Background
Heart failure (HF) is one of the most important health problems in terms of prevalence, morbidity, mortality and health service use [
1]. It affects around 2 to 3 % of the population and the prevalence increases with age, affecting as much as around 10 to 20 % of the population over 65 years old [
1‐
3]. In developed countries, the prevalence of HF is increasing due to population aging, longer survival of patients and effectiveness of secondary prevention [
4,
5]. Projections indicate that the prevalence of HF will increase as much as 46 % from 2012 to 2030 [
1]. In brief, HF is a common disease with a huge impact on the prognosis and lifestyle of patients and a growing challenge for health policy makers [
6].
The health-related quality of life (HRQoL) of patients with HF is an important outcome as it reflects the impact of HF on their daily lives [
7,
8]. Instruments to assess HRQoL provide a way to explore the perceptions of patients about how HF affects their daily lives and wellbeing, providing information that cannot be obtained directly from clinical measurements [
5]. In recognition of this, improving the HRQoL has emerged as an important treatment goal [
4,
9,
10].
Various specific HRQoL questionnaires for patients with HF have become regarded as important assessment tools in recent decades [
2,
5,
11,
12]. Among these, one of the most widely known and used is the Minnesota Living with Heart Failure Questionnaire (MLHFQ) [
10,
11,
13], which has been translated and culturally adapted into at least 34 languages, and has demonstrated good psychometric properties in numerous studies [
2,
5,
7,
11,
14‐
20]. However, there are some concerns about its factor structure and the homogeneity of items [
7,
10,
13,
17]. Several authors have even proposed different factor structures [
4,
7,
15,
18,
21]. When reviewing validation studies, we encountered certain problems and weaknesses. Firstly, while some authors obtained similar factor structures to the original developers of the questionnaire [
18,
22,
23], others obtained two-factor structures but disagreed on certain items [
7], and various even extracted structures of three factors, with a new social subscale [
4,
15,
16,
18,
21], but disagree on the items that make up the third factor. Secondly, in clinical practice, the MLHFQ is commonly used to generate a total score, which assumes that the total scale is unidimensional, but we found only two studies analyzing a single-factor structure [
15,
18], and they differ in their conclusions. Lastly, most studies that analyze the structural validity of the instrument have been carried out from the perspective of classical test theory (CTT) [
4,
7,
16‐
19,
21,
22], and more specifically, using techniques of exploratory factor analysis (EFA) rather than confirmatory factor analysis (CFA). Considering all this, we posed the following questions. Is the total score unidimensional? Does the questionnaire have a two-factor structure? Or is there a third factor representing a social dimension? And if so, which of the social factors proposed is the most appropriate?
Therefore, the objectives of the present study were: 1) to conduct a validation study of the MLHFQ, analyzing the internal structure using both CTT and item response theory (IRT); 2) to compare different factor structures proposed by other authors; and 3) to assess other psychometric properties including known-groups validity, convergent validity, and reliability of the different social factors proposed.
Discussion
The results of the current prospective study with a large cohort of patients hospitalized for HF at different hospitals support the validity and reliability of the MLHFQ, and most importantly, support the unidimensionality of the MLHFQ total score and the existence of a third factor, a social dimension, with good psychometric properties. To the best of our knowledge, this is the first study that compares different MLHFQ factor structures; this approach is a strength of the research in that it helps us to explore whether the original MLHFQ factor structure is valid, and to assess which of the different social factors proposed is the most appropriate.
Another strength is that we have conducted a complete study of the structural validity, using both confirmatory techniques of CTT, such as CFA, and IRT-based Rasch analysis. Most studies have assessed the structural validity of this questionnaire from the perspective of CTT [
4,
7,
16‐
19,
21,
22], and more specifically, using EFA rather than CFA techniques. Once an instrument has been translated into another language and culturally adapted for the target population, its structure should be confirmed by CFA. We only found two studies in which CFA was conducted [
15,
23], one of them using a sample of just 50 patients [
23], and we only found one study on the structural validity of the instrument combining both CTT- and IRT-based methods [
18].
Regarding two-factor structures, reviewing MLHFQ validation studies, we identified several problems and weaknesses. Specifically, several authors have questioned the factor structure of the questionnaire [
2,
5,
7,
11,
14‐
20]. Our CFA results indicate that the original structure of the questionnaire does have adequate structural validity. Considering the results of Rasch analysis for the physical factor, we found only item 6 to be misfitting. Munyombwe et al. [
18], in the only study in which an IRT model is applied to the questionnaire, did not find this item to be problematic in the Rasch analysis, but unlike us, they detected DIF by sex in item 3. None of the other studies that proposed different factor structures [
4,
7,
15,
16,
18,
21] drop item 6 from the physical factor. Further, taking into account the satisfactory results obtained from the rest of the Rasch analysis and the satisfactory CFA results, we do not consider that the identification of this item as misfitting is sufficient reason to conclude that this item should be excluded from the physical dimension. Regarding the emotional MLHFQ dimension, the fit indices from Rasch method support unidimensionality and provide strong evidence of construct validity. Munyombwe et al. [
18] also found satisfactory results in the Rasch analysis applied to this dimension.
Concerning different factor structures that have been proposed [
4,
7,
15,
16,
18,
21], in general, there is consensus about the emotional factor, all but one study agreeing on the constituent items [
21]. The largest discrepancies are related to the items that make up the physical factor, and the fact that three-factor structures have emerged in some studies, the new factor corresponding to a social dimension [
4,
15,
16,
18,
21]. In relation to the two-factor structures considered, Heo et al. [
7] proposed a physical factor which includes the same items as the original developers and adds two more items, item 1 and item 9, maintaining the same emotional factor. Other authors have also proposed that item 1 be included in the physical factor [
4,
15,
16,
18]; however, when comparing our CFA results for Heo’s model with those for the original model [
7], we obtain slightly better results for the latter. Hence, we rule out Heo’s model as an alternative to the original.
In relation to the physical factor suggested by other authors (Appendix
1), Ho and Moon both proposed a factor with somewhat larger discrepancies with the original. Further, we found the worst CFA results for these two proposals. Among the other structures, the composition of the physical factor differs only in one or two items. However, as noted previously [
15], the modification of an instrument is not easy. Besides, in the case of this questionnaire, the new structures that have been proposed are generally obtained from EFA and not CFA [
15], and on the other hand, the widespread use of the questionnaire means that changes would be difficult to implement and would also hinder comparability with existing data. Consequently, and considering that the results from both CTT and IRT for this factor were satisfactory, we see no need to establish a different composition for the physical factor.
With respect to a potential third factor, representing a social dimension, adding a third factor would not be as complicated as changing the composition of existing factors, since it would not involve any change to what was established by the original developers [
13] or affect comparability with other studies. However, it is important to reach a consensus on which of the different social factors proposed is the most appropriate and has the best psychometric properties [
4,
15,
16,
18,
21]. Although several authors have proposed such a third factor, none of them have studied the properties of the factor from the perspective of IRT, or using confirmatory techniques. In our analysis, the Ho and Moon social factors were considered inadequate, having fit indices below the minimum required, and obtained the highest AIC values. Furthermore, they included items of the physical factor proposed by the original developers in their social factor, implying a complete change of structure. The remaining proposals for a social factor only disagree on a few items. All of them considered items 8, 9 and 10; Garin also included item 15; and Munyombwe, besides item 15, includes items 14 and 16. Regarding the results of the CFA, the lowest AIC value was obtained for Garin’s factor. However, to compare the three social factors, it is also necessary to consider the IRT results, because in the CFA we are analyzing the complete structure of the questionnaire and not just the social factor. Rasch analysis results are satisfactory for all three structures, although Munyombwe’s model is the only one that met all the requirements to be considered an acceptable model. Further, regarding convergent validity, known-groups validity, and reliability, the best results were found for Munyombwe’s social dimension, with the highest correlation coefficients with the SF-12 components, the highest ES in known-groups validity, and the highest Cronbach’s alpha coefficient.
Lastly, the MLHFQ is commonly used to generate a total score, which assumes that the total scale is unidimensional. However, we found only two previous studies [
15,
18] that had explored the existence of a single factor, and they differ in their conclusions. The first one [
15] applied CFA within a bifactor model and the results confirmed the unidimensionality of the total score. The other study [
18] applied Rasch analysis to study the dimensionality of the total factor, and authors concluded that there were some misfitting items, namely, items 7, 8, 10, 14 and 15. They also found DIF by age in items 1 and 8, and by sex in item 3. Regarding misfitting items, Heo et al. [
7] concluded that items 8, 10, 14, 15 and 16 were problematic. Another study [
44] also stated that items 8, 10 and 15 were problematic, since they were not applicable to all patients. In our case, we only found two items to be markedly misfitting, items 8 and 10, with infit and outfit values well above the threshold. In item 1, we also found some degree of misfit with an outfit of 1.82, and in item 10, DIF was detected by sex and age, men and younger patients finding this item more difficult than women and older patients. Therefore, we confirm the existence of some problematic items in the composition of the total score, but unlike some previous authors [
18], we did not detect problems in the functioning of the rating scale categories. As Munyombwe et al. [
18] stated, the fact that there are misfitting items does not necessarily imply the need to remove them from the questionnaire, above all when these items would be included in the social factor. Considering a third factor, most of the 21 items would be included in a factor, and hence considering factor scores for the three-factor structure could be an alternative to the total score.
This study has some limitations that should be taken into account. The sample is composed of patients in Spain and we used the Spanish version of the questionnaire, and hence the results may not be generalizable to other populations or other language versions. Moreover, besides having to be valid and reliable, an instrument must also be responsive to changes to be useful. To the best of our knowledge, although there are some studies on the responsiveness of physical, emotional and total scores, the responsiveness of the social factor proposed has not yet been explored.
Conclusions
In conclusion, this comprehensive validation process, which used a large patient sample and combined classical and contemporary methods, supports the validity of MLHFQ physical and emotional subscales in patients with HF, showing good properties from both CTT- and IRT-based perspectives. In addition, the results confirmed the existence of a third factor, and we recommend the use of Munyombwe’s social factor, since it has good psychometric properties, the best among the social factors proposed. On the other hand, we found some problematic items within the total score, implying that it should be used with caution. Moreover, given the validity of the social factor, 19 of the 21 items would be included in a factor, and consequently, factor scores for the three-factor structure could be an alternative to the total factor. In conclusion, this study provides strong evidence that the MLHFQ is useful for measuring HRQoL in patients with HF, and it can be used both in clinical practice and research.
Ethics, consent and permissions
The study was approved by each corresponding institutional review board. All patients were given a letter informing them about the study and asking for their voluntary participation.
Competing interests
The authors declare that they have no competing interests.
Authors’ contributions
AB has participated in the conception and design of the study, in the analysis and interpretation of data, and has been involved in drafting the manuscript; AE, LGP, GN and RQ have participated in the conception, design, acquisition of data, and coordination of the study, have helped to draft the manuscript and have been involved in revising it critically for important content. All authors have read and approved the final manuscript.