Item Generation
With the initiation of a qualitative approach, relevant items have been generated. The experience of physical resilience of hospitalized older adults has been synthesized and summarized using a content analysis, including the inductive-deductive approach in interpreting these perspectives. The detailed information was published [6]. We formulated a 20-item questionnaire based on the previous study [7, 8] and major themes identified in participant interviews (Table 1). Content validity was analyzed using the evaluations from three experts who were experts in geriatric care. They independently assessed the consistency between initial pool items and the construct of physical resilience. Among panel discussion, The item N6 was removed due to redundancy and agreement of the experts; specifically, items N6 and N 16 had the strongest correlation coefficients, which is over 0.9. Finally, a 19-item questionnaire was developed (Supplementary Material), the content validity index (CVI) of which was 0.88. After the qualitative study procedures mentioned above, the present study decided an initial version of the PRIFOR to be a data collection tool in psychometric testing.
Psychometric Testing of the Generated Items on the PRIFOR
Sample
The target participants were enrolled from medical wards of one medical center located in Southern Taiwan. Inclusion criteria of the eligibility were (i) participants were age 65 and over, (ii) they had the ability to communicate independently, and (iii) their scores in the Clinical Frailty Scale (CFS) were between 4 and 6 [9]. The participants were excluded if their admissions were due to the following situations: (i) needing hospice care; (ii) needing surgery; and (iii) needing intensive care. The study was approved by the Institutional Review Board of medical center with the registered number of IRB No. B-ER-108-064. Also, all participants (or their family when the participants were unable to sign) provided written informed consent.
Measures
During the interview, participants completed the PRIFOR, the EuroQoL 5-dimension Questionnaire (EQ5D), the Katz Index of Independence in Activities of Daily Living (Katz ADL), the 5-item Geriatric Depression Scale (GDS-5), and the Short Portable Mental Status Questionnaire (SPMSQ).
The PRIFOR includes 19 items rated on 5-point Likert scale (1 = strongly disagree; 5 = strongly agree) assessing aspects of resilience associated with recovery following acute health stressors. The score range for the PRIFOR was between 19 and 95, where a higher score indicating greater levels of physical resilience [10].
The GDS-5 assesses the depression level of older adults. The scores of the five items are totaled, and a higher score indicates a higher level of depression [11, 12]. The GDS-5 has good psychometric properties, including a negative likelihood ratio of 0.07, a positive likelihood ratio of 4.92, a negative predictive value of 0.94, a positive predictive value of 0.81, a sensitivity of 0.94, and a specificity of 0.81 [13]. The GDS-5 in the present study also had good internal consistency (α = 0.75).
The SPMSQ uses 10 questions to assess how well the cognitive function is for older adults, and a point was given when the older adults provide a wrong answer. An older adult has intact cognitive if answering with fewer than 2 wrong answers; mild cognitive impairment with 3 to 4 wrong answers; moderate cognitive impairment with 5 to 7 wrong answers; and severe cognitive impairment with more than 8 wrong answers. Moreover, educational level is considered for adjusting the SPMSQ score [14]. The SPMSQ Chinese version has good internal consistency (α = 0.70) [15]. Criterion-related validity of the SPMSQ was supported [16]. The SPMSQ in the present study had good internal consistency (α = 0.88).
The Katz ADL determines the function levels of ADL by measuring daily activities of use of incontinence materials, eating, getting up out of a chair, visiting the toilet, dressing, and bathing. Scores between 0 and 2 are given to each daily activity (0 = dependence, 1 = limited assistance, and 2 = independence), and a total score between 0 and 12 can be calculated. Moreover, a higher score indicates better level of ADL independence [17]. The internal consistency of the Katz ADL was satisfactory (α = 0.84–0.94) [18], and the present study also had good internal consistency (α = 0.86).
The EQ5D assesses the quality of life for the general population. It contains five self-administered items with three different descriptions for each item. The three descriptions reflect three levels of health status and are coded as 1 (indicating no problems in health), 2 (indicating moderate problems in health), and 3 (indicating extreme problems in health) [19]. The descriptions in the five items were then converted into a 0–1 scale using the equation generated from a time trade-off technique [20]; a higher score in the 0–1 scale indicates better quality of life. The EQ5D in the present study had good internal consistency (α = 0.72).
Data Analysis
The descriptive statistics was used to analyze the participant’s demographics data, including mean and frequency. Afterwards, the 19 items in the PRIFOR were analyzed to understand their response distribution. At this stage, items without normally distributed responses (ie, skewness > 3 or kurtosis >8) and those with an extremely high ceiling or floor effect (ie, >50%) were removed. The items with normally distributed responses and acceptable ceiling/floor effect were additionally analyzed using the EFA to examine the factor structure of the PRIFOR. The EFA applied the extraction method of principal axis factoring and an oblique rotation (ie, oblimin) to test the PRIFOR factor structure. A factor is extracted when an eigenvalue is larger than 1. After the EFA recommended the factor structure, Rasch analysis with partial credit model was used to reexamine the unidimensionality of each factor recommended by the EFA. More specifically, two types of mean square (MnSq), infit MnSq and outfit MnSq, were applied to examine whether an item fits in its embedded construct. Acceptable infit and outfit MnSq are within the range between 0.5 and 1.5 [21]. If an item has misfit MnSq (either in infit or outfit MnSq), the item is removed and the EFA and Rasch analysis are reanalyzed until satisfactory Rasch fit statistics are achieved.
After the factor structure and the unidimensionality of each factor were verified, Cronbach’s α, Rasch separation reliability, and Rasch separation index were applied to understand the PRIFOR’s scale properties. A value higher than 0.7 in Cronbach’s α and Rasch separation reliability indicates good internal consistency of the PRIFOR. A value higher than 2 in the Rasch separation index indicates good discrimination of the PRIFOR (ie, the PRIFOR can effectively distinguish participants with different levels of physical resilience) [22].
Criterion-related validity was then carried out to understand whether the PRIFOR links well with relevant health-related outcomes. Pearson correlation was used to examine the bivariate correlations between the PRIFOR and each of the following health outcomes: depression (assessed using GDS-5), cognitive function (assessed using SPMSQ), ADL (assessed using Katz ADL), and quality of life (assessed using EQ5D). Except for the correlation with depression, the PRIFOR was expected to have positive correlations with all the health outcomes.
Finally, known-group validity was tested for the PRIFOR using the different levels of frailty among the participants. The participants were first classified into two levels of frailty: vulnerable (ie, scores 4 in the CFS) and mildly to moderately frail (ie, scores 5 to 6 in the CFS). Then, independent t-tests were used to examine whether the PRIFOR can significantly distinguish the participants into the two levels of frailty. Additionally, Cohen’s d was used to understand the effects in distinguishing the two levels of frailty.