Introduction
Fiberoptic endoscopic evaluation of swallowing (FEES) is a reliable tool that allows the dysphagia professional to evaluate the pharyngeal phase of swallowing [
1]. FEES is well tolerated, easily repeatable, and can be performed at the bedside [
1]. During a standardized FEES examination, patients swallow a sequence of boluses of different consistencies in consecutive order [
2,
3]. Usually visuoperceptual ordinal-scale variables are applied to judge FEES images.
FEES data on dysphagic patients are highly heterogeneous, both within and between subjects. This variability is partly due to the diversity of the dysphagic population, reflecting different etiologies of the swallowing dysfunction [Parkinson’s disease (PD), stroke, myopathies, head and neck cancer (HNC), etc.]. However, even within a single etiological group, variability can be substantial. For instance, PD, which is characterized by progressive neuro-degeneration, covers a large group of patients with different levels of disease severity [
2]. Therefore, acknowledging potential reasons for the high variability in FEES response (as recorded by repeated measures over the seven swallows) calls for a different analytical approach, one that capitalizes on the heterogeneity of these responses. In this respect, an alternative approach may shed new light on the dynamics of patients’ swallowing capabilities. The alternative presented here allows discernible patterns of swallowing courses to be extracted from the FEES responses.
Group-based trajectory modeling (GBTM) was developed to identify groups of individuals following a similar progression of a certain behavior in a longitudinal setting. This model-based clustering method is often referred to as a
person-
centered approach [
4‐
6]. It enables researchers to understand how life-course experiences unfold at the individual level and to cluster individuals who share similar developmental patterns. In GBTM, the population of interest is assumed to be heterogeneous—a mixture of groups. Shi et al. applied GBTM to describe the heterogeneity of symptom burden among patients with HNC and to identify subgroups with distinct symptom development trajectories [
7]. Treatment-related symptom burden varies significantly among patients undergoing radiotherapy or chemoradiotherapy, yet such variation is typically not reflected in the results from single-group studies. A two-group GBTM model identified 68 % of patients as having high symptom burden, associated with older age, worse baseline performance status, and chemoradiotherapy treatment [
7]. Another example of how GBTM has been used in the past is the study by Pines et al. [
8]. They described sexual risk trajectories among HIV-negative men who have sex with men. Three sexual risk trajectory groups were identified: low-risk, moderate-risk, and high-risk sexual behavior. The trajectories were significantly associated with earning an income, distress/depression symptoms, and substance use [
8].
However, a FEES examination protocol is of short duration, representing only a snapshot of a patient’s swallowing functional state. Therefore, GBTM is used here primarily for exploratory purposes. The main objective was to identify subgroups of patients with qualitatively distinct responses over the seven swallowing trials. This would allow us to describe the swallowing trajectories’ level, shape, and prevalence and then link these features to the etiology of dysphagia. To our knowledge, GBTM has not yet been considered for analysis of FEES data.
Discussion
Heterogeneity of FEES outcomes makes it difficult for practitioners to pinpoint and characterize regular or irregular patterns of swallowing behavior in population-based studies. There is great demand for evidence-based data that could enhance the clinicians’ interpretation, justify generalizations, and support potential clinical decisions.
This study confirms the hypothesis that GBTM can identify (latent) subgroups of developmental courses for measured FEES variables. Its application uncovered the variability of onset and demonstrated the reactive developmental changes as a function of the swallow trials. This information, in turn, facilitated the recognition of typical or atypical swallowing behaviors, allowing an estimation of their prevalence. The link of the trajectories to the etiology of oropharyngeal dysphagia was found to be significant for most FEES variables. For instance, DM1 patients tend to fall in the higher, impaired trajectories for the variables postswallow vallecular pooling and postswallow pyriform sinus pooling, whereas the opposite applies to the variables piecemeal deglutition and penetration-aspiration [
3]. This finding is consistent with a previous study showing more severe impairment for postswallow vallecular pooling when swallowing thick liquid compared with thin. DM1 patients performed better (i.e., had less impaired FEES outcomes) with the solid bolus consistency (bite-sized cracker) than with the other consistencies for the variable postswallow vallecular pooling [
3].
HNC patients, by contrast, fell in less impaired trajectories for the variables postswallow vallecular pooling and postswallow pyriform sinus. They had a higher risk of being assigned to top, impaired trajectories for the variables piecemeal deglutition and penetration-aspiration compared to DM1 patients [
3,
13]. This makes sense, as (chemo) radiation therapy for HNC can cause sensory impairment, xerostomia, and fibrosis of the upper aerodigestive tract, any of which could increase the risk of aspiration and oral residue [
14].
Furthermore, PD patients fell in higher impaired trajectories for the variable delayed initiation of the pharyngeal reflex [
2]. That finding was not unexpected; similar results have been observed in previous studies using this variable. [
2] Various motor disorders of PD have considerable influence on swallowing. For example, disturbed motility in the oral phase of swallowing is characteristic of PD [
15,
16]. According to Nilsson et al., the prolongation of the oral-pharyngeal transit time is likely to reflect dysfunction caused by rigidity, bradykinesia (slowness of movement), and hypokinesia [
17]. This explanation concurs with the higher impaired trajectories we found for the variable delayed initiation of the pharyngeal reflex in PD. For the other measured FEES variables, PD patients were diversely distributed over the trajectories.
The application of GBTM in an analysis of FEES data is unprecedented, calling for some caution when weighing any new insights it might yield. Under the present circumstances and given the limitations of this study, our conclusions are preliminary and merely indicative. Nonetheless, a few of the interpretations enabled by trajectory clustering are noteworthy, as set forth below.
First, our results lend credence to the hypothesis that to use the
‘one model fits all’ approach, namely to model population trends, leads to an oversimplification of how patients react to the swallow trials. Such techniques may obscure deviations from the norm that are not necessarily linked to known clinical variables. Moreover, analysis of FEES data is often static, focusing on each bolus separately. Note, however, that, if we were to consider each one individually, the detected latent groups would be no longer separable. In our study, trajectories overlapped to a greater or lesser extent during certain parts of the examination. It was only from the developmental perspective that the differences became apparent and here lies the strength of GBTM. The visualization of functional groups may hint at diversity in the nature of change. For instance, for postswallow vallecular pooling, both trajectory nos. 2 and 3 can be considered as lying at an intermediate level of impairment and as being less stable than the other two extremes (upper and bottom trajectories). However, nos. 2 and 3 describe opposite swallowing behaviors. While cluster 2 swallowed better with thin liquids, the transition to thick liquids and a bite-sized cracker was accompanied by a deterioration of the swallowing function. In cluster 3, by contrast, the deglutition of thick liquids and a bite-sized cracker induced substantial improvement in swallowing response of the patients. It is unclear whether these findings reflect different underlying pathophysiological mechanisms affecting the swallowing response (for instance, if trajectory no. 2 would be measuring the effect of fatigue or weakness in the patient or an effect of post radiation xerostomia instead of bolus consistency, contrary to no. 3) [
4,
13]. Trajectory no. 2 applied to a larger proportion of HNC patients (post radiation xerostomia). The improvement described by trajectory no. 3 was observed mainly among PD patients and concerned postswallow vallecular pooling with the ingestion of a bite-sized cracker. The DM1 patients fell mainly in the severe impairment trajectory no. 4.
Second, any observed changes in trajectories were most likely induced by the consistency, since points of inflection often coincided with the transitions in the bolus sequence. However, other explanations cannot be disregarded. For instance, the intermediary trajectory no. 2 of postswallow pyriform sinus pooling shows that patients’ swallowing function consistently deteriorated over the swallows, almost independently of the bolus consistencies. Apparently, the subjects fatigued quickly or suffered from severe post radiation xerostomia, leading to a more impaired swallowing function for thick liquid and a bite-sized cracker (mainly HNC patients in trajectory no. 2). Given these results, researchers might do well to adopt a more dynamic approach when analyzing and interpreting the various processes representing patients’ reactive changes to a swallowing challenge. It is only by taking a developmental perspective that patients’ dysfunctional behavior can be better discriminated and the pathophysiology of swallowing impairment properly understood.
The usefulness of GBTM for the analysis of FEES data goes beyond the visualization and characterization of the latent functional groups. Its impact greatly depends on the interpretational value of the trajectories and their theoretical plausibility. As argued here, GBTM provides an insightful depiction of swallowing behaviors. This conclusion needs to be explored further and replicated in larger-scale studies. Clinicians would then have a more detailed evaluation at their disposal, which they could use to support and guide their choices of rehabilitation programs or interventions.
Limitations of the Study
The present prospective study has some methodological limitations. First, its relatively small sample size posed a major constraint on data analysis with GBTM, as this technique requires larger samples to properly extract latent clusters. Nonetheless, the models converged without difficulty and the indices of model adequacy remained within acceptable bounds. For the analysis of three-point-scale ordinal FEES variables, dichotomization of the measured data may have over-simplified the overall swallow functioning of the patients due to loss of information. Another limitation concerns a well-known misunderstanding linked to GBTM analysis, the fallacy of reification. It occurs once latent trajectories are interpreted as real distinct entities. This should be avoided. The intent behind using GBTM was to describe the heterogeneity of FEES outcomes in such a way as to facilitate their clinical interpretation. The identified trajectories are not meant to represent definite classes of swallowing capabilities. The applied FEES protocol in the current study is the standardized protocol we use in daily clinical practice for many years. However, another FEES protocol might produce different results in capturing swallowing trajectories. Finally, a potential drawback of this study is that healthy controls were not included. However, the observed developmental trajectories as described above revealed the presence of clinically relevant subgroups of dysphagic patients in the study population.