Profiles of Quality of Life Outcomes Observed in Rehabilitation Patients Using Latent Profile Analysis

Abstract

Objective:

To explore patterns of quality-of-life (QOL) outcomes in a diverse group of patients commonly seen in rehabilitation settings, in an effort to understand shared profiles of experiences that cut across diagnostic groupings.

Design:

Cross-sectional study.

Setting:

Academic research center.

Participants:

English-speaking adults with medically confirmed history of traumatic brain injury, spinal cord injury, stroke, or major limb injury and/or limb loss due to sudden-onset cause, recruited from three clinical sites in the U.S.

Interventions:

n/a

Main Outcome Measure:

Standardized factor scores computed for nine outcome cluster factors scores based on patient-reported outcome measures of economic QOL, fatigue/sleep, pain, independence, physical function, negative affect, positive affect, and social health, and cognitive measures from the Brief Test of Adult Cognition by Telephone.

Results:

Optimal model fit for the latent profile analysis was achieved with a four-profile model. These profiles were found to cut across clinical condition, and were distinguished by a number of outcomes beyond symptom severity. Pain, independence, cognition and social participation varied in important ways across the profiles.

Conclusion:

This study supports the use of outcome cluster profiles as part of rehabilitation treatment, in addition to information about one’s clinical condition when conceptualizing patient needs. The findings reported here may be useful for developing distinct rehabilitation management approaches based on these outcome cluster profiles.

Introduction

Rehabilitation providers evaluate and treat individuals recovering from a variety of sudden-onset, debilitating conditions that can affect all bodily systems. Ideally, providers will have a full understanding of potential symptoms involving the nervous system—including motor and sensory functions, cognitive performance, and mental/emotional health—and common co-occurring symptoms that affect, for example, musculoskeletal systems, skin integrity, and genitourinary function.^{1, 2} It can be useful in rehabilitation settings to group patients and conceptualize treatment plans by diagnostic categories of presenting condition, such as compartmentalizing patients into separate stroke,^{3, 4} traumatic brain injury (TBI),^5-7 spinal cord injury (SCI),^{8, 9} and orthopedic injury/amputation^{10, 11} treatment tracks. However, there are meaningful similarities in symptoms and outcomes—particularly in the post-acute phase—that cut across these groups, including psychosocial challenges, cognitive changes, secondary medical and mental health conditions, need for environmental supports, financial burdens, and other barriers to optimal quality of life [QOL].^{2, 12-29} Understanding these commonalities can be useful for diagnostic conceptualization and treatment planning (as proposed by Wong, et al. [2017]³⁰). Indeed, there have been efforts to generate clinical practice guidelines that are applicable across rehabilitation populations, like those by Hornby, et al. [2020]³¹ and Jolliffe, et al. [2018]³².

Clustering patients by symptoms is an area ripe for further research, as identified and encouraged recently by several NIH-funded symptom scientists.^{33, 34} Symptom clustering has been popular in oncology research for decades,^35-39 and this perspective has recently been applied to rehabilitation populations.⁴⁰ To understand symptoms relevant across rehabilitation populations, latent variable modeling statistical approaches can identify meaningful shared profiles of individuals that may not otherwise be apparent.³⁴ Detecting outcome cluster profiles in individuals with disabilities can be useful for identifying modifiable factors and tailoring treatments.⁴¹

Statistical clustering approaches (e.g., latent class analysis, latent profile analysis [LPA]) have been most often used to define profiles of patients that vary by symptom severity. For example, for distinct severity subgroups of children with TBI⁴² and for individuals following stroke, severity categories have been defined for psychoneurological symptoms,⁴³ depression,⁴⁴ anxiety,⁴⁵ degree of interoceptive sensibility,⁴⁶ and overall symptom burden.⁴⁷ For persons with SCI, clustering research has identified levels of functional independence and ability level,^{48, 49} extent of injury and medical outcomes,⁵⁰ profiles of degree of psychological adaptation and mental health needs at time of discharge,^{51, 52} and severity categories of activity of daily living (ADL) impairment and upper-limb dysfunction.⁵³

There is also a large body of symptom clusters research within specific clinical conditions to understand the interplay of symptoms from different domains. For example, in the TBI literature, multiple studies have defined phenotypes of neurobehavioral functioning and neuropsychiatric symptoms,^54-56 and profiles of neurological, cognitive, behavioral, somatic, and systemic symptoms.^{57, 58} This includes a study by Sherer, et al. [2017]⁵⁹—which was cross validated by Sherer, et al. [2017]⁶⁰ and replicated by Ponsford, et al. [2025]⁶¹—that identified five distinct patterns of TBI outcomes on the dimensions of cognitive, emotional, and physical functioning, environmental supports, and performance validity, using several modern patient-reported outcome measures (PROMs).⁵⁹ In the stroke literature, statistical clustering approaches have been used to define profiles of cognitive decline/dementia and symptoms of depression,^{62, 63} and to define distinct mental health and somatic symptom cluster profiles.^{64, 65}

Some symptom science research in rehabilitation populations has focused on understanding a specific set of symptoms or behaviors in depth. For example, SCI researchers have defined classes of patients by urinary tract dysfunction management,⁶⁶ profiles of cognitive functioning,⁶⁷ classes of functional trajectories,⁶⁸ and classes of pain trajectories.⁶⁹ In the musculoskeletal injuries literature, cluster analyses have defined profiles of medication usage after combat-related amputation⁷⁰ and coping strategy profiles for people with limb loss.⁷¹ Less frequently, statistical clustering techniques have defined symptoms that cut across populations. Such studies include the degree of ADL limitation in 8,000 elderly individuals with various diagnoses from the Chinese Longitudinal Health Longevity Survey,⁷² patterns of alcohol use after TBI and SCI,⁷³ and clusters of physical, cognitive, and/or emotional symptoms in individuals with combat-related polytrauma.^74-77

The objective of the present study was to empirically evaluate a large dataset composed of health-related quality of life (HRQOL) and other symptom-focused outcome measures gathered from individuals belonging to four groups commonly seen in rehabilitation settings: SCI, TBI, stroke, and sudden-onset, moderate to severe major limb illness/injury. We adhere to the tripartite model of health as described by the WHO,⁷⁸ as including (1) physical/medical health and symptoms, (2) mental health/psychological symptoms and, and (3) social health and participation. Under this model, HRQOL is more than just life satisfaction, but can instead be quantified into different subdomains that are individually relevant for observation and intervention. We hypothesized that, by using an LPA approach on these HRQOL outcome variables, profiles of patients that vary on important outcomes would be detected, independent of diagnostic groupings; these profiles may be useful for understanding the common experiences and shared treatment needs of these individuals. Associations between these profiles and other outcomes were explored using mean household income, functional status, social participation, and general health ratings of the participants assigned to each profile.

Methods

Participants were 755 individuals who had experienced a sudden-onset disability due to a medically confirmed diagnosis of TBI, major limb injury and/or limb loss, SCI, or stroke. Participants were community-dwelling, English-speaking adults recruited from three clinical centers in the U.S., who provided informed consent prior to participation. The study was approved by the University of Delware IRB. Participants completed a 2- to 3-hour telephone interview with a trained research assistant, which consisted of a demographic and medical history survey, PROMs, and performance-based cognitive tests. The sampling and study methodologies—including information on recruitment, eligibility, data collection procedures, and participant flow—were described previously by [Tulsky, et al. [2026]⁴⁰ (see Supplemental Material for a CONSORT diagram).

Indicator variables for the LPA were standardized factor scores computed for nine outcome cluster factors scores (based on a factor analysis of outcome measures as described by [Tulsky, et al. [2026]⁴⁰): Cognition, Economic QOL, Fatigue/Sleep, Pain, Independence, Physical Function, Negative Affect, Positive Affect, and Social Health (Table 1 contains mean values by diagnosis and details on the component measures from the factor analysis; additional details on the CFA are in the Supplemental Material). LPA was conducted in R (version 4.2.2) with the mclust package (version 6.0.0).⁷⁹ A model comparison approach was used to select an optimal profile classification scheme, such that the selected model provided optimal model-data fit among alternatives and exhibited well-defined, clinically interpretable profiles. Furthermore, of central interest was whether the chosen solution was driven by injury type or outcome cluster presentation that cut across injury subgroups. To select an optimal LPA model, we compared models with differing numbers of latent profiles (between 1 and 8) and different specifications of the variance-covariance matrix for the 9 LPA indicators (the mclust package permits 14 different specifications). Indicator means for the 9 dimensions were freely estimated in each profile for all models. Once estimated, model fit was compared using two information criteria—the Bayesian information criterion (BIC)⁸⁰ and integrated complete-data likelihood (ICL)⁷⁹—along with the bootstrap likelihood ratio test (BLRT)⁸¹ and entropy statistic.⁸² Maximal BIC/ICL values and entropy values greater than .80 were indicative of optimal solution(s).⁸³

Table 1.

Outcome Cluster Indices – Mean (SD) by Diagnosis

Index	Component Measure	TBI (n=185)	Limb Injury (n=188)	SCI (n=190)	Stroke (n=192)
Cognition *	Category Fluency	−0.33 (1.03)	0.00 (0.79)	0.05 (0.86)	−0.61 (0.82)
	Digit Span Backward Immediate Recall
Economic QOL	ECON-QOL	−0.44 (0.92)	−0.01 (0.95)	−0.27 (1.03)	−0.04 (0.93)
	Financial Cutbacks † Material Needs †
Fatigue/Sleep	PROMIS Fatigue	−0.01 (1.01)	−0.52 (0.95)	−0.38 (1.04)	−0.12 (1.05)
	Fatigue Severity Scale PROMIS Sleep Impairment
Pain	PROMIS Pain Interference	−0.11 (1.17)	−0.06 (0.95)	0.15 (0.95)	−0.09 (1.14)
	PROIMS Pain Intensity PROMIS Nociceptive Pain
Independence	TBI-QOL/SCI-QOL Independence	−0.16 (1.08)	−0.01 (1.00)	−0.72 (1.29)	−0.40 (1.16)
Physical Function‡	PROMIS Physical Function	0.70 (1.15)	0.00 (1.00)	−1.03 (1.22)	−0.33 (1.05)
Negative Affect	PROMIS Anger	−0.17 (1.08)	0.00 (0.95)	−0.13 (0.99)	0.09 (0.96)
	PROMIS Anxiety PROMIS Depression PROMIS Social Isolation
Positive Affect	TBI-QOL/SCI-QOL/LIMB-QOL Resilience	−0.31 (1.02)	−0.13 (0.97)	0.01 (0.88)	0.17 (0.82)
	Neuro-QoL Positive Affect and Well-Being
Social Health	Neuro-QoL Ability to Participate in SRA	0.16 (1.02)	0.00 (0.97)	−0.17 (1.00)	0.08 (0.92)
	Neuro-QoL Satisfaction with SRA

Notes: All scores represent factor scores which are on a standardized z metric, computed using an Empirical Bayes Modal approach.

^*From the Brief Test of Adult Cognition by Telephone.

^†From the Conger Economic Pressure Scale.

^‡The conceptualization is different than the Tulsky, et al. [2026]¹ factor analysis, as a lack of invariance was found across conditions for a Physical Function factor indicated by two measures in the factor analysis study; as such, we used a custom 8-item short form derived from the PROMIS Physical Function item bank that assessed global physical function (a copy of this form can be made available upon request). A tenth factor from the Tulsky, et al. [2026]¹ factor analysis, Psychological Adjustment, was excluded from these analyses due to collinearity with the other Emotional Health indicators (r = .9 with Negative Affect and rs = .8 with Positive Affect and Social Health). Factor scores were estimated under a partial scalar-invariant model as described in Tulsky, et al. [2026]¹ which provides full details on the development of these nine factors. Factor scores were available for the full sample (N = 755), therefore, missing data techniques were not required. QOL = quality of life; TBI = traumatic brain injury; SCI = spinal cord injury; PROMIS^® = Patient-Reported Outcomes Measurement Information System; TBI-QOL = Traumatic Brain Injury-Quality of Life; SCI-QOL = Spinal Cord Injury-Quality of Life; Neuro-QoL^™ = Quality of Life in Neurological Disorders; LIMB-QOL = Limb Injury Measurement Battery for Quality of Life; SRA = social roles and activities.

The best solution in terms of model fit and interpretability was further analyzed by comparing five variables across profiles via analysis of variance (ANOVA; p < .05) and post-hoc pairwise comparisons (Tukey’s honestly significant difference [HSD]⁸⁴). These variables were chosen to better understand the experiences of the individuals in each profile: (a) participant age, (b) household income, (c) general overall health rating, (d) Katz ADL scale score,⁸⁵ and (e) Participation Assessment with Recombined Tools-Objective (PART-O) Out and About subscale score.^{86, 87}

Results

Participants

Table 2 presents demographic and injury characteristics for the sample. Age ranged from 19 to 90, with mean of 50.8 years. Average age was highest among individuals with stroke (64.1). Participants were, on average, 9.9 years removed from their injury, with three of the four injury subgroup averages between seven and 10 years; participants with SCI were further removed from injury, on average (14.8 years). Participant gender was split evenly within the stroke subgroup, whereas, as expected, most participants were male (70-80%) in the other three conditions.

Table 2.

Demographic and Injury Characteristics for Sample and by Diagnosis (n = 755)

	TBI (n = 185)	Limb Injury (n = 188)	SCI (n = 190)	Stroke (n = 192)	Overall (N = 755)
	M (SD)	M (SD)	M (SD)	M (SD)	M (SD)
Age	43.8 (15.0)	45.0 (13.5)	50.0 (15.7)	64.0 (12.0)	50.8 (16.2)
Time Since Injury (Years)	8.6 (5.3)	9.2 (8.0)	14.8 (12.0)	7.0 (4.1)	9.9 (8.5)
	n (%)	n (%)	n (%)	n (%)	n (%)
Gender
Male	130 (71)	147 (78)	152 (80)	94 (49)	523 (69)
Female	53 (29)	41 (22)	38 (20)	98 (51)	230 (31)
Not Provided	2 (1)	0 (0)	0 (0)	0 (0)	2 (0)
Race
White	126 (68)	140 (74)	133 (70)	122 (64)	521 (69)
Black	32 (17)	26 (14)	39 (21)	51 (27)	148 (20)
Asian	7 (4)	0 (0)	5 (3)	7 (4)	19 (3)
American Indian / Alaskan Native	0 (0)	0 (0)	0 (0)	1 (1)	1 (0)
Native Hawaiian / Pacific Islander	0 (0)	1 (1)	0 (0.0)	0 (0.0)	1 (0)
Multiracial	3 (2)	8 (4)	5 (3)	6 (3)	22 (3)
Other	17 (9)	12 (7)	8 (4)	5 (3)	42 (6)
Not Provided	0 (0)	1 (1)	0 (0)	0 (0)	1 (0)
Ethnicity
Non-Hispanic	148 (80)	148 (79)	160 (84)	185 (96)	641 (85)
Hispanic	35 (19)	37 (20)	30 (16)	6 (3)	108 (14)
Not Provided	2 (1)	3 (2)	0 (0)	1 (1)	6 (1)

LPA Results

Model Selection

Table 3 presents correlations among LPA indicators. Table 4 contains model selection fit information. Optimal BIC and ICL fit values were found for a 4-profile solution, which also exhibited the largest entropy estimate (.84) among differing numbers of profiles. The BLRT test statistic also supported the 4-profile solution, which was significantly different from zero when increasing the number of profiles from 2 to 4, but not significantly different from zero when a 5th profile was added. Initial review also supported clinical interpretability of the 4 profiles; therefore, the 4-profile solution was selected for further study.

Table 3.

Factor Score Correlations

	Cognition	Economic QOL	Fatigue/Sleep (Reversed) *	Pain (Reversed) *	Independence	Physical Function	Negative Affect (Reversed) *	Positive Affect	Social Health
Cognition
Economic QOL	0.29
Fatigue/Sleep (Reversed) *	0.15	0.40
Pain (Reversed) *	0.12	0.30	0.53
Independence	0.25	0.37	0.42	0.31
Physical Function	0.13	0.21	0.26	0.37	0.61
Negative Affect (Reversed) *	0.08	0.51	0.73	0.52	0.55	0.23
Positive Affect	0.00	0.39	0.64	0.36	0.53	0.25	0.84
Social Health	0.15	0.47	0.71	0.55	0.74	0.53	0.80	0.79

Note:

^{* =}Factor scoring direction was reversed to facilitate interpretation

Table 4.

LPA Model Selection Indices

Profiles	Parameters	BIC	ICL	BLRT	BLRT p	Entropy
1	54	−15629.5	−15629.5
2	73	−15406.5	−15609.6	348.9	0.00	0.63
3	92	−15361.6	−15586.0	170.8	0.00	0.72
4	111	−15212.9	−15371.0	274.6	0.00	0.84
5	130	−15400.3	−15702.0	−61.5	0.97	0.74
6	149	−15399.2	−15718.7	127.1	0.00	0.75
7	168	−15476.8	−15787.1	48.3	0.10	0.77
8	187	−15560.5	−15886.6	42.2	0.27	0.77

Notes: BIC = Bayesian information criterion; ICL = integrated complete-data likelihood; BLRT = bootstrap likelihood ratio test. BLRT values compare the fit of a model with m profiles to a model with m − 1 profiles. The best-fitting 4-profile solution is highlighted in bold. Fit information is provided for the VVE variance-covariance matrix specification in the mclust package, which fit better than all alternative specifications allowed by the software. The VVE specification provides a balanced trade-off between simpler as well as more flexible specifications of the variance-covariance matrix within each profile.

Profile Interpretation

Table 5 presents the number of participants per profile and prevalence by condition. All 4 conditions were represented in each profile, indicating that classification was largely driven by outcome cluster presentation rather than condition. As noted above, entropy for the 4-profile solution was .84, indicating acceptable clarity of profile classification. Visualization of the profiles across all indicators is shown in Figure 1; Table 6 provides indicator mean and SD estimates within profiles. The profiles are labeled and described below.

Table 5.

Frequency Counts for 4-Profile Solution Overall and by Diagnosis

	TBI		Limb Injury		SCI		Stroke		Overall
	n	(%)	n	(%)	n	(%)	n	(%)	n	(%)
Profile 1	51	(52)	25	(25)	5	(5)	18	(18)	99	(13)
Profile 2	14	(24)	13	(22)	13	(22)	19	(32)	59	(8)
Profile 3	30	(12)	95	(39)	62	(25)	57	(23)	244	(32)
Profile 4	90	(25)	110	(16)	110	(31)	98	(28)	353	(47)

Notes: Percentages sum across conditions (row-wise) in the first four columns, and across profiles in the last column (column-wise). Profile assignment was based on each individual’s highest posterior probability estimate.

Figure 1. Profile Plot for 4-Profile Solution.

Means and +/− 1SD bands are shown for each profile across the 9 LPA indicators. Scoring directions for Fatigue/Sleep, Pain, and Negative Affect have been reversed (Rev); therefore, higher scores across all indicators reflect better QOL/fewer symptoms, and lower scores represent worse QOL/more symptoms.

Table 6.

Outcome Cluster Indices – Mean (SD) by Profile

	Profile 1 n = 99	Profile 2 n = 59	Profile 3 n = 244	Profile 4 n = 353
Cognition	0.15 (0.74)	−0.29 (0.74)	−0.02 (0.72)	−0.45 (0.99)
Economic QOL	0.36 (0.62)	0.33 (0.69)	0.18 (0.62)	−0.68 (1.03)
Fatigue/Sleep (Reversed) *	0.80 (0.82)	1.10 (0.61)	−0.03 (0.78)	−0.44 (1.11)
Pain (Reversed) *	0.68 (0.87)	0.58 (0.80)	−0.42 (0.81)	−0.54 (1.07)
Independence	0.72 (0.51)	0.63 (0.65)	0.29 (0.57)	−1.19 (0.98)
Physical Function	1.83 (0.24)	0.34 (0.79)	−0.22 (0.57)	−0.77 (1.30)
Negative Affect (Reversed) *	0.38 (0.83)	0.98 (0.44)	0.10 (0.73)	−0.46 (0.95)
Positive Affect	0.37 (0.84)	1.16 (0.57)	0.06 (0.76)	−0.46 (1.02)
Social Health	0.97 (0.68)	1.67 (0.37)	0.07 (0.57)	−0.56 (0.73)

Notes:

^*= Test scores were inverted for analyses to facilitate profile interpretation. Profile counts based on most likely class membership.

Profile 1.

Profile 1 was marked by above average QOL/function on most indicators and absence of self-reported physical limitations. Specifically, the within-profile mean estimate was close to 2 SDs above the overall sample for Physical Function (at or near the ceiling on the component measures). Individuals in this profile also showed above average emotional well-being (Positive and Negative Affect indicators each approaching +.5 SD) and Social Health. This profile’s lowest score was in Cognition, which was close to the overall mean. This profile captured 13% of the overall sample, with 52% originating from the TBI subgroup.

Profile 2.

Profile 2 exhibited the highest mean levels of emotional well-being (Positive and Negative Affect ≥ +1 SD) and Social Health (≥ +1 SD). Participants exhibited mean estimates that were above average across all indicators except Cognition, which was slightly below the overall overage. This profile is similar to Profile 1, but with higher mean levels of psychosocial health, higher mean ratings of Fatigue/Sleep, and a lower mean level of Physical Function (although still approximately +.5 SD above the overall sample average). Profile 2 was the smallest profile, containing only 8% of participants. A plurality of individuals in this profile were from the stroke subgroup (32%).

Profile 3.

Profile 3 demonstrated near average levels (within .5 SD) on all LPA indicators. Mean estimates were close to the overall group average, and this profile did not differ from the high QOL profiles (Profiles 1 and 2) on Cognition, Economic QOL, or Independence (see Table 7 for profile comparisons). This profile contained 32% of overall sample, with 39% from the limb injury subgroup.

Table 7.

Pairwise Comparisons Across Profiles (Tukey’s HSD)

Criterion Variable	Profile (I)	Profile (J)	Mean Difference (I – J)	p	95% Confidence Interval
					Lower Bound	Upper Bound
Age (range: 19-90)	1	2	−6.29	0.08	−13.11	0.54
		3	−6.13	0.01	−11.07	−1.18
		4	−5.70	0.01	−10.42	−0.98
	2	3	0.16	1.00	−5.86	6.18
		4	0.58	0.99	−5.25	6.42
	3	4	0.43	0.99	−3.03	3.88
Income * (range: 1-7)	1	2	0.41	0.43	−0.28	1.10
		3	0.16	0.82	−0.32	0.64
		4	0.92	0.00	0.45	1.39
	2	3	−0.25	0.73	−0.86	0.37
		4	0.51	0.13	−0.09	1.12
	3	4	0.76	0.00	0.42	1.11
Katz ADL (range: 0-6)	1	2	0.18	0.88	−0.44	0.80
		3	0.24	0.52	−0.21	0.69
		4	1.56	0.00	1.13	1.99
	2	3	0.06	0.99	−0.49	0.60
		4	1.38	0.00	0.85	1.91
	3	4	1.32	0.00	1.01	1.63
Part-O, Out and About (range: 0-26)	1	2	0.12	0.69	−0.16	0.40
		3	0.13	0.34	−0.07	0.34
		4	0.57	0.00	0.37	0.77
	2	3	0.01	1.00	−0.24	0.26
		4	0.45	0.00	0.21	0.69
	3	4	0.44	0.00	0.29	0.58
General Health † (range: 1-5)	1	2	−0.01	1.00	−0.38	0.37
		3	−0.54	0.00	−0.81	−0.26
		4	−0.90	0.00	−1.16	−0.64
	2	3	−0.53	0.00	−0.86	−0.20
		4	−0.89	0.00	−1.21	−0.57
	3	4	−0.36	0.00	−0.55	−0.17

Notes: Pairwise comparison p-values < .05 are highlighted in bold.

^*=Income queried as “What was your total household income (income from all sources, including child support, alimony, disability, SSI, unemployment) before taxes, last year?” and responses were coded as: Less than $5,000 = 1; $5,000 to $9,999 = 2; $10,000 to $19,999 = 3; $20,000 to $39,999 = 4; $40,000 to $74,999 = 5; $75,000 to $99,000 = 6; $100,000 or more = 7.

^†= General Health was queried as “In general, would you say your health is…?” with the responses coded as: Excellent = 1; Very good = 2; Good = 3; Fair = 4; Poor = 5.

Profile 4.

Mean estimates within this profile were below overall sample averages for all indicators, typically by .5 SD or more (See Figure 1 and Table 6), and were the lowest observed for any profile. The mean Cognition estimate for this profile was similar to Profile 2, and the mean estimate for Pain was similar to Profile 3. The lowest mean estimates for this profile were on Independence and Physical Functioning (< −.75 SD). This was the largest profile, containing 47% of participants. The highest proportion of individuals in this profile were from the SCI subgroup (31%).

Profile Comparisons

Table 7 shows pairwise mean differences between profiles and Tukey HSD values for the 5 variables. The results support the interpretability of the profiles, such that profiles reporting fewer impairments and greater QOL (Profiles 1 and 2, and to a lesser extent, Profile 3) reported significantly higher levels of household income, functional status, social participation, and general health. Specifically, (a) Profiles 1 and 3 reported significantly higher household income than Profile 4, (b) Profiles 1, 2, 3 reported higher levels of functional status and social participation on the Katz ADL and PART-O subscale, respectively, than Profile 4, and (c) Profiles 1 and 2 reported better general health than Profiles 3 and 4 (Profiles 1 and 2 did not differ with regard to general health). Moreover, Profile 3 reported higher general health than Profile 4. Regarding age, Profile 1 was younger than Profiles 2 (p = .08), 3, and 4—likely owing to the larger proportions of individuals with TBI or limb injury (the youngest injury subtypes in the sample)—and was characterized by the highest levels of physical function—which is also likely attributable to the younger age of Profile 1. Table 8 presents additional demographic breakdowns of each profile (e.g., gender, ethnicity).

Table 8.

Demographic and Breakdown by Profile (n = 755)

	Profile 1 n = 99	Profile 2 n = 59	Profile 3 n = 244	Profile 4 n = 353
	M (SD)	M (SD)	M (SD)	M (SD)
Time Since Injury (Years)	8.3 (5.4)	10.6 (8.6)	10.7 (9.1)	9.8 (8.8)
	n (%)	n (%)	n (%)	n (%)
Gender
Male	78 (78.8)	43 (72.9)	175 (71.7)	227 (64.3)
Female	20 (20.2)	16 (27.1)	69 (28.3)	125 (35.4)
Not Provided	0 (0.0)	0 (0.0)	0 (0.0)	1 (0.3)
Race
White	67 (67.7)	36 (61.0)	181 (74.2)	237 (67.1)
Black	18 (18.2)	15 (25.4)	41 (16.8)	74 (21.0)
Asian	4 (4.0)	0 (0.0)	7 (2.9)	8 (2.3)
American Indian / Alaskan Native	0 (0.0)	0 (0.0)	0 (0.0)	1 (0.3)
Native Hawaiian / Pacific Islander	0 (0.0)	0 (0.0)	1 (0.4)	0 (0.0)
Multiracial	2 (2.0)	2 (3.4)	8 (3.3)	10 (2.8)
Other	8 (8.1)	5 (8.5)	6 (2.5)	23 (6.5)
Not Provided	0 (0.0)	1 (1.7)	0 (0.0)	0 (0.0)
Ethnicity
Non-Hispanic	83 (83.8)	48 (81.4)	216 (88.5)	294 (83.3)
Hispanic	16 (16.2)	9 (15.3)	26 (10.7)	57 (16.1)
Not Provided	0 (0.0)	2 (3.4)	2 (0.8)	2 (0.6)
Educational Attainment
8th Grade or Less	1 (1.0)	0 (0.0)	1 (0.4)	1 (0.3)
Some High School	0 (0.0)	2 (3.4)	5 (2.0)	12 (3.4)
Completed High School	18 (18.2)	14 (23.7)	46 (18.9)	91 (25.8)
Some College	38 (38.4)	17 (28.8)	67 (27.5)	113 (32.0)
4-Year Degree	21 (21.2)	16 (27.1)	72 (29.5)	81 (22.9)
Some Graduate School	1 (1.0)	1 (1.7)	6 (2.5)	4 (1.1)
Graduate/Professional Degree	0 (0.0)	2 (3.4)	5 (2.0)	12 (3.4)

Discussion

This empirically derived profile analysis of a heterogenous group of individuals from four common rehabilitation populations revealed four distinct patterns of multidomain QOL outcomes that are represented across diagnostic categories. Rather than identifying distinct patterns of QOL outcomes that vary by clinical condition, the membership of these profiles includes a meaningful number of individuals from each clinical group; with one exception, no profile includes more than 50% of members from any one diagnostic group (52% of members in Profile 1 were from the TBI group). This supports our hypothesis of condition-agnostic outcome clusters. These profiles reflect shared experiences and suggest common treatment needs that are independent of diagnosis.

This research makes an important contribution to the literature because we have evaluated a large, diverse array of QOL indicators, which is unlike many existing reports using cluster analytic methods. The included outcome clusters were chosen based on empirical research to understand QOL of individuals with disabilities.^88-90 We believe these results represent an important first step for understanding how symptoms and functioning are affected holistically by a life-altering acquired disability, which could serve as a model to guide comprehensive assessment of QOL outcomes in rehabilitation medicine.

The results identify two groups that could be considered to be thriving post-injury (Profiles 1 and 2). In these groups, physical functioning and independence were preserved/intact, psychosocial health indicators were average to above-average, and participants were relatively unbothered by pain, fatigue, and sleep disturbance. On average, cognitive and economic indicators were at or above the sample mean, except for Profile 2 which exhibited slightly below average cognition (−.25 SD). TBI was the largest diagnosis represented in Profile 1 (52%), which was also the youngest groups. Conversely, Profile 2 contained relatively equal proportions of each condition. The rehabilitation needs of individuals in these profiles appear to be minimal, and follow-up could be conducted as needed. Individuals in these profiles should be encouraged to continue social engagement to maintain psychosocial health.

Profile 3 displayed generally average scores, with above average independence—which was validated by the Katz ADL scale and similar in magnitude to Profiles 1 and 2—and some problems with pain (~.5 SD below mean). Intervention for pain may be beneficial for individuals in Profile 3; otherwise, monitoring as needed should suffice.

Finally, Profile 4, which contained approximately half of the sample (47%), reported lower QOL/higher symptom burden compared to all other profiles, particularly in Independence and Physical Function (−.75 to −1.0 SD below mean), with cognitive, economic, fatigue/sleep, pain, and psychosocial indicators in the low average range (~0.5 SD below mean). Comparisons on external variables showed that Profile 4 generally reported lower income and worse independence, participation, and general health than other profiles.

Profile 4 would likely benefit from routine engagement by rehabilitation therapies aimed at improving independence and physical function, either through assistive technologies, environmental supports, or behavioral/skill training; frequent physical and occupational therapy check-ins may improve independence and other domains of QOL. Social work services may be relevant given lower economic QOL and income observed in this profile. Individuals in Profile 4 will warrant routine monitoring of pain symptoms (like Profile 3), and potentially associated problems with fatigue, sleep disruption, and poor mood (all of which can be exacerbated by pain). This outcome cluster may benefit from combined pharmacological and behavioral interventions (e.g., exercise programs and medication combined with psychotherapy can be useful for comorbid problems with pain, sleep, and mood). For Profiles 3 and 4, assistive technologies for pain should be considered.^91-103 Directed, and potentially intensive, interventions for pain and other symptoms in individuals with functional limitations—given that SCI, limb injury, or stroke can all produce injury-associated pain and are susceptible to pain related to secondary conditions—may lead to improved QOL in other domains known to be negatively influenced by pain.^104-106

Study Limitations and Future Directions

Before our results are used clinically, additional work to validate the profiles and determine their practical meaning will be needed. Study participants were not administered formal measures of symptom validity. Consequently, some participants may have under-reported or over-reported their symptoms and functioning. Future research should include measures of symptom validity, because inaccurate self-reports could result in over- or under-estimation of the need for additional clinical care; for example, it is unknown if individuals in Profiles 1 and 2 may be underreporting symptoms or difficulties due to impaired self-awareness.¹⁰⁷ Predictive validity will also be important to study. For example, future research exploring associations of post-injury employment status with these profiles could be useful—e.g., to determine whether individuals in Profiles 1 or 2 tend to return to work more frequently than individuals in other profiles. We could envision a future program whereby individuals are assigned to separate treatment tracks depending on their rehabilitation needs, although this will require further study and refinement prior to implementation.

The profiles detected in this study are based on measures that are applicable across clinical groups, which was an intentional decision to avoid including measures relevant only to one group. For example, bowel and bladder functioning are relevant almost exclusively to participants with a history of SCI, and not other groups, so these and other condition-specific measures were excluded. This could be considered a limitation regarding the comprehensiveness of the assessments included, although we did consider this a necessary step to avoid biasing the results toward detecting condition-specific profiles (which was antithetical to our goals). Finally, recent work has suggested researchers establish profile similarity across known groups —in this study, injury condition—akin to tests of measurement invariance. Attempts to do so were not possible within mclust (due to lack of support for multiple-group models or inclusion of external predictor features) and computationally intractable in the Mplus package due to the limited sample size within each injury condition (~ 200 per condition) as well as the estimation of indicator covariances, which are handled differently across packages¹⁰⁸ (convergence issues were encountered for all but simple diagonal covariance structures in Mplus). If able to be implemented in the future, the present results would benefit from such tests.

Conclusion

This pioneering study evaluated the presence of holistic outcome patterns that appear across traditional diagnostic and severity groupings, and has revealed four distinct profiles. This includes two groups that are thriving post-injury (one of which is significantly younger than all other groups), an overall average group with worse-than-average pain, and one below-average profile characterized by profound limitations in physical function and independence. We posit that each profile could benefit from a distinct rehabilitation management approach, although further study will be required.

Abbreviations

ADL

activity of daily living

ANOVA

analysis of variance

BIC

Bayesian information criterion

BLRT

bootstrap likelihood ratio test

HSD

honestly significant difference

HRQOL

health-related quality of life

ICL

integrated complete-data likelihood

LPA

latent profile analysis

PART-O

Participation Assessment with Recombined Tools-Objective

PROM

patient-reported outcome measure

PROMIS

Patient-Reported Outcomes Measurement Information System

QOL

quality of life

SCI

spinal cord injury

TBI

traumatic brain injury

Data availability:

The data presented in this article are not readily available because a data use agreement must be signed prior to release. Requests to access the datasets should be directed to the senior author, David Tulsky, at dtulsky@udel.edu.