In our study, the results demonstrated a satisfactory psychometric properties and diagnostic accuracy of the GFI-C. The 15-item three-factor structure GFI-C was a reliable and valid instrument for screening frailty and pre-frailty in community-dwelling older Chinese people.
Regarding internal consistency, the Cronbach’s alpha of the total scores was 0.867, which indicated good internal consistency (Kline, 2000). This result was consistent with a previous study (Xiang et al., 2020; Olaroiu et al., 2014; Metzelthin et al., 2010). Except the subscale of ‘psychological component’, the Cronbach’s alpha values of the total scores and subscales of ‘physical components’ and ‘social component’ were higher than 0.70, indicating satisfactory internal consistency for the GFI-C subscales of ‘physical’ and ‘social’ (Portney et al., 2009). The Cronbach’s alpha of the ‘psychological’ component was slightly lower than 0.70 possibly because the items of ‘psychological component’ were few. The number of items in a scale or subscale considerably contributed to the magnitude of the internal consistency (Portney & Watkins, 2009). The finding was satisfactory in scale level (ICC = 0.865). Frailty would be affected by external and environmental factors. In this study, the plausible reason affecting the retest score might be historical event (Polit, 2014), which was Chinese New Year. The first interview of this study was implemented from November 2017 to February 2018. The Chinese New Year of 2018 was from February 15 to 22, and the retest of the GFI-C was just implemented after Chinese New Year. The percentages of agreement of items 14 and 15 were 74% and 76%, respectively, which were the lowest of the 15 items. Chinese New Year means a new beginning and happiness to all Chinese people, and hence participants for test–retest reliability might possibly provide positive answers for ‘psychological’ subscale in the second interview. These reasons in relation to the impact of Chinese New Year might lower the stability of the GFI-C. Implication for social influence of frailty and refusal to conduct retest after festival were recommended for future research.
The results of correlation matrix between the GFI-C and Fried’s frailty phenotype indicated that the two instruments had a significant and strong correlation, showing the satisfactory concurrent validity of the GFI-C (Kraemer, 1980). In particular, four components of the GFI-C demonstrated significantly strong and moderate correlation with Fried’s frailty phenotype. Such results added credibility to support the concurrent validity of the GFI-C. The theoretical hypothesis stated that the score of the GFI-C is correlated to the degree of cognitive impairment and physical capability of older people. The correlation of score of the GFI-C and SBI was negative, indicating that the participants rated as low physical dependency by the SBI (i.e. higher score) had a lower GFI-C score on average. The correlation of the scores of the GFI-C and AMT was negative, indicating that the participants rated as cognitively impairment (i.e. lower score) by the AMT had a higher GFI-C score on average. The significantly strong and moderate correlation between the GFI-C and AMT and the GFI-C and SBI results supported the hypothesis and indicated the construct validity of the GFI-C.
The CFA was used in examining construct validity of the GFI-C, and a three-factor model of the GFI was identified through principal component analysis in Bielderman’s study with a sample size of 1508 persons. By examining the factor loading of 15 items, all paths were significantly loaded onto the hypothesised subconstructs, and 86.7% items obtained the loading of 0.32 or greater (range of loadings = 0.21–1.49). These items mainly belonged the subconstruct of ‘Health Problems’. With respect to the results of internal consistency, corrected item–total correlation of items 8 and 9 were low, indicating weak homogeneity in the respective ‘physical’ component. Four items were cross-loaded: items 11 and 12 and items 3 and 4 with large modification indices (MIs) of 49.75 and 44.38 separately. As a large MI reveals the presence of factor cross-loadings and error covariance (Byrne, 2013), model re-specification or modification was used, and the model was re-estimated for the improvement of model fit (Kline, 2011). In summary, the goodness-of-fit indices generated by the CFA model for the three-factor structure of the GFI-C were acceptable. All paths were significantly loaded to the hypothesised subconstructs. The evidence supported the construct validity of the GFI-C of 350 older Chinese people and contained three factors: ‘Daily Activities’, ‘Health Problems’ and ‘Psychosocial Functioning’.
By interpreting the results from ROC curves, a cut-off value of 2 (the maximum value of Youden Index) enriched the pre-frailty screening with the GFI-C, with acceptable sensitivity, specificity and AUC. For frailty screening, a cut-off value of 3 provided satisfactory sensitivity (88.2%) and specificity (79.6%) compared with that of a previous study (cut-off value ≥ 4, sensitivity = 66%, specificity = 87%; Baitar et al., 2013). All the results of sensitivity and specificity and the AUC supported a new cut-off value of 3/15 (i.e. a score of ≥3 indicated frailty). A conventional cut-off value of 4 for the GFI has been adopted in many frailty epidemiology studies since the development of the instrument (Steverink et al., 2001; Metzelthin et al., 2010; Peters et al., 2012; Baitar et al., 2013; Bielderman et al., 2013; Drubbel et al., 2013; Olaroiu et al., 2014; Luh, Yu & Yang, 2018; Xiang et al., 2020). However, the current study examined the optimal cut-off value with the well-accepted instrument of frailty (i.e. Fried’s frailty phenotype), and frailty status was diagnosed by a nurse. The satisfactory and comparable sensitivity and specificity and the AUC results indicated a new cut-off value (i.e. 3), which is different from the original one (cut-off value of 4). The plausible reasons for the change in cut-off value are culture differences between older Western and Chinese people, differences in living habits and difference in the purpose of GFI-C screening. In the background of Chinese Confucian ideology, the noun ‘face’ not only means the outside appearance of a person but also represents the self-esteem, dignity and reputation of a person and is the invisible existence of social psychology in Chinese (Yan, et al., 2007; Lam, 2015). For instance, in item 5 of ‘what mark do you give yourself for physical fitness?’, Chinese people provided rate better score than Western people as they want to protect their ‘face’. In other words, they want other people to think that they are worthy of respect. However, in Western culture, people paid more attention to individual feelings than thoughts from other people. Thus, for item 5, older Chinese people might get a lower GFI-C score than older Western people. Another culture difference between China and Western countries is that in Chinese traditional Confucian ideology, ‘standards of filial piety’ is essential to Chinese traditional culture (Feng, 2017; Lam, 2015). Chinese families tended to live together and take care of older people owing to the traditional virtues of showing filial respect for older people. Given that the ‘face’ issue of older Chinese people and the fact of supporting family members, ‘social component’ with items 11–13 might have lower GFI-C scores than in older Western people, who were more likely to live alone or with their partner rather than their sons or daughters. Moreover, owing to the ‘psychological component’, older Chinese people might also answer in a positive way. All these reasons induced a low score of the GFI-C in the measurement of frailty in older Chinese people. A study published in 2010 stated that 22.8% adults in China never measured their body weights, and the lower their education levels are, the higher the proportion of weight gain is (Jiang et al., 2013). As the demographic data of our participants showed the average age of participants of this study was 75, and 27.4% of them were illiterate. However, item 8 of the GFI-C which was ‘during the past 6 months have you lost a lot of weight unwillingly?’ would require our participants to know their weights or have a habit of measuring their body weights regularly or recently. When the study results and the ‘face’ issue of Chinese people were considered, some of our older people participants provided ‘no’ answer to item 8 which presented as a lower score of the GFI-C. In addition, there was an old saying in Chinese is that ‘taking medications is just like taking poison’, which reflected the Chinese culture of not taking medications unless they were really ill. Moreover, Chinese traditional herbal medicine was more acceptable in China than Western medicine. Even though older people in China had to take medications, 91.8% of the community-dwelling older Chinese people did not know the names of medications and 55.6% had forgotten to take medications exactly as prescribed by their doctors (Jin et al., 2005). As item 9 of the GFI-C asked about the medication types of our participants, older people in China might not be able to correctly distinguish types of medications they were taking and sometimes they would not follow prescription and use Chinese herbal medicine or tea instead. Hence, their real medication status may be underestimated, and the score of the GFI-C may be low. As the plausible reason for lower score of the GFI-C rated by Chinese older people, a cut-off value of 3 might be attained. In the literature review, frailty was strongly linked to the adverse outcomes of older people, including fracture, falls, hospital admission and mortality (Cawthon et al., 2007; Topinková, 2008; Feng et al., 2017; Kane et al., 2012; Clegg et al., 2013). The early detection of frailty can reduce adverse outcomes in older people, and thus interventions for improving their health status can prevent them from becoming frail. Moreover, frailty can be detected early with a GFI-C instrument. The results of sensitivity and specificity of the GFI-C (Table 4) showed that at a cut-off value of 3, sensitivity (i.e. 88.2%) was better than that in previous studies (sensitivity = 66%), but specificity (i.e. 79.6%) was lower than that in previous studies (specificity = 87%; Baitar et al., 2013). However, when the cut-off value was 4, the sensitivity of the GFI-C changed to be 76.4%; and specificity, to 89.3%. The high sensitivity indicated few false negative of frailty, and low specificity indicated more false positives of the GFI-C at a cut-off value of 3. In other words, the low sensitivity indicated many false negatives of frailty, and high specificity indicated few false positives at a cut-off GFI-C value of 4. On the basis of the results of the GFI-C of 350 Chinese older people in the current study, the number of older people screened as frail was 169 at a cut-off value of 3, versus 132 at a cut-off value of 4. This result demonstrated that at a cut-off value of 3, the GFI-C can screen more older people as frail than when 4 was used as a cut-off value. As a screening tools, more conservative was important to reduce the chance of missing any frail case.
Apart from the satisfactory results of the current study, two areas of limitations related to the generalisability of sample and methodological issues of this study should be noted. For the generalisability issue, owing to convenience sampling, the results of frailty prevalence rate cannot be generalised to the target population (i.e. all community-dwelling older people in China) because of the potential bias of the sampling method (Bornstein, Jager, & Putnick, 2013). The inclusion criterion that distinguishes frail but non-communicable individuals in communicable older people led were left out. For methodological issues, these limitations hinders the application of supplementary laboratory investigation to diagnosis accuracy tests.