- Email: [email protected]
Factors related to daily life interference in lung cancer patients: A cross-sectional regression tree study
European Journal of Oncology Nursing 16 (2012) 345e352
Contents lists available at ScienceDirect
European Journal of Oncology Nursing journal homepage: www.elsevier.com/locate/ejon
Factors related to daily life interference in lung cancer patients: A cross-sectional regression tree study Hsueh-Hsing Pan a, Kuan-Chia Lin b, Shung-Tai Ho c, Chun-Yu Liang a, Shih-Chun Lee d, Kwua-Yun Wang e, f, * a
Graduate Institute of Medical Science, National Defense Medical Center, No. 161, Sec. 6, Minquan E. Rd., Neihu Dist., Taipei City 114, Taiwan Department of Nursing, National Taipei College of Nursing, No. 365, Ming-te Road, Peitou District, Taipei City, Taiwan Department of Anesthesiology, Taipei Veterans General Hospital, No. 201, Sec. 2, Shipai Rd., Taipei 112, Taiwan d Department of Thoracic Surgery, Tri-Service General Hospital, No. 325, Sec. 2, Chenggong Rd., Neihu District, Taipei City 114, Taiwan e Department of Nursing, Taipei Veterans General Hospital, No. 201, Sec. 2, Shipai Rd., Taipei 112, Taiwan f School of Nursing, National Defense Medical Center, No. 161, Sec. 6, Minquan E. Rd., Neihu Dist., Taipei City 114, Taiwan b c
a b s t r a c t Keywords: Daily life interference Lung cancer Physical activity interference Psychological activity interference Regression tree model
Purpose: To identify the symptom combination patterns and symptom severity levels that induce severe symptom interference in daily life activities, including physical and psychological activity interference in lung cancer patients. Methods: In a cross-sectional descriptive study using convenience sampling, 131 participants were recruited at a medical center in northern Taiwan. The Eastern Cooperative Oncology Group (ECOG) performance status was used to assess performance status, and the Taiwanese version of the M.D. Anderson Symptom Inventory (MDASI-T) was used to assess symptom severity and symptom interference in daily life activities including physical and psychological activities. Regression tree models were applied to examine variable combinations for symptom interference level in daily life activities, including physical and psychological activity interference. Results: Study results revealed that the performance status is the key discriminator of the symptom interference level in daily life and physical activities, but distress severity is the key discriminating factor of the symptom interference level in psychological activities. The performance status and distress severity, plus other factors, further specifically show the discrimination paths and interactions between the risk groups. Conclusions: This study provided an alternative approach to identify low- and high-risk groups of symptom interference among lung cancer patients in Taiwan. Increased awareness and further understanding of the risk combinations and discriminate levels of symptom severity that induced high symptom interference offer different perspectives to develop patient-centered care planning for lung cancer patient rehabilitation. Ó 2011 Elsevier Ltd. All rights reserved.
Introduction Among cancer patients, symptoms usually occur not in isolation, but in pairs, groups, or clusters (Dodd et al., 2004), and also result from a variety of physiological, psychological, behavioral, and sociocultural factors interacting with each other (Parker et al., 2005). Recent research has increasingly focused on the exploration of symptom clusters, and demonstrated that different symptoms may * Corresponding author. Department of Nursing, Taipei Veterans General Hospital, No. 201, Sec. 2, Shipai Rd., Taipei 112, Taiwan. Tel.: þ886 2 28757233, 886 2 87923100x18766; fax: þ886 2 28752932. E-mail address: [email protected] (K.-Y. Wang). 1462-3889/$ e see front matter Ó 2011 Elsevier Ltd. All rights reserved. doi:10.1016/j.ejon.2011.07.011
occur in combination or influence each other (Cheung et al., 2009; Kim et al., 2009; Wang et al., 2008). Studies also indicate that symptom clusters significantly correlate with daily life activities (Wang et al., 2008). Findings show that symptom severity correlates with many aspects of illness, including treatment-related factors, psychosocial factors, physical conditions, comorbidities, and personal profiles (Gift et al., 2004). Symptom severity can seriously impact on patients’ daily lives. However, symptom interference, including physical and psychological activity interference, refers to interference with functional status in terms of general activities, work, walking, mood, enjoyment of life and relationships with others (Cleeland et al., 2000). Previous studies showed that higher levels
346
H.-H. Pan et al. / European Journal of Oncology Nursing 16 (2012) 345e352
of symptom severity could produce greater symptom interference among cancer patients (Tanaka et al., 2002b; Wells, 2000; Wells et al., 2003). Most prior research on the relationship between symptom severity and symptom interference among lung cancer patients has focused on the study of individual symptoms in isolation or clusters rather than on the potential interactions and inter-relationships among various symptoms that can be used to classify subjects into high- and low-risk groups. Existing conceptual models or theories, such as the Symptom Management Model (Larson et al., 1994), the Theory of Unpleasant Symptoms (Lenz et al., 1995), the Symptoms Experience Model (Armstrong, 2003) and the Symptoms Experience in Time Model (Henly et al., 2003) have been applied in a variety of symptom management studies. It has been proposed that these models and theories may reveal important knowledge gaps concerning symptom interactions and the impact of interventions on patient outcomes (Brant et al., 2010). Lung cancer has been the leading cause of cancer-related deaths in Taiwan since 1999, and the incidence has continued on an upward trend over the past decade (Department of Health, 2008). The overall 5-year rate of survival is in the 21%e24% range (Chiang et al., 2008). These statistics reveal that lung cancer is still a significant clinical problem. Hence, the present study sought to identify symptom combination patterns (symptoms coupled with certain demographics, disease- and treatment-related characteristics) and symptom severity levels that induce severe symptom interference in daily life activities, including physical and psychological activity interference in lung cancer patients using regression tree modeling. Methods Study participants This was a cross-sectional study. Based on the procedure related to the conditional or fixed factors model of multiple regression (Erdfelder et al., 1996), we assumed that the dependent variable (daily life interference) was predicted by as a set B of 10e15 predictors. We further assumed that the population R2YjB was 0.15 (moderate effect size) (Cohen, 1988), that the 15e20 predictors would account for 15% of the variance of Y. Then with an alpha of 0.05 and 80% power (Erdfelder et al., 1996), the required sample size was estimated to be 125 subjects. The inclusion criteria were 1) pathologically diagnosed lung cancer; 2) over 18 years of age; 3) free of cognitive impairments/mental illness; 4) able to communicate in Mandarin or Taiwanese and 5) willing to participate. In this study, a total of 132 lung cancer patients were recruited from outpatient hematology/oncology, radiation oncology and thoracic surgery departments in an approx 1700-bed medical center in northern Taiwan between January and April 2004. Twenty to twenty-five lung cancer patients were treated each month in this medical center. Among 132 lung cancer patients, only one patient was too weak to complete the interview. Finally, 131 participants (65 men, 49.6%; 66 women, 50.4%) completed the study. Instruments Demographics, disease- and treatment-related characteristics Demographic characteristics included participant age, gender, years of education (more or less than nine years), marital status (married or other), employment status (i.e., employed, unemployed or retired, unemployed due to disease), religious affiliation (yes or no) and smoking status (i.e., never smoked, has stopped smoking, still smoking). Disease-related characteristics consist of lung cancer cell type (small and non-small cell), stage of disease (IA-IIIA or IIIB-
IV), metastasis, disease duration, complications (yes or no), comorbidity (yes or no) and self-perceived disease severity (i.e., no, moderate, and severe). Treatment-related characteristics collected were cancer-related treatments (i.e., chemotherapy, chemotherapy combined with radiotherapy, and/or operation, others) received since cancer diagnosis and during the most recent one-week period, narcotics and/or analgesics used or not, and self-perceived treatment effectiveness (yes or no). Performance status Performance status was measured by the Eastern Cooperative Oncology Group (ECOG) performance status, referring to the functional status of cancer patients rated by patient self-report. The ECOG was rated on a scale of 0e5, where 0 was being fully active; 1 being restricted in physically strenuous activity but ambulatory; 2 being ambulatory and capable of all self-care; 3 capable of only limited self-care; 4 completely disabled; and 5 being dead (Oken et al., 1982). In view of the single item design of ECOG, equivalence reliability cannot be acquired. In terms of final score, performance status was classified as good (scores of 0e1) performance or poor (scores of 2e4) (Cleeland et al., 2000). In another study, the ECOG was documented as having good predictive validity (Buccheri et al., 1996). In our study, this questionnaire was translated into Chinese, and back-translation was used to verify semantics of the Chinese version. Symptom severity and symptom interference Symptom severity and symptom interference were measured by the Taiwanese version of the M.D. Anderson Symptom Inventory (MDASI-T). The original MDASI included two subscales: the first subscale covered the 13 core symptom items of pain, fatigue, sleep disturbance, distress, shortness of breath, memory difficulties, drowsiness, dry mouth, sadness, numbness, poor appetite, nausea and vomiting. The second subscale mainly addressed daily life activities, which integrated six symptom interference items into two categories of physical activities (general activity, walking, and normal work) and psychological activities (mood, relationships with others, and enjoyment of life) (Cleeland et al., 2000). MDASI is a patient-reported outcome tool validated for use among cancer patients and was developed to measure symptom severity and symptom interference with activities of daily living during the most recent 24-h period in cancer patients. Each symptom severity item was rated on an 11-point numeric scale, from 0 (not severe at all) to 10 (as severe as you can imagine). Each symptom interference item was rated on a similar scale, from 0 (no interference) to 10 (complete interference). The MDASI-T was developed using a standard translation and back-translation procedure, with content validity evaluated by five experts (one oncologist, two clinical oncology specialists and two oncology professors) and five cancer patients. Cronbach’s a of internal consistency was 0.87 for symptom severity items and 0.86 for interference items. Test-retest reliability over a one-week interval was 0.77 for symptom severity items and 0.86 for interference items. In a previous Chinese study, MDASI-T was validated in a sample of 556 Taiwanese patients with multiple diagnoses of cancer. The internal consistency Cronbach’s a was 0.89 for symptom severity items and 0.94 for interference items. The test-retest reliability over a 3-day interval was 0.97 for symptom severity items and 0.86 for the interference items. Construct validity was established by factor analysis and concurrent validity was examined by correlating the MADSI-T scores and scores of the Medical Outcome Study 36-Item Short-Form Health Survey (Lin et al., 2007).
H.-H. Pan et al. / European Journal of Oncology Nursing 16 (2012) 345e352
Data collection procedures The study was approved by the institution and permission was granted by the participating patients. Physicians referred the patients who met the inclusion criteria to the researcher. Then, the researcher provided a verbal explanation of the study. All patients were informed that they could withdraw from the study at any time without penalty, and all information would be kept confidential. Patients completed the questionnaire during a face-to-face interview with the same researcher to ensure the quality of data collection and to avoid missing data. Disease- and treatmentrelated information were obtained from the participants’ medical charts following completion of the questionnaire.
347
(5) As extremely large and complex tree structures would be unexplainable in practical application, we further applied the cross-validation to randomly divide the sample data into a model-building set (90% of the data sample) and a validation set (10% of the data sample). The model-building set was then used to establish a tree that was pruned using the validation set to achieve an estimation of the most appropriate tree through the minimum cost-complexity test. (6) The final model was established using the forward (likelihood ratio based), estimated (using Wald-c2) and 2 log-likelihood change. The P-value was set at 0.05 as the criterion for selecting independent variables. Results
Statistical analysis Participant characteristics Demographics, disease- and treatment-related characteristics were expressed as a mean standard deviation (SD) for continuous variables and as proportions for categorical variables. The type of combinations (symptoms coupled with certain demographics, disease- and treatment-related characteristics) and discriminate levels were used in a regression tree analysis. The regression trees in this study were run by S-Plus 6.2 software (Insightful Inc., Seattle, WA, USA), and the growing, stopping, and pruning of the trees were determined by the Gini improvement measure (Everitt, 2004). Overall, the symptom interference level in daily life activities (SIL-DLA) was segregated into two discrete categories: symptom interference level in physical activities (SIL-Phys) and symptom interference level in psychological activities (SIL-Psy). These activities were adopted as dependent variables in the regression tree models. Some demographic data, disease- and treatment-related variables, and symptom severity were treated as independent variables. As the name implies, a regression tree can be viewed as a classifier in the form of a tree structure, in which each node is either a child node (leaf node) or terminal node (decision node). All terminal nodes have splits, testing the symptom interference level. Each branch from the terminal node corresponds to a distinct outcome of the test. Each child node also has a class label attached to it. Although a variety of regression tree algorithms have been developed with different capabilities and requirements, most are variations of a core learning algorithm that employs a Top-down Greedy Split (TGS) search through the space of a possible regression tree. The strategies for tree construction in this study are as follows (Armitage et al., 2002; Breiman et al., 1984; Li et al., 2010; Teng et al., 2007): (1) Start with an empty tree (parent node) and the entire training set. (2) Select the splitting attribute (risk factor) that was the most appropriate in separating the samples into distinct classes based on statistical goodness measure of the split. This attribute then becomes a child node. In choosing the best splitter, a number of different measures of purity can be selected, simply called “splitting criteria” or “splitting functions.” The most common splitting function is the “Gini.” Therefore, we used “Gini” to make the splits in the data, as opposed to the information gain measure that other tree classifiers use. (3) A branch was created for each distinct value of risk factor and the samples were partitioned accordingly. (4) The process was conducted recursively until the criterion of statistical stop rule was reached, such that all samples for n belonged to the same class (terminal node), no risk factor could be further partitioned, or the improvement was not substantial enough to justify further splitting.
Demographics, disease- and treatment-related characteristics are shown in Table 1. The participant average age was 63.9 years (range: 27e89 years). The mean duration of disease since diagnosis was 12.4 months (range: 1e109 months). Almost three quarters (73.3%) of the participants had at least one co-morbidity and 91.6% were diagnosed with non-small cell lung cancer; 57.3% of the participants had no metastasis and 80.2% had good performance status (ECOG ¼ 0e1), while 77.9% had received chemotherapy or combined cancer-related therapy treatments since diagnosis. Symptom severity and symptom interference Symptom severity and symptom interference are shown in Table 2. The highest ranking five mean levels of symptom severity were fatigue, dry mouth, shortness of breath, sleep disturbance and pain. However, the highest ranking three activities in regard to the symptom interference level were work, walking and enjoyment of life. Mean SD scores for SIL-DLA were 2.76 2.58, while in SILPhys and SIL-Psy they were 3.28 2.95 and 2.23 2.59 respectively. From a statistical point of view, the large SD in several symptom severity and symptom interference level indicators reflect that the data points are not clustered closely around the mean. Regression tree models Fig. 1 displays the regression tree model for overall SIL-DLA, in which the most appropriate tree size was divided into five terminal nodes. The results show that the most important discriminator (the first layer in this tree model) was performance status. The performance status then combined with the secondary layer factors, including distress or sadness, followed by different discriminate levels in regard to symptom severity. In addition, shortness of breath further added into the model as the third layer to induce the different SIL-DLA. Terminal node 5 indicates the highest SIL-DLA such that when performance status was poor and sadness severity score >4.5, the predicted functional outcome for SIL-DLA was 7.87 and the population size was 9. By contrast, terminal node 1 indicates the lowest SIL-DLA such that when performance status was good, distress severity score <4.5, and shortness of breath severity score <7.5, the predicted functional outcome for SIL-DLA was 1.28 and the population size was 78. The other combinatorial paths, inter-relationships, and populations are also illustrated in Fig. 1. SIL-DLA was segregated into the two discrete categories of SILPhys and SIL-Psy. As shown in Fig. 2, the most appropriate tree size for predicting SIL-Phys was divided into six terminal nodes. This study also demonstrated that performance status was the most
348
H.-H. Pan et al. / European Journal of Oncology Nursing 16 (2012) 345e352
important discriminator in the first layer and combined with varying discriminate levels of symptom severity or other variables such as shortness of breath and employment status (the secondary layer), pain (the third layer), and dry mouth (the fourth layer) to produce different SIL-Phys. On the right-hand side, terminal node 6 indicates the highest SIL-Phys such that when the performance status was poor and unemployment was due to disease, the predicted functional outcome for SIL-Phys was 9.07 and the population size was 9. By contrast, terminal node 1 showed that when the performance status was good, shortness of breath severity score
Table 1 Patient demographics and disease-related characteristics (n ¼ 131). Variable
n
Mean age (SD) 63.9 Age range 27e89 Mean disease duration (SD) 12.4 (months) Gender Male 65 Female 66 Years of education < 9 yrs 78 > 9 yrs 53 Marital status Married 93 Other 38 Employment Employed 10 Unemployed or retired 81 Unemployed due to disease 40 Religious affiliation Yes 87 No 44 Smoking status Never smoked 60 Stopped smoking 62 Keeps smoking 9 Cell type SCLC 11 NSCLC 120 Stage of disease IA-IIIA 24 IIIB-IV 107 Metastasis Yes 56 No 75 Complication(s) Yes 63 No 68 Comorbidity(ies) Yes 96 No 35 Performance status Good (ECOG ¼ 0-1) 105 Poor (ECOG ¼ 2-4) 26 Self-perceived disease severity None 50 Moderate 34 Severe 47 Treatment following disease diagnosis Chemotherapy 41 Chemotherapy combined with others 61 Others 29 Treatment administered during the previous week Yes 48 No 83 Narcotics and/or analgesics used or not Yes 41 No 90 Self-perceived treatment effect (n ¼ 122) Yes 70 No 52
(%) (12.8) (14.8) (49.6) (50.4) (59.5) (40.5) (71.0) (29.0)
Table 2 Symptom severity and symptom interference in lung cancer patients (n ¼ 131). Item
Mean
SD
Symptom severity Fatigue Dry mouth Shortness of breath Sleep disturbance Pain Lack of appetite Sadness Distress Drowsiness Difficulty remembering Numbness Nausea Vomiting Symptom interference Physical interference Work Walking General activity Psychological interference Enjoyment of life Mood Relationships with others
2.35 3.48 2.91 2.80 2.63 2.56 2.54 2.51 2.47 2.40 2.02 1.93 1.50 0.81 2.76 3.28 4.22 3.42 2.20 2.23 2.82 2.50 1.38
1.59 3.37 3.38 2.91 3.24 3.05 3.46 3.50 3.36 3.11 3.05 2.80 2.79 2.10 2.58 2.95 3.73 3.33 3.17 2.59 3.69 3.35 2.65
SD ¼ standard deviation.
(42.7) (57.3)
<7.5, and pain severity score <1.5, the predicted functional outcome for SIL-Phys was 1.06 and the population size was 43. The other combinatorial paths, inter-relationships, and populations are also illustrated in Fig. 2. Fig. 3 shows the most appropriate tree size for predicting SIL-Psy was divided into five terminal nodes. This study indicated that distress was the most important discriminator in the first layer, sadness and drowsiness were in the secondary layer, and nausea was in the third layer. Terminal node 5 indicates the highest SIL-Psy such that when the distress severity score was >4.5, and drowsiness severity score >4.5, the predicted functional outcome for SILPsy was 7.03 and the population size was 10. By contrast, terminal node 1 indicates the lowest SIL-Psy such that when the distress severity score was <4.5, and sadness severity score <0.5, the predicted functional outcome for SIL-Psy was 0.9 and the population size was 71. The other combinatorial paths, inter-relationships, and populations are also illustrated in Fig. 3.
(48.1) (51.9)
Discussion
(7.6) (61.8) (30.5) (66.4) (33.6) (45.8) (47.3) (6.9) (8.4) (91.6) (18.3) (81.7)
(73.3) (26.7) (80.2) (19.8) (38.2) (26.0) (35.9) (31.3) (46.6) (22.1) (36.7) (63.4) (31.3) (68.7) (57.4) (42.6)
SD ¼ standard deviation; SCLC ¼ small cell lung cancer; NSCLC ¼ non-small cell lung cancer; ECOG ¼ Eastern Cooperative Oncology Group.
This study provided essential information to identify combination patterns of symptom interaction as well as each cutoff point related to symptom interference level among Taiwanese lung cancer patients. Although a similar study design, sampling procedure, and questionnaire was done in Taiwan before (Wang et al., 2008), the authors used traditional statistical modeling methods such as bivariate analysis, hierarchical cluster analysis, and linear regression to detect the possible related factors of certain health outcomes and to estimate the statistical prediction. Consequently, it was difficult to identify and explore complex data patterns and to model higher order interactions. The regression tree analysis used in this study is different from traditional variable-oriented regression analysis. It is a subject-oriented multivariate statistic that aims to recognize people with the same characteristics by the overlap of many factors. Interestingly, in this study, regression tree analysis pointed out the high- and low-risk patients and determined optimal cutoff points of symptom levels. This makes it possible to classify individuals into distinct groups or categories based on individual response patterns so that individuals within a group are more similar than individuals between groups.
H.-H. Pan et al. / European Journal of Oncology Nursing 16 (2012) 345e352
The findings demonstrated that the symptom interference level may be explained and predicted through the serial interactions of a number of independent variables, which allowed us to find the crucial role in identifying homogeneity subgroups of Taiwanese lung cancer patients. Based on this, we found that performance status was the key discriminator of the SIL-DLA and SIL-Phys, and the distress severity was the key discriminator of the SIL-Psy. The performance status or distress severity plus other factors further showed the discrimination paths and interactions between the risk groups. In other words, there were different tree structures between the SIL-DLA, SIL-Phys, and SIL-Psy. For the SIL-DLA, several previous studies have shown that patients with poor performance status, experienced greater interference with daily life activities (Cleeland et al., 2000; Lin et al., 2007; Tseng et al., 2008), and experienced significantly more anxiety and depressive symptoms and worse emotional and mental functioning (Maric et al., 2010). In addition, many prior studies have indicated that the severity of shortness of breath significantly interfered with activities including physical and psychological activities (Reddy et al., 2009; Tanaka et al., 2002b). Our findings from different tree structures imply that the performance status interacting with distress, sadness and shortness of breath appear to be sensitive to discriminate the high-risk group and relatively lowrisk group SIL-DLA among Taiwanese lung cancer patients. Symptom severity may reduce physical function, but may also manifest itself in decreased mental attentiveness, alertness and motivation. Untreated psychological distress is associated with reduced quality of life and inadequate palliation of physical
349
symptoms (Maric et al., 2010). Patients with poor performance status, and especially greater sadness or distress, may therefore require management of their psychological problems using psychotherapeutic and pharmacological interventions or combined therapies from clinicians in order to reduce SIL-DLA (Keller et al., 2000; Rodriguez Vega et al., 2010). This study revealed that the performance status interacts with shortness of breath, employment status, pain and dry mouth to influence on SIL-Phys. Study has indicated that patients with poor performance status also had reduced physical activity (Sarna, 1994). Patients with shortness of breath were reflected in a higher SILPhys (Reddy et al., 2009; Rogers et al., 2008; Tanaka et al., 2002b), which also tended to coexist with other symptoms such as fatigue, pain, dry mouth, drowsiness, nausea, distress and sadness (Cheung et al., 2009; Lin et al., 2007); pain often interfered with physical and psychological activities (Chow et al., 2007; Tanaka et al., 2002a); and dry mouth was significantly associated with physical activity (Rogers et al., 2008). Studies also indicated that the majority of cancer patients had a work-related disability following cancer diagnosis or cancer-related treatments (Oberst et al., 2010). Loss of employment because of cancer had a negative impact on their quality of life (Kobayashi et al., 2008). Therefore, lung cancer patients with poor performance and with symptoms such as shortness of breath, pain and dry mouth may also require more attention from clinicians in order to prevent a negative impact on physical activities. For the SIL-Psy, our findings suggested that distress severity interacted with sadness, drowsiness or nausea to discern the target
Parent node N=131
Performance status
Good performance status Child node 1 N=105
Poor performance status Child node 2 N=26
Distress
Sadness
Distress < 4.5 Child node 3 N=83 Shortness of breath
Distress > 4.5 N=22 SIL-DLA= 4.17 Terminal node 3
Shortness of breath < 7.5 N=78 SIL-DLA= 1.28
Shortness of breath > 7.5 N=5 SIL-DLA= 4.87
Terminal node 1
Terminal node 2
Sadness < 4.5 N=17 SIL-DLA= 4.37 Terminal node 4
Sadness > 4.5 N=9 SIL-DLA= 7.87 Terminal node 5
Fig. 1. Regression tree for symptom interference level in daily life activities of 131 lung cancer patients in Taiwan. SIL-DLA, symptom interference level in daily life activities; Parent node, start with an empty tree; Child node, the splitting attribute that was the most appropriate in separating the samples into distinct classes based on statistical goodness measure of split; Terminal node, the process was conducted recursively until the criterion of statistical stop rule was reached.
350
H.-H. Pan et al. / European Journal of Oncology Nursing 16 (2012) 345e352
the high- or low-risk groups. A distinctive finding in this study was that some symptoms such as dry mouth (Fig. 2) and nausea (Fig. 3) were not the key discriminators (in the third or fourth layer) in the tree models, but affected other symptoms to generate different levels of symptom interference. This study has a number of limitations. First, the MDASI was a patient-reported assessment tool. Because of conservative folk customs, Taiwanese patients may be embarrassed or reluctant to report their symptom severity and symptom interference due to concern about bothering the nurses or physicians or fear of being stigmatized as a difficult patient. As a result, symptom severity and symptom interference might be under-reported by patients. Second, as our study population was a convenience sample and limited to one medical center, the generalizability of our findings is limited. Third, as this study only addressed lung cancer outpatients, our findings cannot be generalized to seriously ill lung cancer inpatients. Forth, this study used a cross-sectional design, meaning that change in symptom severity and symptom interference over time was not investigated. Fifth, some demographic and diseaseand treatment-related characteristics were dichotomized a strategy that may affect the interactions with symptom interference. Finally,
group (high- or low-risk group) on SIL-Psy among Taiwanese lung cancer patients. These findings are similar to the previous studies showing that distress ranked as the most important symptom interfering with psychological activities (Steinberg et al., 2009; Zabora et al., 2001). Patients with psychological problems expressed higher frequency of drowsiness and nausea and higher intensity of drowsiness (Delgado-Guay et al., 2009). Distress was a component similar to sadness, and these symptoms may provide insights into the underlying mechanisms associated with the occurrence of multiple symptoms (Wang et al., 2008). Thus, in order to reduce SIL-Psy, it is necessary to develop effective interventions to lower or eliminated distress symptoms. It is worth mentioning that SIL-DLA was also segregated into the two discrete categories of SIL-Phys and SIL-Psy in order to identify the most important combination patterns for each. Although most previous studies have demonstrated that individual symptoms or symptom clusters would impact on symptom interference, their findings could not identify the combination patterns and how symptoms interact with each other. The present study used regression tree models to provide a different perspective from which to explore the symptom combinations and to differentiate
Parent node N=131
Performance status
Good performance status Child node 1 N=105
Poor performance status Child node 2 N=26
Shortness of breath
Shortness of breath < 7.5 Child node 3 N=96
Employment
Shortness of breath > 7.5 N=9 SIL-Phys= 6.19
Terminal node 4 Pain
Pain < 1.5 N=43 SIL-Phys= 1.06
Employed, Unemployed or retaired N=17 SIL-Phys= 5.73 Terminal node 5
Unemployed due to disease N=9 SIL-Phys= 9.07 Terminal node 6
Pain > 1.5 Child node 4 N=53
Terminal node 1 Dry mouth
Dry mouth < 8.5 N=46 SIL-Phys= 2.51
Dry mouth > 8.5 N=7 SIL-Phys= 4.86
Terminal node 2
Terminal node 3
Fig. 2. Regression tree for symptom interference level in physical activities of 131 lung cancer patients in Taiwan. SIL-Phys, symptom interference level in physical activities; Parent node, start with an empty tree; Child node, the splitting attribute that was the most appropriate in separating the samples into distinct classes based on statistical goodness measure of split; Terminal node, the process was conducted recursively until the criterion of statistical stop rule was reached.
H.-H. Pan et al. / European Journal of Oncology Nursing 16 (2012) 345e352
351
Parent node N=131
Distress
Distress < 4.5 Child node 1 N=100
Distress > 4.5 Child node 2 N=31
Sadness
Sadness < 0.5 N=71 SIL-Psy= 0.90
Terminal node 1
Drowsiness
Sadness > 0.5 N=29 SIL-Psy= 2.47
Drowsiness < 4.5 Child node 3 N=21
Terminal node 5
Nausea
Terminal node 2
Drowsiness < 4.5 N=10 SIL-Psy= 7.03
Nausea < 1 N=11 SIL-Psy= 2.82
Nausea > 1 N=29 SIL-Psy= 5.57
Terminal node 3
Terminal node 4
Fig. 3. Regression tree for symptom interference level in psychological activities of 131 lung cancer patients in Taiwan. SIL-Psy, symptom interference level in psychological activities; Parent node, start with an empty tree; Child node, the splitting attribute that was the most appropriate in separating the samples into distinct classes based on statistical goodness measure of split; Terminal node, the process was conducted recursively until the criterion of statistical stop rule was reached.
the regression tree models can help to identify risk factors associated with outcome variables, but cannot demonstrate the statistical significance of such factors. Based on the results of this study, future research directions are indicated. First, studies using additional empirical assessment tools or clinical laboratory data are needed to determine the symptoms and symptom interference level. Second, random sampling and large sample research in Taiwan is needed to confirm our findings. Third, further studies using lung cancer inpatients are needed to increase the applicability and generalizability of results to the overall lung cancer population. Forth, studies using a longitudinal design are needed to explore the changes in symptom severity and symptom interference over the course of lung cancer disease. Fifth, the variables of demographics and disease- and treatment-related characteristics such as co-morbidity or complications should be measured and classified into more groups to predict symptom interference for the further study. Results from these kinds of studies, the results may lead to the development of novel interventions for symptom assessment and management. In conclusion, this study used an alternative approach to identify low- and high-risk groups of symptom interference among lung cancer patients in Taiwan. Increased awareness and further understanding of the risk combinations and discriminate levels of symptom severity that induced high symptom interference offer different perspectives to develop patient-centered care planning for lung cancer patient rehabilitation.
Conflict of interest statement None declared. Acknowledgements This study was supported by the grants from the Tri-Service General Hospital, Taiwan (TSGH-C93-49).The authors gratefully thank physicians and nurses helped to recruit patients. References Armitage, P., Berry, G., Matthews, J.N.S., 2002. Statistical Methods in Medical Research. Blackwell, Oxford. Armstrong, T.S., 2003. Symptoms experience: a concept analysis. Oncology Nursing Forum 30, 601e606. Brant, J.M., Beck, S., Miaskowski, C., 2010. Building dynamic models and theories to advance the science of symptom management research. Journal of Advanced Nursing 66, 228e240. Breiman, L., Friedman, R.A., Olshen, R.A., Stone, C.G., 1984. Classification and Regression Trees. Wadsworth, Pacific Grove, CA. Buccheri, G., Ferrigno, D., Tamburini, M., 1996. Karnofsky and ECOG performance status scoring in lung cancer: a prospective, longitudinal study of 536 patients from a single institution. European Journal of Cancer 32A, 1135e1141. Cheung, W.Y., Le, L.W., Zimmermann, C., 2009. Symptom clusters in patients with advanced cancers. Supportive Care in Cancer 17, 1223e1230. Chiang, T.A., Chen, P.H., Wu, P.F., Wang, T.N., Chang, P.Y., Ko, A.M., Huang, M.S., Ko, Y.C., 2008. Important prognostic factors for the long-term survival of lung cancer subjects in Taiwan. BMC Cancer 8, 324. Chow, E., Fan, G., Hadi, S., Filipczak, L., 2007. Symptom clusters in cancer patients with bone metastases. Supportive Care in Cancer 15, 1035e1043.
352
H.-H. Pan et al. / European Journal of Oncology Nursing 16 (2012) 345e352
Cleeland, C.S., Mendoza, T.R., Wang, X.S., Chou, C., Harle, M.T., Morrissey, M., Engstrom, M.C., 2000. Assessing symptom distress in cancer patients: the M.D. Anderson Symptom Inventory. Cancer 89, 1634e1646. Cohen, J., 1988. Statistical Power Analysis for the Behavioral Sciences. Erlbaum, Hillsdale, NJ. Delgado-Guay, M., Parsons, H.A., Li, Z., Palmer, J.L., Bruera, E., 2009. Symptom distress in advanced cancer patients with anxiety and depression in the palliative care setting. Supportive Care in Cancer 17, 573e579. Department of Health, 2008. Statistics of Causes of Death, 2008. Department of Health, Taipei. Dodd, M.J., Miaskowski, C., Lee, K.A., 2004. Occurrence of symptom clusters. Journal of the National Cancer Institute. Monographs, 76e78. Erdfelder, E., Faul, F., Buchner, A., 1996. GPOWER: a general power analysis program. Behavior Research Methods, Instruments, and Computers 28, 1e11. Everitt, B.S., 2004. Modern Medical Statistics. Hodder Arnold, Oxford, New York. Gift, A.G., Jablonski, A., Stommel, M., Given, C.W., 2004. Symptom clusters in elderly patients with lung cancer. Oncology Nursing Forum 31, 202e212. Henly, S.J., Kallas, K.D., Klatt, C.M., Swenson, K.K., 2003. The notion of time in symptom experiences. Nursing Research 52, 410e417. Keller, M.B., McCullough, J.P., Klein, D.N., Arnow, B., Dunner, D.L., Gelenberg, A.J., Markowitz, J.C., Nemeroff, C.B., Russell, J.M., Thase, M.E., Trivedi, M.H., Zajecka, J., 2000. A comparison of nefazodone, the cognitive behavioral-analysis system of psychotherapy, and their combination for the treatment of chronic depression. New England Journal of Medicine 342, 1462e1470. Kim, E., Jahan, T., Aouizerat, B.E., Dodd, M.J., Cooper, B.A., Paul, S.M., West, C., Lee, K., Swift, P.S., Wara, W., Miaskowski, C., 2009. Changes in symptom clusters in patients undergoing radiation therapy. Supportive Care in Cancer 17, 1383e1391. Kobayashi, K., Morita, S., Shimonagayoshi, M., Kobayashi, M., Fujiki, Y., Uchida, Y., Yamaguchi, K., 2008. Effects of socioeconomic factors and cancer survivors’ worries on their quality of life (QOL) in Japan. Psychooncology 17, 606e611. Larson, P., Carrieri-Kohlman, V., Dodd, M., Douglas, M., Faucett, J., Froelicher, E., Gortner, S., Haliburton, P., Janson, S., Lee, K., Miaskowski, C., Savedra, M., Stotts, N.A., Taylor, D., Underwood, P., 1994. A model for symptom management. Image Journal of Nursing Scholarship 26, 272e276. Lenz, E.R., Suppe, F., Gift, A.G., Pugh, L.C., Milligan, R.A., 1995. Collaborative development of middle-range nursing theories: toward a theory of unpleasant symptoms. Advances in Nursing Science 17, 1e13. Li, C.C., Chen, Y.M., Tsay, S.L., Hu, G.C., Lin, K.C., 2010. Predicting functional outcomes in patients suffering from ischaemic stroke using initial admission variables and physiological data: a comparison between tree model and multivariate regression analysis. Disability and Rehabilitation 32, 2088e2096. Lin, C.C., Chang, A.P., Cleeland, C.S., Mendoza, T.R., Wang, X.S., 2007. Taiwanese version of the M. D. Anderson symptom inventory: symptom assessment in cancer patients. Journal of Pain and Symptom Management 33, 180e188. Maric, D.M., Jovanovic, D.M., Golubicic, I.V., Nagorni-Obradovic, L.J., Stojsic, J.M., Pekmezovic, T.D., 2010. Psychological well-being in advanced NSCLC patients in Serbia: impact of sociodemographic and clinical factors. Neoplasma 57, 1e7.
Oberst, K., Bradley, C.J., Gardiner, J.C., Schenk, M., Given, C.W., 2010. Work task disability in employed breast and prostate cancer patients. Journal of Cancer Survivorship 4, 322e330. Oken, M.M., Creech, R.H., Tormey, D.C., Horton, J., Davis, T.E., McFadden, E.T., Carbone, P.P., 1982. Toxicity and response criteria of the Eastern Cooperative oncology group. American Journal of Clinical Oncology 5, 649e655. Parker, K.P., Kimble, L.P., Dunbar, S.B., Clark, P.C., 2005. Symptom interactions as mechanisms underlying symptom pairs and clusters. Journal of Nursing Scholarship 37, 209e215. Reddy, S.K., Parsons, H.A., Elsayem, A., Palmer, J.L., Bruera, E., 2009. Characteristics and correlates of dyspnea in patients with advanced cancer. Journal of Palliative Medicine 12, 29e36. Rodriguez Vega, B., Palao, A., Torres, G., Hospital, A., Benito, G., Perez, E., Dieguez, M., Castelo, B., Bayon, C., 2010. Combined therapy versus usual care for the treatment of depression in oncologic patients: a randomized controlled trial. Psycho-oncology Available at. http://onlinelibrary.wiley.com/doi/10.1002/pon. 1800/pdf. Rogers, L.Q., Courneya, K.S., Robbins, K.T., Malone, J., Seiz, A., Koch, L., Rao, K., 2008. Physical activity correlates and barriers in head and neck cancer patients. Supportive Care in Cancer 16, 19e27. Sarna, L., 1994. Functional status in women with lung cancer. Cancer Nursing 17, 87e93. Steinberg, T., Roseman, M., Kasymjanova, G., Dobson, S., Lajeunesse, L., Dajczman, E., Kreisman, H., Macdonald, N., Agulnik, J., Cohen, V., Rosberger, Z., Chasen, M., Small, D., 2009. Prevalence of emotional distress in newly diagnosed lung cancer patients. Supportive Care in Cancer 17, 1493e1497. Tanaka, K., Akechi, T., Okuyama, T., Nishiwaki, Y., Uchitomi, Y., 2002a. Impact of dyspnea, pain, and fatigue on daily life activities in ambulatory patients with advanced lung cancer. Journal of Pain and Symptom Management 23, 417e423. Tanaka, K., Akechi, T., Okuyama, T., Nishiwaki, Y., Uchitomi, Y., 2002b. Prevalence and screening of dyspnea interfering with daily life activities in ambulatory patients with advanced lung cancer. Journal of Pain and Symptom Management 23, 484e489. Teng, J.H., Lin, K.C., Ho, B.S., 2007. Application of classification tree and logistic regression for the management and health intervention plans in a communitybased study. Journal of Evaluation in Clinical Practice 13, 741e748. Tseng, T.H., Cleeland, C.S., Wang, X.S., Lin, C.C., 2008. Assessing cancer symptoms in adolescents with cancer using the Taiwanese version of the M. D. Anderson Symptom Inventory. Cancer Nursing 31, E9eE16. Wang, S.Y., Tsai, C.M., Chen, B.C., Lin, C.H., Lin, C.C., 2008. Symptom clusters and relationships to symptom interference with daily life in Taiwanese lung cancer patients. Journal of Pain and Symptom Management 35, 258e266. Wells, N., 2000. Pain intensity and pain interference in hospitalized patients with cancer. Oncology Nursing Forum 27, 985e991. Wells, N., Murphy, B., Wujcik, D., Johnson, R., 2003. Pain-related distress and interference with daily life of ambulatory patients with cancer pain. Oncology Nursing Forum 30, 977e986. Zabora, J., Brintzenhofeszoc, K., Curbow, B., Hooker, C., Piantadosi, S., 2001. The prevalence of psychological distress by cancer site. Psycho-oncology 10, 19e28.
Contents lists available at ScienceDirect
European Journal of Oncology Nursing journal homepage: www.elsevier.com/locate/ejon
Factors related to daily life interference in lung cancer patients: A cross-sectional regression tree study Hsueh-Hsing Pan a, Kuan-Chia Lin b, Shung-Tai Ho c, Chun-Yu Liang a, Shih-Chun Lee d, Kwua-Yun Wang e, f, * a
Graduate Institute of Medical Science, National Defense Medical Center, No. 161, Sec. 6, Minquan E. Rd., Neihu Dist., Taipei City 114, Taiwan Department of Nursing, National Taipei College of Nursing, No. 365, Ming-te Road, Peitou District, Taipei City, Taiwan Department of Anesthesiology, Taipei Veterans General Hospital, No. 201, Sec. 2, Shipai Rd., Taipei 112, Taiwan d Department of Thoracic Surgery, Tri-Service General Hospital, No. 325, Sec. 2, Chenggong Rd., Neihu District, Taipei City 114, Taiwan e Department of Nursing, Taipei Veterans General Hospital, No. 201, Sec. 2, Shipai Rd., Taipei 112, Taiwan f School of Nursing, National Defense Medical Center, No. 161, Sec. 6, Minquan E. Rd., Neihu Dist., Taipei City 114, Taiwan b c
a b s t r a c t Keywords: Daily life interference Lung cancer Physical activity interference Psychological activity interference Regression tree model
Purpose: To identify the symptom combination patterns and symptom severity levels that induce severe symptom interference in daily life activities, including physical and psychological activity interference in lung cancer patients. Methods: In a cross-sectional descriptive study using convenience sampling, 131 participants were recruited at a medical center in northern Taiwan. The Eastern Cooperative Oncology Group (ECOG) performance status was used to assess performance status, and the Taiwanese version of the M.D. Anderson Symptom Inventory (MDASI-T) was used to assess symptom severity and symptom interference in daily life activities including physical and psychological activities. Regression tree models were applied to examine variable combinations for symptom interference level in daily life activities, including physical and psychological activity interference. Results: Study results revealed that the performance status is the key discriminator of the symptom interference level in daily life and physical activities, but distress severity is the key discriminating factor of the symptom interference level in psychological activities. The performance status and distress severity, plus other factors, further specifically show the discrimination paths and interactions between the risk groups. Conclusions: This study provided an alternative approach to identify low- and high-risk groups of symptom interference among lung cancer patients in Taiwan. Increased awareness and further understanding of the risk combinations and discriminate levels of symptom severity that induced high symptom interference offer different perspectives to develop patient-centered care planning for lung cancer patient rehabilitation. Ó 2011 Elsevier Ltd. All rights reserved.
Introduction Among cancer patients, symptoms usually occur not in isolation, but in pairs, groups, or clusters (Dodd et al., 2004), and also result from a variety of physiological, psychological, behavioral, and sociocultural factors interacting with each other (Parker et al., 2005). Recent research has increasingly focused on the exploration of symptom clusters, and demonstrated that different symptoms may * Corresponding author. Department of Nursing, Taipei Veterans General Hospital, No. 201, Sec. 2, Shipai Rd., Taipei 112, Taiwan. Tel.: þ886 2 28757233, 886 2 87923100x18766; fax: þ886 2 28752932. E-mail address: [email protected] (K.-Y. Wang). 1462-3889/$ e see front matter Ó 2011 Elsevier Ltd. All rights reserved. doi:10.1016/j.ejon.2011.07.011
occur in combination or influence each other (Cheung et al., 2009; Kim et al., 2009; Wang et al., 2008). Studies also indicate that symptom clusters significantly correlate with daily life activities (Wang et al., 2008). Findings show that symptom severity correlates with many aspects of illness, including treatment-related factors, psychosocial factors, physical conditions, comorbidities, and personal profiles (Gift et al., 2004). Symptom severity can seriously impact on patients’ daily lives. However, symptom interference, including physical and psychological activity interference, refers to interference with functional status in terms of general activities, work, walking, mood, enjoyment of life and relationships with others (Cleeland et al., 2000). Previous studies showed that higher levels
346
H.-H. Pan et al. / European Journal of Oncology Nursing 16 (2012) 345e352
of symptom severity could produce greater symptom interference among cancer patients (Tanaka et al., 2002b; Wells, 2000; Wells et al., 2003). Most prior research on the relationship between symptom severity and symptom interference among lung cancer patients has focused on the study of individual symptoms in isolation or clusters rather than on the potential interactions and inter-relationships among various symptoms that can be used to classify subjects into high- and low-risk groups. Existing conceptual models or theories, such as the Symptom Management Model (Larson et al., 1994), the Theory of Unpleasant Symptoms (Lenz et al., 1995), the Symptoms Experience Model (Armstrong, 2003) and the Symptoms Experience in Time Model (Henly et al., 2003) have been applied in a variety of symptom management studies. It has been proposed that these models and theories may reveal important knowledge gaps concerning symptom interactions and the impact of interventions on patient outcomes (Brant et al., 2010). Lung cancer has been the leading cause of cancer-related deaths in Taiwan since 1999, and the incidence has continued on an upward trend over the past decade (Department of Health, 2008). The overall 5-year rate of survival is in the 21%e24% range (Chiang et al., 2008). These statistics reveal that lung cancer is still a significant clinical problem. Hence, the present study sought to identify symptom combination patterns (symptoms coupled with certain demographics, disease- and treatment-related characteristics) and symptom severity levels that induce severe symptom interference in daily life activities, including physical and psychological activity interference in lung cancer patients using regression tree modeling. Methods Study participants This was a cross-sectional study. Based on the procedure related to the conditional or fixed factors model of multiple regression (Erdfelder et al., 1996), we assumed that the dependent variable (daily life interference) was predicted by as a set B of 10e15 predictors. We further assumed that the population R2YjB was 0.15 (moderate effect size) (Cohen, 1988), that the 15e20 predictors would account for 15% of the variance of Y. Then with an alpha of 0.05 and 80% power (Erdfelder et al., 1996), the required sample size was estimated to be 125 subjects. The inclusion criteria were 1) pathologically diagnosed lung cancer; 2) over 18 years of age; 3) free of cognitive impairments/mental illness; 4) able to communicate in Mandarin or Taiwanese and 5) willing to participate. In this study, a total of 132 lung cancer patients were recruited from outpatient hematology/oncology, radiation oncology and thoracic surgery departments in an approx 1700-bed medical center in northern Taiwan between January and April 2004. Twenty to twenty-five lung cancer patients were treated each month in this medical center. Among 132 lung cancer patients, only one patient was too weak to complete the interview. Finally, 131 participants (65 men, 49.6%; 66 women, 50.4%) completed the study. Instruments Demographics, disease- and treatment-related characteristics Demographic characteristics included participant age, gender, years of education (more or less than nine years), marital status (married or other), employment status (i.e., employed, unemployed or retired, unemployed due to disease), religious affiliation (yes or no) and smoking status (i.e., never smoked, has stopped smoking, still smoking). Disease-related characteristics consist of lung cancer cell type (small and non-small cell), stage of disease (IA-IIIA or IIIB-
IV), metastasis, disease duration, complications (yes or no), comorbidity (yes or no) and self-perceived disease severity (i.e., no, moderate, and severe). Treatment-related characteristics collected were cancer-related treatments (i.e., chemotherapy, chemotherapy combined with radiotherapy, and/or operation, others) received since cancer diagnosis and during the most recent one-week period, narcotics and/or analgesics used or not, and self-perceived treatment effectiveness (yes or no). Performance status Performance status was measured by the Eastern Cooperative Oncology Group (ECOG) performance status, referring to the functional status of cancer patients rated by patient self-report. The ECOG was rated on a scale of 0e5, where 0 was being fully active; 1 being restricted in physically strenuous activity but ambulatory; 2 being ambulatory and capable of all self-care; 3 capable of only limited self-care; 4 completely disabled; and 5 being dead (Oken et al., 1982). In view of the single item design of ECOG, equivalence reliability cannot be acquired. In terms of final score, performance status was classified as good (scores of 0e1) performance or poor (scores of 2e4) (Cleeland et al., 2000). In another study, the ECOG was documented as having good predictive validity (Buccheri et al., 1996). In our study, this questionnaire was translated into Chinese, and back-translation was used to verify semantics of the Chinese version. Symptom severity and symptom interference Symptom severity and symptom interference were measured by the Taiwanese version of the M.D. Anderson Symptom Inventory (MDASI-T). The original MDASI included two subscales: the first subscale covered the 13 core symptom items of pain, fatigue, sleep disturbance, distress, shortness of breath, memory difficulties, drowsiness, dry mouth, sadness, numbness, poor appetite, nausea and vomiting. The second subscale mainly addressed daily life activities, which integrated six symptom interference items into two categories of physical activities (general activity, walking, and normal work) and psychological activities (mood, relationships with others, and enjoyment of life) (Cleeland et al., 2000). MDASI is a patient-reported outcome tool validated for use among cancer patients and was developed to measure symptom severity and symptom interference with activities of daily living during the most recent 24-h period in cancer patients. Each symptom severity item was rated on an 11-point numeric scale, from 0 (not severe at all) to 10 (as severe as you can imagine). Each symptom interference item was rated on a similar scale, from 0 (no interference) to 10 (complete interference). The MDASI-T was developed using a standard translation and back-translation procedure, with content validity evaluated by five experts (one oncologist, two clinical oncology specialists and two oncology professors) and five cancer patients. Cronbach’s a of internal consistency was 0.87 for symptom severity items and 0.86 for interference items. Test-retest reliability over a one-week interval was 0.77 for symptom severity items and 0.86 for interference items. In a previous Chinese study, MDASI-T was validated in a sample of 556 Taiwanese patients with multiple diagnoses of cancer. The internal consistency Cronbach’s a was 0.89 for symptom severity items and 0.94 for interference items. The test-retest reliability over a 3-day interval was 0.97 for symptom severity items and 0.86 for the interference items. Construct validity was established by factor analysis and concurrent validity was examined by correlating the MADSI-T scores and scores of the Medical Outcome Study 36-Item Short-Form Health Survey (Lin et al., 2007).
H.-H. Pan et al. / European Journal of Oncology Nursing 16 (2012) 345e352
Data collection procedures The study was approved by the institution and permission was granted by the participating patients. Physicians referred the patients who met the inclusion criteria to the researcher. Then, the researcher provided a verbal explanation of the study. All patients were informed that they could withdraw from the study at any time without penalty, and all information would be kept confidential. Patients completed the questionnaire during a face-to-face interview with the same researcher to ensure the quality of data collection and to avoid missing data. Disease- and treatmentrelated information were obtained from the participants’ medical charts following completion of the questionnaire.
347
(5) As extremely large and complex tree structures would be unexplainable in practical application, we further applied the cross-validation to randomly divide the sample data into a model-building set (90% of the data sample) and a validation set (10% of the data sample). The model-building set was then used to establish a tree that was pruned using the validation set to achieve an estimation of the most appropriate tree through the minimum cost-complexity test. (6) The final model was established using the forward (likelihood ratio based), estimated (using Wald-c2) and 2 log-likelihood change. The P-value was set at 0.05 as the criterion for selecting independent variables. Results
Statistical analysis Participant characteristics Demographics, disease- and treatment-related characteristics were expressed as a mean standard deviation (SD) for continuous variables and as proportions for categorical variables. The type of combinations (symptoms coupled with certain demographics, disease- and treatment-related characteristics) and discriminate levels were used in a regression tree analysis. The regression trees in this study were run by S-Plus 6.2 software (Insightful Inc., Seattle, WA, USA), and the growing, stopping, and pruning of the trees were determined by the Gini improvement measure (Everitt, 2004). Overall, the symptom interference level in daily life activities (SIL-DLA) was segregated into two discrete categories: symptom interference level in physical activities (SIL-Phys) and symptom interference level in psychological activities (SIL-Psy). These activities were adopted as dependent variables in the regression tree models. Some demographic data, disease- and treatment-related variables, and symptom severity were treated as independent variables. As the name implies, a regression tree can be viewed as a classifier in the form of a tree structure, in which each node is either a child node (leaf node) or terminal node (decision node). All terminal nodes have splits, testing the symptom interference level. Each branch from the terminal node corresponds to a distinct outcome of the test. Each child node also has a class label attached to it. Although a variety of regression tree algorithms have been developed with different capabilities and requirements, most are variations of a core learning algorithm that employs a Top-down Greedy Split (TGS) search through the space of a possible regression tree. The strategies for tree construction in this study are as follows (Armitage et al., 2002; Breiman et al., 1984; Li et al., 2010; Teng et al., 2007): (1) Start with an empty tree (parent node) and the entire training set. (2) Select the splitting attribute (risk factor) that was the most appropriate in separating the samples into distinct classes based on statistical goodness measure of the split. This attribute then becomes a child node. In choosing the best splitter, a number of different measures of purity can be selected, simply called “splitting criteria” or “splitting functions.” The most common splitting function is the “Gini.” Therefore, we used “Gini” to make the splits in the data, as opposed to the information gain measure that other tree classifiers use. (3) A branch was created for each distinct value of risk factor and the samples were partitioned accordingly. (4) The process was conducted recursively until the criterion of statistical stop rule was reached, such that all samples for n belonged to the same class (terminal node), no risk factor could be further partitioned, or the improvement was not substantial enough to justify further splitting.
Demographics, disease- and treatment-related characteristics are shown in Table 1. The participant average age was 63.9 years (range: 27e89 years). The mean duration of disease since diagnosis was 12.4 months (range: 1e109 months). Almost three quarters (73.3%) of the participants had at least one co-morbidity and 91.6% were diagnosed with non-small cell lung cancer; 57.3% of the participants had no metastasis and 80.2% had good performance status (ECOG ¼ 0e1), while 77.9% had received chemotherapy or combined cancer-related therapy treatments since diagnosis. Symptom severity and symptom interference Symptom severity and symptom interference are shown in Table 2. The highest ranking five mean levels of symptom severity were fatigue, dry mouth, shortness of breath, sleep disturbance and pain. However, the highest ranking three activities in regard to the symptom interference level were work, walking and enjoyment of life. Mean SD scores for SIL-DLA were 2.76 2.58, while in SILPhys and SIL-Psy they were 3.28 2.95 and 2.23 2.59 respectively. From a statistical point of view, the large SD in several symptom severity and symptom interference level indicators reflect that the data points are not clustered closely around the mean. Regression tree models Fig. 1 displays the regression tree model for overall SIL-DLA, in which the most appropriate tree size was divided into five terminal nodes. The results show that the most important discriminator (the first layer in this tree model) was performance status. The performance status then combined with the secondary layer factors, including distress or sadness, followed by different discriminate levels in regard to symptom severity. In addition, shortness of breath further added into the model as the third layer to induce the different SIL-DLA. Terminal node 5 indicates the highest SIL-DLA such that when performance status was poor and sadness severity score >4.5, the predicted functional outcome for SIL-DLA was 7.87 and the population size was 9. By contrast, terminal node 1 indicates the lowest SIL-DLA such that when performance status was good, distress severity score <4.5, and shortness of breath severity score <7.5, the predicted functional outcome for SIL-DLA was 1.28 and the population size was 78. The other combinatorial paths, inter-relationships, and populations are also illustrated in Fig. 1. SIL-DLA was segregated into the two discrete categories of SILPhys and SIL-Psy. As shown in Fig. 2, the most appropriate tree size for predicting SIL-Phys was divided into six terminal nodes. This study also demonstrated that performance status was the most
348
H.-H. Pan et al. / European Journal of Oncology Nursing 16 (2012) 345e352
important discriminator in the first layer and combined with varying discriminate levels of symptom severity or other variables such as shortness of breath and employment status (the secondary layer), pain (the third layer), and dry mouth (the fourth layer) to produce different SIL-Phys. On the right-hand side, terminal node 6 indicates the highest SIL-Phys such that when the performance status was poor and unemployment was due to disease, the predicted functional outcome for SIL-Phys was 9.07 and the population size was 9. By contrast, terminal node 1 showed that when the performance status was good, shortness of breath severity score
Table 1 Patient demographics and disease-related characteristics (n ¼ 131). Variable
n
Mean age (SD) 63.9 Age range 27e89 Mean disease duration (SD) 12.4 (months) Gender Male 65 Female 66 Years of education < 9 yrs 78 > 9 yrs 53 Marital status Married 93 Other 38 Employment Employed 10 Unemployed or retired 81 Unemployed due to disease 40 Religious affiliation Yes 87 No 44 Smoking status Never smoked 60 Stopped smoking 62 Keeps smoking 9 Cell type SCLC 11 NSCLC 120 Stage of disease IA-IIIA 24 IIIB-IV 107 Metastasis Yes 56 No 75 Complication(s) Yes 63 No 68 Comorbidity(ies) Yes 96 No 35 Performance status Good (ECOG ¼ 0-1) 105 Poor (ECOG ¼ 2-4) 26 Self-perceived disease severity None 50 Moderate 34 Severe 47 Treatment following disease diagnosis Chemotherapy 41 Chemotherapy combined with others 61 Others 29 Treatment administered during the previous week Yes 48 No 83 Narcotics and/or analgesics used or not Yes 41 No 90 Self-perceived treatment effect (n ¼ 122) Yes 70 No 52
(%) (12.8) (14.8) (49.6) (50.4) (59.5) (40.5) (71.0) (29.0)
Table 2 Symptom severity and symptom interference in lung cancer patients (n ¼ 131). Item
Mean
SD
Symptom severity Fatigue Dry mouth Shortness of breath Sleep disturbance Pain Lack of appetite Sadness Distress Drowsiness Difficulty remembering Numbness Nausea Vomiting Symptom interference Physical interference Work Walking General activity Psychological interference Enjoyment of life Mood Relationships with others
2.35 3.48 2.91 2.80 2.63 2.56 2.54 2.51 2.47 2.40 2.02 1.93 1.50 0.81 2.76 3.28 4.22 3.42 2.20 2.23 2.82 2.50 1.38
1.59 3.37 3.38 2.91 3.24 3.05 3.46 3.50 3.36 3.11 3.05 2.80 2.79 2.10 2.58 2.95 3.73 3.33 3.17 2.59 3.69 3.35 2.65
SD ¼ standard deviation.
(42.7) (57.3)
<7.5, and pain severity score <1.5, the predicted functional outcome for SIL-Phys was 1.06 and the population size was 43. The other combinatorial paths, inter-relationships, and populations are also illustrated in Fig. 2. Fig. 3 shows the most appropriate tree size for predicting SIL-Psy was divided into five terminal nodes. This study indicated that distress was the most important discriminator in the first layer, sadness and drowsiness were in the secondary layer, and nausea was in the third layer. Terminal node 5 indicates the highest SIL-Psy such that when the distress severity score was >4.5, and drowsiness severity score >4.5, the predicted functional outcome for SILPsy was 7.03 and the population size was 10. By contrast, terminal node 1 indicates the lowest SIL-Psy such that when the distress severity score was <4.5, and sadness severity score <0.5, the predicted functional outcome for SIL-Psy was 0.9 and the population size was 71. The other combinatorial paths, inter-relationships, and populations are also illustrated in Fig. 3.
(48.1) (51.9)
Discussion
(7.6) (61.8) (30.5) (66.4) (33.6) (45.8) (47.3) (6.9) (8.4) (91.6) (18.3) (81.7)
(73.3) (26.7) (80.2) (19.8) (38.2) (26.0) (35.9) (31.3) (46.6) (22.1) (36.7) (63.4) (31.3) (68.7) (57.4) (42.6)
SD ¼ standard deviation; SCLC ¼ small cell lung cancer; NSCLC ¼ non-small cell lung cancer; ECOG ¼ Eastern Cooperative Oncology Group.
This study provided essential information to identify combination patterns of symptom interaction as well as each cutoff point related to symptom interference level among Taiwanese lung cancer patients. Although a similar study design, sampling procedure, and questionnaire was done in Taiwan before (Wang et al., 2008), the authors used traditional statistical modeling methods such as bivariate analysis, hierarchical cluster analysis, and linear regression to detect the possible related factors of certain health outcomes and to estimate the statistical prediction. Consequently, it was difficult to identify and explore complex data patterns and to model higher order interactions. The regression tree analysis used in this study is different from traditional variable-oriented regression analysis. It is a subject-oriented multivariate statistic that aims to recognize people with the same characteristics by the overlap of many factors. Interestingly, in this study, regression tree analysis pointed out the high- and low-risk patients and determined optimal cutoff points of symptom levels. This makes it possible to classify individuals into distinct groups or categories based on individual response patterns so that individuals within a group are more similar than individuals between groups.
H.-H. Pan et al. / European Journal of Oncology Nursing 16 (2012) 345e352
The findings demonstrated that the symptom interference level may be explained and predicted through the serial interactions of a number of independent variables, which allowed us to find the crucial role in identifying homogeneity subgroups of Taiwanese lung cancer patients. Based on this, we found that performance status was the key discriminator of the SIL-DLA and SIL-Phys, and the distress severity was the key discriminator of the SIL-Psy. The performance status or distress severity plus other factors further showed the discrimination paths and interactions between the risk groups. In other words, there were different tree structures between the SIL-DLA, SIL-Phys, and SIL-Psy. For the SIL-DLA, several previous studies have shown that patients with poor performance status, experienced greater interference with daily life activities (Cleeland et al., 2000; Lin et al., 2007; Tseng et al., 2008), and experienced significantly more anxiety and depressive symptoms and worse emotional and mental functioning (Maric et al., 2010). In addition, many prior studies have indicated that the severity of shortness of breath significantly interfered with activities including physical and psychological activities (Reddy et al., 2009; Tanaka et al., 2002b). Our findings from different tree structures imply that the performance status interacting with distress, sadness and shortness of breath appear to be sensitive to discriminate the high-risk group and relatively lowrisk group SIL-DLA among Taiwanese lung cancer patients. Symptom severity may reduce physical function, but may also manifest itself in decreased mental attentiveness, alertness and motivation. Untreated psychological distress is associated with reduced quality of life and inadequate palliation of physical
349
symptoms (Maric et al., 2010). Patients with poor performance status, and especially greater sadness or distress, may therefore require management of their psychological problems using psychotherapeutic and pharmacological interventions or combined therapies from clinicians in order to reduce SIL-DLA (Keller et al., 2000; Rodriguez Vega et al., 2010). This study revealed that the performance status interacts with shortness of breath, employment status, pain and dry mouth to influence on SIL-Phys. Study has indicated that patients with poor performance status also had reduced physical activity (Sarna, 1994). Patients with shortness of breath were reflected in a higher SILPhys (Reddy et al., 2009; Rogers et al., 2008; Tanaka et al., 2002b), which also tended to coexist with other symptoms such as fatigue, pain, dry mouth, drowsiness, nausea, distress and sadness (Cheung et al., 2009; Lin et al., 2007); pain often interfered with physical and psychological activities (Chow et al., 2007; Tanaka et al., 2002a); and dry mouth was significantly associated with physical activity (Rogers et al., 2008). Studies also indicated that the majority of cancer patients had a work-related disability following cancer diagnosis or cancer-related treatments (Oberst et al., 2010). Loss of employment because of cancer had a negative impact on their quality of life (Kobayashi et al., 2008). Therefore, lung cancer patients with poor performance and with symptoms such as shortness of breath, pain and dry mouth may also require more attention from clinicians in order to prevent a negative impact on physical activities. For the SIL-Psy, our findings suggested that distress severity interacted with sadness, drowsiness or nausea to discern the target
Parent node N=131
Performance status
Good performance status Child node 1 N=105
Poor performance status Child node 2 N=26
Distress
Sadness
Distress < 4.5 Child node 3 N=83 Shortness of breath
Distress > 4.5 N=22 SIL-DLA= 4.17 Terminal node 3
Shortness of breath < 7.5 N=78 SIL-DLA= 1.28
Shortness of breath > 7.5 N=5 SIL-DLA= 4.87
Terminal node 1
Terminal node 2
Sadness < 4.5 N=17 SIL-DLA= 4.37 Terminal node 4
Sadness > 4.5 N=9 SIL-DLA= 7.87 Terminal node 5
Fig. 1. Regression tree for symptom interference level in daily life activities of 131 lung cancer patients in Taiwan. SIL-DLA, symptom interference level in daily life activities; Parent node, start with an empty tree; Child node, the splitting attribute that was the most appropriate in separating the samples into distinct classes based on statistical goodness measure of split; Terminal node, the process was conducted recursively until the criterion of statistical stop rule was reached.
350
H.-H. Pan et al. / European Journal of Oncology Nursing 16 (2012) 345e352
the high- or low-risk groups. A distinctive finding in this study was that some symptoms such as dry mouth (Fig. 2) and nausea (Fig. 3) were not the key discriminators (in the third or fourth layer) in the tree models, but affected other symptoms to generate different levels of symptom interference. This study has a number of limitations. First, the MDASI was a patient-reported assessment tool. Because of conservative folk customs, Taiwanese patients may be embarrassed or reluctant to report their symptom severity and symptom interference due to concern about bothering the nurses or physicians or fear of being stigmatized as a difficult patient. As a result, symptom severity and symptom interference might be under-reported by patients. Second, as our study population was a convenience sample and limited to one medical center, the generalizability of our findings is limited. Third, as this study only addressed lung cancer outpatients, our findings cannot be generalized to seriously ill lung cancer inpatients. Forth, this study used a cross-sectional design, meaning that change in symptom severity and symptom interference over time was not investigated. Fifth, some demographic and diseaseand treatment-related characteristics were dichotomized a strategy that may affect the interactions with symptom interference. Finally,
group (high- or low-risk group) on SIL-Psy among Taiwanese lung cancer patients. These findings are similar to the previous studies showing that distress ranked as the most important symptom interfering with psychological activities (Steinberg et al., 2009; Zabora et al., 2001). Patients with psychological problems expressed higher frequency of drowsiness and nausea and higher intensity of drowsiness (Delgado-Guay et al., 2009). Distress was a component similar to sadness, and these symptoms may provide insights into the underlying mechanisms associated with the occurrence of multiple symptoms (Wang et al., 2008). Thus, in order to reduce SIL-Psy, it is necessary to develop effective interventions to lower or eliminated distress symptoms. It is worth mentioning that SIL-DLA was also segregated into the two discrete categories of SIL-Phys and SIL-Psy in order to identify the most important combination patterns for each. Although most previous studies have demonstrated that individual symptoms or symptom clusters would impact on symptom interference, their findings could not identify the combination patterns and how symptoms interact with each other. The present study used regression tree models to provide a different perspective from which to explore the symptom combinations and to differentiate
Parent node N=131
Performance status
Good performance status Child node 1 N=105
Poor performance status Child node 2 N=26
Shortness of breath
Shortness of breath < 7.5 Child node 3 N=96
Employment
Shortness of breath > 7.5 N=9 SIL-Phys= 6.19
Terminal node 4 Pain
Pain < 1.5 N=43 SIL-Phys= 1.06
Employed, Unemployed or retaired N=17 SIL-Phys= 5.73 Terminal node 5
Unemployed due to disease N=9 SIL-Phys= 9.07 Terminal node 6
Pain > 1.5 Child node 4 N=53
Terminal node 1 Dry mouth
Dry mouth < 8.5 N=46 SIL-Phys= 2.51
Dry mouth > 8.5 N=7 SIL-Phys= 4.86
Terminal node 2
Terminal node 3
Fig. 2. Regression tree for symptom interference level in physical activities of 131 lung cancer patients in Taiwan. SIL-Phys, symptom interference level in physical activities; Parent node, start with an empty tree; Child node, the splitting attribute that was the most appropriate in separating the samples into distinct classes based on statistical goodness measure of split; Terminal node, the process was conducted recursively until the criterion of statistical stop rule was reached.
H.-H. Pan et al. / European Journal of Oncology Nursing 16 (2012) 345e352
351
Parent node N=131
Distress
Distress < 4.5 Child node 1 N=100
Distress > 4.5 Child node 2 N=31
Sadness
Sadness < 0.5 N=71 SIL-Psy= 0.90
Terminal node 1
Drowsiness
Sadness > 0.5 N=29 SIL-Psy= 2.47
Drowsiness < 4.5 Child node 3 N=21
Terminal node 5
Nausea
Terminal node 2
Drowsiness < 4.5 N=10 SIL-Psy= 7.03
Nausea < 1 N=11 SIL-Psy= 2.82
Nausea > 1 N=29 SIL-Psy= 5.57
Terminal node 3
Terminal node 4
Fig. 3. Regression tree for symptom interference level in psychological activities of 131 lung cancer patients in Taiwan. SIL-Psy, symptom interference level in psychological activities; Parent node, start with an empty tree; Child node, the splitting attribute that was the most appropriate in separating the samples into distinct classes based on statistical goodness measure of split; Terminal node, the process was conducted recursively until the criterion of statistical stop rule was reached.
the regression tree models can help to identify risk factors associated with outcome variables, but cannot demonstrate the statistical significance of such factors. Based on the results of this study, future research directions are indicated. First, studies using additional empirical assessment tools or clinical laboratory data are needed to determine the symptoms and symptom interference level. Second, random sampling and large sample research in Taiwan is needed to confirm our findings. Third, further studies using lung cancer inpatients are needed to increase the applicability and generalizability of results to the overall lung cancer population. Forth, studies using a longitudinal design are needed to explore the changes in symptom severity and symptom interference over the course of lung cancer disease. Fifth, the variables of demographics and disease- and treatment-related characteristics such as co-morbidity or complications should be measured and classified into more groups to predict symptom interference for the further study. Results from these kinds of studies, the results may lead to the development of novel interventions for symptom assessment and management. In conclusion, this study used an alternative approach to identify low- and high-risk groups of symptom interference among lung cancer patients in Taiwan. Increased awareness and further understanding of the risk combinations and discriminate levels of symptom severity that induced high symptom interference offer different perspectives to develop patient-centered care planning for lung cancer patient rehabilitation.
Conflict of interest statement None declared. Acknowledgements This study was supported by the grants from the Tri-Service General Hospital, Taiwan (TSGH-C93-49).The authors gratefully thank physicians and nurses helped to recruit patients. References Armitage, P., Berry, G., Matthews, J.N.S., 2002. Statistical Methods in Medical Research. Blackwell, Oxford. Armstrong, T.S., 2003. Symptoms experience: a concept analysis. Oncology Nursing Forum 30, 601e606. Brant, J.M., Beck, S., Miaskowski, C., 2010. Building dynamic models and theories to advance the science of symptom management research. Journal of Advanced Nursing 66, 228e240. Breiman, L., Friedman, R.A., Olshen, R.A., Stone, C.G., 1984. Classification and Regression Trees. Wadsworth, Pacific Grove, CA. Buccheri, G., Ferrigno, D., Tamburini, M., 1996. Karnofsky and ECOG performance status scoring in lung cancer: a prospective, longitudinal study of 536 patients from a single institution. European Journal of Cancer 32A, 1135e1141. Cheung, W.Y., Le, L.W., Zimmermann, C., 2009. Symptom clusters in patients with advanced cancers. Supportive Care in Cancer 17, 1223e1230. Chiang, T.A., Chen, P.H., Wu, P.F., Wang, T.N., Chang, P.Y., Ko, A.M., Huang, M.S., Ko, Y.C., 2008. Important prognostic factors for the long-term survival of lung cancer subjects in Taiwan. BMC Cancer 8, 324. Chow, E., Fan, G., Hadi, S., Filipczak, L., 2007. Symptom clusters in cancer patients with bone metastases. Supportive Care in Cancer 15, 1035e1043.
352
H.-H. Pan et al. / European Journal of Oncology Nursing 16 (2012) 345e352
Cleeland, C.S., Mendoza, T.R., Wang, X.S., Chou, C., Harle, M.T., Morrissey, M., Engstrom, M.C., 2000. Assessing symptom distress in cancer patients: the M.D. Anderson Symptom Inventory. Cancer 89, 1634e1646. Cohen, J., 1988. Statistical Power Analysis for the Behavioral Sciences. Erlbaum, Hillsdale, NJ. Delgado-Guay, M., Parsons, H.A., Li, Z., Palmer, J.L., Bruera, E., 2009. Symptom distress in advanced cancer patients with anxiety and depression in the palliative care setting. Supportive Care in Cancer 17, 573e579. Department of Health, 2008. Statistics of Causes of Death, 2008. Department of Health, Taipei. Dodd, M.J., Miaskowski, C., Lee, K.A., 2004. Occurrence of symptom clusters. Journal of the National Cancer Institute. Monographs, 76e78. Erdfelder, E., Faul, F., Buchner, A., 1996. GPOWER: a general power analysis program. Behavior Research Methods, Instruments, and Computers 28, 1e11. Everitt, B.S., 2004. Modern Medical Statistics. Hodder Arnold, Oxford, New York. Gift, A.G., Jablonski, A., Stommel, M., Given, C.W., 2004. Symptom clusters in elderly patients with lung cancer. Oncology Nursing Forum 31, 202e212. Henly, S.J., Kallas, K.D., Klatt, C.M., Swenson, K.K., 2003. The notion of time in symptom experiences. Nursing Research 52, 410e417. Keller, M.B., McCullough, J.P., Klein, D.N., Arnow, B., Dunner, D.L., Gelenberg, A.J., Markowitz, J.C., Nemeroff, C.B., Russell, J.M., Thase, M.E., Trivedi, M.H., Zajecka, J., 2000. A comparison of nefazodone, the cognitive behavioral-analysis system of psychotherapy, and their combination for the treatment of chronic depression. New England Journal of Medicine 342, 1462e1470. Kim, E., Jahan, T., Aouizerat, B.E., Dodd, M.J., Cooper, B.A., Paul, S.M., West, C., Lee, K., Swift, P.S., Wara, W., Miaskowski, C., 2009. Changes in symptom clusters in patients undergoing radiation therapy. Supportive Care in Cancer 17, 1383e1391. Kobayashi, K., Morita, S., Shimonagayoshi, M., Kobayashi, M., Fujiki, Y., Uchida, Y., Yamaguchi, K., 2008. Effects of socioeconomic factors and cancer survivors’ worries on their quality of life (QOL) in Japan. Psychooncology 17, 606e611. Larson, P., Carrieri-Kohlman, V., Dodd, M., Douglas, M., Faucett, J., Froelicher, E., Gortner, S., Haliburton, P., Janson, S., Lee, K., Miaskowski, C., Savedra, M., Stotts, N.A., Taylor, D., Underwood, P., 1994. A model for symptom management. Image Journal of Nursing Scholarship 26, 272e276. Lenz, E.R., Suppe, F., Gift, A.G., Pugh, L.C., Milligan, R.A., 1995. Collaborative development of middle-range nursing theories: toward a theory of unpleasant symptoms. Advances in Nursing Science 17, 1e13. Li, C.C., Chen, Y.M., Tsay, S.L., Hu, G.C., Lin, K.C., 2010. Predicting functional outcomes in patients suffering from ischaemic stroke using initial admission variables and physiological data: a comparison between tree model and multivariate regression analysis. Disability and Rehabilitation 32, 2088e2096. Lin, C.C., Chang, A.P., Cleeland, C.S., Mendoza, T.R., Wang, X.S., 2007. Taiwanese version of the M. D. Anderson symptom inventory: symptom assessment in cancer patients. Journal of Pain and Symptom Management 33, 180e188. Maric, D.M., Jovanovic, D.M., Golubicic, I.V., Nagorni-Obradovic, L.J., Stojsic, J.M., Pekmezovic, T.D., 2010. Psychological well-being in advanced NSCLC patients in Serbia: impact of sociodemographic and clinical factors. Neoplasma 57, 1e7.
Oberst, K., Bradley, C.J., Gardiner, J.C., Schenk, M., Given, C.W., 2010. Work task disability in employed breast and prostate cancer patients. Journal of Cancer Survivorship 4, 322e330. Oken, M.M., Creech, R.H., Tormey, D.C., Horton, J., Davis, T.E., McFadden, E.T., Carbone, P.P., 1982. Toxicity and response criteria of the Eastern Cooperative oncology group. American Journal of Clinical Oncology 5, 649e655. Parker, K.P., Kimble, L.P., Dunbar, S.B., Clark, P.C., 2005. Symptom interactions as mechanisms underlying symptom pairs and clusters. Journal of Nursing Scholarship 37, 209e215. Reddy, S.K., Parsons, H.A., Elsayem, A., Palmer, J.L., Bruera, E., 2009. Characteristics and correlates of dyspnea in patients with advanced cancer. Journal of Palliative Medicine 12, 29e36. Rodriguez Vega, B., Palao, A., Torres, G., Hospital, A., Benito, G., Perez, E., Dieguez, M., Castelo, B., Bayon, C., 2010. Combined therapy versus usual care for the treatment of depression in oncologic patients: a randomized controlled trial. Psycho-oncology Available at. http://onlinelibrary.wiley.com/doi/10.1002/pon. 1800/pdf. Rogers, L.Q., Courneya, K.S., Robbins, K.T., Malone, J., Seiz, A., Koch, L., Rao, K., 2008. Physical activity correlates and barriers in head and neck cancer patients. Supportive Care in Cancer 16, 19e27. Sarna, L., 1994. Functional status in women with lung cancer. Cancer Nursing 17, 87e93. Steinberg, T., Roseman, M., Kasymjanova, G., Dobson, S., Lajeunesse, L., Dajczman, E., Kreisman, H., Macdonald, N., Agulnik, J., Cohen, V., Rosberger, Z., Chasen, M., Small, D., 2009. Prevalence of emotional distress in newly diagnosed lung cancer patients. Supportive Care in Cancer 17, 1493e1497. Tanaka, K., Akechi, T., Okuyama, T., Nishiwaki, Y., Uchitomi, Y., 2002a. Impact of dyspnea, pain, and fatigue on daily life activities in ambulatory patients with advanced lung cancer. Journal of Pain and Symptom Management 23, 417e423. Tanaka, K., Akechi, T., Okuyama, T., Nishiwaki, Y., Uchitomi, Y., 2002b. Prevalence and screening of dyspnea interfering with daily life activities in ambulatory patients with advanced lung cancer. Journal of Pain and Symptom Management 23, 484e489. Teng, J.H., Lin, K.C., Ho, B.S., 2007. Application of classification tree and logistic regression for the management and health intervention plans in a communitybased study. Journal of Evaluation in Clinical Practice 13, 741e748. Tseng, T.H., Cleeland, C.S., Wang, X.S., Lin, C.C., 2008. Assessing cancer symptoms in adolescents with cancer using the Taiwanese version of the M. D. Anderson Symptom Inventory. Cancer Nursing 31, E9eE16. Wang, S.Y., Tsai, C.M., Chen, B.C., Lin, C.H., Lin, C.C., 2008. Symptom clusters and relationships to symptom interference with daily life in Taiwanese lung cancer patients. Journal of Pain and Symptom Management 35, 258e266. Wells, N., 2000. Pain intensity and pain interference in hospitalized patients with cancer. Oncology Nursing Forum 27, 985e991. Wells, N., Murphy, B., Wujcik, D., Johnson, R., 2003. Pain-related distress and interference with daily life of ambulatory patients with cancer pain. Oncology Nursing Forum 30, 977e986. Zabora, J., Brintzenhofeszoc, K., Curbow, B., Hooker, C., Piantadosi, S., 2001. The prevalence of psychological distress by cancer site. Psycho-oncology 10, 19e28.