Exercise Adherence and Long-Term Quality of Life After Ischemic Stroke

Clinical Question (PICO)

Among adults discharged after ischemic stroke from a single hospital in Guangdong, China, was greater self-reported adherence to functional exercise, compared with lower adherence, associated with better stroke-specific quality of life over 24 months after accounting for disability and measured baseline characteristics?

This is an exposure comparison, not a randomized treatment comparison. Exercise adherence and quality of life were measured at the same visits. No prespecified treatment-effect subgroups or subgroup interaction estimates were reported; the primary model also did not include an adherence-by-time interaction.

Bottom Line

The authors conclude that greater exercise adherence was independently associated with better quality of life on average over two years. The adjusted association was 0.82 additional SS-QoL points per one-point higher exercise-adherence score (95% CI 0.69–0.95; P < 0.001), based on 355 participants and 1,795 observations; 242 participants completed the two-year assessment.

The result supports attention to sustained rehabilitation engagement, but it does not establish that increasing adherence causes this improvement: depression and motivation were unmeasured, the exposure and outcome were concurrent, and follow-up ended after death or recurrent stroke.

Design

Study type: Prospective observational cohort, reported using STROBE; not a clinical trial.
N: 355 enrolled at discharge from 358 initially screened. No randomized arms or allocated group sizes; adherence was modeled as a continuous exposure.
Randomization/blinding: No randomization; no blinded treatment comparison was reported.
Setting: The First Affiliated Hospital of Shantou University Medical College, Guangdong, China; one hospital.
Enrollment: Patients discharged January–August 2021. This is a 2026 publication analyzing an earlier cohort.
Follow-up: Discharge and 1, 3, 6, 12 and 24 months. Maximum scheduled follow-up was 24 months; mean follow-up duration was not reported.
Analysis: Participant-level random-intercept linear mixed-effects model, estimated with restricted maximum likelihood. Time used a natural cubic spline with three degrees of freedom. Five multiply imputed baseline datasets were combined using Rubin's rules; all 355 participants contributed at least one outcome observation.
Adjustment: Baseline demographic, social and clinical variables, including NIHSS and Barthel ADL scores; concurrent time-varying exercise adherence and modified Rankin Scale (mRS) grade.
Main outcome: Total Stroke-Specific Quality of Life Scale (SS-QoL) score, 49 items, range 49–245; higher is better.
Main exposure: Exercise Adherence Questionnaire (EAQ), 14 items scored 1–4, total range 14–56; higher is better adherence.

Population

Inclusion Criteria

Age at least 18 years with CT- or MRI-confirmed ischemic stroke.
Able to cooperate with questionnaires at discharge and during follow-up.
Possession of a mobile phone for follow-up and voluntary participation.

Exclusion Criteria

Discharge because of critical illness or at the family's request.
Transfer to a rehabilitation hospital at discharge.

Subsequent withdrawal criteria included inability to establish contact after the specified repeated calls, recurrent stroke, death, refusal, admission to a rehabilitation hospital, or hospitalization for another serious illness. Consequently, the observed quality-of-life trajectory is not a complete account of all post-stroke outcomes.

Baseline Characteristics (all 355 enrolled)

Mean age 65.90 years (SD 11.92); 226 men (63.7%).
NIHSS median 4 (IQR 2–6) among 297 with recorded scores; 58 baseline NIHSS values (16.34%) were missing and subsequently imputed.
Barthel ADL mean 72.10 (SD 30.58).
EAQ mean 41.95 (SD 9.08); SS-QoL mean 163.07 (SD 32.96).
Prior stroke in 72 (20.3%); family history of stroke in 56 (15.8%).
mRS 4 in 93 (26.2%) and mRS 5 in 28 (7.9%): substantial disability was represented at discharge, although the authors caution about generalization to severely affected patients because of selection and attrition.

Interventions

Higher versus lower adherence: Naturally occurring differences in self-reported functional-exercise adherence; no exercise regimen, drug, device, dose or supervised training schedule was assigned by the investigators.
Measurement: EAQ captured adherence repeatedly, alongside SS-QoL and mRS. A point on this questionnaire is not a minute of exercise, a session count, or a prescribed exercise intensity.
What was not tested: Motivational coaching, family support, telehealth, wearable monitoring and habit-building were proposed by the authors, not randomized interventions evaluated in this cohort.

Outcomes

Main association: SS-QoL over 24 months

Exposure or covariate	Adjusted difference in SS-QoL points, β (95% CI)	P value
EAQ, per one-point higher score	0.82 (0.69 to 0.95)	<0.001
Baseline NIHSS, per point	−0.80 (−1.22 to −0.38)	<0.001
Baseline Barthel ADL, per point	0.17 (0.10 to 0.23)	<0.001
Family history of stroke, yes versus no	−4.50 (−7.62 to −1.37)	0.005
Prior stroke, yes versus no	−3.16 (−6.02 to −0.31)	0.030

These are unstandardized regression coefficients, not hazard ratios, odds ratios or relative risks. As a calculated rescaling of the fitted association, ten points higher EAQ corresponds to 8.2 points higher SS-QoL (95% CI 6.9–9.5), assuming the model's linear exposure relationship. This is not a predicted benefit from an adherence intervention.

Other reported associations and temporal findings

Compared with concurrent mRS 0, adjusted SS-QoL differences were −3.71 for mRS 1 (95% CI −6.05 to −1.37; P = 0.002), −15.12 for mRS 2 (−18.38 to −11.85), −22.00 for mRS 3 (−25.84 to −18.17), −43.13 for mRS 4 (−47.52 to −38.74), and −70.71 for mRS 5 (−78.02 to −63.39); P < 0.001 for mRS 2–5. These are disability-category associations, not adherence subgroup effects.
The modeled quality-of-life trajectory rose rapidly early after discharge, peaked at six months at a predicted mean of 210.73 (95% CI 206.35–215.13), declined between six and twelve months, then rose slowly. This trajectory does not identify when an adherence intervention would be most effective.
No adherence-by-time interaction was fitted. The 0.82 coefficient is an average association across visits, not evidence of an equal causal effect at every recovery stage.

Missing-data sensitivity analysis

The delta-adjusted pattern-mixture analysis additionally imputed missing QoL. At Δ = 0, its adherence coefficient was 0.78 (95% CI 0.64–0.92; P < 0.001), versus 0.82 in the primary available-outcome model. Coefficients remained positive with confidence intervals excluding zero across the tested dropout-outcome shifts of Δ = ±10 points. This probes specified departures from the missing-data assumption; it does not eliminate confounding or make QoL after death observable.

Adverse Events / Safety

No randomized safety comparison, exercise-attributable adverse-event analysis, symptomatic intracranial hemorrhage comparison or bleeding endpoint was reported. The participant-flow figure records deaths and recurrent strokes as reasons for withdrawal, not as treatment effects. Summing its interval counts gives 36 deaths and 15 recurrent strokes; these are descriptive attrition counts, with no adherence-stratified risk estimate. Overall, 113 of 355 participants did not complete the final assessment (31.8%).

Figures

Participant flow from 358 screened patients to 242 completing two-year follow-up — *Figure 1.* Flow-chart of survivors' enrolment and follow-up. The cohort declined from 355 at discharge to 338, 302, 285, 273 and 242 at the scheduled follow-up visits. Death, recurrence, loss of contact, refusal and rehabilitation admission contributed to attrition.

Criticisms

Association is not treatment efficacy. Exercise adherence and QoL were measured concurrently. Better recovery or mood could enable both higher adherence and higher QoL; prospective enrollment alone does not settle causal direction.
Important confounding remains. Depression, motivation and other psychological factors were not adjusted for. The authors acknowledge this; caregiver category and economic status are not complete measures of available support or rehabilitation access.
Attrition is clinically informative. Non-completers had greater baseline severity and lower support. Baseline differences rule against completely random dropout, but do not prove missing-at-random conditional on the model. Death and recurrence also raise an estimand problem beyond ordinary missing questionnaire values.
The coefficient blends different contrasts. The model did not separately estimate within-person adherence changes versus between-person differences. A random intercept does not by itself turn the pooled adherence coefficient into the effect of changing one patient's behavior.
Disability adjustment is not causally neutral. Concurrent mRS might partly lie on a pathway between exercise and QoL. Without an explicit causal model, the adjusted coefficient should not be called either the total exercise effect or a proven direct effect.
Limited generalizability and intervention specificity. One hospital, mobile-phone access and questionnaire capacity were required; rehabilitation transfers were excluded or withdrawn. The paper supplies no tested exercise dose, adherence target or intervention-specific number needed to treat.
No established timing or subgroup benefit. No adherence-by-time interaction, subgroup treatment effects or formal prospective LMM sample-size calculation was provided. Narrow confidence intervals describe precision conditional on the model, not immunity to systematic bias.

Funding

Supported by the Guangdong Provincial Nurses' Association, grant gdshsxh2021b004. The authors declare no conflicts of interest. The funding statement does not specify the funder's role in design, analysis or publication. No trial registration identifier is reported; this was an observational cohort, not a randomized trial.

The paper

Authors. Jin Ziyuan, Zhang Hong, Liu Zhili, Wu Yanchun and Lin Hongchun, as displayed in the PMC full text.
Title. Association Between Exercise Adherence and Long-Term Quality of Life in Ischemic Stroke Survivors: A 2-Year Prospective Cohort Study.
Journal. Nursing Open, volume 13, issue 11, e70965.
Year. 2026; cohort enrollment in 2021.
DOI. 10.1002/nop2.70965
PMCID. PMC13722777

Deep Dive — click to expand

What this is

[F] This 2026 report follows a 2021 ischemic-stroke cohort across six scheduled assessments and links a stroke-specific adherence questionnaire with a patient-centered outcome. [F] The main signal is β = 0.82 SS-QoL points per EAQ point, with 31.8% final-visit attrition and no randomized adherence intervention. [I] Its best use is to sharpen long-term follow-up and formulate a testable care strategy, not to promise a quantified recovery gain.

The commentary labels facts from the source [F], editorial inferences [I], hypotheses requiring testing [H], and missing evidence [M].

1. Shadow Audit

[F] Follow-up withdrawal included recurrent stroke and death; the flow diagram contains 36 deaths and 15 recurrent strokes when interval counts are summed. [F] QoL was modeled using 1,795 available observations, with 242 final-visit completers. [I] The reassuring average association can therefore coexist with clinically important outcomes that stop further QoL measurement: this is not evidence that adherent survivors avoid death or recurrence.

[F] The paper's dropout sensitivity analysis retained a positive coefficient under QoL shifts of ±10. [I] That robustness concerns a specified missing-QoL model, not a joint survival-and-quality-of-life endpoint. [M] A survivor-specific estimand, a composite accounting for death, or a joint model would clarify what population-level recovery claim is justified. [H] Reanalyzing the existing cohort with death and recurrence handled explicitly could change the apparent long-term trajectory without reversing the observed questionnaire association.

2. Inversion Engine

[F] The fitted coefficient is 0.82, with lower confidence limit 0.69. [I] On that coefficient's scale, an omitted-variable or reverse-causation bias of +0.82 SS-QoL points per EAQ point could account for the entire point estimate; this is an arithmetic scenario, not a formal sensitivity bound. [F] Depression and motivation were unmeasured, and exposure and outcome were assessed concurrently. [H] The opposite causal interpretation—people who feel and function better adhere more—could generate a positive coefficient even if raising adherence alone had little effect.

[H] The cheapest discriminating reanalysis is to test whether earlier adherence predicts later QoL after conditioning on prior QoL, and to separate within-person from between-person adherence. [I] Persistence would strengthen temporal plausibility, not prove causation; disappearance would weaken a coaching-as-treatment interpretation. [M] The available report does not give those estimates.

3. Second-Order Catalyst

[F] The adjusted QoL trajectory peaked at six months, then declined to twelve months; the authors proposed sustained follow-up and habit support. [I] A community stroke-nursing or outpatient rehabilitation service is therefore a more plausible early adopter than an acute reperfusion team. [H] A reversible service pilot could add a brief adherence-and-barriers review to existing later follow-ups, record patient-selected goals and route worsening function, mood or support problems for assessment.

[I] The second-order risk is turning a prognostic questionnaire into a patient-compliance scorecard: low adherence may identify inadequate access or disability, not unwillingness. [H] Audit who receives extra assistance and who is lost to follow-up before expanding the program. [M] This cohort does not establish that six months is the optimal intervention time or that additional calls improve QoL.

4. Asymmetric Leverage

[F] The model translates a ten-point EAQ difference into an associated 8.2-point SS-QoL difference; baseline mean EAQ was 41.95 on a scale ending at 56. [I] That makes the relationship interpretable, but not a feasible ten-point improvement target for every survivor. [F] Disability differences were also large—for example, mRS 4 versus 0 corresponded to −43.13 SS-QoL points after adjustment. [I] Chasing an adherence score while neglecting assistive care, mood or rehabilitation access would mistake a modifiable marker for the whole recovery system.

[H] The low-cost leverage point is to ask what prevents the already-prescribed, individually appropriate activity and address one practical barrier, then track both behavior and patient-reported well-being. [M] Cost-effectiveness, attainable adherence changes and treatment-mediated QoL gains remain unknown; no population payoff or number needed to treat can be calculated from this coefficient.

5. Paradigm Destroyer

[F] The study is longitudinal, but its main time-varying variables are concurrent and no adherence-by-time interaction was fitted. [I] It challenges two reflexes: treating discharge exercise advice as a completed care task, and treating a prospective association as proof of a behavior-change treatment effect. [H] The strongest skeptical explanation is that recovery, mood and social resources jointly drive both questionnaires; adjustment for concurrent disability does not remove that possibility.

[H] A proportionate two-sentence protocol proposal is: “At existing post-stroke follow-ups, document exercise barriers and help patients access an individualized rehabilitation plan. Measure function and quality of life as well as adherence, and do not promise the cohort's 0.82-point coefficient as a causal benefit.” [I] This is an editorial care-process proposal, not a guideline recommendation or a protocol tested by this study.

MVP — Minimum Viable Proof

[F] Higher adherence remained associated with higher QoL after measured covariate adjustment: β = 0.82, 95% CI 0.69–0.95. [I] That is sufficient to treat adherence difficulties as a useful follow-up conversation, not sufficient to mandate a particular adherence intervention. [H] A stronger practice-changing test would randomize feasible adherence support and demonstrate improved patient-centered QoL, while retaining severely affected survivors and accounting for death and recurrence.

Best Combination

[F] Earlier work by Hou and colleagues, first published online in 2017, also reported longitudinal exercise–QoL associations, using exercise frequency/duration and the generic SF-12 rather than this study's EAQ and SS-QoL. [F] A longitudinal study by Wang and colleagues, published online in 2023, followed 75 patients for six months and considered both rehabilitation adherence and depressive symptoms. These contextual findings were checked against the prior studies' indexed abstracts, not independently reviewed full texts.

[I] Read together, the studies support a recurring association across different measures while highlighting why mood cannot be left out of the next analysis. [H] The best analytical sequence is Shadow Audit → Inversion Engine: define whose outcomes are represented after attrition, then test temporal direction and psychological confounding before translating the signal into an intervention effect. [M] Repetition across observational cohorts still does not provide a randomized answer to the adherence-support question.

Overvalue Warning

[I] Do not turn β = 0.82 into a treatment promise. Ten more EAQ points are not ten more sessions, and the calculated 8.2-point QoL association is not a demonstrated within-person benefit.
[I] Do not turn sensitivity robustness into universal applicability. The tested ±10-point missing-QoL shifts do not remove unmeasured confounding, recover QoL after death, or establish the same relationship in excluded rehabilitation transfers, very old adults or survivors unable to complete questionnaires.

Educational use notice. The Time Is Brain Review is dedicated to medical education. Figures from published articles are embedded for teaching purposes, with attribution to the original publication. We do not claim copyright over any embedded figure; all rights remain with the original publisher and authors under the article's published license. If you are a rights holder and would like a figure removed, please contact the editorial team. See our editorial notice for the full policy.