 We conducted a systematic review of studies reporting the direct healthcare costs of treating older adults with diagnosed Alzheimer disease and related dementias (ADRD) within private Medicare managed care plans.

 A systematic review of all studies published in English reporting original empirical analyses of direct costs for older adults with ADRD in Medicare managed care.

 All papers indexed in PubMed or Web of Science reporting ADRD costs within Medicare managed care plans from 1983 through 2018 were identified and reviewed.

 Despite the growth in Medicare managed care enrollment, only 9 papers report the costs of care for individuals with ADRD within these plans, and only 1 study reports data less than 10 years old. This limited literature reports wide ranges for ADRD-attributable costs, with estimates varying from $3738 to $8726 in annual prevalent costs and $8938 to $38,794 in 1-year immediate postdiagnosis incident costs. Reviewed studies also used varied study populations, case and cost ascertainment methods, and analytic methods, making cross-study comparisons difficult.

 The expected continued growth in Medicare managed care enrollment, coupled with the large and growing impact of ADRD on America’s healthcare delivery and finance systems, requires more research on the cost of ADRD within managed care. This research should use more consistent approaches to identify ADRD prevalence and provide more detail regarding which components of care are included in analyses and how the costs of care are captured and measured.

Despite the growth in Medicare managed care enrollment, only 9 studies have reported the cost of Alzheimer disease and related dementias (ADRD) among private health plans now providing care to one-third of Medicare beneficiaries.

Estimates of annual ADRD incremental costs among managed care plans vary widely, with annual prevalent and incident costs ranging from $3738 to $8726 and $8938 to $39,794, respectively.

The extant literature has limited policy relevance to the cost of ADRD among managed care plans due to studies being out of date and reflecting divergent populations and research methods.

As of 2018, estimates suggest that 5.4 million individuals have Alzheimer disease and related dementias (ADRD) in the United States, of whom 96% are 65 years or older.

 If current prevalence rates and population trends continue, the number of older Americans with ADRD will more than double by midcentury, with projections of 11 million to more than 13 million

 individuals with diagnosed ADRD by 2050. Ensuring adequate resources for the health and social services required for a growing population with diagnosed and treated ADRD is a priority for the United States and other countries around the world

; this has led to a growing demand for analyses of the costs of the social, human, and health services required to provide adequate care for cognitively impaired older adults.

Analyses of the direct medical care costs among older adults with ADRD in the United States have relied primarily on diagnostic and healthcare use data from claims submitted by providers and facilities to CMS for services provided through the Medicare and Medicaid programs. This reliance on claims has resulted in most cost evidence for ADRD being based on the experience of the Medicare fee-for-service (FFS) program. Although Medicare beneficiaries have had the option of enrolling in private managed care insurance plans since 1983 when Congress created the Medicare Part C program—later transformed into the Medicare+Choice program, now Medicare Advantage (MA)&mdash;relatively little research has been conducted on the costs of ADRD among older adults within these plans.

The limited focus on ADRD costs within managed care plans has been largely driven by 2 factors. First, MA plans were not required to provide detailed diagnostic and health service use information to CMS until 2013,

 so MA-specific research required investigators to obtain data from individual health plans rather than from a single, national source. Second, enrollment in Medicare managed care plans had been consistently low—between 10% and 15% of eligible adults&mdash;for most of the program’s first 25 years.

 Thus, estimates that relied exclusively on data from Medicare FFS were presumed to produce reliable estimates of national care costs because of the relatively small numbers of older adults who chose to enroll in private managed care plans.

The limited attention that the cost of ADRD within Medicare managed care has received in the extant scientific literature was first highlighted by Rice and colleagues in 2001.

 In discussing the need for private Medicare plans to better understand and develop strategies to address treatment for the growing number of older adults with ADRD, Rice et al

 identified only 1 published paper that estimated the costs of ADRD within Medicare managed care. At that time, 15% of those eligible for Medicare (or 6.2 million individuals) were enrolled in private managed care plans.

 In the almost 20 years following the review published by Rice et al, enrollment in Medicare managed care has increased to 33% of all those eligible (or 19 million individuals). Further, Medicare managed care enrollment is forecast to rise to 41% of eligible individuals by 2027, continuing a steady shift away from FFS Medicare.

Considering the impact that ADRD is likely to have as a driver of health service use and cost among older adults, and the increasing role that managed care plays in health service delivery for this population, we conducted a systematic review of the evidence on the direct medical care costs of ADRD within Medicare managed care. Our goal is to document the evidence of the costs of ADRD within Medicare managed care, identify potential gaps in the evidence base, and propose a research agenda to address these gaps.

We conducted a systematic review of the scientific literature that reports the direct medical care costs of ADRD for the United States within Medicare’s managed care program based on studies reported in either the PubMed or Web of Science citation indices. We included all studies published in English that reported original empirical analyses of either total healthcare costs or key components of costs (eg, inpatient services, skilled nursing care). Although the Medicare managed care program began in 1983, our search did not restrict the years in which papers were published and included papers published from 1983 through 2018. Candidate papers for this review were identified using search terms used in previous literature reviews of ADRD costs, which included combinations of the following terms: “Medicare and managed care,” “Medicare + Choice,” “Medicare Part C,” or “Medicare Advantage” with “Alzheimer’s,” “Alzheimer’s disease,” “dementia,” or “ADRD,” and “economic,” “cost,” “expenditures,” or “spending.” Two members of the study team reviewed the title of each paper to determine relevance for our analysis and then the abstracts of remaining studies to ensure that each met inclusion criteria before identifying the papers for which full reviews were to be conducted. Finally, we reviewed the citation lists for the papers that met all study inclusion criteria, as well as 2 previously published systematic reviews on ADRD cost,

 to ensure that our search criteria did not overlook any previously cited studies. Disagreements regarding the set of papers included in the final review were addressed by the entire study team.

 We assessed each study based on the following criteria: study design, data source and population studied, number of ADRD cases included in the analysis, case ascertainment method, control identification method, source of health service use or cost data, components of reported health service use and cost, statistical methods, and ADRD-attributable and/or incremental cost.

Attributable costs are health service expenditures that are unique to ADRD and are distinct from expenditures that may be incurred by individuals with ADRD. We report costs using 2018 inflation-adjusted US dollars, generated by applying the medical care cost component of the Consumer Price Index to the nominal cost values reported in each paper.

 summarizes the process used to identify the papers included in this review. Our search returned 501 published papers, 432 of which were excluded following an examination of their titles. We reviewed abstracts for the remaining 69 papers, from which we identified 21 for a full-text review. Following this full-text review, we excluded 2 that examined Medicare FFS costs,

 2 that did not address the source of insurance within a specific clinical population,

 and 6 published abstracts for which no full study was available.

 following our full-text analysis because the authors updated the analysis with the data source and methods in a subsequent paper that we did include.

 Nine papers met all inclusion criteria and formed the basis for this review.

 provides a summary of the study design and approach for each of the 9 papers we reviewed, and each study’s cost estimates are reported in 

 Every paper used a retrospective design, with 2 reporting incident ADRD costs

 and the remainder reporting prevalent costs. Of the 2 incident cost analyses, Joyce et al

 report costs for only the first year following an incident ADRD diagnosis, whereas Suehs et al

 report costs for 1 year prior and 2 years following an incident diagnosis. Both papers required that no ADRD diagnosis be present in the data for 12 months prior to the presumed incident diagnosis. The studies by McCormick et al

 are the only studies that report costs over time in their assessment of ADRD-attributable costs in the 3 years preceding death.

Only 3 studies provide specific information about the care setting from which subjects were identified. Richards et al

greatest detail regarding where study subjects received healthcare and also had the smallest number of ADRD cases: 150 subjects recruited from 4 care settings. Joyce et al

 used the PharMetrics database, which includes information about each individual’s source of insurance but provides no further details about individuals with the Medicare managed care program, other than that their insurance source is identified as a health maintenance organization. McCormick et al

 identified subjects participating in the Washington Alzheimer’s Disease Patient Registry, all of whom were members of Group Health Cooperative in Washington State (now Kaiser Permanente Washington). The studies by Leon and Neumann

 are the only studies that relied on methods other than diagnostic or pharmacy data from claims to identify cases: McCormick et al

 based ADRD status on a clinical consensus review of the subject’s neurological examination and performance on a comprehensive neuropsychological battery; Leon and Neumann

 outcomes to assess cognitive impairment. Each of the remaining papers used a series of 

International Classification of Diseases, Ninth Revision, Clinical Modification

) codes and/or prescription drug fills to establish likely ADRD cases. Within the studies relying on diagnostic codes or prescriptions, each study used a different definition to identify ADRD cases (Table 1

). None of the papers that relied on diagnostic data to identify both cases and controls provided any information about the number of fields that were available from which to detect the presence of an ADRD diagnosis.

 did not use controls in their analysis and reported mean costs for older adults with diagnosed ADRD by relative severity based on MMSE scores. McCormick et al

 assessed relative ADRD costs with a random set of individuals with no dementia without any matching on observable characteristics. Richards et al

 included every individual who did not meet their ADRD inclusion criteria as controls without considering any demographic characteristics, eventually analyzing 250 cases compared with more than 13,000 controls. The remaining studies matched on age and sex, with 3 additionally matching on geographic region, to identify non-ADRD controls from the same data source from which cases were identified.

 Estimates of total cost and specific cost components (when reported) are shown in Table 2.

 The variety of data sources, time periods, analytic approaches, and range of cost components for which estimates are provided results in a wide range of ADRD-attributable costs among the 9 studies we reviewed. Five studies

 reported annual prevalent costs, with estimates of total costs ranging from $9010

 and of incremental costs ranging from $3738

producing the lowest and highest estimates both used total healthcare use as the basis for cost estimates. Frytak et al

 distinguished between managed care organization (MCO) and all-payer costs, but it is not clear from their presentation what distinguishes MCO costs from other sources. Hill et al

 focused on the costs of different forms of dementia and report the incremental costs of vascular dementia relative to controls; they found increased annual costs of $16,122 due almost entirely to hospital and skilled nursing facility services, where incremental costs are the marginal or additional expenditures specifically attributable to vascular dementia. Although each study reported statistically significant differences in cost between cases and controls, none of the papers we reviewed provided median costs, which might be a better reflection of the distribution of the underlying data.

An examination of specific cost components reveals that care in institutional settings (ie, hospitals, skilled nursing facilities, long-term care facilities) is the key driver of greater costs among older adults with ADRD. However, there is little consistency—with respect to both which components of healthcare use are reported and how these components are defined&mdash;to support cross-study comparisons. For example, several papers report costs for “inpatient” services, whereas others cite “hospital” costs; however, their Methods sections provide little guidance as to whether these estimates include all facility and professional services or outpatient services provided in hospital settings. Similarly, it is not clear if “physician office”

 services are measuring the same resources.

Two studies report incident costs. Suehs et al

 report $17,712 in total healthcare costs and $8938 in ADRD-attributable costs in the first year post index. Joyce et al

 estimate $48,895 and $38,794 for total and ADRD-attributable costs, respectively, in the incident year.

 which is the only one that examined end-of-life costs attributable to ADRD within Medicare managed care, is also the only study that found lower healthcare costs for individuals with ADRD relative to controls at both 1 and 3 years before death. The authors found that total healthcare costs in the last 3 years of life were $50,387 for individuals with Alzheimer disease and $78,766 for individuals with other dementias compared with $85,876 for controls, with long-term care costs being the main driver of the higher costs among controls.

We also note that, despite the call of Rice et al

 for more research on the cost of ADRD within Medicare managed care, the 2 papers reporting the most recent data are by Suehs et al,

 who published in 2013 using data from 2011, and Frytak et al,

 who published in 2008 using 2002 data. The remaining literature is based on data that precede the growth in MA enrollment.

The scientific literature examining ADRD-attributable costs among older adults in Medicare managed care programs reports a wide range of estimates of the economic consequences of ADRD to the MA program. Across these studies, the different care settings evaluated, methods for identifying cases, ways in which costs of care are captured and measured, and analytic approaches make it difficult to directly compare and draw conclusions about the impact of ADRD on costs within the MA program.

The extant literature is also too dated to be of much relevance to inform current policy regarding the costs of ADRD within Medicare managed care. Only 1 paper includes data that are less than 10 years old.

 Older studies do not reflect a series of important changes to Medicare, including the Part D prescription drug program introduced in 2005,

 a revised payment model for MA plans introduced between 2004 and 2006,

 and incentives created for older adults to enroll in private health plans created by the Affordable Care Act.

 Each of these changes to the Medicare program, as well as market forces such as the economic recession of 2008 to 2009, contributed to the enrollment shift from FFS to MA that is not reflected in the current literature.

 In addition to the need for more up-to-date research on the economics of ADRD within the MA program, future research should pursue standardized approaches to measuring the prevalence and incidence of ADRD, the selection of non-ADRD comparators, inclusion and measurements of costs included in analyses, and the statistical models used to estimate ADRD-specific costs. In support of this goal, we offer the following 6 suggestions for future research on the cost of ADRD:

 Studies should identify the care setting(s) from which individuals with ADRD are identified to allow the reader to understand the type of managed care setting that they examined. Managed care encompasses an increasingly diverse set of service delivery and financial arrangements that vary from closed model and integrated health systems to preferred provider organizations that may operate similarly to FFS. Although researchers may be constrained with respect to how much information a partner organization may allow them to share, the reader must know more about the managed care model that is being examined than what is reported by all but 2 of the papers included in this review.

 More research must be conducted on both the costs of incident ADRD and the way that costs change over time following a diagnosis and continuing until end of life. Several studies have examined these issues within the Medicare FFS program,

 but, as we note in our review, only 2 studies

 examined incident ADRD and only 1 evaluated end-of-life costs.

 Ongoing research should use consistent and replicable ways to establish both incident and prevalent ADRD. As we highlight in our review, each paper identified cases in different ways. With so much variation in case identification, comparing findings across studies is difficult.

 diagnostic codes or pharmacy fills to identify ADRD cases, this information is only available for individuals accessing health services for whom this information is recorded by a healthcare provider. This case identification method ignores the economic impact of ADRD among those whose cognitive impairment is not yet recognized by the healthcare delivery system and is subject to bias on the basis of where and how individuals access healthcare.

 To allow for comparisons between and among different estimates, Methods sections should present clear explanations about which costs are included in analyses, how cost measures are calculated, and from whose perspective costs are being analyzed (eg, payer, provider, society). Similarly, studies should also be clear about which costs are not being included in analyses. ADRD costs are driven by services that are not covered by Medicare or not covered in comprehensive ways—examples of this are long-term and skilled nursing care beyond the postacute care phase. Further, Medicare Part D has made access to prescription drug information easier for researchers to include in analyses, but greater detail is needed about how each component of care is identified, measured, and costed.

 Studies should use appropriate econometric methods to generate cost estimates. In addition to addressing the skewed distribution of healthcare costs, analyses must consider the ways in which length of enrollment within an MA plan and mortality affect both the likelihood that we will observe an ADRD diagnosis and estimates of ADRD-attributable costs. For example, previous research on the cost of cancer has explored the challenge of when a cancer episode begins and the degree to which life expectancy should factor into cost estimates.

 There are likely similarities in the cost trajectories of ADRD and some forms of cancer. Lessons from a more robust literature on the cost of cancer should be applied to analyses of ADRD.

We reviewed the extant literature on the cost of ADRD within Medicare managed care. The few studies we examined were based on diverse methods and populations, making it difficult to draw generalizable conclusions about the economic impact of ADRD within Medicare managed care. Further, with 1 exception, these studies were based on relatively older data and, thus, may not be relevant to the current market, which has seen considerable increases in managed care enrollment. More research is needed to provide public and private policy makers with the evidence that they need to adequately prepare for the expected impact of ADRD on the US health economy.

 University of Washington (PF, LW, PKC, BI), Seattle, WA; University of Pennsylvania (NBC), Philadelphia, PA; National Bureau of Economic Research (NBC), Princeton, NJ; Drexel University (SP), Philadelphia, PA; Kaiser Permanente Washington Health Research Institute (EBL), Seattle, WA.

 The CDC (Direct and Indirect Costs of Dementia Care, SIP 14-005) and the National Institutes on Aging (Current and Future Costs of Alzheimer’s and Dementia Care, AG049815 and the Adult Changes in Thought Study, AG 006781).

 Drs Fishman and Coe report that this work was performed under CDC SIP 14-005 and NIA R01AG049815 (PI: Coe); the funding agencies have no input into the work. Dr Larson reports employment with Kaiser Permanente Washington and receiving royalties as an UpToDate author. The remaining authors report no relationship or financial interest with any entity that would pose a conflict of interest with the subject matter of this article.

 Concept and design (PF, NBC, EBL); acquisition of data (PF, NBC, EBL); analysis and interpretation of data (PF, NBC, LW, PKC, SP, BI, EBL); drafting of the manuscript (PF, SP, BI); critical revision of the manuscript for important intellectual content (PF, NBC, LW, PKC, SP, BI, EBL); obtaining funding (PF, NBC, EBL); administrative, technical, or logistic support (NBC, LW, PKC, EBL); and supervision (PF, NBC, EBL).

 Paul Fishman, PhD, Department of Health Services, University of Washington, 1959 NE Pacific St, Seattle, WA 98185. Email: paulfish@uw.edu.

1. Wimo A, Guerchet M, Ali GC, et al. The worldwide costs of dementia 2015 and comparisons with 2010. 

. 2017;13(1):1-7. doi: 10.1016/j.jalz.2016.07.150.

2. Hebert LE, Scherr PA, Bienias JL, Bennett DA, Evans DA. Alzheimer disease in the US population: prevalence estimates using the 2000 census. 

. 2003;60(8):1119-1122. doi: 10.1001/archneur.60.8.1119.

3. Alzheimer’s Association. 2018 Alzheimer’s disease facts and figures [erratum in 

. 2018;14(5):701. doi: 10.1016/j.jalz.2018.04.001]. 

. 2018;14(3):367-429. doi: 10.1016/j.jalz.2018.02.001.

4. McCarthy M. Cost of dementia care in US to double by 2040. 

. 2013;346:f2175. doi: 10.1136/bmj.f2175.

5. Brennan N, Ornstein C, Frakt AB. Time to release Medicare Advantage claims data. 

. 2018;319(10):975-976. doi: 10.1001/jama.2017.21519.

6. Neuman P, Jacobson GA. Medicare Advantage checkup. 

. 2018;379(22):2163-2172. doi: 10.1056/NEJMhpr1804089.

7. Rice DP, Fillit HM, Max W, Knopman DS, Lloyd JR, Duttagupta S. Prevalence, costs, and treatment of Alzheimer’s disease and related dementia: a managed care perspective. 

8. Deb A, Thornton JD, Sambamoorthi U, Innes K. Direct and indirect cost of managing Alzheimer’s disease and related dementias in the United States. 

. 2017;17(2):189-202. doi: 10.1080/14737167.2017.1313118.

9. Mauskopf J, Mucha L. A review of the methods used to estimate the cost of Alzheimer’s disease in the United States. 

. 2011;26(4):298-309. doi: 10.1177/1533317511407481.

Articles

WEB EXCLUSIVE

A systematic review of the literature reporting the cost of dementia among Medicare managed care plans found a limited and dated body of evidence.

Cost of Dementia in Medicare Managed Care: A Systematic Literature Review

We all recognize the problem of high costs in healthcare, which put a serious burden on all Americans and US employers that provide benefits to their employees and families. The good news is that there is now a rare window of opportunity to meaningfully address this problem through bipartisan Congressional action. There is an urgent need for all of us to engage in these policy debates before the window closes during the upcoming election season.

Many of us believe that healthy competition and market forces can and should spur efficiency, appropriate care, high quality, and better patient experience. But there is growing recognition that our healthcare system has serious problems—including the effects of market consolidation and anticompetitive practices on prices. We know that market power has enabled providers, drug companies, pharmacy benefit managers, and others to raise prices. We also know that market concentration has been growing in recent years. Most hospital markets are already highly concentrated, and hospitals have also been buying up physician practices, limiting purchasers’ ability to fairly negotiate on behalf of patients. In other words, many parts of the healthcare market are fundamentally broken, and policy intervention is required.

Surprise billing is a painful symptom of a dysfunctional and unaffordable healthcare system. This nightmare is all too familiar: Unsuspecting and vulnerable patients are grateful for care but saddled with expensive and unexpected bills, even though they received care at a hospital designated as “in network.” In fact, unexpected medical bills are the number 1 concern among Americans, according to a Kaiser Family Foundation poll (Figure 7).

 More than 67% of Americans said that they were very or somewhat worried about being able to afford a medical bill they received due to no fault of their own. This is a clear example of a market failure. In the specific case of surprise bills, policy makers must take steps to protect consumers 

One of the promising solutions to this problem has been endorsed by the Senate Committee on Health, Education, Labor, and Pensions: a fair benchmark payment derived from existing market prices to avoid price gouging by a small number of providers. As an article in this issue of 

 shows, California’s experience implementing a policy to address surprise medical billing demonstrates that out-of-network payment standards can influence payer—provider bargaining leverage, impacting prices and network breadth. Furthermore, an analysis by Blue Shield of California has shown that the overall number of physicians under contract has increased since the implementation of the law, and the contracted rates for ancillary providers who frequently balance-billed also continued to increase after the law went into effect.

Patients may not care whether Congress uses arbitration or benchmarking in its legislation to eliminate surprise medical bills, but they do care about access to affordable, high-quality healthcare for themselves and their families. Implementation of these and other policies will take time and require significant changes by important stakeholders. Yet the vitality of our economy, the solvency of our nation’s treasury, and the welfare of all Americans depend upon our efforts.

1. Kirzinger A, Wu B, Muñana C, Brodie M. Kaiser Health Tracking Poll—late summer 2018: the election, pre-existing conditions, and surprises on medical bills. Kaiser Family Foundation website. kff.org/health-costs/poll-finding/kaiser-health-tracking-poll-late-summer-2018-the-election-pre-existing-conditions-and-surprises-on-medical-bills. Published September 5, 2018. Accessed August 4, 2019.

From The Editorial Board

From the Editorial Board: Elizabeth Mitchell

 To determine whether self-identified social needs, such as financial assistance with utilities, food programs, housing support, transportation, and medication assistance, collected using a passive social health surveillance system were associated with inpatient readmissions.

 This retrospective observational study linked social service referral data collected from a call center—based passive social health surveillance system with healthcare claims data extracted from a managed care organization (MCO). Mixed-effects logistic regression models calculated the odds of all-cause hospital readmissions within 30, 90, and 180 days among individuals with self-identified social service needs compared with those without.

Individuals who identified social service needs

had 68% (odds ratio [OR], 1.68; 95% CI, 1.51-1.86), 89% (OR, 1.89; 95% CI, 1.74-2.05), and 101% (OR, 2.01; 95% CI, 1.87-2.17) higher odds of readmission within 30, 90, and 180 days, respectively, after controlling for other study variables. Examining each social service need separately, individuals had higher odds of hospital readmission within 30 days of discharge if they identified a financial (OR, 1.19; 95% CI, 1.07-1.33), food (OR, 1.32; 95% CI, 1.17-1.48), housing (OR, 1.31; 95% CI, 1.09-1.57), or transportation (OR, 1.21; 95% CI, 1.08-1.36) need compared with those without those social needs. In all study outcomes, medication assistance was not associated with readmissions.

 An MCO created a passive social health surveillance program to more effectively integrate medical and social care. Understanding individual-level social health needs provides the insights needed to develop interventions to prevent hospital readmissions.

A passive social health surveillance system—a managed care organization (MCO)’s call center—based social service referral program&mdash;enabled examination of whether self-identified social needs are associated with inpatient readmissions.

This study linked referral and claims data for 19,817 individuals to examine the odds of inpatient readmissions for individuals with self-identified social needs, such as food and housing insecurity, compared with those without self-identified needs.

Results showed that individuals with self-identified social needs had higher odds of 30-day inpatient readmissions.

Passive social health surveillance systems could enable MCOs to play an important role in integrating medical and social care by complementing the social health screenings of medical care providers.

Social needs, such as adequate housing, food security, and access to transportation services, affect population health outcomes.

 Neglecting to integrate social care with medical care may burden the healthcare system with unnecessary or unplanned utilization.

 Excess inpatient readmissions represent one such problem that US policy makers seek to address.

 Significant Medicare payment penalties motivate hospitals to develop strategies to curtail these readmissions. One approach includes hospital-defined algorithms that identify individuals with the greatest risk of readmission at the time of discharge.

Social health surveillance systems, which can identify social needs and connect people to community resources, may support medical and social care integration.

 Active social health surveillance uses screening tools to directly identify patient social needs at medical facilities.

 However, medical care providers may lack the time, training, and resources to effectively conduct social health screenings or link patients to community service providers.

A passive social health surveillance system—analogous to public health surveillance systems&mdash;consists of the ongoing collection, storage, and classification of social health needs information identified and reported by individuals or their caregivers. Two nationwide passive social health surveillance systems, the 2-1-1 programs

 and WellCare Health Plan’s Community Assistance Line,

 have been identified as effective approaches for identifying and tracking individual-level social health needs, including financial assistance to pay for utilities, food programs, assistance with medication, housing support, and free or low-cost transportation, among others.

The objective of this study was to determine whether self-identified social service needs collected through a passive social health surveillance system were associated with inpatient readmissions within 30, 90, and 180 days.

WellCare Health Plan’s Community Assistance Line, a part of its Center for CommUnity Impact, provides referrals to community-based social service organizations to individuals who contact the call center and report a social need. This cross-sectional study linked self-identified social needs collected in the social service referral program’s database with the managed care organization (MCO)’s health service claims data.

The study examined claims and social service referral data from January 1, 2013, through October 19, 2017, from 27,189 individuals insured by WellCare Health Plan’s Medicaid managed care and Medicare Advantage programs. The sample included individuals who were at least 18 years of age with at least 1 admission to an acute care hospital. Individuals who were pregnant (n = 638), were younger than 18 years (n = 1469), were transferred to a different hospital (n = 757), had 1 admission but less than 6 months of enrollment and claims data (n = 4210), received chemotherapy (n = 88), lived outside the 13 established states of interest (n = 84), had claims data errors (n = 11), or had incomplete data on study variables (n = 115) were excluded. The final study population included 19,817 individuals from 13 states who were readmitted to 1064 hospitals.

Outcome measures consisted of 3 all-cause hospital readmissions: individuals who were admitted within 30, 90, and 180 days of prior inpatient discharge. An admission was considered a readmission only if it occurred within the stipulated time frame (eg, 30 days) and was not a transfer to a different hospital. To identify a readmission, we calculated the difference between the first and subsequent admission and categorized the individuals as having 30-, 90-, and 180-day readmissions. Individuals with a single admission and who had utilization data for at least 6 months were classified as not having a readmission.

The study population was further classified into 2 groups: (1) individuals with self-identified social service needs reported to the Community Assistance Line and (2) individuals without self-identified social service needs. The study further classified individuals with a self-identified need based on the presence of any of the following 5 social service needs: financial assistance for utilities, food programs, housing support, transportation, and medication assistance. Individuals contacting the Community Assistance Line with self-identified social needs either had exhausted their defined health insurance benefit for services such as food programs, transportation, or medication assistance, or the social service was not a part of the health plan benefit, such as financial assistance and health literacy programs. These 5 domains of social needs were selected based on the recommendation by CMS and their previous associations with health outcomes.

 The study also examined the individuals with each of these 5 aforementioned domains of social needs separately compared with individuals without them.

To account for the maldistribution of established risk factors for hospital readmissions between study groups, several variables were included in our analyses. These variables included state of residence (Arkansas, Florida, Georgia, Illinois, Kentucky, Louisiana, Mississippi, Missouri, New Jersey, New York, South Carolina, Tennessee, or Texas), age at index admission, sex, race (black, white, Hispanic, or other), type of insurance (Medicaid managed care or Medicare Advantage), length of stay of the index admission, and whether the individual received case management services before the date of readmission.

The study also quantified the illness severity of individuals by calculating the Charlson Comorbidity Index (CCI) score using 

International Classification of Diseases, Ninth Revision

) diagnosis codes from claims data within the 12 months before hospitalization.

 diagnosis codes, the study further identified individuals with alcohol abuse, severe mental illness, and anxiety and depression. To account for neighborhood characteristics, the study included 2 measures of the socioeconomic status of the individual’s community: percentage of people 25 years or older with less than a high school diploma and percentage of people 25 years or older living below the poverty level.

The characteristics of individuals who identified a social need and those who did not were described using frequencies, proportions, means, and medians. The differences in these characteristics were tested using χ

 tests of independence for proportions, 2 independent sample 

 tests for differences in means, and Wilcoxon tests for differences in medians. Tables were created to describe the distribution of the identified social needs among applicable individuals.

The study accounted for the hierarchical nature of the data by performing mixed-effects logistic regression with the hospital included as a random intercept to account for clustering of individuals within hospitals. This modeling approach accounts for the variation of hospital-related factors on readmission and controls the type I error rate.

 A test of multicollinearity was performed between variables with potentially common underlying determinants using the variance inflation factor (VIF). All selected covariates had a VIF of less than 10, indicating that the variables included in the model were not strongly correlated (

]). Odds ratios (ORs) and their 95% CIs were used to compare the odds of readmission (within 30, 90, and 180 days) between the 2 study groups before and after adjusting for risk factors. This sequential approach highlights the change in odds of readmission in the presence of known risk factors for readmission.

Three groups of analyses were conducted according to the study groups under observation. First, a model was created to compare the odds of readmission on all study outcomes between individuals with any self-identified social need and those without. Second, the study compared the odds of readmission among individuals with at least 1 of the 5 domains of social needs (financial assistance with utilities, food programs, housing support, transportation, and medication assistance) and individuals with other social needs with those without any social need. The final model compared the odds of readmission on all study outcomes for individuals with a social need except for financial assistance, food programs, housing support, transportation, and medication assistance versus those without any of these needs. All 3 models controlled for the differences in risk factors including age, gender, race, state of residence, type of insurance, length of stay, comorbidities, case management status, severe mental illness, anxiety/depression, alcohol abuse, and, in their neighborhood, percentage of people 25 years or older with less than a high school diploma and percentage of people 25 years or older living below the poverty level.

All statistical tests were 2-sided and conducted using SAS 9.4 (SAS Institute Inc; Cary, North Carolina). 

 values and ORs, along with their 95% CIs, were calculated. 

 values <.05 were considered statistically significant.

The study was classified as exempt by the University of South Florida Institutional Review Board (#Pro00028372), as the data used in the study were previously collected for existing program operations by WellCare Health Plans.

Our study population consisted of 19,817 individuals with a mean (SD) age of 56.6 (16.2) years; 64.4% were female, and 75.3% were white. About 69.2% of the study population identified at least 1 social need, resulting in 24,867 social service referrals. The 10 most frequent social service referrals, accounting for approximately 88.7% of referrals in this population, are shown in 

Compared with members who did not identify a social need, members with a self-identified social need were younger (55.8 vs 58.2 years) and had a higher median CCI score (1.0 vs 0.0). They were more likely to be female (66.4% vs 60.0%), be insured by Medicaid (50.0% vs 41.6%), be in case management (88.8% vs 86.0%), have a severe mental illness diagnosis (9.5% vs 4.9%), have anxiety or depression (22.3% vs 13.6%), abuse alcohol (1.3% vs 0.8%), and live in neighborhoods with a higher proportion of adults with less than a high school diploma (32.8% vs 32.2%). These differences were statistically significant (

Before adjusting for other study variables, individuals with any self-identified social service need had 70%, 90%, and 102% higher odds of readmission within 30, 90, and 180 days, respectively (30 days: OR, 1.70; 95% CI, 1.53-1.88; 90 days: OR, 1.90; 95% CI, 1.75-2.05; 180 days: OR, 2.02; 95% CI, 1.88-2.17), compared with individuals without a self-identified social need. After controlling for other study variables, the odds of readmission attenuated slightly but remained higher in all study periods for individuals with a social need compared with those without one (30 days: OR, 1.68; 95% CI, 1.51-1.86; 90 days: OR, 1.89; 95% CI, 1.74-2.05; 180 days: OR, 2.01; 95% CI, 1.87-2.17) (

 shows the results of the 30-day readmission adjusted mixed-effects logistic regression model, including the independent variables, where the odds of readmission for those with a self-identified need were 68% higher (95% CI, 1.51-1.86). Length of stay of index admission showed increased odds of readmission (OR, 1.04; 95% CI, 1.03-1.05) for every day spent in the hospital. Female sex (OR, 0.90; 95% CI, 0.82-0.98) and case management before readmission (OR, 0.57; 95% CI, 0.48-0.67) were each associated with reduced odds of 30-day readmission in the adjusted logistic regression model.

The odds of readmission remained higher in all study periods for individuals who identified at least 1 of the 5 social service needs (financial assistance for utilities, food programs, housing support, transportation, or medication assistance). The odds of readmission observed in this group were comparable with the effect observed in the main group but slightly higher in most study outcomes than in the group with other social needs (

The results in Table 5 show that individuals needing financial assistance had 19% higher odds of being readmitted within 30 days of discharge (95% CI, 1.07-1.33) after controlling for other study variables. The increased statistical significance was maintained within 90 (OR, 1.18; 95% CI, 1.08-1.29) and 180 (OR, 1.25; 95% CI, 1.16-1.35) days. Similarly, individuals with food needs (30 days: OR, 1.32; 95% CI, 1.17-1.48; 90 days: OR, 1.31; 95% CI, 1.19-1.44; 180 days: OR, 1.31; 95% CI, 1.20-1.42), housing needs (30 days: OR, 1.31; 95% CI, 1.09-1.57; 90 days: OR, 1.28; 95% CI, 1.10-1.48; 180 days: OR, 1.44; 95% CI, 1.26-1.65), and transportation service needs (30 days: OR, 1.21; 95% CI, 1.08-1.36; 90 days: OR, 1.34; 95% CI, 1.23-1.47; 180 days: OR, 1.43; 95% CI, 1.31-1.55) had higher odds of readmission within 30, 90, and 180 days compared with others who did not identify these needs. In all study outcomes, individuals needing medication assistance did not have higher odds of inpatient readmissions compared with individuals who did not identify assistance with medication as a need (30 days: OR, 1.05; 95% CI, 0.89-1.24; 90 days: OR, 1.11; 95% CI, 0.97-1.26; 180 days: OR, 1.08; 95% CI, 0.96-1.22).

This study examined the association between self-identified social needs collected from a passive social service surveillance system and inpatient readmissions extracted from healthcare claims covering almost 5 years. Combining social service referrals with healthcare utilization data supports efforts to integrate medical and social care. Results show that individuals with self-identified social needs have higher odds of 30-, 90-, and 180-day inpatient readmission compared with individuals without a self-identified social need. This underscores the potential of passive social health surveillance systems in identifying individuals at risk of inpatient readmissions.

Although the mechanism by which social needs affect readmissions is not confirmed, it is possible that the self-identified social needs exacerbate the health conditions of the original diagnosis or hinder complete recovery and/or adherence to the postdischarge plan. For example, individuals without transportation services might miss doctor’s appointments and those in need of food might lack the nutritional requirements for adequate recovery. These social needs, if unmet, could potentially lead to a return to the hospital. Although examining the mediating effect of social needs on subsequent inpatient readmission and access to medical care among people of low socioeconomic status is beyond the scope of this paper, we believe that this might be an opportunity for further research.

In addition, although unmet social needs may be associated with readmissions, it is plausible that the result could be influenced by unmeasured confounders related to both self-reported social service needs and readmissions. For example, the apparent association between social needs and readmission may be that patients lack social support, which, in turn, leads to both contacting the call center to report a social need and returning to the hospital for readmission. Thus, our results may be overestimated. If factors related to social support that were unavailable in the present data were added to the models, the observed association reflected in our results may have been attenuated. However, even if self-reported social needs simply act as a spurious marker for readmissions, from a practical standpoint, this is problematic only if interventions focus solely on addressing the social needs, such as housing instability and food insecurity, while ignoring possible underlying determinants, such as shortage of affordable housing and inconsistent access to food. Additional research should be conducted to elucidate the actual nature of the relationship between social needs and readmission rates in order to properly identify the best interventions for preventing inpatient readmissions.

Individual-level social health surveillance information has not been previously evaluated when estimating inpatient readmission outcomes. Most readmission predictive models have incorporated clinical and demographic variables obtained from administrative databases.

 Other research predicting hospital readmissions used community-level factors as proxies for unmeasured individual-level measures of social health.

 The results from this study support the addition of individual-level social health factors collected through social health surveillance—both active and passive&mdash;to models seeking to predict individuals at risk of readmission.

The study also analyzed individuals with each of the following social service needs: financial assistance for utilities, food programs, housing support, transportation, and medication assistance. Compared with individuals without each of those social service needs, the odds of readmission for each of these social service types except medication assistance were higher at 30, 90, and 180 days. These findings are consistent with those of previous research showing that a variety of social determinants of health, such as financial insecurity,

Furthermore, this study highlights the contribution of nonclinical organizations to social health surveillance efforts. MCOs

 and social care coordination programs conduct passive social health surveillance that requires adequate investment to operate effectively.

 For example, the social service referral program should be appropriately promoted to ensure awareness by all stakeholders, including individuals with social health needs and medical care providers seeking resources to address patients’ social determinants of health. In addition, sufficient staffing of social health surveillance programs facilitates the connection of patients to community resources, supports technologies that track social service referrals, and enables robust program evaluation.

 The social health surveillance function highlighted in this study—the Community Assistance Line call center—based social service referral program&mdash;could be a useful tactic for accountable care organizations (ACOs). ACOs represent a risk-based financing model similar to the MCO evaluated in this study. ACOs are a type of alternative payment model that holds healthcare organizations financially responsible for the cost and quality of care of a population.

 Some ACOs also seek to combine their medical care capacity with social service coordination to achieve improved health outcomes, a concept called Accountable Health Communities.

 However, most ACOs lack the social health surveillance functions and community organization relationships necessary to address their patients’ social determinants of health.

 The Community Assistance Line represents 1 potential medical and social care integration model for organizations financially responsible for improving quality and reducing costs of care.

Future studies should examine the benefits of using passive social health surveillance data, such as those presented in this study, to assist in preventing inpatient readmissions. Programs combining social health surveillance with social care interventions, including hospital-based community health workers,

 have demonstrated success at reducing preventable readmissions. In addition, future investigation should build on this retrospective observational study to examine the optimal nature, quantity, and timing of social health surveillance and social care interventions that could prevent readmissions among populations with social care needs.

Several limitations of this study are worth noting. First, the study design precludes any causal and temporal inferences about the relationships between the self-identified social care needs and readmissions. Although observed group differences were accounted for in the analysis, the impact of unmeasured factors unavailable in our claims data, such as discharge location, marital status, and other social support, is unknown.

Also, this study analyzed individuals enrolled in the Medicare Advantage and Medicaid health plans of a single MCO. Although the large sample size drawn from 1064 hospitals in 13 states supports the generalizability of the results, extrapolation of the findings to other social surveillance systems or the broader US population will require further research.

Furthermore, it should be noted that the classification of readmissions was such that an individual readmitted within 30 days of discharge was also classified in the 90- and 180-day outcomes. Therefore, it is expected that the ORs of all study outcomes would follow an increasing linear response. Thus, we caution the interpretation of the dose—response relationship observed in our results.

In addition, the self-identified need does not mean that the referral resulted in access to or satisfaction with social services. Although these metrics were tracked, the focus on this study was to examine the hospitalization pattern of subjects with a social need. Future prospective randomized studies should determine if the fulfillment of social needs leads to a reduction in readmissions or unnecessary healthcare utilization.

MANAGERIAL

By collecting self-identified social needs and linking them to claims data, this study analysis reveals that social needs are associated with inpatient readmissions.

Passive Social Health Surveillance and Inpatient Readmissions

Nurse practitioners (NPs) and physician assistants (PAs) represent a growing aspect of primary care in the United States.

Education, training, licensure, certification, and scope of practice vary between NPs and PAs, with significant variation across US states.

Quality of care is similar among NPs, PAs, and physicians for routine patient presentations, but evidence is less robust for complex patients, and there is little evidence on cost-effectiveness.

Reforming the US healthcare system must include a reprioritization of primary care, significant revisions to medical education, and development of a conceptual framework to promote and critically evaluate interdisciplinary, team-based care.

Nurse practitioners (NPs) and physician assistants (PAs) emerged

as occupations in the 1960s to mitigate shortages in the US primary care workforce.

 Recently, a second wave of primary care physician shortages

resulted from resident duty hour restrictions.

 Presently, 248,000 NPs and 115,500 PAs practice across all sites of care in the United States.

 The term “advanced practice providers” (APPs) refers to NPs and PAs, as well as other licensed, nonphysician providers, including certified nurse midwives, clinical nurse specialists, and certified registered nurse anesthetists.

 This commentary focuses on NPs and PAs, compares their education and training with that of primary care physicians, and reviews their scope of practice. We review literature on quality of care and offer a perspective on restructuring primary care in America.

The authors compare advanced practice providers’ education, training, scope of practice, and quality of care with that of physicians. A framework is essential to promote team-based primary care.

August 2019

Cost of Dementia in Medicare Managed Care: A Systematic Literature Review

From the Editorial Board: Elizabeth Mitchell

Passive Social Health Surveillance and Inpatient Readmissions

Current Evidence and Controversies: Advanced Practice Providers in Healthcare

The Adoption and Spread of Hospital Care Coordination Activities Under Value-Based Programs

The Potential Impact of CAR T-Cell Treatment Delays on Society

Late Diagnosis of Hepatitis C Virus Infection, 2014-2016: Continuing Missed Intervention Opportunities

Access to Chiropractic Care and the Cost of Spine Conditions Among Older Adults

Pediatric Codeine Prescriptions in Outpatient and Inpatient Settings in Korea

Tools to Improve Referrals From Primary Care to Specialty Care

Influence of Out-of-Network Payment Standards on Insurer–Provider Bargaining: California’s Experience