Effects of minocycline on health-related quality of life in patients with acute ischaemic stroke in China: a prespecified secondary analysis of the EMPHASIS trial

Summary

Background

Health-related quality of life (HRQoL) is increasingly recognised as a critical outcome after acute ischaemic stroke (AIS). Minocycline may have neuroprotective potential in AIS; however, its effects on patient-reported HRQoL remain uncertain. Therefore, we aimed to evaluate the effect of minocycline on HRQoL at 90 days.

Methods

This prespecified secondary analysis was based on data from the EMPHASIS trial, a parallel randomised controlled trial of minocycline in AIS at 58 centres in China from May 19, 2023, to May 20, 2024. Eligible participants were aged 18–80 years, were enrolled within 72 h of acute ischaemic stroke onset and had a National Institutes of Health Stroke Scale score of 4–25, and a pre-stroke modified Rankin Scale score of 1 or less. Participants were randomly assigned (1:1) to receive minocycline or placebo plus standard care. HRQoL was assessed using the EuroQol five-dimension three-level questionnaire (EQ-5D-3L) at 90 days. The primary outcome was EQ-5D-3L utility score; secondary outcomes were EuroQol visual analogue scale (EQ-VAS), a favourable health state defined as EQ-5D-3L utility score ≥0.75, and EQ-5D-3L domain responses. Primary analyses were conducted in the prespecified modified intention-to-treat population using linear, log-binomial, and ordinal logistic models for continuous, binary, and ordinal outcomes. Multivariable-adjusted and per-protocol sensitivity analyses were also performed. This trial was registered with ClinicalTrials.gov (NCT05836740).

Findings

Of 1724 participants, 1697 were included in the final analyses. The primary analysis showed no significant between-group difference in EQ-5D-3L utility score (p = 0.224). After multivariable adjustment, the EQ-5D-3L utility score was higher in minocycline group than in placebo group (p = 0.049). The proportion of patients achieving a favourable health state was higher in the minocycline group than in the placebo group (73.4% vs 68.9%; risk ratio 1.07, 95% confidence interval 1.003–1.13; p = 0.039), with improvements in mobility (p = 0.043), self-care (p = 0.030), and usual activities (p = 0.031). EQ-VAS did not differ significantly between two groups (p = 0.420).

Interpretation

Minocycline may improve health state and EQ-5D-3L domains at 90 days, supporting further evaluation as adjunctive therapy and incorporation of patient-reported outcomes into stroke trials.

Funding

National Natural Science Foundation of China, The National Key Research and Development Program of China, Beijing Health Union Cardio-cerebrovascular Disease Prevention and Treatment Foundation, Noncommunicable Chronic Diseases—National Science and Technology Major Project, New Cornerstone Science Foundation, Beijing Scholar Program, Beijing Municipal Science & Technology Commission, and Capital's Funds for Health Improvement and Research.

Research in context.

Evidence before this study

We searched PubMed from database inception to April 15, 2026, for randomised trials assessing the efficacy of minocycline in patients with acute ischaemic stroke, using the terms “minocycline” AND “stroke”, with no language restrictions. We also screened the reference lists of relevant articles. We identified seven randomised trial reports, including five studies of oral minocycline and two studies of intravenous minocycline. The evidence base was heterogeneous in design and quality, comprising three small open-label or evaluator-blinded trials, one single-blinded placebo-controlled trial, two multicentre double-blind placebo-controlled trials, and one pilot intravenous trial that also included haemorrhagic stroke. Outcomes in these trials were largely restricted to clinician-rated neurological and functional measures, including the National Institutes of Health Stroke Scale, modified Rankin Scale, Barthel Index, vascular events, and haemorrhagic transformation. Early small trials suggested possible benefit of minocycline on neurological or functional recovery, whereas later blinded and pilot studies yielded neutral results and highlighted imprecision and between-study heterogeneity. One trial-level meta-analysis suggested a possible improvement in handicap-free survival, but with substantial heterogeneity. None of the identified randomised trials systematically assessed health-related quality of life (HRQoL) using a preference-based patient-reported outcome measure such as the EuroQol five-dimension questionnaire (EQ-5D). Thus, before this study, evidence for minocycline in acute ischaemic stroke was focused almost entirely on clinician-rated outcomes, with little direct evidence on patient-perceived health status.

Added value of this study

This study provides, to our knowledge, the first randomised-trial evidence on health-related quality of life after acute ischaemic stroke in patients treated with minocycline, assessed using the EuroQol five-dimension three-level questionnaire (EQ-5D-3L) in a large, multicentre, randomised, double-blind, placebo-controlled setting. Unlike previous minocycline trials, which focused primarily on neurological impairment, disability, or vascular outcomes, this prespecified secondary analysis evaluated patient-reported health status, including EQ-5D-3L utility score, EuroQol visual analogue scale (EQ-VAS), favourable health state, and domain-specific outcomes. Our findings extend the evidence base beyond conventional clinician-rated recovery by showing that minocycline was associated with a higher likelihood of achieving a favourable health state and with better outcomes in mobility, self-care, and usual activities, indicating potential benefit in aspects of recovery directly meaningful to patients.

Implications of all the available evidence

Previous randomised trials have shown inconsistent effects of minocycline on conventional neurological and functional outcomes after acute stroke. Our findings add a patient-centred dimension to this evidence. In this prespecified secondary analysis, minocycline did not significantly improve continuous EQ-5D-3L utility or EQ-VAS scores at 90 days. However, patients assigned to minocycline were more likely to achieve the prespecified favourable health-state threshold and reported more favourable outcomes in mobility, self-care, and usual activities. These findings suggest that any potential effect of minocycline on post-stroke HRQoL might be more apparent in the distribution of favourable health states and specific functional domains than in overall utility scores. Further studies are needed to confirm these findings, establish their clinical relevance, and identify patients who might derive the greatest benefit.

Introduction

Acute ischaemic stroke (AIS) remains a leading cause of death and long-term disability worldwide.¹ Despite advances in reperfusion therapies, a substantial proportion of survivors experience persistent functional limitations, emotional distress, reduced independence, and impaired social participation, resulting in a profound decline in health-related quality of life (HRQoL).² Beyond individual suffering, poor post-stroke HRQoL imposes considerable burdens on families, caregivers, and healthcare systems. However, large randomised controlled trials (RCTs) in AIS have traditionally focused on neurological and functional endpoints, such as mortality and the modified Rankin Scale (mRS), while patient-centred outcomes have been relatively underexplored. One of the key pathophysiological mechanisms underlying AIS and its poor prognosis is neuroinflammation.³ Experimental and clinical studies have suggested that neuroprotective and anti-inflammatory agents may attenuate secondary brain injury and improve neurological outcome.^4,5 Clinical studies further demonstrated that elevated inflammatory activity has been consistently associated with worse functional recovery and higher disability after stroke.⁶ Nevertheless, whether anti-inflammatory neuroprotective therapy can improve the overall quality of life in patients with AIS remains unknown.

The EuroQol five-dimension three-level questionnaire (EQ-5D-3L) is a widely used, preference-based instrument for assessing HRQoL. It captures patients' self-reported health across five domains: mobility, self-care, usual activities, pain or discomfort, and anxiety or depression, and converts these responses into a single utility score anchored at 1 for full health and 0 for death, with negative values representing health states considered worse than death.⁷ Patient-reported outcomes, such as HRQoL, complement traditional clinician-rated measures by providing direct insight into recovery and treatment benefit from the patient's perspective, including perceived health, daily functioning, independence, and emotional wellbeing.⁸ Prior studies have shown that AIS is associated with substantial and sustained reductions in EQ-5D utility scores, even among patients with mild to moderate functional impairment.^8,9 Some studies with limited sample size have shown that acute stroke reperfusion treatment improved functional outcomes without proportionate gains in HRQoL, underscoring the limitations of relying solely on conventional endpoints.^10,11

To date, few studies have evaluated whether anti-inflammatory neuroprotective therapies can improve quality of life in AIS population. Therefore, we conducted a prespecified secondary analysis of the Efficacy and Safety of Minocycline in Patients with Acute Ischaemic Stroke (EMPHASIS) trial to investigate whether minocycline also confers benefits in terms of patient-reported HRQoL. The EMPHASIS trial was a large, multicentre, parallel-group, randomised controlled study designed to evaluate the efficacy and safety of using minocycline in the early phase of AIS. Minocycline has been extensively studied for its multi-targeted anti-inflammatory and neuroprotective effects including modulating neuroinflammation, oxidative stress and microglial activation, particularly in neurological conditions. The primary results of EMPHASIS trial have been published, which demonstrated that minocycline significantly increased the proportion of patients achieving favourable functional outcomes (mRS 0–1) at 90 days compared with placebo, with a safety profile comparable to that of placebo.¹² In this secondary analysis, we aimed to determine the impact of minocycline on overall utility, domain-specific health states, and clinically meaningful thresholds of quality of life in patients with AIS, using the EQ-5D-3L instrument.

Methods

Study design and patients

This study was a prespecified secondary analysis of the EMPHASIS trial, which is a prospective, multicentre, randomised, double-blind, placebo-controlled, parallel-group trial designed to evaluate the safety and efficacy of minocycline in patients with acute ischaemic stroke. Between May 19, 2023, and May 20, 2024, the trial enrolled patients aged 18–80 years within 72 h of acute ischaemic stroke onset. The diagnosis was confirmed by brain computed tomography (CT) or magnetic resonance imaging (MRI). Eligible patients had a National Institutes of Health Stroke Scale (NIHSS) score of 4–25, a score of 1 or less on the level-of-consciousness item (item 1a) of the NIHSS, and either a first-ever stroke or a pre-stroke mRS score of 1 or less. Written informed consent was obtained from all patients or their legally authorised representatives. Key exclusion criteria included allergy or resistance to tetracycline antibiotics, prior tetracycline use within the past 7 days and severe hepatic or renal insufficiency. A full list of inclusion and exclusion criteria is detailed in the EMPHASIS trial protocol.¹³ The sex of the patients was determined from their identity card, which indicated either male or female. Race and ethnicity data were not collected as they were not prespecified in the trial protocol. This trial was registered with ClinicalTrials.gov (NCT05836740) on April 19, 2023, and is now completed. This Article was reported in accordance with the Consolidated Standards of Reporting Trials (CONSORT) 2025 statement. The full English-language EMPHASIS trial protocol is provided in the Supplementary Materials.

Randomisation and masking

Patients were enrolled at each study site by investigators who had no access to the randomisation sequence and remained masked to treatment allocation throughout the trial. Eligible patients were randomly assigned (1:1) to receive minocycline or matching placebo, in addition to guideline-recommended standard care. Randomisation was stratified by study site and implemented in fixed blocks of four using an allocation sequence generated by an independent statistical team through a centralised computer system. Minocycline and placebo capsules were identical in appearance and taste and supplied in indistinguishable containers labelled only with unique randomisation codes. The containers were distributed to each site in blocks of four and assigned sequentially to eligible participants without investigators having access to the allocation sequence, thereby maintaining allocation concealment. Participants, treating clinicians, investigators, and outcome assessors remained masked to treatment allocation throughout the trial.

Procedure

Participants received an oral loading dose of 200 mg minocycline or matching placebo within 30 min of randomisation, followed by 100 mg every 12 h for 4 days. Study medication was administered via a nasogastric tube in patients with dysphagia. Both groups received guideline-recommended standard care for acute ischaemic stroke, including intravenous thrombolysis or endovascular thrombectomy when clinically indicated. Participants were assessed at baseline, day 6 (±1), and day 90 (±7); detailed trial procedures have been reported previously and are provided in the trial protocol.¹³

Study outcomes

Health-related quality of life was assessed at the 90-day follow-up, through direct patient interviews conducted by trained neurologists using the EQ-5D-3L Chinese (Simplified) version.¹⁴ Permission for its use and publication was obtained from the EuroQol Research Foundation.

The EQ-5D-3L instrument consists of the EQ-5D descriptive system and EQ visual analogue scale (EQ-VAS). The descriptive system comprises five dimensions including mobility, self-care, usual activities, pain/discomfort, and anxiety/depression. Each dimension has three response levels (level 1 = no problems, level 2 = some/moderate problems, level 3 = extreme problems) indicating the severity of problems experienced by patients. Responses across the five dimensions define 243 unique health states, which are converted using country-specific value sets into a single utility score anchored at 1 for full health and 0 for death, with negative values representing states considered worse than death.^14,15 To better reflect the health state preferences of the Chinese population, the Chinese value set for the EQ-5D-3L was used to calculate utility score.¹⁶ In accordance with conventional scoring, patients who died by the 90-day follow-up were assigned a utility value of 0, consistent with the preference-based scoring framework. These patients were excluded from the EQ-VAS analysis. The EQ-VAS records self-rated health on a scale from 0 to 100, with higher scores indicating better perceived health. Among survivors at 90 days, we further analysed the distribution of responses across the five EQ-5D-3L dimensions. The instrument has demonstrated construct and known-groups validity and acceptable test-retest reliability in stroke survivors.¹⁷ Moreover, we defined “a favourable health state” as EQ-5D-3L utility score ≥0.75 based on previous studies and the preliminary results of our pilot study.^8,18

The primary outcome of this secondary analysis was the EQ-5D-3L utility score at 90 days. Secondary outcomes were the proportion of patients achieving a favourable health state, defined as an EQ-5D-3L utility score of at least 0.75; the EQ-VAS score; and the distribution of responses across the five EQ-5D-3L dimensions. Subgroup analyses of the EQ-5D-3L outcomes were exploratory.

Trial oversight

The Steering Committee provided scientific and strategic oversight and was responsible for the trial design, implementation, quality assurance, and publication. The Executive Committee reviewed trial progress, collected available blinded data, guided major operational decisions, and oversaw site management and auditing. The independent Data Safety Monitoring Board regularly monitored trial conduct, ethics, and participant safety. The Clinical Event Adjudication Committee, comprising external experts, independently reviewed clinical outcome events and relevant neuroimaging documentation. The statistical and data management centre was responsible for data management and statistical procedures. All committees were established before participant recruitment, and each study centre completed training and certification before enrolment.

Ethics

The EMPHASIS trial received ethical approval from the Ethics Committee of Beijing Tiantan Hospital, Capital Medical University (KY2023-007-04). This secondary analysis was conducted using data collected under the approved trial protocol and was therefore covered by the original ethics approval. Written informed consent, including the collection and analysis of EQ-5D-3L data, was obtained from all participants or their legally authorised representatives before enrolment. No separate informed consent was obtained for this prespecified secondary analysis. The trial was conducted in accordance with the Declaration of Helsinki.

Statistical analysis

For sample size calculation of this study, we assumed mean utility scores of 0.80 in the minocycline group and 0.75 in the placebo group, with a common standard deviation of 0.25 in both groups based on the pilot study conducted by our research team. Given a two-sided α of 0.05, 90% power and 10% attrition, a total of 1169 participants were required to detect a significant difference in EQ-5D-3L utility score between the minocycline and the placebo groups. The primary EMPHASIS trial required 1672 participants to assess the proportion achieving a modified Rankin Scale score of 0–1 at 90 days; therefore, its planned sample size exceeded the minimum requirement for this secondary analysis.

The prespecified modified intention-to-treat (mITT) population comprised all randomised participants with ascertainable 90-day HRQoL outcomes, analysed according to their assigned treatment group. Complete data were used, and missing data were not imputed. For analysis, deaths before 90 days were assigned a utility score of 0 and classified as not achieving the favourable health state; analyses of EQ-VAS and the five EQ-5D-3L dimensions were restricted to survivors. Demographic and clinical characteristics are presented as n (%) for categorical variables, mean (standard deviation [SD]) for normally distributed continuous variables, and median (interquartile range [IQR]) for skewed continuous variables.

The EQ-5D-3L utility score and EQ-VAS score were summarised as medians with IQRs. Linear regression was used to estimate absolute between-group differences in these continuous outcomes, with treatment effects reported as β coefficients and corresponding 95% confidence intervals (CIs).

For the binary outcome of a favourable health state, defined as an EQ-5D-3L utility score ≥0.75, log-binomial regression was used to directly estimate the risk ratio (RR) and corresponding 95% CI. The RR was reported because it provides a direct and clinically interpretable comparison of probabilities, whereas the odds ratio may overestimate the magnitude of the relative association when the outcome is common.

Responses to each of the five EQ-5D-3L dimensions were presented as counts and percentages across the three ordinal levels. Ordinal logistic regression was used to account for the ordered nature of these responses, with treatment effects reported as common odds ratios (cORs) and corresponding 95% CIs.

Multivariable analyses used linear, log-binomial, and ordinal logistic mixed-effects models for continuous, binary, and ordinal outcomes, respectively, with pooled study centre as a random intercept to account for within-centre clustering and prespecified covariates as fixed effects. If the log-binomial model failed to converge, a modified Poisson regression mixed-effects model with robust standard errors was used. Models were adjusted for age, sex, body-mass index (BMI), hypertension, diabetes mellitus, dyslipidaemia, previous ischaemic stroke, atrial fibrillation, ever smoker, alcohol consumption, baseline NIHSS score, high-sensitivity C-reactive protein (hs-CRP), reperfusion therapy, onset-to-treatment time, and Trial of Org 10172 in Acute Stroke Treatment (TOAST) subtype. As a post-hoc model diagnostic, multicollinearity among the covariates included in the adjusted models was assessed using variance inflation factors (VIFs), with a VIF of 5 or greater considered indicative of potentially meaningful multicollinearity.

As a prespecified sensitivity analysis, baseline characteristics and HRQoL outcomes were analysed in the per-protocol population, defined as participants who received the allocated treatment without major protocol violations, using the same models as in the primary analyses. For the primary outcome of EQ-5D-3L utility score at 90 days, post-hoc analyses included multiple imputation of missing outcomes, analyses restricted to 90-day survivors, sequential mixed-effects models with progressive covariate adjustment, and an additional model adjusted only for baseline NIHSS score.

Exploratory subgroup analyses were conducted to assess whether the effect of minocycline on HRQoL differed across clinically relevant patient subgroups derived from the prespecified subgroup framework of the EMPHASIS trial. Treatment-by-subgroup interaction terms were included to assess heterogeneity in treatment effects. No adjustment was made for multiple comparisons; therefore, these findings were considered hypothesis-generating.

All hypothesis tests were two-sided, with a significance level of 0.05. Statistical analyses were performed using SAS version 9.4 (SAS Institute, Cary, NC, USA).

Role of the funding source

The funders of the study had no role in study design, data collection, data formal analysis, data interpretation, or writing of the report.

Results

A total of 1724 patients were enrolled and randomly assigned across 58 study centres in China (Table S1). Patients were randomly allocated, in a 1:1 ratio, to either minocycline (n = 862) or placebo groups (n = 862). Four patients withdrew their consent, 19 were lost to follow-up, and 4 declined to complete the 90-day EQ-5D-3L assessment. Consequently, 90-day EQ-5D-3L data were available for 1697 patients (846 in the minocycline group and 851 in the placebo group). Of these, 34 patients (14 in the minocycline group and 20 in the placebo group) died, and their EQ-5D-3L scores were assigned a value of 0. Thus, 1663 survivors (832 in the minocycline group and 831 in the placebo group) had complete 90-day EQ-VAS and EQ-5D-3L domain scores (Fig. 1). In total, 75 (4.4%) patients with major protocol deviations were excluded from the per-protocol analysis: 36 (4.3%) in the minocycline group and 39 (4.6%) in the placebo group (Figure S1).

Fig. 1.

Trial profile. EQ-5D-3L, EuroQol five-dimension three-level questionnaire. 1697 patients with available 90-day EQ-5D-3L data were included in this analysis.

Baseline characteristics were well balanced between the two treatment groups (Table 1). The mean age was 63.7 (SD 10.2) years, and 1131 (66.6%) participants were men. The median baseline NIHSS score was 5 (IQR 4–7) in both groups. In the per-protocol population, baseline characteristics were also similarly balanced between the minocycline and placebo groups (Table S2).

Table 1.

Baseline characteristics.

Baseline characteristics	Overall (n = 1697)	Minocycline (n = 846)	Placebo (n = 851)
Age, years	63.7 (10.2)	64.0 (10.2)	63.4 (10.1)
Sex
Male	1131 (66.6%)	562 (66.4%)	569 (66.9%)
Female	566 (33.4%)	284 (33.6%)	282 (33.1%)
Body-mass index, kg/m2	24.9 (3.5)	24.8 (3.7)	24.9 (3.4)
Medical history
Hypertension	1111 (65.5%)	545 (64.4%)	566 (66.5%)
Diabetes mellitus	558 (32.9%)	281 (33.2%)	277 (32.5%)
Dyslipidaemia	730 (43.0%)	370 (43.7%)	360 (42.3%)
Previous ischaemic stroke	491 (28.9%)	235 (27.8%)	256 (30.1%)
Coronary heart disease	205 (12.1%)	91 (10.8%)	114 (13.4%)
Heart failure	11 (0.6%)	5 (0.6%)	6 (0.7%)
Atrial fibrillation	83 (4.9%)	44 (5.2%)	39 (4.6%)
Ever smoker	671 (39.5%)	319 (37.7%)	352 (41.4%)
Alcohol consumption	452 (26.6%)	214 (25.3%)	238 (28.0%)
Medication within 1 month before stroke
Antiplatelet agent	227 (13.4%)	105 (12.4%)	122 (14.3%)
Anticoagulant agent	24 (1.4%)	11 (1.3%)	13 (1.5%)
Statins	196 (11.5%)	100 (11.8%)	96 (11.3%)
Antibiotics	6 (0.4%)	2 (0.2%)	4 (0.5%)
Neuroprotective agent	28 (1.6%)	14 (1.7%)	14 (1.6%)
NIHSS score at baseline	5.0 (4.0–7.0)	5.0 (4.0–7.0)	5.0 (4.0–7.0)
NIHSS score at baseline (classification)
4–5	997 (58.8%)	482 (57.0%)	515 (60.5%)
6–10	564 (33.2%)	296 (35.0%)	268 (31.5%)
≥11	136 (8.0%)	68 (8.0%)	68 (8.0%)
Pre-stroke mRS score
0	1460 (86.0%)	725 (85.7%)	735 (86.4%)
1	233 (13.7%)	120 (14.2%)	113 (13.3%)
≥2	4 (0.2%)	1 (0.1%)	3 (0.4%)
hs-CRP at baseline, mg/La	2.0 (0.9–4.7)	2.0 (0.8–4.8)	2.0 (0.9–4.5)
Stroke subtypes (TOAST classification)
Large-artery atherosclerosis	774 (45.6%)	380 (44.9%)	394 (46.3%)
Cardioembolism	90 (5.3%)	48 (5.7%)	42 (4.9%)
Small-artery occlusion	676 (39.8%)	334 (39.5%)	342 (40.2%)
Other determined aetiology	52 (3.1%)	30 (3.5%)	22 (2.6%)
Undetermined aetiology	105 (6.2%)	54 (6.4%)	51 (6.0%)
Reperfusion therapy	237 (14.0%)	108 (12.8%)	129 (15.2%)
Time from stroke onset to treatment
≤24 h	350 (20.6%)	176 (20.8%)	174 (20.4%)
>24–48 h	756 (44.5%)	364 (43.0%)	392 (46.1%)
>48 h	591 (34.8%)	306 (36.2%)	285 (33.5%)

Data are presented as mean (SD), median (IQR), or n (%), as appropriate. The overall column includes all participants in the modified intention-to-treat population. Percentages may not total 100 because of rounding.

NIHSS, National Institutes of Health Stroke Scale; mRS, modified Rankin Scale; hs-CRP, high-sensitivity C-reactive protein; TOAST, Trial of Org 10172 in Acute Stroke Treatment; SD, standard deviation; IQR, interquartile range.

^ahs-CRP data are missing for 94 and 95 patients in the minocycline and placebo groups, respectively.

EQ-5D-3L utility score

At 90 days, the median (IQR) EQ-5D-3L utility score was 0.96 (0.70–0.96) in the minocycline group and 0.89 (0.68–0.96) in the placebo group (treatment effect 0.01, 95% CI −0.01 to 0.04; p = 0.224, Table 2).

Table 2.

90-Day EQ-5D-3L Outcomes (in the modified intention-to-treat population).

Outcome	Minocycline (n = 846)	Placebo (n = 851)	Treatment effect (95% CI)	p value
Primary outcome
EQ-5D-3L utility score	0.96 (0.70, 0.96)	0.89 (0.68, 0.96)	β 0.01 (−0.01 to 0.04)	0.224
Secondary outcome
EQ-VASa	85.0 (70.0, 90.0)	80.0 (70.0, 90.0)	β 0.67 (−0.97 to 2.31)	0.420
EQ-5D-3L utility score ≥0.75	621 (73.4%)	586 (68.9%)	RR 1.07 (1.003–1.13)	0.039

Data are n (%) or median (IQR).

p values less than 0.05 are shown in bold.

CI, confidence interval; EQ-5D-3L, EuroQol five-dimension three-level questionnaire; EQ-VAS, EuroQol visual analogue scale; β, regression coefficients; RR, risk ratio; IQR, interquartile range.

^aEQ-VAS was analysed among survivors with available EQ-VAS data (n = 832 and n = 831). Continuous outcomes were analysed using linear regression and the binary outcome using log-binomial regression.

In the per-protocol population, findings were consistent with the primary analysis: the median EQ-5D-3L utility score was 0.96 (0.71–0.96) in the minocycline group and 0.89 (0.68–0.96) in the placebo group (β 0.02, 95% CI −0.01 to 0.04; p = 0.184; Table S3).

EQ-VAS

EQ-VAS scores also tended to favour the minocycline group with a median (IQR) of 85.0 (70.0–90.0) compared with 80.0 (70.0–90.0) in the placebo group (β 0.67, 95% CI −0.97 to 2.31; p = 0.420); however, this difference did not reach statistical significance (Table 2). Analyses restricted to the per-protocol population produced comparable findings (Table S3).

Favourable health state

A greater proportion of patients in the minocycline group achieved a favourable health-related quality of life, defined as an EQ-5D-3L utility score ≥0.75, compared with those in the placebo group (621 [73.4%] vs 586 [68.9%]; RR 1.07, 95% CI 1.003–1.13; p = 0.039; Table 2). This pattern persisted in the per-protocol population, in which 601 (74.2%) patients in the minocycline group and 563 (69.3%) in the placebo group achieved the favourable threshold (RR 1.07, 95% CI 1.01–1.14; p = 0.030; Table S3), indicating concordance between mITT and per-protocol analyses.

Domains of EQ-5D-3L

Patients who received minocycline showed better functional outcomes in mobility, self-care and usual activities at 90 days, compared with those received placebo (Table 3). A higher proportion of patients in the minocycline group reported no mobility problems (level 1: 659 [79.2%] vs 627 [75.5%]; cOR 1.27, 95% CI 1.01–1.59; p = 0.043), no limitations in self-care (613 [73.7%] vs 577 [69.4%]; cOR 1.26, 95% CI 1.02–1.56; p = 0.030), and no restrictions in usual activities (502 [60.3%] vs 465 [56.0%]; cOR 1.23, 95% CI 1.02–1.49; p = 0.031). For the domains of pain/discomfort and anxiety/depression, no statistically significant differences were observed between the minocycline and the placebo groups. Similar results were observed in the per-protocol analysis (Table S4).

Table 3.

Distribution of responses across EQ-5D-3L domains at 90 days (in the modified intention-to-treat population).

Domain	Minocycline (n = 832)	Placebo (n = 831)	cOR (95% CI)	p value
Mobility			1.27 (1.01–1.59)	0.043
Level 1	659 (79.2%)	627 (75.5%)
Level 2	129 (15.5%)	169 (20.3%)
Level 3	44 (5.3%)	35 (4.2%)
Self-care			1.26 (1.02–1.56)	0.030
Level 1	613 (73.7%)	577 (69.4%)
Level 2	155 (18.6%)	199 (23.9%)
Level 3	64 (7.7%)	55 (6.6%)
Usual activities			1.23 (1.02–1.49)	0.031
Level 1	502 (60.3%)	465 (56.0%)
Level 2	239 (28.7%)	286 (34.4%)
Level 3	91 (10.9%)	80 (9.6%)
Pain/discomfort			0.85 (0.67–1.07)	0.162
Level 1	656 (78.8%)	630 (75.8%)
Level 2	162 (19.5%)	191 (23.0%)
Level 3	14 (1.7%)	10 (1.2%)
Anxiety/depression			1.09 (0.84–1.40)	0.523
Level 1	695 (83.5%)	685 (82.4%)
Level 2	123 (14.8%)	135 (16.2%)
Level 3	14 (1.7%)	11 (1.3%)

Data are presented as n (%). Percentages may not total 100 because of rounding. Differences in the three response levels of each EQ-5D-3L domain were analysed using ordinal logistic regression. p values less than 0.05 are shown in bold.

EQ-5D-3L, EuroQol five-dimension three-level questionnaire; cOR, common odds ratio; CI, confidence interval.

Multivariable-adjusted analyses

After multivariable adjustment, minocycline was associated with a modestly higher EQ-5D-3L utility score (β 0.02, 95% CI 0.0001–0.05; p = 0.049), a greater likelihood of achieving a utility score of 0.75 or higher (adjusted RR 1.08, 95% CI 1.03–1.14; p = 0.002), and more favourable mobility (adjusted cOR 1.43, 95% CI 1.23–1.65; p < 0.001) and self-care levels (adjusted cOR 1.38, 95% CI 1.14–1.67; p < 0.001). Differences in EQ-VAS, usual activities, pain/discomfort, and anxiety/depression were not significant (Table 4). In a post-hoc assessment, all VIFs were below 2 (range 1.034–1.663), indicating no meaningful multicollinearity among the covariates included in the adjusted models (Table S5).

Table 4.

Multivariable-adjusted EQ-5D-3L outcomes at 90 days.

Outcome	Adjusted Model
	Treatment effect (95% CI)	p value
Primary outcome
EQ-5D-3L utility score	β 0.02 (0.0001–0.05)	0.049
Secondary outcome
EQ-VAS	β 0.90 (−0.63 to 2.43)	0.251
EQ-5D-3L utility score ≥0.75	RR 1.08 (1.03–1.14)	0.002
EQ-5D-3L Domain
Mobility	cOR 1.43 (1.23–1.65)	<0.001
Self-care	cOR 1.38 (1.14–1.67)	<0.001
Usual activities	cOR 1.26 (0.96–1.65)	0.100
Pain/discomfort	cOR 0.86 (0.71–1.05)	0.131
Anxiety/depression	cOR 1.15 (0.95–1.39)	0.148

Estimates were adjusted for age, sex, body mass index, hypertension, diabetes mellitus, dyslipidaemia, previous ischaemic stroke, atrial fibrillation, ever smoker, alcohol consumption, baseline National Institutes of Health Stroke Scale score, baseline high-sensitivity C-reactive protein level, receipt of reperfusion therapy, onset-to-treatment time, and Trial of Org 10172 in Acute Stroke Treatment classification. Continuous outcomes were analysed using multivariable linear mixed-effects models, the binary outcome using a modified Poisson mixed-effects model with robust standard errors, and ordinal outcomes using multivariable ordinal logistic mixed-effects models. All models incorporated a random intercept for pooled study centre. p values less than 0.05 are shown in bold.

CI, confidence interval; EQ-5D-3L, EuroQol five-dimension three-level questionnaire; EQ-VAS, EuroQol visual analogue scale; β, regression coefficients; RR, risk ratio; cOR, common odds ratio.

Subgroup analyses

Exploratory subgroup analyses based on prespecified baseline characteristics are shown in Fig. 2. A significant interaction was observed across TOAST classifications (p for interaction = 0.001). Minocycline was more effective in patients with stroke of undetermined aetiology (β 0.15, 95% CI 0.06–0.24; p = 0.002) but was associated with lower EQ-5D-3L utility scores in those with cardioembolic stroke (β −0.15, 95% CI −0.29 to −0.01; p = 0.039). No significant benefit was observed in the other stroke subtypes. However, the effect of minocycline appeared consistent across different levels of other personal and clinical characteristics (all p for interaction >0.05). Nevertheless, a greater treatment effect of minocycline was observed in patients with alcohol consumption (β 0.05, 95% CI 0.001–0.09; p = 0.043), although there was no significant evidence of interaction (p for interaction = 0.113).

Fig. 2.

Subgroup analyses on the primary outcome. Subgroup analyses were performed according to prespecified baseline characteristics. Squares represent β coefficients for the treatment effect (minocycline vs placebo) on EQ-5D-3L utility score, with horizontal lines indicating 95% CIs. EQ-5D-3L, EuroQol five-dimension three-level questionnaire; β, regression coefficient; CI, confidence interval; TOAST, Trial of Org 10172 in Acute Stroke Treatment; LAA, large-artery atherosclerosis; CE, cardioembolism; SAO, small-artery occlusion; SOE, stroke of other determined aetiology; SUE, stroke of undetermined aetiology.

The subgroup analysis of favourable health state is shown in Fig. 3. The effect of minocycline on favourable health state appears consistent across all prespecified characteristics (p for interaction >0.05), but it might be more effective in patients with stroke of undetermined aetiology (RR 1.42, 95% CI 1.09–1.84; p = 0.008), patients with hypertension (RR 1.09, 95% CI 1.01–1.18; p = 0.029), patients with dyslipidaemia (RR 1.13, 95% CI 1.03–1.23; p = 0.012), and patients reporting alcohol consumption (RR 1.13, 95% CI 1.00–1.27; p = 0.049).

Fig. 3.

Subgroup analyses on favourable health state (EQ-5D-3L utility ≥0.75) at 90 days. Subgroup analyses were performed according to prespecified baseline characteristics. Squares represent the estimated risk ratios for achieving a favourable health state (EQ-5D-3L utility ≥0.75) with minocycline vs placebo, and horizontal lines indicate 95% CIs. EQ-5D-3L, EuroQol five-dimension three-level questionnaire; RR, risk ratio; CI, confidence interval; TOAST, Trial of Org 10172 in Acute Stroke Treatment; LAA, large-artery atherosclerosis; CE, cardioembolism; SAO, small-artery occlusion; SOE, stroke of other determined aetiology; SUE, stroke of undetermined aetiology.

Additionally, prespecified subgroup analyses of EQ-5D-3L domains, including mobility, self-care, and usual activities, are shown in Figure S2. Across these three domains, the effects of minocycline were generally consistent across prespecified characteristics, except for usual activities, for which treatment effects differed by hypertension status. For mobility, greater benefit was observed in patients with hypertension (cOR 1.44, 95% CI 1.09–1.91; p = 0.010); dyslipidaemia (cOR 1.51, 95% CI 1.06–2.14; p = 0.022) and alcohol consumption (cOR 1.58, 95% CI 1.00–2.48; p = 0.048; Figure S2A). For self-care, greater benefit was observed in male patients (cOR 1.39, 95% CI 1.06–1.81; p = 0.018), patients with hypertension (cOR 1.41, 95% CI 1.09–1.83; p = 0.008), alcohol consumption (cOR 1.53, 95% CI 1.02–2.31; p = 0.042), and stroke of undetermined aetiology (cOR 3.51, 95% CI 1.42–8.65; p = 0.007; Figure S2B). For usual activities, treatment effects differed by hypertension status (p for interaction = 0.036), with minocycline associated with more favourable outcomes in patients with hypertension (cOR 1.41, 95% CI 1.12–1.79; p = 0.004). Favourable associations were also observed in patients with diabetes (cOR 1.42, 95% CI 1.02–1.97; p = 0.037), those reporting alcohol consumption (cOR 1.63, 95% CI 1.13–2.36; p = 0.009), and those with stroke of undetermined aetiology (cOR 3.26, 95% CI 1.45–7.34; p = 0.004; Figure S2C).

Sensitivity analyses

In post-hoc sensitivity analyses, multiple imputation showed a higher 90-day EQ-5D-3L utility score with minocycline (β 0.045, 95% CI 0.002–0.088; Table S6), whereas the analysis restricted to participants alive at 90 days showed no significant difference (β 0.01, 95% CI −0.01 to 0.03; p = 0.397; Table S7). In sequential mixed-effects models progressively adjusted for the prespecified demographic and clinical covariates, the treatment-effect estimate showed the largest change after adjustment for baseline NIHSS score and remained significant in the fully adjusted model. However, adjustment for baseline NIHSS score alone did not result in a statistically significant treatment effect (Table S8).

Discussion

In this prespecified secondary analysis of the EMPHASIS trial, minocycline was associated with better health-related quality of life at 90 days after acute ischaemic stroke, as measured with the EQ-5D-3L. Although the overall difference in utility score did not reach statistical significance, patients receiving minocycline were more likely to achieve a favourable health state (EQ-5D-3L utility score ≥0.75) and to report no problems in mobility, self-care, and usual activities compared with those receiving placebo. After multivariable adjustment, the association with the EQ-5D-3L utility score reached statistical significance. Similar patterns were generally observed in the per-protocol population and exploratory subgroup analyses, providing additional support for the overall findings.

We selected the EQ-5D-3L as the primary patient-reported outcome measure to provide a more comprehensive understanding of treatment effects, integrating both objective neurological recovery and patient’ lived experience of health. Several factors may account for this finding. First, the distribution of EQ-5D-3L utility scores was markedly skewed towards higher values, reflecting a ceiling effect in this population with relatively mild stroke severity (median NIHSS 5), which may have reduced the sensitivity to detect between-group differences. Second, EQ-5D-3L utility is a composite index derived from five domains and may be less responsive to domain-specific improvements, particularly when gains are concentrated in physical function rather than across all dimensions. Third, inter-individual variability in patient-reported outcomes, influenced by psychosocial and contextual factors, may further attenuate detectable differences at the group level. To facilitate clinical interpretation of the continuous utility measure, we prespecified an EQ-5D-3L utility value of 0.75 or higher as a favourable health state. This pragmatic threshold corresponded to the anticipated mean utility value in the placebo group used for sample-size estimation. It was also broadly anchored to an mRS score of 2, which represents slight disability with preserved independence. In a pooled analysis of more than 20,000 patients with acute stroke, mRS 2 corresponded to utility values of approximately 0.72–0.74, compared with 0.54–0.57 for mRS 3.¹⁸ A recent multicentre study similarly estimated utility values of 0.74–0.76 for mRS 2 and 0.62–0.63 for mRS 3.¹⁹ Thus, the threshold of 0.75 approximately represents a health state characterised by no more than slight disability and retained functional independence, while distinguishing it from the lower utility typically associated with moderate disability and a need for assistance. The between-group difference in continuous utility scores was small in magnitude, whereas minocycline was associated with a higher probability of reaching this favourable threshold, indicating a distributional shift towards clinically meaningful recovery. This pattern may relate to the non-linear association between utility values and perceived health status, because even a small increase in utility near the threshold can correspond to a meaningful improvement in patient-perceived recovery. Furthermore, after adjustment for prespecified confounding factors, statistically significant differences also emerged in EQ-5D-3L utility scores, indicating that minocycline might confer measurable benefits in health-related quality of life among patients recovering from acute ischaemic stroke. Our findings are broadly consistent with previous studies of minocycline in acute ischaemic stroke, which have demonstrated improvements in neurological and functional outcomes, including NIHSS and mRS scores, but have rarely incorporated patient-reported quality-of-life measures.^20,21 Earlier trials and meta-analyses have suggested that minocycline exerts neuroprotective effects through anti-inflammatory and anti-apoptotic mechanisms, contributing to improved functional recovery.²² The present analysis extends these observations by demonstrating that such benefits may translate into patient-perceived improvements in daily functioning and overall health status.

Notably, minocycline showed a consistent pattern of benefit in domains related to physical autonomy, including mobility, self-care, and usual activities, while effects on pain or discomfort and anxiety or depression remained largely similar between groups. These findings were aligned with the improvements observed in clinician-rated outcomes in the main trial and support the interpretation that minocycline mainly promotes motor recovery and functional independence. The domain-specific pattern also suggests that the treatment effect was concentrated in aspects of recovery related to physical function, with a comparatively smaller influence on affective symptoms and bodily discomfort. This divergence highlights that the neurobiological mechanisms through which minocycline promotes functional recovery may not extend to domains driven by mood, emotional resilience, or chronic pain perception.

This domain-specific pattern of benefit is biologically plausible. In acute ischaemic stroke, minocycline has been shown to attenuate microglial activation, suppress pro-inflammatory cytokine release, and modulate excitotoxic and oxidative injury. These actions could help preserve peri-infarct tissue and support neural reorganisation in motor and association cortices, which may in turn contribute to recovery in domains related to physical function and independence.²³ These processes might contribute to recovery of corticospinal tract function and support the restoration of basic activities of daily living, which could help explain the observed improvements in mobility, self-care, and usual activities.

By contrast, post-stroke anxiety, depression, and pain are likely to arise from not only by focal brain injury and neuroinflammation, but also by widespread alterations in fronto-limbic circuitry, pre-existing vulnerability, psychosocial stressors, and longer-term maladaptive plasticity.^24,25 These broader and more complex influences might be less readily modified by a short-term intervention targeting inflammation and microglial activation in the hyperacute phase. Moreover, mood and pain symptoms often evolve over a longer timeframe than the early recovery period assessed in this trial.^26,27 Meaningful improvement in these domains may therefore require additional approaches that address emotional regulation, coping, and chronic pain processing. Previous stroke cohorts have shown a similar separation between functional recovery and affective outcomes, which supports the consistency of this pattern.²⁸

Across subgroup analyses of EQ-5D-3L utility, favourable health state, and domain-specific outcomes, the overall direction of treatment effect was generally consistent with the primary findings. Minocycline was associated with better patient-reported health status across most demographic and clinical subgroups. Larger effect estimates were seen in patients with hypertension and in those who reported alcohol consumption, although the interaction tests indicated variation of uncertain strength. These findings are biologically credible, because hypertension and alcohol exposure are both linked to enhanced neuroinflammatory responses, endothelial injury, and microglial activation, factors that could increase susceptibility to ischaemic damage and shape the response to anti-inflammatory treatment.^29,30

Variation across TOAST subtypes was also evident, with the clearest signal of benefit seen in patients with stroke of undetermined aetiology. This subgroup may include patients with mixed or covert mechanisms, including small-vessel disease or inflammatory arteriopathy, in whom neuroinflammatory pathways could play a more prominent part. Patients with cardioembolic stroke showed a weaker association with benefit, which suggests that treatment response may vary across different pathophysiological mechanisms. Previous studies of neuroprotective strategies have also described variation in treatment effects across stroke subtypes, supporting the view that underlying aetiology may influence therapeutic efficacy.

From a clinical perspective, these findings suggest that minocycline could serve as an adjunctive treatment to improve quality of life after stroke alongside established reperfusion and secondary prevention strategies. EQ-5D-3L adds further value by capturing aspects of recovery that extend beyond traditional endpoints such as the mRS and thereby provides a more patient-centred assessment of treatment benefit. The subgroup findings also offer hypotheses about populations that may derive greater benefit, and these signals merit further evaluation in future studies.

Several limitations should be acknowledged. First, the EQ-5D-3L was assessed only at the 90-day follow-up, without a baseline assessment. Consequently, we were unable to directly confirm the comparability of baseline HRQoL between the treatment groups or evaluate within-patient changes in HRQoL over time. Second, although we observed signals of benefit, the study lacked detailed mechanistic data, including inflammatory biomarkers, advanced neuroimaging, and other biological measures, limiting our ability to directly elucidate the pathways underlying the observed improvements. Third, although participants were recruited from multiple centres across China, the absence of race and ethnicity data limits assessment of the representativeness and generalisability of the findings across populations with different sociocultural backgrounds. Fourth, participants generally had relatively mild stroke, which may limit the applicability of the findings to patients with more severe stroke. Finally, missing HRQoL data were not imputed, and selection bias cannot be excluded if participants without available 90-day assessments differed systematically from those included.

Future studies should incorporate baseline and repeated longitudinal assessments using instruments with greater discriminatory capacity, such as the EuroQol five-dimension five-level questionnaire (EQ-5D-5L) combined with stroke-specific measures, would enable evaluation of within-patient changes, recovery trajectories, and quality-adjusted survival while reducing the influence of ceiling effects. Recruitment of more diverse populations across different regions and levels of stroke severity would improve generalisability. Integrating inflammatory biomarkers and advanced neuroimaging could also help clarify the mechanisms underlying any patient-perceived benefit and validate the exploratory treatment-effect heterogeneity observed across clinical subgroups.

In this prespecified secondary analysis of the EMPHASIS trial, minocycline showed signals of modest patient-perceived benefit after acute ischaemic stroke, particularly in aspects of daily functioning. These findings support further evaluation of minocycline and underscore the value of incorporating patient-reported outcomes into trials of emerging therapies for acute ischaemic stroke. However, given the absence of a statistically significant effect on the primary continuous HRQoL outcome and the secondary nature of this analysis, the findings should be considered exploratory. Further trials incorporating baseline and longitudinal HRQoL assessments are needed to confirm the potential benefit and establish its clinical relevance.

Contributors

YiW and LG conceptualised the study and were responsible for project administration. QY, YLi, YLu, MZ, DZ, CL, LR, XZho, HQ, and XL conducted the investigation. LL, XZha, DW, and YoW supervised the study. QY developed the methodology and prepared the visualisations. LW, TA, and YP performed the formal analyses, and LW and TA were responsible for the software. QY and YLi drafted the original manuscript, and LG and QY reviewed and edited it. YiW and LG provided resources and acquired funding. LG and QY accessed and verified the underlying data. All authors read and approved the final version of the manuscript.

Data sharing statement

Data from the EMPHASIS study are not currently publicly available. Following submission of a formal written request, including a study protocol and analysis plan, to the corresponding author and approval by the Steering Committee, the statistical code and outputs may be made available to facilitate reproduction of the study findings.

Declaration of interests

We declare no competing interests.