ORIGINAL ARTICLE
Construct Validity and Factor Structure of Survey-based Assessment of Cost-related Medication Burden Mehmet Burcu, MS,* G. Caleb Alexander, MD, MS,w Xinyi Ng, BSPharm,* and Donna Harrington, PhDz
Background: Millions of Americans are burdened by out-of-pocket prescription costs. Although many survey measures have been developed to assess this burden, the construct validity and the factor structure of these instruments have not been rigorously assessed. Objectives: To characterize the factor structure and the construct validity of items assessing cost-related medication burden. Methods: We applied exploratory factor and confirmatory factor analyses to the 2009 Medicare Current Beneficiary Survey, focusing on 10 items assessing cost-related mediation burden among a nationally representative sample of community-dwelling Medicare beneficiaries. The fit of competing models was compared using several indices. Results: The study population (N = 8777) was predominantly aged over 65 years (83.3%), female (54.4%), and white (84.3%). Two distinct factors were present for the medication cost-reduction strategies: (1) cost-related medication nonadherence and (2) drugshopping behaviors, not directly impacting medication compliance. The two factors were moderately correlated (r = 0.55), highlighting the presence of a 2 distinct but related constructs for cost-related medication burden. An item assessing the use of mail or internet pharmacies did not load well on either factor and may not necessarily measure medication-related cost burden. An item assessing reduced spending on basic needs loaded strongly on the same factor with the cost-related medication nonadherence items, suggesting they together may represent extreme compensatory behaviors that may adversely affect health outcomes. Conclusions: Two distinct constructs were derived from these items examining cost-related medication burden. Although costrelated medication burden is often associated with nonadherence,
From the *Pharmaceutical Health Services Research Department, University of Maryland; wDepartment of Epidemiology and Medicine, Center for Drug Safety and Effectiveness, Johns Hopkins University; and zSchool of Social Work, University of Maryland, Baltimore, MD. The authors declare no conflict of interest. Reprints: Mehmet Burcu, MS, Pharmaceutical Health Services Research Department, University of Maryland, Baltimore, 220 Arch Street, 12th Floor, Room #411-G, Baltimore, MD 21201. E-mail: mburc001@ umaryland.edu. Supplemental Digital Content is available for this article. Direct URL citations appear in the printed text and are provided in the HTML and PDF versions of this article on the journal’s Website, www.lwwmedicalcare.com. Copyright r 2014 Wolters Kluwer Health, Inc. All rights reserved. ISSN: 0025-7079/15/5302-0199
Medical Care
Volume 53, Number 2, February 2015
drug-shopping behaviors that do not directly impact adherence are also important measure of this burden. Key Words: cost-related medication burden, medication nonadherence, validity, factor structure, Medicare (Med Care 2015;53: 199–206)
M
illions of Americans are burdened by their out-of-pocket prescription costs. Such costs have been associated with deleterious cost-coping behaviors such as reduced adherence and increased risk of adverse health outcomes and unintended healthcare expenditures.1–7 Although high out-of-pocket costs may burden the young and old alike, they can pose particular hardship for the elderly, given that Medicare beneficiaries often have limited fixed incomes and multiple comorbid conditions requiring chronic medication treatment. In January 2006, the Medicare Part D program was implemented, the largest change to the Medicare Program since its inception. With the Part D program, Medicare beneficiaries can purchase prescription drug insurance, which has improved coverage and enhanced adherence to prescribed medicines.8–11 Despite this, many seniors, in particular those with multiple morbidities and lower incomes, have continued to face coverage limits such as those imposed by the “doughnut hole.”6,8,12–15 The burden from out-of-pocket costs may impact individuals in a variety of ways. For example, cost pressures may lead some patients to cut back on their medications or basic necessities such as utilities or food, whereas others may experience difficulty with self-care and health management.7,16–18 In light of these different coping strategies and compensatory behaviors, a variety of measures has been developed to evaluate individuals’ burden. Since 2004, the breadth of these items has been reflected in the Medicare Current Beneficiary Survey (MCBS), a nationally representative survey of the aged, disabled, and institutionalized Medicare beneficiaries.19 Four of these items—focused on behaviors such as taking smaller doses or skipping doses of prescribed medicines and delaying to fill or not filling prescribed medicines—have been previously validated and widely used to measure cost-related medication nonadherence.2,8,20–25 By contrast, an additional 6 items within the MCBS may or may not reflect changes in medication adherence but rather measure a variety of cost-coping strategies ranging from spending less on basic supplies to comparing medication prices and purchasing medications from the postal mail or internet. www.lww-medicalcare.com |
199
Medical Care
Burcu et al
Dozens of studies have used one or some combination of these 10 measures characterizing medication cost-reduction strategies (MCRS), both based on data from the MCBS and surveys using these items in other contexts.1,2,4,6,8,26–32 Although the reliability and the consistency of these items have been previously explored,33 to our knowledge, no prior investigation of the construct validity of these 10 items together has been undertaken. Such information is particularly valuable given recent policy changes in the United States as a consequence of the Patient Protection and Affordable Care Act. Thus, we sought to evaluate the construct validity and the factor structure of the MCRS items by using exploratory factor analysis (EFA) and confirmatory factor analysis (CFA) among community-dwelling Medicare beneficiaries.
METHODS Data Source We used cross-sectional data from the 2009 MCBS, which is a rotating panel survey conducted by the Centers for Medicare and Medicaid Services.19 The MCBS interviews approximately 12,000 Medicare beneficiaries annually, and provides comprehensive demographic, health status and functioning, healthcare access, and utilization information.19 The study sample was restricted to community-dwelling Medicare beneficiaries enrolled in fee-for-service or managed care programs. We excluded beneficiaries who did not complete the survey due to any reason or had any residence in a long-term care facility. The study was received and approved by the Institutional Review Board at the University of Maryland, Baltimore.
Measures Medication Cost-reduction Strategies In 2004 and 2006, new measures were included in the MCBS to assess medication cost-reduction strategies. Costreduction medication strategies were measured based on the response to the 10 survey questions ascertaining whether respondents have: (1) taken smaller doses of a medicine to make it last longer, (2) skipped doses to make the medicine last longer, (3) decided not to fill or refill a prescription because it was too expensive, (4) delayed to fill or refill a prescription because it was too expensive, (5) spent less money on food, heat, or other basic needs so that the respondent would have money for medicine, (6) asked for the generic form of the prescription, (7) purchased prescription medications by mail or internet, (8) obtained or asked for prescription medicine samples from a healthcare provider, (9) compared prices or shopped for best prices of prescription medicines, and (10) purchased prescription medicines from another country. These items comprised 3-level ordinal response options including “never,” “sometimes,” and “often.” Items that were not ascertained or items for which beneficiaries responded “don’t know” or refused to respond were categorized as missing.
Demographic, Clinical, Administrative Characteristics
Volume 53, Number 2, February 2015
ital status, metropolitan versus nonmetropolitan region of residence, a composite activities of daily living limitation score, a composite functional limitation score, a composite morbidity score adapted from a previous study,8 and healthcare payment type.
Data Analysis and Statistical Methods We used simple random sampling to divide the beneficiaries into 2 analytical samples for the EFA and CFA. We assessed for failure of randomization by using the Pearson w2 tests to compare the 2 groups with respect to their characteristics and the distribution of the response options for each of our 10 cost-related burden items. To examine the factor structure of the 10 MCRS items, we used EFA using geomin rotation to allow correlations among factors, and also using the mean- and variance-adjusted weighted least squares (WLSMV) estimator due to the 3-level ordinal response nature of these MCRS items.34 Of note, we used pairwise deletion to handle missing values on the MCRS items for the EFA and CFA using the WLSMV estimator.35 We hypothesized that there were at least 2 factors describing the 10 MCRS items: (1) cost-related medication nonadherence, corresponding to cost-reduction strategies impacting medication adherence, and (2) drug price shopping, corresponding to medication acquisition cost-reduction strategies without directly impacting medication compliance. As per Kline’s recommendations, a unidirectional structure (1-factor) was also tested.36 In addition, to allow for further exploration, a 3-factor structure was also examined. The number and structure of factors were determined using the eigenvalues, the scree plot, the factor loadings, and the interpretation of factor structures.37 A factor loading of 0.400 or more was required for an item to be identified as loading on a factor.38,39 To substantiate the factor structures generated from the EFA, we conducted CFA using the WLSMV estimator due to the 3-level ordinal response nature of the MCRS items.34 To assess model fit in the EFA and CFA, we used several indices to compare the fit of the competing models including w2 goodness-of-fit index, comparative fit index (CFI), Tucker-Lewis Index (TLI), root mean square error of approximation (RMSEA), the standardized root mean square residual (SRMR; for EFA only), and the weighted root mean square residual (WRMR; for CFA only). A statistically nonsignificant w2 test, CFI value of 0.95 or greater, TLI value of 0.95 or greater, RMSEA value of 0.06 or less, WRMR value of 1.0 or less, and SRMR value of 0.08 or less suggest a well-fitting model.36,40 We used SAS version 9.3 (SAS Institute Inc., Cary, NC) for operationalization of the study measures and bivariate analyses, and we used Mplus Version 7.1 (Muthe´n & Muthe´n, Los Angeles, CA) to conduct our factor analyses.
RESULTS Characteristics of the Study Sample of Medicare Beneficiaries
Study variables for beneficiary characteristics included age group, sex, race/ethnicity, educational attainment, mar-
Among 8777 community-dwelling Medicare beneficiaries, 4416 (50.3%) were randomized to be in the EFA sample and 4361 (49.7%) were randomized to be in the CFA sample. There were no statistically significant differences
200 | www.lww-medicalcare.com
Copyright
r
2014 Wolters Kluwer Health, Inc. All rights reserved.
Medical Care
Volume 53, Number 2, February 2015
Cost-related Medication Burden
between the analytical samples for the study covariates examined (Table 1), and also in the distribution of item responses (Online Appendix 1, Supplemental Digital Content 1, http://links.lww.com/MLR/A841). The sample was predominantly over 65 years of age (83.3%), female (54.4%), white (84.3%), and enrolled in fee-for-service programs. Although most beneficiaries did not have any limitations
related to activities of daily living (69.1%), more than two fifths (42.9%) had 3 or more functional limitations, and one third had Z4 categories of morbidity.
EFA Both the scree plot and the Kaiser’s eigenvaluesgreater-than-one criterion suggested that 2 factors should be
TABLE 1. Sociodemographic and Administrative Characteristics of the Study Sample by Analysis Type, the Medicare Current Beneficiary Survey, 2009, N = 8777 N (%) Characteristics Age (y) r64 65–69 70–74 75–79 Z80 Sex Male Female Race/ethnicity White African American Hispanic Otherw Educationz < High school High school > High school Marital statusy Married Widowed Divorced/separated Never married Region Metro (SMSA) Other8 Payment type Fee-for-service Managed care ADL limitation scorez 0 1–2 Z3 Functional limitation score# 0 1–2 Z3 No. morbidities** 0–1 2–3 Z4
Total Sample (N = 8777)
Exploratory Factor Analysis (N = 4416)
Confirmatory Factor Analysis (N = 4361)
P* 0.33
1469 1219 1757 1476 2856
(16.7) (13.9) (20.0) (16.8) (32.5)
723 637 907 732 1417
(16.4) (14.4) (20.5) (16.6) (32.1)
746 582 850 744 1439
(17.1) (13.4) (19.5) (17.1) (33.0) 0.67
4005 (45.6) 4772 (54.4)
2005 (45.4) 2411 (54.6)
2000 (45.9) 2361 (54.1)
7396 909 225 247
3721 464 108 123
3675 445 117 124
0.87 (84.3) (10.4) (2.6) (2.8)
(84.3) (10.5) (2.5) (2.8)
(84.3) (10.2) (2.7) (2.8) 0.12
2278 (26.0) 2736 (31.2) 3719 (42.4)
1181 (26.8) 1339 (30.4) 1880 (42.7)
1097 (25.3) 1397 (32.2) 1839 (42.4)
4246 2625 1131 770
2156 1320 555 382
2090 1305 576 388
0.75 (48.4) (29.9) (12.9) (8.8)
(48.9) (29.9) (12.6) (8.7)
(48.0) (29.9) (13.2) (8.9) 0.90
6478 (73.8) 2299 (26.2)
3262 (73.9) 1154 (26.1)
3216 (73.7) 1145 (26.3)
6579 (75.0) 2198 (25.0)
3300 (74.7) 1116 (25.3)
3279 (75.2) 1082 (24.8)
6063 (69.1) 1854 (21.1) 853 (9.7)
3072 (69.6) 919 (20.8) 423 (9.6)
2991 (68.7) 935 (21.5) 430 (9.9)
1757 (20.0) 3251 (37.0) 3762 (42.9)
897 (20.3) 1654 (37.5) 1863 (42.2)
860 (19.7) 1597 (36.7) 1899 (43.6)
1766 (20.1) 3853 (43.9) 3158 (36.0)
901 (20.4) 1904 (43.1) 1611 (36.5)
865 (19.8) 1949 (44.7) 1547 (35.5)
0.62 0.64
0.42
0.33
w
Other (race) includes participants of other races and unknown. 44 participants had missing values for education. 5 participants had missing values for marital status. 8 Other (region) includes nonmetropolitan areas and unknown. z 7 participants had missing values for ADL limitation score, which is a composite variable ranging from 0 to 6 representing the limitations on the following ADL(s) with equal weights: (a) bathing with sponge, bath, or shower; (b) dressing; (c) toilet use; (d) transferring (in and out of bed or chair); (e) urine and bowel continence; and (f) eating. # 7 participants had missing values for functional limitation score, which is a composite variable ranging from 0 to 5 representing the limitations on the following functions with equal weights: (a) stooping/crouching/kneeling, (b) lifting/carrying 10 pounds, (c) difficulty extending arms above shoulders, (d) difficulty writing/handling object, and (e) difficulty walking one-fourth miles or 2–3 blocks. **Morbidity categories include cardiac disease, hypertension, cerebrovascular disease, lung disease, cancer, diabetes, arthritis, psychiatric disorder or depression, dementia or other neurological conditions, and osteoporosis or bone fracture. ADL indicates activities of daily living; SMSA, Standard Metropolitan Statistical Area. *Pearson w2 test P-value. z y
Copyright
r
2014 Wolters Kluwer Health, Inc. All rights reserved.
www.lww-medicalcare.com |
201
202 | www.lww-medicalcare.com
0.048 0.184w 0.101 0.005 0.428* 0.596* 0.331w 0.589* 0.715* 0.490* 0.001 0.221w 0.021 0.075 0.007 0.602* 0.326w 0.585* 0.740* 0.531*
0.428* 0.411* 0.045 0.067 0.019 0.009 0.013 0.134w 0.003 0.135w 1.066* 1.080* 0.801* 0.951* 0.706* 0.007 0.021 0.191w 0.119w 0.172w
3 2
3-factor
1 r
0.956* 0.969* 0.781* 0.866* 0.667*
0.009 0.278w 0.033 0.032 0.184
0.947* 0.952* 0.877* 0.921* 0.751*
0.438* 0.013 0.463* 0.554* 0.257w
1
1-factor
1
2-factor
2
Copyright
Volume 53, Number 2, February 2015
EFA item loadings, N = 4381. *Items with statistically significant (at 5% alpha level) geomin rotated loadings that are greater than 0.400. w Items with statistically significant (at 5% alpha level) geomin rotated loadings that are
Construct Validity and Factor Structure of Survey-based Assessment of Cost-related Medication Burden Mehmet Burcu, MS,* G. Caleb Alexander, MD, MS,w Xinyi Ng, BSPharm,* and Donna Harrington, PhDz
Background: Millions of Americans are burdened by out-of-pocket prescription costs. Although many survey measures have been developed to assess this burden, the construct validity and the factor structure of these instruments have not been rigorously assessed. Objectives: To characterize the factor structure and the construct validity of items assessing cost-related medication burden. Methods: We applied exploratory factor and confirmatory factor analyses to the 2009 Medicare Current Beneficiary Survey, focusing on 10 items assessing cost-related mediation burden among a nationally representative sample of community-dwelling Medicare beneficiaries. The fit of competing models was compared using several indices. Results: The study population (N = 8777) was predominantly aged over 65 years (83.3%), female (54.4%), and white (84.3%). Two distinct factors were present for the medication cost-reduction strategies: (1) cost-related medication nonadherence and (2) drugshopping behaviors, not directly impacting medication compliance. The two factors were moderately correlated (r = 0.55), highlighting the presence of a 2 distinct but related constructs for cost-related medication burden. An item assessing the use of mail or internet pharmacies did not load well on either factor and may not necessarily measure medication-related cost burden. An item assessing reduced spending on basic needs loaded strongly on the same factor with the cost-related medication nonadherence items, suggesting they together may represent extreme compensatory behaviors that may adversely affect health outcomes. Conclusions: Two distinct constructs were derived from these items examining cost-related medication burden. Although costrelated medication burden is often associated with nonadherence,
From the *Pharmaceutical Health Services Research Department, University of Maryland; wDepartment of Epidemiology and Medicine, Center for Drug Safety and Effectiveness, Johns Hopkins University; and zSchool of Social Work, University of Maryland, Baltimore, MD. The authors declare no conflict of interest. Reprints: Mehmet Burcu, MS, Pharmaceutical Health Services Research Department, University of Maryland, Baltimore, 220 Arch Street, 12th Floor, Room #411-G, Baltimore, MD 21201. E-mail: mburc001@ umaryland.edu. Supplemental Digital Content is available for this article. Direct URL citations appear in the printed text and are provided in the HTML and PDF versions of this article on the journal’s Website, www.lwwmedicalcare.com. Copyright r 2014 Wolters Kluwer Health, Inc. All rights reserved. ISSN: 0025-7079/15/5302-0199
Medical Care
Volume 53, Number 2, February 2015
drug-shopping behaviors that do not directly impact adherence are also important measure of this burden. Key Words: cost-related medication burden, medication nonadherence, validity, factor structure, Medicare (Med Care 2015;53: 199–206)
M
illions of Americans are burdened by their out-of-pocket prescription costs. Such costs have been associated with deleterious cost-coping behaviors such as reduced adherence and increased risk of adverse health outcomes and unintended healthcare expenditures.1–7 Although high out-of-pocket costs may burden the young and old alike, they can pose particular hardship for the elderly, given that Medicare beneficiaries often have limited fixed incomes and multiple comorbid conditions requiring chronic medication treatment. In January 2006, the Medicare Part D program was implemented, the largest change to the Medicare Program since its inception. With the Part D program, Medicare beneficiaries can purchase prescription drug insurance, which has improved coverage and enhanced adherence to prescribed medicines.8–11 Despite this, many seniors, in particular those with multiple morbidities and lower incomes, have continued to face coverage limits such as those imposed by the “doughnut hole.”6,8,12–15 The burden from out-of-pocket costs may impact individuals in a variety of ways. For example, cost pressures may lead some patients to cut back on their medications or basic necessities such as utilities or food, whereas others may experience difficulty with self-care and health management.7,16–18 In light of these different coping strategies and compensatory behaviors, a variety of measures has been developed to evaluate individuals’ burden. Since 2004, the breadth of these items has been reflected in the Medicare Current Beneficiary Survey (MCBS), a nationally representative survey of the aged, disabled, and institutionalized Medicare beneficiaries.19 Four of these items—focused on behaviors such as taking smaller doses or skipping doses of prescribed medicines and delaying to fill or not filling prescribed medicines—have been previously validated and widely used to measure cost-related medication nonadherence.2,8,20–25 By contrast, an additional 6 items within the MCBS may or may not reflect changes in medication adherence but rather measure a variety of cost-coping strategies ranging from spending less on basic supplies to comparing medication prices and purchasing medications from the postal mail or internet. www.lww-medicalcare.com |
199
Medical Care
Burcu et al
Dozens of studies have used one or some combination of these 10 measures characterizing medication cost-reduction strategies (MCRS), both based on data from the MCBS and surveys using these items in other contexts.1,2,4,6,8,26–32 Although the reliability and the consistency of these items have been previously explored,33 to our knowledge, no prior investigation of the construct validity of these 10 items together has been undertaken. Such information is particularly valuable given recent policy changes in the United States as a consequence of the Patient Protection and Affordable Care Act. Thus, we sought to evaluate the construct validity and the factor structure of the MCRS items by using exploratory factor analysis (EFA) and confirmatory factor analysis (CFA) among community-dwelling Medicare beneficiaries.
METHODS Data Source We used cross-sectional data from the 2009 MCBS, which is a rotating panel survey conducted by the Centers for Medicare and Medicaid Services.19 The MCBS interviews approximately 12,000 Medicare beneficiaries annually, and provides comprehensive demographic, health status and functioning, healthcare access, and utilization information.19 The study sample was restricted to community-dwelling Medicare beneficiaries enrolled in fee-for-service or managed care programs. We excluded beneficiaries who did not complete the survey due to any reason or had any residence in a long-term care facility. The study was received and approved by the Institutional Review Board at the University of Maryland, Baltimore.
Measures Medication Cost-reduction Strategies In 2004 and 2006, new measures were included in the MCBS to assess medication cost-reduction strategies. Costreduction medication strategies were measured based on the response to the 10 survey questions ascertaining whether respondents have: (1) taken smaller doses of a medicine to make it last longer, (2) skipped doses to make the medicine last longer, (3) decided not to fill or refill a prescription because it was too expensive, (4) delayed to fill or refill a prescription because it was too expensive, (5) spent less money on food, heat, or other basic needs so that the respondent would have money for medicine, (6) asked for the generic form of the prescription, (7) purchased prescription medications by mail or internet, (8) obtained or asked for prescription medicine samples from a healthcare provider, (9) compared prices or shopped for best prices of prescription medicines, and (10) purchased prescription medicines from another country. These items comprised 3-level ordinal response options including “never,” “sometimes,” and “often.” Items that were not ascertained or items for which beneficiaries responded “don’t know” or refused to respond were categorized as missing.
Demographic, Clinical, Administrative Characteristics
Volume 53, Number 2, February 2015
ital status, metropolitan versus nonmetropolitan region of residence, a composite activities of daily living limitation score, a composite functional limitation score, a composite morbidity score adapted from a previous study,8 and healthcare payment type.
Data Analysis and Statistical Methods We used simple random sampling to divide the beneficiaries into 2 analytical samples for the EFA and CFA. We assessed for failure of randomization by using the Pearson w2 tests to compare the 2 groups with respect to their characteristics and the distribution of the response options for each of our 10 cost-related burden items. To examine the factor structure of the 10 MCRS items, we used EFA using geomin rotation to allow correlations among factors, and also using the mean- and variance-adjusted weighted least squares (WLSMV) estimator due to the 3-level ordinal response nature of these MCRS items.34 Of note, we used pairwise deletion to handle missing values on the MCRS items for the EFA and CFA using the WLSMV estimator.35 We hypothesized that there were at least 2 factors describing the 10 MCRS items: (1) cost-related medication nonadherence, corresponding to cost-reduction strategies impacting medication adherence, and (2) drug price shopping, corresponding to medication acquisition cost-reduction strategies without directly impacting medication compliance. As per Kline’s recommendations, a unidirectional structure (1-factor) was also tested.36 In addition, to allow for further exploration, a 3-factor structure was also examined. The number and structure of factors were determined using the eigenvalues, the scree plot, the factor loadings, and the interpretation of factor structures.37 A factor loading of 0.400 or more was required for an item to be identified as loading on a factor.38,39 To substantiate the factor structures generated from the EFA, we conducted CFA using the WLSMV estimator due to the 3-level ordinal response nature of the MCRS items.34 To assess model fit in the EFA and CFA, we used several indices to compare the fit of the competing models including w2 goodness-of-fit index, comparative fit index (CFI), Tucker-Lewis Index (TLI), root mean square error of approximation (RMSEA), the standardized root mean square residual (SRMR; for EFA only), and the weighted root mean square residual (WRMR; for CFA only). A statistically nonsignificant w2 test, CFI value of 0.95 or greater, TLI value of 0.95 or greater, RMSEA value of 0.06 or less, WRMR value of 1.0 or less, and SRMR value of 0.08 or less suggest a well-fitting model.36,40 We used SAS version 9.3 (SAS Institute Inc., Cary, NC) for operationalization of the study measures and bivariate analyses, and we used Mplus Version 7.1 (Muthe´n & Muthe´n, Los Angeles, CA) to conduct our factor analyses.
RESULTS Characteristics of the Study Sample of Medicare Beneficiaries
Study variables for beneficiary characteristics included age group, sex, race/ethnicity, educational attainment, mar-
Among 8777 community-dwelling Medicare beneficiaries, 4416 (50.3%) were randomized to be in the EFA sample and 4361 (49.7%) were randomized to be in the CFA sample. There were no statistically significant differences
200 | www.lww-medicalcare.com
Copyright
r
2014 Wolters Kluwer Health, Inc. All rights reserved.
Medical Care
Volume 53, Number 2, February 2015
Cost-related Medication Burden
between the analytical samples for the study covariates examined (Table 1), and also in the distribution of item responses (Online Appendix 1, Supplemental Digital Content 1, http://links.lww.com/MLR/A841). The sample was predominantly over 65 years of age (83.3%), female (54.4%), white (84.3%), and enrolled in fee-for-service programs. Although most beneficiaries did not have any limitations
related to activities of daily living (69.1%), more than two fifths (42.9%) had 3 or more functional limitations, and one third had Z4 categories of morbidity.
EFA Both the scree plot and the Kaiser’s eigenvaluesgreater-than-one criterion suggested that 2 factors should be
TABLE 1. Sociodemographic and Administrative Characteristics of the Study Sample by Analysis Type, the Medicare Current Beneficiary Survey, 2009, N = 8777 N (%) Characteristics Age (y) r64 65–69 70–74 75–79 Z80 Sex Male Female Race/ethnicity White African American Hispanic Otherw Educationz < High school High school > High school Marital statusy Married Widowed Divorced/separated Never married Region Metro (SMSA) Other8 Payment type Fee-for-service Managed care ADL limitation scorez 0 1–2 Z3 Functional limitation score# 0 1–2 Z3 No. morbidities** 0–1 2–3 Z4
Total Sample (N = 8777)
Exploratory Factor Analysis (N = 4416)
Confirmatory Factor Analysis (N = 4361)
P* 0.33
1469 1219 1757 1476 2856
(16.7) (13.9) (20.0) (16.8) (32.5)
723 637 907 732 1417
(16.4) (14.4) (20.5) (16.6) (32.1)
746 582 850 744 1439
(17.1) (13.4) (19.5) (17.1) (33.0) 0.67
4005 (45.6) 4772 (54.4)
2005 (45.4) 2411 (54.6)
2000 (45.9) 2361 (54.1)
7396 909 225 247
3721 464 108 123
3675 445 117 124
0.87 (84.3) (10.4) (2.6) (2.8)
(84.3) (10.5) (2.5) (2.8)
(84.3) (10.2) (2.7) (2.8) 0.12
2278 (26.0) 2736 (31.2) 3719 (42.4)
1181 (26.8) 1339 (30.4) 1880 (42.7)
1097 (25.3) 1397 (32.2) 1839 (42.4)
4246 2625 1131 770
2156 1320 555 382
2090 1305 576 388
0.75 (48.4) (29.9) (12.9) (8.8)
(48.9) (29.9) (12.6) (8.7)
(48.0) (29.9) (13.2) (8.9) 0.90
6478 (73.8) 2299 (26.2)
3262 (73.9) 1154 (26.1)
3216 (73.7) 1145 (26.3)
6579 (75.0) 2198 (25.0)
3300 (74.7) 1116 (25.3)
3279 (75.2) 1082 (24.8)
6063 (69.1) 1854 (21.1) 853 (9.7)
3072 (69.6) 919 (20.8) 423 (9.6)
2991 (68.7) 935 (21.5) 430 (9.9)
1757 (20.0) 3251 (37.0) 3762 (42.9)
897 (20.3) 1654 (37.5) 1863 (42.2)
860 (19.7) 1597 (36.7) 1899 (43.6)
1766 (20.1) 3853 (43.9) 3158 (36.0)
901 (20.4) 1904 (43.1) 1611 (36.5)
865 (19.8) 1949 (44.7) 1547 (35.5)
0.62 0.64
0.42
0.33
w
Other (race) includes participants of other races and unknown. 44 participants had missing values for education. 5 participants had missing values for marital status. 8 Other (region) includes nonmetropolitan areas and unknown. z 7 participants had missing values for ADL limitation score, which is a composite variable ranging from 0 to 6 representing the limitations on the following ADL(s) with equal weights: (a) bathing with sponge, bath, or shower; (b) dressing; (c) toilet use; (d) transferring (in and out of bed or chair); (e) urine and bowel continence; and (f) eating. # 7 participants had missing values for functional limitation score, which is a composite variable ranging from 0 to 5 representing the limitations on the following functions with equal weights: (a) stooping/crouching/kneeling, (b) lifting/carrying 10 pounds, (c) difficulty extending arms above shoulders, (d) difficulty writing/handling object, and (e) difficulty walking one-fourth miles or 2–3 blocks. **Morbidity categories include cardiac disease, hypertension, cerebrovascular disease, lung disease, cancer, diabetes, arthritis, psychiatric disorder or depression, dementia or other neurological conditions, and osteoporosis or bone fracture. ADL indicates activities of daily living; SMSA, Standard Metropolitan Statistical Area. *Pearson w2 test P-value. z y
Copyright
r
2014 Wolters Kluwer Health, Inc. All rights reserved.
www.lww-medicalcare.com |
201
202 | www.lww-medicalcare.com
0.048 0.184w 0.101 0.005 0.428* 0.596* 0.331w 0.589* 0.715* 0.490* 0.001 0.221w 0.021 0.075 0.007 0.602* 0.326w 0.585* 0.740* 0.531*
0.428* 0.411* 0.045 0.067 0.019 0.009 0.013 0.134w 0.003 0.135w 1.066* 1.080* 0.801* 0.951* 0.706* 0.007 0.021 0.191w 0.119w 0.172w
3 2
3-factor
1 r
0.956* 0.969* 0.781* 0.866* 0.667*
0.009 0.278w 0.033 0.032 0.184
0.947* 0.952* 0.877* 0.921* 0.751*
0.438* 0.013 0.463* 0.554* 0.257w
1
1-factor
1
2-factor
2
Copyright
Volume 53, Number 2, February 2015
EFA item loadings, N = 4381. *Items with statistically significant (at 5% alpha level) geomin rotated loadings that are greater than 0.400. w Items with statistically significant (at 5% alpha level) geomin rotated loadings that are
