Differential Impact of Symptom Prevalence and Chronic Conditions on Quality of Life in Cancer Survivors and Non-Cancer Individuals: A Population Study

In 2014, approximately 15 million Americans were cancer survivors, and this number will reach 19 million by 2024 (1). Following anticancer therapy, survivors may develop late effects inclusive of second malignancy, various chronic conditions, and physical and psychological symptoms. Maintaining normal daily activity and quality of life is the primary goal of cancer survivorship.

A recent study based on the National Health Interview Survey (NHIS) found that 25% and 10% of cancer survivors in the United States had suboptimal physical and mental health–related quality of life (HRQOL), respectively (2). Although vulnerable demographic and treatment factors have been associated with impaired HRQOL (2), these factors are antecedent to survivors' HRQOL. In contrast, symptom prevalence and chronic conditions are deemed to be the proximal factors on the impact of daily functioning or quality of life (3).

Previous research suggests that cancer survivors without chronic conditions had equivalent or better HRQOL than general population with chronic conditions (4, 5); yet survivors had decremented HRQOL if they developed chronic conditions. Compared with the general population, cancer survivors with more chronic conditions are more likely to co-occur different symptoms (6). Indeed, the mechanisms between the prevalence of multiple symptoms and chronic conditions are complicated, and the relative impact of symptom prevalence and chronic conditions on HRQOL in cancer survivors are understudied. In an era of patient-centered oncological care, specific attention has been paid to address the issues related to symptom burden that represent one's perceived abnormal physical, emotional, cognitive, or psychosomatic state (7, 8).

Symptom burden has been commonly investigated in cancer patients during and shortly after completion of therapy but not in long-term survivors (5 years after diagnosis). Symptoms resulting from anticancer therapy can persist over time, and new symptoms may emerge decades after therapy completion. A rehabilitation study found that survivors 10 years after diagnosis had symptom prevalence above the severe threshold, including fatigue (65%), sleep disturbance (46%), lack of concentration (45%), and joint/muscle pain (44%; ref. 9). Symptom prevalence in cancer populations, especially fatigue, has a strong predictive validity for survival (10). Our previous study demonstrated that prevalence of 12 symptoms (e.g., pain, perceived cognitive problems, distress, and sensation abnormalities) in adult survivors of childhood cancer was independently associated with decreased HRQOL (11). Prevalence of these symptoms explained 60% and 56% of the variance in the SF-36′s physical and mental summary scores (PCS/MCS), respectively, whereas demographic and clinical variables explained up to 15% and 10% of the variance in PCS and MCS, respectively. However, how socio-demographic factors, cancer experience (i.e., survivors vs. non-cancer individuals), and chronic conditions are related to symptom prevalence, as well as the extent to which these factors collectively impact HRQOL were not evaluated in the previous study.

It is our hope to treat symptoms as an avenue toward improving daily activity and HRQOL of cancer survivors. Clinical implications for this initiative can be intensified if the prevalence of symptom phenotypes is high in cancer survivors versus non-cancer individuals and if the association of decreased HRQOL with cancer experience is significantly explained by symptom prevalence. Previous HRQOL studies in cancer survivorship often used convenience or matched samples as comparison groups (12); however, most studies have not taken symptom prevalence into consideration. It is therefore not surprising that mixed evidence has emerged (13), including worse (12, 14) and equivalent (15, 16) HRQOL between survivors and comparison groups.

This study addressed two specific aims: The first aim was to compare the prevalence of six symptom phenotypes (sensation abnormality, pain, fatigue, cognitive disturbance, depressive symptom, and anxiety) between survivors of adult cancers and non-cancer individuals, and identify the determinants for the prevalence of individual symptoms. The second aim was to examine the association between cancer experience and HRQOL by accounting for the influence of the six selected symptom prevalence, chronic conditions, and socio-demographic factors. We hypothesize that the prevalence of individual symptom phenotypes is higher and HRQOL is worse in cancer survivors than non-cancer individuals. Additionally, we hypothesize that both symptom prevalence and co-occurring chronic conditions will significantly impact HRQOL regardless of the status of survivorship, and the effect of multiple symptoms on HRQOL will be larger than that of chronic conditions. This study used 2010 NHIS data comprised of representative cancer survivors and non-cancer non-institutionalized individuals in the United States; therefore, our findings can be generalized to the non-institutionalized U.S. population.

NHIS is an annual survey conducted by the U.S. Center for Disease Control and Prevention through computer-assisted personal interviews to investigate health status and utilization issues (17). A national representative sample using households was selected by means of a complex sampling framework.

The survey is comprised of three core components (family, sample adult, and sample child) and different supplements focusing on specific public health issues. In this study, family and sample adult components as well as cancer control and quality of life supplements were used. The cancer control supplement is completed every 5 years and the quality-of-life supplement was completed in 2010. All members of the household 18 years of age and above who were at home at the time of the survey were invited to complete the family component. One adult per household was randomly selected to complete the sample adult component and cancer control supplement; one quarter of the sample adults were randomly selected to complete the quality of life supplement.

Given the use of secondary, de-identified data from the U.S. federal government, which are also publicly available, a waiver was granted by Institutional Review Board of University of Florida where the study was conducted.

Because items capturing the concept of symptom prevalence are scattered throughout the NHIS and standard symptom tools are not used by the NHIS, we developed a three-step strategy to identify appropriate symptom items. The first step was to create a list of symptom phenotypes based on the framework of our previous study (11). The second step was to search for the items measuring symptoms listed in the sample adult component and quality-of-life supplement of the questionnaire. Individual items were independently documented using an extraction form by the first author and a research coordinator. Each item was assigned to one of the symptom phenotypes based on the previous study's framework (11), and then classified by the following attributes: symptom presence, symptom intensity, and symptom behavior. Only items measuring the attribute of symptom presence were included in this study. The third step was for the two co-authors (I-C. Huang and K.R. Krull) to evaluate the quality of individual items and select appropriate items. Discrepancies were adjudicated by consensus in a group meeting.

The research team established a priori criteria to set aside items containing a vague concept of symptoms or having no clear time frame. In total, 18 out of 65 items were retained to capture the prevalence of the six symptom phenotypes: sensation abnormality (6 items), pain (6 items), fatigue (1 item), cognitive disturbance (3 items), depression (1 item), and anxiety (1 item; Table 1). Symptom presence was denoted if individuals endorsed “yes,” “some days/most days/every day,” or “some difficulty/a lot of difficulty” for items measuring a specific phonotype.

HRQOL was self-reported using the Global Health Scale of the Patient-Reported Outcomes Measurement Information System (PROMIS). PROMIS applies modern test theory to develop reliable, valid, flexible, and precise tools for assessing HRQOL (18). The Global Health Scale is a 10-item scale measuring physical and mental HRQOL (19). The scores of individuals' physical and mental HRQOL were calculated and normalized to T-scores (mean = 50, SD = 10). Higher scores indicate better HRQOL.

Several important socio-demographic characteristics collected from the NHIS were used in analyses. Race/ethnicity was grouped as Hispanic, non-Hispanic White, non-Hispanic Black, Non-Hispanic Asian, and non-Hispanic other. Marital status was classified as separated, divorced, married, single/never married, and widowed. Education was classified as below high school, high school/GED, some college, bachelor degree, and graduate degree. For clinical variables, chronic conditions were self-reported by cancer survivors and non-cancer individuals; types of cancer and anticancer therapy were self-reported by cancer survivors. Years since cancer diagnosis was calculated based on the difference between year at diagnosis and year at interview. Survivors with <5 versus ≥ 5 years since diagnosis were defined as recent and long-term survivors, respectively. Chronic conditions were selected based on known primary conditions (20) and classified by the absolute number (none, 1, and 2 and above).

Standard psychometric analyses were performed to test measurement properties of the symptom phenotypes used in this study; these methods include the Kuder–Richardson Formula-20 for estimating scale reliability and confirmatory factory analysis for testing construct validity. To test our first aim, raw differences in the prevalence of individual symptom phenotypes between cancer survivors and non-cancer individuals were compared using odds ratios (OR) by conducting logistic regression. The independent effect of cancer experience (survivors vs. non-cancer individuals) on the prevalence of each symptom was compared using ORs by conducting logistic regression adjusted for chronic conditions, age, sex, race/ethnicity, marital status, and education. To test the second aim, raw differences in physical/mental HRQOL associated with the prevalence of each symptom phenotype were tested using t tests. The independent effect of all symptoms and cancer experience (survivors vs. those without cancer) on physical/mental HRQOL was tested using linear regression adjusted for the aforementioned covariates. We performed a two-step approach (i.e., two models) to test the impact of cancer experience on HRQOL. Model 1 included individual symptom phenotypes plus socio-demographic variables as predictors of HRQOL, and Model 2 included number of chronic health conditions in addition to the predictors described in Model 1. Because some individuals who experience symptoms also have co-occurring chronic health conditions that further affect HRQOL (6, 21), this two-step approach helps elucidate whether the inclusion of chronic health conditions influences the estimated association of cancer experience with HRQOL when prevalence of individual symptoms is included in the model. All analyses were performed by STATA 13.1 through the weighted procedures to address unequal probability of selection due to a complex sampling design of the NHIS.

Of the 6,770 participants, 604 were cancer survivors and 6,166 did not have a cancer history (Table 2). The sample is representative of the approximately 230 million U.S. non-institutionalized adults (cancer survivors: 8.9%; non-cancer individuals: 91.1%). Half of the cancer survivors (49.9%) were ≥65 years of age, which was higher than non-cancer individuals (13.7%; P < 0.001). More survivors were female (56.6%) and non-Hispanic White (85.6%) compared with non-cancer individuals (51.2% female, 66.8% non-Hispanic White; P < 0.001). More survivors reported having at least one chronic condition (45.3%) than non-cancer individuals (22.0%; P < 0.001). Approximately 66% of the survivors were long-term survivors.

Psychometric analysis reveals acceptable scale reliability and construct validity (Table 1). Cancer survivors overall reported a higher prevalence of individual symptom phenotypes than non-cancer individuals in sensation abnormality (61.0% vs. 39.3%; P < 0.001), pain (73.5% vs. 56.6%; P < 0.001), fatigue (68.6% vs. 61.0%; P < 0.01), cognitive disturbance (32.2% vs. 21.1%; P < 0.001), depression (43.4% vs. 41.2%; P > 0.05), and anxiety (62.9% vs. 59.4%; P > 0.05). Long-term survivors had a slightly higher prevalence in individual symptom phenotypes (except fatigue) than did recent survivors. Without adjusting for covariates, ORs of symptom prevalence were significantly higher for survivors in sensation (OR = 2.4; 95% CI = 1.9 to 3.0), pain (OR = 2.1; 95% CI = 1.7 to 2.6), fatigue (OR = 1.4; 95% CI = 1.1 to 1.8), and cognitive disturbance (OR = 1.8; 95% CI = 1.4 to 2.3). Compared to non-cancer individuals, survivors reported significantly lower raw physical HRQOL scores (−3.7; 95% CI = −4.7 to −2.6), but equivalent raw mental HRQOL scores (Table 3).

Among the six symptoms, individuals experiencing pain had the largest decrease in physical HRQOL scores when compared with those without pain (−8.9), followed by cognitive disturbance versus no (−8.4), fatigue versus no (−7.8), depressive symptom versus no (−7.1), sensation abnormality versus no (−6.1), and anxiety versus no (−5.9; P < 0.001; Table 4). In contrast, individuals experiencing depressive symptom had the largest decrease in mental HRQOL scores when compared with those without depressive symptom (−8.4), followed by cognitive disturbance versus no (−7.9), anxiety versus no (−6.6), fatigue and pain versus no (both −5.5), and sensation abnormality versus no (−4.7; P < 0.001).

Compared with non-cancer individuals, survivors had significantly higher odds of experiencing sensation abnormality (OR = 1.3; 95% CI = 1.0 to 1.7), pain (OR = 1.5; 95% CI = 1.1 to 1.9), and fatigue (OR = 1.4; 95% CI = 1.1 to 1.8) after adjusting for various covariates. Individuals having more chronic conditions had significantly higher odds of experiencing all six symptoms than those having fewer chronic conditions (P < 0.001). Specifically, having ≥2 and 1 chronic conditions increased the odds of pain by approximately 27-fold (95% CI = 16.8 to 44.3) and 7-fold (95% CI = 5.3 to 8.2), respectively, when compared with having no chronic conditions. Additionally, having an older age at the time of the survey was associated with higher odds of sensation abnormality and cognitive disturbance but lower odds of fatigue and anxiety. Females were more likely to report pain (OR = 1.4; 95% CI = 1.2 to 1.6), fatigue (OR = 1.7; 95% CI = 1.5 to 1.9), depressive symptom (OR = 1.6; 95% CI = 1.4 to 1.8), and anxiety (OR = 1.6; 95% CI = 1.4 to 1.8) than males.

Multivariable analyses revealed that prevalence of six symptoms and chronic conditions independently affected physical HRQOL (Model 1b). Cancer experience affected physical HROQL only when chronic conditions were excluded from the analysis (Model 1a). Pain had the strongest impact on physical HRQOL (β = −4.0; 95% CI = −4.5 to −3.6), followed by fatigue (β = −3.3; 95% CI = −3.8 to −2.9), depressive symptom (β = −2.5; 95% CI = −3.1 to −2.0), cognitive disturbance (β = −2.3; 95% CI = −2.9 to −1.8), sensation abnormality (β = −1.1; 95% CI = −1.5 to −0.7), and anxiety (β = −0.9; 95% CI = −1.4 to −0.4). Individuals having 1 and ≥2 chronic conditions reported decremented physical HRQOL by 4.8 points (95% CI = −5.5 to −4.1) and 9.2 points (95% CI = −10.2 to −8.2) compared with those without chronic conditions, respectively. Across the two models, decremented physical HRQOL was significantly associated with older age, non-Hispanic Black (vs. non-Hispanic White), and lower education (below high school, high school/GED, and some college vs. graduate degree).

Symptom prevalence alone rather than cancer experience significantly affected mental HRQOL regardless of the inclusion of chronic conditions (Models 2a and 2b). After adjusting for chronic conditions, prevalence of depressive symptom had the strongest effect on mental HRQOL (β = −5.2; 95% CI = −5.8 to −4.6), followed by cognitive disturbance (β = −3.3; 95% CI = −4.0 to −2.7), anxiety (β = −1.9; 95% CI = −2.4 to −1.3), pain (β = −1.8; 95% CI = −2.2 to −1.3), fatigue (β = −1.2; 95% CI = −1.7 to −0.7), and sensation abnormality (β = −0.9; 95% CI = −1.4 to −0.4). Individuals occurring with 1 and ≥2 chronic conditions had decremented mental HRQOL by 1.8 points (95% CI = −2.5 to −1.1) and 3.3 points (95% CI = −4.4 to −2.3) compared with those without chronic conditions, respectively. Across the two models, decremented mental HRQOL was significantly related to being currently unmarried (separated, divorced, single/never married, and widowed status vs. married), and having a lower education (below high school, high school/GED, some college, and bachelor degree vs. graduate degree).

This study demonstrates that cancer survivors experienced a significantly higher symptom prevalence, especially in sensation abnormality, pain, and fatigue, than non-cancer individuals. Chronic conditions were significantly related to the prevalence of all six symptoms. Survivors reported significantly worse physical HRQOL than non-cancer individuals, yet survivorship status did not significantly affect mental HRQOL. Instead of cancer experience, both prevalence of symptom phenotypes and chronic conditions independently affected physical and mental HRQOL. Older age, non-Hispanic Black, currently being unmarried, and lower education were related to decremented physical and/or mental HRQOL.

Because survivors of adult cancers tend to be older in age, a higher prevalence of symptom phenotypes may be due to issues related to aging (e.g., more chronic conditions; ref. 22). Indeed, we found that the presence of individual symptoms was linked to a greater number of chronic conditions. We further discovered that the inclusion of chronic conditions alongside socio-demographic factors in the multivariable models indicated a higher prevalence of individual symptom phenotypes (e.g., sensation abnormality, pain, and fatigue) among cancer survivors compared with that reported by non-cancer individuals (Table 5). This finding implies that chronic conditions and socio-demographics alone may not fully explain the variation of symptom burden, and the residual symptom burden in cancer survivors may be caused by symptom-specific mechanisms, particularly in relation to treatment exposures, such as autonomic nervous system activation, systemic inflammation, alteration of hypothalamic–pituitary–adrenal axis, and endothelial dysfunction (23). Longitudinal studies evaluating the onset of individual symptoms beginning with therapy completion and continuing throughout different survivorship stages are required to guide the development of a heuristic model for elucidating the complex nature of symptoms related to toxic therapy, aging, and chronic conditions.

Most impressively, we observed that cancer survivors and non-cancer individuals reported comparable physical and mental HRQOL given the same magnitude of symptom burden (insignificant coefficients −0.65 and −0.23 on cancer experience; Table 6). However, physical and mental HRQOL is substantially decreased (approximately 14 points or 1.4 SD on both domains; Table 6) among cancer and non-cancer individuals who experienced all six symptoms compared with that of asymptomatic individuals. Our study also suggests that the effect of multiple symptoms on HRQOL seems larger than that of chronic conditions, which is consistent with the previous cancer survivorship studies (24, 25). Interestingly, the magnitude of all six symptoms (Table 6) was smaller than the summated effects of individual symptoms on physical and mental HRQOL (approximately 30 and 32 points or 3.0 and 3.2 SD, respectively; results upon request), suggesting the effects of individual symptoms on HRQOL are antagonistic rather than additive. Although our study highlights the influence of symptom prevalence on impaired HRQOL after controlling for chronic conditions and socio-demographics, other factors related to survivorship (e.g., poor coping skill and social support; refs. 26, 27) that are not being measured by the NHIS may explain the residual variance.

We have argued that the mixed results (12–14, 16) of HRQOL between survivors and non-cancer individuals may mislead clinical relevance if symptom burden and chronic conditions are not taken into consideration. Although symptomatology plays a key role in patient-centered oncological research, how symptom manifestation influences our interpretation of health outcomes remains to be determined (28). Cancer survivors often experience two or more concurrent symptoms, known as symptom clusters (29), and multiple symptoms may influence each other and act as mediators to explain the association between cancer experience and HRQOL (3). Cancer patients/survivors with symptom clusters (30) have shown a higher risk of premature death than those without symptom clusters. Although this study focused on the impact of individual symptom prevalence on HRQOL, future studies are warranted to elucidate possible clusters of symptom prevalence and test the association between cluster prevalence and HRQOL. This alternate approach will help identify symptom clusters that may share common etiology for clinical interventions.

Our results have implications for designing prospective research of delayed occurrences of morbidity, mortality, and impaired HRQOL for cancer survivors. If the causal links between symptom burden and future health outcomes are established, symptoms can be used as sentinel indictors of an early diagnosis for adverse events (e.g., unexplained cardiac arrest; ref. 31) and functional status decline. Although survivorship guidelines have noted the importance of screening for symptoms (e.g., fatigue, depression, and anxiety) as an avenue for ascertaining chronic problems (32, 33), cancer survivors have received infrequent screening and inadequate care for symptoms.

This study contains several limitations. First, NHIS collects data from the civilian of non-institutionalized population. Excluding institutionalized individuals who are likely older and severely ill will limit the generalizability of our findings to the entire U.S. population. Second, cancer information (e.g., diagnosis and treatment) was self-reported by participants. Although experiencing cancer is a significant event, self-reporting may cause information recall bias, especially in elderly participants. Third, symptom phenotypes were generated using the items in NHIS, and measurement properties were determined by content and construct validity. Using standard symptom measures with sophisticated measurement properties is warranted to replicate our findings. Fourth, only six symptom phenotypes were investigated. Excluding other important symptoms (e.g., respiratory, sexual, and sleep disturbance) that were not contained in NHIS may underestimate the impact of symptom burden on HRQOL.

Compared with non-cancer individuals, cancer survivors suffer significantly from various symptoms, typically sensation abnormality, pain, and fatigue. However, asymptomatic cancer survivors appear to experience physical and mental HRQOL comparable with individuals who have not had cancer. When survivors develop symptoms and chronic conditions, their HRQOL substantially decreases. Interventions to control symptoms especially for older survivors and survivors with more chronic conditions may improve their HRQOL outcomes.

The dataset generated for statistical analyses during the current study is available through the U.S. CDC's website (http://www.cdc.gov/nchs/nhis/nhis_2010_data_release.htm). The dataset is also available from the corresponding author on reasonable request.

No potential conflicts of interest were disclosed.

Conception and design: I-C. Huang, M.M. Hudson, K.R. Krull

Development of methodology: I-C. Huang, L.L. Robison

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.):I-C. Huang, M.M. Hudson

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis):I-C. Huang, M.M. Hudson, K.R. Krull

Writing, review, and/or revision of the manuscript:I-C. Huang, M.M. Hudson, L.L. Robison, K.R. Krull

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases):I-C. Huang,

Study supervision:I-C. Huang, M.M. Hudson, K.R. Krull

The authors thank Zheng Li, PhD, who was a research coordinator of this study and prepared the dataset for statistical analyses.

This study is supported in part by the U.S. NIH grant R21 CA202210 [I-C. Huang (PI), K.R. Krull], Cancer Center Support grant P30 CA21765 (I-C. Huang, M.M. Hudson, L.L. Robison, K.R. Krull), and ALSAC (I-C. Huang, M.M. Hudson, L.L. Robison, K.R. Krull).

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.