Effectiveness of a Depression Care Management Initiative in Home Healthcare

The goal of this research is to improve depression treatment and outcomes among elderly home healthcare patients. Homecare nursing is a major source of health care for a large and growing number of medically ill or injured older adults who are homebound by illness or disability. Clinically significant depression is twice as prevalent in this patient population compared to similarly aged primary care patients. Depression can be effectively treated in older adults, and treatment guidelines have been developed to help physicians make treatment decisions for their depressed older patients. However, medical home healthcare patients rarely receive guideline-consistent treatment for depression. This research tests the effectiveness of an intervention, Depression Care for Patients at Home" (CAREPATH), on two outcomes: 1. Depression treatment (i.e., initiate treatment or have a change in treatment that is consistent with guidelines), and 2. Depressive symptoms (i.e., reduction in depressive symptoms over time). The CAREPATH protocol was designed in partnership with home healthcare providers. It includes the major elements of depression care management models that have proven effective in primary care but restructures these elements to fit the clinical needs of home healthcare patients and for consistency with home healthcare practice. The intervention itself is designed to be ecologically sensitive to maximize the feasibility and generalizability of the program. The CAREPATH Intervention is being tested within six home healthcare agencies located in Vermont/New Hampshire, New York, Pennsylvania, Michigan, Florida, and Arkansas. The design includes randomization of ~20 teams of nurses to CAREPATH or usual care. The impact of CAREPATH on depression treatment is tested with all eligible patients (N~600) using data collected routinely by all agencies as these are the kinds of data that agencies typically use for quality assurance. Depressive symptoms outcomes are tested using the Hamilton Depression Rating Scale (HDRS) collected by researcher staff from (N=300) patients who consent to in-person baseline and telephone follow-up interviews at 12, 24, and 52 weeks.

Data Plan: H1 Depression Treatment : Patients of CAREPATH home nurses with clinically significant depressive symptoms will be more likely to receive a "guideline-based step" in their treatment of depression than patients of nurses providing usual care. This analysis will be tested using the merged administrative data set. A mixed-effects logistic regression analyses will compare patients in the intervention and usual care groups on change in depression treatment received. The primary independent variable (a fixed effect) is group and the dependent variable is change (from start-of-care to discharge to guideline consistent treatment received (yes/no). The structure of these data from this cluster randomized trial involves three level mixed-effects models in which patients are nested within nurse and nurse within team supervisor. These analyses will be preceded by mixed-effects models that compare groups on sociodemographic and clinical variables. Those variables that differ significantly will be included as covariates in the primary analysis that examines the intervention effect (described above).

H2 Depressive Symptoms: Patients of CAREPATH home nurses with clinically significant depressive symptoms will have greater reduction in depressive symptomatology (HDRS change from baseline) by 3, 6 and 12 months of the baseline interview than patients receiving usual care. This analysis will be tested using data collected from patient research interviews. A mixed-effects linear regression analyses will compare patients in the intervention and usual care groups on change in severity of depressive symptoms from baseline. Covariates in the model will be selected as described in H1.

D9.3 Exploratory Analyses: . S1. Different Outcomes:. Whether the intervention reduces the risk of poor outcomes as measured by Medicare's "Outcome-Based Quality Indicators" (OBQI) and targeted adverse events, including: decline in activities of daily living, discharge to hospital, and/or falls. This analysis will be tested using the merged administrative data set. Mixed-effects analyses will be conducted on the following OBQI outcomes and adverse events. Mixed-effects linear regression will be used for the continuous measures (e.g., ADL decline) whereas mixed-effects logistic regression analyses will be used on binary outcomes (e.g., fall). The choice of covariates and the structure of the data will conform to that described for H1. We anticipate that some of these exploratory analyses will be sufficiently power for statistical tests (e.g., decline in ADL), yet others (e.g., adverse fall events) will be examined for the direction and magnitude of effects rather than statistical significance.

S2 Patient Characteristics as Moderators: Whether the effects of the intervention on patient outcomes and quality of care differ by depression severity, patient location (e.g., rural vs. urban), race/ethnicity (White, Black, Hispanic, Native American), availability of social support (caregiver), health status, or cognitive impairment. Separate models will examine each patient characteristic as a moderator using mixed-effects linear or logistic regression analyses. The independent variables will include intervention and the respective hypothesized mediating (from post baseline) or moderating (from baseline) effects (described below). Initially the main effects will be tested. Then subsequent models will examine the incremental contribution of the interaction of intervention with each of the hypothesized moderating effects.

D10 POWER ANALYSIS Power analyses for the primary hypotheses were conducted based on the following assumptions about sample size: 5 agencies; 4 nurse teams per agency, 5 nurses per team, and 5 patients subjects per nurse. These assumptions result in a patient sample size of 500 patients (5*4*5*5). We estimate that the number of patients who consent to research interviews will be about half of the patients who are eligible based in the agency's database data (i.e., 60% participation at baseline; 85% of baseline patients eligible for follow-up). Thus the number of patients in the agency's database that could be included in analyses using the this source of data will be at least 1,000.

Other assumptions for the power analyses included a two-tailed alpha = 0.05, 12 and 24 week follow-up assessments for each subject, and an attrition rate of 15%. This rate is based on our six month follow-up rates as well as our experience with other samples of community-dwelling frail elders (e.g., home care patients), where we have found that obtaining the first interview is far more difficult than following older adults overtime once they have met and talked with us. Because computer algorithms are not readily available for conducting power analyses for three-level mixed-effects models, power estimates for testing H1 and H2 are based on simulations described below, that involved 1000 simulation runs for each combination of specifications.

H1 Depression Treatment: The simulations considered two intraclass correlations reflecting variations in level 1 (subject-level intraclass correlation within nurse) and level 2 (nurse-level intraclass correlation within team). Statistical power to detect the hypothesized effects with the anticipated sample size, will exceed >80%.

H2 Depressive Symptoms: Power analyses was conducted based on simulation using Mixed-effects models for level 1 and level 2 level random intercepts 3-level linear mixed effects regression model. We hypothesized medium intervention effects (Cohen's d) with a standardized group mean difference in HDRS change from the baseline: 0.5 and 0.6. (These correspond to differences in HDRS changes = 3.43, and 4.11 based on an estimated residual standard deviation = 6.85 of HAM-D changes from the TRIAD study.) The table shows that power to detect effect size > 0.5 is adequate (>80%).