Psychometric performance of the WHO-5 well-being index in a nationwide sample of inpatients discharged from specialised mental health care

Mental health conditions are among the leading causes of health loss worldwide [1], and the number of individuals reporting mental health problems has risen in recent years, increasing pressure on health services [2]. Although treatment can reduce much of this burden [1], quality measurement in mental health care still lags behind that of somatic health services, especially in evaluating service effectiveness [3]. Valid and reliable tools are needed to capture patient-reported outcomes and experiences, which are key to patient-centred care and service improvement.

Patient-reported experience measures (PREMs) and patient-reported outcome measures (PROMs) provide insight into how patients perceive their care and outcomes. These measures complement clinical indicators by giving voice to patients, supporting shared decision-making, quality improvement, and policy development [4]. As such, PREMs and PROMs are central to delivering high-quality, patient-centred services.

In Norway, the Norwegian Institute of Public Health (NIPH) has conducted extensive research on PREMs and PROMs to support a more patient-centred health system. This work includes both methodological investigations into instrument validity and reliability, and broader studies on predictors of patient experiences and social inequalities. While most previous research has focused on PREMs, the shift from cross-sectional to longitudinal data collection in mental health and substance dependence surveys has increased the relevance of PROMs [5]. Longitudinal designs allow for a deeper understanding of how patient experiences and outcomes evolve over time, particularly after discharge.

This perspective enables the tracking of changes in health status and care trajectories, supporting a more comprehensive assessment of care quality and patient well-being. Although validated PREM instruments have long been in use, the integration of PROMs is now gaining importance [6, 7]. Such integration has become a growing international priority, as reflected in the OECD’s Patient-Reported Indicator Surveys (PaRIS) initiative, which recently developed and field-tested PaRIS Patient Questionnaire (PaRIS-PQ) integrating PROMs and PREMs [4, 8‐11]. PROMs capture health outcomes from the patient’s perspective, including how these relate to treatment, and can be either condition-specific or generic, covering a broad range of psychosocial and physical domains [4]. As mental illness often impacts well-being and quality of life, generic PROMs are particularly valuable in mental health care.

As part of the development of the national post-discharge survey targeting adults recently discharged from specialised inpatient mental health care, NIPH conducted a structured selection process to identify suitable PROMs for integration. Several PROMs were considered during this process. As a part of the PaRIS initiative [8], which recently field-tested the PaRIS-PQ in 18 countries [9], the NIPH followed the recommendations of the PaRIS Mental Health Group, which aims to harmonise PREMs and PROMs internationally [4, 10, 11]. Based on these guidelines, the five-item World Health Organization Well-Being Index (WHO-5) was selected for its strong validity, international relevance, and ease of integration into existing survey frameworks. It was used alongside two OECD-developed questions on overall life satisfaction and perceived meaning or purpose in life [10, 11].

The WHO-5 is one of the most widely used generic instruments for measuring subjective psychological well-being [12]. Originally developed for primary care, it has since been widely adopted in general population studies and across a range of clinical conditions, particularly for depression screening. The scale comprises five positively worded items rated on a Likert scale. It has been translated into over 30 languages and validated in many countries [13]. Its psychometric and clinical validity is well documented across age groups, diagnoses, and care settings, including mental health [12, 14‐17]. Importantly, the WHO-5 is one of the few instruments that assesses well-being independently of physical or mental functioning [18].

Scores are obtained by summing the five items (range 0–25) and rescaling to 0–100, with higher values indicating better well-being. Widely used thresholds are ≤ 50, indicating reduced well-being warranting follow-up, and ≤ 28, suggestive of very low well-being [12]. These cut-offs are recommended for screening and monitoring purposes only, not for diagnostic classification [12]. Validation of the WHO-5 therefore concerns the interpretation of scores as indicators of subjective well-being and their intended use for monitoring change over time and comparing subgroups or services [9, 11, 13].

Most previous WHO-5 validation studies have been conducted in selected health care settings [12], with some also including general population samples [13, 19, 20]. The instrument has demonstrated good validity among psychiatric patients [21‐23], and a study confirmed its robust psychometric properties in outpatient mental health care [24]. However, recent literature indicates that measurement properties of the WHO-5 are not entirely invariant across populations and settings [12, 13], and findings from psychiatric samples suggest that score distributions and cut-offs may vary [21, 23]. To date, little is known about its performance among patients recently discharged from specialised inpatient psychiatric care, a clinically vulnerable group with high symptom burden, comorbidity, and heterogeneous recovery trajectories. This may increase the risk of floor effects and reduced separation between mid-range categories. Moreover, few studies have validated the WHO-5 in nationwide samples. While nationwide validation studies are useful for examining subgroup differences in heterogeneous clinical populations, establishing population norms requires a representative population sample and is not attempted here. This study therefore aims to evaluate the reliability, construct validity, and measurement invariance of the WHO-5 in a sample of recently discharged adult psychiatric inpatients, using baseline data from a nationwide longitudinal survey.

Respondent characteristics are shown in Table 1. A total of 2310 patients completed the survey (response rate: 33.5%). Most were women (64.0%), and 47.2% were aged 25–44 years. Regarding education, 31.9% had only primary education, and 33.3% held a university or college degree. The majority were born in Norway (84.9%), and 33.3% lived with a partner. Employment status varied: 23.6% were in paid work, while 53.6% were unable to work due to illness or poor health. Prior hospital experience also varied: 29.4% had never been admitted to psychiatric care, while 26.8% had been admitted five times or more. Length of stay ranged from 1 to 2 days (17.5%) to ≥ 1 month (16.8%). Before admission, 81.9% rated their mental health as “very poor” or “rather poor”, and 5.4% as “very good” or “rather good”. At survey time, 41.6% still rated their mental health as “very poor” or “rather poor”, while 18.3% reported it as “very good” or “rather good”. Current general condition was rated as “very poor” or “rather poor” by 36.6%, and “rather good” or “very good” by 25.9%. Physical health was rated “very good” or “excellent” by 13.5%, and “poor” or “rather good” by 57.0%.

Respondents (n = 2310) differed from non-respondents (n = 4584) on sex, age, and education (Table S1). Women were somewhat more likely to respond than men, response rates were highest among those aged 25–66 and lowest among the youngest and oldest, and patients with higher education responded more often than those with primary education. These differences were statistically significant but moderate.

The distribution of responses and item-level mean scores are presented in Table 2. The proportion of missing responses on the WHO-5 items was low (0.6–0.8%). The mean total score on the WHO-5 was 33.4 (scale 0–100), indicating low well-being. A high proportion of lowest-category responses was observed for items 3 (“I have felt active and vigorous”) and 4 (“I woke up feeling fresh and rested”), with 28.2% and 35.0% of respondents selecting “at no time”, respectively. None of the items were excluded from further analysis. Item-level response distributions and mean scores are shown in Table 2 Subgroup differences by sex, age, and education are presented in Table S2, and score distributions are visualised in Fig. S1.

EFA using a polychoric correlation matrix and minimum residual extraction supported a unidimensional structure of the WHO-5. A single factor accounted for 71.7% of the total variance (eigenvalue = 3.59). Internal consistency was high (Cronbach’s α = 0.910, McDonald’s ω = 0.927), and item-deleted alpha values ranged from 0.881 to 0.899, indicating that all items contributed meaningfully to the scale’s reliability (Table 3).

The one-factor model was confirmed through CFA with the WHO-5 items specified as ordered categorical variables and estimated using WLSMV (Fig. 1). The χ² test yielded χ² (5) = 294.93, p < 0.001. Model fit indices were excellent for CFI (0.990), TLI (0.981), and SRMR (0.024), although the robust RMSEA was 0.149 (90% CI [0.136–0.163]), exceeding conventional thresholds. All item loadings were statistically significant (p < 0.001), with standardised values ranging from 0.802 to 0.905. R² values ranged from 0.643 to 0.819, indicating substantial explained variance. Item 4 had the weakest factor loading, suggesting reduced association with the latent construct, but its removal did not improve model fit, including the RMSEA value.

Measurement invariance was supported across sex, age, and education (Table 4). Across models, CFI was high (0.990–0.994) and SRMR was low (≤ 0.033). RMSEA was high for the configural and metric models (0.129–0.163) but acceptable for the scalar models (0.041–0.052). Changes between nested models were small and within decision thresholds (ΔCFI = − 0.001–0.003; ΔRMSEA = − 0.090 to − 0.021). Although DIFFTEST p-values were often significant, decisions were based on change-in-fit criteria, supporting full measurement invariance and valid mean comparisons across subgroups.

The WHO-5 correlated strongly with the OECD core items on subjective well-being, including overall life satisfaction (ρ = 0.764) and perceived meaning in life (ρ = 0.744) (Table 5). Similarly, strong correlations were found with self-perceived mental health (ρ = 0.727) and current general condition (ρ = 0.739). The correlation with self-rated physical health was moderate and negative (ρ = − 0.471), reflecting the item coding where higher scores indicate poorer health. Correlations with items on treatment-related understanding, coping, and hope were also significant, ranging from moderate to weak (ρ = 0.505–0.342).

IRT analyses provided further evidence of the scale’s performance. Discrimination parameters (a-values) ranged from 2.38 to 3.91, indicating variability in how well items differentiated between levels of well-being (Table 6). All estimated thresholds were ordered, indicating that the response categories functioned as intended. Some adjacent thresholds were located close to one another. Item 1 (“I have felt cheerful and in good spirits”) showed the highest discrimination (a = 3.91), reflecting high sensitivity to differences in the latent trait, while item 4 had the lowest (a = 2.38), indicating weaker differentiation. Threshold parameters (b-values) ranged from − 1.15 to 2.47, suggesting the items captured variation across low to moderate levels of well-being within this inpatient sample. Item 5 “My daily life has been filled with things that interest me” had widely spaced thresholds, whereas item 4 was more compressed, reflecting lower precision, particularly at the lower end of the well-being scale. Item fit was evaluated using the S–X² statistic. Although all item tests were significant (p < 0.001), this is expected in large samples. Item-level RMSEA values ranged from 0.02 to 0.03, supporting acceptable item fit. Global model fit was acceptable: M2 (5) = 147.75, p < 0.001, RMSEA = 0.112 (90% CI [0.097–0.128]), SRMSR = 0.028.

Item information functions (Fig. 2) peaked around θ ≈ 0, reflecting the sample mean. Items 1 and 5 provided the most information overall due to higher discrimination (Table 6), while item 4 showed lower discrimination and more compressed thresholds. Consequently, item 4 contributed less total information, with modestly greater precision toward the lower end of the continuum. Thresholds were ordered for all items, but the middle thresholds (b2–b3, b3–b4) were closely spaced, leading to overlapping category curves across items.

Categorical response curves (Fig. 3) showed that response categories were generally ordered and captured a meaningful gradient of well-being. However, overlap was evident among the adjacent middle categories (2-4), particularly for item 4, where thresholds were closely spaced (b2 = 0.29, b3 = 0.84, b4 = 1.39). Although thresholds were ordered for all items (Table 6), the narrower spacing in the mid-range reduced resolution around moderate well-being.

This study assessed the psychometric properties of the WHO-5 in a large national sample of adults recently discharged from specialised inpatient mental health care. The results support the scale’s reliability and construct validity in this population, demonstrating high internal consistency, a clear unidimensional structure, and acceptable measurement invariance across sex, age, and educational level. Our findings align with prior validation studies [13, 21, 30, 31] and confirm the scale’s performance in a post-discharge psychiatric population, a context that has received limited attention. Moreover, the findings underscore the strengths of the WHO-5 as a brief, generic tool for assessing psychological well-being, well suited for clinical and research settings where time and cognitive demands on patients are key concerns.

The factor structure of the WHO-5 was supported by both EFA and CFA. The one-factor model explained 71.7% of the variance, with strong loadings and low residual variances for most items. Items 1 (“I have felt cheerful and in good spirits”) and 5 (“My daily life has been filled with things that interest me”) showed the strongest associations with the latent construct, while item 4 (“I woke up feeling fresh and rested”) contributed less and showed moderate to high floor effects. Although the RMSEA exceeded conventional thresholds, this may be due to the very low degrees of freedom of the five-item model rather than substantive misfit. RMSEA is known to overstate misfit in such cases [33], and more weight should therefore be placed on CFI, TLI, and SRMR, which indicated acceptable fit.

Measurement invariance findings suggest that the WHO-5 performs consistently across sex, age, and education groups, supporting valid mean comparisons. Configural, metric and scalar invariance were supported, indicating similar conceptualisation, stable item loadings, and comparable item thresholds across groups. These findings support the WHO-5 as a psychometrically sound and broadly applicable PROM in psychiatric care, in line with international standards for equitable cross-group measurement [4, 10, 11, 37].

Strong correlations between WHO-5 scores and measures of subjective quality of life and mental health support the scale’s concurrent validity. Significant but weaker associations with physical health and enablement further confirm its construct validity. These findings align with previous validation studies and reinforce the scale’s utility for monitoring recovery and well-being in psychiatric populations [13, 22, 23].

IRT analyses provided additional insight into item functioning. Discrimination values varied, with item 1 showing the strongest ability to distinguish between levels of well-being. In the GRM, θ is sample-centred by default, with zero representing the mean of the inpatient sample rather than a population-wide norm. This explains why the item information functions peaked at θ = 0 despite patients’ generally low absolute well-being compared with general populations. Within this sample, Items 1 and 5 were the most informative overall, while item 4, though lower in discrimination, provided slightly greater precision at lower levels. Across items, the close spacing of mid-range thresholds produced overlapping category curves and reduced resolution around moderate well-being, a limitation that is common in short PROMs. These findings mirror concerns raised in previous studies [12, 13] and may reflect challenges in capturing nuanced states of well-being after severe mental illness.

Moderate floor effects were observed for items 3 and 4, particularly item 4, where 35.0% of respondents selected the lowest response category. This aligns with previous findings in psychiatric populations, where well-being scores often skew toward the lower end. Previous findings on floor effects have been mixed, likely reflecting differences in study populations [21, 24]. These effects may reduce sensitivity to the level of deterioration at the very low end, since the lowest response category (“At no time”) already represents complete absence of well-being. This insensitivity is not unique to the WHO-5 but reflects a general limitation of brief, generic well-being instruments [12, 13]. In severely impaired populations, more condition-specific PROMs may be required to capture variation at the lowest levels of well-being. Although item 4 showed relatively low discrimination and compressed thresholds in IRT analyses, it still captured an aspect of well-being (feeling fresh and rested) that is not fully represented by the other items. This conceptual contribution supports its retention despite weaker statistical performance. Thresholds spanned a reasonable range on the latent continuum, supporting measurement across low to moderate levels of well-being in this clinical sample. As θ in IRT is sample-centred, item information functions naturally peak around the sample mean rather than a population norm. The observed spectrum thus reflects the clinical characteristics of recently discharged psychiatric inpatients. This implies that sensitivity is reduced at the higher end, and limited precision in the mid-range of the well-being continuum.

The current response categories of WHO-5 are somewhat ‘unbalanced’, i.e., the theoretical distance between the two poorest (“At no time”, “Some of the time”) and the two best categories (“All the time”, “Most of the time”) are different. For all WHO-5 items, 50–60% of patients in our study clustered in the two poorest response categories. Given the clinical severity of mental health inpatients and the unbalanced response categories, there might be room for further differentiating low levels of well-being by introducing an additional category between the two poorest categories (“Rarely/seldom”). Thus, we propose further experimental research in this setting by randomizing patients to either receive the original 6-category format or an alternative response format more adjusted to the patient group (e.g., a 7-category format with “Rarely/seldom), as refinements of response options have been shown to improve measurement precision [54]. A revised format could be relevant and useful for applications involving patient populations with poor mental health, facilitating a more fine-grained measurement of the lowest levels of well-being and the ability to measure differences between individuals/groups and over time.

Including PROMs like the WHO-5 alongside established PREMs supports a more holistic understanding of care quality. A recently published scoping review highlighted the paucity of studies that concurrently assess PREMs and PROMs in psychiatric and substance use disorder services [55]. This dual approach, endorsed by the OECD’s PaRIS initiative [4, 8‐11], reflects international priorities and enables health systems to align patient perspectives on both care experiences and treatment outcomes. The use of standardised instruments like the WHO-5 facilitates international benchmarking, while also offering locally relevant data to support service improvement [4, 8‐11, 34, 35].

The WHO-5 can be applied at both the individual and system levels. At the individual level, it supports screening and monitoring of well-being at discharge, with thresholds of ≤ 50 and ≤ 28 indicating reduced and very low well-being, respectively [12]. Given the 0–100 scaling, differences occur in steps of four points. A change of around 10 points (typically 8–12 points) is commonly regarded as clinically meaningful [12]. At the service and system level, aggregated WHO-5 data can be reported as mean scores with confidence intervals, distributions, and proportions below the 50-point threshold, ideally with case-mix adjustment. This approach is consistent with evidence from large-scale validations [13] and aligns with international initiatives such as the OECD PaRIS programme, which promotes the integration of PROMs and PREMs in quality monitoring [4, 8‐11]. While the WHO-5 is not diagnostic and may show floor or ceiling effects, it provides a valid and efficient indicator of psychological well-being that can complement PREMs, other PROMs, and clinical outcomes [12, 24, 36, 55].

The WHO-5 demonstrated strong reliability and construct validity as a measure of psychological well-being among patients discharged from specialised mental health care. Despite some limitations, including moderate floor effects and overlapping mid-range response categories, the WHO-5 remains a robust and accessible instrument. To enhance precision, future revisions might explore refining the response format or supplementing the WHO-5 with condition-specific PROMs to better capture low levels of well-being.

Integrating PROMs like the WHO-5 alongside established PREMs can support a more holistic assessment of mental health care quality, aligning patients’ experiences with outcomes that matter to them. While this dual approach is a key international priority [4, 8‐11, 35, 36], evidence in psychiatric populations remains limited. Our findings demonstrate the relevance of using the WHO-5 alongside PREMs in this context, thereby adding to the emerging evidence base. Future work should aim to refine PROMs such as the WHO-5 to ensure greater sensitivity to diverse patient needs while also supporting broader system evaluations and international benchmarking. Such integration can strengthen mental health services’ ability to deliver patient-centred care that captures both experiences and outcomes.