Cough predicts prognosis in idiopathic pulmonary fibrosis

ABSTRACT

Background and objective:  The clinical associations and prognostic value of cough in IPF have not been adequately described. The objective of this study was to describe the characteristics and prognostic value of cough in IPF.

Methods:  Subjects with IPF were identified from an ongoing longitudinal database. Cough and other clinical variables were recorded prospectively. Logistic regression was used to determine predictors of cough and predictors of disease progression, defined as 10% decline in FVC, 15% decline in DL_CO, lung transplantation or death within 6 months of clinic visit. The relationship of cough with time to death or lung transplantation was analysed using Cox proportional hazards analysis.

Results:  Two hundred and forty-two subjects were included. Cough was reported in 84% of subjects. On multivariate analysis, cough was less likely in previous smokers (OR 0.07, 95% CI: 0.01–0.55, P = 0.01), and more likely in subjects with exertional desaturation (OR 2.56, 95% CI: 1.15–5.72, P = 0.02) and lower FVC (OR 0.76, 95% CI: 0.60–0.96, P = 0.02). Cough predicted disease progression (OR 4.97, 95% CI: 1.25–19.80, P = 0.02) independent of disease severity, and may predict time to death or lung transplantation (HR 1.78, 95% CI: 0.94–3.35, P = 0.08).

Conclusions:  Cough in IPF is more prevalent in never-smokers and patients with more advanced disease. Cough is an independent predictor of disease progression and may predict time to death or lung transplantation.

INTRODUCTION

IPF is a common form of interstitial lung disease (ILD) that affects approximately 100 000 people in the USA.¹ IPF is generally characterized by relentless progression and an estimated 3-year mortality of up to 50%.^2,3 IPF is a highly morbid disease, with over 90% of patients reporting symptoms at the time of diagnosis.⁴

Cough and dyspnoea are the two most common symptoms described in patients with IPF.⁵ The characteristics and prognostic implications of dyspnoea in patients with IPF have been well described.^5,6 Dyspnoea is associated with important comorbidities including depression and debility,⁷ is strongly correlated with reduced pulmonary function and quality of life,⁶ and is able to predict disease progression and mortality.⁸ Unlike dyspnoea, the clinical associations and prognostic value of cough in IPF have not been adequately described.

We therefore had three main objectives: (i) to determine the prevalence of cough in IPF; (ii) to determine the association of cough with baseline clinical parameters; and (iii) to determine whether cough predicts disease progression and mortality, independent of disease severity. We hypothesized that cough is common in IPF, is associated with disease severity, and that the presence of cough predicts more rapid disease progression and mortality, independent of disease severity.

METHODS

Study subjects

Subjects were identified through the University of California San Francisco (UCSF) ILD Database between 2000 and 2010. Subjects were included if they had a diagnosis of IPF based on multi-disciplinary review according to established criteria.^5,9 There were no exclusion criteria. The UCSF Committee on Human Research approved the UCSF ILD Database and all subjects provided written informed consent.

Measurements

Baseline information was recorded and questionnaires completed at the time of initial consultation. Cough was recorded at baseline as a Yes/No response to the question ‘Do you cough? (This includes any cough, even if there is no phlegm. Do not include clearing your throat)’. Additional data included age, gender, BMI, smoking history and dyspnoea severity. Dyspnoea severity was measured using the self-administered dyspnoea component of the Clinical, Radiographic, and Physiologic scoring system.¹⁰

Conditions that commonly cause cough were identified by both patient questionnaire and chart review. These conditions included current smoking, current treatment with an angiotensin converting enzyme (ACE) inhibitor, any cause of upper airway cough syndrome (UACS), and a diagnosis of asthma, COPD or gastroesophageal reflux disease (GERD). Relevant diagnostic studies were performed according to the discretion of the treating physician. These results were reviewed to improve diagnostic accuracy (e.g. bronchodilator responsiveness, oesophageal studies). A composite dichotomous index was created to indicate the presence of any of these conditions.

Resting oxygen saturation and the need for long-term oxygen therapy were recorded. Exertional desaturation was defined as a decline in pulse oximetry of at least 4% to less than 93% at any time during a 3- or 6-min walk test,¹¹ or by the requirement for supplemental oxygen therapy at rest. Pre-bronchodilator FVC, TLC and DL_CO were measured using standard techniques.^12–14

Date of transplant or death was recorded. Vital status and date of death were verified using the United States Death Registry Index. A composite endpoint of lung transplantation and death was used for survival analysis (i.e. transplant-free survival). This was based on the finding that subjects undergoing lung transplantation had similar or worse measures of disease severity compared with deceased subjects.

Statistical analysis

The association of cough with other variables was tested using the χ²-test, Fisher's exact test, Student's t-test or the Wilcoxon rank-sum (Mann–Whitney) test. The following predefined independent variables were evaluated: age, gender, BMI, diagnosis by surgical lung biopsy, smoking history, other potential causes of cough as described above and measures of disease severity. Measures of disease severity included dyspnoea score, use of long-term oxygen therapy, exertional oxygen desaturation, FVC, TLC and DL_CO. Disease progression was defined as any of the following within 6 months of the initial UCSF ILD Clinic visit: 10% decline in FVC, 15% decline in DL_CO, lung transplantation or death due to any cause.

Logistic regression was used to determine the independent predictors of cough, and the relationship between cough and disease progression. Cox proportional hazards analysis was used to evaluate the independent relationship of cough with transplant-free survival. Predictors with a bivariate P < 0.20 were included in multivariate models. Backward selection was used to determine which variables were retained in each model, using the Akaike Information Criterion to provide a balance between accuracy and complexity in the model.¹⁵ This approach may retain variables with slightly higher P-values if they contribute importantly to the predictor model. Predictor variables were transformed to approximate a normal distribution if necessary. No variables were forced into the model. All data analysis was performed using STATA 11.0 (StataCorp, College Station, TX, USA).

RESULTS

Study subjects

Demographic and clinical characteristics of the 242 subjects are summarized in Table 1. Subjects had a mean age of 70 years and were predominantly men and former smokers. Cough was present in 84% of subjects. A slight majority reported phlegm production (57%) or wheeze (52%). Three of 178 subjects with bronchodilator testing met criteria for a positive bronchodilator response.¹⁶ Pharmacotherapy directed at IPF had been used in 33% of subjects, including prednisone alone in 21% and prednisone with additional pharmacotherapy in 8%. Inhalers were currently being used in 34% of subjects, including bronchodilators in 33% and corticosteroids in 18%. Mean FVC was 70% predicted, mean TLC was 69% predicted and mean DL_CO was 44% predicted.

Table 1.  Subject characteristics

Variable	Value	SD
ACE, angiotensin converting enzyme; GERD, gastroesophageal reflux disease; UACS, upper airway cough syndrome.
Age, years	69.9	8.6
Male gender, %	73.0	—
BMI, cm/kg^2	27.9	4.8
Diagnosis by surgical lung biopsy, %	44.2	—
Cough, %	84.3	—
Smoking history
Ever smoked, %	75.6	—
Pack-years	23.3	25.3
Other potential causes of cough
GERD, %	35.7	—
COPD, %	29.1	—
Asthma, %	17.4	—
UACS, %	17.0	—
Current ACE inhibitor therapy, %	9.1	—
Current cigarette smoking, %	1.7	—
Any potential cause of cough, %	74.0	—
Measures of disease severity
Dyspnoea score	9.9	5.7
Long-term oxygen therapy, %	24.8	—
Exertional desaturation, %	64.1	—
Pulmonary function
FVC, % predicted	69.3	18.2
TLC, % predicted	69.2	15.2
DLCO, % predicted	44.0	17.0

Association of cough with baseline clinical variables

Cough was strongly associated with several variables on bivariate analysis (Table 2, Fig. 1). Smoking history was the strongest predictor of cough, with current or previous smokers less likely to report cough (57 of 59 non-smokers reported cough (97%), compared with 147 of 183 current or previous smokers (80%); OR 0.14, 95% CI: 0–0.56, P = 0.003). This relationship remained significant when comparing never-smokers to previous smokers with as few as 1–5 pack-years (P = 0.03). The association of cough with number of pack-years showed similar findings, with higher number of pack-years associated with a lower prevalence of cough (P = 0.009). There was no association between the date of smoking cessation and the presence of cough (P = 0.52). The presence of cough was not associated with the presence asthma, COPD, GERD, UACS or current ACE inhibitor use. There were too few current cigarette smokers (n = 4) to determine the relationship of current smoking with the presence of cough.

Table 2.  Bivariate association of cough with clinical parameters

Variable	Odds ratio	95% CI	P-value
^†  Odds ratio reported for 10-unit change in age, FVC, TLC and DLCO. ^‡  Odds ratio reported for square root of pack-years (pack-years transformed to normal distribution). ^§  Odds ratio reported for 2-point change in dyspnoea score. ACE, angiotensin converting enzyme; GERD, gastroesophageal reflux disease; UACS, upper airway cough syndrome.
Age, years^†	1.24	0.83–1.84	0.29
Male gender	0.81	0.37–1.80	0.62
BMI	1.05	0.97–1.14	0.24
Diagnosis by surgical lung biopsy	1.11	0.55–2.23	0.78
Smoking history
Ever smoked	0.14	0–0.56	0.003
Pack-years^‡	0.87	0.79–0.97	0.009
Other potential causes of cough
GERD	1.03	0.50–2.12	0.94
COPD	1.65	0.73–3.73	0.24
Asthma	1.46	0.55–3.86	0.46
UACS	0.86	0.35–2.06	0.74
Current ACE inhibitor therapy	1.16	0.35–3.87	0.81
Current cigarette smoking	0.55	0.08–5.45	0.61
Any potential cause of cough	1.38	0.66–2.91	0.40
Measures of disease severity
Dyspnoea score^§	1.16	1.02–1.32	0.02
Long-term oxygen therapy	1.21	0.53–2.76	0.66
Exertional desaturation	2.41	1.19–4.85	0.01
Pulmonary function
FVC, % predicted^†	0.76	0.63–0.93	0.007
TLC, % predicted^†	0.75	0.57–0.99	0.04
DLCO, % predicted^†	0.94	0.76–1.16	0.55

Figure 1. Box plots of continuous variables associated with cough on bivariate analysis.

The current use of systemic pharmacotherapy directed at IPF was not significantly associated with the presence of cough (OR 2.05, 95% CI: 0.91–4.61, P = 0.09). Cough was associated with bronchodilator use (OR 2.52, 95% CI: 1.08–5.87, P = 0.03) and inhaled corticosteroid use (OR 4.53, 95% CI: 1.15–19.6, P = 0.03). Dyspnoea score was increased and FVC was decreased in subjects using inhaled or systemic pharmacotherapy (P ≤ 0.005 for all comparisons).

On multivariate analysis, smoking status remained the strongest predictor of cough, with previous smokers less likely to report cough (OR 0.07, 95% CI: 0.01–0.55, P = 0.01) (Table 3). Increased disease severity, indicated by low FVC and exertional desaturation, also remained strongly predictive of the presence of cough. The strength of these associations was not altered by considering the possibility of threshold effects (e.g. that only a FVC below a certain threshold impacted cough) or when considering the impact of interactions among included variables.

Table 3.  Multivariate model of cough

Variable	Unadjusted odds ratio	Adjusted odds ratio	95% CI	P-value
^†  Odds ratio reported for 10% change in % predicted FVC. Variables included are from stepwise regression, using the Akaike Information Criterion as the criterion for inclusion.
Ever smoked	0.14	0.07	0.01–0.55	0.01
Exertional desaturation	2.41	2.56	1.15–5.72	0.02
FVC, % predicted^†	0.76	0.76	0.60–0.96	0.02

Association of cough with disease progression

Six-month follow-up pulmonary function was available for 159 subjects (66%). There was no difference in baseline characteristics in subjects with and without 6-month data. Disease progression occurred in 66 of these 159 patients (36 with decline in pulmonary function, 27 deaths, 3 lung transplantations). Decline in FVC was present in 30 subjects and decline in DL_CO was present in 15 subjects. Nine subjects had decline in both pulmonary function measures.

On bivariate analysis, disease progression was significantly more common in subjects with higher baseline dyspnoea, long-term oxygen therapy and baseline exertional desaturation (Table S1). Cough had borderline association with disease progression on bivariate analysis (OR 2.40, 95% CI: 0.92–6.23, P = 0.07). Cough predicted disease progression with a sensitivity of 91%, but specificity was only 19%. On multivariate analysis, cough was an independent predictor of disease progression (OR 4.97, 95% CI: 1.25–19.80, P = 0.02) (Table 4). Disease progression was not related to the current use of inhaled bronchodilators or corticosteroids, or to systemic pharmacotherapy for IPF.

Table 4.  Multivariate model of disease progression

Variable	Unadjusted odds ratio	Adjusted odds ratio	95% CI	P-value
^†  Odds ratio reported for 10% change in % predicted TLC. Variables included are from stepwise regression, using the Akaike Information Criterion as the criterion for inclusion. Disease progression defined as 10% decline in FVC, 15% decline in DLCO, lung transplantation or death due to any cause.
Cough	2.40	4.97	1.25–19.80	0.02
Long-term oxygen therapy	3.33	4.85	1.75–13.44	0.002
TLC, % predicted^†	1.24	1.43	1.05–1.94	0.02

Association of cough with transplant-free survival

Death occurred in 100 subjects and 15 subjects underwent lung transplantation. The median time to death or transplant was 3.0 years (Fig. S1). On bivariate analysis, significant predictors of transplant-free survival included dyspnoea, long-term oxygen therapy, exertional desaturation, FVC, TLC and DL_CO (Table S2). Cough had borderline association with decreased time to death or lung transplantation on bivariate (HR 1.59, 95% CI: 0.91–2.79, P = 0.10) (Fig. 2) and multivariate (HR 1.78, 95% CI: 0.94–3.35, P = 0.08) analysis (Table 5). Cough was retained in the multivariate predictive model using Akaike Information Criterion analysis. When censoring subjects who underwent lung transplantation, cough was an independent predictor of time to death.

Figure 2. Kaplan–Meier curve, stratified by cough. Reported as time from date of clinic visit to death or lung transplant. HR, hazard ratio. () Cough absent; () Cough present.

Table 5.  Multivariate model of time to death or lung transplant

Variable	Unadjusted hazard ratio	Adjusted hazard ratio	95% CI	P-value
^†  Odds ratio reported for 10% change in % predicted DLCO and TLC. Variables included are from stepwise regression, using the Akaike Information Criterion as the criterion for inclusion.
Cough	1.59	1.78	0.94–3.35	0.08
Long-term oxygen therapy	4.80	3.28	2.03–5.30	<0.0005
TLC, % predicted^†	0.79	0.86	0.75–0.98	0.03
DLCO, % predicted^†	0.69	0.81	0.69–0.96	0.01

DISCUSSION

We show in a large, well-described cohort of subjects with IPF that cough is common, is increased in never-smokers and patients with more severe disease, is an independent predictor of disease progression, and may predict time to death or lung transplantation.

Association of cough with baseline clinical variables

The finding that cough is reduced in patients with a history of smoking was unexpected. The strength of the association and significant correlation between cough and pack-years makes this finding robust and chance is an unlikely explanation. There are several possible explanations for this finding. First, despite increasing cough frequency,¹⁷ smoking reduces cough sensitivity, typically measured as the amount of inhaled capsaicin that stimulates cough.¹⁸ This indicates that while smokers cough frequently, they require a larger irritant exposure to induce cough. This effect may persist following smoking cessation, resulting in decreased cough sensitivity in previous smokers compared with never-smokers. Second, reporting bias may occur with previous smokers less likely to report this symptom. We feel the simplicity and clarity of the question as asked in this study makes reporting bias unlikely. Third, smoking-related and smoking-unrelated IPF may have some phenotypic differences, such as smoking-related IPF being less likely to induce cough.

Cough was increased in subjects with IPF who had more severe physiological impairment (i.e. reduced FVC and DL_CO). This association is not surprising, as it is likely that the prevalence and severity of pulmonary symptoms are related to disease severity. The association of cough with disease severity in the present study conflicts with two previous studies. However, both of these studies were small and underpowered to detect this relationship.^19,20 The strength and consistency of the relationship between cough and physiological impairment in the current much larger study suggest this is a true association.

The associations of cough with pharmacotherapies likely represent confounding that resulted from increased treatment prescribed for patients who complain of symptoms or who have more severe disease. The association of treatment with worse dyspnoea severity and lower FVC supports this hypothesis. We did not have sufficient longitudinal treatment data to evaluate the effects of these treatments on cough.

We found no association of cough with the presence of other common causes of cough, including asthma, COPD, GERD, UACS, current ACE inhibitor use and current cigarette smoking. These results conflict with a study that showed six of nine subjects with IPF had cough attributed to a diagnosis other than IPF.²¹ This smaller study included patients referred to a cough clinic and this highly selected cohort might not be representative of most patients with IPF. The current study suggests that cough in IPF is commonly secondary to the primary disease and not other conditions. Recent guidelines recommend the exclusion of other causes of cough in patients with IPF.²² While this approach is reasonable, the current study suggests that this strategy may be unsuccessful in many cases.

Association of cough with disease progression and transplant-free survival

Cough was an independent predictor of disease progression and may predict shorter transplant-free survival. Although cough had borderline association with disease progression on bivariate analysis, cough was a significant predictor of disease progression when controlling for potential confounders. Because cough predicts prognosis independent of disease severity, the presence of cough may be able to identify patients with active disease prior to worsening of more typical measures of disease severity. In other words, cough may be an early clinical marker of disease activity and may identify patients at high risk of progression. This would have important implications for disease surveillance, aggressiveness of treatment and potentially design of clinical trials.

Cough in IPF is likely produced by either mechanical or chemical stimulation of peripheral cough receptors. The association of cough with disease progression suggests that patients with active disease have increased stimulation of at least one of these pathways. Stimulation of mechanical cough receptors could be related to worsening fibrosis that results in architectural distortion. Stimulation of chemical cough receptors could be related to subclinical inflammation. Previous studies have found increased eosinophils and other inflammatory markers in sputum and BAL fluid from patients with IPF, supporting this hypothesis.^23–25 We did not have sputum or BAL samples with which to evaluate this possibility.

There are several limitations to this study. First, we defined cough based on the response to an unvalidated question, and additional characteristics of cough (e.g. frequency, intensity, duration) were not available. We could not rule out the possibility that some cases of cough represent acute cough. Quantitative assessment of cough may improve the low specificity of cough for prediction of disease progression that was observed in this study. However, there is no cough questionnaire validated in IPF, and cough is a symptom that is easily represented by a dichotomous response. The presence or absence of cough is readily available in the clinical setting and the associations detected in the current study suggest that these dichotomous data provide valuable prognostic information. Second, we did not use a standardized definition of COPD, asthma, UACS or GERD. However, the finding that cough predicts prognosis is an important and clinically useful finding, regardless of the aetiology of cough. Third, disease progression data were not available in 34% of study subjects. It is possible that subjects with cough were evaluated more frequently for disease progression; however, baseline characteristics, including the presence of cough, were similar in subjects with and without progression data available. We therefore believe it is unlikely that the lack of follow-up data led to a spurious association between cough and disease progression. Finally, subjects were recruited from a tertiary ILD clinic and the characteristics of these subjects may not be representative of the general IPF population.

In summary, we confirm that cough is common in patients with IPF and demonstrate that cough is more prevalent in never-smokers and in patients with more advanced disease. Importantly, cough is an independent predictor of disease progression, and may predict time to death or lung transplantation. Future longitudinal studies of cough in IPF using validated instruments are required to verify these findings and to describe the determinants and impact of change in cough over time. The association of cough with disease progression and possible association with transplant-free survival in the current study suggests that the presence of cough may be an important and clinically useful predictor of poor prognosis in IPF.

ACKNOWLEDGEMENTS

This research is supported by grant HL086516, and the University of British Columbia Clinical Investigator Program. The authors would also like to acknowledge the providers and staff of the UCSF Interstitial Lung Disease Program for their assistance in recruiting subjects for this study, the members of the UCSF Interstitial Lung Disease Consortium for their continued referral of patients to our centre and the patients with ILD who, through their generosity and efforts, allow us to conduct clinical research studies such as this in an effort to improve the lives of patients with ILD.