Advertisement

Effects of Caregiver and Decedent Characteristics on CAHPS Hospice Survey Scores

Open ArchivePublished:July 23, 2018DOI:https://doi.org/10.1016/j.jpainsymman.2018.07.014

      Abstract

      Context

      The Consumer Assessment of Healthcare Providers and Systems (CAHPS®) Hospice Survey assesses the care experiences of hospice patients and their families. Public reporting of hospice performance on these survey measures began in February 2018.

      Objectives

      Develop an appropriate case-mix adjustment (CMA) model to allow for fair comparisons between hospices.

      Methods

      We analyzed CAHPS Hospice Survey data reflecting experiences of 915,442 patients who received care from 2513 hospice programs between April 2015 and March 2016. Decedent and caregiver characteristics were identified for inclusion in CMA based on their variation across hospices (as measured by intraclass correlation coefficients [ICCs]) and how predictive they were of responses to survey questions (as assessed by linear regression).

      Results

      The final CMA model included decedent age, payer for hospice care, primary diagnosis, length of final episode of hospice care, caregiver age, caregiver education, relationship to decedent, survey language/language spoken at home, and response percentile. The characteristics that varied most across hospices were language (ICC = 0.48 for Spanish survey or home language) and payer for hospice care (ICC = 0.42 for Medicare only; ICC = 0.35 for Medicare and private insurance). The characteristics that were most predictive of caregivers' survey responses were payer for hospice care, caregiver education, and survey language/language spoken at home. Lack of appropriate adjustment would incorrectly rank hospices by 1.2–5.4 percentile points.

      Conclusion

      To ensure fair comparisons across hospices, CAHPS Hospice Survey measure scores should be adjusted for several caregiver and decedent characteristics.

      Key Words

      Introduction

      Interest in monitoring care quality in U.S. hospices has grown as the use of hospice care has steadily increased,
      for-profit and chain hospices claim a larger share of this care,
      • Stevenson D.G.
      • Dalton J.B.
      • Grabowski D.C.
      • Huskamp H.A.
      Nearly half of all Medicare hospice enrollees received care from agencies owned by regional or national chains.
      and research has identified considerable variation in hospice care processes.
      • Stevenson D.G.
      • Grabowski D.C.
      • Keating N.L.
      • Huskamp H.A.
      Effect of ownership on hospice service use: 2005-2011.
      • Prsic E.
      • Plotzke M.
      • Christian T.J.
      • Gozalo P.
      • Teno J.M.
      A national study of live hospice discharges between 2000 and 2012.
      • Teno J.M.
      • Plotzke M.
      • Christian T.
      • Gozalo P.
      Examining variation in hospice visits by professional staff in the last 2 days of life.
      In 2012, the Centers for Medicare & Medicaid Services (CMS) introduced the Hospice Quality Reporting Program to measure and compare the quality of care provided by hospices. In the Hospice Quality Reporting Program, hospices must submit data on clinical quality measures
      • Zheng N.T.
      • Li Q.
      • Hanson L.C.
      • et al.
      Nationwide quality of hospice care: findings from the Centers for Medicare & Medicaid services hospice quality reporting program.
      in the Hospice Item Set, and since 2015, collect data on patient and family experiences of hospice care using the Consumer Assessment of Healthcare Providers and Systems (CAHPS®) Hospice Survey to avoid a financial penalty.

      Price RA, Quigley DD, Bradley MA, et al. Hospice experience of care survey. 2014. Available at: https://www.rand.org/pubs/research_reports/RR657.html. Accessed June 10, 2018.

      Centers for Medicaid & Medicare Services
      Hospice Quality Reporting Program (HQRP): Requirements for the Fiscal Year (FY) 2018 Reporting Year.
      Public reporting of hospice performance on CAHPS Hospice Survey measures began in 2018.
      To ensure that publicly reported CAHPS Hospice Survey scores can be fairly compared across hospices, it is important to adjust for case-mix factors that influence how caregivers respond to survey questions but are not consequences of the hospice's care quality or otherwise controlled by the hospice.
      • Rau J.
      Medicare to begin basing hospital payments on patient-satisfaction scores.
      • Zaslavsky A.M.
      • Zaborski L.B.
      • Ding L.
      • Shaul J.A.
      Adjusting performance measures to ensure equitable plan comparisons.
      Furthermore, without such adjustment, hospices would have an incentive to avoid patients and caregivers likely to give lower survey ratings.
      National Academies of Sciences, Engineering, and Medicine
      Accounting for social risk factors in Medicare payment: Identifying social risk factors.
      In the hospice setting, characteristics of both caregiver respondents and hospice patients are important to consider for adjustment. Unlike most other experience of care surveys, the caregiver (usually a family member), rather than the patient, responds to the CAHPS Hospice Survey. In addition, the patient and his/her caregiver together compose the unit of care for hospice, and therefore, the CAHPS Hospice Survey aims to assess the care experiences of both the patient and his/her caregiver. In this article, we identify an appropriate case-mix adjustment (CMA) model for the CAHPS Hospice Survey. We also describe the degree to which this adjustment affects hospice scores.

      Methods

      Study Sample

      The CAHPS Hospice Survey was designed to be completed by the primary caregiver of the hospice decedent; eligibility details are described elsewhere.
      • Price R.A.
      • Parast L.
      • Haas A.
      • Teno J.M.
      • Elliott M.N.
      Black and Hispanic patients receive hospice care similar to that of white patients when in the same hospices.
      Since April 2015, all Medicare-certified hospices are required to participate in the CAHPS Hospice Survey unless they are new (i.e., received their Centers for Medicare & Medicaid Services certification number on or after the first day of the performance year) or served fewer than 50 survey-eligible decedents/caregivers in the prior calendar year. Surveys were administered using either mail-only mode, telephone-only mode, or mixed mode. Details of the mode protocol are specified elsewhere;

      Consumer Assessment of Healthcare Providers and Systems (CAHPS) Hospice Quality Assurance Guidelines, Version 4.0; September 2017. Available at: http://www.hospicecahpssurvey.org/globalassets/hospice-cahps/quality-assurance-guidelines/cahps-hospice-survey-quality-assurance-guidelines-v4.0_september-2017.pdf. Accessed June 10, 2018.

      for all modes, data collection closed within 42 days after the first survey mailing or first phone attempt.
      We analyzed CAHPS Hospice Survey data regarding patients who received hospice care from April 2015 to March 2016. Data from the 900 hospices with fewer than 20 respondents during this interval were excluded from all analyses. Therefore, we report on 915,442 hospice deaths from 2513 hospice programs.

      Survey Instrument, Measures, and Scoring

      The CAHPS Hospice Survey instrument

      Consumer Assessment of Healthcare Providers and Systems (CAHPS) Hospice Quality Assurance Guidelines, Version 4.0; September 2017. Available at: http://www.hospicecahpssurvey.org/globalassets/hospice-cahps/quality-assurance-guidelines/cahps-hospice-survey-quality-assurance-guidelines-v4.0_september-2017.pdf. Accessed June 10, 2018.

      consists of 47 items and is used to calculate eight hospice quality measures (Table 1). Responses for all evaluative questions were transformed from the original response scale to a 0/100 value using top-box coding, where the most favorable response option(s) was coded as 100 and other options as 0. For example, for frequency (how often?), questions with response options never, sometimes, usually, always, where always indicates better quality, a response of always was coded 100 and any other response was coded 0.
      Table 1Dependent Variables for CMA Analyses
      ItemMeasure RepresentedResponse OptionsTop BoxOverall Average Top Box Score
      While your family member was in hospice care, how often did the hospice team keep you informed about when they would arrive to care for your family member?Hospice team communication (composite measure)Never; sometimes; usually; alwaysAlways72.4
      While your family member was in hospice care, when you or your family member asked for help from the hospice team, how often did you get help as soon as you needed it?Getting timely care (composite measure)Never; sometimes; usually; alwaysAlways78.4
      While your family member was in hospice care, how often did the hospice team treat your family member with dignity and respect?Treating family member with respect (composite measure)Never; sometimes; usually; alwaysAlways93.5
      In the weeks after your family member died, how much emotional support did you get from the hospice team?Getting emotional and religious support (composite measure)Too little; right amount; too muchRight amount84.8
      Did your family member get as much help with pain as he or she needed?Getting help for symptoms (composite measure)Yes, definitely; yes, somewhat; noYes, definitely84.7
      Did the hospice team give you the training you needed about what to do if your family member became restless or agitated?Getting hospice care training (composite measure)Yes, definitely; yes, somewhat; noYes, definitely65.1
      Using any number from 0 to 10, where 0 is the worst hospice care possible and 10 is the best hospice care possible, what number would you use to rate your family member's hospice care?Rating of hospice (global measure)0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 109 and 1080.3
      Would you recommend this hospice to your friends and family?Recommend hospice (global measure)Definitely no; probably no; probably yes; definitely yesDefinitely yes84.5
      CMA = case-mix adjustment.

      Dependent Variables: Survey Responses

      Eight items (one from each of the eight measures), scored as described previously, were selected as outcomes in our analyses (Table 1). These items were selected to represent the composite measures because they varied substantially across hospices and had different response scales.

      Independent Variables: Caregiver and Decedent Characteristics

      Caregiver and decedent characteristics were obtained from hospice administrative data and survey response data. Hospice administrative data contained information about the caregiver's relationship to the decedent (e.g., spouse or partner, child or stepchild, parent, other family member, friend, or other), decedent age at death in years, decedent gender, decedent race/ethnicity (white, black, Hispanic, Asian or Pacific Islander, multiracial, or other), payer for hospice care (Medicare only, Medicaid only or Medicaid and private, Medicare and Medicaid, private only or Medicare and private, or other), decedent's final setting of care (home, nursing home, acute care hospital, hospice inpatient unit, or other), decedent's length of final episode of hospice care (the length of the decedent's last stay in hospice, i.e., the stay in which death occurred, relevant when there were multiple experiences with hospice), and decedent's primary diagnosis. The classification of decedent's primary diagnosis into 20 categories (Table 2) was informed by both physician input and analyses that tested the effect of collapsing/combining of various diagnoses on predictive strength.
      Table 2Effect of Case-Mix Adjustors on Eight Selected Evaluative Questions (One Selected From Each Measure)
      Models also control for hospice.
      Case-Mix AdjustorHospice Team Kept You Informed About When They Would ArriveGot Help From the Hospice Team as Soon as You Needed ItHospice Team Treated Your Family Member With Dignity and RespectGot the Right Amount of Emotional Support From the Hospice Team After Your Family Member DiedFamily Member Got Needed Help With PainHospice Team Gave You the Training You Needed About What to Do if Your Family Member Became Restless or Agitated
      Note that this question is only examined among those indicating on the survey that the decedent was in the home and/or assisted living facility.
      Overall Rating of HospiceRecommend Hospice
      Coefficient (SE)
      Decedent age
       Age 18–547.64 (0.67)1.48 (0.62)−0.98 (0.33)−2.82 (0.49)−2.35 (0.61)1.03 (1.01)0.62 (0.54)0.62 (0.52)
       Age 55–645.93 (0.49)0.70 (0.45)−0.68 (0.24)−2.73 (0.36)−0.78 (0.46)1.62 (0.75)0.24 (0.40)0.25 (0.38)
       Age 65–695.40 (0.43)0.70 (0.40)−0.72 (0.21)−2.47 (0.32)−0.77 (0.40)2.38 (0.66)−0.11 (0.35)0.22 (0.34)
       Age 70–745.29 (0.38)0.82 (0.36)−0.43 (0.19)−1.72 (0.28)0.06 (0.36)2.63 (0.60)0.33 (0.31)0.55 (0.30)
       Age 75–794.90 (0.34)1.17 (0.32)−0.18 (0.17)−1.30 (0.25)0.66 (0.33)3.54 (0.54)0.36 (0.28)0.87 (0.26)
       Age 80–843.34 (0.29)0.35 (0.27)−0.26 (0.14)−1.11 (0.22)0.39 (0.29)1.88 (0.48)0.31 (0.24)0.78 (0.23)
       Age 85–892.15 (0.26)0.41 (0.24)−0.18 (0.13)−0.83 (0.19)0.39 (0.25)1.29 (0.43)0.25 (0.21)0.17 (0.20)
       Age 90+ (Reference)REFREFREFREFREFREFREFREF
      Payer for hospice care
       Medicare only (Reference)REFREFREFREFREFREFREFREF
       Medicaid only or Medicaid and private insurance−1.91 (0.71)−1.22 (0.66)−1.31 (0.35)−2.94 (0.52)−1.45 (0.65)−1.53 (1.09)−1.52 (0.58)−1.37 (0.55)
       Medicare and Medicaid−9.98 (0.46)−2.28 (0.43)−1.23 (0.23)−1.44 (0.34)−3.84 (0.45)−7.20 (1.07)−4.44 (0.38)−4.09 (0.36)
       Private insurance only−1.04 (0.49)−0.86 (0.46)−0.49 (0.24)0.93 (0.36)−0.68 (0.45)−0.61 (0.72)−0.67 (0.40)−0.79 (0.38)
       Medicare and private insurance−0.97 (0.59)−0.40 (0.55)−0.04 (0.29)0.17 (0.44)0.28 (0.56)0.69 (0.91)−0.09 (0.48)−0.42 (0.46)
       Other−3.52 (0.41)−0.73 (0.38)−0.70 (0.20)0.23 (0.31)−0.84 (0.39)−1.11 (0.71)−0.57 (0.34)−0.52 (0.32)
      Primary diagnosis
       Alzheimer's and non-Alzheimer's dementias (Reference)REFREFREFREFREFREFREFREF
       Bladder cancer6.54 (0.82)−0.98 (0.77)−0.15 (0.40)−2.15 (0.61)−1.02 (0.75)0.14 (1.22)−0.93 (0.67)−1.63 (0.64)
       Blood and lymphatic cancers7.99 (0.59)−1.77 (0.55)−0.64 (0.29)−1.50 (0.44)−0.76 (0.56)0.00 (0.94)−1.45 (0.48)−1.14 (0.46)
       Brain cancer6.59 (0.80)−2.33 (0.74)0.04 (0.39)−0.17 (0.59)2.37 (0.78)5.13 (1.21)0.22 (0.65)−0.27 (0.62)
       Breast cancer8.16 (0.63)−0.01 (0.59)0.32 (0.31)−0.61 (0.47)0.98 (0.58)2.54 (0.98)−0.21 (0.52)0.10 (0.50)
       Congestive heart failure5.50 (0.38)−1.02 (0.36)−0.40 (0.19)−1.82 (0.28)−1.51 (0.37)0.56 (0.61)−1.11 (0.31)−0.63 (0.30)
       Chronic kidney disease5.03 (0.62)−1.03 (0.58)−0.91 (0.30)−1.26 (0.46)−1.93 (0.61)3.29 (1.09)−1.51 (0.51)−1.71 (0.49)
       Chronic liver disease5.80 (0.74)0.04 (0.69)−0.45 (0.36)−1.31 (0.54)1.72 (0.71)4.44 (1.17)−1.09 (0.60)−0.67 (0.58)
       Colorectal cancer8.81 (0.55)−0.40 (0.52)−0.12 (0.27)−1.51 (0.41)0.08 (0.51)1.63 (0.85)−0.07 (0.45)−0.54 (0.43)
       CVA/stroke3.03 (0.40)−0.12 (0.38)−0.35 (0.20)−0.41 (0.30)0.39 (0.41)3.28 (0.73)0.12 (0.33)−0.08 (0.31)
       Liver cancer8.69 (0.72)−0.84 (0.67)−0.60 (0.35)−1.41 (0.53)−0.22 (0.66)2.13 (1.05)−1.50 (0.58)−2.35 (0.56)
       Lung & other chest cavity cancer9.08 (0.40)−0.15 (0.37)−0.38 (0.19)−1.47 (0.29)−0.29 (0.38)1.90 (0.61)−0.30 (0.32)−0.15 (0.31)
       Noninfectious respiratory6.64 (0.44)−1.51 (0.41)−0.51 (0.21)−2.10 (0.32)−1.09 (0.43)1.03 (0.68)−1.06 (0.36)−1.02 (0.34)
       Other heart disease5.70 (0.41)−0.29 (0.39)−0.29 (0.20)−0.93 (0.30)−0.92 (0.40)2.57 (0.67)−0.21 (0.34)−0.23 (0.32)
       Pancreatic cancer10.22 (0.57)−0.70 (0.53)−0.08 (0.28)−1.06 (0.42)−0.49 (0.52)0.54 (0.85)−0.61 (0.47)−0.45 (0.45)
      Coefficient (SE)
       Parkinson's and other degenerative diseases3.32 (0.60)−0.55 (0.56)0.16 (0.29)−1.22 (0.44)0.02 (0.59)0.85 (0.97)−0.17 (0.49)0.46 (0.47)
       Pneumonias and other infectious lung diseases4.67 (0.57)−0.40 (0.53)−0.47 (0.28)−1.43 (0.42)−0.38 (0.59)2.48 (1.03)−0.98 (0.47)−1.47 (0.45)
       Prostate cancer7.68 (0.65)−1.30 (0.61)−1.16 (0.32)−1.88 (0.48)−1.20 (0.59)0.90 (0.95)−1.52 (0.53)−1.49 (0.51)
       Other, cancer8.43 (0.38)−1.08 (0.35)−0.65 (0.18)−1.58 (0.28)−0.74 (0.36)0.74 (0.60)−0.91 (0.31)−0.90 (0.29)
       Other, noncancer3.63 (0.37)−0.50 (0.35)−0.54 (0.18)−0.70 (0.27)−0.68 (0.37)1.22 (0.65)−0.35 (0.30)−0.39 (0.29)
      Lengths of hospice stay (days)
       2–58.15 (0.26)6.19 (0.24)0.21 (0.13)2.26 (0.19)0.08 (0.26)−2.76 (0.47)1.07 (0.21)0.18 (0.20)
       6–126.44 (0.26)3.23 (0.24)−0.05 (0.13)1.66 (0.19)−0.50 (0.25)−1.52 (0.42)−0.11 (0.21)−0.23 (0.21)
       13–294.41 (0.26)0.43 (0.24)−0.48 (0.13)0.66 (0.19)−1.82 (0.24)−1.76 (0.38)−1.57 (0.21)−1.09 (0.21)
       30–802.51 (0.26)−0.57 (0.24)−0.31 (0.13)−0.30 (0.19)−1.50 (0.24)−1.43 (0.37)−1.73 (0.21)−1.20 (0.20)
       81+ (Reference)REFREFREFREFREFREFREFREF
      Caregiver age
       Age 18–44−0.58 (0.48)−3.10 (0.45)−2.28 (0.24)−5.63 (0.36)−2.12 (0.44)−1.09 (0.72)−4.85 (0.40)−3.30 (0.38)
       Age 45–540.75 (0.30)−1.17 (0.28)−0.79 (0.15)−2.61 (0.22)−0.49 (0.28)0.35 (0.46)−1.42 (0.24)−0.96 (0.23)
       Age 55–64 (Reference)REFREFREFREFREFREFREFREF
       Age 65–74−0.40 (0.23)1.22 (0.22)0.95 (0.11)1.90 (0.17)0.89 (0.22)−0.95 (0.37)1.79 (0.19)1.14 (0.18)
       Age 75–84−2.39 (0.31)0.46 (0.29)0.80 (0.15)2.73 (0.23)−0.91 (0.30)−5.67 (0.49)0.74 (0.26)0.24 (0.24)
       Age 85+−4.26 (0.42)−1.48 (0.39)−0.24 (0.21)1.76 (0.31)−2.69 (0.41)−8.72 (0.69)−1.03 (0.34)−0.87 (0.33)
      Caregiver education
       Eigth grade or less2.81 (0.75)0.61 (0.70)−0.88 (0.37)−2.62 (0.56)−0.83 (0.72)1.97 (1.16)0.13 (0.62)−0.72 (0.59)
       Some high school4.02 (0.45)2.21 (0.41)0.39 (0.22)−1.64 (0.33)1.54 (0.43)2.60 (0.67)1.56 (0.36)0.78 (0.35)
       High school graduate or GED (Reference)REFREFREFREFREFREFREFREF
       Some college−4.41 (0.22)−3.29 (0.21)−1.05 (0.11)−1.15 (0.16)−1.76 (0.21)−2.85 (0.35)−2.22 (0.18)−1.38 (0.17)
       Four-year college graduate−10.60 (0.27)−7.01 (0.25)−1.54 (0.13)−0.56 (0.20)−2.74 (0.26)−7.82 (0.43)−4.52 (0.22)−2.25 (0.21)
       More than four-year college graduate−13.41 (0.26)−9.70 (0.24)−2.46 (0.13)−1.89 (0.19)−4.44 (0.25)−10.31 (0.41)−5.95 (0.21)−3.13 (0.20)
      Caregiver relationship (decedent was the caregiver's)
       Spouse or partner5.62 (0.29)0.41 (0.27)−0.07 (0.14)−4.20 (0.22)0.24 (0.28)−0.10 (0.45)−0.77 (0.24)−0.95 (0.23)
       Parent (Reference)REFREFREFREFREFREFREFREF
       Mother-in-law or father-in-law2.34 (0.57)1.94 (0.53)1.35 (0.28)1.15 (0.42)2.75 (0.54)6.04 (0.88)2.63 (0.47)1.77 (0.45)
       Aunt or uncle−4.33 (0.56)1.99 (0.52)0.82 (0.27)3.09 (0.41)1.34 (0.55)−1.89 (1.00)1.09 (0.45)0.27 (0.43)
       Sister or brother−4.11 (0.48)−0.93 (0.45)−0.71 (0.23)0.84 (0.35)−0.83 (0.45)−3.63 (0.82)−1.31 (0.39)−2.17 (0.37)
       Child1.30 (0.74)0.81 (0.68)−0.11 (0.36)−1.63 (0.54)0.48 (0.69)−1.51 (1.15)−0.60 (0.60)−0.95 (0.57)
       Friend−3.70 (0.68)−1.69 (0.63)−0.37 (0.33)1.60 (0.50)0.55 (0.65)−4.01 (1.12)−0.43 (0.55)−0.17 (0.53)
       Other−0.85 (0.51)0.90 (0.48)0.11 (0.25)0.28 (0.38)1.26 (0.49)0.68 (0.82)−0.44 (0.42)−0.17 (0.40)
      Survey language/respondent's home language
       Spanish survey or home language5.95 (0.91)2.28 (0.85)2.11 (0.45)−11.87 (0.68)6.60 (0.89)12.34 (1.42)5.11 (0.75)3.27 (0.72)
       All others (Reference)REFREFREFREFREFREFREFREF
      Response percentile (scaled from 0 to 1)−5.18 (0.76)−8.77 (0.71)−2.17 (0.37)−2.45 (0.56)−4.07 (0.73)−6.11 (1.21)−8.02 (0.62)−8.24 (0.60)
      SE = standard error; CVA = cerebrovascular accident; GED = general equivalency diploma.
      Bold text indicates statistical significance (P < 0.05).
      a Models also control for hospice.
      b Note that this question is only examined among those indicating on the survey that the decedent was in the home and/or assisted living facility.
      Characteristics obtained from the survey were language of survey completion (English, Spanish, Chinese, Russian, Portuguese, and Vietnamese), decedent education and caregiver education (eighth grade or less, some high school but did not graduate, high school graduate, some college or two-year degree, four-year college graduate, and more than four-year degree), caregiver age (18–44, 45–54, 55–64, 65–74, 75–84, 85, or older), caregiver gender, caregiver's language spoken at home (English, Spanish, Chinese, Russian, Portuguese, Vietnamese, and other), caregiver's relationship to the decedent (as reported by the caregiver rather than obtained in the administrative file), all settings of care where the decedent received care from the hospice, and response percentile (described later).
      For analytic purposes, a variable combining survey language and language spoken at home was defined using the following categories: 1) Spanish survey language or Spanish home language and 2) all others. Only 0.1% of CAHPS Hospice participants, concentrated in very few hospices, responded in Chinese or preferred Chinese at home. Although adequately precise single-quarter Chinese-language CMA is not possible, several hospices would be adversely affected without adjustment, given strong negative associations with CAHPS scores.
      • Mayer L.A.
      • Elliott M.N.
      • Haas A.
      • Hays R.D.
      • Weinick R.M.
      Less use of extreme response options by Asians to standardized care scenarios may explain some racial/ethnic differences in CAHPS scores.
      • Elliott M.N.
      • Edwards W.S.
      • Klein D.J.
      • Heller A.
      Differences by survey language and mode among Chinese respondents to a CAHPS health plan survey.
      We therefore more precisely estimate the Chinese-language marginal effect pooled for more than eight quarters, adjusting hospices with >10% Chinese language responses after CMA (results not discussed).
      Our definition of response percentile parallels that used in the Hospital CAHPS Survey (HCAHPS); this quantity captures both hospice response rate (RR) and how quickly a caregiver responded compared with other caregivers in the same hospice and mode. This variable, scaled from 0 to 1, is often considered in CMA given that previous evidence has shown that lower RRs are often associated with more positive responses and that later responders often provide more negative responses.
      • Barron D.N.
      • West E.
      • Reeves R.
      • Hawkes D.
      It takes patience and persistence to get negative feedback about patients' experiences: a secondary analysis of national inpatient survey data.
      • Zaslavsky A.M.
      • Zaborski L.B.
      • Cleary P.D.
      Factors affecting response rates to the consumer assessment of Health Plans Study survey.
      • Elliott M.N.
      • Zaslavsky A.M.
      • Goldstein E.
      • et al.
      Effects of survey mode, patient mix, and nonresponse on CAHPS® hospital survey scores.
      • Saunders C.L.
      • Elliott M.N.
      • Lyratzopoulos G.
      • Abel G.A.
      Do differential response rates to patient surveys between organizations lead to unfair performance comparisons?: Evidence from the English Cancer Patient Experience Survey.
      Lengths of stay for CMA were categorized as 2–5, 6–12, 13–29, 30–80, and 81 or more days. This categorization was based on several factors: the distribution of lengths of stay among all decedents, the differences in response patterns between these categories, and efforts to address the concern that lengths of stay were within the control of the hospice (endogenous). Although case-mix adjustors should never include characteristics that are substantially influenced by the hospice, lengths of stay, at least up to 90 days, are driven by a patient's admitting condition and severity. That is, except for lengths of stay beyond 90 days, a hospice's actions largely do not influence the lengths of stay before death. We intentionally differentiate between lengths of stay less than 90 days to capture characteristics that patients bring to the hospice but do not differentiate beyond 90 days out of concern that patients with such long lengths of stay may have had an admitting condition that was amenable to treatment/medical care and/or that this may reflect the admission of a patient who does not have the characteristics intended for hospice care.
      Missingness was handled by imputing with the hospice mean for the variable, or if the variable was missing for the entire hospice, by imputing with the overall mean across all hospices.

      Statistical Analysis

      Nonresponse

      We used multivariate logistic regression with hospice fixed effects and indicators for mode of survey administration to examine the odds of responding to the survey by characteristics from administrative data.

      Case-Mix Adjustment

      The importance of adjusting responses for each patient or caregiver characteristic depends on the variation in the characteristics between entities (in this case, hospices) and the strength of the relationships between characteristics and scores. Therefore, CMA was investigated by examining both the variation in each potential case-mix adjustor across hospices and the predictive strength of each potential case-mix adjustor within hospices (i.e., the degree to which responses varied with each potential adjustor). Variation between hospice programs was quantified using the intraclass correlation coefficient (ICC). Predictive strength was assessed using bivariate and multivariate linear regression models where the outcome was the mode-adjusted 0–100 score for the selected item, and the predictors (examined individually and in a multivariate model) included the independent variables described in the previous section, along with hospice fixed effects. All responses were adjusted for mode of administration (mail only, telephone only, or mixed mode) before analysis.
      • Parast L.
      • Elliott M.N.
      • Hambarsoomian K.
      • Teno J.
      • Anhang Price R.
      Effects of survey mode on consumer assessment of Healthcare Providers and Systems (CAHPS) hospice survey scores.
      We evaluated the impact of each case-mix adjustor variable on adjustments by quantifying the relative variance of the adjustments (i.e., the variance of adjustments for that variable relative to the variance of hospice scores fully adjusted using all included variables); we considered values between .01 and .1 hospice-level SDs to indicate moderate impact and values greater than .1 SD to indicate notable impact. This quantity integrates information from the ICC estimates and the regression coefficients to provide an overall assessment of each adjustor's unique impact on hospice-level scores. Only characteristics that vary among hospices and have predictive power at the level of individual respondents can influence hospice-level scores.
      Once an appropriate CMA model was determined, we investigated the overall effect of the adjustment on hospice measure scores by comparing hospice-level estimates without CMA to hospice-level estimates with CMA. We examined this comparison using three metrics: Kendall's tau, the effect size of the adjustments in terms of hospice-level SDs, and the percentile rank shift attributable to adjustments. Kendall's tau is a measure of correlation between two sets of scores; it summarizes the proportion of hospice pairs whose relative rankings would be reversed by adjustment, scaled to range from −1 to 1. A tau of +1 would indicate that adjustment had no effect on the relative standings of any two hospices, whereas a tau of +0.90 would indicate that (1 − 0.90)/2 = 0.05 or 5% of all possible comparisons between two hospices would be reversed by the adjustment. For the percentile rank shift, we examined the percentile rank shift that would be associated with a lack of CMA, for a hospice that was truly at the median (50th percentile), and a hospice that was truly at the 90th percentile of hospice scores for that quality measure.
      With three notable exceptions (decedent race/ethnicity and both setting of care variables), all characteristics described as independent variables previously were tested in CMA analyses. Decedent race/ethnicity was not tested because of sensitivities around such adjustment. With respect to setting of care, prior studies have suggested substantially different care processes and reports of care experiences by setting
      • Teno J.M.
      • Plotzke M.
      • Christian T.
      • Gozalo P.
      Examining variation in hospice visits by professional staff in the last 2 days of life.
      • Teno J.M.
      • Clarridge B.R.
      • Casey V.
      • et al.
      Family perspectives on end-of-life care at the last place of care.
      and have demonstrated associations between both hospice profit status and setting as well as profit status and care processes.
      • Wachterman M.W.
      • Marcantonio E.R.
      • Davis R.B.
      • McCarthy E.P.
      Association of hospice agency profit status with patient diagnosis, location of care, and length of stay.
      True differences in quality of care across setting should be identified and remedied, rather than adjusted away, because hospices are required to provide the same range and quality of services regardless of setting.
      Compliance Program Guidance for Nursing Facilities (2008).
      Therefore, final setting of care (from hospice administrative data) and all settings of care (respondent reported) were not tested here.

      Results

      Of the 915,442 sampled decedents/caregivers, 21,618 decedents (2.4%) were determined to be ineligible after sampling, 599,994 did not respond to the survey, and 293,830 responded to the survey, for an overall RR of 33% (American Association for Public Opinion Research RR1).

      AAPOR. American Association for Public Opinion Research Standard Definitions: Final Dispositions of Case Codes and Outcome Rates for Surveys. 2016. Available at: https://www.aapor.org/AAPOR_Main/media/publications/Standard-Definitions20169theditionfinal.pdf. Accessed June 10, 2018.

      The first two columns of Table 3 present the composition of eligible sampled cases and respondents. Almost half of the sampled caregivers were children of the decedent. Most sampled decedents were older than 65 years, and more than 60% of sampled decedents were described as non-Hispanic white. The most common final setting of care among sampled decedents was home (47%). Medicare was the sole payer for hospice care for two-thirds of sampled decedents.
      Table 3Characteristics of Eligible Sampled Decedents/Caregivers and Respondents
      CharacteristicEligible Sampled (N = 893,824), %All Respondents (N = 293,830), %RR in This Category,
      P-value is from a model examining probability of response, adjusting for this variable, survey mode, and hospice fixed effects only. Missing values are imputed.
      %
      Adjusted Odds Ratio of Responding (95% CI)
      Odds ratios from a multivariate model examining probability of response adjusting for survey mode and hospice fixed effects and all other characteristics in the table. Missing values are imputed.
      ,
      Note that, as is often the case, bivariate and multivariate regression coefficients differ in sign in several instances, so that unadjusted RRs may be higher for a group (e.g., caregivers of male decedents) associated with a lower adjusted (multivariate) odds ratio of responding.
      Decedent characteristics
       Sex
      Female [Reference]5654321.00
      Male444634***0.95 (0.94, 0.96)***
       Age
      18–545321***0.33 (0.32, 0.34)***
      55–6410826***0.47 (0.46, 0.48)***
      65–74171631***0.60 (0.59, 0.61)***
      75–84272734***0.76 (0.75, 0.77)***
      85 or older [Reference]4145361.00
       Race/ethnicity
      Non-Hispanic white [Reference]6369361.00
      Black6319***0.53 (0.52, 0.55)***
      Hispanic5422***0.53 (0.51, 0.55)***
      Asian or Pacific Islander1121***0.6 (0.57, 0.63)***
      Multiracial or other5532***0.77 (0.74, 0.80)***
       Final setting of care
      Home [Reference]4751351.00
      Nursing home201829***0.80 (0.79, 0.81)***
      Acute care hospital6426***0.80 (0.78, 0.82)***
      Hospice inpatient unit141433***1.03 (1.01, 1.04)**
      Other1128***0.80 (0.75, 0.86)***
       Length of final episode of hospice care
      Less than one week282529***0.63 (0.62, 0.64)***
      One week to less than one month333332***0.77 (0.76, 0.78)***
      One month to less than six months293136***0.90 (0.88, 0.91)***
      Six or more months [Reference]1011381.00
       Primary diagnosis
      Alzheimer's and non-Alzheimer's dementias [Reference]1212341.00
      Blood and lymphatic cancers22351.07 (1.03, 1.11)***
      Nonblood cancers
      Includes cancers of the bladder, brain, breast, colon and rectum, liver, lung, ovaries, pancreas, prostate, and stomach.
      232232***1.03 (1.01, 1.05)***
      Chronic end-organ diseases171732***1.01 (0.99, 1.03)
      Other353533***1.03 (1.01, 1.04)**
       Payer for health care services
      Medicare only [Reference]6669351.00
      Medicaid only or Medicaid/private3218***0.75 (0.73, 0.78)***
      Medicare/Medicaid5425***0.80 (0.78, 0.82)***
      Private only or private/Medicare8832***0.98 (0.96, 1.00)*
      Uninsured/no payer or other131332***0.93 (0.91, 0.95)***
      Caregiver characteristics
       Relationship to decedent (caregiver was decedent's)
      Spouse or partner293944***2.51 (2.48, 2.54)***
      Parent2233***2.85 (2.73, 2.97)***
      Child [Reference]4741291.00
      Other family member11826***1.05 (1.03, 1.06)***
      Friend2226***1.02 (0.98, 1.06)
      Other4326***0.97 (0.95, 1)
      RR = response rate.
      Percentages calculated among all cases (including missing); *P < 0.05, **P < 0.01, ***P < 0.001.
      a P-value is from a model examining probability of response, adjusting for this variable, survey mode, and hospice fixed effects only. Missing values are imputed.
      b Odds ratios from a multivariate model examining probability of response adjusting for survey mode and hospice fixed effects and all other characteristics in the table. Missing values are imputed.
      c Note that, as is often the case, bivariate and multivariate regression coefficients differ in sign in several instances, so that unadjusted RRs may be higher for a group (e.g., caregivers of male decedents) associated with a lower adjusted (multivariate) odds ratio of responding.
      d Includes cancers of the bladder, brain, breast, colon and rectum, liver, lung, ovaries, pancreas, prostate, and stomach.

      Odds of Responding to the Survey

      Table 3 presents the RRs by various characteristics and logistic regression model results. All characteristics were associated with the odds of response. For example, caregivers who were the decedent's parent or spouse or partner were more likely to respond to the survey compared with caregivers who were the decedent's child (33% and 44% RRs for parent and spouse/partner, respectively; 29% RR for child). Caregivers of racial/ethnic minority decedents were less likely to respond compared with caregivers of whites (19% and 22% RRs for black and Hispanic, respectively; 36% RR for whites).

      Variation and Predictors of Caregivers' Assessments

      All characteristics varied moderately or substantially across hospices as measured by ICC >0.01, except for decedent gender, caregiver gender, and caregiver age (results not shown). The characteristics with the largest variation across hospices were caregiver's language preference (ICC = 0.48 for Spanish survey or home language) and payer for hospice care (e.g., ICC = 0.42 for Medicare only; ICC = 0.35 for Medicare and private insurance).
      In bivariate models, all examined characteristics, except for caregiver gender and decedent gender, were significantly (P < 0.001) and strongly (regression coefficient greater than five points on a 0–100 scale) associated with respondent assessments of at least one outcome. Both caregiver and decedent education were significantly and strongly predictive of the outcomes. These two variables are moderately correlated with one another (R = 0.38); the coefficients for caregiver education were generally larger, indicating a stronger adjusted relationship with the outcomes. Because of the need to limit survey length in current and future versions of the survey, only caregiver education was retained in the model. We thus examined multivariate models including all characteristics except for caregiver gender, decedent gender, and decedent education. Table 2 presents the predictive results of the multivariate regression analyses for each of the eight selected items. Table 2 regression coefficients are on a 0–100 scale (i.e., the same scale as the measure scores). These results show that, for example, caregivers of decedents that had Medicaid as the payer for hospice care tended to respond more negatively than those of decedents with Medicare only and caregivers who were more highly educated tended to respond more negatively than caregivers with only a high school degree for most measures. Some predictors were significantly associated with several outcomes, but in different directions; for instance, a longer final hospice stay was associated with lower ratings for hospice team kept you informed about when they would arrive, a higher rating for getting training for restlessness or agitation, and had a U-shaped association with Rating of Hospice and Recommend Hospice, with caregivers of decedents with both short and long stays tending to give higher ratings than those with stays of an intermediate length. Spanish-speaking caregivers tended to give more positive responses than the reference group (primarily English speakers) for all measures except for got the right amount of emotional support from the hospice team after your family member died. In these multivariate models, each characteristic was significantly associated with at least one of the outcomes (P < 0.001).
      Several characteristics had a notable impact on at least one of the outcomes, as measured by a relative variance greater than .1 SD: payer for hospice care, caregiver education, and Spanish survey or home language (results provided in the Appendix). All other characteristics in the multivariate model had a moderate impact of at least .01 SD on at least one outcome.
      These results support a CMA model using the following variables: decedent age, payer for hospice care, primary diagnosis, length of final episode of hospice care, caregiver age, caregiver education, relationship to decedent, Spanish survey or home language, and response percentile.

      Impact of Adjustment

      Kendall's tau estimates for all the measures are shown in the first column of Table 4, summarizing the overall impact of the adjustment model. These range from 0.89 for getting hospice training and hospice team communication to 0.95 for recommend hospice, indicating that between 3% and 6% of hospice pairs would switch relative rankings with this adjustment model (e.g., [1–0.89]/2 = 6%).

      Zaslavsky AM. Issues in case-mix adjustment of measures of the quality of health plans. Dallas, TX: Proceedings of the Government and Social Statistics Sections. 1998.

      The effect sizes of the absolute value of the adjustments for quality measure scores were small, ranging from 0.06 to 0.14 hospice-level SDs (Table 4). The percentile rank shift associated with these adjustments ranged from 1.2 to 5.4 “percentile” points when considering hospices at the 50th–90th percentiles of hospice scores. For example, a hospice whose hospice team communication score was truly at the 50th percentile of hospice scores may be incorrectly ranked 5.4 points lower, at the 44.6th percentile, in the absence of appropriate CMA. Although these adjustments are relatively small on average, they can be very meaningful for hospices with very different case-mix distributions. For example, one hospice with a large proportion of responding caregivers whose decedents had Medicare and Medicaid as the payer for hospice care (51%) and a large proportion of highly educated responding caregivers (53% with more education than four-year college) would have a hospice team communication score at the 25th percentile of hospice scores without CMA, but a score at the 59th percentile if this case-mix was accounted for in adjustment (results not shown).
      Table 4Impact of CMA for Each Measure (N = 2514 Hospices)
      Quality MeasureKendall's Tau
      Kendall's tau is a measure of correlation between two sets of scores and is helpful in this setting because it can be interpreted as the proportion of hospice pairs whose relative rankings would be reversed by adjustment. Specifically, Kendall's tau ranges from −1 to 1; a −1 indicates perfect inverse correlation between unadjusted and adjusted scores, which would imply that adjustment had a very dramatic effect and completely reversed the rankings of all possible hospice pairs. Conversely, a one would indicate perfect correlation between unadjusted and adjusted scores, which would imply that adjustment had no effect and that no hospice pairs switched rankings as a result of adjustment. The percentage of hospice pairs that would switch rankings because of adjustment can be calculated as (1 − K)/2%, where K is Kendall's tau.
      Percent of Hospice Pairs That Would Switch Rankings Because of AdjustmentEffect Size of Average Absolute Adjustment (Hospice-Level SD)Percentile Rank Shift
      Expected percentile rank shift shifts for a hospice that was truly at the 50th percentile (second number in range) and a hospice that was truly at the 90th percentile of hospice scores for that quality measure (first number in range). For example, a hospice whose hospice team communication score was truly at the 50th percentile of hospice scores may be incorrectly ranked five points lower, at the 45th percentile, in the absence of appropriate case-mix adjustment.
      Composite measures
       Hospice team communication0.895.142.6–5.4
       Getting timely care0.924.101.9–4.0
       Treating family member with respect0.924.091.7–3.6
       Providing emotional support0.924.122.3–4.9
       Getting help for symptoms0.905.132.4–5.0
       Getting hospice care training
      Note that this measure is only calculated among those indicating on the survey that the decedent was in the home and/or assisted living facility.
      0.896.132.5–5.2
      Global measures
       Overall rating of hospice0.924.101.8–3.9
       Recommend hospice0.953.061.2–2.6
      a Kendall's tau is a measure of correlation between two sets of scores and is helpful in this setting because it can be interpreted as the proportion of hospice pairs whose relative rankings would be reversed by adjustment. Specifically, Kendall's tau ranges from −1 to 1; a −1 indicates perfect inverse correlation between unadjusted and adjusted scores, which would imply that adjustment had a very dramatic effect and completely reversed the rankings of all possible hospice pairs. Conversely, a one would indicate perfect correlation between unadjusted and adjusted scores, which would imply that adjustment had no effect and that no hospice pairs switched rankings as a result of adjustment. The percentage of hospice pairs that would switch rankings because of adjustment can be calculated as (1 − K)/2%, where K is Kendall's tau.
      b Expected percentile rank shift shifts for a hospice that was truly at the 50th percentile (second number in range) and a hospice that was truly at the 90th percentile of hospice scores for that quality measure (first number in range). For example, a hospice whose hospice team communication score was truly at the 50th percentile of hospice scores may be incorrectly ranked five points lower, at the 45th percentile, in the absence of appropriate case-mix adjustment.
      c Note that this measure is only calculated among those indicating on the survey that the decedent was in the home and/or assisted living facility.

      Discussion

      Similar to prior studies of patient experience surveys, we found that spouses and parent caregivers of decedents were more likely to respond; caregivers of racial/ethnic minority decedents and decedents who had Medicaid as the payer for hospice care were less likely to respond to the CAHPS Hospice Survey.
      • Elliott M.N.
      • Edwards C.
      • Angeles J.
      • Hambarsoomians K.
      • Hays R.D.
      Patterns of unit and item nonresponse in the CAHPS® Hospital Survey.
      • Kahn K.L.
      • Liu H.
      • Adams J.L.
      • et al.
      Methodological challenges associated with patient responses to follow-up longitudinal surveys regarding quality of care.
      • Klein D.J.
      • Elliott M.N.
      • Haviland A.M.
      • et al.
      Understanding nonresponse to the 2007 Medicare CAHPS survey.
      • McHorney C.A.
      • Kosinski M.
      • Ware Jr., J.E.
      Comparisons of the costs and quality of norms for the SF-36 health survey collected by mail versus telephone interview: results from a national survey.
      • Zaslavsky A.M.
      • Zaborski L.
      • Cleary P.D.
      Does the effect of respondent characteristics on consumer assessments vary across health plans?.
      • Casarett D.
      • Smith D.
      • Breslin S.
      • Richardson D.
      Does nonresponse bias the results of retrospective surveys of end-of-life care?.
      It will be important for future work to consider how survey efforts might be adapted to encourage greater participation from under-represented groups while also respecting that respondents in these subgroups might have unique difficulties that make survey participation particularly burdensome. Differential survey nonresponse can mean that respondents are not representative of the full eligible sample, and nonresponse bias can result if these differences are also related to the outcomes being measured.
      • Casarett D.
      • Smith D.
      • Breslin S.
      • Richardson D.
      Does nonresponse bias the results of retrospective surveys of end-of-life care?.
      • Groves R.M.
      Survey errors and survey costs.
      • Groves R.M.
      • Couper M.P.
      Nonresponse in household interview surveys.
      Case-mix adjusting for these characteristics can mitigate potential nonresponse bias with less loss of precision than nonresponse weighting.
      • Elliott M.N.
      • Zaslavsky A.M.
      • Goldstein E.
      • et al.
      Effects of survey mode, patient mix, and nonresponse on CAHPS® hospital survey scores.
      Public reporting of the National Quality Forum-endorsed CAHPS Hospice Survey measures, which began in February 2018, heightens the importance of CMA that levels the playing field for hospice comparisons. To ensure that publicly reported scores are comparable across hospices, one must adjust quality measure scores for characteristics that vary between hospices and influence how caregivers respond to survey questions but are not consequences of the hospice's care quality or otherwise controlled by the hospice. Our analyses identified several caregiver and decedent characteristics that meet these criteria. The most important adjustors were payer for hospice care, caregiver education, and survey language/language spoken at home. The association between Medicaid as a payer for hospice care and more negative ratings may reflect the more limited resources available to the hospice organization when Medicaid is the payer and/or more difficulty in caring for Medicaid beneficiaries. Respondent education and language are known to affect how people respond to survey questions, and in particular, use CAHPS response scales.
      • Elliott M.N.
      • Haviland A.M.
      • Kanouse D.E.
      • Hambarsoomian K.
      • Hays R.D.
      Adjusting for subgroup differences in extreme response tendency in ratings of health care: impact on disparity estimates.
      In the absence of CMA, hospice scores would be distorted, and this error can be large for hospices with unusual patient profiles.
      The hospice setting is unique in that characteristics of both caregiver respondents and hospice patients must be considered when adjusting hospice scores. Our results show that characteristics of both are indeed important and play a role in the evaluation of care received. In most other experience of care surveys, including HCAHPS, the respondent is also the patient. Interestingly, many of the respondent characteristics identified in our analyses as being associated with response patterns have exhibited similar patterns in HCAHPS. For example, in both settings, those who respond later to the survey and those with more education tend to respond more negatively.
      • Elliott M.N.
      • Zaslavsky A.M.
      • Goldstein E.
      • et al.
      Effects of survey mode, patient mix, and nonresponse on CAHPS® hospital survey scores.
      Although it is a characteristic of the decedent, our identification of primary diagnosis as a characteristic associated with response patterns parallels the use of service line (surgery, maternity, or other) in the adjustment of HCAHPS scores. Unlike education and language, which affect how people respond, it is reasonable to suggest that associations with diagnosis or service line are a reflection of differences in the inherent difficulty of treating patients with different needs and of characteristics that a patient brings to the setting and, thus, are not under the control of the hospice or hospital.
      Our study has several limitations. First, our analyses are restricted to hospices that were large enough to be required to participate in the CAHPS Hospice Survey and those that had at least 20 respondents for more than four quarters of survey administration. Therefore, our results may not be representative of all hospices nationally, particularly small hospices. In addition, most of the variables used in CMA and the data used to obtain measure scores are self-reported and are thus subject to recall bias, although any such bias should not affect comparisons of hospices. With public reporting of CAHPS Hospice Survey scores, CMA will be critical to ensuring that scores provide fair comparisons between hospices and that changes to patient or caregiver populations do not disrupt trends in scores over time.

      Disclosures and Acknowledgments

      The data collection and analysis on which this publication is based was sponsored under contract number HHSM-500-2016-00022G, entitled, “National Implementation of the CAHPS Hospice Survey,” funded by the Centers for Medicare and Medicaid Services, Department of Health and Human Services. The content of this publication neither necessarily reflect the views or policies of the Department of Health and Human Services nor does the mention of trade names, commercial products, or organizations imply endorsement by the U.S. government. The authors assume full responsibility for the accuracy and completeness of the ideas presented. The authors declare no conflicts of interest.

      Appendix

      Appendix TableImpact of Each Case-Mix Adjustors on Eight Selected Evaluative Questions (One Selected From Each Measure)
      Case-Mix AdjustorHospice Team Kept You Informed About When They Would Arrive
      RV is defined as the variance of adjustments for this variable relative to the variance of hospice scores fully adjusted using all included variables.
      Got Help From the Hospice Team as Soon as You Needed ItHospice Team Treated Your Family Member With Dignity and RespectGot the Right Amount of Emotional Support From the Hospice Team After Your Family Member DiedFamily Member Got Needed Help With PainHospice Team Gave You the Training You Needed About What to Do If Your Family Member Became Restless or Agitated
      Note that this question is only examined among those indicating on the survey that the decedent was in the home and/or assisted living facility.
      Overall Rating of HospiceRecommend Hospice
      RV
      Response percentile0.0260.0520.0290.0180.0290.0270.0540.052
      Decedent age0.0300.0060.0100.0190.0180.0200.0030.006
      Payer for hospice care0.1250.0310.0580.0270.0530.0400.0530.046
      Primary diagnosis0.0740.0120.0160.0220.0210.0230.0130.013
      Lengths of hospice stay0.0710.0670.0110.0330.0200.0190.0250.012
      Caregiver age0.0150.0190.0280.0380.0230.0370.0280.018
      Caregiver education0.1470.1200.0630.0210.0650.1030.0850.041
      Caregiver relationship0.0480.0100.0110.0450.0130.0240.0110.009
      Survey language/respondent's home language0.0530.0240.0500.1570.0830.0890.0610.037
      RV = relative variance.
      Bold text indicates notable impact, RV > 0.1.
      a RV is defined as the variance of adjustments for this variable relative to the variance of hospice scores fully adjusted using all included variables.
      b Note that this question is only examined among those indicating on the survey that the decedent was in the home and/or assisted living facility.

      References

      1. NHPCO facts and figures: Hospice care in America. National Hospice and Palliative Care Organization, Alexandria, VA2015
        • Stevenson D.G.
        • Dalton J.B.
        • Grabowski D.C.
        • Huskamp H.A.
        Nearly half of all Medicare hospice enrollees received care from agencies owned by regional or national chains.
        Health Aff. 2015; 34: 30-38
        • Stevenson D.G.
        • Grabowski D.C.
        • Keating N.L.
        • Huskamp H.A.
        Effect of ownership on hospice service use: 2005-2011.
        J Am Geriatr Soc. 2016; 64: 1024-1031
        • Prsic E.
        • Plotzke M.
        • Christian T.J.
        • Gozalo P.
        • Teno J.M.
        A national study of live hospice discharges between 2000 and 2012.
        J Palliat Med. 2016; 19: 987-990
        • Teno J.M.
        • Plotzke M.
        • Christian T.
        • Gozalo P.
        Examining variation in hospice visits by professional staff in the last 2 days of life.
        JAMA Intern Med. 2016; 176: 364-370
        • Zheng N.T.
        • Li Q.
        • Hanson L.C.
        • et al.
        Nationwide quality of hospice care: findings from the Centers for Medicare & Medicaid services hospice quality reporting program.
        J Pain Symptom Manag. 2017; 55: 427-432
      2. Price RA, Quigley DD, Bradley MA, et al. Hospice experience of care survey. 2014. Available at: https://www.rand.org/pubs/research_reports/RR657.html. Accessed June 10, 2018.

        • Centers for Medicaid & Medicare Services
        Hospice Quality Reporting Program (HQRP): Requirements for the Fiscal Year (FY) 2018 Reporting Year.
        (Available at:)
        • Rau J.
        Medicare to begin basing hospital payments on patient-satisfaction scores.
        Kaiser Health News, Washington, DC2011
        • Zaslavsky A.M.
        • Zaborski L.B.
        • Ding L.
        • Shaul J.A.
        Adjusting performance measures to ensure equitable plan comparisons.
        Health Care Financ Rev. 2001; 22: 109-126
        • National Academies of Sciences, Engineering, and Medicine
        Accounting for social risk factors in Medicare payment: Identifying social risk factors.
        The National Academies Press, Washington, DC2016
        • Price R.A.
        • Parast L.
        • Haas A.
        • Teno J.M.
        • Elliott M.N.
        Black and Hispanic patients receive hospice care similar to that of white patients when in the same hospices.
        Health Aff. 2017; 36: 1283-1290
      3. Consumer Assessment of Healthcare Providers and Systems (CAHPS) Hospice Quality Assurance Guidelines, Version 4.0; September 2017. Available at: http://www.hospicecahpssurvey.org/globalassets/hospice-cahps/quality-assurance-guidelines/cahps-hospice-survey-quality-assurance-guidelines-v4.0_september-2017.pdf. Accessed June 10, 2018.

        • Mayer L.A.
        • Elliott M.N.
        • Haas A.
        • Hays R.D.
        • Weinick R.M.
        Less use of extreme response options by Asians to standardized care scenarios may explain some racial/ethnic differences in CAHPS scores.
        Med Care. 2016; 54: 38-44
        • Elliott M.N.
        • Edwards W.S.
        • Klein D.J.
        • Heller A.
        Differences by survey language and mode among Chinese respondents to a CAHPS health plan survey.
        Public Opin Q. 2012; 76: 238-264
        • Barron D.N.
        • West E.
        • Reeves R.
        • Hawkes D.
        It takes patience and persistence to get negative feedback about patients' experiences: a secondary analysis of national inpatient survey data.
        BMC Health Serv Res. 2014; 14: 153
        • Zaslavsky A.M.
        • Zaborski L.B.
        • Cleary P.D.
        Factors affecting response rates to the consumer assessment of Health Plans Study survey.
        Med Care. 2002; 40: 485-499
        • Elliott M.N.
        • Zaslavsky A.M.
        • Goldstein E.
        • et al.
        Effects of survey mode, patient mix, and nonresponse on CAHPS® hospital survey scores.
        Health Serv Res. 2009; 44: 501-518
        • Saunders C.L.
        • Elliott M.N.
        • Lyratzopoulos G.
        • Abel G.A.
        Do differential response rates to patient surveys between organizations lead to unfair performance comparisons?: Evidence from the English Cancer Patient Experience Survey.
        Med Care. 2016; 54: 45
        • Parast L.
        • Elliott M.N.
        • Hambarsoomian K.
        • Teno J.
        • Anhang Price R.
        Effects of survey mode on consumer assessment of Healthcare Providers and Systems (CAHPS) hospice survey scores.
        J Am Geriatr Soc. 2018; 66: 546-552
        • Teno J.M.
        • Clarridge B.R.
        • Casey V.
        • et al.
        Family perspectives on end-of-life care at the last place of care.
        JAMA. 2004; 291: 88-93
        • Wachterman M.W.
        • Marcantonio E.R.
        • Davis R.B.
        • McCarthy E.P.
        Association of hospice agency profit status with patient diagnosis, location of care, and length of stay.
        JAMA. 2011; 305: 472-479
      4. Compliance Program Guidance for Nursing Facilities (2008).
        (Available at:)
      5. AAPOR. American Association for Public Opinion Research Standard Definitions: Final Dispositions of Case Codes and Outcome Rates for Surveys. 2016. Available at: https://www.aapor.org/AAPOR_Main/media/publications/Standard-Definitions20169theditionfinal.pdf. Accessed June 10, 2018.

      6. Zaslavsky AM. Issues in case-mix adjustment of measures of the quality of health plans. Dallas, TX: Proceedings of the Government and Social Statistics Sections. 1998.

        • Elliott M.N.
        • Edwards C.
        • Angeles J.
        • Hambarsoomians K.
        • Hays R.D.
        Patterns of unit and item nonresponse in the CAHPS® Hospital Survey.
        Health Serv Res. 2005; 40: 2096-2119
        • Kahn K.L.
        • Liu H.
        • Adams J.L.
        • et al.
        Methodological challenges associated with patient responses to follow-up longitudinal surveys regarding quality of care.
        Health Serv Res. 2003; 38: 1579-1598
        • Klein D.J.
        • Elliott M.N.
        • Haviland A.M.
        • et al.
        Understanding nonresponse to the 2007 Medicare CAHPS survey.
        Gerontologist. 2011; 51: 843-855
        • McHorney C.A.
        • Kosinski M.
        • Ware Jr., J.E.
        Comparisons of the costs and quality of norms for the SF-36 health survey collected by mail versus telephone interview: results from a national survey.
        Med Care. 1994; 32: 551-567
        • Zaslavsky A.M.
        • Zaborski L.
        • Cleary P.D.
        Does the effect of respondent characteristics on consumer assessments vary across health plans?.
        Med Care Res Rev. 2000; 57: 379-394
        • Casarett D.
        • Smith D.
        • Breslin S.
        • Richardson D.
        Does nonresponse bias the results of retrospective surveys of end-of-life care?.
        J Am Geriatr Soc. 2010; 58: 2381-2386
        • Groves R.M.
        Survey errors and survey costs.
        John Wiley & Sons, Hoboken, NJ2004
        • Groves R.M.
        • Couper M.P.
        Nonresponse in household interview surveys.
        John Wiley & Sons, New York2012
        • Elliott M.N.
        • Haviland A.M.
        • Kanouse D.E.
        • Hambarsoomian K.
        • Hays R.D.
        Adjusting for subgroup differences in extreme response tendency in ratings of health care: impact on disparity estimates.
        Health Serv Res. 2009; 44: 542-561