An exploration of clinically meaningful definitions of cough bouts

Abstract

Rationale

The measurement of cough frequency is widely used in clinical trials, typically expressed as the number of explosive cough sounds per hour. However, this measure does not capture the clustering of coughs into bouts. Coughing bouts contribute to perceived cough severity and the physical complications of coughing, but an agreed standard definition of cough bouts is lacking. The objectives of the present study were to explore the impact of different definitions of cough bouts on the parameters generated, their relationships with reported cough severity and influence of age and gender in refractory chronic cough (RCC).

Methods

We analysed 24-h acoustic recordings and concurrent cough severity visual analogue scales from 91 RCC patients (62% female, median (interquartile range) age 60.0 (54–67.0) years). A custom-built algorithm calculated cough bouts, defined by the intervals between explosive cough sounds. Bouts defined by inter-cough intervals from ≤0.5 to ≤10 s (0.5 s increments) were explored, and parameters including number of bouts, median/maximum bout length and total bout duration calculated.

Measurements and main results

Using inter-cough intervals of >3 s to define cough bouts made little difference to cough bout parameters. Correlations between cough severity and bout parameters were weak but most likely to be significant when single coughs were removed. Cough-free time/total time spent coughing tended to have more influence on cough severity than the average cough bout length, irrespective of the interval used.

Conclusion

These analyses favour definitions of cough bouts utilising inter-cough intervals of ≤3 s and the exclusion of single coughs from cough bout analysis.

Shareable abstract

A standard method for defining cough bouts is lacking, but these data suggest that applying time intervals of ≤3 seconds and exclusion of single coughs may best reflect patient reported cough severity https://bit.ly/3RJVCpo

Introduction

Objective measurement of cough frequency by acoustic ambulatory monitoring is an established technique used in mechanistic studies and clinical trials evaluating novel treatments for chronic cough [1]. Cough frequency is usually quantified as the total number of explosive cough sounds over 24-h where each individual explosive phase in an audio recording is counted [2] (supplementary figure S1a). In respiratory disease, coughing often occurs as series of multiple explosive phases termed “peals”, “epochs”, “bursts”, “attacks”, “fits” or “bouts”, which may be initiated by a single inhalation or have several inhalations interspersed (supplementary figure S1b) [3]. Qualitative studies have reported that cough attacks, fits or bouts are important to patients with chronic cough and relate to their perception of cough severity [4–6]. Patients can have difficulty controlling prolonged bouts, which can cause physical discomfort and disruption to daily activities, thus affecting overall quality of life. Studies of the mechanics of voluntary coughing in healthy volunteers have demonstrated that during bouts of coughing, neuromuscular drive increases, intrathoracic and abdominal pressures remain high and significant haemodynamic shifts occur away from the trunk [7, 8]. This may explain why “intense cough attacks” are the most important feature of chronic cough for patients [6] and why these episodes might relate to physical complications of coughing, such as urinary incontinence, vomiting, breathlessness, feeling faint, or physical exhaustion [9].

Little attention has been given to studying temporal patterns of coughing and the extent to which coughs are dispersed or clustered together, yet two individuals may have exactly the same 24-h cough frequency but a completely different temporal distribution of cough events (figure 1). Investigating the nature of cough bouts (the number of times bouts are triggered and how long bouts last) may provide additional information about mechanisms driving cough, different patient phenotypes, and insights into perceived cough severity compared with cough frequency alone. A trial of a γ-aminobutyric acid receptor B agonist also suggested that certain medications may impact the number of cough bouts more significantly than individual cough sounds [10].

FIGURE 1.

An illustrative example of how the temporal distribution of cough events between two individual patients (A and B) can differ over a 24-h period despite having the same cough frequency (60 coughs). Vertical lines indicate single coughs. Cough events for patient B are more clustered together with longer latent periods.

One of the main limitations in studying cough bouts is the lack of an agreed and clinically relevant definition of a “bout” [11]. One approach would be to count the number of coughs occurring after an inhalation. However, inspirations are difficult to identify using sound recording alone and simultaneous measurement of breathing pattern is challenging in ambulatory subjects. An alternative approach is to apply an arbitrary time window to group adjacent explosive phases together according to their proximity. Most studies to date have used this technique and defined a cough bout as continuous coughing with pauses of no more than 2 s between explosive cough sounds [2, 10, 12]. However, recent pilot work suggested that a 3 s window may correlate better with subjective cough severity [13]. The aim of this study was to describe the impact of different cough bout definitions on the number of bouts, number of single coughs, bout length and total bout duration in 24-h acoustic recordings from patients with refractory chronic cough (RCC). We also assessed how these different definitions influenced the relationships between cough bout parameters and reported cough severity and the effects of age and gender. A proportion of the data from this study were previously reported in abstract form [13].

Methods

Data collection

Cough recording data used in this study were sourced from an ethically approved research database of 24-h acoustic cough recordings (RaDAR; Research Database of Ambulatory Acoustic Recordings; research ethic committee reference 23/NW/0256) which were adopted from previous cough monitoring studies with patient consent. On entry to the database, recordings were aurally and visually analysed in full by trained cough analysts who placed an electronic tag on the explosive portion of each individual cough sound (supplementary figure S1a). The tag positions of each cough were extracted as a list of time points and then converted to bouts by custom-written, validated software. All suitable baseline recordings from patients with RCC in the database were selected for analysis and all files were from distinct individuals. Recordings were 24-h in duration and collected using the VitaloJAK cough monitor (Vitalograph Ltd, Buckingham, UK).

Patient population

All patients were recruited from a specialist cough clinic and diagnosed with RCC, i.e. chronic cough of >8 weeks duration that persists despite addressing treatable traits [14]. Investigations and treatment trials were carried out according to relevant European Respiratory Society and American College of Chest Physicians guidelines at the time of each original study [15–17]. For all original studies, patients were aged ≥18 years with a normal chest radiograph, normal spirometry (forced expiratory volume in 1 s (FEV₁) and forced vital capacity (FVC) %predicted of >70%) and no recent respiratory tract infection. Those taking medications known to modulate cough such as angiotensin-converting enzyme inhibitors, along with current smokers and ex-smokers with >20 pack-year history, were excluded. Patients taking cough suppressants, e.g. opioids, were included if the medication was discontinued for the study duration following a suitable pre-study washout period.

At the end of the cough recording, all participants completed a visual analogue scale (VAS) to assess cough severity; a 100 mm horizontal line ranging from “no cough” on the far left to “worst cough” on the far right to indicate their perception of cough severity during the day and overnight. No data were available on patient co-morbidities, body mass index or activity.

Cough bout analysis software

A custom-built algorithm was developed to automatically convert text lists of 24-h cough tag positions into cough bouts (for validation, see supplementary material). The software used a window-based approach to group coughs together into bouts depending on their proximity to one another. A predetermined range of fixed time windows were applied to the intervals between the tag positions, i.e. inter-cough intervals, then all adjacent coughs within that time window were grouped (example in figure 2). A total of 20 inter-cough intervals at half-second increments from 0.5 s up to 10 s were tested.

FIGURE 2.

Illustrative example of bout software analysis. Fixed time window or inter-cough interval (2.5 s) is applied to 24-h tagged cough positions and coughs located less than 2.5 s apart in proximity are grouped together to form one bout of eight coughs. Each cough is marked with red dashed line. The start and end times of the bout are recorded; labelled here as “bout 1 start” and “bout 1 end”.

A bout of coughing consisted of two or more coughs depending on the positioning of coughs within the specified inter-cough interval, i.e. bouts of two coughs, three coughs, four coughs and so on. Isolated coughs with no adjacent coughs within the specified interval were classified as “singles”. Based on previous data, a fixed 0.326 s was allocated for the duration of each single cough and to each bout to account for the length of the final cough in the bout [18].

Cough bout parameters

For each inter-cough interval, the following parameters were calculated:

• 1) Number of single coughs

• 2) Total number of bouts with/without singles

• 3) Bout length as number of coughs with/without singles

• 4) Bout length as time (s) with/without singles

• 5) Maximum bout length as number of coughs

• 6) Maximum bout length in time (s)

• 7) Total bout duration, i.e. amount of time spent coughing in bouts (min) with/without singles

• 8) Cough free time (min) with/without singles

Statistical analysis

Data were summarised as medians and interquartile ranges (IQRs) and groups compared using Mann–Whitney U-tests. Spearman's rank correlation coefficients assessed the relationships between cough bout parameters, cough severity VAS and age. All analyses were performed with and without single coughs and were carried out using SPSS statistics for Windows (version 29, IBM) or R V4.3.0 (R Core Team (2023); R: A Language and Environment for Statistical Computing, R Foundation for Statistical Computing, Vienna, Austria). A p-value of ≤0.05 was considered statistically significant.

Results

Participants

Cough recordings from 91 patients with RCC were included in the analysis; baseline measurements are displayed in table 1. The demographics were typical of chronic cough populations; mostly female (62%), with a median (IQR) age of 60.0 (54–67.0) years and cough duration of 572 (26–1040) weeks. Smoking history was minimal across the group; 71 participants had never smoked, and lung function parameters were above normal limits. When compared by sex, females had a longer cough duration, higher objective cough frequency (24-h, day and night) and rated their cough as more severe on VAS than males. Night-time coughing was minimal, therefore daytime VAS was used for bout analysis throughout.

TABLE 1.

Participant demographics and cough measurements

	All	Male	Female	p-value
Participants, n (%)	91	35 (38%)	56 (62%)
Age, years	60.0 (54.0–67.0)	59.0 (52.0–65.0)	60.0 (54.0–67.8)	0.313
Cough duration, weeks	572 (260–1040)	520 (216–1040)	624 (312–1040)	0.002
Smoking history, pack-years	0.0 (0.0–0.0)	0.0 (0.0–2.5)	0.0 (0.0–0.0)	<0.001
FEV1, % predicted	98.0 (86.0–108.0)	96.0 (91.0–111.0)	98.5 (86.0–107.8)	0.595
FVC, % predicted	105.0 (93.0–118.0)	104.0 (93.0–113.0)	105.5 (95.3–118.0)	<0.001
Daytime coughs, n	393.0 (239.0–792.0)	315.0 (157.0–524.0)	464.0 (278.0–842.8)	<0.001
Night-time coughs, n	21.0 (3.0–47.0)	9.0 (2.0–39.0)	27.0 (5.3–68.8)	<0.001
Total coughs 24 h, n	429.0 (274.0–813.0)	329.0 (166.0–545.0)	492.5 (305.3–922.3)	<0.001
Daytime cough severity VAS, mm	49.0 (28.0–63.0)	37.0 (20.0–63.0)	51.0 (39.5–62.5)	<0.001

All data are expressed as median (interquartile range), unless otherwise indicated. Mann–Whitney U test compared measurements between males and females. FEV₁: forced expiratory volume in 1 s; FVC: forced vital capacity; VAS: visual analogue scale.

Total number of bouts by inter-cough interval

The total number of bouts over 24-h and the number of single coughs was calculated for each inter-cough interval (supplementary table S1). As the inter-cough interval widened from 0.5 s to 3.0 s, the total number of bouts decreased, i.e. more coughs were grouped into fewer bouts. At the shorter intervals, 0.5 s and 1 s, there was a much larger, more variable number of bouts. Whether single coughs were included or not, there was little change in the number of bouts produced beyond the 3.0 s interval (figure 3).

FIGURE 3.

Box and whisker plots displaying the total number of bouts for each inter-cough interval a) with and b) without single coughs included. Boxes represent the median and interquartile range (IQR) (Q1, Q3) and whiskers calculated as 1.5×IQR (where IQR is Q3−Q1). Single data points represent outliers beyond the upper or lower whisker.

Bout length by inter-cough interval

Median bout lengths measured by number of coughs and seconds are displayed in supplementary table S2. The average number of coughs in a bout did not change significantly as the interval widened. Bouts mostly comprised two coughs (with singles) or three coughs (without singles), irrespective of the interval (figure 4a and b). For all inter-cough intervals, bouts were on average one cough longer when single coughs were removed. There was a gradual increase in average bout length, measured in seconds, as the interval increased (figure 4c and d); however, this did not exceed 2 s. Even at the 10 s interval, median (IQR) bout length was 1.13 (0.66–3.82) s with singles and 1.7 (0.87–5.08) s without singles (supplementary table S2).

FIGURE 4.

Box and whisker plots displaying a) bout length in coughs per bout with singles, b) bout length in coughs per bout without singles, c) bout length in seconds with singles, and d) bout length in seconds without singles. Boxes represent median and interquartile range (IQR) (Q1, Q3) and whiskers calculated as 1.5×IQR (where IQR is Q3−Q1). Single data points represent outliers beyond the upper or lower whisker. Note log scaled x-axis.

Maximum bout length by inter-cough interval

To evaluate longer bouts, the median (IQR) maximum bout length for each interval was calculated (supplementary table S3). The maximum bout length was similar (12–14 coughs) above the 2 s interval but there was large variability in the data. A maximum bout length of 96 coughs lasting for 103 s was recorded at the larger intervals (≥8 s) (figure 5).

FIGURE 5.

Box and whisker plots displaying maximum bout lengths for each inter-cough interval; measured by a) coughs per bout and b) seconds per bout. Boxes represent the median and interquartile range (IQR) (Q1, Q3) and whiskers calculated as 1.5×IQR (where IQR is Q3−Q1). Single data points represent outliers beyond the upper or lower whisker. There were no patients with a maximum bout length of 1 therefore the with and without single values were the same.

Total bout duration and cough free time by inter-cough interval

All bouts were combined together to give a total bout duration (in minutes), i.e. the total amount of time spent coughing in bouts over 24-h. Inversely, the amount of “cough-free time” over 24-h was calculated by deducting the total bout duration from 24-h. Total bout duration and cough free time for each inter-cough interval is shown in supplementary table S4. As the intervals widened, total bout duration expanded (figure 6) and cough free time reduced (supplementary figure S2), however the median values changed little for intervals ≥3 s.

FIGURE 6.

Box and whisker plots displaying total bout duration (in minutes) for each time interval a) with and b) without single coughs included. Boxes represent the median and interquartile range (IQR) (Q1, Q3) and whiskers calculated as 1.5×IQR (where IQR is Q3−Q1). Single data points represent outliers beyond the upper or lower whisker.

Correlation of cough bouts with cough severity VAS

Number of cough bouts versus VAS

With single coughs included, correlations between the total number of bouts and VAS were weak but significant for all time intervals (supplementary table S5). The 0.5 s and 2.0 s intervals demonstrated the highest correlation coefficients; rho=0.37, p=0.001 and rho=0.35, p<0.001, respectively: similar to the correlation with standard cough frequency (total number of explosive cough sounds in 24-h, rho=0.38, p<0.001; listed as 0.0-s time interval). Widening the inter-cough interval beyond 3 s did not change the correlation coefficients (figure 7). When single coughs were excluded, correlations between number of bouts and VAS were similar for all windows and slightly stronger than with singles. The 1.0, 2.0, 2.5 and 3.0 s intervals yielded a slightly higher correlation coefficient than other intervals (rho=0.37, p<0.001; supplementary table S5).

FIGURE 7.

Dual axis plots showing distribution of total number of bouts a) with and; b) without single coughs, and the correlation between the total number of bouts and cough severity visual analogue scales by time interval. On the left axis, dots represent median values and error bars interquartile range. On the right axis, the triangles represent correlation coefficient (Spearman's rank rho).

Cough bout length versus VAS

There were no significant relationships between cough severity VAS and cough bout length measured either as coughs per bout or seconds (supplementary table S6). Correlation coefficients were similar for all intervals whether single coughs were included or not. Maximum bout length correlated slightly better with VAS than median bout length (supplementary table S7); correlations were statistically significant at all intervals other than 1.5 s but rho values were less than 0.4.

Total bout duration and cough-free time versus VAS

All correlations between total bout duration and cough severity VAS were significant (p<0.001; supplementary table S8). The rho value was relatively stable between 0.34 and 0.42 for all intervals both with and without single coughs. Correlations between cough-free time and VAS were the same as those between VAS and total bout duration but they were inversely related, i.e. lower VAS scores related to higher amounts of cough-free time, with rho values around −0.4 (p<0.001 for all intervals with and without single coughs).

Influence of sex

There were no significant differences between males and females in the number of single coughs generated by each time interval (supplementary table S9). The 0.5 s time interval produced the greatest number of single coughs, which then decreased as the interval widened but stabilised at 3.0 s with no significant change above this.

Female patients exhibited numerically more cough bouts than males for all inter-cough intervals, irrespective of whether single coughs were included (supplementary table S10). The differences became significant when single coughs were removed; however, the group was not well-balanced for sex (males n=35).

Bout length did not differ between sexes when measured either as coughs per bout (supplementary table S11) or in seconds (supplementary table S12). Both males and females mostly coughed in bouts of two or three coughs at all intervals ≥0.5 s. Total bout duration (supplementary table S13) was significantly longer and cough-free time shorter (supplementary table S14) for females at all time intervals, irrespective of the inclusion of single coughs.

Influence of age

There were no significant relationships between age and any of the cough bout parameters tested at any time interval; number of bouts (supplementary table S15), bout length (supplementary table S16), total bout duration and cough-free time (supplemental table S17).

Discussion

To the best of our knowledge, this is the first exploration of the influence of different definitions of cough bouts on cough bout parameters, and their relationships with reported cough severity. In RCC patients, we found that using inter-cough intervals of >3 s to define bouts of coughing made little difference to the median numbers of cough bouts, the maximum number of coughs per bouts, total time spent coughing or cough-free time. The median number of coughs per bout was relatively independent of the interval used to define bouts. However, of note, longer inter-cough intervals did result in very long cough bouts, consisting of over 95 coughs in some individuals. The duration of cough bouts (in seconds) behaved slightly differently, with the median and maximum length of bouts in seconds very gradually increasing with longer inter-cough intervals.

Overall correlations between reported cough severity and cough bout parameters tended to be weak, and similar to those seen with standard cough frequency. Cough severity correlated best with the median numbers of cough bouts with single coughs removed, for shorter inter-cough intervals (<3 s) if single coughs were included. The median coughs per bout and cough bout length (with/without single coughs), however, were not significantly related to cough severity. In contrast, the maximum bout length and total time spent coughing/cough-free time did significantly influence cough severity, irrespective of the interval used, with some correlation coefficients exceeding that with standard cough frequency.

Previous publications, including those from our group, have arbitrarily defined bouts as continuous coughing without a 2 s pause. The handling of single coughs, i.e. whether to exclude these or include them as bouts comprising one cough, has not previously been addressed. These data argue that using intervals of greater than 3 s to define cough bouts makes very little difference to most of the parameters explored. This indicates that once a 3 s pause in coughing occurs, the likelihood of further coughing is much reduced, providing some insights into the mechanisms governing the nature of coughing bouts, and perhaps reflecting inhibitory controls that terminate/inhibit coughing episodes. Moreover, including single coughs in our analysis had little impact on most bout parameters and their removal resulted in slightly stronger correlations with cough severity. This observation implies that single coughs have little impact for RCC patients.

Consistent with qualitative data from RCC patients, the maximum length of cough bouts (and maximum number of coughs per bout) exhibited significant relationships with reported cough severity, not seen with the average measures of bout length. Very prolonged coughing bouts are known to be the most distressing for RCC patients and associated with physical complications of coughing. Such episodes are probably more memorable and hence more likely to be reflected by patient-reported outcomes.

We have studied data from a relatively large number of RCC patients typical of those described in the literature, being predominantly female with a median age of 60 years. We have previously observed that female patients attending specialist clinics tend to have higher cough frequencies than male counterparts [19]. Our data suggest this is a consequence of a greater number of coughing bouts rather than more prolonged coughing bouts. It is difficult to judge whether this reflects differences in the mechanisms driving cough in females compared with males or differences in the severity of patients referred to specialist services with chronic cough. Interestingly, patient age did not appear to influence cough bout parameters. Whether the clustering of coughs in other respiratory conditions will demonstrate similar patterns, sex differences and relationships with cough severity to those described in RCC remains to be explored. Interestingly, a recent study on cough sound annotation suggested females exhibited shorter cough sounds and longer cough bouts than men but this was a small study (n=24) of patients with a range of diagnoses (including COVID-19-associated cough) and only analysing 10% of the cough data collected [20].

For this analysis we used 24-h cough recordings that had undergone manual cough counting, considered the gold standard for quantification of coughing. This avoids the addition of artefacts or removal of coughs by filtering or automatic detection algorithms. The algorithm for generating the number of cough bouts from this data was fully automated and could be applied to data on the location of coughs from semi or fully automated cough detection systems. Fully automated cough detection would be of great use to understand the value of cough monitoring in clinical practice and to reduce costs in clinical trials. However, the accuracy of currently available fully automated cough monitoring systems is poorly defined, as only limited validation studies have been performed.

The methodology described here for defining cough bouts has the advantage of being relatively simple to apply, but has some limitations. While the definitions of cough bouts are based on precise inter-cough intervals, the location of each cough within a 24-h sound recording is dependent upon manual tagging of the explosive part of cough sounds by cough analysts. Agreement between analysts counting explosive cough sounds has been shown to be excellent, suggesting identification of explosive phases of cough sounds is relatively easy to perform. Nonetheless, it is likely that some variability will exist between analysts. Additionally, our cough bout duration parameters were calculated by applying a fixed value to account for the length of coughs at the end of a bout and single coughs. Although this value was derived from previous data on the length of cough sounds, these values are intrinsically an estimate of the true time spent coughing. Generating precise data on individual cough bout duration would be challenging, as the amplitude of cough sounds tends to gradually dwindle, making the exact end difficult to identify. This would also be incredibly time-consuming considering the large number of cough bouts in each 24-h recording. Finally, VAS was the only patient reported outcome measure available in this dataset and it may be useful in future studies to correlate cough bout parameters with additional quality of life tools that capture some of the physical and psychological effects of chronic coughing.

In conclusion, these analyses favour definitions of cough bouts utilising shorter inter-cough intervals (≤3 s) and the exclusion of single coughs for RCC patients. Further work is required to delineate the optimal definition for cough bouts and should include responsiveness of cough bout parameters to interventions alongside a range of patient reported outcomes.