Prediction of Discharge Destination After Stroke Using the Motor Assessment Scale on Admission: A Prospective, Multisite Study

Methods 

Results 

A total of 566 patients across 15 sites were admitted to the study, with discharge destination information collected from 554 subjects. Of these subjects, 410 (74%) were discharged home, 32 (5.8%) were discharged to a hostel, 60 (10.8%) were discharged to a nursing home, 44 (7.9%) were transferred to an acute ward, and 8 (1.4%) died. Most patients were living at home prior to their stroke (96.6%). Demographic characteristics are outlined in table 1.

Table 1. Participant Characteristics

	Characteristic	n	Mean ± SD	n (%)
	Age (y)	563	72.5±13
	Stroke to rehabilitation admission (d)	548	15.3±16
	Length of stay in rehabilitation (d)	546	40.1±35
	Sex (males)	565		306(54.1)
	Side of stroke	560
	Left			269(47.5)
	Right			250(44.6)
	Other			41(7.2)
	Prestroke residence	558
	Home			539(96.6)
	Hostel			15(2.7)
	Nursing home			4(0.7)
	Discharge residence	554
	Home			410(74.0)
	Hostel			32(5.8)
	Nursing home			60(10.8)
	Acute ward			44(7.9)
	Death			8(1.4)

Abbreviation: SD, standard deviation.

Admission functional scores across the 8 items of the MAS are outlined in table 2. The Spearman rank correlation found that all 8 admission MAS scores had a significant (P<.001) correlation with discharge destination, with r values ranging from −.192 to −.35 (see table 2). A more dependent residence after rehabilitation (nursing home) was most associated with lower MAS scores in gait, rolling, lie to sit, sit-to-stand ability, and sitting balance scores on admission. These MAS scores had a correlation with discharge destination of greater than .30 and thus were then included in the logistic regression.

Table 2. Functional Ability (admission MAS scores) Correlation With Discharge Destination and the Admission MAS Scores for Subjects Discharged to Home or Residential Aged Care

	MAS Item	Correlation With Discharge Destination⁎	Score Median (25th–75th percentile)		Mean Score ± SD
			Home	RAC	Home	RAC
	Rolling (MAS-1)	−.341	5(2–6)	1(0–5)	4.18±2.17	2.34±2.36
	Lie to sit (MAS-2)	−.338	6(3–6)	3(1–5)	4.64±1.83	3.14±2.09
	Balanced sit (MAS-3)	−.305	5(3–5)	2(2–4)	4.11±1.63	2.92±1.61
	Sit to stand (MAS-4)	−.315	2(1–5)	1(1–2)	3.23±2.13	1.80±1.81
	Gait (MAS-5)	−.350	2(0–4)	0(0–1)	2.24±1.46	0.74±1.46
	Upper arm (MAS-6)	−.227	5(1–6)	1(0–5)	3.71±2.26	2.28±2.38
	Hand (MAS-7)	−.204	4(0–6)	1(0–4)	3.30±2.49	1.95±2.31
	Advanced hand (MAS-8)	−.192	1(0–2)	0(0–2)	1.81±2.07	0.95±1.65

Abbreviation: RAC, residential aged care.

⁎ Coded as 0=home, 1=hostel, 2=nursing home.

Table 3 presents the adjusted odds ratios (ORs) and 95% confidence interval (CI) result of the multinomial forward stepwise logistic regression model (model A) best able to predict discharge destination to home or residential aged care facility (hostel and nursing home). Admission gait ability (MAS-5), prestroke residential status, age, and admission rolling ability (MAS-1) were included in the final model (Nagelkerke r2=.405, P<.001). This model had a sensitivity (correctly predicting discharge home) of 99% and a specificity (correctly predicting discharge to a residential aged care facility) of 33.3%, with an overall predictive accuracy of 87.3% (table 4). The adjusted ORs indicate that for every 1-point increase in MAS-5 (gait), subjects were 1.66 times more likely to go home (1/.60; 95% CI, 1.28–2.27). Similarly, for every 1 point increase in MAS-1 (rolling), subjects were 1.28 times more likely to go home (1/.78; 95% CI, 1.11–1.49). Receiver operator characteristic (ROC) curves are presented in figure 1, with area under the curve calculations of 75% for gait ability, 73% for rolling, and 67% for age. The cross-validation estimate of accuracy of this model was .864.

Table 3. Prediction of Discharge Home Versus Residential Aged Care

	Characteristics	Adjusted OR	95% CI
	Model A
	Prestroke residence	1.47×104⁎	(0.00–∞)†
	Gait (MAS-5)	0.60	(0.44–0.79)†
	Rolling (MAS-1)	0.78	(0.67–0.90)‡
	Age	1.05	(1.03–1.08)†
	Constant	1.47	NA
	Model B
	Prestroke residence	1.49×104⁎	(0.00–∞)†
	Gait (MAS-5) (0–1 vs 2–6)	0.16	(0.07–0.32)†
	Rolling (MAS-1) (0–5 vs 6)	0.38	(0.17–0.79)‡
	Age	1.05	(1.03–1.09)†
	Constant	0.45	NA

Abbreviation: NA, not applicable.

⁎ The large adjusted ORs are due to 96.6% having a prior residence at home. † P<.001; ‡ P<.01.

Table 4. Actual Versus Predicted Discharge Destinations

		Predicted Destination
	Destination	Home	RAC	% Correct
	Actual
	Home	396	4	99.0
	Destination
	RAC	58	29	33.3
	Overall %			87.3

View Large Image Download to PowerPoint

Fig 1. ROC curve for MAS-5, MAS-1, and age to predict discharge destination of home versus residential aged care.

To define more specifically the predictive functional capabilities for the clinician, the gait and rolling categories were each recoded as binary variables. The gait (MAS-5) was recoded as either unable (scoring 0 or 1) or able (scoring 2–6) to achieve the task with standby help. Rolling (MAS-1) was recoded as either unable (scoring 0–5) or able (scoring 6) to achieve the task in 3 seconds without overbalancing. These new categorical variables were entered along with prestroke residential status and age. Model B (Nagelkerke r2=.386, P<.001) had a similarly high sensitivity (99.5%), but a lower specificity (19.5%), with a slightly lower overall predictive accuracy of 85.2% when compared with the previous model (see table 4). The adjusted ORs indicate that if subjects scored greater than or equal to 2 in MAS-5 (gait), they were 6.25 times more likely to go home than if scoring a 1 or 2 (1/.16; 95% CI, 3.13–14.28). Similarly, if subjects scored a 6 in MAS-1 (rolling), they were 2.63 times more likely to go home than if scoring a 5 or less (1/.38; 95% CI, 1.27–5.88). ROC curves are presented in figure 2, with area under the curve calculations of 70% for gait ability, 66% for rolling, and 67% for age. The cross-validation estimate of accuracy of this model was .835.

View Large Image Download to PowerPoint

Fig 2. ROC curve for MAS-5 and MAS-1 recoded as binary outcomes, and age to predict discharge destination of home versus residential aged care.

Discussion 

We found that MAS scores of gait and rolling ability on admission to rehabilitation, together with prestroke residential status and age were able to predict discharge destination of patients with stroke to home or residential aged care facility with an overall accuracy of 87%. This is important because, even though the MAS is widely used in Australian rehabilitation facilities, this is the first study to investigate its ability to predict discharge destination.

The MAS, previous residential status, and age had a very high sensitivity at predicting discharge to home (99%) but a comparably low specificity in predicting discharge to a residential aged care facility (33.3%). This is likely to reflect the importance of issues other than the factors in our model in the decision to discharge to residential aged care facilities. This is supported by Mokler et al,21 who were more successful in predicting stroke survivors who were not discharged home (76.6% correct) than those who were discharged home (53.4%) using the FIM functional-related groups, with the most significant variables being toilet transfer, memory, and bladder management. Our strong prediction of discharge home, however, indicates that the 4 predictors in our model (prestroke residence, gait ability, rolling ability, age) are factors important in the decision to discharge to home.

Our overall accuracy of predicting discharge destination (87%) was similar to that of Rabadi and Blau,20 who found an overall accuracy of 82% in a model using admission ambulation velocity, age, and admission FIM score. Wee and Wong18 also found an overall accuracy of 85.3% when using family supports and admission BBS score in predicting discharge destination poststroke. Although the FIM instrument is a widely accepted measure of functional outcome, some authors have identified potential limitations to the clinical use of the FIM and BBS. These have included the time taken and specialized training required to complete the scales,31 difficulty with scoring criteria32, 33 and possible item redundancy of the BBS.33 Thus if the FIM or BBS are not routinely performed on patients on admission to rehabilitation, the MAS may prove of similar discharge predictive ability and take a shorter time to perform.

Our ORs of the best predictive model indicate that for every increase in the MAS gait item score by one, the odds of being discharged to home increases by a factor of 1.66. An increase in score in this item reflects independence, gait speed, and ability to multitask. The multiple dimensions of this scale may be a key contributor to its predictive ability for discharge home. In support, gait abilities have previously been demonstrated to contribute to prediction of discharge home.10, 13, 20 Another reason for this item to prove predictive may have been the sensitivity of the scale within this item. English et al34 found the gait item of the MAS to be sensitive to change within the rehabilitation period, showing low floor and ceiling effects.

This is the first report of such a low level task (rolling) significantly contributing to the prediction of discharge destination poststroke. For every increase in the rolling score by 1, the odds of being discharged home increases by a factor of 1.28. The ability to roll in bed may be a key factor for discharge home because for many stroke survivors, the failure to achieve this ability may preclude them from living alone or with a frail aging partner who is unable to provide the necessary physical assistance. Another reason for its importance in this model may be because this task involves multifaceted abilities required for home living that are different from that of gait. It requires complex motor control of the limbs and trunk and an accurate sense of orientation. This premise is supported by findings that perceptual problems14 and poor trunk control10 are factors contributing to discharge to a residential aged care facility. The inclusion of age and previous living arrangement in the final model with best predictive ability confirm previous studies of their importance with similar ORs.10, 13

The recoding of the MAS items into binary cutoff levels did not diminish the overall accuracy of its predictive ability and may prove of greater clinical use. The better cutoff was the description of the need for standby assistance, rather than independence, to best predict discharge home. This finding reflects the reports that most stroke survivors discharged home still require some form of physical assistance.5

The conduct of this study over 15 public and private rehabilitation sites facilitates the generalization of findings of this study. In addition, cross-validation of the predictive models produced a high estimate of accuracy (.83). It is acknowledged that the prediction model should be validated in a second independent group of patients.35 Our discharge rate to home (74%) was similar to other Australian prospective studies (75%)4 and is comparable with another Australian multisite3 study that had a discharge rate to the community of 84%, but a younger cohort (61.5y vs 72.5y).

Conclusions 

By using 2 items of the MAS, age, and previous residential status, discharge to home could be predicted in 99% of cases. Discharge home poststroke frequently requires an adjustment of lifestyle for carers, and the use of community services. Both these factors would benefit from early planning to maximize readiness of the stroke survivor and carer for community living. In demonstrating that simple demographic and clinical measures can contribute to prediction of discharge destination, clinicians may be better able to contribute to discharge planning.