Participant focus groups
Face validity of the PETS treatment burden domains in cancer survivors was assessed using qualitative methods. Focus groups of survivors of women’s cancers who were managing both cancer and a chronic health condition were recruited from multidisciplinary oncology practices at two large healthcare systems to participate in telephone-based group discussion format. Between 4 to 6 patients were recruited per group to call in on a toll-free line at a mutually convenient time. A total of 4 focus groups were held; the groups were led by a trained moderator. Moderated group discussions focused on self-management activities for cancer and concurrent chronic health conditions and were guided by the item domains in the PETS [6]. Participants were asked to consider and describe their major self-management needs and activities (e.g., medication taking, monitoring health, medical appointments, finding health information, exercise, diet, and use of medical devices); difficulties in self-management; and the range of impact of self-management on well-being. Conditions that may influence treatment burden were discussed including financial strain, previous struggles with self-management, and health literacy. Transcripts and meeting notes were prepared and reviewed by the study team and were used to create a brief set of self-management items and to set priorities for inclusion of PETS items and scales in the study questionnaire.
Survey design and measurement
The final study questionnaire assessed demographic characteristics; aspects of cancer diagnosis, treatment burden concepts of self-management, difficulty, and impact; general health; and selected potential modifiers of treatment burden. Demographic status included, date of birth, county of residence, race, marital status, educational attainment, employment status, income level, and health insurance. Treatment burden items assessed self-management activities and difficulties related to cancer and, separately, for the survivor’s other health conditions. The present study focused on self-management associated with cancer by asking participants whether they did specific self-management activities for their cancer. These items included: taking medications, scheduling medical appointments, monitoring health conditions and behaviors (such as exercise, dieting, body weight, blood pressure, blood sugar), finding reliable health information about cancer, having a routine or program for regular exercise, and needing to use medical devices or equipment for health (such as a glucose monitor, blood pressure cuff or wheelchair). We assessed whether the respondent generally performed each activity for her cancer condition (yes, no, don’t know), and summed the total number as cancer self-management activities (see Appendix A). For each activity, we then assessed the level of ease or difficulty (very easy, easy, neither easy nor difficult, difficult, very difficult), which were summed as self-management difficulties. We assessed self-management impact by asking respondents to rate the extent that their self-management influenced their role and social activities and levels of physical and mental exhaustion. The later were assessed using the PETS impact scales developed by Eton et al. [6]. Perceived general health was assessed using 9 of 10 items from the Patient-Reported Outcomes Measurement Information System (PROMIS) Global-10 scale. These items covered areas of general physical functioning, emotional health, social participation, pain, fatigue, and overall perception of quality of life [15].
Potential modifiers of treatment burden assessed included number of current comorbid conditions that require self-management (diabetes, high blood pressure, high cholesterol, depression, anxiety, neuropathy, arthritis, other), a rating of financial security using a single item that asked how comfortably participants lived on their current household income (living comfortably on present income/getting by on present income/finding it difficult on present income/finding it very difficult on present income); and health literacy using a single item that asked how often participants need help reading instructions, pamphlets, or other written material from their doctor or pharmacy (never, rarely, sometimes, often, always) [16]. Cancer characteristics assessed included (cancer type [s], cancer treatment type and end date, year cancer was diagnosed).
Survey population
All adult survivors of women’s cancers who were within a 6-month to 3-year window from date of last treatment were identified from the institutional patient registries at two large cancer centers in Virginia. Patients treated at one of the participating cancer centers are mostly from small urban areas with a small proportion from rural areas, while the other cancer center serves a mostly rural catchment area. Women were selected for contact if they were age 18 years of age or older, had a diagnosis of breast, cervical, ovarian, or endometrial/uterine cancer, stage I, II, or III, and completed active treatment (surgery, chemotherapy and/or radiotherapy) between 6 months and 3 years prior to the lookup date. To insure that adequate numbers of rural cancer survivors were included from each participating cancer center site despite differences in patient volume and mix, at the larger cancer center which sees mostly urban patients, all cancer survivors with rural or non-metropolitan 5-digit Zip codes were selected for contact (N = 411). For the remainder, 50% (N = 698) of those living in urban or metropolitan areas were randomly selected for contact. At the smaller cancer center that sees a high proportion of rural patients, all eligible survivors (N = 688) were selected for contact. To protect patient privacy, staff at each cancer center mailed an invitation to be contacted and cover letter describing the study to their respective patients. All patients who returned the study invitation card approving contact were attempted for follow-up by telephone. During the telephone call, detailed information was provided about the scope of the study and participation, and treatment window eligibility was self-verified. Survivors who remained eligible and affirmed performing self-management for at least one additional chronic health condition were invited to provide oral informed consent and complete the study survey. The interviews and administration of the surveys were conducted by trained, female interviewers by an academic research center using Computer Assisted Telephone Interviewing (CATI) software to facilitate ease of dialing, track call attempts, and to facilitate data entry including skip patterns and item eligibility.
Statistical analysis
Measures
Cancer self-management difficulty item responses were summed by averaging all self-management tasks respondents reported performing (1-very easy to 5-very difficult). A total score for impact of cancer self-management on psychosocial functioning was calculated by determining separate scores for the two PETS impact scales (i.e. role/social activity limitations and physical/mental exhaustion), transforming the scale scores as 0 and 100, with higher scores indicating greater impact, and taking their mean. For the general health items component scores were derived for physical health (GPH) and mental health (GMH) using PROMIS component score algorithms [15]. Because our physical health measure contained fewer items than the standard PROMIS item set, our GPH score is approximate.
Regression and mediation analysis
A model of PETS self-management impact and general health (GPH and GMH) was constructed by testing differences in means for the theoretical covariates using t-tests for regression coefficients and maximum difference in means for multi-categorical predictors using Tukey’s Studentized Range. Covariates examined included age at diagnosis, race, rurality, education, marital status, employment, income, financial security, health literacy, number of comorbidities, cancer type, and chemotherapy. To avoid the risk of over-adjustment, we introduced covariates sequentially according to their hypothesized order of antecedence to self-management impact. Step 1 covariates were age, race, rurality, education, marital status; step 2 added employment and income; step 3 added financial security, and health literacy; step 4 added number of comorbidities, cancer type and chemotherapy, and the final step added self-management difficulties. We expanded this model by including effects of PETS impact score on GPH and GMH. Covariate adjustment was conducted by multivariate regression modeling and tested with the Tukey-Kramer method for multi-category variables. The quantities omega-squared, ω2, an estimate of population proportion of variance explained, and partial ω2 were used as measures of effect size, and estimate the proportion of variance explained by each predictor, independent of the other predictors considered [17]. Estimates of 0.01, 0.06 and 0.14 have been cited as thresholds for small, medium and large effects [17].
Because comorbidity status has been shown to be strongly associated with cancer survivors’ general health and is also likely associated with self-management difficulty and impact, we hypothesized both direct and indirect pathways could exist in its relationship with general health. Direct effects were considered those that can be attributed solely to level of comorbidity (i.e., assumed to operate through functional impairment or disability), while indirect effects were those operating through self-management difficulty or impact by influencing self-management role functioning and physical/mental exhaustion. A path analysis model was fit to the data using the SAS CALIS procedure, regressing GPH and GMH separately as a function of exposure, covariates, and mediators, and regressing the hypothesized mediators (i.e. self-management difficulties and impact) individually as a function of the covariates. All independent, or, exogenous variables, were allowed to co-vary freely with other exogenous variables. In addition to assuming a linear relationship and approximately normal and symmetrical distribution of residuals, the assumption of no moderation of the mediators on direct effect was tested by conducting statistical interactions, and the assumption of no latent confounding between the exposure-outcome and mediator-outcome was made to identify controlled direct effects (CDE). Decomposition of exposure effects into direct and indirect effects (overall and attributable to each mediator), was made using the ‘EFFPART’ statement in the CALIS procedure.
Missing data were present for income (20 cases, 11% of total) and the self-management difficulty variable (8 cases, 4% of total). When included as covariates in the regression, both variables were imputed using a FCS (Fully Conditional Specification) missing data method available in the SAS system MI procedure (v 9.4), assuming data followed the missing at random (MAR) assumption. For the mediation analysis, instead of imputation, the path model was estimated using Full Information Maximum Likelihood (FIML), which incorporates cases with missing data under the MAR assumption [18].