Original Article
Variation in Symptom Distress in Underserved Chinese American Cancer Patients Lara K. Dhingra, PhD1; Kin Lam, MD2; William Cheung, MD3; Theresa Shao, MD4; Zujun Li, MD4; Sandra Van de Maele, NP4; Victor T. Chang, MD5,6; Jack Chen, MBS1; Huiyan Ye, MA7; Rhoda Wong, LCSW1; Wan Ling Lam, MD8,9; Selina Chan, RN8; Marilyn Bookbinder, RN, PhD1; Nathan F. Dieckmann, PhD10,11; and Russell Portenoy, MD1,12
BACKGROUND: Cancer is prevalent in the rapidly growing Chinese American community, yet little is known about the symptom experience to guide comprehensive treatment planning. This study evaluated symptom prevalence and patient subgroups with symptom distress in a large sample of Chinese American cancer patients. METHODS: Patients were consecutively recruited from 4 oncology practices, and they completed a translated cancer symptom scale. Latent class cluster analysis was used to identify subgroups of patients with distinct symptom distress profiles. RESULTS: There were 1436 patients screened; 94.4% were non–English-speaking, and 45.1% were undergoing cancer therapy. The cancers included breast (32.6%), lung (14.8%), head and neck (12.5%), and hematologic cancer (10.1%). Overall, 1289 patients (89.8%) had 1 or more symptoms, and 1129 (78.6%) had 2 or more. The most prevalent symptoms were a lack of energy (57.0%), dry mouth (55.6%), feeling sad (49.3%), worrying (47.5%), and difficulty sleeping (46.8%). Symptoms causing “quite a bit” or “very much” distress included difficulty sleeping (37.9%), a lack of appetite (37.2%), feeling nervous (35.8%), pain (35.2%), and worrying (34.0%). Four patient subgroups were identified according to the probability of reporting moderate to high symptom distress: very low physical and psychological symptom distress (49.5%), low physical symptom distress and moderate psychological symptom distress (25.2%), moderate physical and psychological symptom distress (17.4%), and high physical and psychological symptom distress (7.8%). CONCLUSIONS: Symptom prevalence is high in community-dwelling Chinese American cancer patients, and nearly half experience severe distress (rated as “quite a bit” or “very much” distressing) from physical symptoms, psychological symptoms, or both. These data have important implications for the development of effective symptom control intervenC 2015 American Cancer Society. tions. Cancer 2015;121:3352-9. V KEYWORDS: cancer disparities, Condensed Memorial Symptom Assessment Scale, ethnic Chinese, latent class cluster analysis (LCCA), minority health, symptom distress.
INTRODUCTION Chinese Americans are the largest Asian subgroup in the United States. The population is growing, and more than onethird are recent immigrants.1 Cancer is a leading cause of death among Asian Americans,2 and access to both diseasemodifying therapy and culturally relevant palliative care is essential. Care may be compromised, however, by sociocultural, economic, and other barriers.3 Symptom management is a recognized priority in patient management. However, despite persistent disparities in cancer incidence and mortality for Asian American populations,4,5 little is known about differences in symptom experience across racial/ethnic groups. Studies of cancer patients have noted a high prevalence of multiple symptoms,6 high symptom distress,7,8 and clustering of symptoms,9,10 but few have evaluated the Chinese American population.11 One small study of ambulatory ethnic Chinese patients in the United States (n 5 25) found an average of 14 symptoms during chemotherapy, the most prevalent of which were a lack of energy, hair loss, dry mouth, a lack of appetite, and difficulty sleeping.11 Studies of cancer patients in China, Taiwan, and Hong Kong suggest that the prevalence of multiple symptoms
Corresponding author: Lara K. Dhingra, PhD, MJHS Institute for Innovation in Palliative Care, 39 Broadway, 12th Floor, New York, NY 10006; Fax: (212) 649-5544; [email protected] 1 MJHS Institute for Innovation in Palliative Care, New York, New York; 2Community Oncology Program, Asian Services Center, Mount Sinai Beth Israel, New York, New York; 3Community Private Practice, New York, New York; 4Department of Medical Oncology, Mount Sinai Beth Israel Comprehensive Cancer Center, New York, New York; 5Section of Hematology/Oncology, Veterans Affairs New Jersey Health Care System, East Orange, New Jersey; 6Department of Medicine, Rutgers New Jersey Medical School, Newark, New Jersey; 7University of North New Jersey, Cranford, New Jersey; 8Asian Services Center, Mount Sinai Beth Israel, New York, New York; 9Department of Medicine, Icahn School of Medicine at Mount Sinai, New York, New York; 10School of Nursing, Department of Public Health and Preventative Medicine, Oregon Health and Science University, Portland, Oregon; 11Decision Research, Eugene, Oregon; 12Department of Neurology, Albert Einstein College of Medicine, Bronx, New York
We gratefully acknowledge the office staff of Dr. Kin Yui Lam, Dr. Theresa Shao, Dr. William Cheung, and Dr. Zujun Li for their assistance with participant recruitment; Dr. Tak Kwan (MSBI Asian Services Center) for promoting this community-based research collaboration; and the patients who participated in this study. DOI: 10.1002/cncr.29497, Received: June 30, 2014; Revised: April 14, 2015; Accepted: May 4, 2015, Published online June 8, 2015 in Wiley Online Library (wileyonlinelibrary.com)
3352
Cancer
September 15, 2015
Symptom Distress in Chinese Patients/Dhingra et al
may be high (25%-63%)12-14 and comparable to the prevalence in non-Chinese samples in the United States.15-17 These and other studies suggest that fatigue, sleep disturbance, pain, and dry mouth are the most prevalent symptoms6,12,14,16,18,19 and that pain and fatigue are the most distressing symptoms.7,15,20 Recent symptom-related research has focused increasingly on identifying subgroups of patients with similar symptom experiences (eg, severity or distress) as a means of exploring shared mechanisms or potential strategies for treatment.21 From the clinical perspective, patients with the same symptoms may vary in associated distress,22-24 and the ability to identify a subgroup that experiences high distress is now recognized as an important aspect of cancer care.25 This approach may guide symptom screening or treatment protocols and promote more efficient management.25 One statistical approach to identifying distinct patient subgroups in a population on the basis of their symptom experience is latent class cluster analysis (LCCA). LCCA may offer methodological advantages over traditional methodologies (such as k-means clustering and hierarchical cluster analysis) because of its ability to include variables with different scale types (eg, continuous, ordinal, and binary), efficiently handle missing data, and apply assumptions about the underlying data structure that make subgroup identification less arbitrary than traditional approaches.26 In this study, we evaluated symptom epidemiology and used LCCA to identify patient subgroups with similar symptom distress profiles in a large survey of community-dwelling Chinese American cancer patients. MATERIALS AND METHODS The current study is a screening survey that is part of an ongoing community intervention to test the effectiveness of a rapid-cycle quality improvement model of care for pain and symptom management in Chinese American cancer patients.27 Data were collected between October 2009 and August 2013 from patients at 2 large communitybased oncology practices in Manhattan’s Chinatown, an area that is designated by the US government as economically disadvantaged and medically underserved,28 and at 2 hospital-based practices in Manhattan’s West Village that treat large numbers of underserved ethnic Chinese patients. The institutional review board at Mount Sinai Beth Israel in New York approved the protocol. Patient Selection
The 2 community-based practices that provided patients for the study are private practices unaffiliated with a hosCancer
September 15, 2015
pital; the 2 hospital-based practices are part of a community cancer center affiliated with Mount Sinai Beth Israel in New York. At each practice site, a multilingual research assistant used clinic lists or physician referral to identify ethnic Chinese patients with a history of cancer. At each practice session, the research assistant consecutively screened these patients to determine eligibility for the quality improvement study. Per protocol, individual patient consent was not required for the screening, which included sociodemographic, disease status, and symptom distress data. Eligible patients spoke Cantonese, Mandarin, Fujianese, Toisanese, Hakka, or English as their primary language, had active cancer of any stage or a history of cancer, were at any point in their illness trajectory, and had no evidence of psychopathology or cognitive impairment severe enough to prevent study completion. Measurements
Sociodemographic data and information about disease characteristics and antineoplastic treatments were recorded. Data on current symptom treatments and performance status were not available. Patients reported the prevalence and current level of symptom-related distress for 11 physical symptoms and 3 psychological symptoms with a validated Chinese language instrument, the Condensed Memorial Symptom Assessment Scale–Chinese (CMSAS-C).29,30 Each symptom was described as present or absent during the past week, and those that were reported as present were rated in terms of how much the symptom distressed or bothered the patient. The Chinese translation of this terminology means “bother” or “discomfort.” Distress was rated on a 5-point Likert-type scale ranging from 0 (“not at all”) to 4 (“very much”). If a symptom was not experienced, a distress score of 0 was assigned. Scoring yielded 2 subscales (physical symptom distress and psychological symptom distress) and the total sum of symptom distress. In this study, only the physical symptom items were initially administered to patients to reduce the burden; however, the psychological symptoms were later added. Statistical Analyses
Patient sociodemographics and symptom distress were summarized. LCCA was used to explore latent classes (subgroups) of patients with similar symptom distress profiles. LCCA is a model-based approach in which the observed sample data are assumed to be generated by a mixture of underlying population probability distributions. The distribution of the latent class indicators (outcomes) for the model parameters is assumed to be a 3353
Original Article TABLE 1. Sociodemographic and Disease-Related Characteristics of Ethnic Chinese Cancer Patients Presenting to 4 Oncology Practices (n 5 1436) Age (n 5 1378), mean 6 SD (range), y Female sex, No. (%) Dialect/language spoken (n 5 765), No. (%) Cantonese Mandarin Toisanese Fujianese English Hakka Cancer site (n 5 1381), No. (%) Breast Lung Head, neck, and thyroid Hematologic malignancy Colorectal Gastrointestinal Other Liver Prostate Ovarian Kidney Pancreatic Unknown primary Bone Active treatment within past 3 mo (n 5 1255), No. (%) Cancer treatment since diagnosis (n 5 1222), No. (%) Surgery Chemotherapy Radiotherapy Extent of disease (n 5 259), No. (%) Localized Metastatic NED
62.3 6 13.5 (18-96) 875 (60.9) 583 (76.2) 173 (22.6) 66 (8.6) 61 (8.0) 43 (5.6) 8 (1.0) 450 (32.6) 204 (14.8) 172 (12.5) 140 (10.1) 115 (8.3) 102 (7.4) 44 (3.2) 43 (3.1) 41 (3.0) 27 (2.0) 16 (1.2) 13 (0.9) 13 (0.9) 1 (0.1) 566 (45.1) 705 (57.7) 773 (63.3) 384 (31.4) 147 (56.8) 84 (32.4) 28 (10.8)
Abbreviations: NED, no evidence of disease; SD, standard deviation.
mixture of class-specific densities. A (restricted) multinomial distribution is used to model the ordered (ordinal) latent class indicators. The relations between the categorical latent variables and the ordinal latent class indicators are described by a set of logistic regression equations. The probabilistic parameterization of this model results in parameter estimates for latent class probabilities (the probability that a randomly selected observation in the sample is located in class j) and conditional threshold probabilities (the probability that a member of latent class t will be at a specified level of an indicator variable). Each patient is then assigned a probability of belonging to each latent class (subgroup), which is computed from these parameter estimates and the observed data. Patients are then assigned as members of the class that has the highest membership probability. The maximum likelihood estimation underlying this approach does not require complete data and thus allows the analysis of all available information on each subject. This was a major advantage in this study 3354
because the psychological symptoms were rated only by a subset of the sample. LCCA also simultaneously manages variables that have different scale types (eg, continuous, ordinal, and binary), applies a relatively large number of formal criteria to identify the appropriate number of classes, and flexibly incorporates covariates and predictors into the model.31,32 For this study, the model specification was based on procedures by Ram and Grimm.33 Various strategies, including information criteria (Akaike information criterion, Bayesian information criterion, and sample size–adjusted Bayesian information criterion, where smaller numbers indicate a better fitting model), convergence (entropy > 0.80), the proportion of the sample in each latent class (not
Variation in Symptom Distress in Underserved Chinese American Cancer Patients Lara K. Dhingra, PhD1; Kin Lam, MD2; William Cheung, MD3; Theresa Shao, MD4; Zujun Li, MD4; Sandra Van de Maele, NP4; Victor T. Chang, MD5,6; Jack Chen, MBS1; Huiyan Ye, MA7; Rhoda Wong, LCSW1; Wan Ling Lam, MD8,9; Selina Chan, RN8; Marilyn Bookbinder, RN, PhD1; Nathan F. Dieckmann, PhD10,11; and Russell Portenoy, MD1,12
BACKGROUND: Cancer is prevalent in the rapidly growing Chinese American community, yet little is known about the symptom experience to guide comprehensive treatment planning. This study evaluated symptom prevalence and patient subgroups with symptom distress in a large sample of Chinese American cancer patients. METHODS: Patients were consecutively recruited from 4 oncology practices, and they completed a translated cancer symptom scale. Latent class cluster analysis was used to identify subgroups of patients with distinct symptom distress profiles. RESULTS: There were 1436 patients screened; 94.4% were non–English-speaking, and 45.1% were undergoing cancer therapy. The cancers included breast (32.6%), lung (14.8%), head and neck (12.5%), and hematologic cancer (10.1%). Overall, 1289 patients (89.8%) had 1 or more symptoms, and 1129 (78.6%) had 2 or more. The most prevalent symptoms were a lack of energy (57.0%), dry mouth (55.6%), feeling sad (49.3%), worrying (47.5%), and difficulty sleeping (46.8%). Symptoms causing “quite a bit” or “very much” distress included difficulty sleeping (37.9%), a lack of appetite (37.2%), feeling nervous (35.8%), pain (35.2%), and worrying (34.0%). Four patient subgroups were identified according to the probability of reporting moderate to high symptom distress: very low physical and psychological symptom distress (49.5%), low physical symptom distress and moderate psychological symptom distress (25.2%), moderate physical and psychological symptom distress (17.4%), and high physical and psychological symptom distress (7.8%). CONCLUSIONS: Symptom prevalence is high in community-dwelling Chinese American cancer patients, and nearly half experience severe distress (rated as “quite a bit” or “very much” distressing) from physical symptoms, psychological symptoms, or both. These data have important implications for the development of effective symptom control intervenC 2015 American Cancer Society. tions. Cancer 2015;121:3352-9. V KEYWORDS: cancer disparities, Condensed Memorial Symptom Assessment Scale, ethnic Chinese, latent class cluster analysis (LCCA), minority health, symptom distress.
INTRODUCTION Chinese Americans are the largest Asian subgroup in the United States. The population is growing, and more than onethird are recent immigrants.1 Cancer is a leading cause of death among Asian Americans,2 and access to both diseasemodifying therapy and culturally relevant palliative care is essential. Care may be compromised, however, by sociocultural, economic, and other barriers.3 Symptom management is a recognized priority in patient management. However, despite persistent disparities in cancer incidence and mortality for Asian American populations,4,5 little is known about differences in symptom experience across racial/ethnic groups. Studies of cancer patients have noted a high prevalence of multiple symptoms,6 high symptom distress,7,8 and clustering of symptoms,9,10 but few have evaluated the Chinese American population.11 One small study of ambulatory ethnic Chinese patients in the United States (n 5 25) found an average of 14 symptoms during chemotherapy, the most prevalent of which were a lack of energy, hair loss, dry mouth, a lack of appetite, and difficulty sleeping.11 Studies of cancer patients in China, Taiwan, and Hong Kong suggest that the prevalence of multiple symptoms
Corresponding author: Lara K. Dhingra, PhD, MJHS Institute for Innovation in Palliative Care, 39 Broadway, 12th Floor, New York, NY 10006; Fax: (212) 649-5544; [email protected] 1 MJHS Institute for Innovation in Palliative Care, New York, New York; 2Community Oncology Program, Asian Services Center, Mount Sinai Beth Israel, New York, New York; 3Community Private Practice, New York, New York; 4Department of Medical Oncology, Mount Sinai Beth Israel Comprehensive Cancer Center, New York, New York; 5Section of Hematology/Oncology, Veterans Affairs New Jersey Health Care System, East Orange, New Jersey; 6Department of Medicine, Rutgers New Jersey Medical School, Newark, New Jersey; 7University of North New Jersey, Cranford, New Jersey; 8Asian Services Center, Mount Sinai Beth Israel, New York, New York; 9Department of Medicine, Icahn School of Medicine at Mount Sinai, New York, New York; 10School of Nursing, Department of Public Health and Preventative Medicine, Oregon Health and Science University, Portland, Oregon; 11Decision Research, Eugene, Oregon; 12Department of Neurology, Albert Einstein College of Medicine, Bronx, New York
We gratefully acknowledge the office staff of Dr. Kin Yui Lam, Dr. Theresa Shao, Dr. William Cheung, and Dr. Zujun Li for their assistance with participant recruitment; Dr. Tak Kwan (MSBI Asian Services Center) for promoting this community-based research collaboration; and the patients who participated in this study. DOI: 10.1002/cncr.29497, Received: June 30, 2014; Revised: April 14, 2015; Accepted: May 4, 2015, Published online June 8, 2015 in Wiley Online Library (wileyonlinelibrary.com)
3352
Cancer
September 15, 2015
Symptom Distress in Chinese Patients/Dhingra et al
may be high (25%-63%)12-14 and comparable to the prevalence in non-Chinese samples in the United States.15-17 These and other studies suggest that fatigue, sleep disturbance, pain, and dry mouth are the most prevalent symptoms6,12,14,16,18,19 and that pain and fatigue are the most distressing symptoms.7,15,20 Recent symptom-related research has focused increasingly on identifying subgroups of patients with similar symptom experiences (eg, severity or distress) as a means of exploring shared mechanisms or potential strategies for treatment.21 From the clinical perspective, patients with the same symptoms may vary in associated distress,22-24 and the ability to identify a subgroup that experiences high distress is now recognized as an important aspect of cancer care.25 This approach may guide symptom screening or treatment protocols and promote more efficient management.25 One statistical approach to identifying distinct patient subgroups in a population on the basis of their symptom experience is latent class cluster analysis (LCCA). LCCA may offer methodological advantages over traditional methodologies (such as k-means clustering and hierarchical cluster analysis) because of its ability to include variables with different scale types (eg, continuous, ordinal, and binary), efficiently handle missing data, and apply assumptions about the underlying data structure that make subgroup identification less arbitrary than traditional approaches.26 In this study, we evaluated symptom epidemiology and used LCCA to identify patient subgroups with similar symptom distress profiles in a large survey of community-dwelling Chinese American cancer patients. MATERIALS AND METHODS The current study is a screening survey that is part of an ongoing community intervention to test the effectiveness of a rapid-cycle quality improvement model of care for pain and symptom management in Chinese American cancer patients.27 Data were collected between October 2009 and August 2013 from patients at 2 large communitybased oncology practices in Manhattan’s Chinatown, an area that is designated by the US government as economically disadvantaged and medically underserved,28 and at 2 hospital-based practices in Manhattan’s West Village that treat large numbers of underserved ethnic Chinese patients. The institutional review board at Mount Sinai Beth Israel in New York approved the protocol. Patient Selection
The 2 community-based practices that provided patients for the study are private practices unaffiliated with a hosCancer
September 15, 2015
pital; the 2 hospital-based practices are part of a community cancer center affiliated with Mount Sinai Beth Israel in New York. At each practice site, a multilingual research assistant used clinic lists or physician referral to identify ethnic Chinese patients with a history of cancer. At each practice session, the research assistant consecutively screened these patients to determine eligibility for the quality improvement study. Per protocol, individual patient consent was not required for the screening, which included sociodemographic, disease status, and symptom distress data. Eligible patients spoke Cantonese, Mandarin, Fujianese, Toisanese, Hakka, or English as their primary language, had active cancer of any stage or a history of cancer, were at any point in their illness trajectory, and had no evidence of psychopathology or cognitive impairment severe enough to prevent study completion. Measurements
Sociodemographic data and information about disease characteristics and antineoplastic treatments were recorded. Data on current symptom treatments and performance status were not available. Patients reported the prevalence and current level of symptom-related distress for 11 physical symptoms and 3 psychological symptoms with a validated Chinese language instrument, the Condensed Memorial Symptom Assessment Scale–Chinese (CMSAS-C).29,30 Each symptom was described as present or absent during the past week, and those that were reported as present were rated in terms of how much the symptom distressed or bothered the patient. The Chinese translation of this terminology means “bother” or “discomfort.” Distress was rated on a 5-point Likert-type scale ranging from 0 (“not at all”) to 4 (“very much”). If a symptom was not experienced, a distress score of 0 was assigned. Scoring yielded 2 subscales (physical symptom distress and psychological symptom distress) and the total sum of symptom distress. In this study, only the physical symptom items were initially administered to patients to reduce the burden; however, the psychological symptoms were later added. Statistical Analyses
Patient sociodemographics and symptom distress were summarized. LCCA was used to explore latent classes (subgroups) of patients with similar symptom distress profiles. LCCA is a model-based approach in which the observed sample data are assumed to be generated by a mixture of underlying population probability distributions. The distribution of the latent class indicators (outcomes) for the model parameters is assumed to be a 3353
Original Article TABLE 1. Sociodemographic and Disease-Related Characteristics of Ethnic Chinese Cancer Patients Presenting to 4 Oncology Practices (n 5 1436) Age (n 5 1378), mean 6 SD (range), y Female sex, No. (%) Dialect/language spoken (n 5 765), No. (%) Cantonese Mandarin Toisanese Fujianese English Hakka Cancer site (n 5 1381), No. (%) Breast Lung Head, neck, and thyroid Hematologic malignancy Colorectal Gastrointestinal Other Liver Prostate Ovarian Kidney Pancreatic Unknown primary Bone Active treatment within past 3 mo (n 5 1255), No. (%) Cancer treatment since diagnosis (n 5 1222), No. (%) Surgery Chemotherapy Radiotherapy Extent of disease (n 5 259), No. (%) Localized Metastatic NED
62.3 6 13.5 (18-96) 875 (60.9) 583 (76.2) 173 (22.6) 66 (8.6) 61 (8.0) 43 (5.6) 8 (1.0) 450 (32.6) 204 (14.8) 172 (12.5) 140 (10.1) 115 (8.3) 102 (7.4) 44 (3.2) 43 (3.1) 41 (3.0) 27 (2.0) 16 (1.2) 13 (0.9) 13 (0.9) 1 (0.1) 566 (45.1) 705 (57.7) 773 (63.3) 384 (31.4) 147 (56.8) 84 (32.4) 28 (10.8)
Abbreviations: NED, no evidence of disease; SD, standard deviation.
mixture of class-specific densities. A (restricted) multinomial distribution is used to model the ordered (ordinal) latent class indicators. The relations between the categorical latent variables and the ordinal latent class indicators are described by a set of logistic regression equations. The probabilistic parameterization of this model results in parameter estimates for latent class probabilities (the probability that a randomly selected observation in the sample is located in class j) and conditional threshold probabilities (the probability that a member of latent class t will be at a specified level of an indicator variable). Each patient is then assigned a probability of belonging to each latent class (subgroup), which is computed from these parameter estimates and the observed data. Patients are then assigned as members of the class that has the highest membership probability. The maximum likelihood estimation underlying this approach does not require complete data and thus allows the analysis of all available information on each subject. This was a major advantage in this study 3354
because the psychological symptoms were rated only by a subset of the sample. LCCA also simultaneously manages variables that have different scale types (eg, continuous, ordinal, and binary), applies a relatively large number of formal criteria to identify the appropriate number of classes, and flexibly incorporates covariates and predictors into the model.31,32 For this study, the model specification was based on procedures by Ram and Grimm.33 Various strategies, including information criteria (Akaike information criterion, Bayesian information criterion, and sample size–adjusted Bayesian information criterion, where smaller numbers indicate a better fitting model), convergence (entropy > 0.80), the proportion of the sample in each latent class (not
