Fatigue, Dyspnea, and Cough Comprise a Persistent Symptom Cluster Up to Five Years After Diagnosis with Lung Cancer

Abstract

Context

Aggregates of concurrent symptoms, known as symptom clusters (SxCls), have been described in predominantly cross-sectional samples of lung cancer patients undergoing treatment.

Objectives

The objective of this study was to delineate symptom clusters in lung cancer survivors up to five years following diagnosis, investigate their stability over time, and identify determinants of SxCl development and resolution.

Methods

A sensitivity approach involving multiple exploratory and confirmatory analyses was applied to an eight-year prospective cohort study that annually assessed 2405 patients with lung cancer for symptom burden with the Lung Cancer Symptom Scale (LCSS) and Linear Analogue Self-Assessment (LASA).

Results

A single robust SxCl of fatigue, cough, and dyspnea was identified in 14.6%, 12.9%, 14.1%, 14.6%, and 15.4% of participants at years 1-5 after diagnosis, respectively. Participants with the SxCl (SxCl (+)) were more likely to die than those without it but this tendency diminished over time. SxCl persistence varied, with >40% of surviving patients annually transitioning to or from the SxCl(+) state until year 4, after which the SxCl became increasingly stable. The SxCl was more likely to develop among male survivors who underwent surgery, received radiation, and were current smokers.

Conclusion

A single SxCl comprised of dyspnea, fatigue, and cough has a stable prevalence among lung cancer survivors up to five years following diagnosis but is not stable among individuals. Initially following diagnosis, the SxCl is associated with a greater risk of death; however, after year 2 the SxCl becomes increasingly stable and provides a marker for parenchymal lung injury.

Introduction

Physicians typically confine consideration of symptoms to their role as diagnostic clues or as indicators of the need to initiate or change treatment. However, there is growing interest in the idea that a more encompassing consideration of symptoms, specifically as groups or clusters of symptoms, may provide benefits in the form of improved prognostication, screening and treatment optimization.^1-3 Although, a number of researchers have found empirical links between specific clusters of symptoms and adverse outcomes such as death, explanatory mechanisms remain speculative and the clinical relevance of symptom clusters (SxCl) is unclear.⁴

Lung cancer (LC) engenders a uniquely high symptom burden and, as such, has already received some attention from investigators in this area.^5-7 Previous authors have identified a number of symptom clusters such as anxiety, fatigue, and dyspnea^8-11 that were associated with poor patient performance. These studies, however, have been primarily cross sectional in nature and limited to patients on active treatment.^9-11 Understanding of the persistence of symptom clusters with time is much more limited. Persistence has emerged as an important issue in light of recent reports that symptoms remain problematic throughout the course of LC treatment and continue long into post-treatment survivorship. ^{8, 12}

Given the potential long-term relevance of SxCl to patients with LC, we utilized several complementary analytic techniques to identify SxCls in a cohort of 2405 LC survivors followed longitudinally over eight years. In addition to identifying SxCls using differing analytic approaches, we also examined their stability over time, as well as demographic, disease-based, and treatment-based factors associated with their persistence.

Methods

Subject Recruitment

The Epidemiology and Genetics of Lung Cancer research program maintains a database of all patients with a pathologically confirmed diagnosis of LC who have been evaluated at the Mayo Clinic, Rochester, Minnesota, since January 1, 1999. As of December 31, 2008, this data base contained entries for 2500 subjects who consented to be followed annually. Data collection involved sending patients who agreed to participate a mailed self-report study instrument within six months of their diagnosis and subsequently on a yearly basis. Additional details have been outlined in previous publications.^{13, 14} Patients contacted study personnel with questions but otherwise had no interaction with study personnel. The study was approved by the Mayo Clinic Institutional Review Board.

Study Questionnaire Components Utilized in This Study

The Lung Cancer Symptom Scale (LCSS) is a validated, disease-specific measure of symptom intensity and functionality that emphasizes symptoms related to LC.^{15, 16} Although the LCSS includes two parts, this study confined analyses to six self-report items that assess appetite, fatigue, coughing, dyspnea, hemoptysis, and pain. The intensity of patient responses is measured by 11-point visual analogue scales, with responses converted to a range extending from 0 (worst symptoms/highest burden) to 10 (least symptoms/lowest burden).

Linear Analogue Self-Assessment (LASA) consisted of five items that were rated on a scale from 0 (as bad as it can be) to 100 (as good as it can be). Items included ratings of cognitive, emotional, and social well-being, as well as sleep quality and overall quality of life. Single-item LASAs to assess cancer-related symptoms have been widely utilized in oncology research and evidence strongly suggests that they are valid and reliable.^17-19

Tobacco use at each assessment point was queried with questions regarding current and/or previous use, duration, average number of cigarettes smoked per day, and years since the patient quit smoking.

Information from Electronic Medical Records

Information regarding patients’ demographics, LC histology, staging, location, comorbidities, and treatment was abstracted from their electronic medical records by a research nurse. Copies of LC-related medical records for patients who received all or a portion of their treatment outside the Mayo Clinic were reviewed. ^20-22 The occurrence of disease progression, recurrent disease, and a second primary LC were all captured with a binary “active disease” variable.²³

Vital Status

Patients’ vital status was verified annually through death certificates, the Mayo Clinic’s electronic medical notes and registration database, and next-of-kin reports, as well as through the Mayo Clinic Tumor Registry and Social Security Death Index website.

Statistical Analysis

Analysis extended from the point of entry into the data base until December 31, 2008 or the time of death. Data obtained six months and one year after study enrollment were combined to generate a single score, with scores averaged as appropriate. Separate symptom clusters were created at each evaluation point (less than or equal to one year, two years, three years, four years, and five years).

Three methods were used to construct and validate the symptom clusters. The clusters were identified using variable clustering based on principal components. Cluster and factor analytic methods were based on the correlation matrix of the symptom variables as described by Barsevick et al.²⁴ Exploratory cluster analysis was performed including all available symptom data (poor appetite, fatigue, coughing, dyspnea, hemoptysis, pain, poor sleep quality and subjective cognitive, emotional, and social well-being) at each time point. The clustering criteria used was Euclidean distance and we followed the analytical approach suggested by Romesburg.²⁵ Variables were combined and dendrograms created based on the percent of variation explained by the variables. Factor analysis with varimax rotation was used to determine the variables in each cluster. Factors were selected if their eigenvalues were greater than one and they explained at least 10% of the variation, and the number of factors was verified using scree plots. Individual variables were considered to be part of a factor if their rotated factor loadings were greater than 0.50. The factors were then verified using bootstrapped factor analysis with 5000 replications. Variables with factor loadings over 0.50 in at least 70% of the samples were considered to be validated for that factor.²⁶

The final SxCls were then validated using latent trait analyses and the categorical classification described in the next paragraph.²⁷ Latent trait analysis is a special form of structural equation models wherein categorical variables are used to identify underlying relationships among continuous variables. Like other structural equation models, latent trait analysis involves a detailed modeling of the covariance structure of the variables being studied to identify correlational patterns that suggest pooling of individual observable variables into unobservable latent variables. We followed the methods of Bollen ²⁸ in carrying out the latent trait modeling processes. These latent trait analyses were created using PROC CALIS in SAS (SAS Institute Inc., Cary, NC).²⁹ For this latent trait analysis, the number of clusters and the variables in each cluster were based on the results of the factor analysis and cluster analysis. All three modeling methods used the raw variables.

Following SxCl identification, each patient was subsequently classified as experiencing the symptoms included in the delineated cluster at a problematic intensity or not. A patient was considered to be in a cluster if he/she had scores of 5 or less (on a 0 to 10 scale) on each of the variables in the cluster. A score of 5 was used as the threshold for inclusion in the SxCl subgroup given extensive evidence that a score of 5 or lower on an 11-point numeric rating scale (NRS) indicates a problematic symptom intensity level, and the alignment of this criterion with precedents in the cancer symptom control literature.^{30, 31} Patients were classified, using this scoring algorithm, as either rating all symptoms within the cluster at <5, SxCl(+), or not, SxCl(−). A separate classification category was used to represent death. Examination of the relative likelihood of transitioning among the three mutually exclusive states was accomplished using simple percentage summary statistics.

Markov Chain Monte Carlo

The relative likelihood of transitioning among the three mutually exclusive states was explored using Markov Chain Monte Carlo (MCMC). Separate MCMC models were created for transitioning from SxCl(−) to SxCl(+), transitioning from SxCl(−) to death, from SxCl(+) to SxCl(−), and from SxCl(+) to death. These models treated the proportion of patients transitioned as a Bernoulli process. The models were each run using PROC MCMC in SAS, with 500,000 iterations and 10,000 burn-in iterations. These models adjusted for age at diagnosis, smoking pack years, gender, race, marital status, education, stage of disease, treatment, chronic obstructive pulmonary disease (COPD) status, and whether or not the patient had prior recurrences.

Results

Subjects

Of 3162 patients sent invitations to participate, 65 (2%) died before responding, 208 (7%) refused, 389 (12%) did not respond, and 2500 (79%) consented and returned questionnaires. Among the 2500 patients who returned questionnaires at >1 assessment point, 95 (4%) provided insufficient data for analysis. Therefore, 2405 (76% of eligible subjects) were included in the analyses. The left two columns of Table 1 describe the demographics, LC characteristics and comorbidities of the 2405 patients who were, and the 757 who were not, included in the analyses. Subjects who provided data were more likely to be never or former smokers (P<0.001), to be married (P = 0.011), to have post-high school education (P <0.0001), to have greater medical morbidity and to have been diagnosed with stage I LC (P <0.0001). The number of patients returning questionnaires diminished at each assessment point following diagnosis because of mortality. Data were available for years <1, 2, 3, 4, and 5 from 1828, 1244, 941, 738, and 482 subjects, respectively.

Table 1.

Characteristics of Patients Who Did and Did Not Provide Symptom Data and of Patients with and without the Symptom Cluster at Year ≤1 After Lung Cancer Diagnosis

	Subjects Invited to Participate in Long-term Lung Cancer Follow-up Study			Subjects with Symptom Data at ≤ 1 Year after Diagnosis with Lung Cancer
	Data available	Data Unvailable	P value	SxCl (+)	SxCl (−)	P value*
	N = 2405	N = 757		N = 266	N = 1562
Demographics
Age mean (SD)	67.4 (10.7)	66.6 (10.8)	0.148	67.3 (9.6)	65.7 (11.0)	0.042
Gender, % Female	47.7	51.5	0.069	32	49.9	<0.001
Race, % Caucasian	93.6	92.8	0.445	92.9	94.5	0.102
Marital status, %			0.011			0.672
Married	79.2	72.8		79.8	80
Single	3.8	4.1		2.7	3.7
Divorced	6.9	10.8		9	6.7
Widowed	10	12.2		8.5	9.5
Education beyond high school, %	51.5	42.2	<0.001	42.3	49.7	0.086
Tobacco Exposure
Status at study intake, %			<0.001			0.004
Current smoker	29.9	37.2		35.3	27.3
Former smoker	53.2	41.7		52.6	54.1
Never smoker	16.9	21.2		12	18.6
Pack years mean (SD)	48.6 (30.6)	50.8 (30.6)	0.155	53.1 (31.4)	47.5 (29.5)	0.021
Lung Cancer Characteristics and Treatment
Stage, %			<0.001			0.005
I NSCLC	41.2	29.3		31.4	41.2
II NSCLC	8.8	10.2		7.8	7.4
III NSCLC/Limited SCLC	28.2	34.6		37.1	27.6
IV NSCLC/Extensive SCLC	21.8	28.7		21.2	23.4
Cell Type, %			0.057			0.056
Adenocarcinoma	50.8	47.9		42.9	51.9
Squamous cell carcinoma	22.0	19.9		29.3	21.4
Large cell carcinoma	2.6	2.6		2.6	2.0
Adenosquamous carcinoma	1.3	1.5		1.5	1.5
NSCLC unspecified	8.2	12.8		9.4	8.1
Small cell lung cancer	7.6	9.5		9.0	8.0
Surgery, %	64.1	59.0	0.030	57.1	62.4	0.104
Received chemotherapy, %	54.3	42.6	<0.001	58.3	55.3	0.361
Received radiation therapy, %	35.2	31.5	0.116	42.9	32.5	0.001
Medical Comorbidity
Asthma, %	4.9	4.4	0.568	0.4	3.3	0.009
COPD, %	19.8	12.5	<0.001	10.5	9.5	0.615
Stroke, %	0.7	0.1	0.067	3.8	1.6	0.018
Diabetes, %	1.5	0.5	0.038	1.1	2.2	0.24
Heart disase, %	3.5	0.5	<0.001	4.9	5	0.941
Hypertension, %	6.0	1.7	<0.001	4.9	6.6	0.291

SxCl = symptom cluster; NSCLC = non-small cell lung cancer; SCLC = small cell lung cancer.

^aP-values represent the results of t-tests for continuous variables and Chi-square tests for categorical variables. COMP: PLS REPLACE ASTERISKS IN TABLE WITH SUPERSCRIPT ITALIC a.

Symptom Cluster Delineation

Cluster Analyses

Non-parametric cluster analysis revealed the existence of a single SxCl comprised of dyspnea, cough, and fatigue at each of the first five years after LC diagnosis. Figure 1 presents a dendrogram of the year 1 cluster analysis. Cluster analyses for years 2-5 were very similar. In each year, fatigue and dyspnea combined first, and, excepting year 5 when cough first combined with pain, cough joined the fatigue/dyspnea cluster before any other variables combined.

Figure 1.

Dendrogram of <1 cluster analysis.

Factor Analyses

Exploratory factor analyses were performed for years 1-5 which yielded the integrated scree plot in Fig. 2. The percentage of explained variance, factor loadings, and eigenvalues of all factors identified by the year < 1 analysis are listed in Table 2. Year 2-5 factor analyses yielded similar results. A single dominant factor representing the SxCl of fatigue, dyspnea, and cough was present at each year, with eigenvalues ranging from 2.11 to 2.52. A second factor comprising emotional disturbance and either pain or sleep also was detected in years 1 through 4. However the eigenvalue for this factor was low, 0.28 to 0.49, and this factor did not differentiate substantially from other factors, as depicted in Fig. 2. To further assess the consistency of the SxCl across years following lung cancer diagnosis, 5000 bootstrapped factor analyses were performed with data from each year. Using a factor loading threshold of >0.50, the percentages of bootstrapped samples that yielded a factor containing the SxCl variables are listed in Table 3.

Figure 2.

Cumulative scree plots for factor analyses of data from years <1 through 5.

Table 2.

Results of Year ≤ 1 Exploratory Factor Analysis Including Factor Loadings > 0.5, Eigenvalues, and Percentage of Variance Explained

Variable	Factor 1	Factor 2	Factor 3	Factor 4
Fatigue	0.68
SOB	0.67
Cough	0.5
Sleep		0.51
Emotional		−0.54
Pain
Cognitive Thinking
Appetite
Eigen value	2.11	0.34	0.1	0.04
Percentage of variance explained by the factor	82%	13%	4%	2%

Table 3.

Percentage of Times That Specific Symptoms with Loadings ≥ 0.5 Were Included in the Dominant Factor (Factor 1), Based on 5000 Bootstrapped Samples with Varimax Rotation

	<=1 Year	2nd Year	3rd Year	4th Year	5th Year
Fatigue	91%	98%	95%	91%	98%
Dyspnea	84%	91%	91%	85%	97%
Cough	49%	91%	92%	80%	66%
Pain	26%	21%	17%	12%	30%
Sleep	15%	6%	7%	15%	2%
Emotional Disturbance	0%	0%	0%	0%	0%
Appetite	0%	0%	0%	0%	0%
Cognitive	0%	0%	0%	0%	0%

Latent Trait Analyses

The latent trait analysis confirmed the existence of two factors, as depicted in Fig. 3. The first factor comprised fatigue, shortness of breath, and cough while the second factor comprised emotional well-being and sleep items. The observed coefficients within each factor are between 0.5 and 0.84 indicating that the linkages among the variables within each factor are strong. The two factors are modestly correlated (coefficient of −0.34). These results were consistent with those obtained through factor analysis and cluster analysis.

Figure 3.

Latent variable structure derived from the modeled data.

Prevalence of the Symptom Cluster

As outlined in the Methods section, inclusion in the SxCl subgroup was based on rating all three symptoms at an intensity level < 5. The percentage of patients experiencing the SxCl remained stable after diagnosis and was 14.6%, 12.9%, 14.1%, 14.6%, and 15.4%, in years 1-5, respectively. The right-hand columns of Table 1 list characteristics of the 1828 subjects with data at < year 1 who did (n=266) and did not (n=1562) experience the SxCl. Patients with the SxCl were significantly more likely to be male (P <0.0001), have greater tobacco exposure (P = 0.021) and to be current smokers (P = 0.004). They were also more likely to have had radiation (43% versus 33%, P = 0.001) and a history of stroke (P = 0.02). Similar patterns were noted in the SxCl(+) and SxCl(−) patients in years 2-5.

Transitions Between SxCl Defined States Over Time

Changes in the transition rates between the states of SxCl(+), SxCl(−), and death fluctuated over time as depicted in panels A-D of Fig. 4 and were most marked in the transition between years 1 and 2. The percentage of patients who died was highest between years 1 and 2 (49.8% for those that were SxCl(+) and 34.9% for those that were SxCl(−)) and fell sharply thereafter, as depicted in Fig. 5, panel A. On an overall basis, more patients transitioned from SxCl(+) to death than from SxCl(−) to death.

Figure 4.

Frequencies of transitions between symptom cluster-defined states and death over the first five years following lung cancer diagnosis. SxCl = symptom cluster.

Figure 5.

Alterations over time first five years following lung cancer diagnosis in the probability of dying and transitioning between symptom cluster-defined states. SxCl = symptom cluster.

SxCl stability, as reflected by a lack of in between-year “inter-state” transitions increased steadily over time, most markedly for the SxCl(+) state (Fig. 5, panel B). Between years 1 and 2, 18.8% of patients remained SxCl(+). In contrast, between years 4 and 5, 49.2% of patients remained SxCl(+). The decrease in intra-state transitions over time was less pronounced for the SxCl(−) state yet still notable. Between years 1 and 2, 58.0% of patients remained SxCl(−), in contrast to the interval between years 4 and 5 when the percentage of patients remaining SxCl(−) increased to 81.1%. . The rates of inter-state transitions, i.e., to or from SxCl(+), were remarkably stable over time, as illustrated in Fig. 5, panel C. The percentage of patients transitioning from SxCl(−) to SxCl(+) was consistently low, ranging from 7.1% to 8.6%. Although higher, transition rates from SxCl(+) to SxCl(−) also remained constant over time, ranging from 31.5% to 37.8%.

Determinants of State Transitions

The results of Markov Chain Monte Carlo models of “to death” and inter-state transitions are listed in Table 4. Examining specific transitions between SxCl-defined states across all study years revealed that undergoing surgery significantly lowered a patient’s probability of dying but increased their probability of inter-state transitions, both SxCl(+) to SxCl(−) and visa versa. Male survivors had a higher probability of transitioning to the SxCl(+) state and, if SxCl (+), a lower probability of transitioning to the SxCl(−) state. Being married significantly increased the probability of transitioning from the SxCl(+) state and reduced the probability of death among SxCl(−) patients. Receipt of radiation and current smoking increased the probability of transitions to the SxCl(+) state. Older age and the presence of Stage IV lung cancer increased participants’ probability of dying. Lung cancer recurrence significantly increased the probability of death among SxCl(−) patients.

Table 4.

Odds Ratios and 95% Confidence Intervals of Demographic, Disease and Treatment Characteristics for Transitions Between Symptom Cluster-Defined States and Death, Based on 500,000 Markov Chain Monte Carlo Replications

Variable	SxCl(+) → SxCl(−)	SxCl(+) → Dead	SxCl(−) → SxCl(+)	SxCl(−) → Dead
Any Surgery	4.69 (1.62-11.91)	0.16 (0.05-0.33)	2.91 (1.44-5.09)	0.15 (0.10-0.22)
Stage IV*	0.34 (0.08-0.93)	4.05 (1.21-10.02)	1.34 (0.64-1.74)	5.17 (3.08-7.92)
Male	0.42 (0.18-0.82)	1.63 (0.69-3.28)	1.69 (1.15-2.46)	1.15 (0.84-1.52)
Married	2.44 (1.03-5.06)	0.76 (0.29-1.75)	0.98 (0.63-1.51)	0.70 (0.49-0.99)
Age at Diagnosis†	1.00 (0.97-1.03)	1.05 (1.01-1.09)	1.02 (0.99-1.04)	1.03 (1.02-1.05)
Any Chemotherapy	1.19 (0.46-2.58)	2.84 (1.06-6.63)	1.14 (0.64-1.81)	1.14 (0.74-1.64)
Any Radiation	1.12 (0.47-2.25)	0.99 (0.39-2.04)	2.05 (1.27-3.20)	1.00 (0.66-1.43)
Caucasian Ethnicity	2.09 (0.47-6.62)	1.31 (0.26-4.42)	3.17 (1.10-8.37)	1.98 (0.86-4.09)
Current Smoker	0.99 (0.40-2.05)	1.23 (0.43-2.69)	2.46 (1.34-3.97)	0.89 (0.52-1.38)
Recurrent Lung Cancer	0.69 (0.26-1.42)	1.41 (0.52-3.20)	0.69 (0.37-1.15)	3.93 (2.60-5.84)
Beyond High School Education	1.53 (0.81-2.75)	0.82 (0.52-3.20)	0.81 (0.56-1.13)	0.94 (0.71-1.24)
Pack Years†	1.00 (0.99-1.01)	1.00 (0.99-1.01)	1.01 (1.00-1.01)	1.00 (0.99-1.01)
Stage III*	0.61 (0.25-1.29)	2.15 (0.86-4.63)	1.10 (0.64-1.74)	1.55 (0.99-2.30)

^*Relative to Stage I non-small cell lung cancer.

^†Continuous variables.

COMP: PLS REPLACE ASTERISK and DAGGER IN TABLE & LEGEND WITH SUPERSCRIPT ITALIC a and b.

Discussion

This study identified a SxCl of fatigue, dyspnea and cough that persisted across cluster, factor, and latent trait analyses, and characterized its persistence over time in an eight-year cohort of LC survivors. In so doing, the study yielded a number of observations with potentially important implications for clinical practice and research. First, while presence of the SxCl was not necessarily stable in an individual over time, transition rates between SxCl(+) and SxCl(−) states in the group as a whole remained constant throughout the study period and, after year 4, the SxCl was less likely to resolve in surviving patients. Second, developing the SxCl was more likely among male survivors who underwent surgery, received radiation, and were current smokers. The latter characteristics implicate pulmonary injury as a sustaining factor for the SxCl.

Our results accord with reports of poor prognoses in LC patients manifesting SxCls during the first year. ^{8, 11, 12, 32} A symptom cluster of fatigue, dyspnea, and cough has been previously identified through quantitative³² and qualitative methods³³ among patients with LC up to 12 months following their diagnoses. Our findings add to this work by demonstrating this SxCl’s strong association with death during the first one to two years following a lung cancer diagnosis, as well as its increasing stability and diminishing association with death over time. In addition, our findings may allow more precise estimation of the SxCl’s association with demographics, disease characteristics, treatment and medical comorbidities because our study cohort was not only followed longer but is also significantly larger, n=2405, and more extensively characterized than previously described LC cohorts.

While our analyses suggested that other clusters, e.g., pain or sleep disturbance with emotional disturbance, might be present, these clusters did not meet our a priori criteria of an eigenvalue >1. We assessed survivors for the symptoms described by prior authors, e.g., appetite disturbance, mood, pain, yet failed to detect previously identified clusters.^{10, 12} This may be, in part, because our first sampling point occurred farther along in the LC trajectory than the sampling time frame utilized by many investigators, and many of our participants had completed cancer treatments prior to enrollment. This may suggest that some SxCls resolve following active LC treatment.

Despite a remarkably stable prevalence of roughly 15% among LC survivors in this study, the SxCl was not stable over time and frequent transitions occurred between SxCl-defined states. High transition rates between the SxCl(+) and SxCl(−) states are not surprising given the fact that patients’ LC treatments were frequently changing, particularly after year 1 when some patients recovered from aggressive cure-oriented therapies. Additionally, patients’ disease burden changed over time, with some experiencing disease progression. Given these dynamic processes, the stability of transition rates between the SxCl(+) and SxCl(−) states (collectively 40% - 45% of survivors each year) across all study years is notable and contrasts with the decreasing rates of “to dead” and increasing rates of intra-state transitions.

The patterns and determinants of “to dead” and inter-state transitions may shed some light on the etiology of the SxCl. The fall in the “to dead” transitions and the rise in state stability among both SxCl(+) and SxCl(−) patients with time (see Fig. 5, panels A and B) may indicate that patients with greater disease burden are more likely to die in the first years following diagnosis, which accords with current survival estimates. ^{34, 35} However, the proportion of SxCl (+) patients who died dropped markedly after year 2 and still further after year 4. This finding suggests that the SxCl may indicate a transitional phase prior to LC death experienced by patients with worse prognoses in the first years following an LC diagnosis. However, over time the SxCl becomes increasingly stable, possibly in association with chronic pulmonary compromise. This inference is supported by the increased prevalence of COPD among SxCl(+) patients, as well as the higher proportion of current smokers, radiation recipients, and lung resection patients transitioning to the SxCl(+) state. For clinicians, the implication would be that this SxCl suggests a poor prognosis in the near-term but, with extended survival, the SxCl(+) is increasingly likely to arise from chronic lung injury and that appropriate work-up and therapies should be initiated.

Limitations

Symptom clusters are an evolving construct and there are no specified, optimal analytic methods for their discovery and validation.³⁶ Hence, we adopted a sensitivity approach, running multiple exploratory and confirmatory models to test the consistency of our results in the presence of varying model assumptions. Each approach served overlapping analytic purposes; however, model structures varied substantially. Reassuringly, all yielded consistent results and overwhelmingly support the presence of a single SxCl.

Our 79% response rate was high. Nonetheless, participants who provided symptom data differed in their demographics, tobacco use, LC characteristics and co-morbidities from those eligible subjects who did not. This is a common occurrence in quality of life studies.³⁷ The fact that non-participants were more likely to have higher stage LC and to be current smokers suggests that those who did not provide data may have had greater symptom burden. In this case, our data could underreport the impact of the SxCl. One could argue, however, that patients who were too ill to complete a questionnaire would be readily identified without the need for a complex identification of SxCl, making the present sample more relevant and useful to potential applications in clinical practice. Additionally, our requirement that a symptom be rated at an intensity of 5/10 is admittedly stringent. Prior authors have utilized lower threshold ratings.⁹ Our results must be interpreted in light of the fact that SxCl(+) survivors were severely burdened by their symptoms.

Conclusions

A single SxCl comprised of dyspnea, fatigue, and cough has a stable roughly 15% prevalence among LC survivors up to five years following diagnosis but is not stable among individuals, with the SxCl resolving in approximately 35% and developing in approximately 7% of survivors each year. In the first one to two years following a LC diagnosis, the SxCl appears to be a marker for sicker patients at greater risk of death; however, after year 2 the SxCl becomes increasingly stable and a marker for parenchymal lung injury.