A longitudinal study of pain variability and its correlates in ambulatory patients with advanced stage cancer

Authors


Abstract

BACKGROUND:

Although pain is common among patients with advanced cancer, little is known about longitudinal variability in pain intensity. For this report, the authors examined variability in pain intensity over 24 months among ambulatory patients with advanced stage cancers, associations between patient characteristics and within-patient pain variability, and the relation of pain variability to overall survival.

METHODS:

The sample comprised 949 patients with solid tumors who received care and reported pain scores in at least 3 different months within 24 months of their initial stage IV diagnosis during the period from 2004 to 2006. Pain intensity was measured using a scale ranging from 0 (no pain) to 10 (worst pain). Pain variability was defined as the standard deviation of the maximum monthly pain scores and was dichotomized at the 50th percentile into high and low variability groups.

RESULTS:

Considerable between-patient differences in pain variability (range, 0-5.77) were observed. Nonwhites, patients with a stage IV cancer diagnosed within the previous 3 months, and those with moderate or severe pain at baseline were more likely to experience high pain variability. Although patients with head and neck cancer generally had the highest pain variability, the intensity of their pain typically decreased over the disease course. High pain variability with worsening pain trajectory was associated with increased risk of death.

CONCLUSIONS:

Longitudinally, pain intensity was highly variable among patients with stage IV cancer. Minority patients, newly diagnosed patients, patients with head and neck cancer, and patients with moderate or severe pain at baseline were at higher risk of large fluctuations in pain intensity. Cancer 2012. © 2012 American Cancer Society.

INTRODUCTION

Pain is a common and feared symptom reported by patients with advanced cancer.1 More than 70% of patients with advanced stage cancer reported some pain.2-4 Cancer-related pain is a subjective, multidimensional experience caused by the tumor as well as its treatment.5-7 Consequently, pain intensity may wax and wane throughout the disease and treatment course.8 An understanding of the fluctuating nature of pain, as well as the variables associated with variability in pain intensity among cancer patients, could help patients and providers to anticipate their experience and to target interventions toward those at high risk.

Most pain studies, however, have used cross-sectional designs to examine the prevalence and severity of pain among oncology patients.1, 9 These studies did not take into account the fluctuating nature of cancer-related pain: measuring pain at 1 time point could mask important variability. Although some studies attempted to explore cancer patients' longitudinal experience of pain, those studies selected a limited number of time points to calculate temporal changes in pain severity, pain relief, or incidence.8, 10-12 There has been little empiric work dedicated to characterizing pain variability across cancer types, over an extended period of time, or among those with advanced disease.

To address this knowledge gap, we designed a study to explore longitudinal pain variability in a sample of ambulatory patients with stage IV cancers over 2 years. We sought to examine variability in pain intensity over time, factors associated with pain variability, and the association of pain variability with overall survival.

MATERIALS AND METHODS

Data Sources

Pain intensity score data were drawn from electronic medical records of all office encounters and treatment visits of patients who attended the Dana-Farber Cancer Institute (DFCI) between January 1, 2004 and December 31, 2006. Pain scores were collected routinely with vital signs at the beginning of each patient's appointment. Clinic assistants asked patients to rate their pain intensity from 0 (no pain) to 10 (worst pain), using a 0 to 10 numerical rating scale and a standard script. The pain data along with vital signs were then entered into and maintained in the electronic medical record and were captured for 96% of visits to the cancer center. The information was readily available to clinicians during clinic and treatment visits. The scale is a widely used and well validated measure.13

Pain intensity information was combined with data extracted from 2 electronic databases—the DFCI and Brigham and Women's Hospital Tumor Registries and the DFCI patient registration system. The former provides information on cancer International Classification of Disease (ICD) codes, date of diagnosis, and disease staging; whereas the latter provides patient demographics, such as date of birth, date of death, sex, race, primary language, and insurance type. The study protocol was approved in advance by the Dana-Farber Harvard Cancer Center institutional review board.

Study Population

The study population comprised 2003 adult patients diagnosed with stage IV solid tumors who were evaluated and treated at DFCI from 2004 through 2006. Patients were identified using a combination of the ICD for Oncology, third edition (ICD-O-3) codes14 and the American Joint Committee on Cancer (AJCC) TNM staging system.15 To track fluctuations of pain intensity, we excluded 759 patients who reported pain scores in only 1 or 2 distinct months within 24 months of their initial stage IV cancer diagnoses and 295 patients whose reported pain scores were all 0. The final analytic cohort included 949 patients.

Study Measures

We selected a single pain intensity score per month. To avoid the potential problem of multiple measures internally correlated with a pain exacerbation, we used the maximum pain intensity score from a given month for patients with multiple pain scores in that month. The mean pain intensity score was calculated for each patient as the average of the maximum monthly pain scores. Pain variability was defined as the standard deviation of the maximum monthly pain scores over time for each patient. Because there is no extant, validated measure of pain variability, we used this empirically designed approach to assess the fluctuation of pain.16-18 Both mean pain intensity score and pain variability were calculated using the continuous pain scale from 0 to 10.

Variability was then dichotomized as high (standard deviation, ≥2.12) and low (<2.12) by using the 50th percentile as the cutoff point. High variability may occur when there is an improvement or worsening of pain intensity with time. To differentiate between patients whose pain variability was related to rising or falling pain intensity scores, we performed a bivariable linear regression analysis for each patient of pain intensity (dependent variable) and time (independent variable) to estimate a pain trajectory. We used the slope of this model (negative or positive) as an indicator of the change in pain intensity (decreasing or increasing) over time for each patient in the high variability group. We defined a negative slope as an improvement in pain intensity and a positive slope as worsening pain intensity.

The demographic variables of interest were age, sex, race, primary language, and primary insurance. Information on race mainly was extracted from the patient registration system, which was collected by direct inquiry of patients, and was supplemented with information from tumor registries, which were based primarily on patients' medical records. Clinical variables were cancer type, interval between the date of the initial stage IV diagnosis and the date of the first DFCI visit (ie, time since stage IV diagnosis), and pain intensity score at the first DFCI visit after the diagnosis of stage IV cancer (ie, pain intensity at initial cancer center visit). We classified pain intensity scores at initial cancer center visit into the following categories: none (score, 0), mild (score, 1-3), moderate (score, 4-6), and severe (score, 7-10), according to the National Comprehensive Cancer Network practice guideline.13 We classified cancer diagnosis as breast, thoracic, gastrointestinal (including liver and intrahepatic bile duct, pancreas, esophagus, stomach, colon/rectum), urogenital (ovary, uterus, cervix, prostate, testis, kidney, bladder), head and neck (oral cavity, pharynx, larynx, thyroid), and other (brain and other nervous system, melanoma, and sarcoma) based on ICD-O-3 codes. The date of the stage IV diagnosis referred to the date of the cancer diagnosis first reported by the clinician, regardless of whether the diagnosis was pathologically confirmed.

Statistical Analysis

We compared demographics and clinical characteristics of patients with low and high pain variability using the chi-square statistic. In addition, a multivariable logistic regression model was used to identify patient demographics and clinical attributes associated with high pain variability. Stepwise backward elimination was used, and covariates with P values < .10 were retained in the final model. Among the subgroup of patients who had high variability with either increasing or decreasing pain intensity over time, the chi-square statistic and the multivariable logistic analysis were repeated to examine the risk factors for increasing pain intensity.

To compare pain variability across cancer diagnoses, we presented the 25th percentile, median, and 75th percentile of variability for all patients and for each cancer type using a box-and-whisker plot. In addition, we displayed the overall mean pain intensity and pain variability for each cancer type in a scatter plot chart with pain intensity plotted on the horizontal axis and pain variability plotted on the vertical axis.

Finally, using the log-rank test, we distinguished overall survival among patients with low pain variability, patients who had high variability with decreasing pain intensity, and patients who had high variability with increasing pain intensity; and we graphed Kaplan-Meier curves for a subgroup of 294 patients who had at least 12 months of pain scores. The 3 groups were defined based on the variability and trajectory observed from the first 12-monthly maximum pain scores. Overall survival was defined as the number of months from the 12th monthly pain score to the date of death reported in the DFCI patient registration system, with deaths captured through December 31, 2008. Patients who remained alive at the end of follow-up were censored at that point. P values were 2-sided, and values < .05 were considered statistically significant. SAS version 9.2 was used for all analyses (SAS Institute Inc., Cary, NC).

RESULTS

Patient Characteristics

Patient demographics and clinical characteristics are presented in Table 1. The average age of patients at their first DFCI visit during the study period was 59 years (range, 19-89 years). Slightly more than half of the study participants were men, greater than 90% were white, and approximately 95% spoke English as their primary language. The majority of patients had either Medicare (29%) or private insurance (65%).

Table 1. Characteristics of 949 Patients With Stage IV Cancer (2004-2006 Diagnosis)
CharacteristicNo. of Patients%a
  • a

    Percentages may not sum to 100% because of rounding.

  • b

    Other cancers include brain and other nervous system cancers, melanoma, and sarcoma.

Age, y  
 18-4922723.9
 50-6445748.2
 ≥6526527.9
Sex  
 Men52054.8
 Women42945.2
Race  
 White86290.8
 Nonwhite879.2
Primary language  
 English88793.5
 Other languages515.4
 Missing111.2
Insurance  
 Medicaid/free-care/self-insurance626.5
 Medicare27128.6
 Private insurance61464.7
 Missing20.2
Cancer type  
 Breast646.7
 Thoracic28029.5
 Gastrointestinal28930.5
 Urogenital828.6
 Head and neck16717.6
 Other cancersb677.1
Time since stage IV diagnosis, mo  
 ≤150052.7
 2-325827.2
 4-1215115.9
 13-24404.2
Pain intensity at initial cancer center visit  
 None (0)47149.6
 Mild (1-3)20922.0
 Moderate (4-6)17818.8
 Severe (7-10)919.6

The most common cancer types in our cohort were gastrointestinal cancer (31%) and thoracic cancer (30%), followed by head and neck cancer (18%). More than half of the patients had a newly diagnosed stage IV cancer; that is, they had their first visit to DFCI within 1 month of their stage IV diagnosis. Approximately half of patients reported no pain, whereas approximately 10% reported severe pain at their initial DFCI visit. The number of months with reported pain scores ranged from 3 months to 24 months.

Characteristics of Pain Variability and its Relation With Intensity Across Cancer Diagnoses

Four hundred eighty-two patients (51%) had high pain variability. The overall median pain variability value was 2.12 (range, 0-5.77) (Fig. 1). Across diagnoses, patients with breast cancer had the most consistent levels of reported pain, with a median pain variability of 1.88, and those with head and neck cancer had the greatest median variability of 2.40. Patients with breast cancer had the smallest between-patient difference in pain variability (range, 0.38-4.28), while those with gastrointestinal cancer had the greatest difference in variability across patients (range, 0-5.77).

Figure 1.

This box-and-whisker plot depicts variations in pain variability for all patients and for each cancer diagnosis. Pain variability was defined as the standard deviation of the maximum monthly pain scores over time. Other cancers include brain and other nervous system cancers, melanoma, and sarcoma.

Figure 2 illustrates the mean pain intensity and within-patient pain variability for each of the 6 major cancer types. On average, patients with head and neck cancer had the highest pain variability but ranked second in mean pain intensity, whereas patients with thoracic cancer had the highest mean pain intensity but ranked second in pain variability. In contrast, patients with breast cancer had the lowest mean pain intensity and pain variability.

Figure 2.

This chart compares pain intensity versus pain variability by cancer diagnosis. The horizontal axis represents pain intensity, which was calculated as the average of the maximum monthly pain scores with values from 1.5 to 2.5 on a scale from 0 to 10. The vertical axis represents pain variability, which was defined as the standard deviation of the maximum monthly pain scores over time. Other cancers include brain and other nervous system cancers, melanoma, and sarcoma. GI indicates gastrointestinal.

Factors Associated With Pain Variability

In the multivariable analysis, nonwhites were more likely than whites to have high pain variability (odds ratio [OR], 2.15; 95% confidence interval [CI], 1.32-3.51) (Table 2). The odds of having high pain variability were greatest for patients who had head and neck cancer (OR, 2.06; 95% CI, 1.12-3.79 compared with patients who had breast cancer). There was a significant association between time since stage IV diagnosis and high pain variability, with lower odds for patients whose first visit was between 4 months and 12 months (OR, 0.64; 95% CI, 0.44-0.95) compared with newly diagnosed patients. Finally, patients who had moderate pain (OR, 1.77; 95% CI, 1.24-2.53) or severe pain (OR, 7.70; 95% CI, 4.05-14.60) at baseline were far more likely to have high pain variability compared with those who reported no pain at baseline.

Table 2. Correlates of High Pain Variability Among Patients With Stage IV Cancer
 Low Variability, n = 467High Variability, n = 482High vs Low Variability: OR [95% CI]
CharacteristicNo.%aNo.%aUnadjusted ModelAdjusted Model
  • Abbreviations: CI, confidence interval; OR, odds ratio; Ref, reference category.

  • a

    Percentages may not sum to 100% because of rounding.

  • b

    This variable was dropped because of nonsignificance in the backward selection model.

  • c

    Other cancers include brain and other nervous system cancers, melanoma, and sarcoma.

Age, y     b
 18-4911249.311550.71.0 [Ref] 
 50-6423551.422248.60.92 [0.67-1.27] 
 ≥6512045.314554.71.18 [0.83-1.68] 
Sex     b
 Men26050.026050.01.0 [Ref] 
 Women20748.322251.81.07 [0.83-1.39] 
Race      
 White43850.842449.21.0 [Ref]1.0 [Ref]
 Nonwhite2933.35866.72.07 [1.30-3.29]2.15 [1.32-3.51]
Primary language     b
 English44850.543949.51.0 [Ref] 
 Other languages1733.33466.72.04 [1.12-3.71] 
 Missing218.2981.8  
Insurance     b
 Medicaid/free-care/self-insurance2641.93658.11.0 [Ref] 
 Medicare12546.114653.90.80 [0.46-1.39] 
 Private insurance31651.529848.50.65 [0.38-1.09] 
 Missing00.02100.0  
Cancer type      
 Breast3656.32843.81.0 [Ref]1.0 [Ref]
 Thoracic13247.114852.91.48 [0.86-2.55]1.35 [0.76-2.40]
 Gastrointestinal15352.913647.11.18 [0.68-2.02]1.14 [0.65-2.02]
 Urogenital4352.43947.61.20 [0.62-2.31]1.25 [0.63-2.47]
 Head and neck6538.910261.12.07 [1.16-3.71]2.06 [1.12-3.79]
 Other cancersc3856.72943.31.04 [0.52-2.07]1.00 [0.49-2.06]
Time since stage IV diagnosis, mo      
 ≤123446.826653.21.0 [Ref]1.0 [Ref]
 2-312146.913753.11.00 [0.74-1.35]0.95 [0.69-1.31]
 4-128657.06543.10.67 [0.46-0.96]0.64 [0.44-0.95]
 13-242665.01435.00.47 [0.24-0.93]0.50 [0.24-1.02]
Pain intensity at initial cancer center visit      
 None (0)25854.821345.21.0 [Ref]1.0 [Ref]
 Mild (1-3)12559.88440.20.81 [0.59-1.13]0.80 [0.57-1.12]
 Moderate (4-6)7240.510659.61.78 [1.26-2.53]1.77 [1.24-2.53]
 Severe (7-10)1213.27986.87.97 [4.23-15.03]7.70 [4.05-14.60]

Among the 482 patients who had high pain variability, 212 had a negative slope, 267 had a positive slope, and 3 had a zero slope. The analysis of the subset of 479 patients with nonzero slopes indicated that 71.3% of patients with head and neck cancer had experienced decreasing pain intensity over time, whereas only 40.7% of patients with gastrointestinal cancer and approximately 33% of patients with other types of cancer had their pain intensity reduced over time (Fig. 3). The multivariable analysis also suggested that those who had head and neck cancer were most likely to experience a reduction in pain intensity over time compared with those who had other diagnoses.

Figure 3.

The percentage of patients with high pain variability who experienced improvement in pain intensity over time is illustrated for all patients and for each cancer diagnosis. The denominator includes patients who had high pain variability with either increasing or decreasing pain intensity over time. Other cancers include brain and other nervous system cancers, melanoma, and sarcoma.

Overall Survival

Kaplan-Meier survival curves are provided in Figure 4. We calculated separate curves for patients with low pain variability, those with high variability and a trajectory of improving pain intensity, and those with high variability and a trajectory of worsening pain intensity. The unadjusted 1-year survival probabilities were 39.1% (95% CI, 27.7%-50.4%) for patients with increasing pain intensity compared with 59.2% (95% CI, 50.8%-66.6%) for those with low variability, and 79.5% (95% CI, 68.7%-86.9%) for those with high variability but decreasing intensity. In other words, the combination of high variability and increasing pain intensity was associated with the poorest survival.

Figure 4.

These are Kaplan-Meier survival curves stratified by level of pain variability and trajectory. This analysis was conducted in a subgroup of 294 patients who had at least 12 months of pain scores. Months referred to the interval from the 12th monthly maximum pain score observation to the date of death or the end of the observation period.

DISCUSSION

In this longitudinal study of 949 ambulatory patients with advanced stage cancer, significant between-patient differences in pain variability were observed. Nonwhites, patients with a stage IV cancer diagnosed within the previous 3 months, and those with moderate or severe pain at their initial visit were more likely to experience high pain variability. Although patients with head and neck cancer generally had the highest pain variability, their pain intensity typically decreased over the disease course. Worsening pain over time was associated with an increased risk of death compared with either low pain variability or decreasing pain over time.

Pain intensity is expected to fluctuate over time and to vary depending on the type and stage of cancer, the type of treatment regimen, the quality and timeliness of symptom management, time since diagnosis, psychological functioning, other stressful life events, and physical activity.5, 19, 20 Patients' pain reports typically would reflect the dynamic nature of their disease, treatment, and circumstances. Longitudinal pain variability in clinical practice, however, is a phenomenon that is not well studied, especially among patients with advanced cancer. The absence of research studies in this area may be attributed in part to the lack of suitable data sets. Pain scores routinely solicited and documented in the electronic medical record of patients at each clinic visit in our institute provided a useful resource. Our data highlight the volatile nature of pain among many patients with advanced cancer and underscore the need for careful, ongoing assessment and management.

We observed that nonwhites were more likely than whites to experience high pain variability. Previous studies that addressed the role of racial/ethnic disparities have consistently reported that nonwhites had a higher prevalence and more severe pain and were at greater risk for inadequate pain assessment and treatment than non-Hispanic whites.21-29 Physicians often underestimated the pain severity of nonwhite patients, leading to unintentional variability in decisions about pain management.25-28, 30 In addition, racial/ethnic minorities may experience barriers to accessing pain-management programs and services.25, 28, 31 Furthermore, minorities face language and cultural barriers that affect their ability to communicate effectively with providers, resulting in increased anxiety, pain, and distress and decreased adherence to treatment.28, 32-35 These factors also may explain racial differences in longitudinal pain variability.

Our results did not indicate that age or sex was associated with longitudinal pain variability. Previous research explored age and sex differences in prevalence, severity, or under-treatment of cancer pain yet produced mixed results.9 For example, Cleeland and colleagues reported that women outpatients and those aged ≥70 years were at greater risk for inadequate pain management evaluated by a pain management index they had developed.29 The relation between age and inadequacy of cancer pain treatment was confirmed by Bernabei et al, although they were not able to identify a positive association between women and inadequate pain treatment.36 Nevertheless, a more recent study indicated that there were no age or sex differences in adequacy of pain treatment measured by the pain management index.37 Poor management of cancer pain may lead to large fluctuations in pain intensity over the disease course. How age or sex was associated with pain intensity and variability and the adequacy of pain treatment requires further examination.

Several studies examined and compared the prevalence and severity of pain among different cancer types.1, 38-40 In a 2007 systematic review,1 patients with head and neck cancer had the highest prevalence of pain (70%) compared with the overall prevalence of pain in all cancer types (50%). Our results provided new evidence that these patients also experienced significant pain variability. In addition, more than 7 in 10 patients who had head and neck cancer with high pain variability experienced reduced pain over time; in contrast, less than 4 in 10 patients who had other cancer types with high pain variability had a favorable pain trajectory. This result is consistent with 1 report, which indicated that approximately 50% of patients with head and neck cancer had pain that was reduced through curative treatment,11 and with another study, which documented a significant improvement in pain between diagnosis and 3-year follow-up in patients with head and neck cancer.41

Previous studies have demonstrated that the presence of pain and high pain intensity are important predictors of clinical outcomes in cancer patients.42-48 Our study extends those findings to indicate that high pain variability with worsening pain intensity represents an adverse prognostic factor that may indicate disease progression, such as increased tumor size or more aggressive tumor characteristics. In contrast, improvement in pain intensity over time may suggest proper treatment of pain. Patients who had their pain adequately managed were more likely to maintain activities of daily living (eg, eating, moving); less likely to experience physical symptoms such as nausea, fatigue, and insomnia; and less likely to have emotional distress (eg, depression, anger, frustration, and hopelessness),5, 10, 22 all of which are related to improved survival.45, 49 Scharpf et al emphasized that changes in pain intensity across time were a much better indicator of the association between pain and outcomes than pain intensity at a given point in time; however, relatively few patients had enough data points to support a robust longitudinal analysis.48 The finding that pain is viewed as a “fifth vital sign” in our institute provided an opportunity to study the critical relation between pain and survival. Pain variability also may be related to nonadherence to supportive self-care measures. Many factors, such as education and literacy, poverty, and family presence, can affect self-care and pain outcomes in palliative care, as indicated in the review by Hughes.50 Carefully monitoring pain fluctuations over the disease course may inform treatment selection and palliative care interventions.

The current study has several limitations. First, the measure used to examine pain variability was derived empirically, because there is no existing, validated approach to assess pain fluctuations. We believe this approach was reasonable for this exploratory study. Second, variability is affected by pain intensity at the top and bottom of the scale. This ceiling effect (scores above 10 are not permitted) or floor effect (scores below 0 are not permitted) reduces pain variability for those extremes. Nevertheless, there was very little ceiling effect in our study, because only approximately 0.5% of pain scores were 9 or 10. Third, patients' pain intensity reports may be influenced by their expectations about how the information will be used. For our current study, we relied on routinely collected clinical information in the electronic medical record to ascertain pain intensity. We do not know whether patients may have reported differently had they perceived that this information would be used for research or quality improvement. We have no reason to suspect biased reports in this context. Fourth, the data were collected from a single comprehensive cancer center. Therefore, it would be important to evaluate the generalizability of our findings to the general population of patients with advanced cancer. Fifth, although we obtained complete race information, data on ethnicity were available for approximately 35% of patients, and only 5 nonwhite patients identified themselves as Hispanic. Therefore, we grouped our patients as whites and nonwhites based on the race variable. The white cohort may include both Hispanic and non-Hispanic whites. The nonwhite cohort included only 87 patients. Because race and ethnicity may affect pain intensity and quality of pain care among cancer patients, efforts need to be made to elicit complete ethnicity information. Sixth, our data did not include information about cancer treatment (eg, surgery, radiation therapy, or chemotherapy), use of pain medications, comorbidities, or physical activity, which may influence pain and outcomes. We could not determine whether the volatility of patients' reported pain intensity was attributable to disease progression, cancer treatment, or effectiveness of pain medications and adjunctive therapies. The timing of interventions relative to pain exacerbations and remissions was beyond the scope of the current study, but this is a critical future area of inquiry. Finally, we used maximum monthly pain scores to calculate pain variability. It is possible that other ways of measuring pain variability would produce somewhat different results. Moreover, high pain variability, to some extent, may reflect inadequacy of pain treatment; but pain variability calculated from maximum monthly pain intensity may integrate additional factors, such as concurrent therapy, bony metastases, and physical activity, that decrease the potential association between these 2 variables. In future work, it would be valuable to explore alternative ways to construct longitudinal pain variability; examine its relation with adequacy of pain management, such as the pain management index;29 and test their predictive values for survival or other important clinical outcomes.

In conclusion, pain intensity scores varied greatly among patients with stage IV cancer over time. Nonwhites, patients with a stage IV cancer diagnosed within 3 months, and patients with moderate or severe pain at baseline were at greater risk of fluctuating pain intensity. Although patients with head and neck cancer generally had the highest pain variability across cancer diagnoses, most of these patients experienced reduced pain over time. High pain variability with worsening pain trajectory was associated with an increased risk of death among patients with advanced cancer.

FUNDING SOURCES

This research was supported by a grant (PEP-08-273-01-PC1) from the American Cancer Society.

CONFLICT OF INTEREST DISCLOSURES

The authors made no disclosures.

Ancillary