Cluster Analysis of Vulnerable Groups in Acute TBI Rehabilitation

Abstract

Objective

To analyze the complex relationship between various social indicators that contribute to socioeconomic status and healthcare barriers.

Design

Cluster analysis of historical patient data obtained from inpatient visits.

Setting

Inpatient rehabilitation unit in a large, urban university hospital.

Participants

Adult patients receiving acute inpatient care, predominantly for closed head injury.

Interventions

Not applicable

Main outcome measures

We examined the membership of TBI patients in various “vulnerable group” (VG) clusters (e.g., homeless, unemployed, racial/ethnic minority) and characterized the rehabilitation outcomes of the patients (e.g., duration of stay, changes in Functional Independence Measure [FIM] scores between admission to inpatient stay and discharge).

Results

Analysis revealed four major clusters (i.e., Clusters A-D) separated by VG memberships, with distinct durations of stay and FIM gains during their stay. Cluster B, the largest cluster and also consisting of mostly racial/ethnic minorities, had the shortest duration of hospital stay and one of the lowest FIM improvements among the four clusters despite higher FIM scores at admission. In cluster C, also consisting of mostly ethnic minorities with multiple SES vulnerabilities, patients were characterized by low cognitive FIM scores at admission and the longest duration of stay, and they showed good improvement in FIM scores.

Conclusions

Application of clustering techniques to inpatient data identified distinct clusters of patients who may experience differences in their rehabilitation outcome due to their membership in various “at-risk” groups. Results identified patients (i.e., cluster B, with minority patients and Cluster D, with elderly patients) who attain below-average gains in brain injury rehabilitation. Results also suggested that systemic (e.g., duration of stay) or clinical service improvements (e.g., staff’s language skills, ability to offer substance abuse therapy, provide appropriate referrals or liaise with intensive social work services or plan subacute rehabilitation phase) could be beneficial for acute settings. Stronger recruitment, training and retention initiatives for bilingual and multiethnic professionals may also be considered to optimize gains from acute inpatient rehabilitation following traumatic brain injury.

For brain-injury-related diagnoses, prevalence studies suggest increased risk for various non-White populations. Incidence of traumatic brain injury (TBI) is also higher in young adult males. A study of pediatric TBI patients demonstrated higher incidence of TBI in young African-American children under age five. Among health risk factors associated with stroke (e.g., obesity, hypertension, heart disease, cardiovascular disease), Mexican Americans and Black women had higher prevalence of obesity, while Blacks and Pacific Islanders were more likely to have hypertension¹. Consequently, patients from racial/ethnic and linguistic minority backgrounds, hereafter referred to as “Minorities,” are likely to make up the majority population for rehabilitation facilities in inner cities.

Despite the prevalence of diverse populations within rehabilitation settings, there is a service gap (e.g. lower utilization or provision), as well as general unavailability of evidence-based assessment and treatment for Minorities^2–5. For Minorities who overcome barriers to access health care, research still suggests disparities in quality^6,7 (e.g., duration and intensity of care are lower for non-Whites). Minority patients are less likely to engage in successful community participation after injury and are more likely to be “lost to follow-up” compared to Whites⁸.

This brief report presents a cluster analysis of inpatient data to refine our understanding of health care barriers faced by Minorities. We analyze archival demographic and clinical data from an urban public hospital and examine co-occurring membership of TBI patients in various “vulnerable groups” (e.g., homeless, unemployed, linguistic/racial/ethnic minority), using statistically robust clusters obtained from patient-level data. We illustrate defining characteristics of each patient cluster and compare rehabilitation outcomes.

Methods

Data collection was approved by our medical center’s Institutional Review Board. We used patient-level data (N = 148) collected through retrospective chart review. Inclusion and exclusion criteria followed TBIMS National Database⁹ guidelines, available online. Binary information denoting membership in 11 vulnerable groups were used as the basis of cluster analysis: homelessness, past incarceration, undocumented status, elderly, minority status, non-English speaking, psychiatric history, history of mental institute stays, history of severe medical problems, substance abuse history, and other. A data-driven, two-step cluster approach with log-linear distances (capped at 8 clusters) was used¹⁰. K-means approach was adopted for the 4-cluster solution¹¹. Fit of alternative solutions (e.g., furthest neighbor, 2–4 cluster solutions; resorted and dropped variables) were compared using Akaike Information Criterion obtained from the Two-Step Cluster outputs. Demographic and psychosocial information; length of hospital stay (acute only); FIM score change from admission to discharge were compared using appropriate parametric or nonparametric statistics. A data point indicating a length of stay greater than 5 months was excluded from relevant group-wise statistics with no effect on clustering analysis.

Results

Vulnerable group membership was defined along eleven categories, in line with our group’s past work in this domain¹². Prevalence of vulnerabilities across these groups were as follows: substance abuse history (35.8%), psychiatric history (20.9%), medical history (19.6%), history of psychiatric institutionalization (8.8%), English as a second language (ESL)/non-English speaking (18.2%), racial/ethnic minority (58.1%), elderly (27%), undocumented status (8.8%), history of incarceration (12.2%), history of homelessness (17.6%) and other vulnerabilities (2.7%). Two of the four patients in the “other” vulnerability group sustained injuries from being assaulted with a blunt instrument; one was a homeless male with complex medical and psychiatric history admitted after a fall, the other was a non-English speaking young male admitted after a biking accident with unknown immigration status.

Clustering analysis revealed four main clusters that were stable across 2-step and k-means approaches. Final cluster sizes and characteristics (i.e., cluster centers) are provided in Table 1. Significant group differences were found for the 4-cluster solution (all p’s < .05) in all variables except for “other vulnerability.” Largest F-values were obtained for substance abuse history (F=76.71), medical history (F=45.73), elderly (F=38.25), psychiatric history (F = 33.12), homelessness (F = 27.09), and minority status (F = 18.82), where all p’s < .001. Given the small size of undocumented status and other vulnerability groups, we repeated the analysis with only nine vulnerability groups and still obtained a four cluster solution that highlights cluster differences in terms of minority group membership.

Table 1.

Cluster analysis results and outcomes by cluster

Above half shows cluster centers for each input variable that was used for the clustering analysis. Top defining variables for each cluster are shaded in dark gray. Bottom half shows the breakdown of rehabilitation outcomes for each cluster.

		Cluster Name
		A	B	C	D
INPUT variables
	ethnic/racial minority	.25	.72	.95	.24
	significant medical hx	.04	.01	.45	.72
	elderly	.07	.17	.10	.92
	homelessness hx	.29	.05	.70	.00
	substance abuse hx	.82	.11	1.00	.08
	incarceration hx	.18	.05	.30	.12
	ESL/Non-English Sp	.07	.21	.40	.04
	institutionalization hx	.18	.00	.30	.08
	psychiatric hx	.71	.03	.10	.28
	undocumented	.00	.12	.20	.00
	other	.04	.04	.00	.00
ANALYSIS variables
	Duration of Stay (days)*	21.39	15.95	27.00	18.24
	FIM gains (Total)	39.61	27.86	34.74	26.40
	FIM gains (Cog)*	10.9286	7.7571	9.3158	5.2400
	FIM gains (Motor)	28.6786	21.3784	25.4211	21.1600
	Discharge Location (χ2 = 12.66, p = .049, n = 148)
	Home	32.1%	56.0%	35.0%	36.0%
	Acute / Subacute Care	53.6%	42.7%	55.0%	60.0%
	Other	14.3%	1.3%	10.0%	4.0%
	Primary person at d/c (χ2 = 8.10, p = .231, n = 148)
	Self / Alone / Other		6.6%	5.0%	4.0%
	Family / Friends	32.1%	49.3%	30.0%	32.0%
	Facility Care	67.9%	44.0%	65.0%	64.0%

^*indicates significant group differences between clusters based on omnibus ANOVAs, p < .05.

The clusters in the main analysis were characterized as follows: Cluster A – higher likelihood of problematic substance use and psychiatric history; Cluster B – racial and ethnic minority group membership and somewhat increased incidence of ESL; Cluster C – substance use, incarceration, homelessness history in predominantly minority and ESL group; Cluster D – elderly patients with complex medical comorbidities (see Figure 1 for relative incidences of membership in each cluster). Clusters B (n = 75) and C (n = 20), both defined by ethnic/minority membership, comprised more than half the sample (N = 148). Patients with psychiatric histories were more likely to be in Clusters A and C; those with significant medical histories were more likely to be in Clusters C and D. Age was significantly higher for Cluster D (M = 76), compared to the other clusters, where average age per cluster ranged from 43 to 49.

Figure 1. Clusters of VG Membership.

Radial axes represent the percent of patients belonging to each vulnerability group within each cluster. Different shaded areas connect the percentages that define each cluster. For example, cluster C is shown in light blue, characterized by the high incidence of minority status, substance abuse history and minority status in the cluster.

Analyses of rehabilitation outcomes revealed significant differences between clusters in duration of stay (p = .012) and cognitive FIM score change from admission to discharge, (p = .014). Pairwise comparisons, using least-significance difference to correct for multiple comparisons, indicated statistically significant differences between Clusters B and C in duration of stay; and differences in FIM cognitive gains between Clusters A vs. D, A vs. B and C vs. D. Though not statistically significant, large differences in FIM total score and FIM motor score change were also observed between Clusters A and B in pairwise comparisons.

Chi-squared tests on discharge disposition, χ² (6, N = 148) = 12.657, p = .049, and primary person at discharge, χ² (6, N = 148) = 8.103, p = .231, showed large but statistically insignificant group differences between clusters when using a more stringent cut-off due to multiple comparisons (α = .01).

Discussion

Using cluster analysis to explore the relationship between inpatient rehabilitation outcomes and binary socioeconomic categories, health histories, and demographics, we found complex, multivariate associations between vulnerability factors that impact acute TBI rehabilitation outcomes. Results demonstrated four distinct, relatively stable clusters. Cluster A contained primarily White patients with significant substance abuse and psychiatric histories. Cluster B, the largest cluster, was marked by patients who belonged to a racial/ethnic minority group, with often limited English proficiency. Cluster C was comprised of minority patients with substance abuse, incarceration, and homelessness histories. Cluster D was marked by elderly patients with complex medical comorbidities, and the cluster’s average FIM gains suggest that their functional gains may not be fully optimized during the acute stay window and are likely to continue during their prolonged stay at another acute/subacute facility. Duration of inpatient stay was shortest for patients in Cluster B, who were also likely to make smaller total and cognitive FIM gains, and more likely to be discharged home than any other cluster. Cluster A and C both appeared to respond well to acute rehabilitation with higher FIM gains, which may, in part, also explain the longer duration of stay observed for these groups on average.

Study Limitations

Our study has two main limitations. First, our diverse, urban data from one of the most ethnically diverse metropolitan regions in the U.S., may not be entirely representative of different U.S. regions that are more homogenous demographically. Second, cluster analyses’ micro-level results are often sensitive to type, number and definition of input variables used. That said, our overall results appear consistent with our group’s earlier SES analyses and identifies a complex, cautionary relationship between clusters of vulnerable groups and rehabilitation outcomes that can be replicated by other sites. Indeed, simplistic univariate analyses has produced counter-intuitive or null results by washing-out significant outcome differences of latent groups¹³. In contrast, we challenge the notion of equity in terms of access to services and outcomes in the acute rehabilitation landscape.

Conclusions

Our findings suggest that hospital policy changes (e.g., duration of stay) or improvements in service provision may be needed to address disparities in rehabilitation services rooted in complex SES relationships. Programs that augment staff’s language skills, ability to offer substance abuse therapy, systematically provide appropriate referrals or liaise with intensive social work services may be beneficial in addressing the identified gap in service provision to minority patients¹⁴. Similarly, increased recruitment, training, and retention options for bilingual and multiethnic professionals may be considered to optimize gains from acute inpatient TBI rehabilitation.

Highlight.

• Simultaneous co-occurrence (i.e., clustering) of some demographic and clinical variables may affect the acute rehabilitation outcomes of patients with TBI, rendering ethic/racial minority and elderly groups particularly vulnerable to shorter stays despite small gains in functional variables.

Abbreviations

TBI

Traumatic Brain Injury

FIM

Functional Independence Measure

SES

Socio-Economic Status

VG

Vulnerable Group

TBIMS

Traumatic Brain Injury Model Systems

ESL

English as Second Language

hx

history

MCO

Managed Care Organization

Cog

Cognitive