The Symptom Phenotype of Oncology Outpatients Remains Relatively Stable From Prior to Through 1 Week Following Chemotherapy

Abstract

Evidence suggests that some patients experience a higher symptom burden. However, little is known about whether this high risk phenotype persists over time. Latent transition analysis (LTA) was used to examine the probability that patients remained in the same symptom class when assessed prior to the administration of and following their next dose of chemotherapy (CTX). For the patients whose class membership remained consistent, differences in demographic and clinical characteristics, and quality of life (QOL) were evaluated. The Memorial Symptom Assessment Scale (MSAS) was used to evaluate symptom burden. LTA was used to identify subgroups of patients with distinct symptom experiences based on the occurrence of the MSAS symptoms. Of the 906 patients evaluated, 83.9% were classified in the same symptom occurrence class at both assessments. Of these 760 patients 25.0% were classified as Low-Low, 44.1% as Moderate-Moderate, and 30.9% as High-High. Compared to the Low-Low class, the other two classes were younger, more likely to be female and to report child care responsibilities, and had a lower functional status and a higher comorbidity scores. The two higher classes reported lower QOL scores. The use of LTA could assist clinicians to identify higher risk patients and initiate more aggressive interventions.

INTRODUCTION

Patients receiving chemotherapy (CTX) report an average of ten co-occurring symptoms (Esther Kim et al., 2009). However, in several studies (Pud et al., 2008, Dodd et al., 2011, Dodd et al., 2010, Illi et al., 2012, Miaskowski et al., 2006), a significant amount of inter-individual variability was found in patients’ experiences with various multiple co-occurring symptoms. All five of these studies evaluated for subgroups of patients based on self-reports of the four most common symptoms associated with cancer and its treatment (i.e., pain, fatigue, sleep disturbance, depression). A consistent finding across all five studies was the identification of a subgroup of patients with low levels of all four symptoms and a subgroup of patients with high levels of all four symptoms. Equally important, patients in the “all high” subgroup reported worse functional status and poorer quality of life (QOL) outcomes.

While the aforementioned studies were limited to only four co-occurring symptoms, newer work has used the Memorial Symptom Assessment Scale (MSAS) (Portenoy et al., 1994) to identify high risk patients based on the occurrence of 32 symptoms (Ferreira et al., 2008, Gwede et al., 2008, Miaskowski et al., In press). In two of these studies (Ferreira et al., 2008, Gwede et al., 2008), two distinct symptom subgroups were identified. In the third study (Miaskowski et al., In press), four distinct symptom subgroups were found. Across these three studies, patients in the subgroup with the highest symptom occurrence rates reported decrements in functional status and QOL. Of note, across all eight studies cited above, clinical characteristics were not associated with subgroup membership. Variations in the total number of patient subgroups may be related to the heterogeneous nature of the samples in terms of cancer diagnoses and types of treatments, the number of symptoms evaluated, the dimension of the symptom experience (e.g., occurrence, severity) used to create the subgroups, and the statistical procedures used to identify the subgroups.

Recently, our research team reported on the use of latent class analysis (LCA) to identify three distinct groups of patients (i.e., latent classes) based on the relative occurrence rates for 25 symptoms on the MSAS prior to receiving their next cycle of CTX (Miaskowski et al., 2014a). Of the 582 patients evaluated, 36.1% were categorized in the low class (i.e., mean of 5.7 symptoms); 50.0% were in the moderate class (i.e., mean of 12.9 symptoms); and 13.9% were in the all high class (i.e., mean of 20.3 symptoms). Of note, patients in the all high class were significantly younger and more likely to be female and nonwhite, and had lower levels of social support, lower socioeconomic status, poorer functional status, and a higher level of comorbidity. As noted previously, no other clinical characteristics were associated with latent class membership. At the conclusion of this paper, we suggested that the use of LCA may provide an effective way to identify patients with a higher symptom burden or a high risk symptom phenotype.

A question that remains unanswered is whether or not this high risk symptom phenotype persists following the administration of CTX. In other words, do patients who were classified, using LCA, into the low, moderate, and high classes remain in those symptom classes following the administration of their next dose of CTX? To answer this research question, we used latent transition analysis (LTA) (Collins and Lanza, 2010, Lanza et al., 2003), to examine the probability that patients remained in the same class at their initial (i.e., prior to the administration of CTX) and subsequent (i.e., following the administration of their next dose of CTX) assessment. For the patients whose class membership remained consistent over the two time points, differences in demographic and clinical characteristics, as well as QOL outcomes were evaluated.

PATIENTS AND METHODS

Patients and Settings

This study is part of an ongoing, longitudinal study of the symptom experience of oncology outpatients receiving CTX (Miaskowski et al., 2014a). Eligible patients were ≥18 years of age; had a diagnosis of breast, gastrointestinal (GI), gynecological (GYN), or lung cancer; had received CTX within the preceding four weeks; were scheduled to receive at least two additional cycles of CTX; were able to read, write, and understand English; and gave written informed consent. Patients were recruited from two Comprehensive Cancer Centers, one Veteran’s Affairs hospital, and four community-based oncology programs. A total of 1505 patients were approached and 906 consented to participate (60.2% response rate). The major reason for refusal was being overwhelmed with their cancer treatment.

Instruments

A demographic questionnaire obtained information on age, gender, ethnicity, marital status, living arrangements, education, employment status, and income. The Karnofsky Performance Status (KPS) scale (Karnofsky, 1977) was used to evaluate patients’ functional status. The Self-administered Comorbidity Questionnaire (SCQ) (Sangha et al., 2003) evaluated the occurrence, treatment, and functional impact of common comorbid conditions (e.g., diabetes, arthritis).

The MSAS was used to evaluate the occurrence, severity, frequency, and distress of 32 symptoms commonly associated with cancer and its treatment. The MSAS is a self-report questionnaire designed to measure the multidimensional experience of symptoms. Patients were asked to indicate whether or not they had experienced each symptom in the past week (i.e., symptom occurrence). If they had experienced the symptom, they were asked to rate its frequency of occurrence, severity, and distress. The reliability and validity of the MSAS is well established in studies of oncology inpatients and outpatients (Portenoy et al., 1994).

Quality of life (QOL) was evaluated using generic (i.e., Medical Outcomes Study-Short Form-12 (SF-12)) (Ware et al., 1996) and disease-specific (i.e., Quality of Life Scale-Patient Version (QOL-PV)) (Dow et al., 1996, Ferrell et al., 1995) measures. Both measures have well-established validity and reliability. Higher scores on both measures indicate a better QOL.

Study Procedures

The study was approved by the Committee on Human Research at the University of California, San Francisco and by the Institutional Review Board at each of the study sites. Eligible patients were approached by the research staff in the infusion unit to discuss participation in the study. Written informed consent was obtained from all patients. Depending on the length of their CTX cycles, patients completed questionnaires in their homes, a total of six times over two cycles of CTX (i.e., prior to CTX administration (Time 1 and 4), approximately 1 week after CTX administration (Time 2 and 5), approximately 2 weeks after CTX administration (Time 3 and 6)). For this analysis, symptom occurrence data from the Time 1 (i.e., recovery from previous cycle) and Time 2 (i.e., acute symptoms) assessments were analyzed. Patients were asked to report on their symptom experience for the previous week. Medical records were reviewed for disease and treatment information.

Data Analysis

Latent class analysis

LCA identifies latent classes based on an observed response pattern using categorical variables (Collins and Lanza, 2010, Lanza et al., 2003, Nylund et al., 2007a, Vermunt and Magdison, 2002). Prior to identifying the LTA model that described patients’ transitions from classes at Time 1 to classes at Time 2, separate LCAs were done to identify subgroups of patients with similar symptom experiences at each of the two assessments. These analyses provided an estimate of the number of classes that might be expected at each assessment, so as to inform the LTA modeling of the class transitions.

The LCAs and LTA were done using the symptom occurrence data from the MSAS. In order to have a sufficient number of patients with each symptom to perform the LCA and LTA, we identified the MSAS symptoms that occurred in at least 40% of the patients. This criterion was selected to provide assurance that sufficient information was available to identify classes that were not sample-specific, due to infrequent reports of symptoms. A total of 25 out of 32 symptoms from the MSAS occurred in >40% of the patients. Following the identification of the number of latent classes at each assessment with LCA, the estimation of latent transition classes was performed.

The final number of latent classes for each LCA was selected based on the Bayesian Information Criterion (BIC), the Vuong, Lo, Mendel, and Rubin likelihood ratio test (VLMR), and entropy. Typically, the best fitting LCA model has the lowest BIC. This BIC criterion can be supplemented by an evaluation of the VLMR (Nylund et al., 2007b) which tests whether a model with K classes fits the data better than a model with one fewer class (the K-1 class model). If the VLMR is significant, it supports the K-class model as fitting the data better. If it is not significant, it indicates that too many classes were extracted and that the K-1 class model fits the data better than the K-class model. In addition, well-fitting models produce entropy values of ≥.80 (Celeux and Soromenho, 1996). Finally, well-fitting models “make sense” conceptually and the estimated classes differ as might be expected on variables not used in the generation of the model. Because the VLMR is not available for LTA, the best fitting model was determined based on its BIC and entropy values.

Latent Transition Analysis

LTA allows for the identification of individuals who transition from latent classes at one point in time, to the same latent classes at a subsequent point in time, as well as individuals who move to different classes (Collins and Lanza, 2010, Lanza et al., 2003, Nylund et al., 2007a). Symptom occurrence data from the MSAS was used to identify the latent classes of patients with similar symptom experiences at two time points during their CTX cycle (i.e., Time 1 – week prior to CTX administration and Time 2 – week following CTX administration).

The LCA and LTA models were estimated using Mplus™ Version 7 (Muthen and Muthen, 1998-2014). Estimation was carried out with robust Maximum-Likelihood (MLR) and the Expectation-Maximization (EM) algorithm (Muthen and Shedden, 1999). Given that the observed variables were dichotomies, estimation was carried out with a logit link. To protect against solutions that were identified based on a local maximum, from 800 to as many as 6,000 random starts were used in the estimation of the model. This approach ensured that the best fitting log-likelihood was replicated with multiple models (Muthen and Muthen, 1998-2014).

Differences in demographic and clinical characteristics and QOL outcomes among the LTA classes, were evaluated using analyses of variance, Kruskal-Wallis, or Chi-Square tests with Bonferroni corrected post hoc contrasts, using SPSS version 22 (IBM, Armonk, NY). A p-value of <.05 was considered statistically significant. All comparisons among the classes used actual values. Adjustments were not made for missing data. Therefore, the cohort for each of these analyses was dependent on the largest set of complete data among groups.

RESULTS

Latent Class Analyses

A total of 906 patients completed the MSAS at Time 1 and Time 2. A separate LCA was done for each time point. As shown in Table 1, a three-class solution fit the data best at each time point. For both time points, the BIC was lower for the 3-class compared to the 2-class solution and the VLMR was significant. While the BIC for the 4-class solution was smaller, the VLMR was not significant. These consistent results for the two time points suggested an upper limit to the number of classes that might be found over the two assessments with the LTA. The probability of occurrence of the 25 MSAS symptoms for the three class solutions for the Time 1 and Time 2 assessments are illustrated in Figures 1A and 1B, respectively.

Table 1 –

Latent Class Solutions and Fit Indices for the Time 1 and Time 2 Assessments Using Symptom Occurrence Ratings^a

Time 1 Assessment – Prior to Next Dose of Chemotherapy
Model	LL	AIC	BIC	VLMR	Entropy
2 Class	−13465.34	27032.68	27277.55	2375.10****	.84
3 Classb	−13217.76	26589.51	26959.21	495.17****	.83
4 Class	−13084.89	26375.78	26870.31	265.73ns	.80
Time 2 Assessment – Following the Next Dose of Chemotherapy
2 Class	−13286.83	26675.67	26920.53	2320.59****	.83
3 Classb	−13021.47	26196.94	26566.64	530.73****	.81
4 Class	−12892.35	25990.71	26485.24	258.23ns	.79

^nsNot significant

^*p < .05

^**p < .01

^***p < .001

^****p < .0001

^aIn order to have a sufficient number of patients with each symptom to perform the latent class analyses, the MSAS symptoms that occurred in at least 40% of the patients were identified. This criterion was selected to provide assurance that sufficient information was available to identify classes that were not sample-specific, due to infrequent reports of symptoms. A total of 25 out of 32 symptoms from the MSAS occurred in >40% of the patients.

^bThe 3-class solution was selected because the VLMR was significant for the 3-class solution, indicating that three classes fit the data better than two classes, and the VLMR was not significant for the 4-class solution, indicating that too many classes had been extracted.

Note. LL = log-likelihood; AIC = Akaike Information Criterion, BIC = Bayesian Information Criterion; VLMR = Vuong-Lo-Mendell-Rubin likelihood ratio test for the K vs. K-1 model.

Figure 1.

A – Probability of symptom occurrence for each of the latent classes for the 25 symptoms on the Memorial Symptom Assessment Scale that occurred in ≥40% of the total sample (n=760) at Time 1 (i.e., Prior to next dose of chemotherapy).

B - Probability of symptom occurrence for each of the latent classes for the 25 symptoms on the Memorial Symptom Assessment Scale that occurred in ≥40% of the total sample (n=760) at Time 2 (i.e., Following next dose of chemotherapy).

Latent Transition Analysis

First, the LTA models were fit with two classes at each of the two assessments. Then the models were fit with three classes at each of the two time points. To ensure that the LTA classes at the two times were comparable, the LTA models were fit assuming measurement invariance for latent class indicators at each time.

As shown in Table 2, the BIC was smaller for the 3-to-3-class LTA and entropy remained above .80. As shown in Table 3, an LTA solution with 3 classes at each assessment produces nine classes in the joint distribution. Clearly, an inspection of Table 3 shows that the proportion of cases in the off-diagonal classes is much smaller than for the classes on the diagonal and that five of the off-diagonal classes consist of less than 3% of the sample. Therefore, we did not estimate models beyond three classes at each assessment.

Table 2 -.

Latent Transition Solutions and Fit Indices for Two-to-Two and Three-to-Three Classes Using Symptom Occurrence Ratings^a for Time 1 to Time 2

Model	LL	AIC	BIC	Entropy
2 classes	−26645.54	53397.07	53651.95	.86
3 classesb	−26073.41	52312.82	52711.97	.85

^aIn order to have a sufficient number of patients with each symptom to perform the latent class analyses, the MSAS symptoms that occurred in at least 40% of the patients were identified. This criterion was selected to provide assurance that sufficient information was available to identify classes that were not sample-specific, due to infrequent reports of symptoms. A total of 25 out of 32 symptoms from the MSAS occurred in >40% of the patients.

^bThe 3-to-3-class solution was selected because the BIC was smaller than the 2-to-2-class solution.

Note. LL = log-likelihood; AIC = Akaike Information Criterion, BIC = Bayesian Information Criterion.

Table 3 -.

Latent Transition Class Counts and Proportions for Three-to-Three Classes Using Symptom Occurrence Ratings for Time 1 to Time 2

A. Three-Class to Three-Class Solution
Time	Class	Count	Proportion
1	1	264.73	0.292
	2	374.81	0.414
	3	266.45	0.294
2	1	251.18	0.277
	2	434.76	0.480
	3	220.06	0.243

B. Three-Class to Three-Class Pattern
T1 Class	T2 Class	Count	Proportion
1	1	224.39	0.248
1	2	35.33	0.039
1	3	5.01	0.006
2	1	22.77	0.025
2	2	318.08	0.351
2	3	33.97	0.037
3	1	4.02	0.004
3	2	81.34	0.090
3	3	181.08	0.200

C. Actual Classification of Patients Based on Their Most Likely Latent Class Pattern from Time 1 to Time 2
Class	Latent Class Pattern	Number of Patients	Percentage of Patients	Class Names
1	1 - 1	235	25.94	High-High
2	1 – 2	25	2.76	High-Moderate
3	1 - 3	3	0.33	High-Low
4	2 – 1	13	1.43	Moderate-High
5	2 – 2	335	36.98	Moderate-Moderate
6	2 – 3	27	2.98	Moderate-Low
7	3 – 1	3	0.33	Low-High
8	3 – 2	75	8.28	Low-Moderate
9	3 - 3	190	20.97	Low-Low

As shown in Table 3C, of the 906 patients who had evaluable data at Time 1 and Time 2, 760 (83.9%) were classified in the same symptom occurrence classes at both assessments. Of these 760 patients, based on the relative occurrence rates for the 25 MSAS symptoms across the latent classes, 25.0% were classified as Low-Low, 44.1% as Moderate-Moderate, and 30.9% as High-High ( Figure 2).

Figure 2.

Probability of symptom occurrence for each of the latent transition classes for the 25 symptoms on the Memorial Symptom Assessment Scale that occurred in ≥40% of the total sample (n=760).

Differences in Patient Characteristics Among the Three Consistent LTA Classes

Table 4 summarizes the differences in demographic and clinical characteristics among the three consistent LTA classes. Compared to the Low-Low class, patients in the High-High and Moderate-Moderate classes were significantly younger, more likely to be female, more likely to report having child care responsibilities, and had a lower KPS score and a higher comorbidity score. In terms of specific comorbid conditions, compared to the Low-Low class, a higher percentage of patients in the High-High class reported the occurrence of anemia, depression, and back pain. With the exception of the KPS and comorbidity scores, as well as cancer diagnosis, none of the other clinical characteristics (i.e., time since diagnosis, types and number of prior treatments, presence or number of metastatic sites) differed among the LTA classes. For cancer diagnosis, pairwise contrasts found that compared to the High-High class, a higher percentage of patients in the Low-Low class had a GI cancer. Patients in the High-High class reported the occurrence of a significantly higher number of MSAS symptoms (19.3 ± 4.2) than patients in the Moderate class (12.4 ± 3.0). Patients in the Moderate-Moderate class reported a significantly higher number of symptoms than patients in the Low-Low class (5.5 ± 2.7).

Table 4 –

Differences in Demographic and Clinical Characteristics Among the Three Latent Transition Analysis Classes (n=760)

Characteristic	Low-Low (1) n=190 25.0%	Moderate-Moderate (2) n=335 44.1%	High-High (3) n=235 30.9%	Statistics
	Mean (SD)	Mean (SD)	Mean (SD)
Age (years)	61.4 (10.6)	57.0 (11.9)	54.3 (12.1)	F=19.9, p<.0001 1 > 2 > 3
Education (years)	15.7 (3.2)	16.5 (2.8)	16.1 (3.0)	F=4.34, p=.013 1 < 2
Body mass index (kg/m2)	26.1 (5.6)	26.3 (5.8)	25.9 (5.6)	F=0.30, p=.741
Karnofsky Performance Status score	87.4 (9.7)	79.7 (11.8)	74.8 (12.0)	F=58.25, p<.0001 1 > 2 > 3
Self-administered Comorbidity Questionnaire score	4.5 (2.5)	5.5 (3.1)	6.4 (3.3)	F=19.85, p<.0001 1 < 2 < 3
Time since diagnosis (mean in years)	1.9 (3.2)	2.1 (3.7)	2.1 (4.1)	KW, p=.871
Time since diagnosis (median in years)	0.44	0.44	0.44
Number of prior cancer treatments	1.6 (1.6)	1.7 (1.5)	1.8 (1.5)	F=1.34, p=.263
Number of metastatic sites including lymph node involvementa	1.3 (1.2)	1.2 (1.2)	1.2 (1.2)	F=0.42, p=.656
Number of metastatic sites excluding lymph node involvement	0.9 (1.1)	0.8 (1.1)	0.8 (1.0)	F=0.74, p=.477
Mean number of MSAS symptoms (out of 32)	5.5 (2.7)	12.4 (3.0)	19.3 (4.2)	F=880.63, p<.0001 1 < 2 < 3
	% (n)	% (n)	% (n)
Gender (% female)	68.9 (131)	78.2 (262)	89.8 (211)	X2=31.96, p<.0001 1 < 2 < 3
Self-reported ethnicity
White	71.0 (130)	72.9 (240)	68.4 (158)	X2=1.37, p=.505
Non-white	29.0 (53)	27.1 (89)	31.6 (73)
Married or partnered (% yes)	70.1 (131)	67.1 (222)	62.7 (146)	X2=2.64, p=.266
Lives alone (% yes)	17.0 (32)	19.9 (66)	23.9 (56)	X2=3.16, p=.206
Currently employed (% yes)	37.2 (70)	37.4 (125)	28.6 (67)	X2=5.43, p=.066
Annual household income				KW=9.47, p=.230
Less than $30,000	16.7 (27)	15.9 (49)	22.0 (46)
$30,000 to $70,000	24.1 (39)	22.7 (70)	18.7 (39)
$70,000 to $100,000	17.9 (29)	13.6 (42)	19.6 (41)
Greater than $100,000	41.4 (67)	47.9 (148)	39.7 (83)
Child care responsibilities (% yes)	15.7 (29)	24.0 (79)	29.1 (67)	X2=10.40, p=.006 1 < 2 and 3
Elder care responsibilities (% yes)	6.9 (12)	9.9 (30)	7.0 (15)	X2=1.95, p=.378
Common comorbidities (% yes)
Heart disease	4.7 (9)	5.7 (19)	4.3 (10)	X2=0.62, p=.733
High blood pressure	34.7 (66)	31.9 (107)	27.2 (64)	X2=2.92, p=.233
Lung disease	12.6 (24)	11.0 (37)	11.9 (28)	X2=0.31, p=.857
Diabetes	8.4 (16)	9.3 (31)	7.7 (18)	X2=0.45, p=.797
Ulcer or stomach disease	2.6 (5)	4.2 (14)	6.4 (15)	X2=3.58, p=.167
Kidney disease	0.0 (0)	0.9 (3)	1.7 (4)	X2=3.34, p=.188
Liver disease	7.4 (14)	3.3 (11)	7.7 (18)	X2=6.34, p=.042 No significant pw contrasts
Anemia	6.3 (12)	13.1 (44)	18.3 (43)	X2=13.32, p=.001 1<3
Depression	5.8 (11)	14.9 (50)	39.1 (92)	X2=82.81, p<.0001 1<2<3
Osteoarthritis	10.5 (20)	9.6 (32)	16.2 (38)	X2=6.214, p=.045 No significant pw contrasts
Back pain	15.8 (30)	24.8 (83)	37.4 (88)	X2=26.19, p<.0001 1 and 2 <3
Rheumatoid arthritis	3.2 (6)	3.9 (13)	3.4 (8)	X2=0.21, p=.902
Cancer diagnosis				X2=23.36, p=.001
Breast cancer	34.7 (66)	41.5 (139	46.0 (108	1>3
Gastrointestinal cancer	35.8 (68)	27.5 (92)	21.3 (50)
Gynecological cancer	14.7 (28)	16.7 (56)	24.7 (58)
Lung cancer	14.7 (28)	14.3 (48)	8.1 (19)
Prior cancer treatment				X2=11.25, p=.081
No prior treatment	27.6 (51)	23.4 (77)	15.9 (37)
Only surgery, CTX, or RT	35.1 (65)	41.6 (137)	45.9 (107)
Surgery and CTX, or surgery and RT, or CTX and RT	23.8 (44)	20.1 (66)	21.5 (50)
Surgery and CTX and RT	13.5 (25)	14.9 (49)	26.7 (39)
Metastatic sites				X2=6.33, p=.387
No metastasis	29.1 (55)	33.3 (111)	36.1 (84)
Only lymph node metastasis	20.1 (38)	23.1 (77)	21.0 (49)
Only metastatic disease in other sites	28.6 (54)	23.7 (79)	19.7 (46)
Metastatic disease in lymph nodes and other sites	22.2 (42)	19.8 (66)	23.2 (54)

^aTotal number of metastatic sites evaluated was 9.

Abbreviations: CTX = chemotherapy, kg = kilograms, KW = Kruskal Wallis, m² = meters squared, pw= pair-wise, RT = radiation therapy, SD = standard deviation

Differences in Quality of Life Scores Among the Latent Classes

As shown in Table 5, except for the spiritual well-being subscale, post hoc contrasts revealed that patients in the High-High class reported significantly lower scores on the QOL-PV subscale and total scores than patients in the Moderate-Moderate class. Patients in the Moderate-Moderate class reported significantly lower QOL-PV scores than patients in the Low-Low class.

Table 5 –

Differences in Disease Specific and Generic Quality of Life Scores Among the Three Latent Transition Analysis Classes (n=760)

Characteristic	Low-Low (1) n=190 25.0%	Moderate- Moderate (2) n=335 44.1%	High-High (3) n=235 30.9%	Statistics
	Mean (SD)	Mean (SD)	Mean (SD)
MULTIDIMENSIONAL QUALITY OF LIFE SCALE
Physical well-being	8.1 (1.3)	6.6 (1.6)	5.5 (1.7)	F=153.26, p < .001 1 > 2 > 3
Psychological well-being	6.9 (1.5)	5.7 (1.7)	4.1 (1.5)	F=170.35 p < .001 1 > 2 > 3
Social well-being	7.1 (1.7)	5.8 (1.8)	4.4 (1.7)	F=119.97, p<.0001 1 > 2 > 3
Spiritual well-being	5.4 (2.2)	5.3 (2.1)	5.2 (2.1)	F=0.58, p = .562
Total QOL score	6.9 (1.2)	5.8 (1.3)	4.6 (1.2)	F=183.88 p < .001 1 > 2 > 3
SF12 SCORES
Physical functioning	65.6 (34.0)	48.0 (32.8)	42.3 (32.5)	F=25.89, p < .001 1 > 2 and 3
Role physical	70.6 (26.2)	51.1 (27.6)	40.5 (26.8)	F=63.54, p < .001 1 > 2 > 3
Bodily pain	91.3 (17.9)	76.4 (25.9)	64.1 (29.2)	F=58.64, p<.0001 1 > 2 > 3
General health	73.1 (22.6)	64.1 (27.3)	51.9 (28.8)	F=32.58, p < .001 1 > 2 > 3
Vitality	62.9 (22.6)	42.0 (24.5)	35.0 (24.8)	F=72.61, p < .001 1 > 2 > 3
Social functioning	85.9 (21.8)	67.1 (29.5)	52.1 (30.3)	F=73.36, p<.0001 1 > 2 > 3
Role emotional	87.7 (22.0)	78.2 (25.5)	62.8 (27.8)	F=50.89, p < .001 1 > 2 > 3
Mental health	83.9 (16.9)	74.6 (18.2)	57.7 (20.2)	F=108.83, p < .001 1 > 2 > 3
Physical Component Summary score	46.2 (9.2)	40.0 (10.3)	38.3 (10.0)	F=33.07, p<.0001 1 > 2 and 3
Mental Component Summary score	55.3 (8.6)	50.2 (9.4)	42.5 (9.8)	F=92.45, p<.0001 1 > 2 > 3

Abbreviations: SD = standard deviation

In terms of the SF-12 subscale and physical component summary (PCS) and mental component summary (MCS) scores, except for the physical functioning and PCS scores, post hoc contrasts revealed the same pattern of between group differences in QOL scores (i.e., High-High < Moderate-Moderate < Low-Low). For the physical functioning and PCS scores, the pattern of between class differences was Moderate-Moderate class and High-High class < Low-Low class.

DISCUSSION

To our knowledge, this study is the first to use LTA to evaluate for a stable symptom phenotype based on patients’ experiences with 25 common symptoms before and following a dose of CTX. Consistent with our previous report that used LCA to evaluate 582 patients in the current sample prior to their next dose of CTX (Miaskowski et al., 2014a), three distinct subgroups of patients with consistent symptom experiences were identified by LTA. In addition, the mean number of symptoms reported by each latent class is relatively consistent across both studies. While in the previous study, as well as in the Time 1 LCA done with this larger sample (Figure 1A), approximately 14% of the patients were categorized in the All High class, in the current LTA study, using both the pretreatment and post-treatment assessments, 30.9% of the patients were categorized in the High-High class (Figure 2). This finding suggests that two or more assessments may be warranted to categorize those patients who are in a “stable” lower or higher risk symptom phenotype. In addition, it should be noted that 75% of the patients in this study experienced moderate to high occurrence rates for 25 common symptoms from prior to through the first week following their dose of CTX.

As shown in Table 2, for 83.9% of the patients (n=760), their most likely latent class pattern remained the same from Time 1 to Time 2. In terms of patients who transitioned from a lower to a higher symptom class, 9.7% went up one class and 0.3% went up two classes. In terms of patients whose transition pattern was the opposite, 5.7% went down one class and 0.3% went down two classes. Given the small sample sizes for each of these groups, one cannot readily evaluate the exact reasons for these different patterns of transition.

As shown in Table 6, the six symptoms that were among the top eleven in occurrence rates across the three LTA classes were lack of energy (0.512 to 0.976), difficulty sleeping (0.401 to 0.946), pain (0.341 to 0.855), hair loss (0.338 to 0.733), feeling drowsy (0.255 to 0.877), and nausea (0.184 to 0.727). Of note, lack of energy, difficulty sleeping, anxiety, and pain are symptoms commonly associated with cytokine-induced sickness behavior (Illi et al., 2012, Wang et al., 2010, Myers, 2008, Dantzer et al., 2008). However, while depression is often reported to be a symptom associated with sickness behavior (Walker et al., 2014, Harrison et al., 2009, Dantzer, 2006), it (i.e., feeling sad on the MSAS) was only found in the top eleven occurring symptoms in patients in the High-High class. In fact, in the High-High class, worrying (0.946), feeling irritable (0.895), and feeling nervous (0.758) had very high occurrence rates. None of the four psychological symptoms were found in the top 11 occurring symptoms in the other two LTA classes. This finding suggests that in addition to interventions to treat physical symptoms, patients in the High-High class require more in depth mental health evaluation and more proactive and aggressive management of their psychological symptoms. This approach is warranted given the substantial body of evidence that has documented the negative long-term sequelae of ongoing and high levels of psychological distress in cancer patients (Stanton et al., 2015).

Table 6 –

Probability of Occurrence for the 25 MSAS Symptoms For Each of the Three Latent Transition Class in Descending Order of Occurrence

Rank Order	High-High	P	Moderate-Moderate	P	Low-Low	P
1	Lack of energy	.976	Lack of energy	.966	Lack of energy	.512
2	Worrying	.946	Difficulty sleeping	.735	Difficulty sleeping	.401
3	Difficulty sleeping	.903	Feeling drowsy	.690	Pain	.341
4	Feeling sad	.895	Pain	.676	Hair loss	.338
5	Difficulty concentrating	.892	Nausea	.575	Numbness and tingling in hands/feet	.331
6	Feeling drowsy	.887	Difficulty concentrating	.563	Changes in the way food tastes	.284
7	Feeling irritable	.859	Numbness and tingling in hands/feet	.556	Feeling drowsy	.255
8	Pain	.855	Changes in the way food tastes	.535	Constipation	.186
9	Feeling nervous	.758	Lack of appetite	.490	Cough	.185
10	Hair loss	.733	Hair loss	.487	Nausea	.184
11	Nausea	.727	Dry mouth	.482	Dry mouth	.172
12	I don’t look like myself	.721	Constipation	.454	Hot flashes	.166
13	Numbness and tingling in hands/feet	.712	Worrying	.391	Difficulty concentrating	.163
14	Changes in the way food tastes	.695	Feeling sad	.376	I don’t look like myself	.162
15	Lack of appetite	.675	Dizziness	.368	Worrying	.159
16	Constipation	.642	Feeling irritable	.350	Changes in skin	.146
17	Changes in skin	.622	Changes in skin	.339	Diarrhea	.136
18	Dry mouth	.612	I don’t look like myself	.324	Lack of appetite	.124
19	Feeling bloated	.598	Diarrhea	.315	Feeling irritable	.113
20	Sweats	.552	Cough	.304	Feeling sad	.111
21	Dizziness	.529	Hot flashes	.288	Sweats	.110
22	Hot flashes	.509	Feeling bloated	.270	Problems with sexual interest	.109
23	Problems with sexual interest	.501	Sweats	.269	Feeling bloated	.099
24	Cough	.452	Problems with sexual interest	.251	Dizziness	.086
25	Diarrhea	.405	Feeling nervous	.226	Feeling nervous	.077

Consistent with our previous report in the same sample (Miaskowski et al., 2014a), as well as reports by others (Gwede et al., 2008, Ferreira et al., 2008), KPS and SCQ scores were associated with LTA class membership. While associations between a higher symptom burden and a higher level of comorbidity, as well as poorer functional status, are reported consistently in oncology patients (Pud et al., 2008, Dodd et al., 2011, Dodd et al., 2010, Miaskowski et al., 2006, Ferreira et al., 2008, Gwede et al., 2008), additional research is warranted to further explicate these relationships. For example, the most common comorbid conditions in this sample were high blood pressure (31.2%), back pain (26.4%), and depression (20.1%). Many of the chronic conditions listed in Table 4 are associated with both acute and chronic symptoms. Therefore, future studies need to assess the impact of the symptoms associated with cancer and its treatment, as well as the symptoms associated with other chronic conditions, on latent class membership. In addition, future longitudinal studies need to evaluate, using statistical procedures like parallel process growth modeling (Cheong et al., 2003, Rose et al., 2009), whether increases in symptom burden are associated with decreases in functional status or vice versa. Similar approaches could be used to evaluate for changes in patients’ symptom burden in relationship to changes in their comorbidity profiles.

While the majority of the characteristics associated with cancer and its treatment did not predict LTA class membership, compared to the High-High class, a relatively higher percentage of patients with GI cancer were in the Low-Low class. The exact reasons for this difference are not readily apparent and warrant investigation in future studies.

Compared to the Low-Low class, patients in the Moderate-Moderate and High-High classes were almost a half or a whole decade younger, respectively. While the association between younger age and higher symptom burden is reported in previous studies (Illi et al., 2012, Miaskowski et al., 2014a, Cataldo et al., 2013, Ritchie et al., 2014), the underlying physiologic and psychological mechanisms for this association remain to be determined. However, because recent evidence suggests that an overlap exists between molecular mechanisms that govern both aging and cancer (Coppede, 2013, Kong et al., 2013, Menck and Munford, 2014, Teschendorff et al., 2013), patients with cancer may experience “premature biological aging” that is associated with a higher symptom burden. Alternatively, “chronologically” older patients may receive lower doses of CTX (Townsley et al., 2005, Kumar et al., 2007) or have a “response shift” in their perception of symptoms (Sprangers and Schwartz, 1999, Schwartz and Sprangers, 1999).

While female gender, years of education, and child care responsibilities were associated with a higher symptom burden in the current study, findings regarding these characteristics are inconsistent across studies (Dodd et al., 2011, Illi et al., 2012, Miaskowski et al., 2014b, Baldwin et al., 2010, Cheung et al., 2011, Miaskowski, 2004). In addition, while other studies found that being a member of an ethnic minority and reporting a lower socioeconomic status were associated with more severe symptoms (Miaskowski et al., 2014a), these associations were not found in the current study. Additional research is warranted to confirm or refute these inconstant findings.

For both the generic (SF-12) and disease-specific (MQOLS-PV) measures of QOL, as symptom burden increased, QOL decreased. The decrements in QOL among the three latent classes represent not only statistically significant, but clinically meaningful decreases in QOL with effect sizes ranging from 0.44 to 1.54 (Sloan et al., 2003, Osoba, 1999). Taken together and consistent with previous reports (Illi et al., 2012, Miaskowski et al., 2014a, Ferreira et al., 2008, Gwede et al., 2008), these findings provide evidence of the significant negative impact that multiple co-occurring symptoms have on patients’ ability to function and other QOL outcomes.

Several study limitations need to be acknowledged. Because patients were recruited at different points in their CTX treatment, symptom occurrence rates prior to the initiation of CTX are not available. In addition, the CTX drugs used varied based on the patients’ diagnoses and stages of disease. While we cannot rule out the potential contributions of these clinical characteristics to the patients’ symptom experiences, the relatively similar percentages of cancer diagnoses, evidence of metastatic disease, time since cancer diagnosis, and types of previous treatments suggest that the three LTA classes were relatively similar in terms of disease and treatment characteristics. While it is possible that patients in the Low-Low class were receiving more aggressive symptom management interventions, the occurrence rates for fatigue, sleep disturbance, and pain were high across the three LTA classes.

In the era of precision medicine (National Research Council, 2011) and big data (Yoo et al., 2014), coupled with the use of electronic medical records and smart phone technology (e.g. the ASyMS/eSMART© system that is being evaluated as part of a grant from the European Commission (Kearney et al., 2009, Maguire et al., 2015)), it is conceivable that symptom data will be collected in “real time” from oncology patients receiving CTX. The use of analytic approaches like LTA, or the development of more sophisticated algorithms using techniques like machine learning (Bastanlar and Ozuysal, 2014, Yoo et al., 2014), will allow clinicians to analyze patients’ phenotypic and molecular data on an ongoing basis. The integration of these types of information across multiple patients will assist clinicians to identify patients at highest risk for the most severe symptom profiles and to pre-emptively or more aggressively treat their most common and severe symptoms. This type of risk profiling and aggressive symptom management should reduce oncology patients’ symptom burden and improve their QOL.