Automated Pain Intervention for Underserved Minority Women with Breast Cancer

Abstract

BACKGROUND

Minority patients with breast cancer are at risk for undertreatment of cancer-related pain. We evaluated the feasibility and efficacy of an automated pain intervention for improving pain and symptom management of underserved African American and Latina women with breast cancer.

METHODS

Sixty low-income African American and Latina women with breast cancer and cancer-related pain were enrolled in a pilot study of an automated, telephone-based interactive voice response (IVR) intervention. The intervention group patients were called twice per week by the IVR system and asked to rate the intensity of their pain and other symptoms. The patients’ oncologists received e-mail alerts if reported symptoms were moderate to severe. The patients also reported barriers to pain management and received education regarding any reported obstacles.

RESULTS

The proportion of women in both groups reporting moderate to severe pain decreased during the study, but the decrease was significantly greater for the intervention group. The IVR intervention was also associated with improvements in other cancer-related symptoms, including sleep disturbance and drowsiness. Although patient adherence to the IVR call schedule was good, the oncologists treating the patients rated the intervention as only somewhat useful for improving symptom management.

CONCLUSIONS

The IVR intervention reduced pain and symptom severity for underserved minority women with breast cancer. Additional research on technological approaches to symptom management is needed.

INTRODUCTION

Inadequate treatment of pain among minority patients with cancer has been documented in multiple studies.^1,2 When minority patients are underserved due to limited financial resources, they are particularly at risk. Although most minority populations have a lower incidence of breast cancer than non-Hispanic white populations, minority patients are more likely to be diagnosed with advanced disease and may be at particular risk for pain and related symptoms.³ Although the results of descriptive studies suggest that underserved minority patients might benefit from education about cancer pain, the results of a randomized clinical trial found that education alone did not improve pain management.⁴ We concluded that an intervention targeting multiple patient, provider, and health care system barriers is needed to optimize pain management.

Multicomponent interventions, however, are difficult to implement in understaffed public hospitals. Although traditional telephone communication is not feasible for repeated assessment, combining telephones with computerized assessment may be an effective way to follow patients’ pain. Interactive voice response (IVR) and other web-based systems have been used effectively to monitor symptoms associated with chemotherapy^5–14 and stem cell transplantation.^15,16 No previous studies, however, have evaluated an IVR system specifically designed for underserved minority women with breast cancer. In the present study, we evaluated an automated, telephone-based IVR system that measures patients’ pain and symptom levels, alerts providers when a symptom is moderate to severe, and assesses barriers to pain control. Our specific aims were: (1) to pilot test the efficacy of the IVR intervention for improving pain and pain-related symptoms of minority patients with breast cancer, as compared with the current standard of care, and (2) to evaluate the feasibility of the IVR intervention. We hypothesized that minority patients in the intervention group would demonstrate reduced severity of pain and related symptoms.

PATIENTS AND METHODS

Setting and Patients

The patients were recruited in the outpatient medical oncology clinic of a large public hospital in Houston, Texas that treats underserved patients. Patients treated at this hospital are required to meet income-based eligibility criteria based on federal poverty-level guidelines. The patients were approached by bilingual female research staff in private clinic rooms. Eligibility criteria included: (1) outpatient in the oncology clinic, (2) breast cancer diagnosis confirmed by pathology, (3) self-report of being Black/African American or Latina/Hispanic, (4) English or Spanish-speaking, (5) 18 years of age or older, (6) chronic cancer-related pain, and (7) a “pain worst” score ≥4 on a 0–10 scale. The patient’s oncologist confirmed that the pain was related to cancer or cancer treatment¹⁷ but did not participate in the recruitment process. If an eligible patient agreed to participate, the staff obtained written informed consent and scheduled the baseline assessment, to be conducted in English or Spanish depending on patient preference. The patients were stratified by minority group and randomly assigned to the intervention or control group by an electronic protocol management system. The Institutional Review Boards of The University of Texas MD Anderson Cancer Center and the Harris Health System approved the study.

IVR Intervention

The IVR system was demonstrated at baseline to patients in the intervention group who practiced using the system. The research staff explained that the system would call 2 times per week for 8 weeks at preferred times. The system asked patients to identify themselves using a study number, and then the IVR symptom script began. Patients reported their responses using the touch-tone keypad. If a patient did not have telephone access during the study, then a cell telephone was provided. If a patient did not answer an IVR call, the system repeated the call up to 3 times on the same day, spaced 45 minutes apart. If not answered, the system notified the research staff who contacted the patient by telephone and verbally administered the symptom items.¹⁸

The IVR intervention consisted of: (1) assessment of patients’ pain and related symptoms; (2) determination of pain or other symptoms that exceeded a severity threshold; (3) feedback of information about suprathreshold symptoms to the physician; and (4) assessment of patient-related barriers. The barriers assessed included (1) nonadherence to analgesic medications, (2) difficulty obtaining analgesic medications, (3) side effects from medications, (4) concerns about opioids, such as fear of addiction, (5) reliance on alternative strategies (e.g., herbs) for pain management, and (6) lack of family support for pain management. When a patient reported a barrier, a member of the research staff contacted the patient by telephone and provided educational information regarding the barrier and how to overcome it.⁴ A standardized English/Spanish script was used, followed by the opportunity for questions.

Pain

When the patient’s reported pain level was 5 or greater on the 0–10 scale, the IVR system immediately forwarded this information by e-mail to the patient’s physician. The patients in the clinic are treated by oncology fellows who are supervised by attending oncologists. The e-mail alerts were sent to the patient’s oncology fellow, with a copy to the attending oncologist. The research staff monitored the IVR alerts on a web-based intranet site and sent an e-mail to the physician who received the alert that contained a form for documentation of response.

Other Symptoms

The IVR assessed 12 additional symptoms contained in the M. D. Anderson Symptom Inventory (MDASI).¹⁹ In addition to pain, 6 symptoms (nausea, vomiting, emotional distress, sadness, drowsiness, shortness of breath) were identified by the oncologists as symptoms that should generate e-mail alerts. The symptoms nausea, vomiting, distress, sadness, and drowsiness triggered an alert when the level was ≥5. For shortness of breath, the alert threshold was designated as ≥3. No specific thresholds were set for fatigue, sleep disturbance, lack of appetite, dry mouth, difficulty remembering, and numbness or tingling. The physicians were provided with the intervention group’s most recent IVR symptom ratings prior to the patients’ clinic visits.

Control Group

Patients in the control group received the pain and symptom management interventions usually given by the clinic physicians. The patients completed the paper-and-pencil assessments at baseline and the 2 follow-up assessments. They were informed that their symptom data would be used for research purposes only and would not be provided to their clinicians. All patients were instructed to report any symptoms to their physicians.

Outcomes Assessment

The outcome measures were administered to all patients in the study during clinic visits at time point 1 (4–6 weeks after enrollment in the study) and time point 2 (8–10 weeks after enrollment).

Instruments

M. D. Anderson Symptom Inventory

The MDASI contains 13 symptom items that are rated on 0–10 numeric scales. The reliability and validity of the MDASI have been demonstrated.^19,20 The Spanish language version of the MDASI has been validated.²¹

Barriers Questionnaire-II

The Barriers Questionnaire-II (BQ-II) was developed to measure patient beliefs that are barriers to optimal pain treatment. The BQ-II has demonstrated excellent reliability and validity.^22–24 The Spanish language version of the questionnaire has been validated.²⁵

Other Measures

Performance Status

The ECOG Performance Status Scale was used to assess the physician’s estimate of the patient’s functional status at baseline.²⁶ The ECOG scale is an observational measure of functional ability that has demonstrated excellent reliability and validity.^27,28

Biomedical Variables

Biomedical information was recorded from the patients’ electronic medical records.

Pain Management Index

The analgesic data were used to compute a Pain Management Index (PMI) at each assessment.²⁹ Pain management is considered adequate when there is congruence between the patient’s pain severity and the appropriateness of the prescribed analgesic. Negative PMI scores indicate under medication, and scores ≥0 are considered to be a conservative indicator of acceptable treatment.

Patient Demographic Form

Patients completed a demographic questionnaire, including questions about age, race, and ethnicity.

Feasibility Measures

The feasibility of the intervention was evaluated by determining rates of recruitment and retention, patient adherence to the call schedule, rates of physician receipt and response to the alerts, and physician evaluation of the IVR system.

Statistical Analysis

For this pilot study, we planned to enroll a total of 60 patients. Our goals were to pilot the intervention, obtain an estimate of its effect, and collect feasibility measures. Our power analysis indicated that we could detect a standardized difference of 0.74 in mean pain severity ratings between the control and intervention group using a 2-tailed test at an alpha level of .05 and 0.80 power.

We used descriptive statistics to describe how patients rated pain and symptom severity. Proportions of patients rating their symptoms as moderate to severe (≥5 on a 0–10 scale) were reported.³⁰ Potential differences between patient groups based on demographic variables were explored using a chi-square test or t test. If the normality assumption was not met for t test, the Wilcoxon test was used. Nominal P values were reported.

To evaluate the effectiveness of the IVR intervention, linear mixed models were used to evaluate changes in mean pain and symptom severity across time. A group-by-time interaction term was included in each model. In addition, a random intercept was included to account for baseline variations in pain and symptom severity. Performance status and the PMI were entered as covariates to control for possible effects of function and pain management. The PMI was not included in the model for pain severity. Generalized estimating equation analysis was used to determine the likelihood of observing higher number of symptom threshold events in the intervention group. To address our hypothesis that the severity of pain would be reduced in the intervention group, we used McNemar’s test to compare the proportions of patients with moderate to severe pain across time. In addition, we calculated the PMI for each patient and performed a chi-square test to compare the proportions of patients whose pain was inadequately managed in the IVR and control groups. To address our hypothesis that the severity of pain-related symptoms such as sleep disturbance and emotional distress would be reduced in the intervention group, we used linear mixed models to compare pain-related symptoms between the IVR and control groups. McNemar’s test was used to compare the proportions of patients with moderate to severe symptoms across time.

Secondary Analyses

We used t test or its nonparametric counterpart in comparing BQ-II scores between the groups at baseline and time point 1, and between baseline and time point 2.

The primary endpoint for evaluation of feasibility was recruitment and retention rates; secondary variables were patient adherence to the symptom assessment, frequency of alerts and responses to alerts, and physician evaluation of the intervention.

RESULTS

Patient Characteristics

Of the 449 patients with breast cancer screened for eligibility, 73 patients were eligible and approached regarding participation in the study (Fig. 1). Sixty patients agreed to be enrolled. Reasons for refusal included lack of time (n = 2), reluctance to complete questionnaires (n = 2), financial concerns (n = 1), lack of interest (n = 1), and unknown (n = 7). Eighty-four percent of the intervention group and 72% of the control group patients completed the first follow-up assessment, and 68% of the intervention and 86% of the control group completed the second follow-up assessment.

Figure 1.

The CONSORT flow diagram illustrates patient flow throughout the study.

Demographic and clinical characteristics of the 60 participants are listed by group assignment in Table 1. The mean time since cancer diagnosis was 2.0 + 3.3 years, with no significant differences between groups. No significant baseline differences were found between the intervention and control groups or between African American and Latina women, younger vs. older women or by disease stage. Most of the patients were receiving active treatments, but 16 patients (27%) had recently finished treatment or were waiting for treatment to begin. Among the Latina patients, 74% described Spanish as their primary language.

Table 1.

Baseline Demographic and Disease Variables Among African American and Latina Women with Breast Cancer

Characteristic	Intervention Group (n = 31)	Control Group (n = 29)	Pa
Years of age, mean (SD)	49.6 (9.9)	50.5 (11.0 )	.76
Years of education, mean (SD)	10.6 (4.1)	10.0 (2.9)	.51
Ethnic group			.97
African American	13 (42%)	12 (41%)
Latina	18 (58%)	17 (59%)
Marital status			.78
Married	15 (48%)	13 (45%)
Unmarried	16 (52%)	16 (55%)
Employment status			.31
Unemployed	16 (52%)	15 (52%)
Employed	2 (6%)	4 (14%)
Homemaker	7 (23%)	8 (28%)
Retired	4 (13%)	2 (7%)
Other	2 (6%)	0 (0%)
Disease stage			.91
II A–B	7 (23%)	7 (24%)
III A–B	11 (35%)	12 (41%)
IV	11 (35%)	10 (34%)
Unknown	2 (6%)	0 (0%)
Disease status			.91
No evidence of disease	4 (13%)	4 (14%)
Local/regional	14 (45%)	12 (41%)
Metastatic	11 (35%)	12 (41%)
Unknown	2 (6%)	1 (3%)
Treatment status			.44
Chemotherapy	14 (45%)	17 (59%)
Radiotherapy	1 (3%)	0 (0%)
Hormone therapy	5 (16%)	3 (10%)
Immunotherapy	2 (6%)	2 (7%)
Noneb	9 (29%)	7(24%)
Patients with severe painc	21 (68%)	23 (79%)	.83
Good performance statusd	14 (45%)	15 (45%)	.82

Abbreviation: SD, standard deviation.

^aP value for comparisons between intervention and control groups.

^bNo chemotherapy, radiotherapy, immunotherapy, or hormone therapy.

^cSevere pain intensity is defined as a MDASI “pain worst” score in the severe range (7–10).

^dGood performance status is defined as a score of 0–1 on the 5-point Eastern Cooperative Oncology Group scale.

Symptom Alerts

During the study, 71% of the IVR assessments were completed successfully. A total of 221 symptom alerts were generated for patients in the intervention group for pain (n = 164), drowsiness (n = 140), emotional distress (n = 109), shortness of breath (n = 106), sadness (n = 92), nausea (n = 64), and vomiting (n = 28). During the study, 28 of the 31 patients in the intervention group reported symptoms severe enough to generate an alert, and 100% of the symptom threshold events were detected. E-mail alerts were sent to the physicians, who acknowledged receipt of 161 (73%) via e-mail response to the research staff. The physicians’ most frequent responses to the alerts were to use current symptom treatments (46%), see the patient at next scheduled appointment (33%), or prescribe a new symptom treatment (8%).

Primary Analyses

Differences in Pain Severity

Linear mixed models for mean pain scores on the MDASI revealed a significant time effect (P < .001) and a significant group-by-time interaction (P < .05), with both groups reporting decreased pain severity from baseline to the time point 2 assessment (Fig. 2). There were significant differences between groups in their pain change scores from baseline to time point 1 (0.6 vs 2.3, P = .034, 95% confidence interval (CI), 0.13, 3.3). A similar significant difference was observed between the groups from baseline to time point 2 (1.2 vs 3.5, P = .015, 95% CI, 0.47, 4.2). For both comparisons, larger pain reductions were observed in the intervention group. For patients in the intervention group, the proportion of women reporting moderate to severe pain (≥5) decreased significantly from baseline (86%) to the time point 2 assessment (43%; P = .004). Although there was a decrease in the proportion of patients reporting moderate to severe pain in the control condition from baseline (80%) to the time point 2 assessment (56%), this change was not statistically significant (P = .07). The 43% difference (from 86% to 43%) in the proportion reporting moderate to severe pain in the intervention group was significantly greater than the 24% difference (from 80% to 56%) in the control group (P = .04).

Figure 2.

Mean pain severity over time for the intervention and control groups is depicted.

Differences in Mean Symptom Severity

Controlling for baseline sleep scores, the mean scores for sleep disturbance decreased significantly across time for the intervention but not the control group (P < .01) and were significantly lower for the intervention as compared to the control group (P < .01) at the time point 1 and time point 2 assessments. Fig. 3 shows that mean scores across time for fatigue were not significantly lower for the intervention group than for the control group (P = .07), and there was no significant group-by-time interaction (p = 0.06). Mixed models for sadness found no significant differences between the two groups. Analysis of the patients’ distress scores revealed a significant time effect (P < .03) but no significant differences between groups (P = .06).

Figure 3.

Mean severity for disturbed sleep, fatigue, sadness, and emotional distress over time for the intervention and control groups is depicted.

Probability of Symptom Threshold Events by Group across Time

Generalized estimating equation analysis showed that there were no significant differences in the odds ratios (ORs) of observing lower symptom threshold events for the intervention group compared with the control group at baseline (95% confidence interval (CI), 0.52–2.8), at the time point 1 assessment (95% CI, 0.32–2.5), and at the time point 2 assessment (95% CI, 0.14–1.2). Compared with the baseline assessment, the odds of observing a lower number of threshold events for the intervention group was twice as likely at time point 1 assessment (OR, 2.1; P < .009) and about 6 times as likely at the time point 2 assessment (OR, 5.8) (P < .001). The ORs comparing baseline against time point 1 and time point 2 assessments for the control group were not significant.

Proportion of Moderate to Severe Symptoms across Time

Differences within groups

The proportion of women in the intervention group reporting moderate to severe distress decreased significantly from baseline (57%) to the time point 2 assessment (19%) (P = .008). In contrast, the proportion of women in the control condition reporting moderate to severe distress did not change significantly from baseline (40%) to the time point 2 assessment (40%). For patients in the intervention group, the proportion reporting moderate to severe sadness also decreased significantly from baseline (52%) to the time point 2 assessment (19%) (P = .04). In the control group, the proportion of women reporting moderate to severe sadness decreased from baseline (56%) to the time point 2 assessment (36%), but the change was not significant (P = .27). The proportion of women in the intervention group reporting moderate to severe drowsiness decreased significantly from baseline (65%) to the time point 2 assessment (30%) (P = .04). Although the proportion of women in the control group reporting moderate or severe drowsiness decreased from baseline (52%) to the time point 2 assessment (36%), this change was not significant (P = .29).

Differences between groups

However, while there were significant differences in individual symptoms within groups over time, there were no significant differences between groups over time in the proportion of moderate to severe symptoms.

For the symptoms nausea, vomiting, and shortness of breath, there were no significant changes across time in symptom severity within or between groups or in the proportions of patients in either group reporting moderate to severe symptoms.

Pain Management Index

Table 2 shows that 33% of the intervention group and 28% of the control group received adequate analgesics at baseline. At the time point 1 and time point 2 assessments, some improvement in the PMI was noted, with no significant differences between the 2 groups.

Table 2.

Percentages of Patients in IVR and Control Groups With Negative PMI

		Negative PMIa
Group	Baseline	Time point 1 (4–6 weeks)	Time point 2 (8–10 weeks)
IVR	67%	56%	37%
Control	72%	54%	44%

Abbreviations: IVR, interactive voice response system (intervention); PMI, pain management index.

^aThe PMI is calculated by subtracting the intensity of the patient’s pain from the strength of the analgesic medication. A negative PMI indicates undermanagement of the patient’s pain.

Secondary Analyses

IVR Evaluation

Only one third of the physicians reported that the symptom feedback influenced their clinical decisions. The symptom information was rated as somewhat useful, with a mean rating of 5.0 ± 2.5 on the 0 (“not at all useful”) to 10 (“extremely useful”) scale. The physicians also were asked if they thought the alerts should be incorporated into patient care. The mean response on the 0 (“definitely should not be incorporated”) to 10 (“definitely should be incorporated”) scale was 5.2 ± 3.2.

Barriers

The BQ-II scores did not change significantly during the study, and there were no significant differences between the groups. The total BQ-II score for the intervention group decreased from baseline (10.45) to the time point 2 assessment (8.41), but this decrease was not significant (P = .11).

DISCUSSION

This study evaluated the efficacy of an IVR intervention for improving pain and other cancer-related symptoms in underserved African American and Latina patients with breast cancer. The number of women in both groups reporting moderate to severe pain decreased during the study, but the decrease was statistically significant only in the intervention group. Although both groups reported decreases in pain severity over time, the decreases in the intervention group from baseline to the follow-up assessments were significantly greater than the decreases in the control group. The decrease in reported pain severity for the control group may have been associated with the generalized effects of conducting a pain study in the clinic. The improvements in the PMI during the study support this possibility.

In addition to improving pain intensity, the IVR intervention was associated with improvements in other cancer-related symptoms. The patients in the intervention group demonstrated a significant decrease in symptom threshold events during the study. When examining individual symptoms, the women in the intervention group demonstrated significant decreases across time in moderate to severe distress, sadness, and drowsiness, whereas the women in the control group did not. The feedback to the physicians may have increased their awareness of symptoms not previously reported by patients, and thus may have improved symptom management. It is also possible that the improvement in pain severity in the intervention group may have contributed to reduced sadness and distress.

The IVR intervention was also associated with improvements in sleep disturbance, a symptom that did not trigger alerts. At the end of the study, the patients in the intervention but not the control group reported significantly less sleep disturbance than at baseline. It should be noted that the physicians did receive a summary of the intervention group patients’ most recent IVR assessment prior to clinic visits. In addition, the decreases in the patients’ pain severity may have contributed to the improvements in sleep disturbance.

Although the intervention group patients’ scores on the BQ-II decreased during the study, this improvement was not statistically significant. A more intensive educational intervention may be needed to change patient beliefs and attitudes that are barriers to pain management.

Although patients previously have rated the IVR system highly,^7,18 the physicians in the present study did not, and only 8% of alerts led to changes in symptom treatment. Feedback from the physicians indicated that they typically received the alerts when they were away from the hospital, and contacting the patient was often a challenge. They expressed a preference for receiving symptom feedback when meeting with patients in the clinic. Further adaptations of the IVR intervention are needed to examine strategies for improving provider–patient communication. In addition, other forms of technology such as Web-based or smart phone applications may prove more acceptable to providers.

Our study had several limitations. It was designed as a pilot study with a small sample size and short duration. Another issue is the multiple components included in the intervention. The efficacy of the intervention may be related to regular symptom assessment, alerts, and/or other variables such as improved communication with providers. For example, reporting symptoms may have been reassuring to patients and may have encouraged discussion of symptoms with their oncologists. Given that underserved minority women with cancer are at risk for inadequate pain and symptom management, further research is required to evaluate which components are most effective for improving pain and symptom outcomes.

In conclusion, the IVR intervention was successful in reducing pain and symptom severity for underserved minority women with breast cancer. The feasibility of the automated system also was supported, and additional research on this technological approach to symptom management is warranted.