The Challenge of Proving the Value of Medical Home Care

Federman et al. reported on their randomized controlled trial (RCT) testing the impact of home-based primary care (HBPC) on patient satisfaction and other health outcomes for seriously ill patients with limited mobility.^[1] This study highlights some of the challenges of this type of research. Care was delivered at one of the largest and most experienced academic house call programs in the country, in a large city, so there is little doubt that there were many eligible subjects nearby and there were capable workers to deliver care. Yet, this study was hindered by slow enrollment in the intervention, and early termination on advice of safety monitors mainly because of a numerically small but noteworthy difference in mortality among patients randomized to the intervention arm. The result was an underpowered study (only 112 per group), a lower dose of the intervention—only 65% of the intervention arm received HBPC—, and a problem with randomization that was adverse to the intervention. Despite these limitations, the study reported a 17.8% relative decrease in hospitalization, and significantly higher patient satisfaction with care for those who received home-based team care.

The clinical benefits of home-based care have been many times documented. These include more timely access, better continuity, and improved ability of providers to accurately assess patients and their support systems and then to match care plans to patient and caregivers needs. These consistently result in higher patient and caregiver satisfaction, as was again documented in this study.

A variety of U.S. studies have also pointed in the direction of economic benefits coming from models that provide team-based medical care in the home. In the VA health care system, with many thousands of HBPC patients, combined Medicare and VA costs were lower by 12% when served by an HBPC team than the costs predicted by a carefully calibrated prospective model.^[2] In a matched case analysis, costs of caring for patients in HBPC at Medstar health system were 17% lower than controls.^[3] In the GRACE program RCT, where structured quarterly home visits by nurse practitioners augmented office-based primary care, a planned sub-analysis of the highest risk subgroup showed reduced acute care use in Year 2.^[4] And at the University of Pennsylvania, in another RCT, relatively intensive short-term post-hospital medical care at home by hospital-based nurse practitioners resulted in 50% reduction in total costs and readmissions.^[5] Reduced hospital use by patients receiving home-based team care is repeatedly observed during these studies, validated in data syntheses,^[6] and is again demonstrated in the present RCT, making a likely basis for achieving savings.

Of note, in the current study, we do not see reported the extent of other in-home services, such as skilled home health care, received by subjects in the two groups during the study period. This might skew the results if one group received disproportionately more of those services. Note that in Naylor's RCT there was an equivalent total dose of in-home services by nursing professionals, comparing usual care with the intervention which intermixed visits from mobile nurse practitioners with agency nurse visits, resulting in far better outcomes. It is difficult in usual care models to effectively manage these complex patients from a distance working through a home health agency.

Given the logical appeal of a better clinical care model, and suggestive evidence of cost containment, the dearth of high-quality RCT data testing the longitudinal HBPC model is striking. It is undeniable that there is a subpopulation of individuals in the U.S., immobile and chronically ill, numbering around 2–3 million, whose regular access to care, continuity, and satisfaction with care, all could be helped by longitudinal home-based health care. As noted by the authors, there are only two published U.S. RCTs of HBPC, both conducted three or four decades ago. One trial with 150 persons per arm was in Rochester, NY, showing much improved patient satisfaction plus lower costs at the end of life. The other was conducted in the VA system, again showing improved satisfaction but no economic benefit, implemented at 16 heterogeneous sites where HBPC model fidelity between sites was in question.

The difficulty and cost of formally studying this care model are again exemplified in this current study, with slow subject accrual, small sample size leading to the study being under-powered, and many potential confounding variables. The latter issue mandates extensive, and thus expensive data collection to accurately define the population, and to exclude potential confounding influences on the results. These barriers undoubtedly have contributed to the paucity of rigorous published trials. In addition to having extensive clinical heterogeneity, home-limited patients tend to have highly complex interactions between social supports and health outcomes, as documented in a recent qualitative study.^[7] Thus, along with medical co-morbidity, it is necessary to measure social vulnerabilities, plus trajectory of functional status and service use over time to avoid adverse selection biasing study results.

With these things in mind, we would take note of the published data (Table 1) from the CMMI Independence at Home (IAH) demonstration of HBPC, selectively delivered to a targeted high-cost population.^[8] The demonstration's targeting criteria have been shown to be a reliable and practical means of identifying a high-cost, high-need population, similar to the population in the current RCT. Table 1 displays the actual costs incurred by thousands of enrolled Medicare beneficiaries compared to the expected costs based on a well-calibrated, risk-adjusted predictive model. This model was iteratively improved in the initial 2 years of the demonstration based on new insights gained from studying the population and the care model. Over time there was some site dropout, particularly of less experienced and less successful sites. There was accordingly some reduction in the total enrolled population, though still a substantial sample. The data indicate consistent, substantial cost reductions across numerous practice sites over a span of several years.

A somewhat different interpretation of these results has been offered, relying on an analysis that focused not on whether HBPC as a model of care would produce high-quality care at lower cost, but rather on whether new incentives with shared savings had changed the behavior of HBPC providers.^[9] Careful reading of the appendices from this evaluation report suggests that the costs associated with the intervention cohort across all the varied participating sites were 7% to 9% ($300 to $400 per month) lower than the costs for the matched comparison group in the later years of the demonstration.^[9] These results mirrored CMS reports of considerable savings. Several sites had much larger reductions in costs, around 20%–30%. Ultimately, some uncertainty remains about what total costs and service utilization would be in the absence of the HBPC intervention that costs about $400 a month. That said, many other expensive interventions, including drugs and other types of care, that are now considered standards of care in the U.S., have undergone far less intensive scrutiny.

In the present study, the imbalance in deaths, the main reason for early termination of the trial, raises more questions about the effectiveness of the randomization than about the intervention. The observed mortality rate for the intervention arm (21%) is similar to the carefully matched control group of the IAH demonstration. The mortality rate for patients that received HBPC in this trial (15%) was similar to the IAH intervention group, and below the national mean mortality for patients receiving HBPC in 2018 (18%). What is unusual is not the observed mortality in Federman's intervention group but the low control group mortality (10%), suggesting a failure of randomization to account for population complexity in this small sample, disadvantaging the intervention arm and biasing against finding an intervention benefit.

Further considering the high mortality, we note as the authors did that MOLST forms were completed by substantially more intervention participants than controls. Peaceful death at home may have been a preferred outcome compared with a continued sequence of ultimately futile, often traumatic acute care experiences. The authors carefully investigated looking for signs that the patients who died received substandard care and found none.

Despite an apparent unfavorable distribution of underlying complexity, the results of the study by Federman et al. confirm two decades of studies finding lower hospitalization rates under HBPC. This would again argue for definitive policy action by CMS to make the team-based HBPC model more sustainable and more widely available for selected individuals who are most likely to benefit. Careful reading of CMS's evaluation of the IAH demonstration, and CMS's announced savings, both suggest sufficient cost reductions to fund team-based HBPC. Or if more proof is really needed, perhaps this would argue for a larger and more definitive RCT, if this could be done. Which payer would want to fund care in a model with more hospitalizations? And, which patients would want to be in the control group, we wonder, if they were sick, homebound, and could have well-trained providers deliver team-based medical care at home?