Guided Care and the Cost of Complex Healthcare: A Preliminary Report
Published Online: August 07, 2009
Bruce Leff, MD; Lisa Reider, MHS; Kevin D. Frick, PhD; Daniel O. Scharfstein, ScD; Cynthia M. Boyd, MD, MPH; Katherine Frey, MPH; Lya Karm, MD; and Chad Boult, MD, MPH, MBA
Healthcare for older persons with chronic conditions is very expensive.1 Growing recognition of these costs has led to the development and promulgation of numerous new disease and care management strategies designed to provide high-quality, costeffective chronic care. One such approach is Guided Care (GC).2 Guided Care is provided by a practice-based team that includes a registered nurse, 2 to 5 physicians, and the other members of the office staff. This team provides 8 clinical services to a panel of 50 to 60 of the practice’s highest-risk older patients. For each patient, the Guided Care nurse (GCN) (1) performs a comprehensive assessment at home, (2) creates an evidence-based care guide, (3) monitors and coaches the patient monthly, (4) coordinates the efforts of all of the patient’s providers of healthcare, (5) smooths the patient’s transitions between sites of care, (6) promotes patient self-management, (7) educates and supports family caregivers, and (8) facilitates access to appropriate community resources.
In a pilot study during 2003-2004, GC appeared to improve the quality of chronic care3 and reduce its costs.4 In the initial months of the present cluster-randomized controlled trial (cRCT), patients and physicians rated GC more highly than usual care (UC).5 The aim of the present analysis was to evaluate the effects of GC on the use and costs of health services during the first 8 months of this cRCT. We report these preliminary results now to help inform business and policy decisions related to improving chronic care that must be made before the publication of the study’s final results in 2011.
In 2006, we launched a cRCT of GC in 8 community-based primary care practices located in urban and suburban neighborhoods in the Baltimore, MD–Washington, DC, metropolitan area. Three practices were operated by Kaiser Permanente, a group-model managed care organization; 4 were operated by Johns Hopkins Community Physicians, a statewide network of community-based practices; and 1 was operated by Medstar Physician Partners, a multisite group practice. The study population comprised older persons insured by traditional fee-for-service Medicare (34%) or managed care plans (45% Kaiser Permanente, 21% Tricare/USFHP, a federal health insurance program for retired military personnel and their dependents). Additional details about the study, which was approved by 3 relevant institutional review boards, have been published previously.5
Recruitment of Participants
We screened the insurance claims of all patients of the 8 practices to identify those who were age 65 years or older and at high risk of using health services heavily during the following year, as estimated by the claims-based Hierarchical Condition Category (HCC) predictive model.6 “High risk” was equated with HCC scores of 1.2 or higher, which identifies persons in the highest quartile of risk. Eligible high-risk patients who provided informed consent completed in-home baseline interviews.
Within the 8 practices, we identified 14 “pods,” each of which consisted of 2 to 5 primary care physicians and their consenting high-risk patients. The study’s statistician, blinded to the identities of the pods, randomly allocated each pod (physicians and their patients) to either GC (7 pods) or UC (7 pods).
Registered nurses who had completed a course in Guided Care Nursing joined their assigned pods in May 2006. During the following 6 to 8 months, each GCN established a caseload of 50 to 60 GC patients. The date on which each patient’s care guide was developed was his/her start date for receiving GC.
Patients in the UC group continued to receive the same care they had received before enrolling in the study. We randomly assigned their start dates for receiving UC to reflect the GC patients’ distribution of start dates.
Information about baseline characteristics was obtained from the prerandomization in-home interviews. Data about health service utilization were obtained from paid insurance claims for services used during the period from 1 year before the person’s start date through June 30, 2007. To maximize the capture of most health services, we waited until late November 2007 to begin the process of querying, cleaning, and merging the 3 insurers’ claims databases. We calculated participants’ baseline HCC scores from claims for services used during the 12 months before their start dates. We grouped these health services into 9 categories: hospital, skilled nursing facility, primary care physician, specialty physician, home healthcare, emergency department, durable medical equipment, tests (radiologic and laboratory), and nonmedication treatments (eg, physical therapy, outpatient procedures).
As described in detail previously,5 we imputed values for missing baseline interview questions. We compared the GC and UC patients’ use of health services during the intervals between their start dates and June 30, 2007. For each category of healthcare service, we constructed a marginal regression model to estimate the effect of GC (vs UC) on the mean units of service used per person per year (Table 1). For each category, we modeled the logarithm of the mean as a linear function of treatment group, age, race, sex, education, finances, HCC score, self-rated health, activities of daily living, instrumental activities of daily living, and practice site, plus an offset term for exposure period. Regression parameters were estimated using generalized estimating equations with a working independence covariance structure. The estimated variance-covariance matrix of all the regression estimators was obtained using the sandwich variance technique. The adjusted treatment effect for each outcome was interpreted as the ratio of the mean units of service used per GC recipient (vs UC recipient) over a common exposure period.
To facilitate computation of the difference in costs associated with GC, we first quantified the average units of service used annually by the recipients of UC and multiplied these units by 55 (the average number of patients in a GCN’s caseload). Then we computed the analogous average units of service used by 55 recipients of GC by multiplying the UC units of service by the treatment effects estimated by the marginal regression models. Finally, we calculated the difference between the 2 groups’ aggregate healthcare costs by multiplying the differences between the groups’ annualized use of services by Medicare’s average payment per unit of service.7 Excluded from these estimates are the costs of durable medical equipment, emergency department visits, tests, and treatments, which are extremely heterogeneous and for which average Medicare payment rates are not available.
Confidence intervals for the annualized use of services and cost differences were derived by using a 2-stage bootstrap procedure. In the first stage, individuals were resampled within practices. In the second stage, treatment effect estimates (on a log scale) were sampled from a multivariate normal distribution in which the mean was equal to the estimated treatment effect (on a log scale) and the variance-covariance matrix was equal to that obtained from the generalized estimating equation procedure. A total of 100,000 bootstrap samples were generated, and confidence intervals were formed using the percentile method.
As described in detail previously,5 13,534 older patients were screened for eligibility, 1763 were offered the option of participating, and 904 were randomized by pod to receive either GC (n = 485) or UC (n = 419). We excluded from these analyses utilization by participants whose insurance claims were unavailable (1.4%) and those who rescinded consent (1.8%), died before their start dates (2.2%), or did not have start dates before June 30, 2007 (2.2%). We analyzed data on the remaining 835 participants (n = 433 in the GC group, n = 402 in the UC group).
At baseline, the demographic features, comorbidity, and HCC risk scores of the 2 groups were similar, but the GC participants were less likely to experience economic challenge (10.4% vs 15.5% lacked “enough money at the end of the month”; P = .05) or functional impairment (21.7% vs 28.3% reported difficulty with 2+ instrumental activities of daily living; P = .03), and they were more likely to have better health (23.1% vs 17.3% reported “excellent” or “very good” health; P <.001) and to be insured by Tricare/USFHP (26.8% vs 16.5%, P = .001). Other details of these baseline characteristics have been reported previously.5 The mean period of observation—from participants’ start dates through June 30, 2007—was longer for the GC group (8.3 months vs 7.8 months; P <.05); 83% of all participants were observed for at least 6 months.
PDF is available on the last page.