Fall Risk, Supports and Services, and Falls Following a Nursing Home Discharge

Marwa Noureldin, PharmD, MS, PhD; Zachary Hass, MS, PhD; Kathleen Abrahamson, PhD, RN; Greg Arling, PhD


Gerontologist. 2018;58(6):1075-1084. 

In This Article

Design and Methods

Study Sample

The analytic sample included NH residents who were transitioned from the NH to the community by the Minnesota RTCI between April 2014 and October 2016 (N = 1,459). Data came from the comprehensive Community Planning Tool (CPT), completed by CLS prior to discharge for all NH residents who participated in RTCI. The CPT is a comprehensive assessment and includes demographic information, medical diagnoses, health, functional, and cognitive status of the residents, medication use and medication management, discharge location, as well as caregiver availability and frequency of assistance. The CLS personnel use various sources to collect information for the CPT, including Minimum Data Set (MDS) assessments, NH charts, and NH resident and family caregivers. CLS staff conduct follow-up interviews with older adults (in person or by phone) at 3, 10, and 30 days post-discharge. The follow-up assessments provide information regarding fall occurrence and health care utilization.


Study variables were derived from the CPT and follow-up assessments. The outcome variable of interest was occurrence of falls within 30 days of discharge, dichotomously coded (yes/no). Independent variables included age, gender, presence of at least one musculoskeletal condition (arthritis, hip fracture, osteoporosis, etc.), or presence of at least one neurological condition (dementia, stroke, seizures, etc.). Medical diagnoses were collected from MDS assessments and were based on NH records. Additional variables included presence of moderate-to-severe cognitive impairment (moderate-to-severe score ≤12, cognitively intact 13–15), based on the Brief Interview for Mental Status (BIMS; Saliba et al., 2012) and presence of moderate-to-severe depression based on the Patient Health Questionnaire-9 (PHQ-9 score; moderate-to-severe score ≥10, mild to no depression score 0–9; Kroenke, Spitzer, & Williams, 2001). Prior history of NH falls (yes/no) as well as resident concerns at discharge regarding falling in the community (yes/no) and concerns about balance/vertigo affecting daily activities (yes/no) were also included in the analysis. Home environmental safety issues were defined as older adult concern about getting around within at least one of seven areas in the home, including the basement, bathroom, bedroom, kitchen, laundry room, stairs, and entrances/exits (yes/no). Functional variables included three items assessing whether assistance is sometimes needed with ADLs, specifically toileting, walking, and bed mobility (yes/no). Toileting was defined as getting to and on the toilet, adjusting clothes, and cleaning after toilet use. Walking referred to the ability to walk short distances around the house. Bed mobility was defined as sitting up in bed or moving around in bed. Use of medications considered inappropriate or high risk in the elderly was obtained from medication lists provided to the CLS staff by the NH at discharge. Medications were categorized as psychotropics, analgesics, and anticholinergics (American Geriatric Society, 2015). Psychotropics encompassed use of antidepressants, hypnotics or sedatives, and anti-psychotics; analgesics included use of opioid medications; and varying types of medications with known anticholinergic effects that can lead to dizziness comprised the third category. Variables related to various supports and services included assistance with medication management (independent, somewhat dependent, or dependent); older adult use of durable medical equipment (yes/no); receipt of at least one of the following HCBS: skilled nursing, home health, or personal care assistants; discharge location (alone vs with someone else), and caregiver frequency of support (once weekly or less vs daily or several times a week).

Data Analysis

Descriptive statistics provide an overview of RTCI participant characteristics and 30-day post-discharge outcomes. In the SEMs, we tested the relationships between: (a) fall-related risk factors and receipt of supports and services, (b) fall-related risk factors and the occurrence of falls, and (c) moderating effect of supports and services on the relationship between fall-related risk factors and the occurrence of falls. The SEM approach allows us to model latent constructs from observed measures and to examine complex relationships between observed variables and latent constructs simultaneously (Lei & Wu, 2007; Weston & Gore, 2006). Data management, descriptive statistics, and preliminary regression analyses were conducted using SAS version 9.3 (SAS Institute, Cary, NC), whereas factor analysis and SEM were conducted using Mplus version 7.4 (Muthen & Muthen, Los Angeles, CA).

In developing the SEM models, we first conducted bivariate logistic regression analyses to examine associations between variables identified in the study's conceptual framework with the outcome of falls, to provide preliminary assessment of the relationships, and to assist in SEM model specification. Next, exploratory factor analyses (EFAs) and confirmatory factor analyses (CFAs) were used sequentially to estimate individual latent variable models (measurement models). Initially, we hypothesized the presence of two latent constructs, supports and services, and fall-related risk factors. Results of factor analyses indicated that fall-related risk factors were better represented by three constructs instead of one: fall concerns and fall history, ADLs impairment, and use of high-risk medications. Finally, two SEM models were estimated, one without interactions and one with interactions, including the support and services and fall risk constructs as well as the outcome of falls. We tested all direct and indirect effects and interaction terms between each fall-related risk construct and the supports and services construct; however, for parsimony, the final model included only the significant interaction term. Model co-variates for the full SEM model included age (>85 years vs ≤85 years), gender, diagnosis with at least one musculoskeletal condition, diagnosis with at least one neurological condition, depression, and cognitive status. Results from the model without interactions are included in the Supplementary Figure 2.

Supplementary Figure 2.

SEM initial model (no interactions). The three fall risk factor latent variables are correlated (not shown for simplicity). Standardized coefficients presented. χ2(185)=994.23, p<0.01; CFI=0.61; RMSEA=0.055 (90% CI 0.051, 0.058). Significance bolded.** p<0.05, ***p<0.001

SEM model fit is typically assessed based on several indices including the comparative fit index (CFI), the root mean-square error of approximation (RMSEA), and the maximum likelihood χ 2 test (Lei & Wu, 2007; Weston & Gore, 2006). The χ 2 test is a measure of how well the models fit the observed data with a nonsignificant χ 2 indicating good fit; however, it is extremely sensitive to large sample sizes (Weston & Gore, 2006). The CFI is an incremental fit index that measures improvement in fit with values more than 0.9 or 0.95 indicating improved fit. The RMSEA is an index that corrects for model complexity with values less than 0.06 indicating good fit between the hypothesized model and sample data (Lei & Wu, 2007; Weston & Gore, 2006). Model fit indices were used to assess the individual latent variable models and SEM model with no interaction terms. Due to the dichotomous nature of some variables in the SEM model, Mplus uses maximum likelihood to estimate a model with interaction terms. This estimation technique does not provide traditional fit statistics. We compared our SEM models using the receiver operating curve (ROC) to assess which model was the most predictive of falls (closer to 1.0 indicates more predictive accuracy). The research was approved by the Institutional Review Board at Purdue University.