Symptom Experience in the Last Year of Life Among Individuals with Cancer

Abstract

Individuals with cancer often experience many symptoms that impair their quality of life at end of life. This study examines symptom experience at end of life among individuals with cancer, and determines if symptom experience changes with proximity to death, or differs by depressive symptomatology, sex, site of cancer, or age. A secondary analysis of data from three prospective, descriptive, longitudinal studies (n = 174) was performed, using a three-level hierarchical linear model. Fatigue, weakness, pain, shortness of breath, and cough were the five most prevalent symptoms in the last year of life. The symptom experience in the last year of life was significantly associated with site of cancer, depressive symptomatology, dependencies in activities of daily living, and independent activities of daily living at the start of the study. These findings shed light on the symptom experience in the last year of life for individuals with cancer. With greater understanding of the symptom experience, intervention strategies can be targeted to achieve the desired outcome of increased quality of life at end of life.

Introduction

As cancer is one of the leading causes of death among Americans (1), it is important to understand the symptom experience among individuals with cancer approaching end of life. Individuals with cancer often suffer from many symptoms that impair their quality of life at end of life (2). Correspondingly, one goal of palliative end-of-life care is to provide symptom management; however, symptoms have been reported as poorly managed at the end of life (3). This inability to manage symptoms at the end of life may be due in part to a limited understanding of symptoms and the factors that are associated with symptoms at the end of life.

Improvements in symptom management techniques have been hampered by a dominant research focus on a single symptom (4). Symptoms, however, do not occur in isolation; rather, individuals with cancer have multiple, coexisting symptoms. To better understand the experience at end of life, a more complete picture of the symptom experience needs to be assessed, which in turn requires appropriate statistical methodology.

Current understanding of symptom experience at end of life derives from studies recruiting individuals with cancer, designated as terminally ill, having a prognosis of less than six months, or receiving hospice or palliative care (5–7). Recruitment of terminally ill individuals with cancer has given us great insight into the symptom experience at end of life; however, the current focus on individuals designated as terminally ill and receiving hospice or palliative care provides only a partial understanding of the symptom experience at end of life among individuals with cancer. The results of extant studies can neither be generalized to those whose terminal illness is either unrecognized or unacknowledged, nor to those whose response to a terminal prognosis involves non-palliative, non-hospice approaches.

Furthermore, methods for the collection of symptom experience data at end of life have included proxy and retrospective interviews with caregivers. Proxy interviews are administered to caregivers of individuals nearing end of life, to relieve the respondent burden for those who are terminally ill (8,9). Proxy reports have been shown to be quite accurate for observable symptoms such as vomiting; however, the agreement between self and proxy reporting has been shown to be unreliable for symptoms that are less observable, such as pain or depression (8). Retrospective interviews with caregivers of the deceased individuals with cancer also have been used to understand the symptom experience at end of life (10), but the use of retrospective interviews may be subject to a significant recall bias. Thus, the use of proxy or retrospective interviews may not provide the most accurate understanding of the symptom experience at end of life.

In prior research (involving 1,000 individuals having advanced cancer upon initial referral to a Palliative Medicine Program), age, sex, and activities of daily living (ADLs) / instrumental ADLs (IADLs) were related to symptom experience (12). Previous research has also found a relationship between depression and symptom severity (13). Additionally, certain sites of cancer are known to have a shorter life expectancy than others, for example, lung cancer compared to breast cancer (1). The shorter life expectancy may relate to higher symptom experience in the last year of life.

Prior research is lacking information on how symptom experience changes in the last year of life among individuals with cancer who are not enrolled in a specialized palliative care or hospice setting. As death approaches, the symptom experience may evolve differently for each individual with cancer, and may be related to personal and health/illness characteristics (such as age, sex, ADL/IADLs, depression, and site of cancer).

This paper examines the symptom experience trajectory during the last year of life among individuals with cancer and whether it differs by depressive symptomatology, dependencies in ADLs or IADLs, sex, site of cancer, or age. Participants were enrolled in one of three longitudinal studies and were asked to respond to a common list of 21 symptoms. This exploration extends our understanding of the symptom experience at end of life in several important ways. First, the method of data collection is patient self-report. Second, the study focuses on a sample of individuals with cancer who were followed prospectively, were receiving chemotherapy, and were not enrolled in hospice. Finally, it recognizes the importance of looking at a multiplicity of symptoms at end of life. Thus, this paper, by taking a prospective view of symptoms, provides a different picture of the symptom experience at end of life among individuals with cancer than what is currently available and broadens the understanding of the symptom experience at end of life.

According to the Symptom Management Conceptual Model, which guided this inquiry (11), three domains (person, health/illness, and environment) affect and modify the three dimensions of the model (symptom experience, components of symptom management strategies, and outcomes). In this study, two domains are examined: person, and health/illness. The person domain that may influence the symptom experience includes individual characteristics such as age and sex. Characteristics from the health/illness domain include depressive symptomatology, site of cancer, and ADL/IADL. Characteristics from the environment domain were not included in this analysis as they were not present in the data sets. The outcome for this study was the individual’s perception of the presence of symptoms, a component of the symptom experience dimension.

Methods

Data and Participants

This secondary data analysis employs pooled data from cancer patients who died (n = 174) during one of three descriptive longitudinal studies. Inclusion criteria for all studies required that the individuals have a diagnosis of cancer, be cognitively intact, and be able to speak, read, and write English. Individuals under the care of a psychologist or psychiatrist, or with a diagnosed emotional or psychological disorder, were excluded.

Specific inclusion criteria for Rural Partnership Linkage For Cancer Care (CA56338, Rural study, n = 159) included patients over 21 undergoing treatment for a solid tumor cancers, who resided in rural areas served by a National Cancer Institute (NCI) designated Community Cancer Oncology Program (CCOP).

Specific inclusion criteria for Family Home Care of Cancer – A Community-Based Model (NR01915, Community study, n = 1,150) were cancer patients newly diagnosed with breast, colon, lung or prostate cancer, who were 65 years of age or older. For this study, participants were recruited from 24 community facilities in Michigan.

Specific inclusion criteria for Family Home Care for Cancer Patients (#PBR-32, Cancer study, n = 192) were patients between 20 and 80 years of age who had at least one dependency in an ADL or an IADL. Patients could be newly diagnosed or have recurring cancer. Participants were recruited from six community-based cancer treatment centers covering cities ranging in size from 20,000 to 500,000 and their surrounding rural areas in lower Michigan.

For all studies, nurse recruiters approached individuals who met the inclusion criteria, explained the studies, and obtained written consent. Data collection occurred by telephone and by mailed survey. The timing of the interviews varied by study and can be seen in Table 1.

Table 1.

Timing of Interview Data by Study

Study	Baseline interview or date of diagnosis of cancer	2 months	3 months	4 months	6 months	9 months	12 months
Cancer study			x		x	x	x
Community study		x		x	x		x
Rural studya	x		x		x	x

^aRural study recruitment was not at diagnosis of cancer

Dates of death were confirmed by matching names, addresses, and social security numbers with death certificates obtained from the state Division of Vital Records. Dates of death were collected up to one year after the end of the study. Thus, the span of time from final interview to death could extend up to one year.

Measures

Symptoms were assessed using the self-report Symptom Experience Tool (14). Participants responded regarding the presence of 21 symptoms commonly experienced by individuals with cancer, indicating whether they experienced the symptom (= 1) or not (= 0).

ADLs and IADLs were assessed using the modified index of ADL/IADL (15,16). Nine questions were asked regarding participants’ independence (= 0) or dependence (= 1) with activities such as dressing, eating, walking, and transportation. A summary score was then created, with higher scores indicating more dependence in ADL/IADLs.

Depressive symptomatology was assessed using the Center for Epidemiologic Studies Depression (CES-D) measure, a 20-item assessment tool. Scoring of this instrument is on a 0 to 3 scale for each item, with the sum across the 20 items representing the level of depressive symptomatology. Sums thus range from 0 to 60, with a score of 16 or higher representing clinically significant depressive symptomatology (17).

Time was coded in days from the last interview until death. Other information included age, sex, and site of cancer. The sites of cancer in participants consisted of lung, breast, colon, prostate, or other solid tumor cancers. For analysis, cancer sites were combined into two groups (lung and other), where other was the reference category.

Analysis

Preliminary Analysis: Testing of Assumptions

First, we examined a set of individuals who dropped out of the studies before completing the interview prior to death (n=38) to those who completed the final interview (n=136) using two-sample t-tests for continuous variables and χ²-tests for categorical variables, to determine if there were any discernable differences between the two groups. No significant differences in age, sex, site of cancer, depressive symptomatology, or ADL/IADLs were found. As no differences were found between groups, the missing at random assumption appears to be reasonable (18).

Second, we examined the assumptions of the Item Response Theory (IRT) Rasch model, including equal discrimination and the unidimensionality (19). In order to determine if symptoms were equally discriminating, a 2-parameter IRT model, which estimates discrimination as well as the difficulty of each symptom, was fit to the full sample (all individuals at all time points), as well as four subsamples based on the time of the observation until death (0–99, 100–199, 200–299, and 300–400 days until death) using BILOG-MG. The fit of the 2-parameter model was compared to that of the 1-parameter (Rasch) model using the Bayesian Information Criteria (BIC). In all five comparisons, the 1-parameter model had a smaller BIC value than the 2-parameter model, indicating that the 1-parameter model was a better fit. Note that for the purpose of the assumption testing, if symptom presence or absence was not indicated, it was assumed that the symptom was not present.

To examine the unidimensionality assumption, we looked at differential symptom functioning over time. First, a non-parametric approach was used by examining the ordering of symptoms due to their estimated difficulty in the Rasch model for the five samples mentioned above. Second, a parametric approach was used by comparing the fit of two hierarchical linear models (HLM): one model in which time was included in each equation at level 2, and another model in which time was included only in the intercept equation at level 2. No statistically significant difference between the two models was found (χ²(df=20) = 24.00, P = 0.24).

Hierarchical Linear Modeling Analysis

The main analysis for this study embedded a one-parameter IRT (Rasch) model into a hierarchical linear model. A detailed description of how to set up an IRT Rasch model in an HLM framework has been discussed previously (20, 21, 23). Embedding IRT into an HLM framework produced several benefits, including the creation of a latent symptom experience variable, examination of the latent symptom experience trajectory and its relationship with covariates at the individual level, as well as handling the data on individuals who were missing one or more observations. Application of these analytic techniques advances our abilities to examine longitudinal symptom outcomes of individuals at the end of life.

Two models, unconditional and conditional, were estimated using HLM 6.21 (24). The unconditional model has 20 dummy variables for all but the reference symptom (fatigue) at level 1 and no covariates at level 2 or level 3, thereby yielding a readily interpretable ordering of symptoms as well as unadjusted symptom experience estimates for each person on each occasion that at least one symptom’s presence or absence was recorded. The symptom experience over time is examined by testing linear and quadratic trajectories at level 2. To test the relationships between individual variables and symptom experience, these variables were incorporated into the model at level 3. In the final model, level 1 remains the same as in the unconditional model; however, a time-varying variable (days from interview until death) is added at level 2, and the time-invariant individual variables (age, sex, ADL/IADL at baseline, depressive symptomatology at baseline, site of cancer, and study) are added at level 3. To make the intercept more readily interpretable, age, depressive symptomatology, and dependency in ADL/IADL were centered on their corresponding grand means (70.79, 15.06, 2.31, respectively). Differences in study membership were controlled by creating dummy variables for each study and entering them in the model. The community study was used as the reference, as it had the greatest number of individuals who died.

Results

Sample

Descriptive information on the 174 individuals who died during or within one year after completion of one of the three prospective, descriptive, longitudinal studies can be seen in Table 2 by study membership. Study participants were similar with respect to sex (overall 64% (n = 111) of the sample were male) and depressive symptomatology distributions. Sixty-two percent of the sample had a diagnosis of lung cancer (n = 108). Overall, individuals ranged in age from 38 to 91 years, with a mean of 71 years. There were differences in study membership with respect to age, with the Rural study having a greater age range and a lower mean age than the other two studies; this was in part due to the inclusion criteria of the Community study being age 65 or older. ADL/IADLs were significantly lower for the Cancer study participants compared to the Rural and Community studies.

Table 2.

Descriptive Characteristics at Entry into Study

	Rural study n = 27		Community study n = 112		Cancer study n = 34
	Min - Max	Mean (SD)	Min - Max	Mean (SD)	Min - Max	Mean (SD)
Age	38 – 82	62.6 (11.7)a	65–91	72.6 (5.63)	62–91	71.1 (6.66)
Depressive symptomatology	1 – 37	15.3 (8.10)	0–37	14.9 (7.93)	2–35	15.6 (9.13)
ADL/IADL	0 – 7	2.6 (2.31)	0–8	2.4 (2.15)	0–5	1.5 (1.68) a
	n	%	n	%	n	%
Sex
Male	16	59	73	65	22	65
Female	11	41	39	35	12	35
Site of cancer
Breast	3	11	2	2	1	3
Colon	5	19	15	13	1	3
Lung	10	37	87	78	11	32
Prostate	1	4	8	7	2	6
Other cancers	8	30	0	0	11	32

^aP < 0.05 difference compared with other studies

Of the 174 participants, 60 completed one interview, 70 completed two interviews, 36 completed three interviews, and 8 completed all four interviews. Participants’ final interview prior to death ranged in time from two days to 365 days prior to death, with a mean of 102 days (Figure 1). Thirty-four percent of the participant’s last interviews occurred within 60 days of death, and 12% occurred within 30 days of death.

Figure 1.

Histogram for the day of last interview prior to death.

Results of the Unconditional Model

The unconditional model described the symptom experience at end of life. Fatigue, the most common symptom in the raw data, was used as the reference symptom. The results of the unconditional model yield a readily interpretable ordering of symptoms. Figure 2 shows the symptoms organized by their prevalence; the more prevalent symptoms appear at the top (high values), while less prevalent symptoms appear at the bottom (low values). Symptoms appearing close together in Figure 2 have similar prevalence. The most prevalent symptom, fatigue, was followed by weakness, then pain. The least prevalent symptom during the year before death among participants was dehydration.

Figure 2.

Symptom prevalences according to the unconditional model.

Results of the Conditional Model

Since our focus was to determine the symptom experience trajectory as well as its association with important individual variables (depressive symptomatology, ALD/IADLs, sex, site of cancer, and age), partial output for level 2 and 3 is reported in Table 3. Days from an interview until death ranged from 2 days to 365 days. Both the linear and quadratic trajectories were tested; however, the quadratic term did not significantly improve the model fit. Therefore, a linear trajectory was used in the final model. After controlling for other covariates, no significant difference was detected in symptom experience with proximity to death (γ^₀₁₀ = −0.076, P = 0.681).

Table 3.

Partial Output of Estimates for the Conditional Model (Excluding Symptoms’ Prevalence)

Variable	Coefficient	Standard Error	P-value
Intercept, γ000	1.199	0.194	< 0.001
Female, γ001	0.028	0.117	0.808
Age, γ002	−0.005	0.009	0.558
Lung, γ003	0.411	0.132	0.003
Rural study, γ004	0.318	0.204	0.121
Cancer study, γ005	−0.308	0.163	0.061
Depressive symptomatology, γ006	0.047	0.007	<0.001
Dependencies in ADL/IADL, γ007	0.082	0.035	0.022
Proximity to death, γ010	−0.076	0.184	0.681

After controlling for other covariates, significant differences in symptom experience by site of cancer were detected. Individuals with lung cancer experienced significantly higher symptom experience than individuals with other solid tumor cancers (γ^₀₀₃ = 0.411, P = 0.003). Thus, individuals with lung cancer experienced more symptoms in their last year of life than those who had other solid tumor cancers. Moreover, controlling for other covariates, there was a significant difference (P < 0.001) in symptom experience between participants who differed in depressive symptomatology at baseline. Higher depressive symptomatology at baseline was associated with greater symptom experience during the last year of life (γ^₀₀₆ = 0.047). Controlling for other covariates, dependency on ADL/IADL scales at baseline were found to be significantly associated with symptom experience (P = 0.022). In particular, individuals with higher levels of dependency at baseline tended to have worse symptom experience in the last year of their life (γ^₀₀₇ = 0.082). After controlling for other covariates, no significant difference was detected in symptom experience between males and females (γ^₀₀₁ = 0.028, P = 0.808), nor was age found to be significantly associated with symptom experience (γ^₀₀₂ = −0.005, P = 0.558).

Discussion

The purpose of this study was to describe the symptom experience at end of life among individuals with cancer and to determine if symptom experience differs with proximity to death, depressive symptomatology, sex, site of cancer or age, after controlling for study membership. A contribution of this study is the successful application of a new technique of analysis for longitudinal studies where symptom data are collected. Additionally, as individuals of this analysis were recruited at the time of diagnosis or during chemotherapy treatment, and had not been referred to hospice, this research provides a view of the symptom experience at end of life among individuals with cancer not previously well explored.

The five most prevalent symptoms in this population of cancer patients at end of life were fatigue, weakness, pain, shortness of breath, and cough. Direct comparisons are difficult to make with what has been reported in the extant literature describing individuals with cancer in palliative care settings, due to differences in the symptom assessment instruments used. However, Stromgren and colleagues (7), using the European Organization for Research and Treatment of Cancer Quality-of-Life Questionnaire (EORTC QLQ-C30), reported that among 176 individuals with advanced cancer admitted into palliative care, the most prevalent symptom was fatigue, followed by inactivity, and pain. This suggests that the symptoms experienced among individuals with cancer who are not recruited from hospice and palliative care services may be comparable to those receiving hospice and palliative care.

Most extant longitudinal research has been conducted with individuals receiving hospice or palliative care and has assessed symptoms only in the last weeks or months before death (6, 25–27). These studies report that there was a worsening of symptom distress in the last month or weeks before death. The results of our study indicate that there was not a significant worsening of symptom experience as individuals with cancer approached the end of life. These results may appear counterintuitive at first glance, but may be due to two factors. First, studies used for this secondary data analysis had in common only the measure of the presence or absence of a symptom and did not have common measures of symptom severity or symptom distress. While symptoms may not be significantly increasing in prevalence at end of life, their severity or the distress may change as end-of-life approaches. Second, 12% of the interviews were conducted within one month of death. Since previous research has reported that changes in symptoms are seen only very close to death, there may not have been enough interviews or interviews were not frequent enough close to death to detect significant changes in symptom experience. Future research is needed among individuals with cancer, who are approaching end of life but not admitted to palliative care services, to ascertain if other symptom dimensions such as severity or distress along with more frequent symptom assessment may yield richer or different results.

Characteristics such as sex and age were not significantly associated with the overall end-of-life symptom experience. These results are contradictory to previous findings, which reported that being older or female were associated with higher symptom prevalence for specific symptoms (12). The Symptom Management Conceptual Model states that person variables are intrinsic to the way an individual views and responds to the symptom experience; thus, only measuring symptom presence or absence may not have been adequate to assess how an individual views and responds to the symptom experience. Future research is needed to assess if, when additional dimensions of the symptom experience such as symptom severity or limitation are added, the person variables will be seen to impact the symptom experience in the last year of life.

In this study, health and illness variables included site of cancer, ADL/IADLs, and depressive symptomatology. As conceptualized by the Symptom Management Conceptual Model, health and illness variables had a direct effect on the symptom experience. Significant differences were seen between individuals with a diagnosis of lung cancer and those with other solid tumor cancers in the symptom experience. These findings further support previous findings by Donnelly and colleagues (28). Examining 37 symptoms among 1,000 individuals with advanced cancer admitted to palliative care services, they found significant differences in symptom prevalence between cancer sites. Thus, it is likely that health care providers will see differences in end-of-life symptom experience in individuals with different diagnoses.

Past work has demonstrated a relationship between symptoms and physical functioning (27, 29). Our findings extend this understanding of the relationship between symptoms and physical function at end of life. The finding that increased dependencies in physical functioning at initial assessment are associated with increased symptom experience further suggests that there is a reciprocal relationship between symptom experience and physical functioning which should be explored more fully.

Our results support and extend other research findings that suggest a strong relationship between depressive symptomatology and symptoms (30). This study extends previous findings on the relationship between depressive symptomatology and symptom experience to individuals with solid tumor cancers receiving chemotherapy near end of life, a group at high risk for worsening symptom experience. This study provides evidence that depressive symptomatology at the start of the study was associated with poorer symptom experience. Furthermore, individuals with high initial depressive symptomatology continued to report worse symptom experience than individuals with low depressive symptomatology as death approached. The participants of this study were recruited while receiving chemotherapy and not enrolled in hospice; however, we do not know whether or not they were told if they had a terminal prognosis. The knowledge of the terminal prognosis may have influenced depressive symptomatology in the participants. Future research is needed to clarify the directionality of the relationship between depression and symptoms in order to provide guidance to palliative care practitioners on how to improve the quality of life at the end of life for individuals with cancer.

Limitations

Although this analysis pooled 3 longitudinal descriptive studies, the sample size (n = 174) remains small. Increasing the sample size would allow greater differentiation between the various cancer diagnoses. Additionally, having a greater number and frequency of time points, especially closer to death, would allow for greater precision in the determination of the symptom experience trajectory.

Underestimation of the symptom experience at end of life may be present in this report, as those with the worst physical function deficits may not have enrolled in the studies or been able to respond to the interview closest to death. However, the potential response bias for those who did not respond to the interview closes to death can be viewed as minimal, since the 38 individuals who did not complete the interview prior to death were not significantly different in either personal or illness variables from those who did complete the interview prior to death.

As this was a secondary analysis of a data set, not all variables of interest that may influence symptom experience were available for analysis. Availability of additional covariates in the models tested (e.g., co-morbidities and stage of cancer) would enhance our understanding of symptom experience at end of life; however, those that were used are pertinent and further our understanding of the symptom experience at end of life among individuals with cancer not receiving hospice care.

Conclusions

This exploration extends our understanding of the symptom experience at end of life in several important ways. It examined a multiplicity of symptoms using data from three prospective studies. This previously unexplored sample consists of individuals with cancer in the last year of life who were receiving chemotherapy and not receiving hospice care. Thus, these findings allow us to be more confident about generalizing to the larger population of all cancer patients at end of life. Future research is needed to include greater detail in the symptom experience dimension, such as the assessment of symptom severity. With greater understanding of the symptom experience, intervention strategies reflected in the model as the domain of symptom management can be targeted to achieve the desired outcome of increased quality of life at end of life.