Cost effectiveness of non-drug interventions that reduce nursing home admissions for people living with dementia

Abstract

INTRODUCTION:

Six million Americans live with Alzheimer’s disease and Alzheimer’s disease and related dementias (AD/ADRD), a major health-care cost driver. We evaluated the cost effectiveness of non-pharmacologic interventions that reduce nursing home admissions for people living with AD/ADRD.

METHODS:

We used a person-level microsimulation to model the hazard ratios (HR) on nursing home admission for four evidence-based interventions compared to usual care: Maximizing Independence at Home (MIND), NYU Caregiver (NYU); Alzheimer’s and Dementia Care (ADC); and Adult Day Service Plus (ADS Plus). We evaluated societal costs, quality-adjusted life years and incremental cost-effectiveness ratios.

RESULTS:

All four interventions cost less and are more effective (i.e., cost savings) than usual care from a societal perspective. Results did not materially change in 1-way, 2-way, structural, and probabilistic sensitivity analyses.

CONCLUSION:

Dementia-care interventions that reduce nursing home admissions save societal costs compared to usual care. Policies should incentivize providers and health systems to implement non-pharmacologic interventions.

1 |. INTRODUCTION

Alzheimer’s disease and Alzheimer’s disease and related dementias (AD/ADRD) affect more than 6 million Americans.¹ AD/ADRD is one of the most costly diseases in the United States,¹ and most AD/ADRD-related costs are for long-term care services, which are paid by families and Medicaid.^2,3

Although people living with AD/ADRD have a high risk of entering a nursing home, many prefer to age with quality of life in their community.^4,5 Helping people safely age in place is also the goal of national policy initiatives.^4,6–8 The long-term care system depends on family caregivers to help people living with AD/ADRD age at home.^9–11 Caregivers may help with activities of daily living, dementia-related behavioral management, and care coordination, among other tasks.¹²

Dyadic non-pharmacologic interventions that provide family caregivers with knowledge, skills, and support tailored to their care challenges have been shown to improve the caregiver’s and person living with AD/ADRD’s quality of life, and reduce nursing home admissions.^13–17 Compared to pharmacologics, non-pharmacologic interventions are not associated with adverse events including hospitalizations and mortality.^18–20 For these reasons, non-pharmacologic interventions are recommended as first-line therapies for the management of AD/ADRD.^21–24

Several challenges have limited the implementation of non-pharmacologic AD/ADRD interventions in routine care.^25,26 First, there are limited mechanisms for providers to be reimbursed for delivering all or key components of non-pharmacologic interventions (e.g., counseling to caregivers).^27–29 In addition, these interventions are often delivered by multiple providers (e.g., community health workers) and across multiple systems (e.g., primary care and community agencies);²⁶ yet, not all of these providers/settings can bill payers for their services. Second, non-pharmacologic AD/ADRD interventions may reduce health-care costs, but in a fee-for-service model or fragmented care network, some providers/organizations who implement the intervention may not receive financial benefits from derived outcomes. For example, savings from reduced hospitalizations do not financially benefit the community-based providers delivering the intervention. Finally, there are no requirements to evaluate cost effectiveness in the United States, but these data are still useful to health-care payers (e.g., state Medicaid programs or Medicare Advantage organizations) to establish mechanisms for reimbursement.

The importance of understanding the cost effectiveness of non-pharmacologic AD/ADRD interventions is further highlighted by changes in Medicare payment models and emerging AD/ADRD therapeutics.^20,27 Accountable Care Organizations and Medicare Advantage are responsible for increasing numbers of Medicare beneficiaries.³⁰ These systems are incentivized to implement interventions that reduce health-care service use, even if they are not directly paid for the service. Furthermore, as the Centers for Medicare & Medicaid Services (CMS) determines coverage for new AD/ADRD drugs, it should consider the benefits of non-pharmacologic interventions.²⁰

To address these issues, we used an AD/ADRD microsimulation model to evaluate societal, health-care payer and family costs, benefits, and cost effectiveness of select dyadic non-pharmacologic interventions that were found in randomized trials to reduce the rate of nursing home placement of people living with AD/ADRD.

2 |. METHODS

2.1 |. Model design and study population

We adapted our previously developed discrete-time (monthly cycle) microsimulation model of the natural history of AD/ADRD from diagnosis to death to evaluate the cost effectiveness of dyadic non-pharmacologic interventions (see Appendix A for key model parameters).³ The model predicts place of residence (community or nursing home), Medicaid enrollment, costs, survival and quality-adjusted life years (QALYs) of people living with AD/ADRD based on cognition, number of functional activity limitations, and number of problematic behaviors. To better inform policy, we calibrated the parameters of the microsimulation so that the model-predicted time people spend in a nursing home is similar to the time Medicare beneficiaries with an AD/ADRD diagnosis, stratified by survival, sex, and race, spend in a nursing home (see Appendix B for details of the calibration procedure).^31,32 The model has five states: (1) community dwelling and not enrolled in Medicaid, (2) community dwelling and enrolled in Medicaid, (3) nursing home resident and not enrolled in Medicaid, (4) nursing home resident and enrolled in Medicaid, and (5) death.

2.2 |. Identifying intervention

We searched two sources to identify dyadic non-pharmacologic interventions that were found in randomized controlled trials or non-randomized controlled studies to reduce the rate of nursing home admission for community-dwelling people living with AD/ADRD: (1) the Best Practice Caregiving website, an online database of evidence-based dementia care interventions (Appendix C) and (2) a published list of evidence-based interventions that were tested for efficacy in randomized trials and which are currently being further tested for their translation in practice settings.²⁹ Four interventions met our inclusion criteria (Table 1).

TABLE 1.

Description of intervention.

Intervention, study design, and description	Hazard ratio (HR) for entering nursing home: intervention versus control	Other intervention effects modeled	Mean (SD) intervention cost per person per month
Maximizing Independence at Home (MIND)13,33 Randomized controlled trial (n = 303); Baltimore, Maryland. Delivered at home and as needed, consists of interdisciplinary care coordination to address unmet needs, care planning, linkage to services, dementia education, skill-building, care monitoring for 18 months.	Intervention effect reported over 18 months. HR: 0.63 (95% CI: 0.42,0.94)	Family time caregiving: −16.90 hours/week (95% CI: −33, −0.72)	$102 (35)a
New York University Caregiver (NYU)34–36 Randomized controlled trial (n = 406); New York, New York. Delivered within an outpatient clinic, consists of counseling and support for caregivers. Two individual and four family counseling sessions within 4 months of enrollment, caregivers encouraged to participate in weekly local support groups, and indefinite ad hoc telephone counseling.	Intervention effect for children caregivers reported over 42 months. HR for spouse caregiver: 0.717 (95% CI: 0.537, 0.958)c HR for adult child caregiver: 0.53 (95% CI: 0.28,0.99)	n/a	Months 1–12: $162 (20)b; Months 13+: $52 (10)
Alzheimer’s and Dementia Care (ADC)16 Non-randomized controlled study (n = 3249); Los Angeles, California. Delivered within a health-care system and as needed, consists of a nurse practitioner and physician that co-manage patients, conduct needs assessment, develop individual care plans, 24/7/365 access to a dementia care manager.	Intervention effect reported over 36 months. HR: 0.60 (95% CI: 0.59,0.61)	Medicare costs: −$601 per patient per quarter (95% CI: −$1198, —$5)	$106 (20)d
Adult Day Service Plus (ADS Plus)14 Non-randomized controlled study (n = 129); Philadelphia, Pennsylvania. As needed care management service embedded within an adult day program. Services include identifying caregiver concerns and needs, developing and implementing a care plan involving disease education, skills training, referral and linkages, stress reduction techniques, and periodic follow-up. Participants receive about 1 hour of face-to-face support from staff per month.	Intervention effect reported over 12 months. HR: 0.44 (95% CI: 0.20,0.94)	n/a	$74 (20)e

^aTrial reports two FTE bachelor prepared social workers (estimated annual wage $48,140) and one 0.5 FTE master’s trained social worker (estimated annual wage $51,760) were hired to deliver the intervention; total FTE annual cost $122,160 or $1221 per trial participant (n = 100) per year or $102 per person per month.

^bEstimate obtained from a prior evaluation of the intervention.

^cSpouse caregiver effect was reported over 12 years, but the child caregiver effect was evaluated over 42 months.

^dTrial reports program cost per person per quarter to be $317 or $106 per person per month.

^eEstimate based on 0.2 FTE of a Family Service Director (estimated annual wage $120,000); total FTE annual cost $24,000 or $889 per trial participant (n = 27 people per site) per year or $74 per person per month.

Abbreviations: CI, confidence interval; FTE, full-time equivalent; HR, hazard ratio; SD, standard deviation.

2.3 |. Interventions

Below we provide a brief description of the interventions (see Appendix D for additional details). Maximizing Independence at Home (MIND) is an at-home, care coordination intervention that consists of care planning, skill building, referral/linkage to services, and care monitoring.^13,33 Intensity of visits with a care coordinator was based on an individualized care plan. MIND was evaluated in an 18-month randomized trial that enrolled 303 dyads of caregivers and people with AD/ADRD.

The New York University (NYU) Caregiver intervention is implemented in an outpatient clinic and provides caregivers with six counseling sessions over 4 months plus lifetime ad hoc support, and encourages caregivers to participate in a weekly support group.^34–36 The NYU intervention was originally evaluated in a randomized controlled trial that enrolled spouse caregivers and had 9.5 years of follow-up. A second randomized trial evaluated the NYU intervention with adult child caregivers over 3.5 years. In our modeling analysis, we assumed the treatment effect of NYU was 3.5 years regardless of the caregiver’s relationship to the person with dementia.

The Alzheimer’s and Dementia Care (ADC) intervention is implemented in a health-care system and provides people living with AD/ADRD and their caregivers a needs assessment, individual care plans, and round-the-clock access to a care manager.¹⁶ Patients engage with providers as needed. ADC was evaluated using a case-control design in which 1083 intervention participants were statistically matched to a comparison cohort of 2166 people with dementia. Outcomes were evaluated over 3 years.

Adult Day Services Plus (ADS Plus) augments adult day care services with staff providing face-to-face caregiver support, disease education, care management, and skill building.¹⁴ ADS Plus was evaluated in a 12-month non-randomized controlled study that enrolled 129 caregivers.

2.4 |. Modeling intervention effects

We simulated individuals once they were diagnosed with AD/ADRD. However, the interventions we modeled do not explicitly target people at the time of diagnosis and instead are recommended based on eligibility criteria. We used each study’s sample characteristics to determine the time from diagnosis a simulated person would be eligible for the intervention (Appendix E).

We modeled each intervention’s effect as a reduction of the rate (i.e., hazard ratio [HR]) of a community-dwelling person living with AD/ADRD entering a nursing home (Table 1). We assumed the interventions were provided in addition to usual care. Time to nursing home placement was the primary outcome in the original MIND, NYU, and ADC studies and a secondary outcome in ADS Plus. In the base-case analysis, we applied a HR to the hazard of entering a nursing home for the follow-up period in the source studies. Once the treatment effect ended, we assumed the rate of entering a nursing home (or any other modeled treatments) returned its natural history. Simulated patients could also stop treatment if they died or once they entered a nursing home. A key modeling assumption is the effects observed in the literature generalize to the simulated population, which is representative of an average Medicare beneficiary with AD/ADRD.

The studies of the four interventions also reported statistically significant effects on other outcomes, which we modeled based on their availability in the literature and amenability to inclusion in our model. Specifically, we modeled MIND’s effect on reducing family caregiving hours (−16.90 hours/week; 95% confidence interval [CI]: −33, −0.72) and ADC’s effect on reducing Medicare expenditures (−$601 per patient per quarter; 95% CI: −$1198, −$5).

2.5 |. Health-related quality of life

We incorporated published health-related quality-of-life weights for people living with AD/ADRD into the microsimulation.³⁷ The quality-of-life weights are reported as a function of a person with AD/ADRD’s dementia stage (mild, moderate, and severe) and place of residence (community or nursing home). On average, people with AD/ADRD and living in a nursing home have a lower quality of life than those in the community (Appendix A).

2.6 |. Cost

The microsimulation model predicts monthly Medicare expenditures, Medicaid expenditures, out-of-pocket medical care expenditures, out-of-pocket nursing home expenditures, time (and value [$19.71/hour]) of family caregiving and time (and value [$23/hour]) of paid caregiving.^38,39 In our primary analysis we evaluated costs from a societal perspective, which includes all the above expenditures. In secondary analyses, we evaluated costs separately from a family (out-of-pocket expenditures and the value of family caregiving) and health-care payer (Medicaid and Medicare expenditures) perspectives.Costs are reported in 2015 dollars.

Per the original studies, ADC costs $106 per dyad per month and NYU costs $162 per dyad per month in the first year and then $52 per dyad per month.^16,36 The cost of delivering MIND and ADS Plus were not reported in source data. We used study data on the type and number of personnel needed to deliver each intervention and estimated the monthly cost of delivering MIND and ADS Plus to be $102 and $74, respectively (Table 1).

2.7 |. Cost-effectiveness analysis

Following current guidelines, we conducted the cost-effectiveness analysis base case using the expected discounted costs and QALYs across all simulations.⁴⁰ First, we sampled 1000 sets of the model parameters from distributions reflecting their uncertainty. For each set, we simulated 5000 individuals who received the intervention and 5000 who received usual care and computed the mean lifetime costs and QALYs for each intervention (10,000,000 total simulations). We applied a 3% annual discount rate to future costs and benefits.⁴¹ We calculated the total mean costs and benefits across all 1000 parameter sets and then calculated the incremental cost effectiveness of each intervention relative to its corresponding usual care. The incremental cost effectiveness is the additional cost of a strategy divided by the additional QALYs compared to usual care. In a secondary analysis and consistent with a recent Institute for Clinical and Economic Review report on the cost effectiveness of aducanumab, we estimated the incremental cost-effectiveness ratio with benefit measured as years in the community.²⁰ These interventions are implemented in different settings (e.g., adult day services or the person with dementia’s home), target people at different stages of dementia and with different background mortality and nursing home rates, so we compared each intervention to their own standard of care.

We conducted one-way, two-way, structural, and probabilistic sensitivity analyses from each perspective.⁴¹ In one-way sensitivity analyses we independently varied the point estimate for the intervention effects and the cost of intervention. To account for variation in implementation success, we varied the proportion (50% or 75%) of simulated people who participated in the intervention. In this scenario, all people in the intervention group, regardless of whether they participated or not, incurred an intervention cost. We varied parameters between optimistic (would favor the intervention) or pessimistic (would favor usual care) assumptions. For MIND and NYU, we varied the intervention effects based on the study-reported CI bounds. We obtained the intervention effect of ADC and ADS Plus from a quasi-experimental study, so we applied a bias modeling approach described by Turner et al., and varied intervention effects based on our adjustment for internal biases (Appendix D.3 and D.4).⁴² We varied the cost of the intervention between their CI bounds. In two-way sensitivity analyses, we simultaneously varied the HR of entering the nursing home and the proportion of simulated people in the intervention group who actually participate in the intervention. Separately, we conducted a structural sensitivity analysis in which we modeled treatment effects assuming it took 12 months to reach maximum effectiveness and the effect extended 12 months after the last reported follow-up. Finally, we plotted the cost-effectiveness acceptability curves (CEAC) and frontiers (CEAF) for each perspective resulting from the probabilistic sensitivity analyses.⁴¹

Statistical analyses were conducted using R 3.6.3 and an interactive version of our analyses is available at http://tiny.cc/dementiacal.^43,44 The study follows the Consolidated Health Economic Evaluation Reporting Standards (CHEERS).

2.8 |. Role of the funding source

The National Institute on Aging had no role in the design, analysis, or reporting of results.

3 |. RESULTS

3.1 |. Base-case results

As shown in Table 2, MIND compared to usual care reduced lifetime societal costs by $13,282, and yielded 0.002 more QALYs and 0.057 more years (21 days) in the community than usual care. Thus, MIND cost less and was more effective (i.e., dominated or was cost-saving) than usual care from a societal perspective. MIND costs more but was more effective than usual care from a payer perspective (incremental cost effectiveness: $268,476/QALY) and it dominated usual care from a family perspective (Appendix F).

TABLE 2.

Societal and health-care payer perspective: intervention cost, incremental lifetime costs ($), quality-adjusted life-years, and incremental cost-effectiveness ratio of each intervention compared to usual care.

Comparison	Intervention cost, $	Incremental benefit		Societal perspective			Health-care payer perspective
		QALYs	Years (days) in the community	Incremental lifetime costs, $a	ICER	ICER, $/year in the community	Incremental lifetime costs, $a	ICER, $/QALY	ICER, $/year in the community
Maximizing Independence at Home (MIND) versus usual care	$1445	0.002	0.057 (21)	−$13,282	Cost saving	Cost saving	$617	$268,476	$10,901
NYU Caregiver (NYU) versus usual care	$2571	0.005	0.118 (43)	−$5297	Cost saving	Cost saving	$12	$2224	$98
Alzheimer’s and Dementia Care (ADC) versus usual care	$2345	0.003	0.061 (22)	−$3668	Cost saving	Cost saving	−$720	Cost saving	Cost saving
Adult Day Service Plus (ADS Plus) versus usual care	$762	0.002	0.038 (14)	−$2813	Cost saving	Cost saving	$192	$126,148	$5015

Abbreviations: Cost saving, the intervention costs less and is more effective than usual care; ICER, incremental cost-effectiveness ratio; QALY, quality-adjusted life-year.

^aIncremental cost is the difference in treatment [inclusive of treatment cost] – usual care cost.

From a societal perspective, NYU reduced costs ($5297 net savings) and generated 0.005 more QALYs and 0.118 additional years (43 days) in the community compared to usual care (Table 2). From a health-care payer perspective NYU cost more and was more effective than usual care (incremental cost effectiveness: $2224/QALY). NYU dominated usual care from a family perspective.

ADC reduced lifetime societal costs ($3668 net savings) and generated 0.003 additional QALYs and 0.061 additional years (22 days) in the community compared to usual care (Table 2). ADC also dominated usual care from health-care payer and family perspectives (Appendix F).

Compared to usual care, ADS Plus decreased societal costs by $2813 and generated 0.002 additional QALYs and 0.038 additional years (14 days) in the community (Table 2). ADS Plus had an incremental cost effectiveness of $126,148/QALY from a health-care payer perspective (Table 2), and dominated usual care from a family perspective (Appendix F).

3.2 |. Sensitivity analysis

Figure 1 present results of the one-way sensitivity analyses from the societal perspective (Appendix G reports one-way sensitivity analyses from the health-care payer and family perspectives). The results for MIND (Figure 1A) and NYU (Figure 1B) did not materially change in one-way sensitivity analyses from a societal perspective. ADC cost more but was still more effective than usual care when the cost to Medicare per month (base case $200 cost savings) was set to the pessimistic assumption of $33 per month (net increase; Figure 1C). Under this assumption, the intervention had an incremental cost-effectiveness ratio from a societal perspective of $9611/QALY. Finally, when the HR of ADS Plus was set to the pessimistic assumption, the intervention cost more and was less effective than usual care (Figure 1D).

FIGURE 1.

One-way sensitivity analysis from a societal perspective. A, Maximizing Independence at Home (MIND) versus usual care. B, New York University Caregiver (NYU) versus usual care. C, Alzheimer’s and Dementia Care (ADC) versus usual care. D, Adult Day Service Plus (ADSPlus) versus usual care.

Appendix H presents the results of the two-way sensitivity analyses. These sensitivity analyses indicate that our results for NYU and ADS Plus were sensitive to the simultaneous changes of the hazard of nursing home entry and the proportion of people who participate in the intervention. For example, when both the hazard of nursing home entry and the proportion of people who participate in the intervention were set to their pessimistic assumption, ADS Plus was estimated to cost more and be less effective than usual care. Results from the structural sensitivity analysis show that a delay in the time it takes for the intervention effect to reach maximum effectiveness reduces the amount of cost savings associated with each intervention, but the overall conclusions remain consistent with the base-case analysis (Appendix I). Finally, Figure 2 illustrates the CEAC and CEAF for each strategy compared to usual care.

FIGURE 2.

CEAC and CEAF. The CEAC displays each strategy’s probability of being cost effective across all the probabilistic sensitivity analysis simulations over a range of cost-effectiveness thresholds. The CEAF illustrates the strategy with the highest expected net monetary benefit at each cost-effectiveness threshold; the net monetary benefit is defined as the difference between the product of QALYs and the cost-effectiveness threshold minus the costs for each of the simulations of the probabilistic sensitivity analyses. ADC, Alzheimer’s and Dementia Care; ADS, Adult Day Service Plus; CEAC, cost-effectiveness acceptability curves; CEAF, cost-effectiveness acceptability frontier; MIND, Maximizing Independence at Home; NYU, New York University; QALY, quality-adjusted life-year.

4 |. DISCUSSION

For this microsimulation study, we identified four evidence-based interventions that reduce the rate of nursing home admission and improve quality of life of people living with AD/ADRD.^{13,14,16,33–35} Our study contributes to the literature by providing policy makers and health-care administrators with additional data on the potential long-term costs, benefits, and cost effectiveness of these interventions at a population level. In the base-case analyses, we did not compare these interventions because they targeted different populations and were implemented in different settings. The interventions are complementary because they target different segments of the population.

All four interventions had small QALY improvements and increased time in the community compared to usual care. The impact of all four interventions on costs depends upon whether a societal, health-care payer, or family perspective is considered. From a societal perspective, the interventions saved costs compared to usual care by reducing out of-pocket and Medicaid nursing home expenditures. From a healthcare payer perspective, ADC cost less than usual care, whereas NYU, MIND, and ADS Plus cost more for delivery but were more effective than usual care. The extra costs of NYU, MIND, and ADS Plus were due to the cost of delivering the programs. The incremental cost effectiveness per year in the community for NYU, MIND, and ADS Plus was substantially less than the median annual cost ($93,075) of a nursing home and less than the incremental cost per QALY of aducanumab.^20,45,47

With regard to a family perspective, all four interventions cost less and were more effective (i.e., cost savings) compared to usual care. The large cost saving to families was primary attributed to reducing out-of-pocket nursing home expenditures. AD/ADRD family caregivers experience significant financial pressures with dementia care, paying on average $33,000 per year (including the value of informal care) to care for a person with AD/ADRD.¹ These costs are particularly challenging for families experiencing financial strain, and minority populations with disproportionate AD/ADRD risk.⁴⁶ Financial strain contributes to the burden that families experience in dementia care.⁴

The model predicted days in the community are lower than those previously reported for MIND and NYU. ADC and ADS Plus reported the risk of transiting to a nursing home, but not days in the community. We calibrated the average time a person with AD/ADRD spends in a nursing home to Medicare data. In the simulated population, which is representative of the entire Medicare population, there is a lower overall risk of transitioning to a nursing home than those observed in the MIND and NYU studies, and this difference is likely responsible for the modeling approach resulting in lesser effects compared to the original MIND and NYU studies. A key modeling assumption is that the treatment effects observed in the literature generalize to the average population of US adults with AD/ADRD.

There are challenges to implementing non-pharmacologic AD/ADRD interventions in diverse care settings.⁴⁸ Intervention fidelity can be difficult to monitor, the interventions may require extensive adaptation to fit within workflows, and families may not fully engage in the programs. Our one-way sensitivity analyses provide insight into the potential effect of imperfect implementation. First, even when assuming that only 50% of people enrolled in an intervention actually receive the intervention, all four interventions were still cost saving from a societal perspective. Second, results were mostly insensitive to variation in treatment effects.

Several study limitations should be noted. We indirectly infer health-related quality-of-life benefits which may result in under- or overestimating the QALY benefits of the selected interventions. The studies showed improved quality of life for participants who received the intervention relative to usual care. Still, we did not directly model these benefits due to differences in the quality-of-life measures used by the studies and microsimulation. Although we modeled time caregiving, we did not model the effect of interventions on other spillovers, including caregiver QALYs. Several of the interventions reported improved well-being for caregivers, meaning our modeling approach may underestimate intervention benefits. Excluding caregivers’ QALYs may not affect our conclusion from the family and societal perspective because the interventions dominate usual care. However, including caregivers’ QALYs would likely make the interventions more cost effective (i.e., lower the incremental cost-effectiveness ratio) from a payer perspective.⁴⁹ Also, we estimated MIND and ADS Plus delivery costs based on the studies’ descriptions of personnel needed to provide the intervention and thus our assumptions may over- or under-inflate program costs. Another limitation is that outcomes for each intervention were reported over different and relatively short time frames. We modeled effects over longer periods of time, but the actual long-term effects of these interventions has not been quantified in studies using real-world data. We analyzed the interventions using a similar approach, but the interventions differ in terms of setting and target population. We modeled interventions using a summary effect size, and our analysis does not allow for understanding the most effective components of these interventions. Our findings may not generalize to health systems outside the United States due to differences in financing and delivery structures of long-term care. Finally, we limited our analysis to interventions that reduced nursing home admissions and this analysis did not evaluate potentially promising interventions that reduce cognition, function, or behavior.

As CMS derives a reimbursement determination for a new class of AD/ADRD drugs, it should also recognize the value of non-pharmacologic interventions and offer a reimbursement structure. The Institute for Clinical and Economic Review concluded that the incremental cost-effectiveness ratio of aducanumab, a new pharmacologic that aims to slow the rate of cognitive decline by removing amyloid beta in people with mild cognitive impairment or mild AD/ADRD, from a societal perspective, was $1,270,000/QALY.²⁰ In contrast, we found that four non-pharmacologic interventions that improve clinical outcomes for people with moderate AD/ADRD including reducing the rate of nursing home admission cost less and were more effective than usual care from societal perspectives. Our findings support advancing payment models to cover evidence-based AD/ADRD non-pharmacologic interventions.

RESEARCH IN CONTEXT.

1. Systematic review: Several non-pharmacologic interventions that provide family caregivers with knowledge and support tailored to their care challenges have been shown to reduce nursing home admissions. The cost effectiveness of these programs is unknown, but such data are needed to establish reimbursement mechanisms.

2. Interpretation: We used a microsimulation to evaluate societal costs and quality-adjusted life years of four interventions that reduce nursing home admission compared to usual care: Maximizing Independence at Home, NYU Caregiver, Alzheimer’s and Dementia Care, and Adult Day Service Plus. These interventions cost less and are more effective from a societal perspective compared to usual care.

3. Future directions: This work uses simulation modeling to predict costs and benefits of evidence-based interventions. First, pragmatic randomized trials and well-designed retrospective studies using real-world data are needed to evaluate the implementation of non-pharmacologic interventions in diverse settings. Second, policies are needed to incentivize providers and health systems to implement non-pharmacologic interventions.

APPENDIX A: KEY MODEL PARAMETERS

A full description of the model and parameters is described in Jutkowitz et al.³ (Table A.1 and Figure A.1).

FIGURE A.1.

Model Structure.

*Individuals can transition to death (not shown) from any state.

TABLE A.1.

Model Parameters

Parameter	Description	Key Reference
Clinical features: cognition, function, and behaviors	Linear mixed effects models estimated with data from the National Alzheimer’s Coordinating Center.	Jutkowitz et al.50
Risk of transition from community to nursing home	Weibull proportional hazards model estimated with data from the National Alzheimer’s Coordinating Center. We calibrated selected model parameters to Medicare data on time in a nursing home (see Appendix B).
Risk of transition from nursing home to community	We linked 2007–2015 Medicare Beneficiary Summary File to the Residential History File (Medicare claims and MDS assessments; see Intrator et al. 2011) to trace a beneficiary’s location of residence (community or nursing home). We then identified beneficiaries ≥70 years of age who were diagnosed with AD/ADRD (CCW definition) in the community. Among this cohort, we identified 416,569 unique nursing home stays representing 260,669 unique beneficiaries. We used a competing risk Weibull proportional hazard model to estimate the time until a beneficiary returned to the community, died in the nursing home, or was censored. Race, sex, age of AD/ADRD onset, and an interaction term between race and sex were included as covariates.a	Intrator et al.54
Medicaid enrollment	We used 2007–2015 Medicare Beneficiary Summary File linked to the Residential History File to identify beneficiaries ≥70 years of age who were diagnosed with AD/ADRD (CCW definition) in the community. We then determined whether these beneficiaries ever enrolled in Medicaid after their AD/ADRD diagnosis. Among this cohort, we identified 615,304 unique beneficiaries with an AD/ADRD diagnosis. We used a Weibull proportional hazards model to estimate the time to Medicaid enrollment. Race, sex, age of dementia onset, number of months spent in a nursing home, and an interaction term between race and sex were included as covariates.a	Intrator et al.54
Costs	Time receiving family caregiving	Jutkowitz et al.51
	Time receiving paid caregiving	Jutkowitz et al.51
	Medicare expenditures	Jutkowitz et al.52
	Out-of-pocket expenditures	Jutkowitz et al.50
	Long-term care facility expenditures	MetLife Mature Market Institute38
	Medicaid expenditures	Garfield et al.55 Bharmal et al.56
Mortality	Age-, sex-, and race mortality rates based on age of dementia diagnosisa	Mayeda et al.57

^aThe data sources used to predict these outcomes did not include information on dementia clinical features of cognition, function, and behavior.

Abbreviations: AD/ADRD, Alzheimer’s disease and Alzheimer’s disease and related dementias; CCW, Chronic Conditions Data Warehouse; MDS, Minimum Data Set.

TABLE A.2.

Preference weights for people living with AD/ADRD^a.

Stage	Community	Nursing home
Mild (MMSE = 21–25)b	0.68	0.71
Moderate (MMSE = 11–20)	0.54	0.48
Severe (MMSE = 0–10)	0.37	0.31

Abbreviations: AD/ADRD, Alzheimer’s disease and Alzheimer’s disease and related dementias; MMSE, Mini-Mental State Examination.

^aPreference weights obtained from Neumann et al.³⁷

^bOur model uses the MMSE to measure cognitive impairment and does not incorporate the Clinical Dementia Rating Scale. We used a published crosswalk (Perneczky R et al. Am J Geriatr Psychiatry. 2006) to map MMSE scores to the Clinical Dementia Rating Scale to arrive at quality-of-life weights.

APPENDIX B: MODEL CALIBRATION

B.1 |. Calibration problem and target data

The microsimulation model predicts the probability of non-Medicare reimbursed nursing home admission using data from National Alzheimer’s Coordinating Center Uniform Data Set and a Weibull proportional hazards regression. The regression includes coefficients for cognition, functional activity limitations, behavior, and demographic characteristics (age, sex, race, and education). The Weibull cumulative hazard function is $\mathrm{H}\left(\mathrm{t}\right)=\mathrm{e}\mathrm{x}\mathrm{p}\left(\mathbf{X}\beta \right){\mathrm{t}}^{\mathrm{a}}$ where $\mathbf{X}$ is the covariates, $\beta$ is the regression coefficients, and a is the Weibull shape parameter.

Thus, AD/ADRD microsimulation links a person with AD/ADRD’s clinical features with the risk of entering a nursing home. The Weibull proportional hazards regression was estimated with data from the National Alzheimer’s Coordinating Center Uniform Data because it has information on AD/ADRD clinical features. However, the National Alzheimer’s Coordinating Center Uniform Data may not generalize to the broader population with AD/ADRD. Therefore, we calibrated the coefficients of the Weibull regression to the time Medicare beneficiaries (stratified by survival, sex, and race) with AD/ADRD spend in a nursing home.

We used Medicare data to determine the average time a person with AD/ADRD spends in a nursing home. Specifically, we used the Chronic Conditions Warehouse to identify fee-for-service Medicare beneficiaries with an AD/ADRD diagnosis for the first time between 2012 and 2015. Among this denominator, we extracted a random sample of 250,000 beneficiaries. We then used Medicare Part A and Part B claims, the Minimum Data Set (v2.0), the Inpatient Rehabilitation Facility Patient Assessment Instrument, and the home health agency Outcome and Assessment Information Set to trace a beneficiary’s location of residence on each day after their initial AD/ADRD diagnosis. Disenrollment from Medicare and death are absorbing states. We calculated the duration of days spent in the community or in the nursing home as a non-Medicare reimbursed resident.

Following the calibration procedure, the microsimulation model–predicted time people spend in a nursing is now similar to values observed for Medicare fee-for-service beneficiaries. This procedure yielded a posterior distribution from which we obtained 1000 samples used for our projections and analyses.

B.2 |. Calibration target

Table B.1 shows the average time spent in a nursing home for people with an AD/ADRD diagnosis stratified by sex, race (White, Black, and other), and survival after a diagnosis (2–3 years, 3–4 years, 4–5 years, and 5–6 years). The root mean square error for the difference between original model prediction and target is 3.774.

TABLE B.1.

Original model prediction and calibration target.

Race and sex	Years to death after AD/ADRD diagnosis	Average months in the nursing home
		Original model prediction	Target outcome from Medicare data
White men	2–3	1.93	2.36
	3–4	4.71	3.56
	4–5	9.49	4.62
	5–6	15.41	6.11
White women	2–3	2.46	3.28
	3–4	6.14	5.03
	4–5	11.75	7.23
	5–6	19.02	9.65
Black men	2–3	1.09	2.91
	3–4	2.74	4.01
	4–5	5.59	6.08
	5–6	10.49	7.30
Black women	2–3	1.50	2.71
	3–4	3.38	4.69
	4–5	6.79	6.74
	5–6	12.44	8.61
Other men	2–3	1.19	2.15
	3–4	2.95	3.59
	4–5	6.21	4.85
	5–6	10.71	5.46
Other women	2–3	1.74	2.26
	3–4	4.19	3.44
	4–5	8.59	5.09
	5–6	13.36	6.79

Abbreviation: AD/ADRD, Alzheimer’s disease and Alzheimer’s disease and related dementias.

B.3 |. Calibration parameters and prior distribution

Table B.2 shows the calibrated parameters, their prior value, and their prior distribution. We calibrated the shape, intercept, sex, and race coefficient of the Weibull regression.We calibrated the shape and intercept because they represent the population’s background risk. In addition, we calculated sex and race coefficients because these data are also available in Medicare data (i.e., target outcomes). We did not calibrate the dementia clinical feature parameters because the target data do not include these measures and we did not want to over/underestimate their values to achieve a better fit for the target outcome.

TABLE B.2.

Calibration parameters.

Calibration parameter	Prior value	Prior distribution
Weibull shape	1.80	Unif (1,3)
Intercept coefficient	−10.32	−3.58 + (−3.84* Weibull shape)a
Sex coefficient (male = 1; female = 0)	−0.19	Unif (−1.7,0.7)
Race coefficient (ref = White)
Black	−0.59	Unif (−1.7,0.7)
Other	−0.52	Unif (−1.7,0.7)

^aThere is a linear relationship between the Weibull shape and Intercept coefficient, so intercept values are proposed as a linear function of the shape parameter.

B.4 |. Calibration algorithm, sample size of posterior distribution, and distance measure

We applied two approximate Bayesian calibration (ABC) algorithms: ABC rejection sampler and Markov chain Monte Carlo (MCMC). We simulated a cohort of 100 people from each stratification for a total of 2400 simulated people. These 2400 people were simulated once for each set of parameters tested, for a total of 12,000 simulations per step.

In the ABC rejection sampler, 100,000 steps were performed and the top 1000 parameters were accepted. The ABC-MCMC chain was run for 25,000 steps, with a burn-in period of 5000 and a thinning of 20, for a sample of 1000 parameters from the approximate posterior distribution (Figure B.1). The distance function used was root mean square error (RMSE).

FIGURE B.1.

Approximate posterior distribution of calibrated coefficients. ABC, approximate Bayesian calibration; MCMC, Markov chain Monte Carlo.

B.5 |. Posterior predictive distribution and calibration target

Table B.3 shows the posterior predictive distribution for each calibration target.

TABLE B.3.

Posterior predictive distribution and calibration target.

	Years to death after AD/ADRD diagnosis	Average months in the nursing home before calibrationa		Average months (90% credible interval) in the nursing home after calibration: ABC rejection samplerb	Average months (90% credible interval) in the nursing home after calibration: ABC MCMCc
		Original model prediction	Target outcome from Medicare data
White men	2–3	1.93	2.36	1.30 [0.68, 2.15]	1.33 [0.67, 2.15]
	3–4	4.71	3.56	2.71 [1.64, 4.02]	2.72 [1.70, 4.00]
	4–5	9.49	4.62	4.71 [3.11, 6.51]	4.59 [2.73, 6.43]
	5–6	15.41	6.11	7.44 [4.95, 9.95]	7.16 [4.72, 9.73]
White women	2–3	2.46	3.28	1.67 [0.92, 2.60]	1.68 [0.93, 2.64]
	3–4	6.14	5.03	3.72 [2.46, 5.24]	3.73 [2.44, 5.14]
	4–5	11.75	7.23	6.52 [4.66, 8.61]	6.43 [4.58, 8.41]
	5–6	19.02	9.65	9.66 [6.99, 12.43]	9.43 [6.81, 12.34]
Black men	2–3	1.09	2.91	1.31 [0.69, 2.16]	1.40 [0.67, 2.15]
	3–4	2.74	4.01	2.85 [1.76, 4.32]	2.94 [1.70, 4.25]
	4–5	5.59	6.08	4.9 [3.14, 6.92]	4.96 [3.23, 6.75]
	5–6	10.49	7.30	7.69 [5.15, 10.35]	7.62 [5.05, 10.29]
Black women	2–3	1.50	2.71	1.66 [0.95, 2.65]	1.76 [0.94, 2.68]
	3–4	3.38	4.69	3.73 [2.42, 5.33]	3.84 [2.38, 5.26]
	4–5	6.79	6.74	6.35 [4.44, 8.48]	6.46 [4.56, 8.51]
	5–6	12.44	8.61	9.56 [6.68, 12.30]	9.47 [6.77, 12.07]
Other men	2–3	1.19	2.15	1.03 [0.48, 1.75]	1.12 [0.54, 1.82]
	3–4	2.95	3.59	2.08 [1.13, 3.28]	2.23 [1.21, 3.41]
	4–5	6.21	4.85	3.99 [2.35, 5.92]	4.22 [2.51, 6.08]
	5–6	10.71	5.46	5.68 [3.33, 8.13]	5.97 [3.64, 8.69]
Other women	2–3	1.74	2.26	1.28 [0.63, 2.15]	1.43 [0.73, 2.23]
	3–4	4.19	3.44	2.66 [1.55, 3.99]	2.93 [1.82, 4.31]
	4–5	8.59	5.09	4.71 [2.90, 6.68]	5.08 [3.13, 7.01]
	5–6	13.36	6.79	7.53 [4.78, 10.48]	7.84 [5.21, 10.76]

Note: RMSE is the average difference (months) between the target and prediction.

Abbreviation: ABC, approximate Bayesian calibration; MCMC, Markov chain Monte Carlo; RMSE, root mean square error.

^aAverage RMSE for the difference between targets and original model predictions: 3.774.

^bAverage RMSE for the difference between accepted parameter predictions and target (ABC Rejection Sampler): 1.47.

^cAverage RMSE for the difference between accepted parameter predictions and target (ABC-MCMC): 1.40.

B.6 |. Distributions of predictions and targets

Figure B.2 shows the distributions of predictions from the ABC Rejection Sampler. Figure B.3 shows the distributions of predictions from the ABC MCMC.

FIGURE B.2.

Distributions of predictions and targets ABC rejection sampler. ABC, approximate Bayesian calibration.

FIGURE B.3.

Distributions of predictions and targets ABC MCMC. ABC, approximate Bayesian calibration; MCMC, Markov chain Monte Carlo.

APPENDIX C: SEARCH STRATEGY TO IDENTIFY INTERVENTIONS

We searched the Best Practice Caregiving Registry (https://bpc.caregiver.org/#home) on August 3, 2021 using the “Study Findings” filter for “Nursing Home Admission Long/Short-Term Quantity.” This search identified two interventions that reduce nursing home admissions.

APPENDIX D: DETAILED DESCRIPTION OF INTERVENTIONS

D.1 |. Maximizing Independence at Home (MIND)

1. Outcome Paper(s) Used for Modeling:

   1. Samus et al.¹³
   2. Tanner et al.³³

2. Intervention Name: MIND

3. Intervention Description: Participants, care partners, and primary care physician receive results of care needs assessment and 18 months of care coordination. The care coordination team is comprised of a non-clinical community worker (Coordinator), RN, and geriatric psychiatrist. Care coordination is manualized and aims to identify needs, care planning, dementia education, skill building, coordination/referral/linkage to services, and care monitoring.

4. Comparison Name: Augmented usual care.

5. Comparison Description: Participants, care partners, and their primary care physicians in augmented usual care receive written results of a needs assessment, recommendations to address unmet needs, and a resource guide with contact information on local and national aging service organizations.

6. Setting: Community/home

7. Target Population: Community-dwelling people living with dementia and their care partner. To participate in the trial, community-dwelling people with dementia had to be 70+ years, have a study partner who was willing to participate in study visits, met Diagnostic and Statistical Manual of Mental Disorders, 4th edition, Text Revision criteria for dementia of cognitive disorder not otherwise specified, and have at least one unmet care need.

8. Intervention Intensity: Up to 18 months of active treatment. Two in-home visits and monthly encounters were pre-specified. All other encounters were determined based on individual need.

9. Study Design/Unit of Randomization: Randomized controlled trial (RCT)/People.

10. Outcome Measure Modeled:

    1. 18-month HR of leaving the home: 0.63 (95% CI: 0.42, 0.94).

       • We sampled the HR from a lognormal (0.63, 0.21) distribution.
    2. Time caregiving: change in augmented usual care—change in intervention from baseline to 18 months in time spent with care recipient: – 16.90 hours/week (95% CI: −33, −0.72)
       • The microsimulation model predicts hours of family caregiving based on a two-part regression model. Part 1 is a logistic regression model that predicts whether a person receives any family caregiving. Part 2 is a log-link gamma distribution regression that predicts the amount of family caregiving among those who received any caregiving. We modeled the effect of MIND on time spent caregiving using a sample importance replacement calibration (SIR) approach. The target outcome was the difference of −16.90 (95% CI: −33, −0.72) caregiving hours in a week or 73.43 (95% CI: −143.39, −3.13) fewer monthly caregiving hours. We calibrated the intercept of part 2 of the informal caregiving regression.

         Calibration parameter	Original parameter value	Prior distribution	Target outcome from trial difference in caregiving hoursa
         Intercept	2.53	Unif (1,3)	−73.43 hours/month

         ^a

         Trial reports 18-month difference. We assume the difference was constant for the duration of the trial.
    3. Posterior Distributions
       

11. Cost of Delivering the Intervention (mean, SD): $102 (35)

12. Structural Assumptions: Intervention has no effect on mortality. The intervention improved patient quality of life on the Quality of life in AD (QOL-AD) which we did not directly model.

D.2 |. NYU caregiver intervention

1. Outcome Paper(s) Used for Modeling:

   1. Mittelman et al.³⁴
   2. Gaugler et al.³⁵

2. Intervention Name: NYU Caregiver Intervention

3. Intervention Description: Intervention consist of two individual and four family counseling sessions, opportunity to participate in weekly support group, and ad hoc telephone counseling.

4. Comparison Name: Usual care.

5. Comparison Description: Care as usually provided by the NYU Alzheimer’s Disease Research Center, which consists of linkages to services and help upon request.

6. Setting: Community/home

7. Target Population: Caregivers of community-dwelling people diagnosed with AD/ADRD.

8. Intervention Intensity: Counseling occurs within 4 months of enrollment. Support was provided to caregivers for an unlimited time.

9. Study Design/Unit of Randomization: RCT/Participants.

10. Outcome Measure Modeled:

    1. 11-year nursing home placement HR for when caregiver is a spouse: 0.717 (95% CI: 0.537, 0.958).*We sampled the HR from a lognormal(log[0.717], 0.1387) distribution for people with a spouse caregiver.
    2. 3.5-year nursing home placement HR for when caregiver is an adult child: 0.53 (95% CI: 0.28, 0.99). We sampled the HR from a lognormal(log[0.53], 0.32217228) for people with Child caregiver.

    *We modeled both the spouse and adult child caregiver effect over 3.5 years.

11. Cost of Delivering the Intervention (mean, SD): Months 1–12: $162 (20); Months 13+: $52 (10). Obtained from Foldes et al.³⁶

12. Structural Assumptions: Intervention has no effect on mortality. The intervention has an effect on reducing behavioral symptoms, which we did not model due to differences in model versus trial measures.

D.3 |. Alzheimer’s and Dementia Care

1. Outcome Paper(s) Used for Modeling:

   1. Jennings et al.¹⁶

2. Intervention Name: Alzheimer’s and Dementia Care (ADC)

3. Intervention Description: Participants with dementia were comanaged by nurse practitioner dementia care managers and physicians. Components of the intervention include a needs assessment, individual care plans, monitoring and updating care plans, 24/7/365 access to a dementia care manager.

4. Comparison Name: Augmented usual care.

5. Comparison Description: Matched comparison cohort of Medicare fee-for-service beneficiaries living in the same zip code as people in the treatment group. Propensity scores used to match.

6. Setting: Community/home and primary care.

7. Target Population: Community-dwelling people with an International Classification of Diseases 9th Revision dementia code. Had to be a Medicare fee-for-service beneficiary.

8. Intervention Intensity: Not reported, but intervention effects are reported to 3 years.

9. Study Design/Unit of Randomization: quasi-experimental.

10. Outcome Measure Modeled:

    1. 3-year HR of being admitted to a long-term care facility: 0.60 (95% CI: 0.59, 0.61).

       • In sensitivity analysis we applied Turner et al.’s⁴² method to adjust for internal bias. Following Turner et al.’s proposed method to account for internal biases, we answered the following question for each domain: “Even if there were no intervention effect in this study, what apparent effect might be induced by this bias? Risk lower in intervention group [or higher in control group] or risk lower in control group [or higher in intervention group].
         Each reviewer is asked to mark a 67% range on the following elicitation scale for each type of bias. The aggregated results from each reviewer are used to calculate a total bias. The following correspondence between qualitative judgments of severity and the range limits are given: none (1), low (0.9–1), medium (0.7–0.9), and high (less than 0.7).”

         	Selection bias	Performance bias	Attrition bias	Detection bias	Total bias
         Reviewer 1	(0.8, 1.0)	(0.9, 0.9)	(0.95, 0.95)	(0.99, 0.99)
         Reviewer 2	(0.7L, 0.9R)	(0.9, 0.9)	(0.9, 1.0)	(0.95, 0.95)
         Reviewer 3	(0.9, 1.0)	(0.99, 0.99)	(0.9, 1.0)	(0.99, 0.99)
         Aggregated	(0.8, 1.0)	(0.9, 0.9)	(0.9, 1.0)	(0.99, 0.99)
         mu_j ± sigma_j	−0.11 ± 0.11	0 ± 0.11	−0.05 ± 0.05	0 ± 0.01	−0.16 ± 0.28

         Note: mu_j is the mean of the bias = [log(average lower bound) + log(1/average upper bound)]/2. sigma_j is standard deviation of the bias = [log(1/upper bound) − log(lower bound)]/2.

         	Original hazard ratio	Estimated total bias	Bias adjusted hazard ratio
         Alzheimer’s and Dementia Care (ADC)	0.60 [0.59, 0.61]	−0.16 ± 0.28	0.76 [0.71, 0.82]

    2. Difference-in-difference estimate for total costs of care for Medicare (treatment—control): −$601 per patient per quarter (95% CI: −$1198, −$5).

       • The microsimulation model predicts monthly Medicare expenditures via a log link gamma distribution regression model. We modeled the effect of the Alzheimer’s and Dementia Care (ADC) on Medicare expenditures using sample importance replacement calibration (SIR) approach. The target outcome was the difference of monthly Medicare cost between treatment and control: −$601 per patient per quarter (95% CI: −$1198, −$5) or −$200.33 per patient per month (95% CI: −$399.33, −$1.67). We calibrated the intercept of regression predicting Medicare expenditures.

         Calibration parameter	Prior parameter value	Prior distribution	Target outcome from difference in Medicare expenditures
         Intercept	6.37	Unif (5.37, 7.37)	−$200.33 (95% CI: −$399.33, −$1.67)

    3. Posterior Distributions
       

11. Cost of Delivering the Intervention (mean, SD): $106 (20)

12. Structural Assumptions: Intervention has no effect on mortality.

D.4 |. Adult Day Services Plus (ADS Plus)

1. Outcome Paper(s) Used for Modeling:

   1. Gitlin et al.¹⁴

2. Intervention Name: Adult Day Services Plus

3. Intervention Description: Program provides care management to caregivers to address dementia-related behaviors, and physical, mental, and social health of family caregivers. Consists of face-to-face session with service director to identify concerns and needs, develop care plan, and implement a targeted plan that consists of counseling, education, referral, and contact with family service director. The service director remained in contact with families to address ongoing concerns.

4. Comparison Name: Usual adult day services.

5. Comparison Description: Control Adult Day Service site part of the same care network and was comparable in terms of programming, staffing, and client characteristics. Families in Adult Day Services received usual care.

6. Setting: Community/home and adult day service center.

7. Target Population: Community-dwelling people and their primary caregiver.

8. Intervention Intensity: Approximately 1 of contract with case manager per month.

9. Study Design/Unit of Randomization: quasi-experimental.

10. Outcome Measure Modeled:

    1. 1-year HR of being admitted to a long-term care facility: 0.43 (95% CI: 0.20, 0.94).

    • In sensitivity analysis we applied Turner et al.’s⁴² method to adjust for internal bias. Following Turner et al.’s proposed method to account for internal biases, we answered the following question for each domain: “Even if there were no intervention effect in this study, what apparent effect might be induced by this bias? Risk lower in intervention group [or higher in control group] or risk lower in control group [or higher in intervention group].
      Each reviewer is asked to mark a 67% range on the following elicitation scale for each type of bias. The aggregated results from each reviewer are used to calculate a total bias. The following correspondence between qualitative judgments of severity and the range limits are given: none (1), low (0.9–1), medium (0.7–0.9), and high (less than 0.7).”

      	Selection bias	Performance bias	Attrition bias	Detection bias	Total bias
      Reviewer 1	(0.9, 0.9)	(0.99, 0.99)	(0.9, 1.0)	(0.99, 0.99)
      Reviewer 2	(0.9, 1.0)	(0.9, 0.9)	(0.8, 1.0)	(0.95, 0.95)
      Reviewer 3	(0.7, 0.9)	(0.99, 0.99)	(0.7, 0.9)	(0.99, 0.99)
      Aggregated	(0.85L, 0.95R)	(0.99, 0.99)	(0.8, 1.0)	(0.99, 0.99)
      mu_j ± sigma_j	−0.06 ± 0.11	0 ± 0.01	−0.11 ± 0.11	0 ± 0.01	−0.17 ± 0.24

      Note: mu_j is the mean of the bias = [log(average lower bound) + log(1/average upper bound)]/2. sigma_j is standard deviation of the bias = [log(1/upper bound) − log(lower bound)]/2.

      	Original hazard ratio	Estimated total bias	Bias adjusted hazard ratio
      Adult Day Service Plus (ADS Plus)	0.44 [0.20, 0.94]	−0.17 ± 0.24	0.60 [0.26, 1.36]

11. Cost of delivering the intervention (mean, SD): $74 (20).

12. Structural Assumptions: Intervention has no effect on mortality. Intervention significantly reduced caregiver depression, and dementia-related behaviors, which were not modeled.

APPENDIX E: SIMULATED TIME TO TREATMENT

Modeling the time to start an intervention for a simulated person is important because the background risk of transitioning to a nursing home increases over time. In addition, other outcomes, such as time caregiving and Medicare expenditures, are not accumulated at a constant rate. Therefore, to determine the incremental benefit of an intervention, it is important to accurately model the time that simulated people receive treatment. We modeled the time a simulated person receives an intervention based on months since a dementia diagnosis. We used months since a dementia diagnosis as the eligibility criteria because it is a measure that can be derived for all the included studies. MIND, NYU Caregiver, and ADS Plus report the sample’s cognition at baseline, but do not report the time people had been living with AD/ADRD. MIND and ADS Plus report the sample’s cognition at baseline on the Mini-Mental State Examination (MMSE), which is the same measure of cognition in the AD/ADRD microsimulation. The NYU Caregiver intervention reports cognition using the Global Deterioration Scale (GDS). We converted GDS to MMSE scores (Solomon et al.⁵³). We simulated 10,000 people to determine the distribution of time in months till they reached each study’s baseline MMSE score. In the AD/ADRD microsimulation, the month a person is assumed to start treatment is sampled from these 10,000 values. Alzheimer’s and Dementia Care (ADC) reports that people included in the study had been living with dementia on average 3 years before participating in the intervention. Therefore, we sampled the entrance month for the ADC intervention from a truncated normal distribution with a mean of 3 years and a SD of 1 year, truncated at 0.

To determine how many people would be simulated for each set of model parameters tested, we performed a stability test on the model. For each of the values N = 100, 500, 1000, 2500, 5000, 10,000, we ran 1000 simulations of the N patients. We then looked at the SD of nursing home time (in months) over these 1000 simulations.

N	100	500	1000	2500	5000	10,000
Standard deviation	4.886	4.142	4.089	4.065	4.041	4.040

Because there was little improvement in reduction of variance when increasing the number of simulations per set of model parameters from 5000 to 10,000, we chose 5000 to minimize computation time while still performing enough simulations to achieve an estimate with low Monte Carlo variance.

APPENDIX F: COSTS, BENEFITS, AND INCREMENTAL COST-EFFECTIVENESS RATIOS FROM A SOCIETAL, FAMILY, AND HEALTH-CARE PAYER PERSPECTIVES

Table F.1, F.2.

TABLE F.1.

Net cost and benefits.

Comparison	Treatment cost	Net difference in lifetime costs, $a			Difference in people living with AD/ADRD’s quality-adjusted life-years	Difference in years (days) in the community
		Family	Health-care payer	Societal
Maximizing Independence at Home (MIND) versus usual care	$1445	−$12,454	$617	−$13,282	0.002	0.057 (21)
NYU Caregiver (NYU) versus usual care	$2571	−$3177	$12	−$5297	0.005	0.118 (43)
Alzheimer’s and Dementia Care (ADC) versus usual care	$2345	−$603	−$720	−$3668	0.003	0.061 (22)
Adult Day Service Plus (ADS Plus) versus usual care	$762	−$2244	$192	−$2813	0.002	0.038 (14)

^aNet cost is the difference in treatment [inclusive of treatment cost] – usual care cost.

TABLE F.2.

Incremental cost-effectiveness ratios.

Comparison	Incremental cost-effectiveness ratio
	Cost per quality-adjusted life-year gained			Cost per additional year in the community
	Family	Health-care payer	Societal	Family	Health-care payer	Societal
Maximizing Independence at Home (MIND) versus usual care	Cost saving	$268,476	Cost saving	Cost saving	$10,901	Cost saving
NYU Caregiver (NYU) versus usual care	Cost saving	$2224	Cost saving	Cost saving	$98	Cost saving
Alzheimer’s and Dementia Care (ADC) versus usual care	Cost saving	Cost saving	Cost saving	Cost saving	Cost saving	Cost saving
Adult Day Service Plus (ADS Plus) versus usual care	Cost saving	$126,148	Cost saving	Cost saving	$5015	Cost saving

APPENDIX G: RESULTS OF ONE-WAY SENSITIVITY ANALYSIS

Figure G.1 reports the results of the one-way sensitivity analysis from a health-care payer perspective. Figure G.2 reports results of the one-way sensitivity analysis from a family perspective.

FIGURE G.1.

Healthcare payer: one-way sensitivity analysis.

FIGURE G.2.

Family perspective: one-way sensitivity analysis.

APPENDIX H: RESULTS OF TWO-WAY SENSITIVITY ANALYSIS

MIND versus usual care (Difference in Cost, Difference in Quality-Adjusted Life-Years).

Societal perspective		Highly effective (HR = 0.42)	Base-case treatment effect (HR = 0.63)	Less effective (HR = 0.94)
Proportion of people who participate in treatment	Base-case 100%	−$15,500, 0.0037	−$13,282, 0.0023	−$10,027, 0.0003
	75%	−$11,444, 0.0028	−$9809, 0.0017	−$7385, 0.0002
	50%	−$7915, 0.0019	−$6773, 0.0012	−$5044, 0.0002

Health-care payer perspective		Highly effective (HR = 0.42)	Base-case treatment effect (HR = 0.63)	Less effective (HR = 0.94)
Proportion of people who participate in treatment	Base-case100%	$131, 0.0037 (ICER: $35,347)	$623, 0.0023 (ICER: $268,476)	$1348, 0.0003 (ICER: 4894,893)
	75%	$13, 0.0028 (ICER: $4543)	$430, 0.0017 (ICER: $250,220)	$1009, 0.0002 (ICER: $4,664,881)
	50%	$63, 0.0019 (ICER: $32,370)	$310, 0.0012 (ICER: $251,922)	$686, 0.0002 (ICER: $4,484,376)

Family perspective		Highly effective (HR = 0.42)	Base-case treatment effect (HR = 0.63)	Less effective (HR = 0.94)
Proportion of people who participate in treatment	Base-case 100%	−$14,185, 0.0037	−$12,454, 0.0023	−$9930, 0.0003
	75%	−$10,381, 0.0028	−$9163, 0.0017	−$7319, 0.0002
	50%	−$7239, 0.0019	−$6344, 0.0012	−$4991, 0.0002

NYU Caregiver versus usual care (Difference in Cost, Difference inQuality-Adjusted Life-Years).

Societal perspective		Highly effective (child caregiver HR = 0.28; spouse caregiver HR = 0.54)	Base-case treatment effect (child caregiver HR = 0.53; spouse caregiver HR = 0.72)	Less effective (child caregiver HR = 0.99; spouse caregiver = 0.96)
Proportion of people who participate in treatment	Base-case100%	−$10,028, 0.0093	−$5297, 0.0052	$1270, 0.0010 (ICER: $1,531,048)
	75%	−$8893, 0.0067	−$5152, 0.0038	$2237, 0.0002 (ICER: $11,185,795)
	50%	−$5905, 0.0044	−$3589, 0.0025	$2509, 0.0001 ($25,091,590)

Health-care payer perspective		Highly effective (child caregiver HR = 0.28; spouse caregiver HR = 0.54)	Base-case treatment effect (child caregiver HR = 0.53; spouse caregiver HR = 0.72)	Less effective (child caregiver HR = 0.99; spouse caregiver = 0.96)
Proportion of people who participate in treatment	Base-case100%	−$1251, 0.0093	−$52, 0.0052	$1636, 0.0010 (ICER: $1,972,703)
	75%	−$1001, 0.0067	$12, 0.0038 (ICER: $3158)	$3359, 0.0002 (ICER: $16,795,795)
	50%	−$554, 0.0044	$45, 0.0025	$3652, 0.0001 (ICER: $18,260,795)

Family perspective		Highly effective (child caregiver HR = 0.28; spouse caregiver HR = 0.54)	Base-case treatment effect (child caregiver HR = 0.53; spouse caregiver HR = 0.72)	Less effective (child caregiver HR = 0.99; spouse caregiver = 0.96)
Proportion of people who participate in treatment	Base-case100%	−$6645, 0.0093	−$3278, 0.0052	$1765, 0.0010 (ICER: $2,128,968)
	75%	−$6070, 0.0067	−$3177, 0.0038	$2959, 0.0002 (ICER: $14,795,795)
	50%	−$3841, 0.0044	−$2124, 0.0025	$3250, 0.0001 (ICER: $16,250,795)

Alzheimer’s and Dementia Care (ADC) versus usual care (Difference in Cost, Difference inQuality-Adjusted Life-Years).

Societal perspective		Highly effective (HR = 0.71)	Base-case treatment effect (HR = 0.76)	Less effective (HR = 0.82)
Proportion of people who participate in treatment	Base-case 100%	−$4575, 0.0032	−$3668, 0.0026	−$2773, 0.0019
	75%	−$3480, 0.0025	−$2799, 0.0020	−$2117, 0.0015
	50%	−$2308, 0.0016	−$1857, 0.0013	−$1409, 0.0010

Health-care payer perspective		Highly effective (HR = 0.71)	Base-case treatment effect (HR = 0.76)	Less effective (HR = 0.82)
Proportion of people who participate in treatment	Base-case100%	−$904, 0.0032	−$720, 0.0026	−$534, 0.0019
	75%	−$806, 0.0025	−$667, 0.0020	−$541, 0.0015
	50%	−$566, 0.0016	−$464, 0.0013	−$383, 0.0010

Family perspective		Highly effective (HR = 0.71)	Base-case treatment effect (HR = 0.76)	Less effective (HR = 0.82)
Proportion of people who participate in treatment	Base-case100%	−$1326, 0.0032	−$603, 0.0023	$106, 0.0019 (ICER: $54,902)
	75%	−$934, 0.0025	−$392, 0.0020	$165, 0.0015 (ICER: $108,778)
	50%	−$542, 0.0016	−$194, 0.0013	$174, 0.0010

Adult Day Service Plus (ADS Plus) versus usual care (Difference in Cost, Difference inQuality-Adjusted Life-Years).

Societal perspective		Highly effective (HR = 0.26)	Base-case treatment effect (HR = 0.60)	Less effective (HR = 1.36)
Proportion of people who participate in treatment	Base-case 100%	−$5573, 0.0032	−$2813, 0.0015	−$2616, −0.0017
	75%	−$4116, 0.0025	−$2119, 0.0013	−$2219, −0.0013
	50%	−$2861, 0.0017	−$1545, 0.0009	−$1392, −0.0009

Health-care payer perspective		Highly effective (HR = 0.26)	Base-case treatment effect (HR = 0.60)	Less effective (HR = 1.36)
Proportion of people who participate in treatment	Base-case 100%	−$464, 0.0032	$192, 0.0015 (ICER: $126,148)	$1417, −0.0017
	75%	−$469, 0.0025	$79, 0.0013 (ICER: $63,321)	$1091, −0.0013
	50%	−$262, 0.0017	$77, 0.0009 (ICER: $87,041)	$755, −0.0009

Family perspective		Highly effective (HR = 0.26)	Base-case treatment effect (HR = 0.60)	Less effective (HR = 1.36)
Proportion of people who participate in treatment	Base-case100%	−$4347, 0.0032	−$2244, 0.0015	$1961, −0.0017
	75%	−$3080, 0.0025	−$1632, 0.0013	$1695, −0.0013
	50%	−$2208, 0.0017	−$1232, 0.0009	$1027, −0.0009

Abbreviations: HR, hazard ratio; ICER, incremental cost-effectiveness ratio.

APPENDIX I: STRUCTURAL SENSITIVITY ANALYSIS—RESULTS OF EACH INTERVENTION COMPARED TO USUAL CARE

In the structural sensitivity analysis, we modeled treatment effects assuming it took 12 months to reach maximum effectiveness and the effect extended 12 months after the last reported follow up. Table I.1 reports the net costs, difference in quality-adjusted life years, and difference in years in the community for each treatment compared to usual care. Table I.2 reports the incremental cost-effectiveness ratios from the structural sensitivity analysis.

TABLE I.1.

Difference in lifetime costs ($) and quality adjusted life years (structural sensitivity analysis).

Comparison	Treatment cost	Net difference in lifetime costs, $a			Difference in people living with AD/ADRD’s quality-adjusted life-years	Difference in years (days) in the community
		Family/Individual	Health-care payer	Societal
Maximizing Independence at Home (MIND) versus usual care	$1445	−$10,062	$743	−$10,764	0.002	0.046 (16.79)
NYU Caregiver (NYU) versus usual care	$2571	−$2491	$446	−$4176	0.004	0.096 (35.04)
Alzheimer’s and Dementia Care (ADC) versus usual care	$2345	$1	−$148	−$2492	0.002	0.050 (18.25)
Adult Day Service Plus (ADS Plus) versus usual care	$762	−$1468	$266	−$1964	0.001	0.031 (11.32)

^aNetcost is the difference in treatment [inclusive of treatment cost] – control cost.

Abbreviation: AD/ADRD, Alzheimer’s disease/Alzheimer’s disease and related dementias.

TABLE I.2.

Incremental cost-effectiveness ratios (structural sensitivity analysis).

Comparison	Incremental cost-effectiveness ratia
	Cost per quality-adjusted life-year gained			Cost per additional year in the community
	Family	Health-care payer	Societal	Family	Health-care payer	Societal
Maximizing Independence at Home (MIND) versus usual care	Cost saving	$371,419	Cost saving	Cost saving	$16,149	Cost saving
NYU Caregiver (NYU) versus usual care	Cost saving	$111,546	Cost saving	Cost saving	$4648	Cost saving
Alzheimer’s and Dementia Care (ADC) versus usual care	$1380	Cost saving	Cost saving	$28	Cost saving	Cost saving
Adult Day Service Plus (ADS Plus) versus usual care	Cost saving	$265,619	Cost saving	Cost saving	$8568	Cost saving

^aThe perspective dictates who is responsible for the cost of the intervention.