Currently Viewing:
The American Journal of Managed Care December 2014
Quality of End-of-Life Care for Cancer Patients: Does Home Hospice Care Matter?
Netta Bentur, PhD; Shirli Resnizky, MA; Ran Balicer, MD; and Tsofia Eilat-Tsanani, MD
Out-of-Plan Pharmacy Use: Insights Into Patient Behavior
Thomas Delate, PhD; Alexander P. Block, PharmD; Deanna Kurz, BA; and Sarah J. Billups, PharmD
Paying for Telemedicine
Robert S. Rudin, PhD; David Auerbach, PhD; Mikhail Zaydman, BS; and Ateev Mehrotra, MD
Validating Electronic Cancer Quality Measures at Veterans Health Administration
Jeremy B. Shelton, MD, MSHS; Ted A. Skolarus, MD, MPH; Diana Ordin, MD, MPH; Jennifer Malin, MD, PhD; AnnaLiza Antonio, MS; Joan Ryoo, MD, MSHS; and Christopher S. Saigal, MD
Did They Come to the Dance? Insurer Participation in Exchanges
Jean M. Abraham, PhD; Roger Feldman, PhD; and Kosali Simon, PhD
ACO Contracting With Private and Public Payers: A Baseline Comparative Analysis
Valerie A. Lewis, PhD; Carrie H. Colla, PhD; William L. Schpero, MPH; Stephen M. Shortell, PhD, MPH, MBA; and Elliott S. Fisher, MD, MPH
Reference-Based Pricing: An Evidence-Based Solution for Lab Services Shopping
L. Doug Melton, PhD, MPH; Kent Bradley, MD, MPH, MBA; Patricia Lin Fu, MPH; Raegan Armata, BS, MBA; and James B. Parr, BA
Addressing Cost Barriers to Medications: A Survey of Patients Requesting Financial Assistance
David Grande, MD, MPA; Margaret Lowenstein, MD, MPhil; Madeleine Tardif, BA; and Carolyn Cannuscio, ScD
Preconsultation Exchange in the United States: Use, Awareness, and Attitudes
Justin L. Sewell, MD, MPH; Katherine S. Telischak, MSc; Lukejohn W. Day, MD; Neil Kirschner, PhD; and Arlene Weissman, PhD
Medicare Star Excludes Diabetes Patients With Poor CVD Risk Factor Control
Julie Schmittdiel, PhD; Marsha Raebel, PharmD; Wendy Dyer, MS; John Steiner, MD, MPH; Glenn Goodrich, MS; Andy Karter, PhD; and Gregory Nichols, PhD
There's More Than One Way to Build a Medical Home
Manasi A. Tirodkar, PhD, MS; Suzanne Morton, MPH, MBA; Thomas Whiting, MPA; Patrick Monahan, MD; Elexis McBee, DO; Robert Saunders, PhD; and Sarah Hudson Scholle, DrPH, MPH
Improving Medication Understanding Among Latinos Through Illustrated Medication Lists
Arun Mohan, MD, MBA; M. Brian Riley, MA; Brian Schmotzer, MS; Dane R. Boyington, PhD; and Sunil Kripalani, MD, MSc
Currently Reading
Predicting Nursing Home Placement Among Home- and Community-Based Services Program Participants
Melissa A. Greiner, MS; Laura G. Qualls, MS; Isao Iwata, MD, PhD, EdM; Heidi K. White, MD; Sheila L. Molony, PhD, APRN, GNP-BC; M. Terry Sullivan, RN, MSW, MSN; Bonnie Burke, MS; Kevin A. Schulman, MD; and Soko Setoguchi, MD, DrPH

Predicting Nursing Home Placement Among Home- and Community-Based Services Program Participants

Melissa A. Greiner, MS; Laura G. Qualls, MS; Isao Iwata, MD, PhD, EdM; Heidi K. White, MD; Sheila L. Molony, PhD, APRN, GNP-BC; M. Terry Sullivan, RN, MSW, MSN; Bonnie Burke, MS; Kevin A. Schulman, MD; and Soko Setoguchi, MD, DrPH
The authors developed a model to identify participants in a home- and community-based services program who are at highest risk for long-term nursing home placement.
ABSTRACT
Background
Several states offer publicly funded–care management programs to prevent long-term care placement of high-risk Medicaid beneficiaries. Understanding participant risk factors and services that may prevent long-term care placement can facilitate efficient allocation of program resources.

Objectives
To develop a practical prediction model to identify participants in a home- and community-based services program who are at highest risk for long-term nursing home placement, and to examine participant-level and program-level predictors of nursing home placement.

Study Design
In a retrospective observational study, we used deidentified data for participants in the Connecticut Home Care Program for Elders who completed an annual assessment survey between 2005 and 2010.

Methods
We analyzed data on patient characteristics, use of program services, and short-term facility admissions in the previous year. We used logistic regression models with random effects to predict nursing home placement. The main outcome measures were long-term nursing home placement within 180 days or 1 year of assessment.

Results
Among 10,975 study participants, 1249 (11.4%) had nursing home placement within 1 year of annual assessment. Risk factors included Alzheimer's disease (odds ratio [OR], 1.30; 95% CI, 1.18-1.43), money management dependency (OR, 1.33; 95% CI, 1.18-1.51), living alone (OR, 1.53; 95% CI, 1.31-1.80), and number of prior short-term skilled nursing facility stays (OR, 1.46; 95% CI, 1.31-1.62). Use of a personal care assistance service was associated with 46% lower odds of nursing home placement. The model C statistic was 0.76 in the validation cohort.

Conclusions
A model using information from a home- and community-based service program had strong discrimination to predict risk of long-term nursing home placement and can be used to identify high-risk participants for targeted interventions.

Am J Manag Care. 2014;20(12):e535-e546
A model using information from a home- and community-based service program had strong discrimination to predict risk of long-term nursing home placement and can be used to identify high-risk participants for targeted interventions.
  • Approximately 11% of participants in a home- and community-based service program were placed in a nursing home within 1 year.
  • Risk factors for long-term nursing home placement include Alzheimer's disease, money management dependency, living alone, and the number of prior short-term skilled nursing facility stays.
  • Use of a personal care assistance service is associated with significantly lower odds of nursing home placement.
Lifetime risk of nursing home use is estimated at more than 40% and is projected to increase with greater life expectancy among Baby Boomer retirees.1 Medicaid is the primary payer of nursing home services in the United States at an average annual cost of $84,000 per beneficiary.2 In 2010, long-term care services for older patients accounted for more than one-third of state Medicaid spending.3 At a total annual cost of over $140 billion, Medicaid costs for long-term care will likely be part of ongoing discussions about state and federal deficit reduction.2,3

In efforts to curb these costs, many states have moved towards home- and community-based services (HCBS)—programs that aim to prevent placement of high-risk Medicaid beneficiaries into long-term nursing facilities. These programs account for 45% of Medicaid long-term care spending,3 and research has shown that such programs may be effective.4-8 The Patient Protection and Affordable Care Act (PPACA) provides monetary incentives to states that implement HCBS programs as alternatives to nursing homes.9

Although a fair amount is known about the predictors of nursing home use,10-19 validated prediction models for long-term nursing home placement in high-risk, HCBS populations have not been studied.20,21 Understanding the risk factors for nursing home placement and identifying services that may prevent such placement can facilitate efficient allocation of resources among HCBS program participants. We used clinical and administrative databases for elderly participants in a state- and waiver-funded HCBS program to develop a practical model to predict the risk of long-term nursing home placement and to examine associated participant characteristics and program services.

METHODS

Data Sources


We used clinical and administrative data from the Connecticut Home Care Program for Elders (CHCPE) provided by Connecticut Community Care, Inc (CCCI). CHCPE is a publicly funded care-management program that provides preventive home care services to older Connecticut residents who are at risk for permanent nursing home placement. CCCI provided deidentified data for all clients from 2005 through 2011. The data included baseline eligibility evaluations and annual reassessments conducted by CCCI primary care managers, which contain demographic characteristics, medical history, functional ability, social support, and financial assistance data elements in the “Modified Community Assessment Tool” published by the Connecticut Department of Social Services (CDSS).22 Additional data included information on program status, funded and unpaid program services, hospital visits, short-term skilled nursing facility stays, and medications.

Study Population

The study population included at-risk residents who were referred to and deemed eligible for the CHCPE program. Program eligibility is based on the number of critical needs, income, and total assets. The state defines critical needs as functional dependencies in specific activities of daily living (ADLs) and instrumental activities of daily living (IADLs) and/or cognitive impairment requiring supervision. 23 We included CHCPE participants 65 years and older who completed an annual assessment between January 1, 2005, and December 31, 2010. If a participant had multiple assessments, we used the earliest reassessment for analysis.

Potential Predictors of Long-Term Nursing HomePlacement

Potential predictors included demographic characteristics, clinical characteristics, social support, living arrangements, financial assistance, and program-level variables. Demographic information included age, sex, race, marital status, and primary spoken language. We categorized participants’ race/ethnicity as black, Hispanic, white, or other. Clinical information included medical diagnoses (coded 0 [none], 1 [secondary], or 2 [major]); ADLs and IADLs (coded 0 [independent], 1 [requires assistance], or 2 [total dependence]); mental status quotient (0 to 10 errors); behavioral and psychological issues; vision and hearing assessments; and medications. A dichotomous variable for “meets nursing home level of care” was based on 3 or more critical care needs as defined by the CDSS.23

Program-level variables included healthcare utilization, program services, and the patient’s primary care manager and team. We assessed hospital admissions, emergency department visits, and short-term skilled nursing facility stays during the year before the assessment. We grouped services into categories (eAppendix A, available at www.ajmc.com); identified the services in place at the time of and in the 12 months before the assessment; and calculated average monthly total costs, medical costs, and social service costs. The personal care assistant service pilot offered during the period of our study gave participants authority to hire a single person, including a family member, to perform services that might otherwise be provided by multiple persons. The assessment year, time since the initial assessment, and time since program activation were used to account for variation in subjects’ program participation time.

Most variables had low rates of missingness (ie, less than 2%). We imputed missing values as follows: “no” for dichotomous variables, the most frequent level or category for multichotomous variables, and median values for continuous variables.24 Since missing values for the mental status quotient (3%) were likely attributable to pronounced cognitive or communication impairments, we imputed missing values to 11.

Outcomes

The primary outcome of interest was placement in a long-term nursing home within 180 or 365 days after the assessment. We calculated the days from assessment to nursing home placement based on a termination record in the program status file.

Statistical Analysis

We present patient characteristics at the time of the annual assessment, using proportions for categorical variables and using means with standard deviations or medians with interquartile ranges for continuous variables. We calculated the frequency and Kaplan-Meier estimates of 1-year or 6-month nursing home placement.

Since the goal of this study was to develop a prediction model that would be useful in practice, we carefully preselected potential predictors based on clinical knowledge and previous literature,10,13,15,25,26 and adhered to the rule of 10 events per examined variable to avoid overfitting.27-30 We used logistic regression models to predict 1-year and 6-month nursing home placement. We chose logistic regression to facilitate both internal- and external-model validation using well-established methods and practical implementation of the prediction model scoring mechanism in the electronic data systems of 2 HCBS programs. In all models, we incorporated random effects to account for variance in nursing home placement among primary care-manager teams. As a sensitivity analysis, we used Cox proportional hazards models with robust standard errors to account for clustering of participants within primary care-management teams. For the Kaplan-Meier and Cox survival analyses, we censored data for participants if they terminated the program and at the time of death.

We randomly selected a derivation sample consisting of 66% of the study cohort and a validation sample consisting of the remaining 34% of the study cohort. We developed the logistic regression models in the derivation sample and then applied the results from these models to the validation sample. We evaluated the calibration and discrimination of all models in both samples and refit the models for the entire study cohort.31,32 Using the derivation model estimates, we generated predicted probabilities of nursing home placement in the derivation and validation samples. To assess the clinical usefulness of the prediction models, we considered patients who were in the top 10%, 15%, 25%, or 50% of the predicted probabilities to be those whom the model predicted would have nursing home placement.33 At each threshold, we calculated the model’s sensitivity, false-negative rate, specificity, false-positive rate, and positive predictive value. We calculated the area under the curve (C statistic) to assess overall model discrimination. To assess model calibration, we plotted percent of predicted nursing home placement versus percent of observed nursing home placement by decile of predicted probability and calculated Eavg.30 Additional details about methods are provided in eAppendix B.

RESULTS

Among 10,975 CHCPE participants who completed an annual assessment between 2005 and 2010, the mean time since program activation was 425 days. The mean available follow-up for the study outcome was 320 days (range, 1-365). The median age was 75 to 79 years, 74% of participants were women, and 74% were white (Table 1). More than two-thirds of participants were eligible for Medicaid. Common medical diagnoses included hypertension (59%), diabetes (33%), and Alzheimer's disease (20%). Common functional dependencies included bathing (80%), meal preparation (92%), housework (97%), and money management (75%). Most participants were able to perform eating/feeding (89%) and toileting (86%) activities without assistance. Over half of participants lived alone, and most had regular contact with their support system.

The most common services received were: housekeeping or shopping (95%); skilled nursing care or medication administration (80%); emergency response system (72%); and adult care, companionship, or supervision (62%) (Table 2). In addition to formal services paid for through the program, most participants also received unpaid services typically performed by someone in their support system, including household/shopping, safety checks, and hands-on care. The pilot personal care assistance service was used by 160 (1.5%) participants, for an average of about 90 hours per month. The mean total monthly cost for paid program services was $2296, the mean non-Medicare medical service cost was $714, and the social service cost was $1549.

Within 1 year of assessment, 1249 (11.4%) participants had nursing home placement, 836 (7.6%) died, and 521 (4.7%) terminated the program for other reasons. The Kaplan-Meier 1-year nursing home rate was 12.2%. Among participants with nursing home placement within 1 year, the median time to placement was 170 days.

 
Copyright AJMC 2006-2019 Clinical Care Targeted Communications Group, LLC. All Rights Reserved.
x
Welcome the the new and improved AJMC.com, the premier managed market network. Tell us about yourself so that we can serve you better.
Sign Up