The Chinese population is on the brink of an aging era, and the percentage of individuals aged ≥65 years is projected to increase from 13% to 27.9% (180–380 million) between 2020 and 2050 (World Health Organization, 2018; China Development Research Foundation, 2020). The lifespan of individuals is correlated with low levels of physical activity and increased dependency. This eventually leads to a higher demand for long-term care needs, which was estimated at 45.30 million in 2020 and is expected to increase by 39.0% to 59.32 million by 2030 (World Health Organization, 2015; Zapata-Lamana et al., 2021; Gong et al., 2022; Musa et al., 2022; Parra-Rizo et al., 2022). However, family-based non-professional care is becoming less feasible due to the China's historical one-child policy, the prevalence of empty nesters amid urbanization, and the widespread migration of the population (Qian et al., 2018). Consequently, care provided by nursing home staff is emerging as an indispensable alternative. Long-term care resource allocation (i.e., conditions of access to nursing home and clinical care needs) is identified by standardized grading assessments (Zhang et al., 2018), which typically centers on the assessment of activities of daily life (ADLs) (Fortinsky et al., 1999; Hébert et al., 1999; Holanda et al., 2022; Jeppestøl et al., 2022). The original 10-item Barthel Index (BI) has gained recognition and been integrated into the China's standardized grading assessment system for evaluating general ADL function in older adults due to its communicability, simplicity, and ease of scoring (Dickinson, 1992; Bouwstra et al., 2019; Medical Administration and Medical Authority, 2020). Studies have highlighted favorable psychometric properties of BI when used for patients with stroke, Parkinson's disease, and older adults (Pashmdarfard and Azad, 2020). However, one study indicated that the reliability of BI fluctuated when it was used for evaluating severe disability (Sainsbury et al., 2005). Given that nursing home residents, typically above 70 years of age, often exhibit serious disability or dependence (Muszalik et al., 2021; Kashiwagi and Morioka, 2022; Zhao et al., 2022), questions arise regarding the suitability of BI for their evaluation. A previous study using BI-10 to assess 644 patients with dementia across 19 long-term care facilities in Thailand, Japan, and South Korea revealed compromised item fit, including item bias, redundancy, and narrow threshold widths, casting doubt on the suitability of BI for assessing dementia and indicating a need for modification (Yi et al., 2020). However, to date, no study has been conducted to validate the use of the 10-item BI for nursing home residents in mainland China. Therefore, it is imperative to promptly validate the psychometric properties of BI for assessing nursing home residents in mainland China.

The classical test theory (CTT) ha been widely used to assess the psychometric properties of scales, focusing on assessing construct validity and internal consistency. However, it overlooks measurement precision, leading to an incomplete understanding of scale's psychometric properties (Hunsley and Mash, 2008). In contrast, the item response theory (IRT) has emerged as a more comprehensive framework, offering deeper insights into internal consistency, factor structure, and measurement precision. This theory provides a precise psychometric framework that delineates interactions between individuals and items to determine if items effectively measure the intended population. Additionally, precision is typically demonstrated through item information curves or test information, highlighting the optimal discrimination location of an item or a scale across individual latent traits (Reckase, 2009). Moreover, IRT enables the detection of item- and test-level biases via differential item functioning (DIF) analysis, allowing for adjustments to mitigate or eliminate these biases (Teresi et al., 2012). Moreover, IRT assessed item functions, such as the discrimination ability and difficulty of each item, without relying on the sample, a capability that CTT lacks (Reckase, 2009; Teresi et al., 2012). Hence, compared to CTT, IRT offers more comprehensive information (e.g., item function, measurement precision) for instrument development and validation.

Therefore, the objective of this study is to validate the psychometric properties of the 10-item BI, tailoring their use to suit the specific context of nursing homes. The current study hypothesizes that the 10-item BI may have compromised psychometric properties for evaluating nursing home residents in mainland China and a modified version of the BI may serve as a suitable instrument.

The present study employed the IRT to validate the psychometric properties of the BI for nursing home residents, aiming to gather more detailed item information compared to the CTT (Reckase, 2009). This study represents the first large-sample examination of BI's psychometric properties among nursing home residents in mainland China. The findings confirmed our hypothesis that the 10-item BI showed unsatisfactory psychometric properties (violation of local independence and item misfit). While the modified BI demonstrated optimal psychometric properties: the unidimensional structure indicated a good model fit, the items possessed robust discriminative power, the thresholds displayed an orderly trend, and no item was misfit. Thus, the modified BI may present as the better instrument for evaluating nursing home residents. It retained nine items but still assessed ten aspects of ADLs, as some items and categories were consolidated, instead of being discarded, to achieve the optimal psychometric properties and preserve the maximum information from the original scale. Moreover, the modified BI showed high measurement accuracy for residents with moderate functional impairment rather than those with minimum or maximum function levels.

While the fit indices in the validation samples displayed inconsistencies regarding model fitting (CFI and TLI >0.95 indicated a good model fit, while RMSEA >0.1 suggested a poor fit), the one-factor GPCM remained the optimal model. A study conducted by Keke et al. suggested that strong correlations among observed variables might yield a “good” CFI but a “bad” RMSEA, emphasizing the importance of balance among various fit indices (Lai and Green, 2016). In this study, item pairs exhibited strong correlations (r1 = 0.51–0.93), contributing to the unfavorable RMSEA performance. Although making potential modifications to the model, such as excluding highly correlated item pairs, might yield a “better” RMSEA, it would result in a loss of valuable item information. Therefore, despite the compromised RMSEA due to substantial item correlations, the one-factor GPCM remained the optimal model in this study.

Ensuring adherence to the local independence hypothesis is essential for accurate parameter calculation. In this study, item pairs (items 5 and 6) violated the local independence hypothesis (>0.1) and were combined into a new item (bowel and bladder control). This finding aligns with previous research indicating that the mean thresholds for “controlling bladder” and “controlling bowel” were nearly identical, suggesting potential redundancy between these items (Yi et al., 2020). Furthermore, physiological mechanisms suggest that bladder and bowel contraction are controlled by the pelvic musculature, potentially leading to concurrent urinary and bowel dysfunction (Hubscher et al., 2021; Kajbafzadeh et al., 2021). Among nursing home residents, who are typically aged over 81 years and have decreased contraction ability of the pelvic musculature, there exists a likelihood of concomitant urinary and bowel dysfunction. Hence, consolidating “bladder control” and “bowel control” into a single item is warranted.

Regarding item fit, items 5, 6, and 9 displayed misfit. This misfit can stem from inappropriate item definitions, poor category definitions, or an excessive number of choices, leading to discordant participant responses (Andrich and Luo, 2003). Common strategies to address this issue involve eliminating misfit items or collapsing categories. Collapsing categories is the primary option to preserve more information. Following the collapsing of categories, item 56 comprises two categories “incontinence” and “voluntary”. This action was corroborated by prior research using the BI for assessing long-term care residents, which indicated the limited utility of the second category (“occasional incontinence”) within items related to bowel and bladder control for discriminating latent traits among individuals (Liu et al., 2015). In nursing home settings, residents experiencing complete incontinence typically use incontinence products and are assessed under the “complete incontinence” category. Conversely, those with intact sphincter control manage themselves and are assessed as having “voluntary” bowel and bladder control. Occasional incontinence poses a challenge as it often goes unnoticed until caregivers discover wet bed sheets (Sayabalian et al., 2019). Such instances might be classified as complete incontinence due to heavy caregiving burdens and workforce shortages in nursing homes, or they might go undetected, leading to categorization as voluntary bowel and bladder control. Consequently, the category of occasional incontinence becomes ambiguous. Thus, “incontinence” (consolidating categories of occasional and complete incontinence) and “voluntary” emerge as more suitable options for item 56 (bowel and bladder control) in nursing home assessments. Similarly, item 9 (mobility) exhibited misfit. The adjacent categories of “wheelchair independent” and “walking with help” might be ambiguous as many nursing home residents prefer wheelchairs despite being capable of walking short distances. Collapsing these two categories into a new category (“some help from others or device”) rendered item 9 compatible with the GPCM. This aligns with previous studies employing the Rasch analysis, which also suggested the potential ambiguity of the “mobility” item's second (wheelchair independent) and third (walking with help) categories leading to model misfit (van Hartingsveld et al., 2006; Yi et al., 2020). Therefore, collapsing categories aligns with the GPCM analysis and practicality within nursing home settings.

In terms of item functions, all nine items exhibited excellent discrimination, implying their robust ability to distinguish between varying functional levels. No disorderly thresholds were observed, indicating the suitability of the categories within the BI. Moreover, the maximum information was concentrated within the latent trait (θj) ranging from −1 to 1, while minimal information function was detected at latent trait (θj) values near −3 or 3. This indicates that the BI possesses a narrow assessment precision width, effectively discriminating among nursing home residents with moderate functional impairments but lacking the capacity to differentiate residents with minimal functional limitations or severe functional impairments. Similarly, the 15-item BI also demonstrated a narrow assessment precise width (−1 < θ < 1) (Zhao et al., 2022). Furthermore, reports of ceiling effects (>15% of the sample attaining the maximum score) and floor effects (>15% of the sample scoring the minimum) have been reported across different populations, such as patients with stroke (Balu, 2009), patients in the intensive care unit (Reis et al., 2021), and older hospitalized patients with cognitive spectrum disorders (Braun et al., 2021). These observations suggest an inherent limitation in the BI's measurement due to its focus on a limited range of ADLs and the absence of assessment beyond basic ADLs (Bouwstra et al., 2019). Consequently, this restricted range of individual locations within the BI might lead to a narrow precision width. The findings of the current study align with the observation that the BI comprises ADL items encompassing individual locations ranging from−1 to 1, offering precise measurement specifically for individuals experiencing moderate functional impairment. In our study, items such as “grooming”, “dressing”, “bowel and bladder control”, “toilet usage”, and “mobility” exhibited closely aligned difficulty levels, indicating potential redundancy among these items, prompting consideration for their removal. However, when a 5-item BI was developed from the original 10-item BI by eliminating redundant items, a significant loss of item information occurred, leading to unsatisfactory recommended reliability criteria (Hobart and Thompson, 2001). Hence, retaining these items with similar difficulty levels is essential to maintain the integrity of the scale structure and to preserve item information.

Regarding measurement bias, prior reports have not addressed DIF between the dementia and normal groups concerning the BI. DIF highlights a potential bias in measurement for each item. In this study, only non-uniform DIF was identified in item 2 (bathing). This suggests a potential disparity in bathing abilities between the dementia and normal groups, a trend supported by previous studies indicating a higher dependency on bathing among individuals with dementia (Liu et al., 2015). Despite this, minimal differences surfaced between the dementia and normal groups in the total expected score across all items, including those with DIF. This implies that any distortion in the scale due to the DIF item is likely insignificant.

The modified BI demonstrated favorable psychometric properties and proved to be suitable for evaluating nursing home residents experiencing moderate functional impairment, which may provide precise evaluation for long-term care resource allocation. Future studies could explore integrating supplementary measurements, such as objective indices, to assess a broader spectrum of functional statuses to potentially enhance the limited precision width observed in the BI.

The original contributions presented in the study are included in the article/supplementary material, further inquiries can be directed to the corresponding author/s.

The studies involving humans were approved by the First Affiliated Hospital of Guangzhou University of Chinese Medicine Ethics Committee. The studies were conducted in accordance with the local legislation and institutional requirements. The participants provided their written informed consent to participate in this study.

ML: Conceptualization, Data curation, Formal analysis, Investigation, Methodology, Project administration, Software, Writing – original draft. MY: Conceptualization, Data curation, Investigation, Project administration, Resources, Writing – review & editing. BG: Conceptualization, Data curation, Investigation, Project administration, Resources, Writing – review & editing. YP: Conceptualization, Data curation, Formal analysis, Methodology, Writing – review & editing. TZ: Conceptualization, Data curation, Writing – review & editing. JW: Conceptualization, Data curation, Writing – review & editing. ZY: Conceptualization, Supervision, Validation, Writing – review & editing.