Fatigue in chronically critically ill patients following intensive care - reliability and validity of the multidimensional fatigue inventory (MFI-20)

N = 195 patients were consecutively enrolled in a rehabilitation hospital (Bavaria Clinic Kreischa) where they were weaned from long-term ventilation. For study participation they had to fulfill the diagnosis of a Critical Illness Polymyopathy (CIP, ICD-10: G62.80) or Critical Illness Myopathy (CIM, ICD-10: G72.80). A convenience sample of patients (see Fig. 1) with the following further inclusion criteria participated: sufficient German language skills, ICU stay of at least six days, alert and able to understand the questionnaires, transfer from ICU at acute care hospital during the weeks before inclusion in the present study. The patients were asked for study participation orally at the Bavaria Clinic Kreischa. A short cognitive test, named Confusion Assessment Method for the Intensive Care Unit (CAM-ICU) [14, 15], was applied in order to preclude cognitive impairment. The patients were informed about the study protocol and informed consent for study participation was received within four weeks following the transfer from ICU at acute care hospital to the post-acute ICU at the rehabilitation hospital (t1).

A detailed description of the study procedure has been already published elsewhere [16]. At t1, basic medical and sociodemographic data (marital status, educational level, work status, age, gender) were obtained from the patient record forms. Clinical data contained the clinical diagnoses, the severity of medical illnesses, the occurrence of sepsis, the sites of infections, the length of mechanical ventilation and the length of ICU stay. The latter two were completed at t2 and t3. The severity of the medical illnesses was assessed indirectly via the Barthel-Index (BI) [17]. The BI is a measure of performance in activities of daily living and of the early rehabilitation state (e.g. intensive care supervision, tracheostomy tube management, mechanical ventilation, confusion, severe impairment of communication, and dysphagia). It was assessed at admission and discharge from rehabilitation hospital by a trained study nurse. A minimum value of − 325 and a maximum value of 100 could be reached. Higher values indicate a better performance.

At t2 (three months post-transfer) and at t3 (six months post-transfer), the patients were contacted via telephone. During the telephone interviews, the following instruments with relevance for the present research question were applied: the MFI-20 [12], the questionnaire Euro-Quality of Life (EQ-5D-3 L) [18] and the Structured Clinical Interview for the Diagnostic and Statistical Manual of mental disorders DSM-IV (SCID-I) [19].

The severity of fatigue was assessed with the Multidimensional Fatigue Inventory-20 (MFI-20) [12, 20] three months and six months following the transfer from acute care ICU via telephone interview. The MFI-20 is a 20-item self-report measurement of fatigue. It covers the five dimensions General Fatigue (GF), Physical Fatigue (PF), Mental Fatigue (MF), Reduced Motivation (RM) and Reduced Activity (RA). Each subscale contains two positively (e.g. „I feel very active.“) and two negatively (e.g. „I tire easily.“) formulated items. Items are rated on a 5-point Likert scale (range 1 „yes, that is true “to 5 „no, that is not true“) which are summed up to a simple total score with a minimum value of 4 (absence of fatigue) and a maximum value of 20 for each subscale. A total fatigue score is calculated as the sum of the subscale scores (range 20–100). Higher total scores indicate higher levels of fatigue. Validity has been shown for different participant populations e.g. cancer patients, army recruits, chronic fatigue syndrome. Internal consistency has been shown to be good for the GF, PF and MF dimensions (Cronbach’s α .84) [12] and adequate for the subscales RA and RM (Cronbach’s α > .65) [12].

The health-related quality of life was measured with the questionnaire Euro-Quality of Life (EQ-5D-3 L) [18] at t2 and t3. The EQ-5D-3 L assesses five dimensions (mobility, self-care, usual activities, pain/ discomfort and anxiety/ depression) which are rated within three severity levels (no problems, some or moderate problems, extreme problems or unable). A single one-dimensional index value is generated based on a simple sum score according to Hinz et al. [21]. This sum score was subjected to a linear transformation leading to values between 0 and 100. Higher values indicate a higher health-related quality of life. In the present study Cronbach’s α for the EQ-5D-3 L was .74 at t2 and .75 at t3.

The diagnosis of major depression was ascertained via SCID-I at t2 and t3. A clinical psychologist with at least five years of clinical practice applied the SCID-I via telephone contact.

The present study protocol was in consent with the Declaration of Helsinki and a positive votum was received by the Ethics Committee (No 3278–10/11) of the Friedrich-Schiller-University, Jena, Germany.

Sociodemographic and item characteristics were displayed as frequencies. Mean values, standard deviations, medians and interquartile ranges were reported depending on the distribution of the dependent variables. Chi-squared tests or Fisher’s exact tests were applied to compare categorical data. Mann-Whitney U tests were calculated to compare ordinally or non-normally distributed data. Standardized Cronbach’s α, item-total subscale (corrected-to-total) and inter-item correlations were calculated as parameters of internal consistency. Cronbach’s α values >.70, item-total subscale (corrected-to-total) correlations ≥ .30 and inter-item correlations of .30 to .70 were acceptable [22, 23]. Measurement error, defined as systematic or random error not attributed to true changes in fatigue between the two consecutive assessments of the MFI, was calculated according to Elbers et al. [22] using Bland and Altman plots [24]. The limits of agreement were determined using the mean difference values (MFI-total, MFI subscales) between t2 and t3 ± 1.96 x standard deviation (SD) of the difference. A linear regression between difference and mean scores was applied in order to detect a proportional bias. MFI subscales were intercorrelated controlling for age and gender [12, 20, 23]. Confirmatory factor analyses (CFA) was applied using the maximum likelihood estimation (MLE) in order to assess the structural validity of the MFI-20 in the present sample of CCI patients at t2 and t3 (see Table 4, Additional file 1: Table S3). Separate CFAs were applied in the sample of CCI patients at t2 (n = 113) and t3 (n = 91). The MFI-20 may be regarded as structurally valid if the items adequately represent the proposed latent factors or measurement model. The following models were evaluated: model A assuming the original five latent fatigue factors (GF, PF, MF, RA, RM), model B assuming one single underlying latent factor, model C with a 2-factor model combining model A and B. Since the subscales representing the three factors GF, PF and RA showed high intercorrelations (see Table 3), these factors were summarized and tested in model D (3-factor model). In models A, C, and D, the latent fatigue factors were allowed to covary with mean values fixed to 0 and variances to 1. The fitness of each model with the data were evaluated using comparative fit indices (CFI, cut-off > .95; Tucker-Lewis Index, TLI,cut-off > .95) and absolute fit indices (Chi² goodness-of-fit, Root Mean Square Error of Approximation, RMSEA, cut-off <.08) [25, 26]. A Chi² value smaller than the number of degrees of freedom (at p > .05), can be regarded to be good. The CFA was carried out using IBM® SPSS® AMOS 24.0.0. The convergent validity was assessed via Spearman’s and point-bisearial correlation with the health-related quality of life and the SCID-based diagnosis of major depression. Above, the convergent validity was evaluated by composite reliability (CR > .7) and average variance extracted (AVE > .5) value of each factor. Discriminant validity was evaluated by maximum shared variance (MSV < AVE), average shared variance (ASV < AVE) and square root of AVE greater than inter-factor correlations [27]. CR, AVE, MSV, ASV and square root of AVE were calculated using stats tool package. Floor and ceiling effects were ascertained if more than 15% of the CCI patients had either the lowest or highest possible score on the MFI subscales at t2 or t3. For the statistical analyses SPSS 24.0 was used. All results were considered significant at p ≤. 05 (two-tailed).

The internal consistency values were reasonsable for the MFI-total (Cronbach’s α = .91) and four of the five subscales (GF, PF, RA, MF, Cronbach’s α range: .69–.86). An inadequate value (Cronbach’s α = .50) was received for the subscale RM (Table 2). Regarding the inter-item correlations, a mean value of ≥ .30 was ascertained for the MFI-total and all but the RM subscale (.20, range .06–.32). The lowest correlation was present between item 15 („I have a lot of plans.“) with item 18 („I don’t feel like doing anything.”) (.06) and item 12 („I feel rested.“) (.02), suggesting item redundancy for item 15. Item-total correlations were ≥ .30 for all but the subscale RM (range: .22–.37) as well as the MFI-total (range: .23–.71) (Table 2). Pearson item subscale correlation coefficients ranged from .11 to .58 (absolute values) with the lowest and non-significant values for items 6, 11 and 20. Of them, only the removal of item 20 (“Physically I feel I am in an excellent condition.”) would lead to an increase of Cronbach’s α for the respective subscale (Additional file 3: Table S2). Values at t2 and t3 were not systematically different as shown in the Bland and Altman plots (Additional file 4: Figure S1).

There was no effect of gender for all the MFI subscales. Only the MFI subscale GF was positively associated with age (Spearman’s rho = .24, p = .01). Nevertheless, partial correlations were calculated controlling for age and gender. At t2, the MFI subscales showed medium-sized to high correlations between each other, ranging from .37 to .77. At t3, the pairwise correlations ranged from .55 to .76 (see Table 3).

Fit statistics revealed a poor model fit for all models at both time points (Table 4, Additional file 1: Table S3). According to the original 5-factor model (A), all but one standardized regression weights β (corresponding to factor-item correlation coefficients) were > .05 at p < .001. The lowest estimates were observed for item 15 for models A and B (t2/ t3, model A: β = .363/ .465, model B: β = .242/ .358). For model C, only the items of the MF subscale showed highly significant β coefficients at both t2 and t3. The best fit to the data (RMSEA ≤ .089) could be obtained for the 3-factor solution both at t2 and t3. All but items 9 and 16 were highly significantly correlated to the subscales. When each of the five factors was evaluated in separate CFAs, the factor RM turned out to show the poorest fit indices at both t2 and t3.

According to model A, the composite reliability (CR) was appropriate (CR > .7) for all MFI subscales at both time points. AVE was acceptable (> .5) besides for RM. Discriminant validity (MSV < AVE, ASV < AVE) could not be ascertained at both time points. The square root of AVE was greater than the inter-factor correlations only for the subscale MF.

According to model C, the CR was appropriate (CR > .7) only for MF at both time points. AVE was inappropriate and only acceptable (> .5) for MF at t2. Discriminant validity (MSV < AVE, ASV < AVE) could be ascertained at both time points.

Referring to model D, the CR was acceptable for all subscales besides RM. AVE was only appropriate for MF. Discriminant validity (MSV > AVE) was only appropriate for MF as well. Square root of AVE was smaller than inter-factor correlations besides for MF at t3.

Since no unequivocal factor model of the MFI-20 could be confirmed in the present sample of CCI patients, correlations were calculated only with the MFI-total but not with the single MFI subscales. The relationship between the MFI-total and the health-related quality of life (EQ-5D-3 L) yielded significant high-sized correlations at both t2 and t3 (range: − 65-(−)66). Significant small- to medium-sized correlations (range: .27–.37) could be obtained with the diagnosis of major depression (see Table 5).

There was neither a floor nor a ceiling effect for the subscales GF, PF, RA and RM. For the MF subscale, floor effects were detected. N = 30 patients (26.5%) had the lowest possible test score at t2, n = 19 patients (20.9%) at t3.