The iCanCare study is a population-based mailed survey of women with early-stage breast cancer. In partnership with the Los Angeles County and Georgia Surveillance Epidemiology and End-Results (SEER) programs, the iCanCare study identified 3,880 women of ages 20 through 79 years who were diagnosed with early-stage breast cancer determined by a definitive breast surgery date between July 1, 2013 and December 31, 2014. Women were sent surveys about 2 months after surgery and completed the survey on average about 7 months after diagnosis. To enable meaningful analyses across racial and ethnic groups, African Americans and Latinas were oversampled in Los Angeles County. The following women were excluded from the iCanCare study sampling protocol: stage III or IV cancer (as the overall project was focused on early-stage patients), Paget’s disease, or tumors > 5cm in size. In Los Angeles County, non-Hispanic whites and African Americans aged <50 were excluded due to a competing study in these populations.

The study was approved by the Institutional Review Boards of the University of Michigan and partnering institutions. Informed by Dillman’s methods,¹² we solicited participation with a $20 cash incentive. Study coordinators in respective geographic areas continued to follow up with non-responders, including up to 9 attempted phone calls and 2 repeated mailings. Participants received survey materials to their home address with a statement that their answers would not be shared individually with their providers. Study materials were printed in English; women with Spanish surnames received Spanish and English materials.¹³

Of 3,880 originally-identified women, 249 were ineligible. From these 3,631 women, 1,053 women were not reached or did not return questionnaires, resulting in an overall response rate of 71% (n=2,578). After excluding 694 women with DCIS or bilateral disease our analytic sample included 1,884 observations in the observed data and 1,945 observations after multiple imputation. SEER registries linked surveys to standardized tumor registry data.

Except where indicated, measures were collected from patient questionnaires. The primary outcome was treatment-associated toxicities. Informed by the PRO-CTCAE working group¹⁴ and our pilot work,¹⁵ participants rated the severity of seven toxicities – at their worst during cancer treatment – using a 5-point Likert scale (0 = none, 1 = mild, 2 = moderate, 3 = severe, 4 = very severe). The toxicities measured were nausea/vomiting, diarrhea, constipation, pain, arm edema, dyspnea, and breast skin irritation. These toxicities were selected after interviews with survivors and analysis of toxicity reports in pilot work.¹⁵

As few studies have investigated patient-reported, treatment-associated toxicities, we measured toxicities in three ways. First, we examined the range of severity ratings across toxicities. Next, we constructed a scale by multiplying the number of toxicities reported by severity. For example, a score of 3 might reflect one toxicity rated as severe or three toxicities rated as mild, and a score of 28 would reflect that a patient reported all seven toxicities as very severe.

To examine toxicity burden, we examined physical health and healthcare service use. To measure physical health, we used the 4-item physical function subscale of the Patient Reported Outcome Measurement Information System (PROMIS) global health measure. The scale is a brief, valid, reliable, precise, and clinically interpretable measure of physical health.¹⁶ Respondents rated each item on a 5-point ordinal scale. The score was standardized and normalized according to the scoring manual; scores below 40 reflected poor physical health.¹⁷ To measure healthcare service use, we asked patients whether they 1) did not seek help, 2) called/emailed their provider, 3) discussed at a routine visit, 4) discussed at an unscheduled visit, or 5) visited the emergency department/hospital. We classified unscheduled care as either an unscheduled clinic visit, emergency department visit, or inpatient hospitalization for toxicity management.

Patients reported their age, race/ethnicity (white, black, Latina, Asian), education (high school or less, some college, college graduate or more), and prior comorbidity diagnosis– chronic lung disease, heart disease, diabetes, or stroke – (no diagnosis, one condition, two or more conditions). We included four separate variables to capture treatment factors: primary breast surgery (lumpectomy, unilateral mastectomy, bilateral mastectomy), radiation therapy (yes/no), systemic chemotherapy (yes/no), and receipt of both radiation and chemotherapy (yes/no). SEER registries provided tumor information: stage (I or II), grade (1, 2, or 3), and lymph node status (N0 or N1). We calculated the difference between the date of patient survey completion and the cancer diagnosis date.

First, we used descriptive statistics to examine patient, disease, and treatment factors in our analytic sample and then examined these factors in the subset of women who rated at least one toxicity as severe/very severe, as well as in the subset of women who reported unscheduled care for toxicity management. Next, for each of the seven toxicities and corresponding severity rating, we calculated the proportion of women who also reported healthcare service use (phone call, scheduled visit, unscheduled visit, emergency department visit/hospitalization) for that toxicity. Using the multiplied scale of number of toxicities reported by their severity, we next plotted the corresponding PROMIS physical function scores. Using multivariable regression, we examined two dependent variables – unscheduled care and PROMIS physical function scores - by toxicity score, controlling for patient, tumor, and treatment factors. Finally, we used multivariable ordinal logistic regression with design weights reflecting the probability of selection and non-response to examine the relationship of patient, tumor, and treatment factors to higher levels of toxicity severity.

Unless specified, analyses controlled for geography (Los Angeles County and Georgia) and were weighted to account for differing probabilities of sample selection and non-response.¹⁸ We identified small amounts of missing data (range of 0-3.9% across variables, 93% of observations had complete data). To minimize biased estimates from missing data, we applied a sequential regression multiple imputation framework.¹⁹ We generated five independently-imputed data sets and computed inferential statistics that combined analyses across datasets.²⁰ Imputation results were indistinguishable from the complete case analysis. Table 1 is based on complete case analysis (N identified in table for each variable) and all subsequent figures and regression results are based on multiply-imputed data (N=1,945).

Women with early-stage invasive breast cancer reported a number of toxicities during treatment, many of which were rated as severe or very severe. 132 patients (7%) reported that none of the seven toxicities occurred during treatment. 1,810 women (93%) reported at least one toxicity and 866 of the women in the analytic sample (45%) rated at least one toxicity as severe/very severe. Among the seven toxicities, pain was most frequently reported as severe/very severe (23%), followed by constipation (14%), and breast skin irritation (13%).

Figure 1 shows patient reports of healthcare service use by each toxicity studied and the corresponding severity rating. Across all seven toxicities, 2-4% of patients did not endorse a toxicity rating, but discussed the problem during a routine office visit. Most patients sought help during an office visit (range between 22% and 77% across the seven toxicities); telephone calls/emails and emergency department visits/hospitalizations were less frequently reported. For women who experienced at least one toxicity, 9% sought care through a previously unscheduled clinic visit and 5% visited an emergency department or hospital.

Nausea/vomiting and diarrhea were frequent sources of telephone calls/emails; 29% of patients with very severe nausea/vomiting and 27% of patients with very severe diarrhea called or emailed their provider. Severe arm edema (77%) and very severe skin irritation (71%) were the primary reasons for unscheduled clinic visits. Patients with severe/very severe dyspnea most frequently visited emergency departments or hospitals for toxicity management (28%), followed by patients with severe/very severe arm edema (27%), severe/very severe diarrhea (18%), and severe/very severe pain (18%).

The mean (SD) physical functioning score on the PROMIS measure was 14.5 (3), reflecting substantial deficits from the optimal score of 50. Figure 2 shows the relationship between the multiplied toxicity rating (number of toxicities and toxicity severity rating) and PROMIS physical scores estimated by a regression model, with corresponding 95% confidence intervals. Higher PROMIS scores reflect better physical functioning and higher toxicity scores reflect more frequent and/or severe toxicity ratings These scores were averaged across age, comorbid conditions, chemotherapy receipt, employment, marital status, and race/ethnicity. PROMIS-physical functioning scores correlated linearly, negatively, and significantly with toxicity ratings (β = −0.2, 95% CI −0.3 - −0.2). Patients without toxicity had the highest scores, whereas patients who reported all seven toxicities as severe reported scores at the lowest possible score of 10 on the scale

Figures 3A-C show the unadjusted differences in toxicity reporting by breast cancer treatment. Toxicity severity varied by chemotherapy receipt (Figure 3A). For example, 29% of chemotherapy recipients reported severe/very severe pain, compared with 19% of women who did not receive chemotherapy. Severe/very severe constipation was reported by 24% of chemotherapy recipients, compared with rates of 9% for women who did not receive chemotherapy. Radiation therapy recipients reported more severe/very severe skin irritation than women who did not receive radiation (22% versus 7%), but did not differ on other toxicities (Figure 3B).

Toxicity severity varied by surgical treatment (Figure 3C). For five of seven toxicities studied, women who received bilateral mastectomy were more likely to report more severe/severe toxicities (nausea/vomiting, diarrhea, constipation, pain, and shortness of breath). More bilateral mastectomy recipients (37%) reported severe/very severe pain than those receiving unilateral mastectomy (25%) or lumpectomy (18%).

Figure 4 shows the results of a multivariable logistic regression model, which shows significant associations between the category of toxicity, plus patient and treatment factors associated with the severity toxicity. We also included a variable to reflect patient receipt of both chemotherapy and radiation therapy. Three toxicities were more frequently associated with more severe ratings; pain (OR 4.7, 95% CI 4.2-5.3), skin irritation (OR 2.1, 95% CI 1.8-2.5), and constipation (OR 1.5, 95% CI 1.4-1.7). Women who received systemic adjuvant chemotherapy were more likely to report more severe toxicity (OR 2.0, 95% CI 1.7-2.4). Patients who received both chemotherapy and radiation therapy had an additional 30% higher odds of more severe toxicity (OR 1.3, 95% CI 1.0-1.7) over those receiving only chemotherapy. Patients who had bilateral mastectomy were more likely to report higher toxicity (OR 1.2, 95% CI 1.0-1.4) than unilateral mastectomy recipients, but the difference did not reach statistical significance.

Older patients were significantly less likely to report higher toxicity (OR 0.8, 95% CI 0.7-0.8). Patients with more comorbidities were more likely to report higher toxicity (OR 1.4 95% CI 1.3-1.5 for the first comorbidity). Latinas were more likely than white women to report higher toxicity (OR 1.3, 95% CI 1.1-1.5). Compared with college graduates, women with some college education were more likely to report higher toxicity (OR 1.2, 95% CI 1.0-1.3).