Cognitive function at admission predicts amount of gait speed change in geriatric physical rehabilitation

Older patients tend to exhibit less improvement over the course of physical rehabilitation than younger patients [1], yet reasons for this reduced “responsiveness” to training in older adults are poorly understood. Preliminary research in stroke and Parkinson disease suggests that cognitive impairments may limit responsiveness to physical rehabilitation [2, 3], but it is currently unknown whether this relationship holds true for a more inclusive and diverse older cohort, representative of a typical rehabilitation clinic.

Gait speed is an important and common outcome measure of geriatric physical rehabilitation, in part because it is a robust predictor of mobility and mortality in older adults [4]. However, as with other outcomes, the degree to which gait speed improves through rehabilitation can be highly variable across patients [5]. Identifying individuals who are more or less likely to improve over the course of therapy will help to personalize care plans and optimize outcomes. Preliminary work suggests that cognitive factors may be related to training responsiveness in upper extremity learning [6]. Whether responsiveness to gait/mobility training is related to cognitive function is unknown. In fact, many physical rehabilitation studies a) exclude patients with low cognitive scores or b) do not report any cognitive data [7]. Moreover, concerns have been raised recently about excluding participants from clinical studies based on age or co-morbidities without clear justification [8], thereby limiting generalizability to real-world care.

Our study tests whether cognitive function predicts rehabilitative responsiveness (change in gait speed from admission to discharge) within transitional care. Over a four-month period, anyone admitted into a single, non-profit transitional care facility requiring physical therapy, and who completed the Medicare Minimum Data Set (MDS) assessment was invited to participate in sharing data for research purposes. There were 113 individuals admitted into transitional care during this time. Ninety-three persons completed the MDS and consented to the study; 14 individuals (of the 93) had missing data for SLUMS cognitive test, and were excluded from the data analysis. Thus, 70% of those admitted to the facility were included in this retrospective study (N=79; mean±SD age = 75.2±9.7 years; 67% female). In order to include a real-world sample within a community-based clinic and to avoid exclusions that could limit generalizability, no inclusion/exclusion criteria were applied based on participant characteristics. Instead, data were a convenience sample of a cohort of transitional care patients. This study was approved by our Institutional Review Board in compliance with the Declaration of Helsinki, and conforms to all STROBE guidelines (checklist available upon request).

Cognitive function was measured within seven days of admission with the Saint Louis University Mental Status Exam (SLUMS), an 11-item screen assessing memory, attention, orientation, and overall executive function. The SLUMS was administered and scored by a staff speech-language pathologist who was blinded to the rest of the study. SLUMS scores of 27–30 indicate normal cognitive function, 21–26 indicate Mild Neurocognitive Disorder (MNCD), and 1–20 indicate probable dementia. Depressive symptoms were assessed via the Minimum Dataset (MDS) item 0300, which is a sum score of nine questions related to frequency of poor mood over the past 2 weeks. The length of stay quantified therapy dose. Due to the limited nature of the dataset provided by the transitional care facility, no information regarding therapy type or primary diagnosis was provided, although co-morbidities were noted and were included in the analytic model as number of diagnoses. The range of co-morbidities was 1–14 ($\overline{x}$=7.7; SD=3.1). The most common among the sample were hypertension (65%), hyperlipidemia (43%), diabetes (31%), and arthritis (31%). Others, such as stroke (2%) or congestive heart failure (14%), were uncommon, and no participants had Alzheimer’s or Parkinson’s disease diagnoses.

The dependent variable was change in gait speed, measured with the Two-Minute Walk Test (2MWT). The 2MWT was scored by the number of feet a patient walked in two minutes. 2MWT was chosen because 1) it is a reliable measure of gait function in geriatric populations in therapy [9], thereby minimizing potential bias, and 2) 75% of the sample were noted in therapy as having “difficulty walking”, and 96% were noted as having “muscle weakness”. It was administered within seven days of admission and then again at discharge by staff physical therapists who were blinded to all cognitive and psychosocial variables. Any assistive devices used during the 2MWT were documented in therapy notes, but were not analyzed statistically since all but eight of the participants maintained the same type of assistive device from pre- to post-testing. Of these eight, seven moved toward less support (e.g., from a four-wheeled walker to a single point cane). Changes in gait ability were calculated as the difference in 2MWT performance at discharge, compared to admission, such that positive scores indicated improvement; negative scores indicated decline in performance.

Two linear regression models (SPSS version 23) were run, each predicting gait speed improvement. Model 1 included the length of stay with the potential control variables. Control variables from Model 1 that were statistically significant at a level of p<.05 were carried forward into Model 2. Model 2 was run to determine if cognitive function at admission was associated with gait speed improvement, after controlling for length of stay and any variables pulled forward from Model 1.

Average length-of-stay was 22.5 (SD=15.9) days (max=84), which is consistent with other reports of transitional care [10]. 2MWT performance was 100.2±92.5 and 274.0±108.1 feet at admission and discharge, respectively, with an average improvement of 164.1 feet (SD=115.6; range −150 to +350). Depressive symptoms at admission were low ($\overline{x}$=4.1; SD=4.5). Average SLUMS score at admission was 20.3 (SD=6.8). Categorically, 21% of patients were in the normal SLUMS range (score 27–30), 33% indicated MNCD (score 21–26), and 46% indicated dementia (score <21). However, given that the SLUMS is only a brief cognitive screen, it cannot be used to diagnose dementia; thus, no categorization of ‘dementia’ was added to the number of co-morbidities.

Model 1 results suggested that age was the only statistically significant demographic variable related to gait speed improvement (R² =.186), and therefore was the only control variable carried forward to the final model. Length of stay did not predict improved outcomes. Model 2 indicated that lower cognitive function at admission predicted less improvement in gait speed (R²=.105), even after accounting for age and length of stay (Table 1). Figure 1 illustrates the mean change in 2MWT performance for each SLUMS category, with less change associated with lower SLUMS scores.

Table 1.

Linear regression results for predicting change in 2MWT with cognitive function and control variables.

		B	Std. Error	Beta	p-value
Model 1	(Constant)	690.00	204.95		.002
	Age	−5.62	2.61	−.32	.037
	Gender	−58.18	41.75	−.21	.172
	Co-morbidity (# diagnoses)	−6.75	6.34	−.16	.294
	Depressive symptoms	−4.59	4.10	−.18	.270
	Length-of-stay (days)	9.45	8.91	.15	.294
Model 2	(Constant)	64.12	174.45		.715
	Age	−.38	2.02	−.03	.852
	Length-of-stay (days)	.54	.98	.07	.587
	SLUMS Score	5.74	2.46	.32	.023

Notes: Dependent Variable for both models = change in feet (discharge minus admission) in 2 Minute Walk Test. Model 1 R² = .186; Model 2 R² = .105

Figure 1.

Mean change in the 2MWT for the three categories of SLUMS scores: Probable dementia (0–20) (red), Mild Neurocognitive Disorder (21–26) (blue), and Normal (27–30) (green). Error bars indicate standard error.

In short, these findings suggest that cognitive impairments may predict patients’ progression through physical rehabilitation. While additional factors may explain remaining variance in rehabilitative responsiveness, there are practical implications for the current findings. Age- or disease-related changes in cognition may, in part, explain why older patients are less responsive to physical interventions than younger patients. These data are consistent with previous findings from inpatient stroke rehabilitation settings [2] and experimental studies [6] showing that brief cognitive screens (like the SLUMS) can predict functional outcomes and motor learning. Thus, by not limiting our analyses to a single diagnosis, inclusion criterion, or therapy type, our study follows new recommendations for inclusion in geriatric research [8]. As a result, our results generalize previous findings to a more clinically-representative population. We also note that the improvements were in gait speed, an outcome with high functional relevance for mobility and mortality [4].

Although this study demonstrates associations between global cognitive performance and changes in walking speed, one cannot infer 1) the direction or causality of this relationship, nor 2) the potential mechanisms linking these outcomes. Nevertheless, there is a growing body of literature linking cognitive function and motor learning [e.g., 11], which is not surprising given that motor learning utilizes a broad network of cortical and subcortical structures that overlaps with cognitive function. Thus, it is plausible that impaired cognition may impact attention when following therapists’ instructions. Poorer global cognitive function may also serve as a proxy for frailty, such that persons with both physical and cognitive impairment may be unable to recover as effectively, compared to those with physical impairment alone.

Regardless of the underlying mechanism, it may benefit geriatric physical therapists to integrate cognitive scores at admission into their therapy plan in order to tailor their interventions accordingly. The fact that the majority of our sample had SLUMS scores indicating probable dementia is particularly important, given that none had a formal diagnosis of Alzheimer’s or Parkinson’s disease listed. Without cognitive screening, therapists might be unaware of cognitive deficits and may not adjust treatment accordingly. Future studies can assess alternative therapies for improving rehabilitation success in more cognitively impaired populations. Further, given that the amount of therapy (proxied here by length of stay) was not significantly related to the patients’ improvements in therapy, these findings also shed light on recent equivocal findings on dose-response relationships in rehabilitation [12] by suggesting that cognitive function may mediate rehabilitative responsiveness, particularly for geriatric populations. In summary, this inclusive, real-world study may be useful in bringing awareness to clinicians about the importance of cognitive function during physical rehabilitation.