INTRODUCTION

A small proportion of patients are responsible for a large majority of health care costs, often through frequent hospitalization.1 These patients are typically older and suffer from multimorbidity2,3, – 4 and functional limitations.5 , 6 Such patients may require more frequent primary care visits, care from multiple subspecialists, home care services, or assistance from social workers, pharmacists, or physical therapists to optimize the management of their medical conditions and their overall health. Additionally, as they use more services, the risk of care fragmentation increases, as does the burden of coordinating care. In a traditional primary care setting, the capability to coordinate complex care may be lacking, leading to preventable hospitalizations.

An intuitive approach to caring for these vulnerable patients is to broaden the scope, increase the intensity, and improve the coordination of outpatient care, in the hope that this will lead to reduced hospital use and lower costs.7 , 8 Increased access to primary care is associated with decreased rates of hospitalizations for certain conditions,9 , 10 and discontinuity of primary care is associated with increased risk of hospitalization.11 Outpatient care redesign efforts include patient-centered medical homes (PCMH),12 , 13 hospital-based care transition programs,14 , 15 and external case management.16 , 17 However, the effects of such efforts on utilization and clinical outcomes have been inconsistent.18,19,20,21,22, – 23

An alternative approach is intensive primary care (IPC), also referred to as ambulatory intensive care.7 , 24 In such programs, patients at high risk for hospitalization or death are enrolled in a separate program (that may be within an existing primary care setting) that addresses a spectrum of medical and social needs, and coordinates care across settings. While PCMH and IPC both use interdisciplinary teams, IPC programs differ in that they specifically target populations at high risk of hospitalization, and provide a higher intensity of care and a broader range of services. IPC programs vary substantially in structure, function, and site of care, but the impact of these different approaches on hospitalization or mortality is unknown.

The objective of this work was to classify interdisciplinary, multicomponent IPC programs according to program characteristics,25 and to evaluate the effectiveness of these programs in reducing hospitalizations, emergency department (ED) visits, and mortality among patients at high risk for hospitalization or death.

METHODS

This review was initially requested for the Department of Veterans Affairs Patient Aligned Care Team (PACT) Health Delivery Committee to inform the creation of an IPC program.26

Data Sources and Searches

We searched for systematic reviews, controlled clinical trials, and observational studies in PubMed (1946 through 3/01/2017), CINAHL (1981 through 3/01/2017), the Cochrane Central Register of Controlled Trials (first quarter 2017), and the Cochrane Database of Reviews of Effects (first quarter 2017) using standard search terms for high utilizing complex patients, IPC, and hospital use. Additional citations were identified from reference lists, hand searching, and consultation with content experts. We limited the search to articles involving human subjects and available in the English language.

Study Selection

A description of study selection criteria is presented in Table 1. Two investigators reviewed titles, abstracts, and full text articles. One reviewer had methodological expertise, and the other had clinical content expertise. There was a high level of agreement, and all disagreements were resolved by consensus.

Table 1 Study Selection Criteria
Full size table

Data Extraction and Quality Assessment

We abstracted data from all included studies on population, intervention, comparator, and timing characteristics and results for each included outcome. All data abstraction and internal validity ratings were first completed by one reviewer and then checked by another. All disagreements were resolved using consensus.

We used predefined criteria to rate the internal validity of all individual studies. Risk of bias in controlled clinical trials was assessed using the Cochrane Collaboration’s tool and was judged as low, unclear, or high.28 We rated the internal validity (quality) of controlled observational studies as good, fair, or poor, using methods developed by the Drug Effectiveness Review Project (DERP) and based on the adequacy of the patient selection process, completeness of follow-up, adequacy of outcome ascertainment, use of acceptable statistical techniques to minimize potential confounding factors, and whether the duration of follow-up was reasonable to capture investigated events. Ratings of low risk of bias and good quality indicate high confidence in study findings. Ratings of unclear risk of bias and fair quality indicate the presence of plausible bias or some flaws that raise some doubt about study findings. Ratings of high risk of bias and poor quality indicate the presence of plausible bias or serious flaws that seriously weaken our confidence in study findings.29 We graded strength of a body of evidence based on the guidance established for the AHRQ Evidence-based Practice Centers program.30 This approach incorporates four key domains: risk of bias (includes study design and aggregate quality), consistency, directness, and precision of the evidence.

Program Classification

We classified programs as primary care replacement (home-based), primary care replacement (clinic-based), or primary care augmentation, and assessed the impact of outcomes separately for each category. We derived our approach primarily from a classification developed by Yee et al.,25 which grouped programs based on similarities in patient selection methods, scope of provided services, and organization of service delivery. As themes emerged during the review, we considered alternative ways to classify programs, but found that these categories were the most effective and consistent with other literature.7

RESULTS

The literature flow diagram (Fig. 1) summarizes the results of the search and study selection processes. Searches resulted in 4151 potentially relevant articles. Of these, we included 11 randomized controlled trials31,32,33,34,35,36,37,38,39,40, – 41 and seven observational studies.42,43,44,45,46,47, – 48

Figure 1
figure 1

Study identification and selection.

Full size image

Tables 2 and 3 summarize the study and intervention characteristics. The majority of the studies involved mostly elderly, non-white females, but four studies focused on veterans who were predominantly male.36 , 37 , 45 , 48 Most studies were small (N < 300), and ranged in size from 59 to 3889 participants, with one observational study that included 364,972 participants. Follow-up duration ranged from 6 months to 4 years. Three studies selected patients based on high utilization of inpatient or emergency department services.39 , 44 , 47 The remaining studies selected patients who were at least moderately disabled in activities of daily living 31 , 35,36,37, – 38 , 48 , 57; homeless46; aged ≥ 65, 31 , 33 , 34 , 48 , 58 ≥ 67,45 ≥ 70,32 , 35 or ≥ 75 years42; had income below 200% of the federal poverty level34 , 58; and/or who met nursing home eligibility criteria.43 Detailed patient selection criteria are described in Table 2.

Table 2 Study and Intervention Characteristics (Stratified by Program Type)
Full size table
Table 3 Intervention Characteristics (Stratified by Program Type)
Full size table

Among the programs, five were home-based primary care replacement models (referred to as home-based care models), where a dedicated interdisciplinary primary care team delivered all needed services in patients' homes.36,37, – 38 , 44 Five were clinic-based primary care replacement models (referred to as clinic-based IPC), where patients transferred care from their regular primary care physicians to dedicated clinics that provided only high-intensity care,31 , 32 , 43 , 57 and eight used primary care augmentation, where additional high-intensity services enhanced ongoing care from their regular primary care physician.33,34, – 35 , 39 , 42 , 58 Because of the clinical heterogeneity in design, patient population, interventions, and outcome assessment methods, we did not pool any data across studies.

Most studies were fair quality/unclear risk of bias or poor quality/high risk of bias. Common methodological limitations among RCTs were lack of blinding, unclear allocation concealment, and unclear methods for addressing incomplete outcome data. Among observational studies, common methodological limitations were lack of unbiased patient selection and lack of control for potential confounding variables. Study results are summarized in Table 4.

Table 4 Summary of Results (Stratified by Program Type, Intervention vs. Control)
Full size table

Home-Based Models

There is limited evidence that home-based care models reduce hospital utilization. VA HBPC enrollment was associated with fewer ACSC hospitalizations over 3–4 years in a recent national VA cohort study of 364,972 elderly veterans with diabetes (Table 2).45 VA HBPC enrollment did not reduce hospitalization in two earlier studies, but this may be due to their smaller sample sizes (N = 233 to 1966), shorter follow-up (6–12 months), or changes in HBPC over time.36 , 37 Other home-based programs implemented by single centers in Barcelona and Stockholm had mixed findings regarding hospital utilization, but our confidence in these findings is very low due to their serious limitations.38 , 44 For example, it is possible that the reduced hospital utilization in Hospital de Barcelona’s home care program was exaggerated, because the urban setting characteristic of this home care program is typically associated with higher health care utilization than the rural setting of the control group.59 Similarly, in Stockholm’s in-home hospital care program, the reduction in long-stay hospital utilization at 6 months may have been exaggerated, because participants in the home care group had a higher number of medical diagnoses than the control group (4.5 vs. 3.9; P = 0.003).

Clinic-Based Intensive Primary Care

The SIPA (French acronym for System of Integrated Care for Older Persons)31 , 57 in Montreal and the PACE (Program of All-Inclusive Care for the Elderly) in the U.S.43 were both associated with reduced hospital utilization at 19 to 24 months.31 , 43 , 57 Additionally, PACE demonstrated reduced risk of mortality in the first 18 months compared to usual care, but it increased between months 18 and 24. Both focused on assisting frail elderly individuals with functional limitations. The main methodological limitation of these studies was possible attrition bias. For example, in the PACE study, there were five times as many exclusions from the intervention group due to incomplete data.

Geriatric evaluation and management (GEM), ambulatory geriatric units (AGe-FIT), and Homeless PACT (H-PACT) were generally not associated with significant reductions in mortality or hospital utilization over a period of 12–25 months.32 , 41 , 46 The exception was AGe-FIT, which reduced the mean number of inpatient days over 24 months (11.5 vs. 15.2, p = 0.035) in 382 elderly individuals in Sweden who had been hospitalized at least three times during the past year.41 Among these models, H-PACT is unique in that it focuses on homeless individuals at high risk of hospitalization and offers a separate high-intensity primary care clinic that includes resources such as improved walk-in capacity, co-located mental health, housing, hygiene, clothing, and food vouchers, and intensive case management. While a 6-month pre–post assessment of utilization demonstrated a 19% decrease in ED visits and a 34.7% reduction in hospitalizations, there was no control group, no statistical test of significance, and no attempt to control for pre-existing trend or regression to the mean.43

Primary Care Augmentation

Three studies using the Geriatric Resources for Assessment and Care of Elders (GRACE) model34 , 47 , 48 , 58 demonstrated reductions in hospitalization rates and/or emergency room visits. A GRACE randomized trial was performed among 951 low-income seniors across six community-based health centers in Indianapolis, Indiana.34 , 58 The trial’s main methodological limitation was that it compared groups of physicians who served different patient populations, as evidenced by the significantly lower rate of county medical assistance use in the GRACE intervention group at baseline (83.7% vs. 89.0%; P = 0.02). GRACE’s impact on hospital utilization was limited to a subgroup of 226 patients with the highest risk of baseline hospitalization (Probability of Repeated Admission score of 0.4 or higher), and only in the second year. Also, as baseline characteristics were not provided for the high-risk subgroup, we could not rule out that potential differences in patient characteristics may have mediated the positive effect of the GRACE intervention on outcomes. In a study of GRACE in a VA Medical Center in Marion County, Indiana, the new model was implemented in four clinics, while one clinic that declined to participate served as a control group. Utilization was compared in the 12 months prior to and after implementation, in both GRACE and control groups, using a difference-in-differences approach. There were no significant differences in 30-day readmission or hospital admission, but there was a significant decrease in inpatient days associated with GRACE (−221.4 annual bed days/100 veterans, p = 0.01).47 In a second study, GRACE was adapted to an academic medical center in San Francisco, and named Care Support. Comparison of utilization in the 6 months pre- and post-Care Support enrollment demonstrated a decline in the mean number of ED visits (5.5 to 0, p = 0.015) and hospitalizations (5.5 to 0, p < 0.001). This study included no comparison group, and no method to control for regression to the mean or pre-existing trend.47

None of the other primary care augmentation interventions succeeded in reducing mortality or hospital utilization outcomes. Among the remaining studies, ImPACT is the largest, most recent, and highest-quality randomized controlled trial. In ImPACT, patients were identified as the 5% highest-cost patients or those with the 5% highest risk of hospitalization using a risk prediction algorithm. A multidisciplinary team consisting of a nurse practitioner, physician, social worker, and recreational therapist performed a comprehensive patient assessment, provided care management for medical and social service needs, and coordinated care with VA and non-VA providers. Notably, in ImPACT, both the control group and the intervention group saw reductions in acute and extended care utilization and cost of care, and there was no significant difference between groups.

DISCUSSION

In this review, we identified 18 studies of IPC interventions and categorized them into three major types: home-based programs, clinic-based IPC, and primary care augmentation programs. The programs varied in the way they identified and screened patients for enrollment, though most focused on older adults with functional limitations. Formalized care planning was present in all interventions, many relying upon the comprehensive geriatric assessment. All programs utilized multidisciplinary staff to meet a range of patient needs, and most commonly included physicians, nurses, social workers, physical therapists, mental health providers, and pharmacists. Some clinic-based IPC programs and primary care augmentation programs also included home visits. Overall, we found moderate- to low-strength evidence that some programs (VA-HBPC, PACE, SIPA, and GRACE) were associated with fewer hospital admissions, but no evidence of impact on ED visits or mortality.

While there has been increasing interest in IPC,60 relatively few studies met our inclusion criteria. Notably, many of the studies we did identify were older and were completed before recent changes in health care such as the widespread adoption of electronic health records. Additionally, several studies used weak study designs—for example, pre–post assessment of utilization without a control group. This design is susceptible to both temporal trends, such as decreasing hospitalization rates,61 and regression to the mean. In the ImPACT trial, reductions in utilization and cost were seen in both intervention and control patients, demonstrating the need for study designs that include control groups.40 Additionally, our review lacked some often-cited IPC interventions including the Camden Coalition or the AtlantiCare Special Care Center,62,63, – 64 as evaluations of these programs do not appear in the peer-reviewed literature. This is likely because programs often do not include evaluations, and among those that do, evaluations are difficult to perform and may take years to complete. Future IPC interventions should include integrated evaluations with rigorous study designs, ideally including randomization.

The current evidence is not adequate to reach a conclusion as to what the right population is for IPC interventions. Patient selection criteria were not well described in many of the studies we reviewed, and no clear patterns of impact on outcomes emerged. Some programs used risk prediction algorithms that included prior diagnoses, laboratory values, and utilization.33 , 40 Others focused on specific subpopulations and selected patients based on older age, an inability to perform activities of daily living, or homelessness. One chose patients based purely on prior health care utilization.47 A potential strategy for new programs would be to identify their specific high-risk patients and assess their needs early in program design, to ensure that provided services fit population needs.

Most of the programs we reviewed used large multidisciplinary teams to deliver a range of specialized services and to coordinate the care of complex patients. However, for complex patients, adding more providers to their care may increase discontinuity with their primary provider and increase the burden of care coordination. Given the negative results of many of these studies, it is possible that attempts to manage complex care using large multidisciplinary teams may be ineffective for some high-needs patients, as the burden of coordination may outweigh the benefits of the specialized skills of each team member. An alternative approach would be to explicitly simplify care to minimize discontinuities and coordination burden by using fewer providers who have a wider scope of practice to care for patients in all settings, such as both in and out of the hospital. In such models, the role of “care coordinator” may be irrelevant.65

Among the studies reviewed, success in reducing hospitalization was achieved in some programs and not others, and success was context-dependent. We had hoped to identify key program features, such as patient selection criteria, that may have contributed to the success or failure of these programs. Unfortunately, reporting of key intervention characteristics was inconsistent, which is a common problem in complex multicomponent intervention studies.66 In addition, the data collected on intervention fidelity, implementation process, and contextual factors at individual intervention sites varied among studies. Future works could use both qualitative and quantitative methods to examine intervention fidelity, implementation, and context.67,68, – 69 Such data could provide insight into what makes IPC programs work, for whom, and why.70 The VA’s PACT Intensive Management Program has incorporated many of these strategies in their implementation and evaluation of a multisite IPC program within PACT.71 , 72

Our work has several limitations. First, while we attempted to create a precise definition of IPC and differentiate it from other efforts to intensify ambulatory care for high-risk patients (such as care management), we found significant overlap in approaches, both between our identified care models and with other care intensification strategies. Additionally, the lack of a standard taxonomy made this topic difficult to search for. Although we attempted to use an exhaustive list of search terms, our search may have missed some relevant studies. Second, our criteria primarily identified geriatric care models focused on older adults with multiple chronic conditions and functional limitations. Fewer identified programs focused on other at-risk populations, and thus our review may have more limited generalizability. However, this highlights the fact that older adults with multiple chronic conditions and functional limitations are an at-risk population that may be an appropriate target for IPC interventions. Finally, some efforts by accountable care organizations to improve care coordination for complex patients spanned the fields of primary care, specialty care, hospital care, and post-acute care, combining features of the PCMH, transitional care programs, and IPC models.73 The principle underlying these approaches is that in order to achieve the triple aim of improving quality, reducing costs, and improving patient experience, one must go beyond innovations in primary care delivery. However, this review does not address such integrated approaches. IPC programs that had unimpressive results in our review might turn out to be valuable components of a more integrated approach.73

We identified studies representing each of three models of IPC: home-based models, clinic-based IPC, and primary care augmentation. Programs varied in their success in reducing hospitalizations. Future IPC interventions should include rigorous evaluations that include consistent and standardized reporting of intervention characteristics and that collect data on implementation and contextual factors, to further our understanding of what makes such programs successful.