This systematic review and meta-analysis was conducted using Preferred Reporting Items for Systematic Reviews and Meta-Analysis (PRISMA) guidelines (Page et al., 2021) and registered with PROSPERO (Registration ID: CRD42025638467), an international prospective registry for systematic reviews.

Two authors (HL and JG) independently searched Web of Science, PubMed, MEDLINE, SPORT-Discus, Cochrane Library, Scopus, and PsycINFO from inception to August 5, 2025; Studies were searched in the electronic databases using the following key terms combined by Boolean logic (“AND”, “OR”): (“Nordic walking” OR “Nordic pole walking” OR “pole walking”) AND (“Cognitive function” OR “cognition” OR “Cognitive performance”). A secondary search strategy was also used, which involved a manual search in the reference lists of eligible studies (i.e., citation tracking). Searches were limited to publications in English. Any disagreements arising during this process were resolved through discussion between the two authors (HL and JG), with additional input provided by a third author (JN). Detailed search strategies are provided in Supplementary Table 1.

All included studies must be published articles. The inclusion criteria were carried out according to the PICOS principle: (1) Population: All the participants included in this study were at least 60 years old. None of them had used any drugs known to significantly affect cognitive function, or had discontinued such drugs for more than 4 weeks, or had indicated their actual medication status at baseline to ensure the comparability of drug exposure between the two groups; (2) Interventions: the interventions used were NW only or NW combined with other interventions. When NW was used in combination with other interventions, the control group was supposed to receive other interventions alone to ensure that the observed changes were caused by NW; (3) Comparisons: Each group is characterized as either active (e.g., interventions involving physical activities other than Nordic walking, NW) or inactive (e.g., non-intervention, or routine treatment for the diseases the subjects suffer from); (4) Outcomes: Outcome measures reflecting cognitive function were employed. (5) Study design: The study employed randomized controlled trials or randomized crossover trial designs. Articles with the following conditions will be excluded: (1) did not investigate cognitive function outcomes or provide specific data of outcome measures (e.g., reporting only p-values without means/SDs); (2) review papers, conference abstracts, and articles; (3) those with duplicate publications; (4) non-English publications.

The process of data extraction was conducted independently by two authors (HL and JG) according to the Cochrane Collaboration Handbook. The extracted information of the publications included: study (authors, year), participants (age, sex, physical condition), grouping and sample size, interventions (type, frequency, number of sessions, duration of each session, duration of intervention), auxiliary means (e.g., Nordic walking poles or Hiking pole, etc.), and outcome measures. Any outcome measures on which the two authors disagreed were discussed with the other two authors (JN and KZ) until a consensus was achieved. For each study, extract the mean and standard deviation (SD) of the post-intervention indicator results. For studies that do not report changes in results before and after, or those presenting results in the form of “Mean ± SE/SEM (Standard Error/Standard Error of the Mean)”, use the following formula for calculation (Borenstein et al., 2013):

When the full-text article data were presented only in a figure format, WebPlotDigitizer (Ankit Rohatgi, 2019, V.4.2; WebPlotDigitizer, Pacifica, CA, USA) was used to extract the data from the figures. In the absence of any relevant data, the first author or the corresponding author of the article would be contacted via email to obtain the required data.

The quality of the included studies was independently evaluated by two authors (HL and JG) according to the principles of the Physiotherapy Evidence Database (PEDro). The PEDro scale, specifically designed to assess the methodological quality of randomized controlled trials in physical interventions, is highly suitable for evaluating the studies in this research (de Morton, 2009; Cashin and McAuley, 2020). This scale examines crucial elements such as randomization, blinding, and allocation concealment, which are essential for ensuring the internal validity of the studies included in this systematic review and meta-analysis (Cashin and McAuley, 2020).

Specifically, the PEDro scale consists of 11 items, to which we were required to respond with “no” or “yes”. For each “no” response, a value of 0 was assigned, and for each “yes” response, a value of 1 was assigned. The total score for each study ranged from 0 to 11. Since blinding (especially of participants and investigators) is difficult to implement in exercise intervention trials (Sherrington et al., 2010), the methodological quality classification of each article was adjusted, taking into account the eligibility criteria as previously described [sum scores: ≥6 (“high quality, low risk of bias”); scores: 4–5 (“acceptable quality, moderate risk of bias”); scores: ≤ 3 (“low quality, high risk of bias”)].

The quality of the evidence was also independently appraised by two authors (HL and JG) based on the Grading of Recommendations Assessment, Development and Evaluation (GRADE) criteria. The GRADE criteria characterize the evidence in terms of study limitations, imprecision, inconsistency, indirectness, and publication bias. In cases where the two authors disagreed on any score, a third author (JN or KZ) was consulted for discussion until a consensus was reached.

Meta-analysis was carried out using Stata/MP 17.0 (STATA Corp, College Station, TX, USA) and R 4.2.0 (R Core Team, R Foundation for Statistical Computing, Vienna, Austria). Given the diverse measurement units of outcome measures across studies, such as time, score, number of stimuli, etc., a random-effects model was employed to calculate Hedges’ g and the 95% confidence interval (CI), which served as the indicator of the effect size for the difference in pre-post changes between the intervention and control groups (Fritz et al., 2012). The effect sizes were categorized as follows: trivial (Hedges’ g < 0.2), small (0.2 ≤ Hedges’ g < 0.5), moderate (0.5 ≤ Hedges’ g < 0.8), or large (Hedges’ g ≥ 0.8) (Cohen, 2013). Additionally, prediction intervals were computed to reflect heterogeneity in comparison with confidence intervals. The between-study variance was estimated using the restricted maximum likelihood estimator with Hartung-Knapp adjustment (Veroniki et al., 2016). Statistical heterogeneity was evaluated by means of the heterogeneity chi-squared (χ²) and I² values. The degree of heterogeneity was interpreted in accordance with the guidelines of the Cochrane Collaboration: 0%–40% might not be of significance; 30%–60% may indicate moderate heterogeneity; 50%–90% may represent substantial heterogeneity; and 75%–100% implies considerable heterogeneity (Higgins, 2003). In the event of substantial or considerable heterogeneity (I² > 50%), subgroup analyses were conducted to explore the impact of study characteristics (e.g., the physical condition of the participants). Additionally, we performed meta-regression to explore the dose-response relationship of NW on cognitive function. The intervention duration, session number, and total time (i.e., session number × session duration) were used as the effect moderators. Subsequently, sensitivity analyses were performed to assess the stability of the pooled estimates and to determine whether any study influenced the overall effect size (Veroniki et al., 2016). Moreover, publication bias was assessed through the generation of funnel plots and the conduct of Egger’s test. If significant asymmetry was detected, the Trim and Fill method was utilized to adjust for publication bias (Duval and Tweedie, 2000).

The results of the study selection process are summarized in a PRISMA flowchart (Figure 1). The study selection commenced with the identification of 336 records through database searches, which were distributed among Web of Science (n = 84), PubMed (n = 64), MEDLINE (n = 29), SPORT-Discus (n = 47), Cochrane Library (n = 46), Scopus (n = 61), and PsycINFO (n = 3). Additionally, two further studies were identified through citation searching. After removing 174 duplicate records, a total of 162 records were available for the screening phase. During the title and abstract screening stage, 143 records were excluded. Subsequently, a full-text eligibility assessment was conducted on 18 articles, and 10 studies were excluded. Eventually, 8 studies were included in the meta-analysis.

This meta-analysis included 23 effect sizes from 8 studies. These effect sizes encompassed multiple domains of cognitive ability, including global cognitive function (Passos-Monteiro et al., 2020; Miyazaki et al., 2022; Angiolillo et al., 2023; Ploydang et al., 2023; Haas et al., 2024), memory function (Angiolillo et al., 2023), executive function (Miyazaki et al., 2022; Angiolillo et al., 2023; Kettinen et al., 2023; Ploydang et al., 2023), perceptual abilities (Guszkowska et al., 2022; Angiolillo et al., 2023), information processing speed (Guszkowska et al., 2022; Kettinen et al., 2023), and attention (Ebersbach et al., 2014; Guszkowska et al., 2022; Angiolillo et al., 2023). Specific cognitive tests administered and their corresponding domain classifications are detailed in Supplementary Table 3.

The overall effect size for cognitive function was moderate and statistically significant [Hedges’ g = 0.56, 95% CI (0.29, 0.84), PI (−0.57, 1.70), p < 0.01, Figure 2], with substantial heterogeneity (I² = 70.75%, p < 0.01). Meta-regression analysis indicated no significant association between effect size and the number of intervention sessions (b = 0.042, p = 0.62, Figure 3A), while it revealed a significant positive association between effect size and the total intervention duration (b = 0.0003, p = 0.049, Figure 3B).

For specific cognitive abilities, the effect size for executive function was large and significant [Hedges’ g = 0.89, 95% CI (0.27, 1.50), p = 0.01], with substantial heterogeneity (I² = 73.04%, p < 0.01). The effect size for memory function was moderate but not statistically significant [Hedges’ g = 0.77, 95% CI (−0.86, 2.39), p = 0.11], with non-significant heterogeneity (I² = 0.00%, p = 0.68). The effect size for global cognition was moderate but not statistically significant [Hedges’ g = 0.73, 95% CI (−0.08, 1.53), p = 0.07], with substantial heterogeneity (I² = 80.46%, p < 0.01). The effect size for attention was small and not statistically significant [Hedges’ g = 0.20, 95% CI (−0.48, 0.88), p = 0.41], with moderate heterogeneity (I² = 40.81%, p = 0.17). The effect size for information processing was trivial and not statistically significant [Hedges’ g = 0.18, 95% CI (−2.57, 2.93), p = 0.56], with non-significant heterogeneity (I² = 7.50%, p = 0.30). The effect size for perceptual ability was trivial and not statistically significant [Hedges’ g = −0.07, 95% CI (−3.70, 3.56), p = 0.85], with non-significant heterogeneity (I² = 20.61%, p = 0.26). The funnel plot (Figure 4) and Egger’s test (t = 2.47, p = 0.02) indicated a potential risk of publication bias, but the Trim and Fill method for sensitive analysis showed that the pooled effect size (Hedges’ g = 0.56, p < 0.01) was robust.

To examine characteristics that may contribute to the impact of Nordic walking on cognitive function, we conducted a subgroup analysis using a random effects model on studies that reported large effects involving cognitive function (Table 2). The analysis was based on the following study characteristics: participants’ health status (i.e., healthy population vs. diseased population) and control type (i.e., active vs. inactive).

The subgroup analysis based on participants’ health status revealed a moderate and statistically significant effect size for the disease group [Hedges’ g = 0.70, 95% CI (0.04, 1.35), p = 0.04], while the healthy group showed a large but non-significant effect size [Hedges’ g = 1.06, 95% CI (−1.28, 3.40), p = 0.19]. The analysis by control type indicated that NW had a large but non-significant effect size compared to the active intervention control group [Hedges’ g = 1.06, 95% CI (−1.28, 3.40), p = 0.19]. In contrast, NW showed a moderate and statistically significant effect size when compared to the inactive control group [Hedges’ g = 0.56, 95% CI (0.20, 0.92), p < 0.01].

Of the eight included studies, one study was with a randomized crossover controlled design. To evaluate the impact of this study on the results, we performed a sensitivity analysis by excluding it. The overall effect size remained significant after excluding it [Hedges’ g = 0.49, 95% CI (0.23, 0.75), p < 0.01], and was consistent with the previous overall effect size [Hedges’ g = 0.56, 95% CI (0.29, 0.84), p < 0.01]. It was noted that the heterogeneity between studies assessing executive function decreased significantly after excluding this study (before exclusion: I² = 73.04%, p ≤ 0.01; after exclusion: I² = 8.42%, p = 0.27), suggesting the difference in the duration of the intervention contributed to the inter study heterogeneity. However, the effect size for executive function also remained significant [Hedges’ g = 0.56, 95% CI (0.06, 1.06), p = 0.04]. These results confirm the robustness of our findings. Furthermore, by eliminating one effect size at a time, we explore the robustness of the pooled results across different cognitive classifications. The sensitivity analysis showed that the results in terms of attention, executive function, information processing, memory and perception were relatively robust (Supplementary Figure 1). In terms of global cognition alone, after excluding a study where the control group engaged in deep-water exercise, the effect size of the meta-analysis showed a significant change. Before exclusion [Hedges’ g = 0.73, 95% CI (−0.08, 1.53), p = 0.07], after exclusion [Hedges’ g = 0.94, 95% CI (0.23, 1.66), p = 0.02]. The above results indicate that the differences in the control conditions might be the reason for the relatively sensitive global cognitive effect. Therefore, these results should be treated with caution.

The methodological assessment details of each included study are presented in Table 3. The overall quality rating of all included studies is high. The overall mean score is 7.13 ± 1.25; high-quality studies account for 100% (8/8). The level of certainty of the evidence was downgraded by one level due to limitations in study bias (Supplementary Table 2).

The results of this work still need to be taken with caution. First, the number of included studies was relatively limited (n = 8), which may potentially affect the statistical power of our meta-analysis and meta-regression results. Additionally, due to the limited number of studies, subgroup analyses to characterize the influences of protocol settings of NW (e.g., appropriate number of sessions, the frequency of intervention, and Intervention intensity) cannot be completed. Furthermore, interpretations regarding underlying mechanisms remain to be validated in future studies due to the small sample size and the limited number of existing studies. Lastly, although sensitivity analyses showed robust results, heterogeneity among the studies may have a potential influence on the interpretation of outcome measures. Nevertheless, the knowledge obtained from this work will help inform the appropriate design of intervention protocols of NW.