with 25% of the items on this dimension. Most items of the NEL were assigned to Connectedness (27.5%). Hope and Optimism was underrepresented in the NEL (7.5%).

TABLE 2.

Content validity rating of the RAS, MHRM, and NEL items as measured with the CHIME framework

CHIME categories	RAS (24 items) items (%)	MHRM (30 items) items (%)	NEL (40 items) items (%)
Connectedness	4 (16.7)	4 (13.3)	11 (27.5)
Hope and optimism	3 (12.5)	4 (13.3)	3 (7.5)
Identity	3 (12.5)	7 (23.3)	4 (10.0)
Meaning and purpose	5 (20.8)	7 (23.3)	7 (17,5)
Empowerment	6 (25.0)	5 (17.7)	8 (20.0)
Item does not map	6 (12.5)	3 (10.0)	6 (17.5)
Variance	29.5	25.6	64.4

MHRM, Mental Health Recovery Measure; NEL, Netherlands Empowerment List; RAS, Recovery Assessment Scale.

With the content validity analysis, items of measures were mapped to the CHIME framework. This resulted in an alternative subscale division of the measures in which each domain of CHIME could serve as a subscale. Significant correlations were found between the measures mapped by their respective CHIME dimension. Only the RAS and MHRM did not correlate on Connectedness. Other correlations ranged from r (50) = 0.335, P < 0.05 to r (50) = 0.717, P < 0.05. See Table S1 for the subscales based on CHIME that resulted from the content validity analysis. Cronbach's alpha was applied to measure the internal consistency of the alternative CHIME subscales and ranged from α = 0.442 to α = 0.836. Acceptable Cronbach's alphas (>0.70) were found for Connectedness on all three measures, for Identity on the MHRM and the NEL and for Hope and Empowerment on the NEL. See Table S2 for the internal consistency of all total and subscale measures.

Convergent validity

A significant medium correlation was found between the SSL12‐I and both the MHRM (r(50) = 0.414, P < 0.01) and NEL (r(50) = 0.418, P < 0.01). The correlation between the SSL12‐I and the RAS did not reach significance. Two subscales of the RAS and the NEL with a focus on connectedness were relevant for individual analyses. These subscales significantly correlated with the SSL12‐I with, respectively, a medium correlation of r(50) = 0.406, P < 0.01 (Reliance on others (RAS)) and a high correlation of r(50) = 0.607, P < 0.01 (Social support (NEL)). All recovery measures correlated significantly with each other (RAS and MHRM r(50) = 0.697, P < 0.01, RAS and NEL r(50) = 0.692, P < 0.01, MHRM and NEL r(50) = 0.821, P < 0.01).

Internal consistency

All three measures showed acceptable homogeneity (α ≥ 0.70) on total scores. The total score on the RAS (α = 0.838), as well as four out of five subscales of the RAS, showed an adequate internal consistency. The mean interitem correlation of the RAS was sufficient with r = 0.192. The subscales No domination by symptoms did not reach the lower boundary (α = 0.671). An analysis on alpha if item deleted showed that by deleting the item My symptoms seem to be a problem for shorter periods of time each time they occur alpha reached an adequate score (α = 0.771). The MHRM marginally exceeded the threshold for adequate internal consistency (α = 0.903) as well as for the mean interitem correlation of r = 0.245. The subscales Self‐empowerment and Learning and new potentials showed an adequate alpha. The subscales Spirituality consist of only two items and showed an alpha of α = 0.921. The NEL total score exceeded the upper threshold for both Cronbach's alpha (α = 0.920) and the mean interitem correlation (r = 0.223). Five out of six subscales of the NEL have an adequate internal consistency. The subscale Self‐management scored just under the threshold (α = 0.628). An analysis on alpha if item deleted did not result in α > 0.70 on a single‐item deletion. See Table S2 for the internal consistency of all measures.

Floor and ceiling effects

None of the measures exceeded the 15% threshold for floor and ceiling effects. Floor effects were absent in total mean scores and subscale mean scores in all three measures. Also, ceiling effects were absent in total mean scores, but not in subscales. Ceiling effects on subscales ranged between 0% and 7.7% for the RAS, between 1.9% and 11.5% for the MHRM, and between 0% and 9.6% for the NEL.

Item interpretability

A little more than half (57.5%) of the items of the NEL was scored as straightforward to interpret by participants. Both the RAS and the MHRM scored low on item interpretability scores with >60% of items containing weak properties. For example, ‘I feel at peace with myself' (MHRM) was considered an ambiguous item. Table 3 shows the number and percentage of properties with respectively a good, fair, or poor rating.

TABLE 3.

Item interpretability scores of RAS, MHRM, and NEL

	RAS (24 items) Items (%)	MHRM (30 items) Items (%)	NEL (40 items) Items (%)
Good	9 (37.5)	7 (23.3)	23 (57.5)
Fair	11 (45.8)	16 (53.3)	11 (27.5)
Poor	4 (16.7)	7 (23.3)	6 (15.0)

Good: no weak properties; Fair: one weak property; Poor: ≥2 weak properties; MHRM, Mental Health Recovery Measure; NEL, Netherlands Empowerment List; RAS, Recovery Assessment Scale.

Ease of administration

All three measures received a positive evaluation on ease of administration by the participants, with similar scores for the RAS (m = 3.71 ± 0.87), the MHRM (m = 3.51 ± 1.05), and the NEL (m = 3.67 ± 0.96). There were no significant differences (χ ² = 2.346, P = 0.309) between the measures. The RAS showed a significant positive correlation between the total recovery score and the ease of administration (r = 0.339, P = 0.014). Table 4 shows the summary of the six aspects that were measured.

TABLE 4.

Summary of the evaluation of three personal recovery measures

	RAS	MHRM	NEL
Content validity	1	2	0
Convergent validity	0	1	1
Internal consistency	1	1	1
Floor and ceiling effects	2	2	2
Item interpretability	0	0	1
Ease of administration	1	1	1
Overall scoring	5	7	6

0, inadequate, 1, moderate, 2, adequate; MHRM, Mental Health Recovery Measure; NEL, Netherlands Empowerment List; RAS, Recovery Assessment Scale.

Discussion

Main findings

Of the three included personal recovery measures (RAS, MHRM, and NEL), only the MHRM showed good content validity with a balanced distribution of the items covering the complete CHIME framework. Unlike the NEL and MHRM, the RAS did not show convergent validity with the SSL12‐I, despite the content validity analysis showing a reasonable number of items allocated to Connectedness. However, on subscale level, a medium significant correlation between the SSL12‐I and the Connectedness subscale of the RAS (Reliance on others) was found. On interpretability, only the NEL achieved adequate scores on more than half of the items (57.5%). The MHRM showed the best result across all six criteria included in this evaluation due to the higher rating on content validity. This minimal difference does not suggest that the MHRM is convincingly better than the other measures. As noted in the introduction, the evaluation of measures resulted in an overall adequate score. Each instrument demonstrates a similar number of strengths and weaknesses, and as such, specific application may be the most useful guide in determining which instrument to use. If a greater focus on Connectedness is desired, the NEL offers the strongest properties. If Empowerment is of more interest, then we suggest using the RAS.

In our introduction, we hypothesized better performance on the NEL on measures of applicability. Although the NEL scored better on interpretability, it did not perform better on ease of administration compared to the other measures, thus rejecting our hypothesis.

Interpretation of the results

The CHIME framework was used in this study for analysing content validity. Items were allocated to one of the CHIME dimensions. For each measure, this resulted in alternative subscales based on the CHIME framework. Significant correlations were found between subscales of the same CHIME dimension. This outcome suggests that CHIME could be used for evaluation of personal recovery measures. Additionally, CHIME could also be used for the development of new scales based on its singular dimensions.

Similar to previous studies, we found evidence for internal consistency in all three measures (Boevink et al. 2017; van Nieuwenhuizen et al. 2014; Salzer & Brusilovskiy 2014). However, the MHRM and the NEL showed a Cronbach's alpha of >0.90, indicating redundant items in the measure. Our finding of a Cronbach's alpha of >0.90 on the subscale spirituality of the MHRM is in line with the study of van Nieuwenhuizen et al. (2014). Also, in line with our study, the study of Boevink et al. (2017) found an alpha of >0.90 on the total measure on the NEL. In the current study, acceptable alphas were found for all but one subscale of the NEL (Self‐management, α = 0.628). This difference might be due to differences in sample characteristics: unlike the current study, the study of Boevink et al. (2017) was not limited to patients with psychosis. Furthermore, in the current study we found a possibility for a slight improvement on alpha in the RAS by deleting the item My symptoms seem to be a problem for shorter periods of time each time they occur. This will increase the internal consistency from α = 0.671 to α = 0.771 in the subscale No domination by symptoms. This finding is consistent with the theory on personal recovery, placing less emphasis on clinical symptoms.

The criteria for item interpretability were only partly met by the NEL and not by the RAS and the MHRM. Different populations might have a different understanding of the concept of personal recovery (Gopal & Henderson 2015) or the items or wording of a measure (Streiner & Norman 2008). For example, aspects of spirituality are differently conceptualized in diverse cultures (Brijnath 2015). As a possible result, interpretation of the items of the NEL was more straightforward for a Dutch team of researchers. This shows the importance of language in personal recovery measures. Furthermore, it reaffirms the influence of culture in the personal recovery construct (Brijnath 2015).

All measures showed moderate scores on ease of administration. In line with previous research, no differences were found in ease of administration (Cavelti et al. 2012; Sklar et al. 2013). Only the RAS showed a significant correlation with personal recovery scores, indicating that people with higher recovery scores finding it easier to complete this questionnaire.

Measuring personal recovery

The complexity and manifold dimensions of the construct of personal recovery are shown in this study and are discussed in several previous studies (Mathew et al. 2018; van Weeghel et al. 2019). For example, Liberman (2012) comments on the personal recovery construct by comparing it with self‐efficacy measures that often leave ‘much ambiguity about what is being measured' (Bandura 2006). Furthermore, Liberman (2012) argues that personal recovery is not a scientific construct and should therefore not be measured as an outcome in research. In an opinion paper, Bellack (2006) observed that the validity of self‐assessment of personal experiences can be argued in patients with psychosis because of impaired reasoning and reality distortion (Bellack 2006; de Jong et al. 2018).

A large part of the complexity and ambiguity can be attributed to the wide variety of views on personal recovery. The lack of consensus on the personal recovery construct leads to many measures with different accents. Concrete examples are the subscale Caring community in the NEL, which is not seen in other personal recovery measures (Boevink et al. 2017). Furthermore, there is debate if personal recovery should be viewed as a process or an outcome (Law et al. 2012; Liberman 2012; Wood & Alsawy 2018). A recent scoping review of van Weeghel et al. (2019) supports the viewing of recovery as a process rather than an outcome. Also, Davidson (2012) argue's that supporting patients who want to have a good life is at best a process with intermediate outcomes. The view of recovery as a process is reflected in measures that include stages of personal recovery. The Self‐Identified Stages of Recovery (SISR; Andresen et al. 2003) and the Stages of Recovery Instrument (Andresen et al. 2006) are examples of questionnaires that measure stages in the process of recovery. Other questions are to what extent personal recovery is influenced by culture (Brijnath 2015; Ellison et al. 2018; Gopal & Henderson 2015) and which themes are most important in the personal recovery construct (e.g. social relationships and social activity, self‐agency, or hope) (Ellison et al. 2018; Law et al. 2012). The CHIME framework offers a good starting point for consensus on these themes. However, even using the same framework can lead to different interpretations: in analysing content validity, based on CHIME, we came to different results than Shanks et al. (2013), who used the same framework. This review had much higher rates for the representation of the dimension Hope and Optimism (RAS: 17%, MHRM: 30%) and more items that did not match the CHIME framework in both the RAS (RAS: 29%, MHRM: 17%) compared to our evaluation.

Limitations

Several limitations should be considered. First, in this review, content validity and item interpretability were only evaluated by researchers and a researcher/clinician. The absence of other stakeholders (e.g. patients, family) limits the validity of these analyses.

Second, the use of the CHIME framework for analysing content validity was confirmed by high correlations of items across measures mapped by the CHIME dimensions. However, a confirmatory factor analysis would show a more definite answer on the question whether CHIME is a suitable framework for analysing content validity. This was, however, not possible in this study due to a limited sample size. Furthermore, there are other frameworks available (Ellison et al. 2018) as well as adaptations on the CHIME framework (Bird et al. 2014; Stuart et al. 2017). This study only tested one framework for personal recovery and was therefore not able to analyse dimensions that are not part of CHIME such as Difficulties (Stuart et al. 2017). Although CHIME was shown to be a suitable framework for personal recovery in the current study, and several others, it is important to note that it was developed based on Western scientific literature (Leamy et al. 2011).

Third, all measures were administered to the participants in the same order. This might have biased the outcome on ease of administration as systematic differences could appear from a primacy or recency effect. A counterbalanced design, in which measures are administered in a random sequence, would therefore have been preferable.

Fourth, the evaluation of convergent validity was limited by only assessing it with social support. This was shown by measuring convergent validity in personal recovery on subscale level (i.e. subscales on Connectedness), which resulted in higher correlations compared to correlations on total measures. Due to the multidimensional nature of the personal recovery construct, evaluating convergent validity with other (CHIME‐related) constructs such as empowerment, self‐esteem, and hope, as well as divergent validity with self‐stigma, loneliness, and depression, would provide additional insight.

Future Research

Personal recovery as defined by consumer views is still in need to transform vague, politically correct concepts into empirically reliable and valid scientific concepts (Liberman 2012). Critical evaluation of the construct, especially on the dimensions to be included in the construct, is therefore needed. An important aspect in reaching this consensus is social validation (Liberman & Kopelowicz 2005). With this form of validation, different stakeholders (patients, family, clinicians, researchers, and the broader public) are able to give input on the criteria for recovery. Furthermore, comprehensive construct validation could support the evidence for the personal recovery construct. However, it will not unambiguously prove the existence of such a construct (Streiner & Norman 2008).

The construct of personal recovery has gained much influence in mental health care. The call for the personal recovery model originated from a mental healthcare system with a primary focus on symptom reduction (Anthony 1993). The consequences of a psychiatric disorder such as loneliness or loss of identity were underexposed in this medical paradigm. As a consequence, the patient movement advocated for a more holistic approach, considering the personal needs of patients with a psychiatric disorder. This raises the question whether the personal recovery construct should be used to guide mental health policy and practice rather than be used for evaluating mental health outcomes (Bellack 2006).

However, the emphasis placed on it by the patient movement demonstrates its great meaning to the clients themselves. As such, it is important that efforts are continued to further operationalize the construct. However, currently the idiosyncratic (Mathew et al. 2018) and culturally sensitive (Brijnath 2015) aspects of the construct impede the development of broad generic measures. In addition, models of recovery are primarily based on concepts of the Western world, which limits generalization to and international use of these measures in Non‐Western cultures (Slade et al 2012). Until more clarity is established on the construct of personal recovery, an option might be to disentangle the personal recovery construct to the CHIME dimensions or other common constructs (i.e. social support, hope, and goal orientation) and measure them as separate constructs (Law et al. 2012). Often these constructs have a longer history in scientific literature and have shown psychometric validity (Silverstein & Bellack 2008) as well as the before suggested social validity.

Conclusions

A comparison of three personal recovery measures (RAS, MHRM, and NEL) on six criteria showed the MHRM to have the best rest results. However, there are several concerns with all personal recovery measures. The review process showed that measuring personal recovery is complex as there is little scientific agreement on the construct. In this regard, it should be considered that ‘not everything that can be counted counts, and not everything that counts can be counted' (Camaron 1967). However, as an alternative to broad generic measures, unidimensional measures that are part of the personal recovery construct (e.g. hope, connectedness) could be used to measure specific aspects of personal recovery.

Relevance for clinical practice

Personal recovery has gained an important role in mental health nursing care. In recent decades, many personal recovery measures were developed. This study evaluated three recovery measures (RAS, MHRM, and NEL). The MHRM showed the best result across all six criteria. However, the multidimensional and idiosyncratic nature of the personal recovery construct impedes a strong conclusion of a preferred measure. For use in mental health nursing practice, CHIME guided specific use of the NEL for a focus on Connectedness and the RAS for a focus on Empowerment.

Supporting information

Data S1. Original Subscales.

Data S2. Recovery Assessment Scale (Dutch version).

Table S1. CHIME Subscales.

Table S2. Internal consistency of the RAS, MHRM and NEL.