Functional Independence Measure Subtypes among Inpatients with Subacute Stroke: Classification via Latent Class Analysis.

Objectives: Using Functional Independence Measure (FIM) records, this study used latent class analysis (LCA) to clarify the structure of activities of daily living (ADL) status in patients following stroke.
Methods: In this retrospective, single-center study, we extracted the medical records of patients with stroke who were admitted to a rehabilitation hospital in Japan between April 2018 and March 2020. LCA was used to determine classes of ADL status based on response patterns in FIM items converted from the original seven levels to three levels: Complete Dependence, FIM1-2; Modified Dependence, FIM3-5; Independence, FIM6-7. We compared the length of stay and discharge destinations among subgroups of patients with different ADL status at admission.
Results: From 373 patients, 1592 FIM records were analyzed. These were classified into six ADL status classes based on "Complete Dependence," "Modified Dependence," and "Independence" in the motor and cognitive domains. Significant differences were observed among the six admission ADL subgroups for the length of stay (median values in patient subgroups based on admission ADL status: 126, 146, 90, 65, 44, and 29 days in the Motor Complete/Cognitive Complete, Motor Complete/Cognitive Modified, Motor Modified/Cognitive Modified, Motor Modified/Cognitive Independent, Motor Independent/Cognitive Modified, and Motor Independent/Cognitive Independent groups, respectively) and discharge destinations (patients discharged home: 27%, 62%, 81%, 92%, 95%, and 98%, respectively, and to acute care hospitals: 18%, 14%, 8%, 8%, 2%, and 2%, respectively). Conclusions: LCA successfully stratified ADL status in patients with stroke undergoing rehabilitation and may aid in determining an appropriate treatment regimen. 2022 The Japanese Association of Rehabilitation Medicine.

INTRODUCTION

Studies on activities of daily living (ADL) are crucial for rehabilitation in patients with stroke. Previous studies on ADL have employed linear regression and outcome prediction methods with the goal of improving the quality of therapeutic interventions and providing information that can aid patients in providing consent.[1],[2],[3],[4]) Additionally, the relationship between ADL ability at admission and important variables, such as discharge destination and length of stay, is critical for developing individualized treatment regimens and for management in hospitals and by insurance systems. However, few studies have addressed these aspects.

The primary scientific aim in most clinical trials is to identify the best treatment for a given disease, with any heterogeneity in patient characteristics or outcomes being viewed as an impediment to the research process.[5]) However, precision medicine has attracted increased attention in recent years, utilizing patient heterogeneity to improve treatment decisions and ensure that the right treatment is given to the right patient at the right time.[5]) As such, considering the heterogeneity of ADL status among patients with stroke is important when attempting to improve the efficacy of intervention.

The Functional Independence Measure (FIM) has been used in several studies related to ADL. FIM items are broadly classified into motor and cognitive items, and the total score and scores for each category (FIM-motor, FIM-cognition, and FIM-total) are often used as continuous variables. Evaluating item-level FIM scores is crucial when applying treatment regimens to individual patients. However, previous studies using Rasch analysis have demonstrated the relative difficulty of item-level FIM assessment.[6],[7]) Koyama et al.[8]) investigated variability in the performance profiles of motor items by applying ordinal logistic modeling analysis using the FIM-motor score as the independent variable and the independence levels of single motor items as the dependent variables. Based on their study, clinicians can now assess which of the motor items should receive more intensive training as per the patient’s FIM-motor score. Although studies have reported a relationship between a single motor item and the FIM-motor score, a similar relationship for FIM cognitive items has not been described. However, in stroke, where both higher brain and motor dysfunction serve as inhibitory factors, the relationship between motor and cognitive items is significant.[9])

Latent class analysis (LCA) assumes that the sample of subjects includes a number of latent subgroups (latent classes) and aims to classify subjects into the same latent class based on homogeneous response patterns to variables of interest (category variables).[10],[11],[12]) LCA is a form of unsupervised learning without predetermined classifications or class characteristics. When there is a relationship between observed variables (manifest variables), LCA assigns a subject to a latent class (latent variable), and the relationship disappears within the class. At this time, the law of “local independence” is established, and subjects within the same latent class are considered homogeneous based on observed variables. Although it was first introduced for use in social sciences research, LCA has recently attracted attention in healthcare fields[10],[12]) and has been applied in studies across various disciplines.[13],[14],[15]) We hypothesized that LCA could be used to classify FIM records into ADL subgroups that consider both motor and cognitive items based on the pattern of responses to FIM items, which are originally represented as ordinal categorical variables.

Therefore, in the current study, we aimed to characterize the structure of ADL status among inpatients with subacute stroke by using LCA to classify FIM records based on the responses to both cognitive and motor items. Furthermore, we applied these classifications to compare significant variables such as the length of stay and discharge destination based on ADL status at the time of admission.

MATERIALS AND METHODS

This retrospective, single-center study was performed at Saiseikai Higashikanagawa Rehabilitation Hospital in Yokohama, Japan. Because this study used medical records containing real-world data, we followed the Reporting of studies Conducted using Observational Routinely collected Data (RECORD) checklist.[16]) We retrospectively enrolled 613 patients with cerebrovascular disease who were admitted to the hospital between April 2018 and March 2020. From these patients, we selected those who met the following inclusion criterion: ischemic or hemorrhagic stroke diagnosed on neuroradiological examination (i.e., computed tomography or magnetic resonance imaging). We excluded those who met one or more of the following exclusion criteria: subarachnoid hemorrhage, subdural hematoma, cerebellar and brainstem lesions, comorbidity of acute locomotor disorders, and hospitalization for less than 1 day. These inclusion and exclusion criteria were based on the criteria used in a previous study[8]); however, patients with a history of stroke were not excluded to account for the actual situation in research facilities in Japan, in which patients with recurrent stroke are relatively common. As a result, the number of patients entered into the final database determined the sample size.

In accordance with the procedures outlined by the Japanese Health Insurance System (2018–2020), patients were referred from acute care medical services within 60 days after the stroke occurred and subsequently could receive inpatient care in our rehabilitation hospital for up to 180 days. During long-term hospitalization, they received physical therapy, occupational therapy, and speech therapy for a total of 120–180 min every day. Similar clinical settings are used across rehabilitation hospitals in Japan.

Measures

To evaluate ADL status, the FIM[17]) was administered at the time of admission, at monthly intervals, and at the time of discharge. The FIM is one of the most widely used, reliable, and valid tools for evaluating ADL status after stroke.[6],[7],[18],[19],[20]) The FIM consists of 13 motor items (eating, grooming, bathing, dressing the upper body, dressing the lower body, toileting, bladder control, bowel control, transfer to bed/chair/wheelchair, transfer to toilet, transfer to tub/shower, walking or wheelchair use, and climbing stairs) and five cognitive items (comprehension, expression, social interaction, problem-solving, and memory). Each item is evaluated on a seven-point scale (1, total assistance; 2, maximal assistance; 3, moderate assistance; 4, minimal assistance; 5, supervision; 6, modified independence; 7, complete independence). In the present study, the FIM assessments were performed by the nurse treating the patient. In LCA, collapsing multiple response options into two or three options makes it easier to interpret the class solution.[21]) Therefore, in this study, FIM response levels were grouped into three levels (1–2, Complete Dependence; 3–5, Modified Dependence; 6–7: Independence) to interpret the major features of each class in accordance with previously described methods.[22])

In addition to obtaining scores for the individual FIM items, the total-summation score (FIM-total), subtotal-summation score of motor components (FIM-motor), and subtotal-summation score of cognitive components (FIM-cognition) were extracted on a seven-point scale. Moreover, data related to the length of stay (days) and discharge destination (home, nursing home/long-term hospital, acute care hospital in case of deterioration of the patient’s condition, or other rehabilitation facilities) were extracted as important variables.

Statistical Analysis

To determine the ability of the FIM to capture ADL status rather than classify patients, we analyzed multiple FIM records under the assumption of independent and identical distribution. All FIM records were classified into latent classes (latent ADL status) based on item responses using LCA, which allowed for assumption regarding the order of the categorical variables observed in this study.[23]) FIM records were classified according to responses to the ordinal categorical variables that were integrated into a three-level classification system.

The number of classes was selected based on the following selection criteria while favoring the most parsimonious model[12],[21],[24],[25]): (a) statistical measures of model fit, (b) number of subjects in the class, (c) certainty of individual membership, and (d) interpretability. The model fit indices relied on the Bayesian information criterion (BIC) and Akaike information criterion (AIC), decreases in which indicated improvements in the model. Based on these criteria, the number of FIM records should be 5% or more for each class, and the certainty of individual membership should have an entropy value of 0.8 or greater. Interpretability should also be confirmed by several physiatrists based on the class characteristics. We interpreted the characteristics of the classes using the item-response probability output of the LCA. These probabilities indicate the probability that a FIM record in a class will respond to the observed measure (i.e., the FIM items). In this study, the profile of item-response probabilities was visualized using a heat map for the three FIM levels. Unlike cluster analysis, in which the subjects are completely classified into one class, LCA calculates the probability of each subject being assigned to each class. In this study, patients had different numbers of FIM records depending on their length of stay. Therefore, the probability of belonging to each class was given for each FIM record that the patient had completed at some point of evaluation. For notational purposes, G represents the latent class including the FIM record (g; 1, ..., G), N represents the patient (n; 1, …, N), J represents the time of evaluation (j; 1, …, J), and X1–X18 represent the responses to the 18 items of the FIM. The probability that a FIM record at time j in patient n belongs to the latent class g is represented by (see Equation 1 and Fig. 1):

Fig. 1.

 Probability that the FIM record at time j in patient n belongs to the latent class g.

Using the probability of belonging to each class (Equation 1) given to all FIM records as a weight, the total FIM score (FIM-motor, FIM-cognition, and FIM-total) was calculated for each class. represents the total FIM score of the FIM record at time j in patient n.

We then compared the length of stay and discharge destination according to the class of ADL status of the patients at admission. FIM records at admission were extracted from all FIM records, following which the length of stay in each subgroup was calculated using the probability (Equation 1) as a weight. Differences among the groups were compared using the Steel–Dwass test (significance level, α=0.05) in consideration of multiplicity. In Equation 3, L represents the length of stay in patient n.

Similarly, the discharge destinations of patient subgroups based on ADL status at admission were examined using the chi-square test (significance level, α=0.05). Finally, using the characteristics of the class (ADL status), we suggested a treatment regimen for that ADL status at each time point. Additionally, we utilized findings regarding the length of stay and discharge destination for each ADL status at admission to update the suggested treatment regimen. All statistical analyses were conducted using LatentGOLD®5.1.0 (Statistical Innovations, Arlington, MA, USA) and JMP® 15.1.0 (SAS Institute, Cary, NC, USA). The heat map was drawn using Python 3.9.0 (Python Software Foundation, DE, USA). LCA was performed using a dataset that excluded FIM records with missing values.

Ethics

This study was approved by the Saiseikai Higashikanagawa Rehabilitation Hospital Ethics Committee (No. 21–04). Informed consent was obtained using the opt-out method, and information on this study was published on the hospital bulletin board and website.

RESULTS

Patients

Based on the inclusion and exclusion criteria, 373 patients were selected for this study (Fig. 2). Of 1626 extracted FIM records, 34 were excluded because of missing values for FIM items, leaving 1592 FIM records for analysis. The characteristics of the study population are summarized in Table 1. The patients varied widely in terms of age, lesion site, and lesion side, and the study population contained a larger proportion of men than women. The population included 70 patients with a history of stroke and 48 with a history of locomotor disorders.

Fig. 2.

 Flowchart showing the scheme used for patient selection.

Table 1.

 Patient characteristics

Patient characteristic		Total n=373
Sex	Female	115
Age, years	Median (IQR)	70 (57–78)
Lesion type	Hemorrhage	154
	Infarction	215
	Infarction with hemorrhage	4
Lesion side	Left	201
	Right	155
	Both	17
Acute treatment	Operative	67
Previous stroke	(+)	70
Previous locomotor disorders	(+)	48
Lesion site	Internal carotid artery	15
	Anterior cerebral artery	8
	Middle cerebral artery	66
	Posterior cerebral artery	3
	Frontal lobe	18
	Parietal lobe	12
	Temporal lobe	14
	Occipital lobe	8
	Corona radiata	30
	Thalamus	56
	Basal ganglion and internal capsule	113
	Others	15
	Combined lesions	8
	Disseminated	4
	Missing value	3

IQR, interquartile range.

Classification

A total of 1592 FIM records were classified into latent classes (ADL status) using LCA, based on combinations of the 18 FIM items. To select the best latent class solution, we examined latent class models with one to seven classes. Table 2 lists the model fit indices. The difference in the BIC/AIC fit between models indicated that the improvements in model fit with each additional class began to slow at the six-class model. Hence, based on reductions in BIC and AIC and other confirmed criteria after verifying the interpretability of the class characteristics with two physiatrists, we selected the six-class model for this study.

Table 2.

 Model fit indices of the latent class analysis of 18 FIM items

	BIC	AIC	df	Entropy	ΔBIC	ΔAIC
1-Class	58,717	58,523	1556	1.00	-	-
2-Class	41,168	40,873	1537	0.98	17,549	17,650
3-Class	35,491	35,093	1518	0.97	5677	5781
4-Class	34,078	33,578	1499	0.95	1413	1515
5-Class	33,110	32,508	1480	0.95	968	1070
6-Class	32,534	31,830	1461	0.95	576	678
7-Class	32,024	31,218	1442	0.94	510	612

Table 3 shows the proportion of FIM records belonging to each class (class size), as well as the item-response probabilities of the FIM items within each class; for example, the class size of class 1 was 0.16 (254.7 records, 1592×0.16=254.7), and the ‘probability of independence in eating’ within class 1 was 0.02 (5.1 of 254.7 records of class 1 indicating independence, 254.7×0.02=5.1). Figure 3 shows a heat map of the item-response probabilities listed in Table 3. As a general trend, independence improved from class 1 to class 6. Items such as eating, grooming, and bladder and bowel control, for which most responses indicated independence (FIM6–7), could be achieved relatively easily. Conversely, items such as transfer to the tub/shower and climbing stairs, for which most responses indicated complete dependence (FIM1–2), were more difficult to achieve. Table 4 presents the data for the total FIM scores of each ADL status according to all FIM records. When assessing the total FIM scores, we observed that the FIM-motor and FIM-total scores of each class were arranged in order, whereas the FIM-cognition scores were arranged differently.

Table 3.

 Class response probability across three levels for each FIM item

FIM item	Grouped response	Class 1	Class 2	Class 3	Class 4	Class 5	Class 6	Total
		MCCC	MCCM	MMCM	MMCI	MICM	MICI
	Class size	0.16	0.11	0.22	0.13	0.10	0.28	1.00
Eating	Independence	0.02	0.15	0.33	0.80	0.87	0.98	0.56
	Modified Dependence	0.40	0.67	0.60	0.20	0.13	0.02	0.31
	Complete Dependence	0.58	0.18	0.07	0.00	0.00	0.00	0.13
Grooming	Independence	0.00	0.00	0.08	0.46	0.85	0.98	0.45
	Modified Dependence	0.21	0.69	0.89	0.54	0.15	0.02	0.39
	Complete Dependence	0.79	0.31	0.03	0.00	0.00	0.00	0.16
Bathing	Independence	0.00	0.00	0.00	0.03	0.48	0.65	0.24
	Modified Dependence	0.06	0.31	0.87	0.94	0.52	0.35	0.51
	Complete Dependence	0.94	0.69	0.13	0.03	0.00	0.00	0.25
Dressing the upper body	Independence	0.00	0.00	0.02	0.15	0.80	0.97	0.38
	Modified Dependence	0.13	0.20	0.90	0.84	0.20	0.03	0.38
	Complete Dependence	0.87	0.80	0.08	0.01	0.00	0.00	0.24
Dressing the lower body	Independence	0.00	0.00	0.02	0.11	0.77	0.96	0.37
	Modified Dependence	0.04	0.07	0.70	0.82	0.23	0.04	0.31
	Complete Dependence	0.96	0.93	0.28	0.07	0.00	0.00	0.32
Toileting	Independence	0.00	0.00	0.02	0.27	0.93	0.99	0.42
	Modified Dependence	0.08	0.05	0.78	0.72	0.07	0.01	0.29
	Complete Dependence	0.92	0.95	0.20	0.01	0.00	0.00	0.29
Bladder control	Independence	0.00	0.02	0.23	0.71	0.94	0.98	0.52
	Modified Dependence	0.03	0.14	0.31	0.22	0.06	0.02	0.13
	Complete Dependence	0.97	0.84	0.46	0.07	0.00	0.00	0.35
Bowel control	Independence	0.00	0.05	0.34	0.88	0.92	0.99	0.57
	Modified Dependence	0.05	0.25	0.40	0.11	0.08	0.01	0.15
	Complete Dependence	0.95	0.70	0.26	0.01	0.00	0.00	0.28
Transfer to bed/chair/wheelchair	Independence	0.00	0.00	0.01	0.23	0.86	0.99	0.40
	Modified Dependence	0.38	0.44	0.99	0.77	0.14	0.01	0.44
	Complete Dependence	0.62	0.56	0.00	0.00	0.00	0.00	0.16
Transfer to toilet	Independence	0.00	0.00	0.00	0.22	0.89	0.99	0.40
	Modified Dependence	0.24	0.29	0.99	0.78	0.11	0.01	0.40
	Complete Dependence	0.76	0.71	0.01	0.00	0.00	0.00	0.20
Transfer to tub/shower	Independence	0.00	0.00	0.02	0.03	0.32	0.49	0.18
	Modified Dependence	0.06	0.02	0.51	0.56	0.64	0.49	0.40
	Complete Dependence	0.94	0.98	0.47	0.41	0.04	0.02	0.42
Walking or wheelchair use	Independence	0.00	0.00	0.18	0.34	0.70	0.87	0.40
	Modified Dependence	0.15	0.15	0.55	0.53	0.28	0.13	0.30
	Complete Dependence	0.85	0.85	0.27	0.13	0.02	0.00	0.30
Stairs	Independence	0.00	0.00	0.02	0.03	0.31	0.47	0.17
	Modified Dependence	0.05	0.01	0.39	0.45	0.60	0.49	0.36
	Complete Dependence	0.95	0.99	0.59	0.52	0.09	0.04	0.47
Comprehension	Independence	0.00	0.31	0.12	0.91	0.24	0.99	0.48
	Modified Dependence	0.38	0.67	0.80	0.09	0.73	0.01	0.40
	Complete Dependence	0.62	0.02	0.08	0.00	0.03	0.00	0.12
Expression	Independence	0.02	0.42	0.21	0.95	0.36	0.93	0.52
	Modified Dependence	0.39	0.54	0.66	0.05	0.58	0.07	0.35
	Complete Dependence	0.59	0.04	0.13	0.00	0.06	0.00	0.13
Social interaction	Independence	0.01	0.51	0.43	0.93	0.65	0.99	0.62
	Modified Dependence	0.41	0.47	0.54	0.07	0.34	0.01	0.28
	Complete Dependence	0.58	0.02	0.03	0.00	0.01	0.00	0.10
Problem-solving	Independence	0.00	0.09	0.01	0.56	0.07	0.90	0.35
	Modified Dependence	0.01	0.85	0.70	0.44	0.85	0.10	0.42
	Complete Dependence	0.99	0.06	0.29	0.00	0.08	0.00	0.23
Memory	Independence	0.00	0.20	0.08	0.76	0.25	0.98	0.45
	Modified Dependence	0.07	0.74	0.75	0.24	0.70	0.02	0.36
	Complete Dependence	0.93	0.06	0.17	0.00	0.05	0.00	0.19

Fig. 3.

 Heat map of response probability for the 18 FIM items. Independence, white; Modified dependence, light gray; Complete dependence, dark gray.

Table 4.

 Total FIM scores in each class

		Class 1	Class 2	Class 3	Class 4	Class 5	Class 6
		MCCC	MCCM	MMCM	MMCI	MICM	MICI
FIM-motor	Mean	20.0	26.3	47.5	60.7	79.1	83.8
	Median	18	26	48	62	79	85
	(IQR)	(13, 25)	(21, 32)	(40, 54)	(56, 67)	(73, 87)	(80, 89)
FIM-cognition	Mean	10.1	23.6	20.7	30.9	24.7	33.3
	Median	10	24	21	31	26	34
	(IQR)	(7, 13)	(20, 27)	(17, 25)	(29, 33)	(23, 27)	(32, 35)
FIM-total	Mean	30.0	49.9	68.2	91.6	103.8	117.2
	Median	30	49	68	92	104	118
	(IQR)	(22, 37)	(44, 56)	(60, 76)	(86, 97)	(99, 112)	(112, 123)

As shown in the heat map in Fig. 3, class 1 responded to most motor and cognitive items with “Complete Dependence”; therefore, we referred to this ADL status as “Motor-Complete Dependence, Cognition-Complete Dependence” (MCCC). MCCC tended to be completely dependent for general FIM items but exhibited improvements in motor items such as eating, grooming, and transfer to bed/chair/wheelchair. Similarly, in the assessments of cognitive items, we observed improvements in comprehension, expression, and social interaction.

Class 2 responded to most of the motor items with “Complete Dependence” and to most of the cognitive items with “Modified Dependence.” Therefore, we referred to this ADL status as “Motor-Complete Dependence, Cognition-Modified Dependence” (MCCM).

Class 3 responded to most motor and cognitive items with “Modified Dependence.” Therefore, we referred to this ADL status as “Motor-Modified Dependence, Cognition-Modified Dependence” (MMCM). In the detailed evaluation of response characteristics, modified dependence was generally common; however, independence was observed for motor items such as eating, bladder control, and bowel control.

Class 4 responded to most of the motor items with “Modified Dependence” and to most of the cognitive items with “Independence.” Therefore, we referred to this ADL status as “Motor-Modified Dependence, Cognition-Independence” (MMCI). In a detailed evaluation of responses to FIM items, the findings for motor items were similar to those observed in MMCM; however, an increase in independence was noted, especially in eating, grooming, bladder control, and bowel control.

Class 5 responded to most of the motor items with “Independence” and to most of the cognitive items with “Modified Dependence.” Therefore, we referred to this ADL status as “Motor-Independence, Cognition-Modified Dependence” (MICM). In the detailed evaluation of responses to FIM items, independence was generally common in most of the motor items, but modified dependence remained in bathing, transfer to tub/shower, and climbing stairs. Modified dependence was generally common for cognitive items such as comprehension, expression, problem-solving, and memory.

Class 6 responded to most of the motor and cognitive items with “Independence.” Therefore, we referred to this ADL status as “Motor-Independence, Cognition-Independence” (MICI). In the detailed evaluation of responses to FIM items, independence was generally common; however, modified dependence remained in bathing, transfer to tub/shower, and climbing stairs.

Length of Stay and Discharge Destination

Table 5 shows the data related to the length of stay for each patient subgroup according to ADL status (class 1–6) at admission, while Table 6 shows the data related to the discharge destination. Patients belonging to MCCC at the time of admission are represented as MCCCa, and the same convention was used to designate the other subgroups. In the MCCCa subgroup, 27% of patients achieved home discharge, and the median length of stay was 126 days; 18% of patients were transferred to an acute care hospital because of unfavorable outcomes. In the MCCMa subgroup, 62% of patients achieved home discharge, and the median length of stay was 146 days. In the MMCMa subgroup, 81% of patients achieved home discharge, and the median length of stay was 90 days. In the MMCIa subgroup, 92% of patients achieved home discharge, and the median length of stay was 65 days. In the MICMa subgroup, 95% of patients achieved home discharge, and the median length of stay was 44 days. In the MICIa subgroup, 98% of patients achieved home discharge, and the median length of stay was 29 days. Comparisons indicated that greater independence was associated with a relatively shorter length of stay (although this trend was reversed between the MCCCa and MCCMa subgroups) and that the length of stay differed significantly among the subgroups, except between MCCCa and MCCMa, MMCMa and MMCIa, and MMCIa and MICMa (Steel–Dwass test, MCCCa–MCCMa, P=0.7005; MMCMa–MMCIa, P=0.0526; MMCIa–MICMa, P=0.511; MICMa–MICIa, P=0.0052; MCCCa–MMCMa, P=0.0163; MMCMa–MICMa, P=0.0007; others, P <0.0001). The mean length of stay was smaller than the median in the MCCCa and MCCMa subgroups and higher than the median in the other classes. Furthermore, the discharge destinations differed significantly among the subgroups [chi-square test, χ2(15)=149.32; P <0.0001]: the proportion of patients in the MCCCa subgroup that were discharged to home was small and the proportion discharged to an acute care hospital was large when compared with the other subgroups.

Table 5.

 Length of stay in each patient subgroup according to ADL class at admission

	MCCCa	MCCMa	MMCMa	MMCIa	MICMa	MICIa	Total
Class size	87.9	70.7	72.2	51.9	38.9	51.4	373
Mean	119.8	133.6	96.1	75.6	61.2	34.8	93.8
(95% CI)	(110.9–128.6)	(123.72–143.4)	(86.4–105.9)	(64.1–87.1)	(47.9–74.5)	(23.2–46.3)	(88.4–99.3)
Median	126	146	90	65	44	29	91
(IQR)	(78, 166)	(109, 168)	(64, 127)	(44, 106)	(29, 93)	(21, 37)	(44, 141)

Data given in days.

CI, confidence interval.

Table 6.

 Discharge destination in each patient subgroup according to ADL class at admission

	MCCCa	MCCMa	MMCMa	MMCIa	MICMa	MICIa	Total
	Freq.	Freq.	Freq.	Freq.	Freq.	Freq.	Freq.
	(χ2)	(χ2)	(χ2)	(χ2)	(χ2)	(χ2)
Home	24.1	43.5	58.3	47.7	36.9	50.4	261
	(22.8)	(0.7)	(1.2)	(3.6)	(3.5)	(5.8)
Nursing home, long-term hospital	43.9	17.1	6.9	0.2	1.0	0.0	69
	(46.8)	(1.3)	(3.1)	(9.3)	(5.4)	(9.5)
Acute care hospital	16.0	10.0	6.0	4.0	1.0	1.0	38
	(5.5)	(1.1)	(0.2)	(0.3)	(2.3)	(3.4)
Other rehabilitation facilities	4.0	0.0	1.0	0.0	0.0	0.0	5
	(6.7)	(0.9)	(0)	(0.7)	(0.5)	(0.7)
Discharged to home	27%	62%	81%	92%	95%	98%	70%
Discharged to acute care hospital	18%	14%	8%	8%	2%	2%	10%

Freq., frequency.

DISCUSSION

In this study, we characterized latent ADL status (class) by classifying FIM records based on response patterns to motor and cognitive items. FIM records were then classified into six latent classes of ADL status based on patterns of complete dependence, modified dependence, and independence in the motor and cognitive domains. To our knowledge, this is the first study to utilize LCA to stratify ADL patterns based on FIM results in patients with stroke. As a result, we were unable to compare these patterns with those obtained using similar methods. However, the response probability for FIM items indicated that items such as eating, grooming, and bladder and bowel control were relatively easy, whereas items such as transfer to the tub/shower and climbing stairs were more difficult. This finding is consistent with that of a previous study that utilized a different method.[8]) In the analysis of the total FIM score, the order in which the classes were arranged was different depending on whether the value represented the FIM-total, FIM-motor, or FIM-cognition score, suggesting the multidimensionality of the FIM. Therefore, LCA represents an appropriate strategy given the multidimensional nature of the FIM.[26]) Furthermore, our findings indicated that the length of stay and discharge destination differed depending on the subgroup into which patients were categorized based on ADL status at admission. In the Japanese Health Insurance System, the optimal length of stay is based on the classification of diseases, and criteria such as the optimal proportion of discharge to home are set for facilities. This study suggests the necessity and possibility of setting different recommended values that focus on ADL status at admission. In addition, further studies are required to consider socio-environmental factors such as the presence of support at home, living with others, and marital status. These factors were reported as crucial for the consideration of home discharge in a previous systematic review[27]) and in another Japanese study.[28]) The paragraphs below include suggestions for the selection of treatment regimens based on the latent ADL status and patient subgroups according to ADL status at admission.

MCCC responded to the motor and cognitive items with “Complete Dependence”; however, higher levels of independence were seen in eating, grooming, and transfer activities, which was consistent with the findings of previous studies showing that these activities are less difficult.[8]) Additionally, MCCC showed improved independence in comprehension, expression, and social interaction in cognitive items. Considering the interventions employed at rehabilitation hospitals in Japan, promoting interactions with other patients and staff (e.g., in the cafeteria or in common areas) and group exercise may lead to improvements in cognitive items and FIM scores in these patients. In MCCCa, a patient subgroup that belonged to MCCC at the time of admission, the proportion of discharge to home was 27% and the median length of stay was 126 days; therefore, an intervention plan that considers the discharge destination from an early stage is important. Although the overall proportion of discharge to an acute care hospital in this study was 10%, the corresponding proportion in the MCCCa subgroup was 18%. In a study of stroke patients admitted to rehabilitation hospitals in Japan, Tokisato et al. noted that 6.9% were transferred to acute care hospitals.[29]) In the USA, Stineman et al. reported that 9.1% of patients were transferred from stroke rehabilitation hospitals to acute care hospitals.[30]) In this study, the overall proportion of patients discharged to acute care hospitals was similar; however, when patients were classified by ADL status at the time of admission, the proportion reached 18% in the MCCCa group. This result suggests that close attention must be paid to clinical deterioration in patients classified as MCCCa.

MCCM responded to most of the motor items with “Complete Dependence” and to most of the cognitive items with “Modified Dependence.” Although ADL status was better among the MCCM than the MCCC, the length of stay was longer in the MCCMa than in the MCCCa group. Unlike in other subgroups, the mean value for the length of stay in both the MCCCa and MCCMa groups was smaller than the median. This indicates the presence of outliers with shorter lengths of stay. It was considered that patients with severe conditions but with a short length of stay would have been discharged to an acute care hospital because of clinical deterioration. In fact, there were several acute care transfers in the MCCCa and MCCMa groups (especially in the MCCCa group, which may have resulted in shortened length of stay). Alternatively, the length of stay may have been extended because the proportion of discharge to home was higher in the MCCMa group (62%) than in the MCCCa group (27%), and the preparation for home discharge required more time for patients in the MCCMa group. These findings indicate that patients in both the MCCCa and MCCMa groups should be monitored for clinical deterioration, and that systematic support should be provided from an early stage while considering the discharge destination and length of stay.

MMCM responded to most of the motor and cognitive items with “Modified Dependence.” Considering that the MMCM generally showed improved independence for motor items, treatment in this class should focus on expansion of ADL function (e.g., providing guidance to the toilet under assistance). However, when attempting to increase independence for motor-related items, the risk of falls should be carefully evaluated. Previous studies have indicated that unilateral spatial neglect is a serious risk factor for falls and is an impediment to the improvement of FIM-motor scores and favorable outcomes in terms of discharge destination and length of stay.[31],[32],[33],[34]) Furthermore, for cognitive items, the degree of independence was low for problem-solving and memory in the MMCM. Although the current study did not investigate factors associated with ADL improvement, it is nonetheless important to carefully increase independence through training and environmental settings that address higher brain dysfunction, such as unilateral spatial neglect. Characteristically, the findings for bladder and bowel control were mixed in the MMCM (independence and complete dependence). Although the MCCM responded to items related to bladder and bowel control with complete dependence, the MMCI responses were close to independence. This suggests that MMCM represents the boundary between independence and dependence for bladder and bowel control, and an intensive intervention approach may be effective in this class. In this study, the median FIM-motor score was 62 in the MMCI and 48 in the MMCM. A previous study reported that more than 50% of patients with FIM-motor scores of 40–50 had bladder and bowel control scores of 5 or greater.[8]) This is consistent with the results of the present study. In the MMCMa group, the proportion of discharge to home exceeded 80%, and the median length of stay was 90 days. A treatment regimen aimed at home discharge and a short length of stay is recommended for the MMCMa group.

MMCI responded to most of the motor items with “Modified Dependence” and to most of the cognitive items with “Independence.” MMCI showed some independence in grooming, dressing, toileting, transfer, and walking or wheelchair use, whereas MICM could perform many of these tasks independently. Therefore, with MMCI representing the boundary of independence for these items, we suggest that an intensive intervention approach for these items may be effective for patients with an ADL status of MMCI. In the MMCIa group, the proportion of patients discharged to home exceeded 90%, and the median length of stay was 65 days. Considering this length of stay for patients in this group, we recommend a treatment regimen aimed at home discharge.

MICM responded to most motor items with “Independence” and to most cognitive items with “Modified Dependence.” Although the cause of decreased independence for cognitive items was not examined in this study, stroke presents with dysfunction in various higher-order processes in the brain, such as unilateral spatial neglect and aphasia. The decreased expression and comprehension abilities in the MICM may be related to aphasia. In the MICMa group, the proportion of patients discharged to home exceeded 95%, and the median length of stay was 44 days. Previous studies have shown that aphasia is a major factor for extended length of stay in rehabilitation hospitals,[35]) which is consistent with longer length of stay in the MICMa group than in the MICIa group. For patients with higher brain dysfunction, such as those with aphasia, an appropriate length of stay should be established, and appropriate training and adjustment should be provided to facilitate home discharge.

MICI responded to most motor and cognitive items with “Independence”; however, “Modified Dependence” was noted for difficult activities such as bathing, transfer to tub/shower, and climbing stairs. As a result, MICI was considered to represent basic ADL independence. In the MICIa group, the proportion of patients discharged to home exceeded 98%, and the median length of stay was 29 days. This suggests that a treatment approach for these patients should aim for early discharge and incorporate higher goals, including advanced movements that are not evaluated in the FIM.

Study Limitations and Future Directions

In the analysis of FIM data, most studies consider the seven-level FIM response for each component of the test.[7],[36]) However, based on the purpose and the limitations of analysis in the current study, we considered a three-level FIM to be suitable. Further consideration will be needed to yield any findings regarding analysis using the seven-level FIM or the categories verified for level integration.

In our analysis, missing data may not have been missing at random. However, because the percentage of missing data was small (2%), it was not considered to have a significant effect on the results. In this study, our analysis aimed to capture FIM response patterns to reflect ADL status in all clinical situations at different time points, although the number of time points differed in each case. Given that this study aimed to classify ADL status rather than to stratify individual patients, the multiple FIM records for each patient were regarded as independent and identically distributed, representing different ADL states at different time points. In addition, when comparing the length of stay among patient subgroups classified based on the class to which they belonged at admission, our preliminary analysis using only FIM records at the time of admission did not disagree with our conclusion. Furthermore, the length of stay examined in this study was based on the stay in the convalescent hospital and did not consider the length of stay in acute care hospitals. However, we chose to focus on convalescent hospitals to improve treatment regimens in this setting. We considered that more severely affected patients with poor ADL status would require a longer stay in an acute care setting. This means that an analysis of length of stay in an acute care hospital may show significant differences in patients subdivided according to ADL status at admission. However, the length of stay in an acute care hospital may not be entirely based on the severity of symptoms, highlighting the need for similar studies to consider the effects of the length of stay in this setting. Finally, this study was performed at a single center in Kanagawa; therefore, the data used in this study may be different from those obtained from other regions in Japan and/or the world. Conversely, in a data-driven society, it is crucial to assume that treatment results will vary depending on the treatment environment and region. Ultimately, it is critical to develop a general analytical method and system that can propose clinical treatment options based on easily accessible data at each facility.

CONCLUSIONS

The current study demonstrated that LCA could be used to clarify ADL structure in patients with stroke. LCA identified six latent classes of ADL status based on the pattern of responses to FIM items on an ordinal scale, incorporating both motor and cognitive items. Furthermore, this analysis revealed that the length of stay and discharge destination differed depending on ADL status at admission. These findings highlight the importance of clarifying the structure of ADL status when designing an appropriate treatment regimen.