Study Sample and Data Collection
The target sample was women aged 20–79 who were diagnosed withbreast cancer within the previous 18 months. The sample included patients with American Joint Committee on Cancer (AJCC) stage 0 – III breast cancer, with the intention to evaluate the full spectrum of experiences in patients who might have many treatment options, as well as those with more advanced disease in whom treatment recommendations might be more prescriptive. Participants were recruited from Memorial Sloan-Kettering Cancer Center (MSKCC) in New York and from Emory University Hospital Midtown, the Winship Cancer Institute of Emory University, and Grady Memorial Hospital in Georgia between June and September 2013. Based on an estimated sample size for adequacy of psychometric analyses, we set a quota sample of 200 completed surveys.
At MSKCC, eligible breast cancer patients were approached in clinic and asked to complete the survey. Patients who met the inclusion criteria were identified by examining the clinic schedule for the upcoming day and approaching all eligible patients on a selected day. These women were given the option to take the survey home to complete if requested. A $10 incentive was provided to respondents upon completion of their survey.
At the Georgia sites, eligible breast cancer patients were identified by clinic records and mailed a survey packet which included a $10 pre-incentive. The institutional review boards of the University of Michigan, MSKCC, and Emory University approved all study procedures and materials.
The response rate was 83.8 % (93 of 111) in New York and 54.0 % (67 of 124) in Georgia, for a combined response rate of 67.2 % (158/235). For factor and internal consistency analysis by physician specialty, which required all items to be completed, the final analytical sample size was 155, 157, 138, and 106, for the overall treatment experience (hereafter referred to as “overall”), surgeon, medical oncologist, and radiation oncologist scales, respectively. For hierarchical factor analysis that required all provider-specific items, the analytical sample size was 106.
Measures
Patient characteristics
Participants were asked about their age, race, ethnicity, and level of education as well as the amount of time (in months) since their breast cancer diagnosis. We also asked yes/no questions to ascertain whether or not they had received various treatments, specifically lumpectomy, mastectomy, radiation therapy, and chemotherapy, and whether they experienced moderate or severe toxicity during their treatment (defined as nausea, vomiting, diarrhea, shortness of breath, pain, or arm swelling).
Perceived autonomy support was assessed with six questions that measured patients’ perceptions of the degree to which their physicians were autonomy supportive. Patients responded to the six questions for their overall treatment experience, followed by questions about their surgeon, medical oncologist, and radiation oncologist, in that order. The six questions, which were provided by the scale’s developer (GW) were asked as follows:
I feel that my (insert breast cancer treatment doctors, surgeon, medical oncologist, or radiation oncologist)…
-
1)
…provided me with choices and options for my breast cancer treatment.
-
2)
…understood how I saw things with respect to my breast cancer.
-
3)
…expressed confidence in my ability to make decisions.
-
4)
…listened to how I would like to handle my breast cancer treatment.
-
5)
…encouraged me to ask questions.
-
6)
…tried to understand how I saw things before offering an opinion.
Responses were on a 7 point scale anchored with: not at all true (1), somewhat true (4), and very true (7).
Analyses
Exploratory factor analysis (EFA) using principal components was used to explore the factor structure, i.e. the number of underlying constructs measured by the items. We began by retaining factors with Eigenvalues near to 1 (indicating that approximately 16.7 % of the variance was explained) and required item-loadings of > 0.45 as indication that the items should be retained. After scales were formed from the factor(s), we measured their internal consistency using Cronbach’s alpha and reported the correlation between the scales as calculated for the four groups (3 provider groups and the overall rating) assessed using Spearman’s rank coefficient. We also evaluated the correlation between provider level scales and the overall scale. We then explored the association of scales stratified by the surgery, chemotherapy, and radiation received, and by patient characteristics using the Kruskal-Wallis test. When the Kruskal-Wallis test suggested a significance difference among the groups, pairwise Wilcoxon Rank-sum tests were performed.
In order to explore whether assessment was needed on the provider level rather than simply asking about overall treatment experience, we examined inter-item correlations for 18 items measured at the provider-specific level using Spearman’s rank coefficient and hierarchical EFA with first- and second-order factors. The first order EFA was used to determine the common constructs/factors measured by the 18 items. It was hypothesized if patients did differentiate between providers that the 6 items asked about each specific provider would compose common factors, with three common factors, one for each provider type discovered. Conversely, if patients did not differentiate between providers, then it was hypothesized that the same questions asked across the three provider types would compose common factors, resulting in 6 factors, each with three items. Once the first-order factor solution was decided, those factor solutions were then used as the inputs into the second-order EFA. If patients did not differentiate between provider types, it was hypothesized that a single common second-order factor would explain the majority of the variance of the provider-level first-order factors. If, however, patients did differentiate between providers, a single common second-order factor would leave considerable variance unexplained. Additionally the feasibility of the single second-order factor was determined by considering the interpretability of the first-order factors loadings. The SAS System version 9.3 (SAS Institute Inc., Cary, NC, USA) was used for all statistical analyses. For statistical tests, p-values at or below 5 % were considered significant.