Sex Differences in the Role of Multimorbidity on Poststroke Disability

Abstract

Background and Objectives

Multimorbidity is common in patients who experience stroke. Less is known about the effect of specific multimorbidity patterns on long-term disability in patients with stroke. Furthermore, given the increased poststroke disability frequently seen in female vs male patients, it is unknown whether multimorbidity has a similar association with disability in both sexes. We assessed whether specific multimorbidity clusters were associated with greater long-term poststroke disability burden overall and by sex.

Methods

In the Taiwan Stroke Registry, an ongoing nationwide prospective registry, patients with first-ever ischemic stroke were enrolled; this analysis is restricted to those individuals surviving to at least 6 months poststroke. Using a hierarchical clustering approach, clusters of prestroke multimorbidity were generated based on 16 risk factors; the algorithm identified 5 distinct clusters. The association between clusters and 12-month poststroke disability, defined using the modified Rankin Scale (mRS), was determined using logistic regression models, with additional models stratified by sex. The longitudinal association between multimorbidity and functional status change was assessed using mixed-effects models.

Results

Nine-thousand eight hundred eighteen patients with first-ever ischemic stroke were included. The cluster with no risk factors was the reference, “healthier” risk group (N = 1,373). Patients with a cluster profile of diabetes, peripheral artery disease (PAD), and chronic kidney disease (CKD) (N = 1882) had significantly greater disability (mRS ≥ 3) at 1 month (OR [95% CI] = 1.36 [1.13–1.63]), 3 months (OR [95% CI] = 1.27 [1.04–1.55]), and 6 months (OR [95% CI] = 1.30 [1.06–1.59]) but not at 12 months (OR [95% CI] = 1.16 [0.95–1.42]) than patients with a healthier risk factor profile. In the sex-stratified analysis, the associations with this risk cluster remained consistent in male patients (OR [95% CI] = 1.42 [1.06–1.89]) at 12 months, who also had a higher comorbidity burden, but not in female patients (OR [95% CI] = 0.95 [0.71–1.26]), who had higher proportions of severe strokes and severe disability (p-interaction = 0.04).

Discussion

Taiwanese patients with multimorbidity, specifically the concurrent presence of diabetes, PAD, and CKD, had higher odds of a worse functional outcome in the first 6 months poststroke. Clusters of multimorbidity may be less informative for long-term disability in female patients. Further studies should evaluate other mechanisms for worse disability in female patients poststroke.

Introduction

Multimorbidity, defined as having two or more chronic health conditions in the same individual, is a major public health burden, being prevalent in most people older than 65 years.^1,2 Clinical events such as experiencing a stroke can further aggravate these clinical conditions in patients with multimorbidity, and the multimorbidity, in turn, may worsen poststroke outcome or interfere with recovery.^3-6 Mechanisms of slowing poststroke recovery in patients with multimorbidity include being unable to physically participate in (or being insufficiently referred for) rehabilitative therapies or not taking a secondary prevention medication because of risks of drug-drug interactions causing severe adverse events.³ Furthermore, caregivers of patients with multimorbidity may have difficulty meeting additional demands that arise from stroke-related disability, in addition to the multimorbidity for which they are already providing care.⁷

The construct of multimorbidity may be more meaningful in stroke-related functional outcome than are individual risk factors, which are common in patients with stroke. In one Swedish study, prestroke multimorbidity burden had a strong long-term prognostic influence on predicting mortality and functional outcomes up to 5 years.⁸ This study, however, did not include a measure of stroke severity, an important confounder of the observed association, nor did it consider ischemic stroke etiology. Finally, it did not systematically evaluate demographic differences in the role of multimorbidity on poststroke functional outcome, which remains one important unanswered question in the field.

Sex differences in poststroke disability have been noted previously, with worse disability poststroke in female than in male patients; these differences have been attributed to disparities in the frequency of comorbid conditions as well as in differences of baseline functional status at the time of an individual's first stroke.⁹ Compared with male patients, female patients have poorer functional prestroke status, are on average older at the stroke event, and have a higher prevalence of atrial fibrillation and incidence of cardioembolic strokes.⁹ Better understanding how these health conditions cluster overall and by sex, and if they may explain some of the observed sex differences in functional outcome after stroke, may allow for more targeted rehabilitative therapies to mitigate poststroke functional impairment.

In this study, we assess the association between multimorbidity and long-term disability burden while accounting for age, sex, education, and stroke severity in first-ever ischemic stroke survivors. We use here a machine learning–informed approach that summarizes prevalent risk factor patterns in patients with stroke and tests the hypothesis that certain patient groups defined by cluster membership have higher disability burden poststroke. We also test whether the associations between these multimorbidity clusters and poststroke disability differs by sex, to evaluate a possible mechanism for sex disparities in poststroke outcome, and evaluate whether differences in ischemic stroke etiology may account for the association between multimorbidity clusters and poststroke disability.

Methods

Patients

A subset of patients with first-ever recorded ischemic stroke (N = 9,818) from an ongoing nationwide prospective registry, the Taiwan Stroke Registry (TSR), was included in the analysis.¹⁰ The registry was established to assess the quality of stroke care in the country, and cases were systematically collected based on predetermined registry protocols. Patients were enrolled when meeting all the following criteria: (1) presenting within 10 days of symptom onset to a TSR hospital; (2) being clinically examined involving CT and/or MRI; and (3) meeting one of the 5 stroke type definitions as determined by the institution's clinical team, namely ischemic stroke, transient ischemic attack (TIA), intracerebral hemorrhage, subarachnoid hemorrhage, and cerebral venous thrombosis. Informed consent was signed by each participant giving permission for follow-up by TSR. Only patients with a first-ever ischemic stroke were included in this analysis. The selection of patients with ischemic stroke is shown in a flowchart in eFigure 1 (links.lww.com/WNL/D379). Additional clinical data from CT/MRI, as well as carotid ultrasonography, ECG/echocardiography, and any other clinically indicated tests to evaluate more atypical causes of stroke, were collected.¹⁰ Ischemic stroke was subcategorized into one of the following 5 major subtypes according to the Trial of ORG 10172 in Acute Stroke Treatment criteria by the patient's attending physician^10,11: large artery atherosclerosis, small vessel occlusion, cardioembolism, stroke of other etiology, and undetermined etiology. Preadmission data were obtained through patient interview, patient's or family's self-report, pre-existing medical records, and through stored information provided by the national insurance system. These preadmission data, details about the inpatient hospitalization, discharge information, and follow-up data were recorded prospectively, in real time, as were multiple poststroke functional status assessments at 1 month, 3 months, 6 months, and 12 months. Post-stroke functional assessments are part of routine clinical practice in Taiwan. In this analysis, only those patients with complete modified Rankin Scale follow-up data were included, which by definition meant that it included only those individuals surviving for at least 6 months poststroke (because an mRS of 6, consistent with death, was possible at the 12-month visit, but the mRS was not repeated after having previously been coded as an mRS of 6).

Comorbidity, Stroke Severity, and Demographic Data

The TSR collected information on 16 major risk factors and conditions that are evaluated in this study. These include hypertension, heart failure, previous TIA, atrial fibrillation, ischemic heart disease, peripheral artery disease (PAD), diabetes, hypertriglyceridemia, hypercholesterolemia (fasting total cholesterol ≥200 mg/dL, or low-density lipoprotein cholesterol ≥130 mg/dL, or with current lipid-lowering drug treatment), obesity (BMI ≥30 kg/m²), daily alcohol use, current smoker (of any amount in the past 2 years), past smoker (prior smoking but with no smoking in the past 2 years), any history of cancer, chronic kidney disease (CKD [stage 5, defined as eGFR <15 mL/min/1.73 m², or requiring dialysis], and polycythemia (increased concentration of RBC in blood based on the report of a blood test). Comorbid risk factors included those known at the time of admission, either from medical interview and examination or from medical records, and new diagnoses made during hospitalization, such as atrial fibrillation, diabetes, hypertension, hyperlipidemia, and hypertriglyceridemia. Occasionally, if any risk factors were still missing, their presence or absence was abstracted from outpatient clinical records. Sex-specific stroke risk factors were not evaluated. The participant's sex was self-reported.

To capture the extent of comorbidity burden in each group, a measure of total comorbidity was used. Total comorbidity burden was defined as the sum of individual medical conditions and categorized into none (0), mild (1), moderate (2–3), and high (≥4). The severity of ischemic stroke at enrolment and discharge was assessed by the clinical team using the NIH Stroke Scale (NIHSS).¹² Investigators rating the NIHSS had to be certified by the Taiwan Stroke Society. Stroke severity at discharge was categorized into minor stroke (NIHSS ≤ 4), moderate stroke (NIHSS = 5–15), and severe stroke (NIHSS ≥ 16). Age, sex, and maximum educational level were added as additional covariates in the analysis.

Poststroke Functional Status

Follow-up assessments were performed by TSR-trained neurologists and study nurses on patients. Functional status was recorded at 1 month, 3 months, 6 months, and 12 months after hospital discharge using the modified Rankin Scale (mRS).^13,14 The assessments were normally completed in-person at the outpatient department. For patients who were not followed up in-person by the original hospital, disability was recorded by phone at the central site or by the research team at the respective hospital where the patient was treated. Case managers were provided with incentives if a patient was completely followed for 12 months. Greater poststroke disability was defined as mRS ≥ 3.¹⁵

Statistical Analysis

Sex differences in age, education, NIHSS, and total comorbidity burden were analyzed using the Wilcoxon rank-sum and Pearson χ² tests. Using multiple imputation by chained equations, 325 missing educational levels and 12 missing values for age were imputed.¹⁶ Functional mRS scores of 6 (deceased) (N = 67) at 12 months were excluded from the analysis because mRS scores of 6 were only recorded in the data at this time point but not at any other follow-up assessments.

An agglomerative hierarchical clustering analysis, which is an unsupervised machine learning method, was used to analyze the co-occurrences of disease patterns (eFigure 2, links.lww.com/WNL/D379).¹⁷ We chose the agglomerative instead of the divisive approach because it has been used more often in prior multimorbidity literature and because it starts by assigning each patient profile to its own cluster initially before iteratively joining the most similar clusters afterward.^18,19 This bottom-up approach allows for initially grouping together homogeneous risk factor profiles of patients with smaller overall prevalence rates in the population. The Gower method with the Dice distance measure was used to assess the similarity of the patients' risk and disease profiles. Clusters were merged based on the minimum between-cluster distance using the Ward squared minimum variance clustering method, and a cluster dendrogram was built. To assess how close patient records in one cluster were to records in neighboring clusters, the average silhouette width was computed. The optimal number of hierarchical clusters (N = 5) was primarily determined based on ocular inspection of the dendrogram and by determining the highest average silhouette width (σ = 0.21) in a range of 1–7 clusters. Differences in age, sex, education, NIHSS, and ischemic stroke subtypes were tested between the 5 clusters using the Pearson χ² and Kruskal-Wallis rank-sum tests. To characterize the multimorbidities associated with each cluster, the observed/expected ratios were calculated by dividing the prevalence of a given condition within a cluster by its prevalence in the overall population. A condition was associated with a cluster when the observed/expected ratio was ≥1.5 and when the intracluster prevalence of a condition was ≥0.25. Clusters were characterized in the overall ischemic cohort and in the 2 sex groups separately (eFigure 3, A–C).

The associations between clusters of multimorbidity and poststroke disability adjusted by age, sex, education, and NIHSS were tested using logistic regression models, overall and stratified by sex, and the interaction between sex and the disability clusters was formally assessed. Testing for sex differences was an a priori hypothesis. The associations of the clusters with greater disability were evaluated conditional on the number of hierarchical clusters that were specified beforehand. The reference cluster in the factor was determined based on the most healthy disease profile before the logistic regression analysis. Specifically, we assessed whether any significant cluster effect observed at the first follow-up time point, that is, 1 month, was consistent across the remaining follow-up mRS assessments in the TSR. To compare the accuracy of the model's prediction with a model without multimorbidity or without considering the NIHSS as covariates, the receiver-operating curve and area under the curve were computed. It was also tested whether any significant effect observed between multimorbidity clusters and greater disability remained significant when including ischemic subtype as an additional factor in the model. In 2 sensitivity analyses, it was tested whether the significant associations would change when excluding patients older than 70 years or patients with high stroke severity (NIHSS≥ 16).

Using a mixed-effects model, it was also assessed whether there was a significant effect of time on mRS and whether functional status trajectory depended on multimorbidity clusters. In a subsequent mixed-effects analysis, it was also tested whether the interaction between time and cluster type on mRS varied by sex membership (using cluster typeXtimeXsex interaction term). The different models were adjusted by the confounders age, sex, education, and stroke severity. The model's underlying normality assumption for the random effects was checked using a QQ plot.

We compared the results of the longitudinal analysis when replacing multimorbidity clusters with comorbidity burden as the model predictor and assessed whether functional status recovery (mRSXtime) varied by comorbidity burden (comorbidity burdenXtime) and additionally by sex (comorbidity burdenXtimeXsex). We furthermore stratified the mixed model analysis by sex. The different models were adjusted by the demographic confounders and stroke severity.

The statistical analysis was conducted using R (version 3.6.2).

Standard Protocol Approvals, Registrations, and Patient Consents

This TSR study was approved by the Joint Institutional Review Board of CMUH102-REC l-086(CR-5) at China Medical University Hospital. The data that supported the findings of this study were approved by individual hospital IRBs from contributing TSR investigators listed in Appendix 2. Informed consent to participate in the stroke registry study was obtained from all patients.

Data Availability

Restrictions may apply to the public availability of these data. However, processed data sets can be requested and made available from the authors with permission from the TSR central IRB at China Medical University Hospital, Taichung, Taiwan.

Results

Participant Characteristics

The median age of the patient cohort was 67 years (Table 1). Small vessel occlusion was the most prevalent ischemic stroke etiology (48%), followed by large vessel atherosclerosis (26%), undetermined (14%), cardioembolic (10%), and specific other (1.2%) etiology. Most patients had a moderate or high total comorbidity burden (2 or more individual health conditions) (81%) but only experienced a stroke with minor severity (77%) (Table 1). Although they were significantly older than male patients (70 years (female) vs 64 years (male) (p < 0.001), female patients had high prestroke comorbidity (4+ health conditions) less frequently than male patients (19% vs 36%, p < 0.001). Ischemic stroke etiology distribution also differed by sex: Female patients were more likely to have a cardioembolic stroke (12% vs 8.7%) than male patients, but with a similar percentage of small vessel ischemic stroke (47% vs 50%) as male patients. Moderate or severe strokes were more common in female than in male patients (27.3% vs 20.4%). Greater poststroke disability was significantly more common among female than male patients at 1 month (41% vs 27%) and at all 3 follow-up time points (Table 1). Prevalence estimates for each individual condition stratified by sex are presented in eTable 1 (links.lww.com/WNL/D379).

Table 1.

Demographic and Clinical Characteristics in the Ischemic Stroke Cohort Overall and Stratified by Sex

Characteristics	Overall (N = 9818)a	Female (N = 3955)a	Male (N = 5863)a	p Valueb
Age	67 (76–57)	70 (77–61)	64 (74–56)	<0.001
Education				<0.001c
Less than elementary school	2,399 (24)	1,611 (41)	788 (13)
Elementary school	3,593 (37)	1,443 (36)	2,150 (37)
Junior high school	1,283 (13)	334 (8.4)	949 (16)
Senior high school	1,478 (15)	362 (9.2)	1,116 (19)
University college	982 (10)	193 (4.9)	789 (13)
Graduate school	83 (0.8)	12 (0.3)	71 (1.2)
Ischemic stroke etiology				<0.001c
Large vessel atherosclerosis	2,539 (26)	1,020 (26)	1,519 (26)
Small vessel occlusion	4,741 (48)	1,842 (47)	2,899 (50)
Cardioembolism	991 (10)	479 (12)	512 (8.7)
Stroke of other etiology	118 (1.2)	45 (1.1)	73 (1.2)
Undetermined etiology	1,420 (14)	568 (14)	852 (15)
Unknown	9 (0.1)	1 (0.1)	8 (0.1)
NIHSS				<0.001c
NIHSS ≤ 4	7,547 (77)	2,856 (72)	4,691 (80)
NIHSS = 5–15	2,016 (21)	929 (23)	1,087 (19)
NIHSS ≥ 16	255 (2.6)	170 (4.3)	85 (1.4)
Total comorbidity burden				<0.001c
None	302 (3.1)	185 (4.7)	117 (2.0)
Mild	1,558 (16)	840 (21)	718 (12)
Moderate	5,124 (52)	2,192 (55)	2,932 (50)
High	2,834 (29)	738 (19)	2,096 (36)
Disability 1 mo				<0.001c
mRS < 3	6,636 (68)	2,348 (59)	4,288 (73)
mRS ≥ 3	3,182 (32)	1,607 (41)	1,575 (27)
Disability 3 mo				<0.001c
mRS < 3	7,204 (73)	2,581 (65)	4,623 (79)
mRS ≥ 3	2,614 (27)	1,374 (35)	1,240 (21)
Disability 6 mo				<0.001c
mRS < 3	7,444 (76)	2,692 (68)	4,752 (81)
mRS ≥ 3	2,374 (24)	1,263 (32)	1,111 (19)
Disability 12 mo				<0.001c
mRS < 3	7,488 (77)	2,698 (69)	4,790 (82)
mRS ≥ 3	2,263 (23)	1,226 (31)	1,037 (18)
Unknown	67 (0.6)	31 (0.8)	36 (0.6)

NIHSS: minor stroke ≤ 4; moderate stroke 5–15; severe stroke 16 or greater; total comorbidity burden: none = 0; mild = 1; moderate = 2–3; high ≥ 4; disability: lower disability mRS < 3; greater disability mRS ≥ 3.

^aMedian (IQR); n(%).

^bWilcoxon rank-sum test; Pearson χ² test.

^cSignificant (p < 0.05) associations.

Cluster Classification and Cluster Differences

The heatmap illustrating the similarities (lower distances (more similarity) marked in red patches) between patients' medical histories is presented in Figure 1A. One large-sized and 4 medium-sized clusters were derived from the patients' risk factor records when using the agglomerative hierarchical clustering method (Figure 1B).¹⁷ None of the 16 conditions were attributed to cluster 1, so this cluster, labeled as the healthier risk factor group, was used as the reference group for subsequent analyses. In cluster 2, previous TIA, atrial fibrillation, heart failure, polycythemia, past smoking, and cancer were significant characteristics (Figure 1C). Whereas cluster 3 was marked by diabetes, CKD, and PAD, cluster 4 had current smoking and daily alcohol use as its main multimorbidity attributes. Cluster 5 had obesity and hypertriglyceridemia as its main attributes. When stratifying by sex, the clusters' risk factor profiles changed minimally, except for heart failure, which was previously only attributed to cluster 2 in the total cohort, but in male patients, it was also classified as part of cluster 3. The proportions of the conditions attributed to each cluster in the overall ischemic cohort and in the sex-stratified cohort are displayed in eFigure 3A and eFigure 3B-C (links.lww.com/WNL/D379), respectively.

Figure 1. Hierarchical Clustering Analysis.

Panels: (A) Heatmap graph visualizing the dissimilarity matrix. The heatmap graph shows strong (red) vs low-risk factor similarity (blue) between patient records. (B) The hierarchical cluster tree with the squared Ward minimum variance clustering method showing the creation of the 5 clusters. (C) Network plot showing the risk factors attributed to each cluster based on the observed/expected ratio ≥ 1.5 and the intracluster prevalence ≥0.25.

The patients' ages between the clusters differed significantly (Table 2). Patients in cluster 2 were the oldest (median age 71, Q1, Q3 61, 78) and had the highest proportion of cardioembolic (14%), undetermined ischemic (16%), and severe (3.8%) strokes. By contrast, patients in cluster 4 were the youngest (median age 60, Q1, Q3 52, 69) and mostly male (94%) and had the highest proportion of minor strokes (82%) and lowest prevalence of severe strokes (1.0%). Both cluster 4 (52%) and cluster 5 (53%) had the highest prevalence of patients with strokes from small vessel occlusion. By contrast, the proportion of large vessel atherosclerosis was highest in cluster 3 (30%). Cluster 5 was the only group where the proportion of female patients (66%) was higher than of male patients.

Table 2.

Characterizing the Multimorbidity Clusters Based on Sex, Age, Ischemic Stroke Subtype, NIHSS, and Disability

Characteristics	Overall (N = 9,818)b	Cluster 1a (N = 1,373)b	Cluster 2a (N = 3,454)b	Cluster 3a (N = 1882)b	Cluster 4a (N = 1,650)b	Cluster 5a (N = 1,459)b	p Valuec
Age	67 (57–76)	68 (57–76)	71 (61–78)	66 (58–74)	60 (52–69)	66 (57–74)	<0.001f
Sex							<0.001f
Female	3,955 (40)	623 (45)	1,503 (44)	771 (41)	96 (5.8)	962 (66)
Male	5,863 (60)	750 (55)	1,951 (56)	1,111 (59)	1,554 (94)	497 (34)
Ischemic stroke etiology							<0.001f
Large vessel atherosclerosis	2,539 (26)	358 (26)	801 (23)	561 (30)	434 (26)	385 (26)
Small vessel occlusion	4,741 (48)	638 (47)	1,575 (46)	908 (48)	850 (52)	770 (53)
Cardioembolism	991 (10)	147 (11)	482 (14)	167 (8.9)	100 (6.1)	95 (6.5)
Specific etiology	118 (1.2)	33 (2.4)	38 (1.1)	12 (0.6)	23 (1.4)	12 (0.8)
Undetermined	1,420 (14)	196 (14)	553 (16)	234 (12)	241 (15)	196 (13)
Unknown	9 (0.1)	1 (0.1)	5 (0.2)	0 (0)	2 (0.1)	1 (0.1)
NIHSSd							<0.001f
NIHSS ≤ 4	7,547 (77)	1,098 (80)	2,596 (75)	1,372 (73)	1,349 (82)	1,132 (78)
NIHSS = 5–15	2,016 (21)	248 (18)	727 (21)	457 (24)	284 (17)	300 (21)
NIHSS ≥ 16	255 (2.6)	27 (2.0)	131 (3.8)	53 (2.8)	17 (1.0)	27 (1.9)
Disability 1 moe							<0.001f
mRS < 3	6,636 (68)	962 (70)	2,257 (65)	1,166 (62)	1,287 (78)	964 (66)
mRS ≥ 3	3,182 (32)	411 (30)	1,197 (35)	716 (38)	363 (22)	495 (34)
Disability 3 moe							<0.001f
mRS < 3	7,204 (73)	1,031 (75)	2,440 (71)	1,292 (69)	1,378 (84)	1,063 (73)
mRS ≥ 3	2,614 (27)	342 (25)	1,014 (29)	590 (31)	272 (16)	396 (27)
Disability 6 moe							<0.001f
mRS < 3	7,444 (76)	1,063 (77)	2,517 (73)	1,338 (71)	1,422 (86)	1,104 (76)
mRS ≥ 3	2,374 (24)	310 (23)	937 (27)	544 (29)	228 (14)	355 (24)
Disability 12 moe							<0.001f
mRS < 3	7,488 (77)	1,055 (77)	2,525 (74)	1,362 (73)	1,428 (87)	1,118 (77)
mRS ≥ 3	2,263 (23)	310 (23)	895 (26)	503 (27)	220 (13)	335 (23)
Unknown	67 (0.6)	8 (0.6)	34 (0.1)	17 (0.1)	2 (0.1)	6 (0.4)

^aCharacteristics in group clusters: 1—(reference group) none; 2—polycythemia, previous TIA, cancer, atrial fibrillation, heart failure, past smoker; 3—diabetes, PAD, CKD; 4—current smoker, daily alcohol use; 5—obesity, hypertriglyceridemia.

^bn (%); median (IQR).

^cPearson χ² test; Kruskal-Wallis rank-sum test.

^dNIHSS: minor stroke ≤ 4; moderate stroke 5–15; severe stroke 16 or greater.

^eDisability: lower disability mRS < 3; greater disability mRS ≥ 3.

^fSignificant (p < 0.05) associations.

The comorbidity burden also differed considerably between the 5 clusters. All patients allocated to clusters 3, 4, and 5 had at least moderate comorbidity, by definition, and more than one-third of them had a high comorbidity burden, meaning that they had at least 4 clinical conditions (eTable 2, links.lww.com/WNL/D379).

Association Between Multimorbidity Cluster and Disability

Patients in cluster 3, characterized by diabetes, CKD, and PAD, had a significantly higher odds than patients in cluster 1 of greater disability at 1 month (OR [95% CI] = 1.36 [1.13–1.63]), 3 months (OR [95% CI] = 1.27 [1.04–1.55]), and 6 months (OR [95% CI] = 1.30 [1.06–1.59]) after adjusting for age, education, and NIHSS (Table 3, eFigure 4, links.lww.com/WNL/D379). No other cluster of multimorbid risk factors was associated with greater disability at any time point poststroke, compared with the healthy risk factor cluster (cluster 1).

Table 3.

Associations Between Clusters of Multimorbidity and Poststroke Disability at 1, 3, 6, and 12 Months, Overall and Stratified by Sex

		Disability (mRS ≥3) 1 ma	Disability (mRS ≥3) 3 ma	Disability (mRS ≥3) 6 ma	Disability (mRS ≥3) 12 ma
Overall, with cluster group as main effectb	Group cluster	OR (95% CI)	OR (95% CI)	OR (95% CI)	OR (95% CI)
	1	1 (Ref)	1 (Ref)	1 (Ref)	1 (Ref)
	2	0.97 (0.82–1.15)	0.94 (0.79–1.13)	0.95 (0.79–1.14)	0.87 (0.72–1.05)
	3	1.36 (1.13–1.63)g,h	1.27 (1.04–1.55)h	1.30 (1.06–1.59)g,h	1.16 (0.95–1.42)f
	4	0.92 (0.74–1.13)	0.86 (0.68–1.08)	0.78 (0.62–0.99)h	0.81 (0.64–1.03)
	5	1.15 (0.94–1.40)	1.03 (0.83–1.28)	1.02 (0.82–1.26)	0.92 (0.74–1.14)
Female onlyc,d,e
	1	1 (Ref)	1 (Ref)	1 (Ref)	1 (Ref)
	2	0.93 (0.73–1.18)	0.94 (0.73–1.21)	0.90 (0.70–1.17)	0.79 (0.61–1.03)
	3	1.06 (0.81–1.39)	1.08 (0.81–1.44)	1.04 (0.78–1.39)	0.95 (0.71–1.26)
	4	0.60 (0.30–1.14)	0.68 (0.32–1.35)	0.75 (0.35–1.51)	0.78 (0.37–1.55)
	5	1.01 (0.78–1.31)	1.04 (0.79–1.37)	0.98 (0.74–1.30)	0.83 (0.63–1.10)
Male onlyc,d,e,f
	1	1 (Ref)	1 (Ref)	1 (Ref)	1 (Ref)
	2	1.03 (0.81–1.31)	0.95 (0.73–1.22)	1.01 (0.78–1.32)	0.97 (0.75–1.27)
	3	1.67 (1.30–2.16)h	1.45 (1.10–1.91)h	1.59 (1.20–2.11)h	1.42 (1.06–1.89)h
	4	1.03 (0.80–1.33)	0.88 (0.67–1.16)	0.83 (0.63–1.11)	0.89 (0.67–1.19)
	5	1.36 (0.99–1.87)	1.00 (0.71–1.41)	1.04 (0.73–1.49)	1.07 (0.74–1.54)

^aAdjusted by age, sex, education, and NIHSS.

^bCharacteristics in group clusters: 1—(reference group) none; 2—polycythemia, previous TIA, cancer, atrial fibrillation, heart failure, past smoker; 3—diabetes, PAD, CKD; 4—current smoker, daily alcohol use; 5—obesity, hypertriglyceridemia.

^cFemale comorbidity burden: none (4.7%), low (21%), moderate (55%), high (19%); male comorbidity burden: none (2.0%), low (12%), moderate (50%), high (36%).

^dFemale stroke severity: minor (72%), moderate (23%), severe (4.3%); male stroke severity: minor (80%), moderate (19%), severe (1.4%).

^eFemale disability at 12 mo: mRS < 3 (69%), mRS ≥ 3 (31%); male disability at 12 mo: mRS < 3 (82%), mRS ≥ 3 (18%).

^fCharacteristics in group clusters: 1—(reference group) none; 2—polycythemia, previous TIA, cancer, atrial fibrillation, heart failure, past smoker; 3—diabetes, PAD, CKD, heart failure; 4—current smoker, daily alcohol use; 5—obesity, hypertriglyceridemia.

^gThe interaction effect sex * multimorbidity was statistically significant (p < 0.05) in the overall analysis.

^hSignificant (p < 0.05) associations.

Sex-Stratified Analysis

When the interaction between sex and multimorbidity clusters was evaluated, there was evidence of significant effect modification by sex for cluster 3 at 1 month (p = 0.02), 6 months (p = 0.03), and 12 months (p = 0.04) (Table 3). When analyses were further stratified by sex, the associations between cluster 3 and greater disability remained significant only in male patients but not in female patients across all follow-up assessments (Table 3). Thus, in male patients, having a cluster 3 multimorbid pattern was associated with an elevated odds of poststroke disability (vs cluster 1, OR 1.59, 95% CI 1.20–2.11 at 6 months), but in female patients, presence of the same cluster 3 multimorbid pattern, compared with cluster 1, was not associated with elevated poststroke disability at 6 months (OR 1.04, 95% CI 0.78–1.39) (with similar patterns at all follow-up time points).

Sensitivity Analyses

Adding ischemic stroke etiology as an additional covariate to the existing multimorbidity model did not attenuate the significant association between cluster 3 and greater disability within the first 6 months (Table 4). The association between cluster 3 and greater disability up to 3 months also remained significant when only including patients younger than 70 years (Table 4). When excluding patients with severe strokes, the association between cluster 3 and greater disability remained significant within the first 6 months but not at 12-month follow-up assessment (Table 4).

Table 4.

Associations Between Multimorbidity Clusters and Poststroke Disability at 1, 3, 6, and 12 Months With Adjustment for Ischemic Stroke Etiology as an Additional Covariate, in Patients Younger Than 70 Years and in Patients With Minor-to-Moderate Stroke Severity

		Disability (mRS ≥3) 1 m	Disability (mRS ≥3) 3 m	Disability (mRS ≥3) 6 m	Disability (mRS ≥3) 12 m
Ischemic stroke etiology includeda	Group clusterb	OR (95% CI)	OR (95% CI)	OR (95% CI)	OR (95% CI)
	1	1 (Ref)	1 (Ref)	1 (Ref)	1 (Ref)
	2	0.98 (0.83–1.16)	0.94 (0.79–1.13)	0.95 (0.79–1.15)	0.87 (0.72–1.05)
	3	1.36 (1.13–1.63)d	1.26 (1.04–1.54)d	1.29 (1.05–1.58)d	1.15 (0.94–1.41)
	4	0.92 (0.74–1.13)	0.85 (0.68–1.07)	0.78 (0.61–0.99)d	0.80 (0.63–1.02)
	5	1.16 (0.95–1.42)	1.04 (0.84–1.29)	1.02 (0.82–1.27)	0.92 (0.74–1.15)
Age below 70 (N = 5,589)c
	1	1 (Ref)	1 (Ref)	1 (Ref)	1 (Ref)
	2	0.99 (0.76–1.29)	0.95 (0.71–1.28)	0.92 (0.68–1.25)	0.92 (0.68–1.27)
	3	1.50 (1.14–1.97)d	1.42 (1.05–1.92)d	1.29 (0.95–1.76)	1.24 (0.90–1.71)
	4	0.96 (0.72–1.28)	0.87 (0.63–1.20)	0.68 (0.49–0.95)d	0.74 (0.52–1.05)
	5	1.33 (0.99–1.77)	1.17 (0.85–1.63)	1.08 (0.78–1.52)	1.06 (0.75–1.49)
Minor-to-moderate stroke (N = 9,563)c
	1	1 (Ref)	1 (Ref)	1 (Ref)	1 (Ref)
	2	0.97 (0.82–1.15)	0.94 (0.78–1.12)	0.95 (0.79–1.14)	0.87 (0.72–1.04)
	3	1.36 (1.13–1.64)d	1.27 (1.04–1.55)d	1.30 (1.06–1.59)d	1.16 (0.95–1.42)
	4	0.92 (0.74–1.13)	0.85 (0.68–1.07)	0.78 (0.61–0.99)d	0.80 (0.63–1.02)
	5	1.14 (0.94–1.39)	1.03 (0.83–1.27)	1.01 (0.81–1.26)	0.92 (0.74–1.14)

^aAdjusted by age, sex, education, NIHSS, and stroke etiology.

^bCharacteristics in group clusters: 1: 1—(reference group) none; 2—polycythemia, previous TIA, cancer, atrial fibrillation, heart failure, past smoker; 3—diabetes, PAD, CKD; 4—current smoker, daily alcohol use; 5—obesity, hypertriglyceridemia.

^cCohort sample size for sensitivity analysis in patients younger than 70 years or with minor-to-moderate stroke.

^dSignificant (p < 0.05) associations.

Longitudinal Analysis

The mRS significantly decreased over the 12 months implying that functional status overall improved in the ischemic stroke population (Figure 2, eTable 3, links.lww.com/WNL/D379). Testing the two-way interaction between time and sex on mRS, however, showed that the rate of functional recovery varied by sex (eTable 3). In addition to a main effect with better functional outcome in male patients, male patients also had a significantly steeper 12-month functional improvement rate than female patients independent of age, education, and stroke severity.

Figure 2. Functional Disability Status Over Time Stratified by Sex and Multimorbid Cluster Type.

Panels: (A) Line graph showing mean mRS over time stratified by cluster type and sex. (B) Bar plot showing greater disability (mRS ≥ 3) vs less disability (mRS < 3) over time stratified by cluster type and sex.

There was also a significant 2-way interaction effect between time and cluster 4 indicating that the decline (improvement) in mRS over time in this group was stronger than in cluster 1 (Table 5, Figure 2). When testing the three-way interaction of whether time and cluster group association with mRS varies by sex (timeXclusterXsex), no significant association was found (eTable 4, links.lww.com/WNL/D379).

Table 5.

Mixed-Effect Model of Change in mRS by Cluster From 1 to 12 Months Poststroke

	mRS
Predictors	Estimates	95% CI	p Value
Intercept	−0.024	−0.168 to 0.121	0.747
Time (mo)	−0.032	−0.036 to −0.027	<0.001a
Age	0.025	0.023 to 0.026	<0.001a
NIHSS
NIHSS ≤ 4	Ref	—	—
NIHSS = 5-15	1.701	1.652 to 1.750	<0.001a
NIHSS ≥ 16	3.172	3.047 to 3.297	<0.001a
Sex
Female	Ref	—	—
Male	−0.198	−0.244 to −0.153	<0.001a
Education
Less than elementary school	Ref	—	—
Elementary school	−0.086	−0.140 to −0.033	0.002a
Junior high school	−0.041	−0.112 to 0.031	0.269
Senior high school	−0.035	−0.106 to 0.036	0.335
University college	−0.015	−0.096 to 0.065	0.709
Graduate school	0.005	−0.214 to 0.225	0.964
Clusters
Cluster 1	Ref	—	—
Cluster 2	−0.061	−0.128 to 0.006	0.074
Cluster 3	0.115	0.041 to 0.190	0.002a
Cluster 4	−0.012	−0.091 to 0.067	0.764
Cluster 5	0.042	−0.037 to 0.122	0.296
Interaction time × clusters
Time × cluster 1	Ref	—	—
Time × cluster 2	−0.002	−0.007 to 0.003	0.480
Time × cluster 3	−0.002	−0.008 to 0.004	0.442
Time × cluster 4	−0.008	−0.014 to −0.001	0.015a
Time × cluster 5	−0.006	−0.012 to 0.000	0.064

Random effects: σ² (variance of the residuals indicating the within-subject variance): 0.180; ^τ00 ID (random intercept variance or between-subject variance): 1.018; ^τ11 ID.TP (random slope variance): 0.005; ^ρ01 ID (the random-slope-intercept correlation): -0,297; ICC (intra-class correlation): 0.850; N ID (number of patients): 9,818; Marginal R² (variance of the fixed effects only) / Conditional R² (variance of fixed and random effects): 0.436/ 0.915.

Abbreviations: mRS = modified Rankin Scale; NIHSS = NIH Stroke Scale.

^aSignificant (p < 0.05) associations.

The results also showed that a high and moderate comorbidity burden was associated with a significantly higher mRS score and that individuals with a high burden had a steeper 12-month functional improvement than people without or with mild comorbidity (eTable 5, links.lww.com/WNL/D379). The 3-way interaction between comorbidity, time, and sex (timeXcomorbidityXsex) furthermore indicated that the steeper 12-month improvement rate by comorbidity varied by sex (eTable 6). In the subsequent sex-stratified analysis, we saw that in both sexes, a high comorbidity burden was associated with greater disability (eTable 7). However, only male but not female patients with a high comorbidity burden had a significantly greater functional improvement over a 12-month period compared with those patients without or with mild comorbidity (eTable 7; eFigure 5).

Discussion

In this registry study, specific clusters of multimorbid risk factors were associated with greater poststroke disability in first-ever ischemic stroke survivors who survived at least 6 months poststroke. Patients in cluster 3, characterized by a higher than expected prevalence of diabetes, CKD, and PAD, had greater disability than patients in a healthier risk factor group. Furthermore, the association between multimorbidity cluster and disability remained significant only in male patients, who had a higher overall comorbidity burden, but not in female patients, who had a higher proportion of severe strokes and greater disability. Longitudinally, there was a significant improvement in functional outcomes over 12 months which was more pronounced in male than in female patients, but the interaction effect between functional recovery rate and cluster type did not vary by sex.

Our primary finding is in line with recent evidence showing that a multimorbidity cluster characterized by diabetes and chronic kidney disease is associated with increased functional impairment in patients with first-ever stroke.²⁰ This study showed that this association does not only hold true during the early poststroke phase, that is, the first 2 weeks, but up to 6 months poststroke. There are multiple explanations why a cluster characterized by PAD, diabetes, and CKD might be associated with worse poststroke functional outcomes. One explanation is that this cluster is associated with greater stroke severity, which translates into a higher disability burden. We, however, found only limited evidence that would support this because the proportion of patients with severe strokes (1%–3.8%) did not markedly vary between the clusters. Another explanation is that patients in cluster 3 already suffer from greater prestroke disability, which likely significantly affects poststroke functional impairment at the follow-up assessments. It is known that PAD in combination with diabetes can cause significant disability.^21,22 The effect of prestroke functional level has previously been shown to be an important predictor for adverse functional outcomes and higher levels of care at discharge after stroke.²³ More recent evidence in the Oxford Vascular Study has demonstrated that multimorbidity in patients with ischemic stroke is strongly associated with premorbid disability but is not independently associated with greater stroke severity.²⁴

Although other investigators have described the increased presence of disability in female patients poststroke,²⁵ it was previously unknown whether discrepancies in prestroke multimorbid conditions may explain these sex differences. Although we confirmed the worsening disability overall poststroke in female patients, with worse longitudinal trajectory of recovery of this disability, our findings indicate that comorbidity patterns do not adequately explain the greater poststroke functional impairment in female patients. Thus, it raises the question of what other factors may instead contribute to the worse long-term clinical outcome observed in female patients.

One explanation may be that female patients already have greater prestroke disability and are, therefore, less physically and cognitively resilient after the stroke event,²⁵ although our findings on poststroke disability suggest that if this is the case, prestroke disability in female patients might not be explained by multimorbidity either. While prestroke dependency is a key predictor of poststroke functional status,²⁵ a combination of other factors such as advanced age, stroke severity, marital status, and poststroke depression is as important to bridge the gap.^26-28 The prevalence of mental illnesses such as depression has been shown to be particularly high in elderly female but not in male patients who suffered from less multimorbidity burden in a large population-based study.²⁶ More long-term poststroke disability and poorer outcomes may furthermore be explained by reduced access to rehabilitation therapies for female patients because they are less likely to have a caregiver because of widowhood or social isolation.^29,30

Worse long-term poststroke disability outcomes in female patients may also be explained by a more widespread cerebrovascular lesion burden before and after the acute phase of the ischemic stroke.^31-33 Population-based studies have shown that elderly female patients have significantly greater white matter lesion burden than male patients after adjusting for age, midlife risk factors, and white matter volume.³⁴ Significant differences were also found for white matter lesion progression, which paralleled the observed decline in cognitive and functional outcomes.^31,35-37

This study has several strengths. First, this study is characterized by a large cohort size of 9,818 patients and had multiple monthly follow-up functional assessments with comprehensive risk factor assessment. Second, by including the NIHSS in the analysis, we were able to consider the effect of stroke severity on the effect of multimorbidity on disability. Third, in line with previous evidence, we did not assume that clinical conditions co-occur at random but that stroke risk factors cluster based on specific patterns where they interact with each other, increasing the risk of negative events beyond the sum of the risk of each risk factor or disease.^19,38 Fourth, we analyzed both the cross-sectional and longitudinal associations between multimorbidity clusters, sex, and functional outcome.

This study also has limitations. First, a disability measure before the ischemic stroke event was not available. It is, therefore, not possible to distinguish the effects of multimorbidity clusters on poststroke functional outcome while accounting for prestroke disability. Second, inpatient elements such as clinical care, medications, or in-hospital complications were not accounted for in this study. It is likely that these factors may explain some of the variations in poststroke disability. Third, imaging measures were not included in this analysis, which could explain some of the observed associations: For instance, differences in cerebrovascular disease burden may explain why female patients suffer from greater stroke severity and disability independently of multimorbidity. Fourth, this study only included those 16 risk factors available in the TSR, and it is likely that other comorbid conditions such as psychiatric conditions explain some of the variations in greater poststroke disability and might cluster with those evaluated risk factors. Previous evidence has shown that in addition to the metabolic-kidney cluster, a heart-gastrointestinal-psychiatric cluster was associated with greater physical disability in first-ever ischemic stroke survivors.²⁰ One particularly important risk factor that was not included is dementia, which is known to be strongly associated with multimorbidity.^39-41 Furthermore, because dementia risk and cognitive decline have been associated with cardiometabolic multimorbid clusters, a cluster we identified to be associated with greater poststroke disability,^40,42 and dementia may be more prevalent in female patients,⁴³ it is possible that the nonsignificant associations between cardiometabolic clusters and poststroke disability in female vs male patients in this study may partly be explained by dementia diagnoses that were not available. Fifth, we cannot exclude the possibility that there is more stigma associated with reporting certain risk factors in male vs female patients, and because many of these were self-reported, they might reflect different severity of those risk factors and comorbidities in male vs female patients. For instance, if male patients only reported having risk factors if they were more severe, this could explain why multimorbidity cluster has a significant effect on disability in male but not in female patients. We are also aware that sex-specific risk factors can influence stroke risk, but these were not available.⁴⁴ Social determinants of health, also generally not available, may further explain some of the observed variation in poststroke disability.⁴⁵ We also note that this population is relatively healthier than might be seen in some stroke units because inclusion in this analysis was conditional on surviving for at least 6 months poststroke. Thus, it is possible that the overall level of disability, stroke severity, and multimorbidity was lower in our sample, which might make associations weaker than were the entire sample evaluated poststroke. Finally, the results obtained were from an ethnically Chinese population and may be different in more racially diverse populations.

This study demonstrates the value of considering clustering of risk factors and morbidities, to define an overall multimorbidity pattern, in understanding poststroke disability and its recovery in first-ever ischemic stroke patients. This study also emphasizes that although multimorbidity pattern is highly associated with poststroke functional outcome in male patients, it is not in female patients, despite their having overall greater poststroke disability (and more severe strokes). Further large-sized stroke registry multiethnic studies are needed to confirm these findings and to evaluate other explanations for the sex differences in disability noted after stroke.

Glossary

CKD

chronic kidney disease

mRS

modified Rankin Scale

PAD

peripheral artery disease

TIA

transient ischemic attack

TSR

Taiwan Stroke Registry

Appendix 1. Authors

Name	Location	Contribution
Marco Egle, PhD	National Institute of Neurological Disorders and Stroke, Intramural Research Program, National Institutes of Health, Bethesda, MD	Drafting/revision of the manuscript for content, including medical writing for content; study concept or design; analysis or interpretation of data
Wei-Chun Wang, MD	National Institute of Neurological Disorders and Stroke, Intramural Research Program, National Institutes of Health, Bethesda, MD; Department of Neurology, China Medical University Hospital, Taichung, Taiwan	Drafting/revision of the manuscript for content, including medical writing for content; study concept or design
Yang C. Fann, PhD	National Institute of Neurological Disorders and Stroke, Intramural Research Program, National Institutes of Health, Bethesda, MD	Drafting/revision of the manuscript for content, including medical writing for content; study concept or design; analysis or interpretation of data
Michelle C. Johansen, MD, PhD	Department of Neurology, The Johns Hopkins University School of Medicine, Baltimore, MD	Drafting/revision of the manuscript for content, including medical writing for content; study concept or design
Jiunn-Tay Lee, MD	Department of Neurology, Tri-Service General Hospital, National Defense Medical Center, Taipei, Taiwan	Major role in the acquisition of data
Chung-Hsin Yeh, MD, PhD	Department of Nursing, College of Nursing and Health, Da-Yeh University; Department of Neurology, Yuan Rung Hospital, Changhua, Taiwan	Major role in the acquisition of data; study concept or design
Chih-Hao Jason Lin, MD	Director of Stroke Center, Department of Neurology Stroke Center, Lin Shin Hospital, Taichung	Major role in the acquisition of data
Jiann-Shing Jeng, MD, PhD	Stroke Center and Department of Neurology, National Taiwan University Hospital	Major role in the acquisition of data
Yu Sun, MD, PhD	Department of Neurology, En Chu Kong Hospital, New Taipei City, Taiwan	Major role in the acquisition of data
Li-Ming Lien, MD, PhD	Department of Neurology, Shin Kong Wu Ho-Su Memorial Hospital, Taipei, Taiwan	Major role in the acquisition of data
Rebecca F. Gottesman, MD, PhD	National Institute of Neurological Disorders and Stroke, Intramural Research Program, National Institutes of Health, Bethesda, MD	Drafting/revision of the manuscript for content, including medical writing for content; study concept or design; analysis or interpretation of data

Appendix 2. Coinvestigators

Coinvestigators are listed at links.lww.com/WNL/D380.

Study Funding

R.F. Gottesman, T.C. Fann, W.C. Wang, and M. Egle were supported by the NINDS Intramural Research Program. W.C. Wang is supported by China Medical University for the NIH visiting researcher program. M.C. Johansen receives funding from the NINDS (K23NS112459).

Disclosure

The authors report no relevant disclosures. Go to Neurology.org/N for full disclosures.