This cluster analysis, which was performed in the middle of Covid-19 pandemic among 1109 elderly, retired subjects with current cough, validates our previous cluster analysis among working-age, employed subjects . Again, two clusters were found. The cluster B, consisting of 24.8% of the subjects, was especially characterized by several triggers of cough, many cough background disorders, and poor cough-related quality of life. These features fit to the cough phenotype TBQ, which was identified in our previous study. The cluster A, lacking these features, may be called as the ‘common’ cough phenotype.
Clustering is a hypothesis-generating method, with the assumption that the inherent patterns within the data may be a reflection of different underlying pathophysiological mechanisms or genetic basis, i.e., that the identified phenotypes may also represent distinct cough endotypes or genotypes . Both cluster analyses identified the number of reported cough triggers as the most important variable to separate the two phenotypes. Both analyses also identified the chemical triggers like paints, fumes, strong scents and automobile exhaust fumes, as the most important types of cough triggers . Several studies have shown that subjects with lower airway symptoms induced by chemical irritants are especially sensitive to the cough-provocation test with capsaicin [21–30]. Therefore, we hypothesize that the main underlying pathophysiological feature separating the phenotype TBQ from the common phenotype is hypersensitivity of the cough reflex arc. Thus, this phenotype might also represent a distinct endotype. To test this hypothesis, studies applying cough provocation tests to subjects with TBQ and the common phenotype should be carried out. To investigate whether the phenotype TBQ is a distinct cough genotype, studies applying genome analyses on well-characterized subjects with cough should be carried out. Of note, family history of chronic cough was far more common in phenotype TBQ than in the common phenotype.
Though the phenotype TBQ was more common in chronic cough than in acute and subacute cough, the length of the current cough episode was far from the most important variables defining the clusters. This was also the case in our previous cluster analysis . This finding speaks against the classification of cough according to the length of the episode, which is widely utilized in cough guidelines [3–11]. The phenotype TBQ was much more strongly associated with doctor’s visits due to cough and the use of cough medicines than the presence of chronic cough. Furthermore, our previous study showed a strong tendency for cough prolongation in the phenotype TBQ . Therefore, cough phenotype TBQ has clinical significance, and identification of it may serve as an indication for prompt and comprehensive clinical evaluation regardless of the duration of the cough episode.
Cough phenotype TBQ resembles the previously described concept ‘cough hypersensitivity syndrome’ . Both emphasize the enhanced cough response to various external triggers. However, there seems to be two major discrepancies between the entities. First, ‘cough hypersensitivity syndrome’ has been connected to chronic cough  but the present study shows that features of cough hypersensitivity can be present in acute and subacute cough as well. Second, it has been postulated that the ‘cough hypersensitivity syndrome’ is present in the majority of subjects with chronic cough [31–33] whereas in the present cluster analysis just 27.4% of subjects with chronic cough showed the features of the phenotype TBQ. These discrepancies may be best explained by the fact that ‘cough hypersensitivity syndrome’ has been described among subjects attending special cough clinics  whereas our cluster analyses are based on community-based populations. Given the documented high tendency of the subjects with the phenotype TBQ to seek medical attention, they are probably overrepresented in the population attending special cough clinics. ‘Cough hypersensitivity syndrome’ is mainly based on expert opinion . Despite the above-mentioned differences, it is highly interesting that unsupervised, data-driven analyses lead to similar conclusions to those made by experienced clinicians.
A recent study from Australia supports the present phenotype analysis. In that study, two clusters could be identified among adult subjects with various respiratory symptoms. The smaller cluster, including 32% of the subjects, was characterized by symptoms of laryngeal hypersensitivity and a strong cough response to mannitol . The characteristics of that cluster resemble those of the phenotype TBQ. To our knowledge, there are no cluster analyses in non-Caucasian cough populations.
For clinical purposes, we calculated the best cut-off values to separate the clusters. They were almost identical to those reported in our previous cluster analysis among working-age subjects . Presence of at least two of the three main TBQ determinants was found to be a very sensitive and modestly specific indicator of the cough phenotype TBQ. It gave the largest sum of sensitivity + specificity and is thus the most suitable clinical criterion for the cough phenotype TBQ. Reliable clinical demonstration of the phenotype TBQ requires that a comprehensive list of triggers is presented to the patient in a written form since patients often forget some triggers when asked openly. The list of the 15 triggers asked in the present study can be found in the questionnaire (Supplementary file).
There were slight differences in the questionnaires between the present and the previous study , which did not affect the main results. Others and we have shown that there are more cough background disorders in elderly than in younger subjects [35–38]. Therefore, bronchiectasis, pulmonary fibrosis, and obstructive sleep apnea were added to the variable ‘number of cough background disorders ‘. All six background disorders, which were utilized in this analysis, have been shown to be statistically significantly associated with cough in the present population . Questions about Covid-19 infection and vaccination, and symptoms of flu at the beginning of the current cough episode were added due to the current pandemic. The recurrence of cough was now investigated by asking how many cough episodes the subject had had in the last 12 months. The number of cough episodes was significantly higher in the phenotype TBQ than in the common phenotype. Of note, the recurrence of cough episode was more strongly associated with the TBQ phenotype than the length of the current cough episode.
The present study has several limitations. The participation rate was relatively low, which is typical for e-mail-based surveys. However, the age and gender distribution of the responders did not differ significantly from the original population, though females were slightly more willing to respond than males. High age of the population may have hindered the use of e-mail in some individuals. However, the participation rate in our previous survey among working-age subjects was of similar magnitude . It is possible that patients with severe cough have been more willing to participate than patients with mild cough. This may have led to an over-representation of the TBQ phenotype. The proportion of current smokers was small, which may have reduced the impact of smoking on the analysis. The prevalence of short, infection-associated cough subtypes was low in the present population, probably due to personal protective and social measures that were highly recommended during the pandemic era. Finally, the analysis was based on the questionnaire data only; spirometry, laboratory, and x-ray data were missing.
The strength of the present study included a large population, which was missing in our previous cluster analysis: Elderly, retired subjects. In addition, the geographic distribution of the present population was wider than that in our previous study . The survey was not directed to subjects complaining of cough but to a community-based population. Therefore, even those coughing subjects who would never complain their cough to doctors were also included. The questionnaire was originally planned and further completed to investigate cough and associated conditions, including also the highly validated LCQ and a comprehensive cough trigger list. The interindividual variation in how subjects recognise and report symptoms was controlled by the variable ‘symptom sum’. Furthermore, there were symptom questions that are recommended for epidemiologic studies, to define important background disorders of cough. All raw data plus the derived variables were included in the cluster analysis without hypotheses using prior knowledge. By this way, all relevant features of cough and even undiagnosed but symptomatic background disorders were equally considered in the cluster analysis.