Archives

  • 2019-10
  • 2019-11
  • 2020-03
  • 2020-07
  • 2020-08
  • br Methods br Study Design and Sample br

    2020-08-12


    2. Methods
    2.1. Study Design and Sample
    This was a cross-sectional analysis of data from surveys among older Medicare beneficiaries with cancer administered by the University of Alabama at Birmingham (UAB) Health System Cancer Community Net-work (CCN). The UAB CCN is comprised of twelve academic and com-munity hospitals in Alabama, Georgia, Tennessee, Mississippi, and Florida [23]. Survey collection was conducted from November 2013 to June 2015. The Institutional Review Boards of UAB and other participat-ing CCN sites approved the study.
    Eligible participants identified through hospital registries included Medicare beneficiaries aged 65 and older diagnosed with cancer after January 1, 2008. Telephone surveys were administered by trained inter-viewers from the UAB Recruitment and Retention Shared Facility. Pa-tients who were from minority populations, had severe disease, were within one year of diagnosis, or were treated at smaller volume cancer centers (b4000 cancer cases/year) were oversampled to ensure a di-verse group of respondents. A total of 5723 patients were contacted, of which 24% were deceased and 22% had disconnected phones. Of the remaining 3090 contacts, 1460 (47%) completed the survey. In this Doxorubicin analysis, we included 637 individuals with the three most common cancer types (breast [n = 347, 23.8%], prostate [n = 189, 13%], or lung [n = 169, 11.6%] cancer) and no missing data on HRQoL or satisfaction (n = 68 [0.9%] excluded due to missing data).
    The main outcomes in this study are domain-specific patient satisfac-tion scores. Respondents were asked eighteen questions [24] about their personal satisfaction with cancer-related care. All eighteen items (listed
    in Table 2) had a five-level ordinal response choice, from “strongly agree” to “strongly disagree”. To obtain domain-specific scores, we followed a scoring approach used in the Consumer Assessment of Healthcare Providers and Systems (CAHPS) Clinician & Group Survey Da-tabase [25] for similar likert-type items related to patient experience. Re-sponses were scored dichotomously using a top-box method, with “strongly agree” indicating the highest level of satisfaction with care, and “agree”, “neither agree nor disagree”, “disagree”, or “strongly dis-agree” indicating lower satisfaction with care. Dichotomized satisfaction item responses were then grouped into five conceptual domains: (1) Ac-cess to Care, (2) Coordination of Care, (3) Patient-Provider Communica-tion, (4) Quality of Care, and (5) Patient Engagement. For a given patient, domain scores were computed as the proportion of domain-specific items for which the respondent indicated the highest level of satisfaction (i.e., with “strongly agree” responses), ranging from zero (no questions in the domain indicated high-satisfaction) to one (all questions in the do-main indicated high-satisfaction). Each question within a domain con-tributed an equal weight based on the number of questions included in that domain.
    2.3. Health-related Quality of Life (HRQoL)
    HRQoL was measured using the physical (PCS) and mental compo-nent summary (MCS) scores of the SF-12.v2 Health Survey [26]. PCS and MCS scores range from 0 to 100, with higher scores indicating bet-ter HRQoL. In the general U.S. adult population, PCS and MCS scores have a mean of 50 and standard deviation of ten.
    2.4. Statistical Analysis
    Characteristics for all respondents were described using means and standard deviations (SD) for continuous variables and frequencies and percentages for categorical variables. Individual item and overall satis-faction domain scores were tabulated for the entire sample and by can-cer type subgroups (breast, lung, prostate). The associations between satisfaction domain scores and HRQoL scores were estimated linearly using fractional logit regression, a modeling approach robust to distri-butional assumptions for numerical outcomes bounded to the range zero-one, such as fractions or proportions [27,28]. Adjusted models included as additional predictors: cancer type (breast, prostate, lung), race (white, non-white), education (less than high school, high school graduate, some college, college graduate), phase of care (initial [less than one year from diagnosis], survivor [at least one year from diagno-sis]), and National Cancer Institute comorbidity index score (zero, one, two or more). Comorbidities were assessed from Medicare administra-tive claims data for the period 2012–2015 [27]. Odds ratios and their corresponding 95% confidence intervals were estimated for all unad-justed and adjusted models. To aid in interpretation, selected log odds ratios were transformed [29] to Cohen's r effect sizes [30] (small~0.1, medium~0.3, large~0.5), and the unadjusted and adjusted model-predicted satisfaction domain scores were computed at the mean HRQoL scores and at the mean scores ± one standard deviation. Analy-ses were performed using SAS© software, version 9.4 (SAS Institute, Cary, NC).