Professional agency vs consumer directed care workers: Outcomes in managed care

Abstract

Direct care workers are a major part of the long-term services and supports (LTSS) needed to address the health of individuals and accounted for $112 billion in United States spending in 2015. Direct care workers are hired within professional agency models (PAMs) or consumer-directed models (CDMs) where workers (including family) are contracted by the individual to obtain services. We sought to identify differences in cost and utilisation outcomes between PAM and CDM participants. Data were obtained from the University of Pittsburgh Medical Center (UPMC) Insurance Services Division from the participants enrolled in UPMC Community HealthChoices in Pennsylvania during 2018. A retrospective, observational cohort study design was performed using claims data. Utilisation outcomes were assessed using multivariate logistic regression and cost outcomes by linear regression. The 3,232 participants met the inclusion criteria. Of these, 69% (N = 2,217) were in a PAM, 23% (N = 752) were in a CDM, and 8% (N = 263) used a combination of services. PAM groups were older (mean 62.4 years vs 54.1 years), more likely to be women (69.0% vs 62.8%), and had more healthcare needs. Hospital utilisation was the same among groups. However, total cost was lower in CDM groups due to differences in LTSS costs between CDM and PAM services. Among dually eligible Medicare and Medicaid beneficiaries receiving LTSS, there are significant differences in age, gender, race and health needs. While hospital utilisation was not different between groups, CDM groups had lower total costs of care compared to PAM. These findings have implications for families, policymakers and insurers in helping to govern community LTSS while supporting member autonomy.

1 |. INTRODUCTION

American adults are increasingly receiving long-term care services within the home compared to institutions (Kitchener et al., 2005). In the United States, over 50% of these services are paid for by Medicaid and accounted for $115.8 billion in federal and state Medicaid spending in 2016 with a projected increase to $158.7 billion in 2026 (Actuarial Report on Medicaid, 2017; Thatch & Wierner, 2018). Much of this spending funds long-term services and supports (LTSS) that address the health, health-related and social needs of individuals to facilitate optimal functioning in the home and community. As America’s population ages and the number of disabled people increases, LTSS will nearly double by 2040 (Thomas & Applebaum, 2015). Moreover, 52% of Americans over age 65 should expect to pay for some of these services, whether anticipated or not (Favreault & Dey, 2015).

Among all LTSS, payments to direct-care workers (DCWs) make up the largest portion of spending among those participants receiving home and community-based services (HCBS). DCWs, also known as personal care attendants or ‘caregivers’, provide crucial services that meet the non-medical needs of program participants (LeBlanc et al., 2001). These services include assisting with activities of daily living (ADLs), such as bathing, dressing, toileting, ambulating and feeding, as well as instrumental activities of daily living (IADLs) such as meal preparation, grocery shopping and finance management (Vulnerabilites in CMS Personal Care Services, 2018). Since 1975, states have had the option of providing DCW services to participants as part of LTSS. Historically, DCWs were trained individuals hired through local or regional professional licensed home care agencies who provided services concordant with a participant’s needs. However, beginning in the late 1990s, as a series of Centers for Medicare & Medicaid Services pilot studies, the landscape of DCW services changed with the introduction of consumer-directed models (CDMs). These models allowed the participant to appoint their own DCW who would receive payment for HCBS. Unlike the PAM, where DCWs were specially trained and licensed, the CDM allowed for untrained DCWs including friends, neighbours and relatives. (Benjamin et al., 2000).

Despite the implications for participants, families and policymakers, research gaps persist, particularly regarding outcomes between PAM and CDM. Prior literature reviews have examined demographic characteristics as well as participant and care attendant satisfaction, with findings generally supporting the choice of CDM or PAM services (Wysocki et al., 2015). However, many of these studies occurred within fragmented state or national waiver programs and were limited by population heterogeneity, poorly defined outcomes or weak study designs. Importantly, cost and acute care outcomes, important to legislatures and payers, were rarely assessed and deemed to be insufficient to draw any conclusions.

With this in mind, we sought to assess the relationship between CDMs and PAMs. We assessed medical utilisation including hospitalisations, emergency department visits, readmissions and cost of care. Informed by prior research, local experience and participant attitudes with CDMs, we hypothesised that the participants receiving services through CDMs will have both decreased utilisation and cost compared to those using PAMs.

2 |. METHODS

We conducted a retrospective cohort study. Study participants included adults age 18 years and older who were dually eligible for Medicaid and Medicare and enrolled in UPMC Community HealthChoices (CHC) in Southwestern Pennsylvania (PA), USA, from January 2018 through December 2018. This region encompasses a geographically and socioeconomically diverse population. CHC is UPMC’s Medicaid managed care program that includes physical health and long-term services and supports. UPMC is currently one of three contractors providing managed LTSS, which was implemented in 2018 in Southwestern Pennsylvania with continued regional expansion into Southeastern Pennsylvania in 2019 and the remainder of the state in 2020. Our participants lived within 14 counties in Southwestern Pennsylvania and received HCBS during every month of the study period. The participants lived within the home during the entire study period. The participants were excluded if they lived in a nursing facility or no longer received HCBS.

For our analysis, our predictor of interest was DCW status. We defined three mutually exclusive groups using billable CPT codes collected for the 12-month study period. Group 1 received HCBS exclusively from a PAM. Group 2 received services exclusively in a CDM. A small proportion of participants received HCBS from both professional and self-directed DCWs during the study period, named afterwards as the combination group (PAM/CDM). The reference category was the combination group (PAM/CDM).

We aimed to measure two outcomes of interest: (1) healthcare utilisation (hospitalisations, emergency department visits, and 7- and 30-day readmissions) and (2) cost (total cost of care, LTSS cost, physical health/medical cost and drug cost). Due to the sensitive nature of cost outcomes in a competitive United States insurance market, the authors and institutional business advisors recommended comparing cost outcomes using standardised references as opposed to presenting actual cost numbers. Utilisation outcomes were assessed as binary outcomes (e.g., patient was either hospitalised or not hospitalised) given a low overall occurrence of events. All study outcomes occurred in 2018, and claims were collected through July 2019 to allow for delay in claims reporting. There was no correction for missing data secondary to contested or addended claims, though we suggest these events occur similarly among groups.

We adjusted regression models for baseline demographic data including age, race, ethnicity, living arrangement (single, married and divorced), help needed in ADLs, IADLs and the Cave Care Needs Index (CNI) grouper, which is a predictive risk score that incorporates current care gaps, chronic medical conditions and clinical lab results to predict health risk (https://cavegroup.com/).

The baseline participants characteristics were compared using one-way analysis of variance for continuous, normally distributed variables; Kruskal–Wallis test for non-normally distributed variables; and Chi-square tests for proportions. Variables were assessed for normality using histograms. Logistic regression was used for assessing binary utilisation outcomes. Linear regression was used for continuous cost outcomes, with standardised β coefficients and 95% confidence intervals reported to compare results across all models. We set a level of confidence of α < 0.05 for statistical conclusions. The study received ethics approval by the University of Pittsburgh Institutional Review Board with low participant risk (STUDY 19010276). Statistical analysis was performed using SAS, STAT version 15.1 (11/2018).

3 |. RESULTS

The 9,013 participants were identified for consideration in the sample; 3,232 met the inclusion criteria. Of the 3,232 participants with DCWs, 69% (N = 2,217) were in the PAM, 23% (N = 752) were in the CDM and 8% (N = 263) used a combination of services. Statistical tests comparing baseline characteristics revealed that age, race and sex were significantly different between agency-only and self-directed only groups, with agency-only groups being older (mean 62.4 years vs 54.1 years) and predominantly female (69.0% vs 62.8%) (Table 1). Agency-only participants were less frequently African American compared to self-directed participants (24.6% vs 26.3%). Agency-only participants had higher health risk as assessed by the Cave CNI.

TABLE 1.

Participant characteristics by direct care worker status

Participant characteristics, mean ± SD, median (IQR), or N (%)	Total N = 3,232	Agency only N = 2,217	Self-directed only N = 752	Both N = 263
Demographics
Agea,*	59.91 ± 16.17	62.38 ± 15.68	54.09 ± 15.98	55.68 ± 15.65
	60.00 (19.00)	62.0 (20.0)	55.0 (20.0)	60.00 (17.00)
Sex, femalea,*	2,186 (67.64)	1,530 (69.01)	472 (62.77)	184 (69.96)
Racea,*
Caucasian	2,216 (68.56)	1,486 (67.03)**	532 (70.74)	198 (75.29)
African American	802 (24.81)	545 (24.58)**	198 (26.33)	59 (22.43)
Asian/Pacific Islander	158 (4.89)	145 (6.54)**	8 (1.06)	5 (1.09)
American Indian/Eskimo	3 (0.09)	3 (0.14)**	0 (0.00)	0 (0.00)
Unknown	53 (1.64)	38 (1.71)**	14 (1.86)	1 (0.38)
Ethnicity
Non-Hispanic	3,172 (99.31)	2,179 (99.45)	735 (98.92)	258 (99.23)
Hispanic	22 (0.69)	12 (0.55)	8 (1.08)	2 (0.77)
Living arrangement, lives with otherŝ	1,511 (46.75)	933 (42.08)**	441 (58.64)**	137 (52.09)
ADL help, yes	1,088 (33.66)	740 (33.38)	252 (33.51)	96 (36.50)
IADL help, yes	1,335 (41.31)	920 (41.50)	312 (41.49)	103 (39.16)
Health Status Indicators
CAVE Care Need Indexa,*	2.23 ± 1.66	2.28 ± 1.66	2.08 ± 1.69	2.24 ± 1.63
	2.25 (2.75)	2.25 (2.75)	2.00 (3.00)	2.25 (2.75)
Utilisation
Utilized PAS hoursa,*	2,372.01 ± 1,613.57	2,301.59 ± 1,689.61	2,508.35 ± 1,400.48	2,575.82 ± 1,484.31
	2,079.75 (1,711.13)	1,984.50 (1,866.50)**	2,181.75 (1,372.13)	2,270.00 (1,783.75)

Abbreviations: ADL, activities of daily living; IADL, instrumental activities of daily living; IQR, interquartile range.

^aFor agency only versus self-directed only.

^*p < 0.05 for comparison across all three groups.

^**p < 0.05 for agency only and self-directed only versus both.

There were no statistically significant differences in hospitalisations, emergency department use or 7- and 30-day readmissions between the PAM or CDM groups (Table 2). For total cost of care, though, the CDM was significantly less expensive than the reference group (β = −0.05). The PAM was not significantly different than the reference. For LTSS cost, CDM was significantly less expensive than the reference (β = −0.09), and PAM was significantly more expensive than the reference (β = 0.07). For physical health cost, CDM was not significantly different than the reference. The PAM was significantly less expensive than the reference (β = −0.05). There was no statistically significant difference in pharmacy cost among CDM, PAM and combination groups.

TABLE 2.

Utilisation and cost outcomes, by direct-care worker (DCW) status

	Utilisation outcomes (yes/no) odds ratio (95% confidence interval)
	Hospitalisations	Emergency room	7-day readmission	30-day readmission
Agency	0.92 (0.78, 1.07)	1.03 (0.84, 1.26)	0.72 (0.49, 1.04)	0.99(0.74, 1.32)
Self-directed	1.00 (0.83, 1.21)	1.08 (0.86, 1.36)	1.08 (0.70, 1.65)	0.88 (0.62, 1.25)
Both	referent	referent	referent	referent
	Cost outcomes (dollars) standardized βs
	Total	LTSS	Physical health	Pharmacy
Agency	0.01	0.07 *	−0.05 *	−0.008
Self-directed	−0.05 *	−0.09 *	0.01	−0.004
Both	referent	referent	referent	referent

Notes.: All models adjusted for age, race, ethnicity, living arrangement, ADL help, IADL help and CAVE Care Need Index.

Abbreviations: ADL, activities of daily living; IADL, instrumental activities of daily living.

^*p < 0.05 for comparison across all three groups.

4 |. DISCUSSION

This study suggests that in participants receiving DCW services at home, there is no significant difference in utilisation between consumer-directed and professional agency groups. However, total cost of care in the CDM was lower than the reference group mainly due to lower LTSS costs. This partially refutes our hypothesis that CDM would have lower utilisation and lower cost. While these outcomes are not the sole factors in evaluating program success, they are surrogate indicators of quality of care (utilisation) and waste, fraud and abuse (cost).

We found notable differences in participants using each type of DCWs. The participants enrolled in PAMs were older and more likely to be female. This likely reflects the gender disparity in caregiver services in the United States, where 57%–81% of care is delivered by women (Heller et al., 2012; Sharma et al., 2016). Older female participants may not have access to willing partners or family members for engagement in CDM services. We also observed small racial differences between groups, with more African American participants receiving self-d irected care. While racial heath disparities are evident and continue to be explored, we suggest additional research is needed to unearth participant’s motivation and health outcomes related to this work. Agency-only participants also had higher baseline health risk measured by the Cave CNI, which suggests that more medically complex participants may enroll in PAM services, possibly out of the perception of more skilled training. We do recognise that the true differences in training and hiring practices, shown in other studies, vary among states and should be evaluated locally (Benjamin et al., 2000).

Both UPMC’s CHC and Pennsylvania’s managed LTSS were created to provide a person-driven, long-term support system in which participants have choice, control and access to services that provide independence, health and quality of life with similar costs. Regarding utilisation and cost outcomes, our findings complement the existing LTSS literature. Prior work, including a systematic narrative review of 17 studies, assessed participant and caregiver outcomes related to CDM in both the United States and international settings (Ottomann et al., 2013). That review suggested that CDMs were generally associated with improved satisfaction outcomes and could be used to empower participants and caregivers alike. Another analysis of three pilot CDM programs in Arkansas, New Jersey and Florida estimated health outcomes related to a ‘cash and counseling’ program where participants paid for their own services (Carlson et al., 2005). That study showed that participants in CDM programs had no adverse health outcomes compared to the control group and similarly had no increased medical needs for falls, injury due to caregiver or overnight hospitalisations. Importantly, the study estimated a significantly lower incidence of worsening shortness of breath and bedsores, likely as a result of increased mobility and DCW attention. Those findings help address the common misconception that an untrained caregiver is not able to adeptly provide services afforded by a trained caregiver. Yet other studies showed the effectiveness of DCW services among 491 community dwelling adults enrolled in a managed LTSS program (Neumann & Documet, 2019). Using a mix of enrollment and claims data, CDM participants were 93% less likely than participants with no DCWs to have a skilled nursing facility (SNF) admission. In addition, CDM participants were also 23.5% less likely to have SNF admission compared to PAM participants. Allowing participants to choose their DCWs, whether CDM or PAM, ensures that participants have choice, control and access to services while potentially leading to lower costs and improved health. Our study importantly adds cost and utilisation outcomes related to home care programs in the United States. It also addresses the newly implemented Pennsylvania CHC program, which since 2018 has been administered by managed care organisations, as opposed to state Medicaid fee for service programs, to deliver care to vulnerable populations.

There are limitations to our study. First, CDM, as previously mentioned, comprises multiple different models that may vary between state and managed care organisation (Benjamin, 2001). There is variability in participant eligibility, provider supervision, financial proxy and range of benefits. There are also variabilities in the number of eligible DCWs within the state, which we are unable to quantify within the realms of our dataset. These limitations may impact generalisability in other states and LTSS populations. Second, some professional agencies may in fact hire family members as DCWs, either as employees or independent contractors, to provide services to the participant. This obfuscates strict interpretation of the results of the PAM, as we were unable to adjust for this effect within the scope of this study. Third, the outcomes measured were concurrent with the study period of this analysis. Therefore, we were unable to predict if future utilisation or costs are impacted by caregiver type. Additional prospective studies may allow for predictive assessment as enrollment within Pennsylvania’s managed LTSS increases. Fourth, we only address payments through United States Medicaid. These payments may not reflect out of pocket expenses nor charges to a participant and therefore may not reflect the actual cost of care to a participant. Future research may reveal surprising findings when assessing payment through private insurance, individual savings plans, Veterans Affairs coffers or direct contracting through online services. Finally, this study represents only a single region within a state and occurred during the first year of implementation, when operational challenges may have unknowingly skewed enrollment into CDM or PAM groups. However, given the size and diversity of Pennsylvania, many states may benefit from the outcomes gleaned from this transition.

Home care services in general remain understudied, perhaps due to imprecision related to the most appropriate medical, political or psychosocial research domain. While our study included a wide age range, the majority of services in the United States are provided to the elderly people and are greatly influenced by contextual personal and societal beliefs and practices, including filial piety and normative gender roles, which may influence a family or participant’s decision in selecting a DCW (McCarty et al., 2008; Sharma et al., 2016). Moreover, there is a growing interest in waste and fraud related to personal care services. A report from the Office of Inspector General identified hundreds of cases of fraud and abuse related to personal care services (Wachino, 2016) The 21st Century Cures Act of 2016 includes a mandate for electronic visit verification for all Medicaid personal care services by 2023 and may include geotagging and GPS locators for participants and caregivers (US House of Representatives Bill 34, 2016).

While not the aim of this study, it is also prudent to recall that the vast majority of all care delivered at home comes in the form of unpaid family contributions and is estimated to have been $470 billion in 2017—nearly 10 times the amount spent on caregivers by Medicaid (Reinhard et al., 2019). Additionally, we do not formally address the shortage of available caregivers both within CDM and PAM. As Americans have fewer children and family members move away from home, available caregivers are projected to steadily decline (Redfoot & Houser, 2013). We encourage further research and practices aimed at continued training, education and advocacy for workforce development and retention of agency frontline caregivers.

In conclusion, in a group of dually eligible Medicare and Medicaid individuals receiving DCW services at home as part of a managed LTSS program, the participants in a CDM had no significantly different utilisation than the participants in a PAM. However, total cost of care in the CDM was lower, mainly due to lower LTSS costs. This study highlights the similarities in outcomes for participants in different care models and supports the shifting landscape regarding individual choice in home care.

What is known on this topic?

• Little is known about cost and utilisation outcomes among Medicare and Medicaid managed care participants who appoint a direct-care worker or have one assigned from an agency.

What this paper adds?

• There are baseline differences in participants who appoint their own care worker compared to professional assignment and may reflect illness severity, gender or availability of support in one’s life.

• Policymakers and insurers should support a participant’s choice in caregiver assignment as there does not appear to be a significant difference in utilisation outcomes.

DATA AVAILABILITY STATEMENT

The data that support the findings of this study are available on request from the corresponding author. The data are not publicly available due to privacy or ethical restrictions.