Fatigue and anxiety mediate the effect of dyspnea on quality of life in amyotrophic lateral sclerosis

Abstract Introduction: Dyspnea (or breathlessness) due to progressive neuromuscular respiratory failure is common in amyotrophic lateral sclerosis (ALS). It is associated with anxiety, depression and reduced quality of life (QoL). For effective treatment, it is essential to understand the relationships between dyspnea, anxiety, depression and QoL. Methods: The UK Trajectories of Outcomes in Neurological Conditions-ALS study (TONiC-ALS) collected self-report measures from patients with ALS. Ordinal scales were transformed to interval-scaled estimates by the Rasch Measurement model. They were subsequently included in a series of path models where the focal relationships were dyspnea to QoL and dyspnea to depression. Results: Path analyses using 1022 participants showed that 60.5% of the variance of QoL was explained by fatigue, anxiety, dyspnea and disability. For depression, 54.1% of the variance was explained by a model of these factors. Dyspnea played an important but mostly indirect role in influencing QoL and depressive symptoms. Disability was dominated by all other factors in the model. Discussion: Dyspnea in ALS influences quality of life and depression largely through indirect effects, principally acting via anxiety and fatigue. Recognition of this is essential for clinicians to understand where to intervene for greatest benefit. Researchers must be aware that studies of the effect of dyspnea on QoL and depression require path models, measuring both direct and indirect effects, as the impact of dyspnea is likely to be significantly miscalculated if only direct effects are assessed.


Introduction
Amyotrophic lateral sclerosis (ALS) is characterized by progressive dysfunction of motor neurons in the motor cortex, brainstem and spinal cord anterior horns. Most patients present with either focal limb or bulbar muscle weakness, which spreads at a variable rate to other body regions, including thoracic muscles and diaphragm (1). Respiratory failure develops insidiously and may present with dyspnea (breathlessness), fatigue and other symptoms associated with hypoventilation. Median survival in ALS/MND is 30 months from symptom onset, with most deaths associated with type II respiratory failure (2).
Dyspnea is a distressing aspect of the trajectory of symptoms in ALS and correlates significantly at a weak to moderate level with emotional wellbeing (3). Management of nocturnal hypoventilation using noninvasive ventilation has been shown to improve dyspnea and quality of life (QoL), as Supplemental  well as survival, in a randomized controlled trial (4). Interventions improving dyspnea, such as noninvasive ventilation, have been shown to also relieve fatigue (4)(5)(6). Recent work on the impact of dyspnea on QoL in ALS reported "dyspnea as a major driver of emotional distress" measured by the anxiety domain of the Hospital Anxiety and Depression Scale (7). In a model reporting data from 636 people with ALS, anxiety was the second greatest influence on QoL after disability, with both dyspnea and fatigue affecting the causal pathway (8). It has been shown that anxiety affects the psychological domains of QoL in ALS (9). Earlier work showed that as disability worsens, QoL deteriorates (10). These findings highlight the importance of a model exploring how dyspnea, fatigue, disability and anxiety inter-relate to affect QoL in ALS, in order to identify targets for most effective clinical intervention.
Longitudinal research in the general population has shown that the onset of dyspnea significantly increases the risk of developing depression and anxiety (11). Qualitative studies of dyspnea among people living with ALS highlight its psychological impact, because dyspnea was "a constant reminder that the illness was a dangerous threat to patients' lives", but there has been little investigation of the association between dyspnea and depression (12). It has been shown that depression is associated with increased disability in ALS (13,14). Depression is also often associated with anxiety (15). The aim of the current study is to determine the relationships between dyspnea and QoL, and dyspnea and depression, in ALS, considering other aspects such as fatigue, anxiety and disability.

Patient recruitment and study conduct
The Trajectories of Outcomes in Neurological Conditions (TONiC) collaboration, an ongoing longitudinal study in the UK, recruited patients with ALS across many centers and asked them to complete a questionnaire pack containing a variety of patient reported outcome measures (PROMs). All participants received written information, and informed consent was obtained prior to enrollment into the study. Ethical approval was granted from the relevant local research committees (reference 11/NW/0743).
Eligibility criteria included adults who had been diagnosed with ALS according to El Escorial World Federation of Neurology criteria for the diagnosis of ALS, and who were capable of informed consent and of answering questionnaires (16). Cases with a family history of ALS were eligible as were patients with only lower motor neuron (LMN) signs in two or more regions, or with progressive primary lateral sclerosis without spinal LMN signs, provided a consultant neurologist specializing in ALS had confirmed the diagnosis.
Data were collected in sequential phases via a multi-stage consent process: Phase 2: Registry involved consent for recording of demographic and clinical data from the patient's medical records, such as onset type (limb, bulbar, respiratory) for people with physician-verified ALS. Registered participants could then consent to Phase 3: Baseline and complete a paper questionnaire pack of PROMs along with additional demographic and health economic data (for additional information on PROMS, see Supplementary Appendices). Data from those who did not consent beyond the registry were used to assess "non-responder" bias. The pack could be completed at home and returned by post. Participants with writing difficulty were allowed to have the help of a friend or family member to complete the packs, provided that individual acted only as a scribe, marking the answers chosen by the patient. Consent could be taken verbally in the presence of an independent witness if the participant could not sign.
The majority of participants joined following invitation during a clinical contact, typically a multidisciplinary clinic visit. Some people with ALS contacted the research team having heard about the study through patient support organizations or the internet. Such subjects were able to complete a self-referral form with basic clinical information and contact details for a healthcare professional, to enable the collection of data from the medical records. The study benefited from widespread support from the patient organization (MND Association), UK MND Clinical Studies Group and NIHR Clinical Research Network as well as many ALS and palliative care teams.
Participants who returned a completed baseline questionnaire pack could consent to Phase 4: Longitudinal follow-up study. The follow-up packs had the same questionnaires, and a change questionnaire with the stem "Since the last time I completed the questionnaire … ". This was followed by a series of items asking about change in various factors like spasticity, coping, pain, overall health, disability, quality of life, etc. assessed on a three point Likert scale of Better/Same/Worse.
Follow-up packs were sent by post to participants at approximately the following intervals: 4, 9, 14, 18, and 27 months ± 2-month window for patient convenience; each pack was preceded by a contact to see if the participant still wanted to continue in study. The start of trajectories analysis was based on date of diagnosis as pilot work showed imprecision in date of onset, whether collected from medical records or patient recall. Only baseline data is used in this analysis. A schematic summarizing the study and the patient PROMs is shown in Figure 1.
The following questionnaires from the pack were used for the path analysis:-1. Dyspnea-12, to measure breathlessnessscored 0-36 where a high score represents extreme breathlessness (17). 2. Neurological Fatigue Index-MND (NFI-MND), to measure fatigueeight item summary scale scored 0-24 with a high score represents greater fatigue (18).

Amyotrophic Lateral Sclerosis Functional
Rating Scale-Revised limb domain (ALSFRS-R_L), as a measure of disability -six items asking about self-care and mobility from the original 12, confirmed as being a separate domain by earlier Rasch analysis (19). With a range of 0-24 for this domain, higher scores indicate less disability.

Anxiety subscale from the modified Hospital
Anxiety and Depression scale (M-HADS-A), to measure anxiety -six items scored 0-18 where a high score represents extreme anxiety, with cut points of 7 & 9 for "possible" and "probable" anxiety (20,21).

Depression subscale from the modified Hospital
Anxiety and Depression scale (M-HADS-D), to measure depression -six items scored 0-18 where a high score represents extreme depression, with cut points of 5 and 8 for "possible" and "probable" depression (20,21).
6. World Health Organization Quality of Life-Bref (WHOQoL-Bref), to measure QoLtotal score (range 0-96) where a high score represents a good QoL (22). The intervalscaled estimate for the total score is obtained through a bi-factor equivalent solution within the Rasch measurement framework.

Rasch analysis
Data from each of the scales subsequently used in the path analysis were fitted to the Rasch polytomous model in a calibration sample of 900 cases taken from three time points in the ongoing longitudinal study (23,24). The parameters from this sample were then exported into the full data set to generate person estimates on the various PROMs.
Only the baseline data are used in the current cross-sectional analysis. Ideal fit values to the Rasch model are shown at the bottom of Table 1. In addition, local response dependency was assessed based upon residual correlations of 0.2 above the average (25). Where this was present, post hoc "super items" were created (26). Differential Item Functioning (DIF) was tested for onset type, age and gender. Should DIF be detected, the difference between adjusted (splitting for DIF) estimates and unadjusted estimates is compared and the effect size reported. An effect size of this difference >0.1 is considered to indicate substantive DIF (27). Reliability is reported both as a Person Separation Index (PSI) and as a Cronbach's Alpha. When the data are normally distributed these produce a similar value, but otherwise, especially with floor and ceiling effects, they diverge. Further detail on Rasch analysis is provided in Supplemetary appendices.

Path analysis
Path analysis is a development of multiple regression which provides estimates of hypothesized relationships between sets of variables. Commonly this is presented in diagrammatic form. In the current study, using Rasch-transformed interval level latent estimates, the focal relationship (primary hypothesis) is that between dyspnea and QoL, followed by dyspnea and depression, hypothesizing that the worse the dyspnea, the worse the QoL and depressive symptoms.
The models are operationalized through single indicator latent variables, the indicators being the estimates derived from the above Rasch analysis. Regression weights and errors for these estimates are pre-specified (28). The overall sample is randomized into two sets: a development sample and a validation sample for replication. The path model fit is determined by a nonsignificant v 2 (p > 0.01), and approximate fit statistics which include the Root Mean Square Error of Approximation (RMSEA) which should be below 0.6, and the Confirmatory and Tucker-Lewis Fit Indices, both of which should be above 0.95 (29). The influence of dyspnea upon QoL, and subsequently depression, is also ascertained when fatigue is added as an additional mediator. This conceptualization is based upon the Wilson and Cleary model with a pathway from symptoms to QoL (30). Further detail on the underlying conceptual model is provided in appendices.

Patients
This analysis uses data from 1022 patients with ALS who consented and returned a completed baseline questionnaire between November 2013 and the end of 2019. Mean age was 64.9 years (SD 10.6), 60.4% were male, and median duration since diagnosis was 9 months (IQR 3.76-22.9; range 0.296-295.7 months). Mean ALSFRS-R score was 25.7 (SD 7.2). The majority had limb onset (70.5%), a quarter had bulbar onset (27.3%), and only 2.2% were recorded as having a respiratory onset. Most were married (77.7%), and two-fifths were able to complete the questionnaire independently (61%). Just 15.4% were employed, either full time or part time. Over a quarter (27.9%) were indicated to be at least 'possible' anxiety on the modified M-HADS-A, and 30.6% were indicated to be at least 'possible' depression. No significant difference was observed between onset type and the level of anxiety (v 2 3.42;df(4) p ¼ 0.489) or depression (v 2 2.32; df(4) p ¼ 0.677). Comparison of the development and validation samples showed no difference between samples for age, duration, onset type or gender (v 2 p > 0.01).

Rasch analysis
In the calibration sample, data from the Dyspnea-12 scale were fitted to the Rasch model. Initial fit was poor (Table 1: Analysis 1). However, the hierarchical ordering of items was consistent with the previously observed findings that the physical items (1-7) would show early impact (i.e., affirmed first), and the psychological items, later impact. The item "I feel short of breath" was the most likely item to be affirmed, whereas the item "My breathing makes me feel miserable" the least likely, representing the upper level of impact. There was no differential item functioning. However, a substantial breach of the local independence assumption was observed, with dependent items clustering within the domains of physical (items 1-7) and mood aspects (items 8-12). Grouping alternating items (i.e., 1,3,5 etc.) into two "super items" to absorb the local dependency and reflect the fact that items are summated across the two domains, showed adequate fit to the model and unidimensionality (Table1: Analysis 2). Fit of all other scales were found to be satisfactory, ( Table 1: Analyses 3-7).

Path analysis
Dyspnea and QoL. The first model hypothesized that dyspnea influences QoL both directly and also indirectly, mediated by anxiety (for illustration, see Supplementary Appendix Figure 2). The model was satisfactory (v 2 1.433 (df2); p ¼ 0.488); ancillary fit statistics indicated perfect fit (Figure 2). At this initial stage, just 34.9% of the variance of QoL was explained by the model (R 2 ). As dyspnea increases, so does anxiety, and both diminish QoL. In this simple model, the magnitudes of direct and indirect effects of dyspnea upon QoL were similar ( When fatigue was added to the model, the dynamics changed (Table 2: Model 2; Figure 3 in Supplementary Appendices). The model was accurately supported by the data (v 2 7.06 (df4); p ¼ 0.133; RMSEA 0.039; CFI 0.994; TLO 0.984) and fully replicated on the validation sample (v 2 7.330 (df4); p ¼ 0.119; RMSEA 0.040; CFI 0.994; TLI 0.98; R 2 0.60). The model paths were invariant by onset type (bulbar vs limb), gender and duration (grouped by quartiles) (Wald test >0.01). Now 53.2% of the variance of QoL was explained by the model (R 2 ). Fatigue was shown to have the strongest influence upon QoL, followed by anxiety and then dyspnea. The effects of dyspnea became fully indirect.
A third model sought to determine if dyspnea retained its influence when disability was added to  the hypothesized pathway between dyspnea and QoL. Thus, Figure 3 adds the limb domain from the ALSFRS-R, essentially a set of Activities of Daily Living covering 6 items on self-care and mobility, and so in this model representing disability. This model was also satisfactory (v 2 5.092 (df5); p ¼ 0.405; RMSEA 0.006; CFI 1.000; TLI 1.000). The model replicated (v 2 4.901 (df5); p ¼ 0.428; RMSEA 0.000; CFI 1.000; TLI 1.000; R 2 0.67). All parameters were invariant by gender, duration, and onset type (Wald test >0.01). Now 60.5% of the variance of QoL was explained by the model, and once again the effects of dyspnea were fully indirect. Fatigue retains the most powerful influence on QoL followed by anxiety, dyspnea and disability ( Dyspnea and depression. Considering the focal relationship between dyspnea and depression, Figure 4 hypothesizes that symptoms such as dyspnea and fatigue, as well as disability, contribute to the level of depression. The data accord with the hypothesized model (v 2 Figure  4 in Supplementary Appendices). The model fit was satisfactory (v 2 6.91 (df4); p ¼ 0.141; RMSEA 0.038; CFI 0.995; TLI 0.982) and replicated (v 2 8.77 (df4); p ¼ 0.067; RMSEA 0.048; CFI 0.993; TLI 0.973; R 2 0.59). Once again there was some variation in the path coefficients across gender. For onset type, the path fatigue ! disability lacked invariance; it was significant in both groups, but with a much higher coefficient in bulbar than in limb onset. The model explained 54.1% of the variation in depression. Anxiety, then fatigue, followed by dyspnea dominated the model (Table 2: Model 5). Disability had a small effect. As before, most of the effect of dyspnea was indirect.

Discussion
In ALS, QoL reflects a complex interplay of many factors such as disease characteristics, symptoms and activity limitations (31)(32)(33)(34)(35). Analysis of the relationships between these factors requires conceptual models, such as the widely cited conceptual framework of the influence of symptoms and functioning upon QoL from Wilson and Cleary (30,36,37). The current study uses the Wilson and Cleary model to test the impact of dyspnea upon QoL, and upon depression.
No prior study has tested an integrative model of the effect of dyspnea upon QoL along with physical impacts of disability and fatigue as well as psychological influences, such as anxiety, in an extensive sample of more than 1000 people living with ALS. We found that the role of dyspnea was 'behind the scenes', via indirect effects through anxiety and fatigue. The influence of dyspnea on anxiety is likely to be multifactorial. Previous work highlights that dyspnea involves more than feedback from the lungs to sensory cortex (38,39). ALS is known to be associated with cortical and brain stem pathology, including prefrontal cortex; in COPD functional imaging shows that dyspnearelated cues engage the prefrontal cortex (40). Acute and sustained dyspnea have been shown to activate regions of the anterior insula, anterior cingulate and prefrontal cortices (41). There is increasing attention on the compensatory, as well as pathological, changes in respiratory motor neurons in ALS but limited understanding of their drivers (42,43).
While this study supports previous evidence that dyspnea influences fatigue, in the current model only 12% of the variation in fatigue was explained by dyspnea, indicating that further work is required to understand factors influencing fatigue in ALS (6). Nevertheless, the treatment of dyspnea has been reported to have the potential for synergies in symptom management arising from the association between fatigue and dyspnea (44).
Adding disability to the model saw only modest improvement in the explained variance of QoL, from 53.2% to 60.5%, further emphasizing the dominant role of symptoms and psychological factors over disability. Disability also had modest influence on depression, assessed using the Modified HADS which omits the item showing misfit ("I feel as though I am slowed down"), which is likely influenced by paralysis due to motor neuron loss (21). Depression was strongly affected by anxiety, followed by fatigue. As with QoL, the effect of dyspnea was mostly indirect, through anxiety and fatigue.
There are limitations to the study. For example, low numbers in the respiratory onset group precluded inclusion in the invariance testing for onset type (the model would not converge with such low numbers). Likewise, the large number of centers involved with the collection of data resulted in small numbers for some centers, which precluded the testing of invariance across centers. There was considerable attrition in the study, in that while 1700 people had given consent 39% did not proceed to complete the baseline questionnaire. However, there was no significant difference in onset type, nor for the duration of ALS between those that dropped out of the study between the initial consent, and those proceeding to subsequent completion of the baseline questionnaire. Also, the models were not adjusted for whether participants were treated with antidepressants or anti-anxiety drugs, or used NIV for dyspnea. The models which assessed disability used the limb domain of the ALSFRS-R. However, more than a quarter of the sample had bulbar onset, and this should be considered in future studies. The specification of the model in the current study is just one of many that could be considered. Future work could consider these relationships more fully, including the potential for a two-way relationship (i.e., a nonrecursive model).
There are also strengths to the study. The conceptual basis of the approach gives a strong framework for analysis. The derivation of key factors to be measured from in-depth work with those with ALS gives a strong ecological framework to what is included in the model. The sample size is large for ALS and the transformation of ordinal scales via Rasch analysis also allows for access to parametric statistics, as well as an elegant single-indicator exposition of a path model. Randomization of the sample into two groups to allow for a validation sample for replication is also a strength.
Dyspnea is an important factor in ALS; path analysis shows it has a fundamental influence on QoL and on depression, mediated through anxiety and fatigue. These clinical targets of dyspnea, anxiety, and fatigue all have greater effects than disability for both QoL and depression. This study suggests that deteriorating QoL and increased depressive symptoms are not untreatable consequences of the progressive disability of ALS. Interventions to improve QoL and depression in ALS should include treatments for anxiety and fatigue as well as appropriate management of respiratory symptoms.