Medical and nursing home costs: From cognitively unimpaired through dementia

Abstract

INTRODUCTION:

Efforts to model the cost-effectiveness of managing/modifying cognitive-impairment lack reliable, objective, baseline medical- and nursing-home-costs.

METHODS:

A stratified-random sample of Olmsted County, MN, residents age 70–89 years (N=3,545), well-characterized as cognitively-unimpaired, mild-cognitive-impairment, or dementia, were followed forward ≤1-year in provider-linked billing-data and Centers-for-Medicare-and-Medicaid-Services nursing-home assessments. Direct medical/nursing-home/medical+nursing-home costs were estimated. Costs were stratified by vital status and nursing-home-use intensity (nursing-home-days/follow-up-days [0%, 1–24%, 25–99%, and100%]). Between-category mean-annual-cost differences were adjusted for patient characteristics and follow-up-days.

RESULTS:

Costs/follow-up-day distributions differed significantly across cognitive categories. Mean costs/follow-up-day were 2.5–18 times higher for decedents vs survivors. Among all persons with mild-cognitive-impairment, the <9% with any NH-use accounted for 18% of all total-annual medical+NH-costs. Adjusted-between-category comparisons revealed significantly higher medical- and medical+NH-costs for mild-cognitive-impairment vs cognitively-unimpaired.

DISCUSSION:

Cost-effectiveness for managing/modifying both mild-cognitive-impairment and dementia should consider end-of-life costs and nursing-home-use intensity. Results can help inform cost-effectiveness models, predict future-care-needs, and aid decision-making by individuals/providers/payers/policy-makers.

1.0. INTRODUCTION

Dementia is characterized by reduced functioning, increased dependency, and greater likelihood of institutionalization.^1,2 The costs of institutionalization, i.e., nursing home (NH) costs, contribute greatly to the substantial economic burden that accompanies dementia.^1,3 As the population ages, the prevalence and economic burden of dementia on individuals and society will increase dramatically.¹ Results from pharmacologic interventions to delay disease progression among persons with dementia have been disappointing.¹ Non-pharmacological approaches proposed by ourselves and others, e.g., enhanced caregiver supports^4,5,6 and primary-care-based interventions⁷ have been shown to delay permanent NH placement and reduce institutionalization rates among persons with dementia. The focus now includes modeling the potential impact of hypothetical interventions at pre-dementia stages to prevent/postpone dementia onset.^8,9 The importance of cost-effectiveness modeling to assess costs and outcomes that might result from interventions at early disease stages, together with challenges facing such modeling efforts, are increasingly recognized.^2,8,9,10,11 Future efforts will greatly benefit from reliable objective data on the demand for and costs of acute- and long-term-care at predementia stages, as well as more rigorous characterization of cognitive impairment for these stages.^2,9,10,11 Such estimates are currently in short supply.

The principal goal of this investigation was to provide baseline direct medical and NH cost estimates to help populate cost-effectiveness models and potentially inform modeling assumptions. Objective estimates were provided from the health-system perspective, across all payers, for persons well-characterized as either cognitively-unimpaired, mild-cognitive-impairment (MCI), or dementia. Estimates may also help identify factors that contribute to the economic burden, facilitate future-cost forecasts, and assist with budgeting decisions by individuals, providers, and policy makers.

2.0. METHODS

2.1. Setting/Study Design

This population-based historical-cohort study was set in Olmsted County, MN, (2010 census 144,248) and was approved by Mayo Clinic and Olmsted Medical Center (OMC) Institutional Review Boards.

2.2. Resources

2.2.1. Rochester Epidemiology Project (REP):

Rochester, Olmsted County seat, is approximately 80 miles from the nearest major metropolitan area and home to Mayo Clinic, a large tertiary-care-referral center. Mayo Clinic, OMC, and affiliated hospitals provide >95% of care received by Olmsted County residents for hospital inpatient/hospital outpatient/ER, office, and NH visits/laboratory/radiology/pathology.^12,13,14 However, some direct medical care is provided outside these institutions; costs for such care were either unavailable (e.g., outpatient pharmaceuticals/home healthcare) or incomplete (e.g., dentistry/optometry/counseling). Each County resident seen at Mayo and/or OMC is assigned a unique identifier. Provider-linked medical records contain line-item detail from every contact.¹³ REP affords vital and residency status for essentially all County residents 1965-present.¹³

2.2.2. REP Cost Data Warehouse (CDW):

Through a data-sharing agreement between Mayo Clinic and OMC, patient-level administrative data on all hospital- and professional-billed services are shared and available for approved research studies. Electronically-linked data contain line-item detail on date/type/frequency/and billed charge for each good or service. Data are available for all ages and payer-types, including uninsured. A costing algorithm uses widely-accepted valuation techniques to generate standardized inflation-adjusted direct medical-cost estimates in constant U.S. dollars and assign a nationally-representative calendar-year-specific dollar cost to each line item (Appendix A).¹⁵ Present-study estimates are provided in 2010 dollars.

2.2.3. Mayo Clinic Study of Aging (MCSA):

REP resources afforded construction of age- and sex-stratified sampling frames of Olmsted County residents aged 70–89 years on 10/01/2004 and 03/01/2008. Initial determination of cognitive status among sampled individuals involved neurologist’s medical-record review to identify persons who met Diagnostic and Statistical Manual of Mental Disorders, Fourth Edition (DSM-IV)¹⁶ dementia criteria.¹⁷ Remaining individuals were invited to participate in prospective in-person evaluations involving 1) study coordinator’s semi-structured interview, including memory and functional-status assessments; 2) physician’s clinical evaluation, with full neurological examination to assess cognition and cerebrovascular-disease history; and 3) extensive cognitive testing by a psychometrist to assess memory, executive function, language, and visuospatial skills. Cognitive status was based on consensus of study physicians, coordinator, and neuropsychologists, using all available information to determine whether individuals met criteria for cognitively-unimpaired, MCI, or previously-unrecognized dementia.¹⁴ Individuals who declined in-person evaluation were invited to participate in telephone interviews using the Modified Telephone Interview for Cognitive Status (TICS-M).^14,18 Validated TICS-M cut-off scores of ≤31 and ≤27 were used to characterize MCI and dementia respectively.¹⁸

2.2.4. Centers for Medicare and Medicaid Services (CMS) Minimum Data Set (MDS):

Information on NH use and costs was facilitated using Minnesota CMS-MDS version 2.0 NH assessments.¹⁹ All Medicare- and Medicaid-certified long-term-care facilities are required to assess each resident (regardless of payer) at admission and specified intervals throughout their stay. Each individual is assigned a unique identifier that is maintained for all admissions across all facilities. Assessments include resident and NH identifiers/resident’s pre-admission ZIP-code/admission, discharge, and assessment dates/admission source/discharge disposition/reason for assessment/and Resource Utilization Groups (RUG) case-mix classification.²⁰ RUG case-mix is based on resident’s level of acuity and need for care and is used for NH-management and reimbursement; we used RUG version III, which consists of 53 distinct case-mix categories.²⁰ Per-diem NH costs were estimated for each individual by identifying every assessment conducted during each stay and employing the RUG-III value for every day the assessment applied. CMS-MDS details are provided in Appendices B and C. NH-cost estimates included all provided services, regardless of payer. Costs for long-term-care provided elsewhere (e.g., assisted living and privately-owned memory-care centers) were unavailable.

2.3. Participants/Data collection

Appendix Figure A provides inclusion/exclusion criteria for the 6,682 MCSA subjects identified from 2004/2008 sampling frames; participants were limited to 3,545 individuals. Appendices B and Figure B provide detailed processes for identifying all local NH residents and merging with all 3,545 MCSA study participants. All participants were followed for medical and NH use/costs from (and including) index forward until death, emigration from Olmsted County, end of availability of CMS NH data (09/30/2010), or maximum 366 days. For prospectively-assessed participants, index was defined as initial-assessment date. For participants retrospectively identified as meeting DSM-IV dementia criteria, record review often included multiple clinical contacts, sometimes over years; thus, index was defined as neurologist’s review date.

2.3.1. Instrumental Activities of Daily Living (IADLs)

Measurements used the 10-item Functional Activity Questionnaire (FAQ).²¹ Measurements were limited to prospectively-assessed cognitively-unimpaired individuals and persons with MCI. Scores were based on informant’s perception of participant’s ability to conduct each item (Appendix Table A). Informants were instructed to only consider limitations related to cognitive-impairment.

2.3.2. Comorbidity.

For each individual, we obtained all International Classification of Diseases, Ninth Revision, Clinical Modification (ICD-9-CM) diagnosis codes²² assigned the year before index. After excluding dementia and MCI codes, we used Johns Hopkins software (Appendix D)²³ to calculate a Resource Utilization Band (RUB) summary measure of comorbidity. Advantages of RUB include that it was developed to examine associations between comorbid conditions and healthcare costs and does not rely on self-reported number or presence of specified conditions.²³

RUB adjusts for multiple medical conditions, including psychiatric conditions. Psychiatric problems are recognized as contributing to dementia-associated costs.³ To explore the extent of that contribution, we again used Johns Hopkins software but excluded relevant codes within ICD-9-CM Chapter 5 (Mental Disorders) from the RUB calculation. Thus, psychiatric conditions were not adjusted for, affording comparison with models that adjusted for these conditions.

2.4. Statistical analysis

2.4.1. Subject characteristics:

Across-cognitive-category comparisons were conducted using chi-square tests for dichotomous variables, Mantel-Haenszel chi-square tests for ordinal variables, and Kruskal Wallis for continuous variables (see table footnotes). Testing used a two-tailed alpha level of 0.05 to determine statistical significance.

2.4.2. Costs:

Estimates were limited to direct medical, NH, and medical+NH costs. Indirect costs were unavailable. Observed results were provided for all individuals (N=3,545), including decedents, and for the subset who were alive with access to cost data over the full year (N=3,142). Results were also stratified by vital status at last-follow-up and by NH-use intensity (defined for each individual as days as a NH resident divided by follow-up-days). Groupings of 0% (no NH use), 1–24%, 25–99%, and 100% (full-time NH resident) were informed by Medicare NH-reimbursement rules for co-pay charges within each 100-day benefit period.²⁴ Across-cognitive-category differences in costs/follow-up-day distributions were assessed using Kruskal Wallis tests.

Multivariable generalized linear models were used to predict mean-annual costs, adjusted for age-at-index, sex, education, length-of-follow-up, and comorbidity. Models assumed a gamma distribution with log link for the error term to account for skewed cost distributions. This approach enabled coefficients to be directly back transformed into the original dollar scale.^25,26 Predicted-mean-cost-differences between cognitive categories and bootstrapped 95% confidence intervals (CI) were calculated for significance tests.^27,28 Analyses were performed using SAS version 9.4 (SAS Institute, Cary, NC).

3.0. RESULTS

3.1. Patient characteristics

Table 1 reveals significant across-cognitive-category differences for each variable. For persons categorized as cognitively-unimpaired, MCI, and dementia respectively: 1.0%, 1.9%, and 17% died; 3.8%, 8.7%, and 35% had NH use. Distributions of NH-use intensity, RUB comorbidity, and FAQ IADL limitations, revealed increasing values with increasing cognitive-impairment.

Table 1.

Participant Characteristics by Cognitive-Impairment Category

Characteristics	Cognitively Unimpaired (N=2,438)	Mild Cognitive Impairment (N=528)	Dementia (N=579)*	P Value†
Male sex, n (%)	1,137 (47%)	274 (52%)	260 (45%)	0.045
Age at index ‡ , years				<0.001
Mean (SD)	79 (5.2)	81 (5.0)	83 (4.6)
Median (IQR)	79 (74, 83)	82 (79, 85)	84 (81, 87)
White race, n (%)	2,401 (98%)	517 (98%)	561 (97%)	0.04
Married, n (%) §	1,505 (62%)	307 (58%)	75 (50%)	0.006
Education, years				<0.001
Mean (SD)	14 (3.0)	13 (3.0)	12 (3.2)
Median (IQR)	13 (12, 16)	12 (12, 14)	12 (10, 14)
Activity: Index‡ through end of follow-up¶
Days of follow-up				<0.001
Mean (SD)	349 (59)	342 (71)	322 (93)
Median (IQR)	366 (366, 366)	366 (366, 366)	366 (366, 366)
Lost-to-follow-up, n (%)	206 (8.4%)	60 (11%)	137 (24%)	<0.001
Died, n (%)¶	24 (1.0%)	10 (1.9%)	98 (17%)	<0.001
Utilization, n (%)
Medical	2,382 (98%)	509 (96%)	553 (96%)	0.009
NH	93 (3.8%)	46 (8.7%)	205 (35%)	<0.001
Nursing-use intensity, n (%)#				<0.001
0% (no NH use)	2,345 (96%)	482 (91%)	374 (65%)
1–24%	75 (3.1%)	35 (6.6%)	37 (6.4%)
25–99%	13 (0.5%)	6 (1.1%)	38 (6.6%)
100% (full-time resident)	5 (0.2%)	5 (1.0%)	130 (22%)
RUB, n (%) **				<0.001
No, Healthy, or Low use	142 (5.8%)	20 (3.8%)	34 (5.9%)
Moderate use	892 (37%)	144 (27%)	162 (28%)
High use	781 (32%)	195 (37%)	161 (28%)
Very High use	623 (26%)	169 (32%)	222 (38%)
FAQ, n (%) ††				<0.001
0	1,499 (75%)	146 (38%)	NA
1–10	479 (24%)	202 (53%)	NA
≥10	6 (0.3%)	31 (8.1%)	NA

SD = Standard deviation, IQR = Interquartile range, NH = Nursing home, RUB = Resource Utilization Band FAQ = Functional Activities Questionnaire, NA = Not available

^*The dementia category includes 461 individuals identified as meeting Diagnostic and Statistical Manual of Mental Disorders, 4^th Edition[17] (DSM-IV) criteria based on retrospective review of detailed provider-linked medical records and 118 individuals who did not meet DSM-IV criteria for dementia based on retrospective record review but subsequently met DSM-IV criteria based on in-person assessment.

^†P values were calculated using chi square tests for dichotomous variables (male sex, white race, married, lost-to-follow-up, died, medical utilization, NH utilization); Mantel-Haenszel chi-square tests for ordinal variables (RUB, NH-use intensity, FAQ) and Kruskal Wallis tests for continuous variables (age at index, education, days of follow-up).

^‡Index dates ranged from 2004–2010

^§For cognitively unimpaired individuals, persons with mild cognitive impairment, and the 118 persons who met DSM-IV criteria for dementia based on prospective assessment, marital status was available from data collected during the assessment. For the 461 persons who met DSM-IV criteria for dementia based on retrospective record review, marital status was obtained electronically from Rochester Epidemiology Project medical records, and was missing for all but 32 individuals, 18 (56%) of whom were married. Calculations of the percent of married individuals for the 579 persons with dementia was limited to the 150 individuals for whom marital status was available.

^¶Individuals were followed from index date to the earliest of death, emigration from Olmsted County, end of availability of NH data from Centers for Medicare and Medicaid Services (9/30/2010), or 1-year after index date (maximum of 366 days of follow-up, including index date). Death was included in lost-to-follow-up measures.

^#All individuals were divided into distinct groupings of NH-use intensity (0%, 1–24%, 25–99%, or 100%), defined as the number of days as a NH resident (including index and/or during the year after index) divided by his/her total number of follow-up days.

^**Johns Hopkins RUB is a summary measure of comorbidity, defined using aggregations of diagnostic groupings that have similar expected resource use, with values ranging from 0 (no encounters) to 5 (diagnosis codes associated with very high use).[23] We calculated RUB values using diagnosis codes assigned in the year before index and excluding codes for dementia and mild cognitive impairment (ICD-9-CM codes 290.XX, 331.0, 331.82, 331.83).

^††The Functional Activities Questionnaire (FAQ)[21] was used to measure Instrumental Activities of Daily Living; FAQ summary scores range from 0 (independent in all 10 activities) to a maximum of 30 (dependent for all 10 activities) FAQ measurements for this study were limited to the subset of 2,363 participants who were assessed in-person as either cognitively unimpaired or MCI and for whom informants responded to at least seven of the ten FAQ activities listed in Appendix Table A. Informants were instructed to only consider limitations related to cognitive impairment.

3.2. Observed costs (all individuals, N=3,545)

3.2.1. When stratified by NH-use intensity (Table 2A)

Table 2.

Observed medical, NH, and medical plus NH costs by cognitive category, obtained for each of 3,545 individuals followed from index until earliest of death, emigration from the area, end of availability of NH data from CMS, or maximum of 366 days^*.

Estimates are provided as mean annual costs and distribution of costs/follow-up day^†

Table 2A. Cost estimates are provided for all individuals (regardless of NH use) and stratified by NH-use intensity (defined for each individual as number of days as a NH resident divided by his/her total follow-up days)
	Cognitively unimpaired (N=2,438)	Mild cognitive impairment (N=528)	Dementia (N=579)	P value‡
MEDICAL COSTS †
All individuals, N (%)	2,438 (100%)	528 (100%)	579 (100%)
Annual costs, mean (SD)	$7,253 (14,373)	$9,542 (21,703)	$10,561 (19,796)
Cost/follow-up day§				0.002
Mean (SD)	$23 (84)	$31 (84)	$39 (85)
Median (IQR)	$7 (3–18)	$9 (4–29)	$7 (2–37)
Range	$0–2,696	$0–1,189	$0–763
0% NH use, N (%)	2,345 (96%)	482 (91%)	374 (65%)
Annual costs, mean (SD)	$6,141 (11,890)	$6,582 (9,931)	$7,351 (14,501)
Cost/follow-up day§				0.004
Mean (SD)	$20 (82)	$21 (50)	$28 (75)
Median (IQR)	$7 (3–16)	$8 (3–23)	$5 (2–22)
Range	$0–2,696	$0–723	$0–763
1–24% NH use, N (%)	75 (3.1%)	35 (6.6%)	37 (6.4%)
Annual costs, mean (SD)	$33,702 (27,979)	$45,002 (62,947)	$28,538 (26,557)
Cost/follow-up day§				0.48
Mean (SD)	$97 (81)	$132 (197)	$97 (91)
Median (IQR)	$73 (45–117)	$92 (46–148)	$62 (32–154)
Range	$2–425	$13–1,189	$1–328
25–99% NH use, N (%)	13 (0.5%)	6 (1.1%)	38 (6.6%)
Annual costs, mean (SD)	$55,904 (48,107)	$42,280 (46,663)	$31,282 (31,324)
Cost/follow-up day§				0.08
Mean (SD)	$183 (133)	$238 (322)	$106 (126)
Median (IQR)	$136 (106–305)	$101 (23–358)	$58 (23–135)
Range	$42–474	$2–842	$2–520
100% NH use, N (%)	5 (0.2%)	5 (0.9%)	130 (22%)
Annual costs, mean (SD)	$5,533 (6,485)	$7,339 (11,258)	$8,622 (20,414)
Cost/follow-up day§				0.86
Mean (SD)	$15 (18)	$20 (31)	$36 (81)
Median (IQR)	$3 (3–28)	$8 (6–10)	$4 (2–29)
Range	$3–40	$1–75	$0–493
NURSING HOME COSTS †
All individuals, N (%)	2,438 (100%)	528 (100%)	579 (100%)
Annual costs, mean (SD)	$770 (6,073)	$2,045 (9,915)	$19,625 (32,571)
Cost/follow-up day§				<0.001
Mean (SD)	$2 (19)	$6 (28)	$65 (101)
Median (IQR)	$0 (0–0)	$0 (0–0)	$0 (0–176)
Range	$0–400	$0–244	$0–375
0% NH use, N (%)	2,345 (96%)	482 (91%)	374 (65%)
Annual costs, mean (SD)	--	--	--
Cost/ follow-up day§				--
Mean (SD)	--	--	--
Median (IQR)	--	--	--
Range	--	--	--
1–24% NH use, N (%)	75 (3.1%)	35 (6.6%)	37 (6.4%)
Annual costs, mean (SD)	$10,503 (8,180)	$12,879 (9,860)	$11,693 (8,437)
Cost/follow-up day§				0.18
Mean (SD)	$30 (22)	$37 (27)	$36 (23)
Median (IQR)	$25 (14–42)	$28 (16–57)	$30 (19–48)
Range	$2–105	$3–116	$3–99
25–99% NH use, N (%)	13 (0.5%)	6 (1.1%)	38 (6.6%)
Annual costs, mean (SD)	$54,786 (29,245)	$37,938 (28,093)	$55,552 (21,523)
Cost/follow-up day§				0.13
Mean (SD)	$188 (98)	$120 (63)	$171 (60)
Median (IQR)	$144 (118–238)	$110 (91–127)	$166 (122–226)
Range	$93–400	$45–235	$53–277
100% NH use, N (%)	5 (0.2%)	5 (0.9%)	130 (22%)
Annual costs, mean (SD)	$75,257 (11,084)	$80,259 (5,447)	$67,842 (27,254)
Cost/follow-up day§
Mean (SD)	$206 (30)	$219 (15)	$230 (39)	0.44
Median (IQR)	$207 (181–224)	$213 (209–222)	$214 (201–248)
Range	$171–244	$208–244	$167–375
MEDICAL + NH COSTS †
All individuals	2,438 (100%)	528 (100%)	579 (100%)
Annual costs, mean (SD)	$8,022 (17,368)	$11,586 (26,155)	$30,186 (41,010)
Cost/follow-up day§				<0.001
Mean (SD)	$25 (89)	$37 (95)	$105 (141)
Median (IQR)	$7 (3–18)	$9 (4–31)	$25 (3–206)
Range	$0–2,696	$0–1,285	$0–778
0% NH use, N (%)	2,345 (96%)	482 (91%)	374 (65%)
Annual costs, mean (SD)	$6,141 (11,890)	$6,582 (9,931)	$7,351 (14,501)
Cost/follow-up day§				0.004
Mean (SD)	$20 (82)	$21 (50)	$28 (75)
Median (IQR)	$7 (3–16)	$8 (3–23)	$5 (2–22)
Range	$0–2,696	$0–723	$0–763
1–24% NH use, N (%)	75 (3.1%)	35 (6.6%)	37 (6.4%)
Annual costs, mean (SD)	$44,205 (32,283)	$57,881 (66,882)	$40,230 (30,301)
Cost/follow-up day§				0.43
Mean (SD)	$127 (92)	$169 (209)	$134 (101)
Median (IQR)	$99 (64–170)	$117 (79–196)	$115 (48–204)
Range	$5–512	$21–1,285	$17–363
25–99% NH use, N (%)	13 (0.5%)	6 (1.1%)	38 (6.6%)
Annual costs, mean (SD)	$110,690 (68,253)	$80,219 (56,034)	$86,834 (44,259)
Cost/follow-up day§				0.26
Mean (SD)	$371 (204)	$358 (326)	$277 (162)
Median (IQR)	$275 (224–456)	$282 (131–450)	$228 (189–330)
Range	$174–797	$47–954	$55–765
100% NH use, N (%)	5 (0.2%)	5 (0.9%)	130 (22%)
Annual costs, mean (SD)	$80,790 (12,918)	$87,597 (16,309)	$76,464 (38,279)
Cost/follow-up day§				0.73
Mean (SD)	$221 (35)	$239 (45)	$266 (99)
Median (IQR)	$221 (210–227)	$221 (219–223)	$225 (207–282)
Range	$174–272	$215–319	$169–778

TABLE 2B. Cost estimates for all 3,545 individuals stratified by vital status as of last follow-up*
	Cognitively unimpaired (N=2,438)	Mild cognitive impairment (N=528)	Dementia (N=579)	P value‡
MEDICAL COSTS †
Individuals who died during follow-up, N (%)	24 (1.0%)	10 (1.9%)	98 (17%)
Annual costs, mean (SD)	$27,613 (26,643)	$62,297 (111,710)	$17,014 (23,070)
Cost/follow-up day§				0.007
Mean (SD)	$350 (689)	$349 (393)	$105 (155)
Median (IQR)	$101 (29–238)	$182 (85–551)	$40 (5–148)
Range	$1–2,696	$5–1,189	$0–763
Individuals alive at follow-up, N (%)	2,414 (99%)	518 (98%)	481 (83%)
Annual costs, mean (SD)	$7,050 (14,060)	$8,523 (14,424)	$9,246 (18,818)
Cost/follow-up day§				<0.001
Mean (SD)	$20 (38)	$25 (50)	$26 (53)
Median (IQR)	$7 (3–18)	$9 (3–27)	$5 (2–25)
Range	$0–524	$0–723	$0–493
NURSING HOME COSTS †
Individuals who died during follow-up, N (%)	24 (1.0%)	10 (1.9%)	98 (17%)
Annual costs, mean (SD)	$3,751 (7,679)	$5,331 (9,775)	$24,079 (29,633)
Cost/follow-up day§				0.001
Mean (SD)	$33 (92)	$28 (43)	$131 (122)
Median (IQR)	$0 (0–18)	$0 (0–43)	$199 (0–232)
Range	$0–400	$0–112	$0–375
Individuals alive at follow-up, N (%)	2,414 (99%)	518 (98%)	481 (83%)
Annual costs, mean (SD)	$740 (6,049)	$1,981 (9,917)	$18,718 (33,093)
Cost/follow-up day§				<0.001
Mean (SD)	$2 (17)	$5 (27)	$52 (91)
Median (IQR)	$0 (0–0)	$0 (0–0)	$0 (0–56)
Range	$0–322	$0–244	$0–347
MEDICAL + NH COSTS †
Individuals who died during follow-up, N (%)	24 (1.0%)	10 (1.9%)	98 (17%)
Annual costs, mean (SD)	$31,364 (28,279)	$67,628 (120,194)	$41,093 (37,560)
Cost/follow-up day§				0.45
Mean (SD)	$382 (690)	$377 (427)	$235 (182)
Median (IQR)	$116 (29–280)	$198 (85–551)	$214 (81–331)
Range	$1–2,696	$5–1,285	$0–765
Individuals alive at follow-up, N (%)	2,414 (99%)	518 (98%)	481 (83%)
Annual costs, mean (SD)	$7,790 (17,075)	$10,505 (19,594)	$27,964 (41,363)
Cost/follow-up day (SD)§				<0.001
Mean (SD)	$22 (47)	$31 (62)	$78 (114)
Median (IQR)	$7 (3–18)	$9 (4–29)	$14 (3–171)
Range	$0–797	$0–723	$0–778

Abbreviations: CMS, Centers for Medicare and Medicaid Services; NH, nursing home; IQR, interquartile range; SD, standard deviation.

^*Individuals were followed both in and out of the NH from (and including) index date to the earliest of death, emigration from the area, end of availability of NH data from the Center for Medicare and Medicaid Services (09/30/2010), or 1-year after index date (maximum of 366 days of follow-up).

^†Estimates of costs/follow-up day were calculated for each individual by dividing his/her total costs within each cost type separately (medical, NH, or medical + NH)) by his/her total days of follow-up. For each individual, values for Medical + NH costs were estimated by adding his/her costs from each source.

^‡P values were calculated using Kruskal Wallis tests for distribution of costs/follow-up day.

^§Values included persons with zero medical and/or NH costs. By definition, for cost types NH and medical + NH costs, all individuals in the grouping 0% NH use had zero NH costs.

Significant across-cognitive-category differences were observed for both medical and medical+NH costs/follow-up-day distributions for all individuals (regardless of NH-use) and those with no NH-use; mean-costs/follow-up-day appeared to increase across cognitive categories. NH costs/follow-up-day distributions differed significantly for all individuals; mean NH-costs/follow-up-day were $2 for cognitively-unimpaired vs $65 for dementia. No significant differences in costs/follow-up-day distributions were observed for NH-use intensities 1–24%, 25–99%, or 100%, likely reflecting small numbers of NH residents in predementia categories. The Figure provides costs/follow-up-day distributions by cognitive category, cost-type and NH-use intensity.

Figure.

Observed objective direct estimates of costs/follow-up-day distributions (diamond=mean, horizontal line=median, box=interquartile range [25–75%]) are provided in 2010 U.S. constant dollars for all 3,545 study participants, well-characterized as either Cognitively unimpaired (white), Mild cognitive impairment (blue), or Dementia (orange). Cost estimates for each individual were obtained forward from (and including) index date until earliest of death, emigration, end of availability of Centers for Medicare and Medicaid Services Minimum Data Set NH assessments (09/30/2010), or maximum of 366 days. Estimates are provided for Medical (top panel), Nursing home (middle panel), and Medical plus Nursing home (bottom panel) costs. Analyses were stratified by NH-use-intensity groupings, calculated for each individual as number of days as a NH resident divided by his/her total follow-up-days (0% [no NH use], 1–24%, 25–99%, and 100% [full-time NH resident]). Separate distributions are also provided for all individuals, regardless of NH use. Estimates included persons with zero costs. To facilitate interpretation, numeric values for mean-costs/follow-up-day are also provided for each distribution

Descriptive comparisons of within-cognitive-category differences revealed mean medical-costs/follow-up-day for persons with dementia and 100% NH-use intensity ($36) were ~1/3 of those for 1–24% ($97) or 25–99% ($106) NH-use intensity. However, mean medical+NH-costs/follow-up-day for persons with dementia and 0%, 1–24%, 25–99%, and 100% NH-use intensity were $28, $134, $277, and $266 respectively. Because 35% of persons with dementia experienced some NH-use, total-annual NH costs for all 579 persons with dementia ($11,363,123) accounted for 65% of total-annual medical+NH costs ($17,477,669). For persons with MCI, although <9% (N=46) experienced any NH use, total-annual NH costs ($1,079,685) for all 528 persons accounted for 18% of total-annual medical+NH costs ($6,117,629).

3.2.2. When stratified by vital status at last-follow-up (Table 2B)

Across-cognitive-category comparisons revealed significant differences in costs/follow-up-day distributions for both survivors and decedents, with one exception (medical+NH costs for decedents). Depending on cognitive category and cost-type, mean-costs/follow-up-day for decedents appeared 2.5–18 times those for survivors. Decedents constituted 1.0%, 2.0%, and 17% of all individuals in cognitively-unimpaired, MCI, and dementia categories respectively; however, decedents contributed 4.0%, 12%, and 30% of total-annual medical+NH costs in these respective categories.

For survivors, descriptive comparisons of across-cognitive-category differences in mean-costs/follow-up-day appeared similar (medical) or higher (NH and medical+NH) for those with vs without dementia. For decedents, mean NH-costs/follow-up-day for persons with dementia, ($131) were approximately four-fold those for persons without dementia (cognitively-unimpaired=$33; MCI=$28); however, mean medical-costs/follow-day for decedents with dementia ($105) were only 30% of those without dementia (cognitively-unimpaired=$350; MCI=$349).

3.3. Observed costs (366-day survivors, N=3,142)

Appendix Table B was stratified by NH-use intensity. Although the format was the same as that used in Table 2A, Appendix Table B was limited to the subset of individuals who were alive, residing locally, and for whom we had access to medical and NH cost data for the full year (i.e., 366-day survivors). Significant across-cognitive-category comparisons of costs/follow-up-day distributions seen for Table 2A were maintained in Appendix Table B. Of the 42 values for mean-costs/follow-up-day shared by both tables, descriptive comparisons revealed that point-estimates appeared the same (N=10) or lower (N=31) for 366-day survivors vs all individuals. The 366-day survivors were likely healthier, required less care, and were essentially absent end-of-life costs.

3.4. Adjusted costs (all 3,545 individuals)

Predicted-mean-annual-cost estimates were adjusted for baseline patient characteristics and follow-up-days (Table 3). Estimates were limited to medical and medical+NH costs. Separate estimation of NH costs was impeded by limited NH use for predementia categories. However, for persons with dementia, medical+NH costs appeared much higher than medical costs in every instance; reinforcing the substantial contribution of NH to medical+NH costs for persons with dementia in Table 2A.

Table 3.

Between-Cognitive-Category Predicted Mean Cost Differences^* for Medical^† and Medical plus NH Costs^† in the Year after Index^‡ for all 3,545 Participants. Estimates of Between-Cognitive-Category Cost Differences are Adjusted for Age, Sex, Education, and Follow-Up Days (Table 3A) and Additionally Adjusted for Two Summary Measures of Comorbidity^§, One Including (Table 3B), the Other Excluding (Table 3C) Adjustment for Between-Category Differences in Psychiatric Conditions^#

A. Adjusted for Age, Sex, Education, and Follow-up Days#
Referent Category*	Contrast Category*	Medical Costs			Medical plus NH Costs
		Referent, predicted mean	Contrast, predicted mean	Contrast minus referent, predicted mean difference (95% CI)	Referent, predicted mean	Contrast, predicted mean	Contrast minus referent, predicted mean difference (95% CI)
Cognitively Unimpaired	MCI	$7,424	$9,474	$2,050 (410–4,117)	$8,319	$11,362	$3,043 (1,021–5,480)
	Dementia	$7,448	$10,881	$3,433 (1,615–5,617)	$8,244	$28,960	$20,716 (17,392–25,206)
MCI	Dementia	$10,042	$11,336	$1,294 (−1,169–3,915)	$12,179	$31,359	$19,188 (14,933–23,888)
B. Adjusted for Age, Sex, Education, Follow-Up Days, and Measure of Comorbidity§ that Included Adjustment for Psychiatric Conditions #
Referent category *	Contrast category *	Medical Costs			Medical plus NH Costs
		Referent, predicted mean	Contrast, predicted mean	Contrast minus referent, predicted mean difference (95% CI)	Referent, predicted mean	Contrast, predicted mean	Contrast minus referent, predicted mean difference (95% CI)
Cognitively unimpaired	MCI	$7,439	$9,015	$1,576 (143–3,371)	$8,325	$10,710	$2,385 (555–4,624)
	Dementia	$7,383	$10,500	$3,117 (1,426–5,318)	$8,208	$27,787	$19,579 (16,664–24,301)
MCI	Dementia	$9,714	$11,200	$1,487 (−912–3,924)	$11,706	$30,634	$18,928 (15,161–23,798)
C. Adjusted for Age, Sex, Education, Follow-up Days and Measure of Comorbidity§ that Did Not Adjust for Psychiatric Conditions
Referent category *	Contrast category *	Medical Costs			Medical plus NH Costs
		Referent, predicted mean	Contrast,predicted mean	Contrast minus referent,predicted mean difference (95% CI)	Referent, redicted mean	Contrast, predicted mean	Contrast minus referent, predicted mean difference (95% CI)
Cognitively Unimpaired	MCI	$7,404	$9,016	$1,612 (176–3,399)	$8,245	$10,680	$2,435 (606–4,682)
	Dementia	$7,459	$10,821	$3,361 (1,545–5,611)	$8,438	$29,275	$20,837 (17,234–25,000)
MCI	Dementia	$9,494	$11,070	$1,576 (−824–3,979)	$11,910	$31,627	$19,717 (15,571–24,275)

NH = Nursing home, RUB = Resource Utilization Band, CI = Confidence interval, MCI = Mild Cognitive Impairment,

^*Between-cognitive-category differences in predicted mean annual costs were estimated using the method of recycled predictions, setting all individuals as referent or contrast, while keeping constant all other individual characteristics.^27,28 Predicted mean annual cost estimates for the less cognitively impaired category (i.e., referent category) were subtracted from those for each of the more cognitively impaired category (i.e. contrast category). Mean values and bootstrapped 95% confidence intervals of the mean difference were calculated. Separate models were run for estimating each between-category difference. Slight variability in predicted mean costs for the same category results from the methodology; recycled predictions sets all individuals to the referent category or the contrast category, although all other individual characteristics remain as observed.

^†The source for medical cost estimates was the Olmsted County Cost Data Warehouse.¹⁵ The source for estimating NH costs was the Centers for Medicare and Medicaid Services (CSM) Minimum Data Set (MDS).¹⁹ MDS assessments are required for all individuals residing in CMS-certified nursing homes; residents are assessed at admission and at defined periods during the stay. Detailed descriptions of each of these data sources are provided in Supplementary texts Appendices A, B, and C.

^‡For persons who met Diagnostic and Statistical Manuel of Mental Disorders¹⁶ criteria for dementia based on retrospective record review (N=461), index date was defined as record review date. For persons identified as cognitively unimpaired (N=2439), MCI (N=528), or who met DSM-IV criteria for dementia based on prospective assessment (N=118), index data was defined as assessment date; cognitive status was that determined at their baseline (i.e., first) assessment. For analytic purposes, individuals identified using either approach for dementia were combined into a single category of dementia (N=579).

^§Comorbidity was assessed using Johns Hopkins Resource Utilization Band (RUB) software.²³ RUB is a summary measure of comorbidity, defined using aggregations of diagnostic groupings that have similar expected resource use, with values ranging from 0 (no encounters) to 5 (diagnosis codes associated with very high use); we collected all diagnosis codes assigned an individual within the year before index to calculate RUB values. RUB calculations for adjustment in Table 3B excluded codes for dementia and MCI (ICD‐9‐CM codes 290.XX, 331.0, 331.82, 331.83) to avoid over-adjustment; RUB calculations for adjustment in Table 3C additionally excluded psychiatric diagnosis codes, thus allowing between-cognitive-category differences in psychiatric diagnoses to contribute to between-cognitive-category cost differences.

^#Psychiatric conditions were defined as relevant diagnosis codes contained within Chapter 5 (Mental Conditions) of The International Classification of Diseases, Ninth Revision, Clinical Modification.²²

Table 3 estimates were adjusted using three approaches; 3A adjusted for age, sex, education, and follow-up-days; 3B added RUB comorbidity that adjusts for multiple medical conditions, including psychiatric conditions; 3C excluded psychiatric conditions from the RUB calculation, allowing psychiatric conditions to contribute to between-category cost differences. Between-category comparisons revealed medical and medical+NH costs were each significantly higher for MCI vs cognitively-unimpaired; differences reached significance in all three adjustment models. Medical+NH costs were significantly lower for MCI vs dementia in all three models; however, medical costs did not differ between MCI and dementia in any of the models.

Significant between-category cost differences appeared attenuated (reduced) following additional adjustment for RUB comorbidity (3B vs 3A results). The extent of attenuation differed depending on the categories compared and cost-types considered. When psychiatric conditions were excluded from the RUB-adjustment calculation (3C), significant between-category cost differences appeared to revert toward values without comorbidity adjustment (3A).

4.0. DISCUSSION

MCSA participants in this study were well-characterized as cognitively-unimpaired, MCI, or dementia. Participants were followed forward up to 1-year from (and including) index for objective estimates of direct medical, NH, and medical+NH costs. Several key findings and possible explanations/implications emerged.

4.0.1 IADL values appeared higher for persons with MCI vs cognitively-unimpaired individuals (Table 1). Results support previous findings,²⁹ and reports by MCSA investigators and others showing MCI is associated with biomarker evidence of brain changes and subtle problems with memory and thinking.^1,30 Important questions of whether IADL screening may enhance earlier diagnosis, establish risk of progression, help direct interventions, or inform clinical practice were beyond the scope of the present investigation.

4.0.2 For each cognitive category, mean medical-costs/follow-up-day appeared lower for persons with 100% vs 1–24% or 25–99% NH-use intensity (Table 2A, Figure). Potential explanations include: full-time NH residents 1) received medical care within the NH;⁴ 2) had lower medical costs compared with patients admitted to NH for short-term rehabilitation following inpatient surgery; 3) were exempt from Medicare’s requirement of three hospital days before NH readmission outside the benefit-period;³¹ or 4) had a higher proportion with ‘do-not-hospitalize’ orders.³² Each explanation has implications for policy and budgeting decisions that require further investigation.

For persons with dementia, despite low mean medical-costs/follow-up-day for full-time NH residents, total-annual NH costs for all 579 persons with dementia accounted for 65% of total-annual medical+NH costs. The well-recognized contribution of NH costs to the economic burden of dementia^1,3,4,5 is extended here by considering full-time NH residents and NH-use intensity. Findings speak to efforts investigating consequences of non-pharmacologic interventions for NH costs, e.g., caregiver supports,⁵ primary-care-management programs,⁷ and long-term-care options.¹

Far less attention has been paid to NH costs among persons with MCI. Although <9% of individuals with MCI experienced any NH use, total-annual NH costs for all persons with MCI accounted for 18% of total-annual medical+NH costs. This important finding suggests that, when modeling healthcare costs for persons with MCI, it may be inappropriate to assume NH costs are negligible.

4.0.3 Descriptive analyses suggesting that survivors differed from decedents with respect to medical-cost comparisons between persons with vs without dementia (Table 2B) may reflect low medical costs for full-time NH residents with dementia and the greater proportion of full-time NH residents among persons with vs without dementia (Table 2A); further investigation is needed to assess if and how medical services for persons near death differ between persons with and without dementia.³³

4.0.4 Adjusted analyses of between-cognitive-category differences in predicted-mean-annual-costs revealed significantly higher medical and medical+NH costs for persons with MCI vs cognitively-unimpaired individuals; differences remained significant for all three adjustment models (Table 3). This finding contrasts with previous findings by ourselves³⁴ and others^29,35 showing no significant cost differences between persons with MCI and cognitively-unimpaired individuals. Possible explanations for between-study differences include that previous estimates were based on cross-sectional analyses of costs collected before assessment for persons who survived to assessment. This explanation is supported by Table 2B descriptive findings showing lower costs for persons alive at last-follow-up vs persons who died during the year. Further support comes from descriptive findings that mean-costs/follow-up-day appeared lower for cross-sectional analyses limited to 366-day survivors (Appendix Table B) vs analyses of all individuals followed until lost-to-follow-up, including death (Table 2A). All findings support following individuals forward to capture high near-death costs.³³ Investigations limited to survivors, including certain approaches to modeling cost-effectiveness, may underestimate the economic burden for both dementia and MCI.

4.1. Comparisons with previous findings

Comparisons are limited by relatively few published cost estimates along the disease continuum^{29,35.36,37,38,39,40} and by between-study differences in objectives and design. Some investigators used administrative data to first identify individuals with and without ICD-diagnosis codes for dementia and/or MCI and then obtain objective medical- and/or NH-cost estimates before and after diagnosis.^37,38,39 Most were limited to either Medicare and/or Medicaid expenditures. We estimated direct costs using a health-systems perspective across all payers.

Few cost studies assessed persons prospectively for MCI.^29,35,36,40 Luppa et al. applied International Working Group on MCI criteria using psychological-test-battery cut-points.³⁵ Zhu et al.³⁶ and Robinson et al.⁴⁰ enrolled individuals referred by providers as having self-reported memory concerns and prospectively confirmed whether they met screening-test cut-points for MCI vs no MCI³⁵ or MCI vs early dementia.⁴⁰ Ton et al.²⁹ enrolled Aging, Demographics, and Memory Study participants. Individuals were categorized as cognitively normal; MCI; and mild/moderate/severe dementia using Clinical Dementia Rating Sum of Boxes. Present-study participants were drawn from a randomly-sampled population, reducing potential bias associated with referral-based studies.⁴¹ Participants were well-characterized for cognitive status, addressing limitations of discharge diagnosis codes, questionnaires, or test scores alone for determining MCI.^42,43

For almost all studies that prospectively identified persons with MCI, direct costs were estimated using self- or proxy-reported service use and applying average costs/service unit for the general population.⁴⁴ The present study’s access to administrative data afforded objective medical- and NH-cost estimates. Several investigations of healthcare costs for pre-dementia categories enrolled community-dwelling individuals and/or were largely absent objective NH costs; few studies reported NH costs separately. Ton et al.²⁹ reported NH costs as part of total-medical spending, thus approximating our medical+NH costs. In adjusted between-category comparisons, findings revealed only one significant difference, i.e., MCI versus severe AD.²⁹ In contrast to investigations limited to survivors, the present study followed individuals forward, affording cost-comparisons between survivors and decedents.

4.2. Limitations

The present study has several limitations. Indirect costs, known to contribute substantially to costs for cognitively-impaired individuals,^1,3,40 were unavailable. The study did not consider health outcomes, quality of life, or costs of interventions/unintended consequences/or caregivers’ healthcare costs. There is increasing recognition that cost-effectiveness models need to incorporate such considerations.¹¹ The setting was Olmsted County, MN (86% white in 2010). Compared with MN and other upper mid-west states, County residents exhibit similar demographics and chronic-disease rates but higher income and education.⁴⁵ Generalizability of findings could be compromised by low racial diversity, higher income/education, and relatively few medical- and NH-care providers.¹ Importantly, however, compared with Medicare-eligible residents of MN, Olmsted County and MN exhibited similar mean hospital-inpatient stays (0.33 versus 0.30) and days (1.5 versus 1.4).¹² And the 4.1% of Olmsted County residents aged 65+ in a nursing facility/skilled-nursing facility in 2010⁴⁶ was similar to 3.7% of MN residents age 65+ in a NH in 2011⁴⁷ and the 3.9% for U.S. white Medicare beneficiaries in a long-term-care facility in 2011.⁴⁸ Potential generalizability of our cost estimates for dementia is suggested by similar estimates by Minnesota Department of Health, using a state-wide all-payers claims database⁴⁹ and by Hurd et al.³ using a nationally-representative Health and Retirement Survey sample.

Prospective assessments were limited to eligible individuals who agreed to participate; non-participants were older, more likely male, and exhibited higher comorbidity.¹⁴ However, non-participants who did not refuse use of medical records for research⁵⁰ were followed forward in REP records for median 3.9 years to estimate dementia incidence using DSM-IV criteria; rates were similar to rates for prospectively-assessed participants.⁵¹

For persons with dementia, measures of severity and IADLs were not available; their absence could limit the extent to which our estimates can inform modeling efforts that require such information.⁴⁷ However, because refusal rates for use of REP medical records in research are typically <5%,^13,50 it is possible that, compared with intensive-prospective assessment for dementia, record-review afforded less participation bias. The present study did not include reasons for NH use, biomarkers, duration of cognitive impairment at time of assessment, or temporal changes in cognition, comorbidity, or functional status. Cost data ended in 2010; with one exception,⁴⁰ previously-published costs for prospectively-identified MCI ended earlier.^29,34,35,36 The effect of subsequent changes in practice styles and availability of long-term-care options is unknown. Construction of future cost-effectiveness models would benefit from more recent baseline cost estimates, biomarker data, and serial assessments as well as incidence rates and transition probabilities of the sort collected in ongoing MCSA studies.^30,51,52

4.3. Conclusions/Implications

Efforts to model the cost-effectiveness of interventions, including disease management and potential disease-modifying-therapies, are currently hampered by a shortage of objective, reliable, baseline data on direct medical and NH costs for persons well-characterized across the cognitive spectrum. Findings provided here add to existing literature and challenge assumptions that pre-dementia medical and NH costs are negligible. Findings may help inform budgeting decisions by individuals, payers, providers, and policy-makers. The value of our estimates for models investigating cost-effectiveness of disease modification at earlier disease stages remains to be seen.¹¹