Boston NamingTest performance in mild cognitive impairment: a meta-analysis

Abstract

Background

Naming impairment is frequently observed in cognitive decline, but evidence in mild cognitive impairment (MCI) remains inconsistent. This study aimed to systematically evaluate naming ability in individuals with MCI.

Methods

A systematic search of PubMed, Web of Science, EMBASE, and Cochrane Library (2015–2025) identified 20 cross-sectional studies comparing Boston Naming Test (BNT) performance between MCI patients and healthy controls. To ensure methodological comparability, only studies using the BNT were included. Standardized mean differences (SMDs) were calculated using random-effects models. Heterogeneity, publication bias, and potential moderators (age, sex, education, global cognition) were assessed.

Results

The meta-analysis included 1,306 MCI patients and 3,877 controls. MCI patients performed significantly worse on naming tasks than controls (SMD = -0.841, 95% CI: -1.001 to -0.675; p < 0.001), with consistent findings across BNT versions. Meta-regression revealed no moderating effects of age, sex, or education, while global cognitive function significantly affected naming performance. Substantial heterogeneity was observed.

Conclusion

Reduced performance on the Boston Naming Test is common in individuals with MCI and appears to be partially related to global cognitive status rather than demographic factors. Future studies should explore subtype-specific patterns and longitudinal associations to clarify the clinical significance of naming impairments in MCI.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12883-026-04773-0.

Introduction

Mild cognitive impairment (MCI) is widely recognized as a transitional stage between normal aging and Alzheimer’s disease (AD) [1]. Substantial evidence indicates that individuals with MCI are at significantly increased risk of progressing to dementia, with reported annual conversion rates ranging from 2% to 31% and an average of approximately 10.2% per year [2]. These findings highlight MCI as a critical clinical window for the early detection of dementia and the initiation of timely interventions.

The primary feature of MCI is a mild decline in one or more cognitive domains, including memory, executive function, language, visuospatial ability, calculation, and comprehension and judgment, while the ability to perform activities of daily living remains relatively preserved [3]. Among these domains, language impairment has been identified as an early and frequently observed feature of cognitive decline. Specifically, naming impairment, which is difficulty in retrieving and producing the correct names for objects or concepts, is a manifestation of language dysfunction that plays a crucial role in evaluating cognitive status and is frequently reported in individuals with MCI [4, 5]. Although often subtle, naming impairment can be reliably detected using standardized assessments, with the Boston Naming Test (BNT) being the most widely used tool in both clinical and research settings, especially in MCI studies [6, 7].

Importantly, naming performance reflects not only language ability but also the integrity of multiple cognitive processes. Naming is a complex cognitive process involving multiple processing stages, including semantic access, lexical selection, and phonological encoding [8, 9]. Impairments in naming performance may therefore arise from language-specific dysfunction, such as degraded semantic representations or impaired lexical retrieval, or may reflect broader domain-general deficits, including executive control and attentional processes required for selection and inhibition. In individuals with MCI, it remains unclear whether reduced naming performance primarily reflects a specific linguistic deficit or is secondary to more global cognitive decline.

Despite the widespread application of the BNT in MCI research, existing findings regarding naming performance in this population remain inconsistent. While many studies report that individuals with MCI perform significantly worse on BNT-based naming tasks than cognitively healthy older adults, others have found no significant differences between the groups [9–12]. The considerable heterogeneity observed across studies may be attributed to variations in sample characteristics, BNT versions, scoring methods, and statistical approaches. Furthermore, demographic and clinical factors such as age, education level, and disease severity have been suggested to influence naming ability in MCI [13]. In addition, the rapid expansion of research in this area, coupled with evolving diagnostic criteria, assessment tools, and population characteristics, raises concerns regarding the generalizability of earlier findings [14].

MCI represents a heterogeneous clinical construct, encompassing amnestic and non-amnestic subtypes, some of which may involve prominent language impairment. However, most existing studies do not consistently classify MCI subtypes or report subtype-specific naming outcomes. Accordingly, the present synthesis adopts a broader MCI framework, while recognizing subtype heterogeneity as a potential source of variability.

To address these inconsistencies and provide a more comprehensive and up-to-date understanding of naming ability in MCI, the present study conducted a meta-analysis of literature published over the past decade, integrating recent evidence to clarify the extent and characteristics of naming impairment in individuals with MCI. By restricting inclusion to studies using the Boston Naming Test, this meta-analysis aimed to enhance methodological comparability and facilitate quantitative synthesis across studies.

Materials and methods

The meta-analysis was conducted according to the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines [15]. This study was not prospectively registered in any systematic review registry.

Eligibility criteria

BNT is the most widely used and standardized confrontational naming task in MCI research worldwide, with multiple validated short forms (15-item, 30-item, and 60-item). To ensure comparability and consistency of the naming measure across studies, we restricted inclusion to studies using the BNT or its standardized short forms. This review is limited to studies using the BNT, which may have excluded evidence from other naming measures.

Studies were included if they met the following criteria: (1) cross-sectional study design; (2) inclusion of cognitively healthy older adults and patients with MCI diagnosed according to internationally recognized criteria; (3) use of the BNT to assess naming ability; and (4) availability of sufficient data for meta-analysis, including sample size, mean, and standard deviation of BNT scores for both groups.

Exclusion criteria were as follows: review articles, pre-post intervention studies, single-case studies, studies without a control group, studies involving individuals with multiple disabilities, use of naming tests other than the BNT, or lack of reported BNT scores.

Search strategy

A systematic review was conducted by two independent reviewers (ZS and WLX) who searched the Cochrane Library, PubMed, Web of Science, and EMBASE databases for relevant studies published between January 1, 2015, and March 17, 2025. The time range was chosen to focus on recent evidence consistent with current MCI diagnostic criteria and standardized BNT versions. Search terms included both controlled vocabulary (MeSH terms for PubMed, Emtree terms for EMBASE) and free-text words for mild cognitive impairment, Boston Naming Test, language disorders, naming, and naming impairment, along with relevant synonyms. Full search strategies for each database are provided in Supplementary Table 1. In addition, the reference lists of all eligible studies and relevant review articles were manually screened to ensure comprehensive coverage. Grey literature sources, such as dissertations, conference proceedings, and preprints, were not systematically searched, which may limit inclusion of unpublished data.

Study selection and data extraction

Data extraction was performed independently by two reviewers (ZS and WLX), with any discrepancies resolved through discussion. If consensus could not be reached, a third reviewer (HYS) was consulted for arbitration. The following information was extracted from each study: first author’s surname, year of publication, version of the BNT used, sample sizes of the MCI and cognitively healthy control groups, proportion of male participants, and the means and standard deviations for age, education level, cognitive test scores, and BNT performance.

Quality assessment and publication bias

The methodological quality of each eligible study was assessed using an adapted version of the Newcastle-Ottawa Scale (NOS) for cross-sectional studies, which evaluates three domains: participant selection, comparability of study groups, and outcome assessment. Quality assessment was independently performed by two reviewers, with disagreements resolved through discussion or consultation with a third reviewer when necessary. Potential publication bias was explored using visual inspection of funnel plots and Egger’s linear regression test [16]. Given the relatively small number of included studies and the presence of substantial between-study heterogeneity, these publication bias assessments were considered exploratory and were interpreted with caution. Sensitivity analyses were conducted to examine the robustness of the pooled effect sizes, incorporating leave-one-out analyses in which each individual study was sequentially excluded. When indications of potential publication bias were detected, the trim-and-fill method was employed as an exploratory technique to estimate the possible influence of missing studies on the pooled standardized mean differences (SMDs) and their corresponding 95% confidence intervals (CIs) [17]. It is important to note that the trim-and-fill method depends on specific distributional assumptions and may yield unreliable results in the presence of substantial heterogeneity; consequently, findings derived from this method were interpreted with caution and not regarded as definitive adjustments.

Statistical analyses to conduct the meta-analyses

All meta-analyses were performed using Stata version 18. Pooled SMDs, estimated using Hedges’ g with 95% CIs, were calculated to compare BNT performance between MCI patients and cognitively healthy controls. Different versions of the Boston Naming Test (15-item, 30-item, and 60-item) were used across studies, which precluded direct comparison using raw mean scores. Given the expected clinical and methodological heterogeneity across studies, all analyses were conducted using random-effects models. Univariable meta-regression analyses were performed to examine the influence of potential moderators, including mean age, proportion of male participants, mean years of education, and global cognitive performance. Each moderator was tested in a separate model to minimize the risk of overfitting and multicollinearity, given the limited number of available studies. Subgroup analyses were further performed to explore whether different BNT versions affected the overall effect size.

Results

Study selection

A total of 9,092 records were identified through electronic database searches. After removing duplicates, 7,012 records remained for title and abstract screening. Based on relevance, 77 full-text articles were assessed for eligibility. Of these, 57 studies were excluded for the following reasons: full text not available (n = 6), unclear MCI diagnosis (n = 2), participants without MCI (n = 2), absence of a cognitively healthy control group (n = 5), and missing BNT results or standard deviations (n = 42). Ultimately, 20 [7, 9–12, 18–32] studies involving 5,237 older adults (1,306 individuals with MCI and 3,877 cognitively healthy controls) were included in the final analysis. The study selection process is summarized in Fig. 1.

Fig. 1.

Flowchart of the literature search

Study characteristics

Studies are detailed in Table 1. All studies compared naming ability between individuals with MCI and cognitively healthy control groups. All included studies employed versions of the BNT to assess naming performance. Specifically, 11 studies utilized the BNT-15, 5 studies used the BNT-30, and 4 studies applied the BNT-60. The sample size for the MCI groups ranged from 12 to 350 participants, while the control groups ranged from 13 to 959 participants. The mean age of participants was generally comparable between the MCI and control groups. Minor differences were observed in the proportion of male participants and levels of education.

Table 1.

Demographic Characteristics and BNT Test Performance of MCI Patients and Healthy Controls

	Mild cognitive impairment					Health Controls
	Gender		Age	Education	BNT test	Gender		Age	Education	BNT test
Study	n	(%male)	Mean ± SD	Mean ± SD	Mean ± SD	n	(%male)	Mean ± SD	Mean ± SD	Mean ± SD	BNT version
Drummond 2015 [11]	22	50%	72.1 ± 4.4	13.1 ± 2.3	13.4 ± 1.3	41	36.59%	69.6 ± 5.8	14.5 ± 2.6	14.2 ± 1	BNT-15
Choi 2016 [20]	12	41.7%	72.28 ± 6.96	5.4 ± 3.81	8.17 ± 2.37	13	46.15%	75.75 ± 7.74	5.7 ± 6.18	10.54 ± 2.76	BNT-15
Kim 2017 [32]	27	44.4%	73 ± 8.87	12.48 ± 4.15	11.37 ± 2.51	30	56.67%	71.2 ± 8.71	10.07 ± 2.93	12.57 ± 1.83	BNT-15
Byeon 2020 [30]	106	34.9%	71.13 ± 5.81	5.61 ± 4.53	8.76 ± 3.66	203	25.12%	71.38 ± 6.65	5.93 ± 4.95	10.72 ± 3.09	BNT-15
Liampas 2023 [23]	146	44.5%	74.84 ± 5.2	7.2 ± 5.44	8.92 ± 3.13	1607	40.01%	73.31 ± 5.19	8.17 ± 4.86	10.88 ± 2.88	BNT-15
Macoir 2019 [26]	20	45%	71.05 ± 6.1	13.45 ± 3.3	12.2 ± 1.9	20	30%	70.8 ± 7.1	14.9 ± 2.95	13.65 ± 1.1	BNT-15
Macoir 2021 [25]	14	50%	74.5 ± 6.7	12.5 ± 3.8	12.3 ± 1.7	227	36.56%	68 ± 9.3	14.6 ± 3.5	13.5 ± 1.3	BNT-15
Folia 2023 [18]	118	41.52%	74.13 ± 5.36	6.72 ± 4.52	9.12 ± 2.97	959	39%	72.46 ± 4.7	8.74 ± 4.86	11.21 ± 2.75	BNT-15
Kim 2019 [12]	12	NA	81.75 ± 3.84	11.33 ± 3.75	9 ± 4.09	12	NA	81.83 ± 4.11	7.83 ± 3.71	10.33 ± 3.45	BNT-15
Li 2022 [28]	51	94.1%	67.2 ± 6.7	12.3 ± 2.8	23.1 ± 3.2	101	26.73%	65.6 ± 5.9	12.7 ± 2.6	25.3 ± 2.8	BNT-30
Chasles 2024 [19]	33	42.4%	79.62 ± 5.59	14.14 ± 4.24	25.67 ± 4.37	39	25.64%	77.1 ± 5.94	14.28 ± 2.78	29.18 ± 1.36	BNT-30
Marco 2023 [21]	350	66.3%	74.9 ± 7.24	15.76 ± 2.91	25.24 ± 3.99	197	50.76%	76.07 ± 4.93	16.14 ± 2.93	27.43 ± 2.68	BNT-30
Jin 2019 [27]	48	52.1%	69 ± 6	11.7 ± 3.4	25 ± 2.6	105	47.62	66 ± 6	12.5 ± 3	27.4 ± 1.7	BNT-30
Devora 2024 [7]	18	NA	73.61 ± 8.6	17.44 ± 2.15	54.61 ± 3.57	33	NA	70.83 ± 7.09	16.74 ± 2.62	56.29 ± 2.87	BNT-60
Hirsch 2021 [22]	93	NA	73.8 ± 8.7	16 ± 2.7	52.4 ± 6.2	48	NA	73 ± 6.5	15.9 ± 2.83	56.4 ± 2.47	BNT-60
Won 2017 [29]	35	NA	73.65 ± 5.84	11.53 ± 4.89	40.42 ± 9.44	40	NA	71.27 ± 5.79	11.62 ± 4.01	51.3 ± 4.01	BNT-60
Serrao 2015 [10]	61	NA	68.92 ± 6.49	9.8 ± 5.38	49.55 ± 1.1	38	NA	67.37 ± 5.89	11.89 ± 5.07	52.84 ± 1.54	BNT-60
Taler 2020 [9]	38	55.3%	75.76 ± 7.16	15.21 ± 3.11	47 ± 6.8	41	31.71%	73.32 ± 3.72	15.51 ± 2.67	47.56 ± 6.69	BNT-60
Higes 2021 [24]	81	45%	74.45 ± 4.18	NA	42.78 ± 7.68	84	24.7%	73.42 ± 4.75	NA	47.6 ± 5.9	BNT-60
Taler 2016 [31]	21	52.4%	75 ± 5.41	16 ± 3.5	45.62 ± 10.87	39	46.15%	70.76 ± 5.67	16.15 ± 2.94	53.05 ± 4.93	BNT-60

Compare group performance differences based on BNT type

A total of 1,306 individuals with MCI and 3,877 cognitively healthy older adults were included in the analysis. When results were pooled across all BNT versions, MCI patients demonstrated significantly poorer naming performance compared to cognitively healthy controls, with a standardized mean difference (SMD) of -0.841 (95% CI: -1.001 to -0.675, p < 0.001) (Table 2).

Table 2.

Summary of effect sizes for BNT score between MCI and normal elderly

Type	Effect size and 95% Cl				Homogeneity tests
	SMD	Lower limit	Upper limit	p	Q	df(Q)	p	I²
Overall	-0.841	-1.008	-0.675	< 0.001	76.21	19	< 0.001	75.1%
BNT-15	-0.693	-0.796	-0.589	< 0.001	3.42	8	0.905	0.0%
BNT-30	-0.877	-1.174	-0.579	< 0.001	9.86	3	0.020	69.6%
BNT-60	-1.013	-1.546	-0.479	< 0.001	58.57	6	< 0.001	89.8%

Subgroup analyses based on BNT versions further confirmed this pattern. Studies utilizing the BNT-15 reported an SMD of -0.693 (95% CI: -0.796 to -0.589, p < 0.001), indicating significantly reduced naming performance among MCI patients. Similarly, studies using the BNT-30 revealed an SMD of -0.877 (95% CI: -1.174 to -0.579, p < 0.001). In studies employing the BNT-60, the largest effect size was observed (SMD = -1.013, 95% CI: -1.546 to -0.479, p < 0.001)(Fig. 2). These results consistently demonstrate significant naming impairments in individuals with MCI across all BNT versions.

Fig. 2.

Forest plot comparing BNT naming test scores between MCI patients and healthy controls

Meta-regression analyses

Meta-regression analyses were performed to explore the potential moderating effects of age, years of education, cognitive function, and sex on the effect size of naming impairment between individuals with MCI and cognitively healthy controls. The Knapp-Hartung adjustment was applied to all models to improve the robustness of the estimates. The results showed that age (p = 0.516, adjusted R² = -2.95%; Fig. 3A), years of education (p = 0.165, adjusted R² = 6.58%; Fig. 3B), and sex (logRR-based, p = 0.695, adjusted R² = -55.83%; Fig. 3C) were not significant moderators. In contrast, cognitive function demonstrated a nominally significant moderating effect (p = 0.033, adjusted R²= 33.09%; Fig. 3D). However, even after accounting for these covariates, the adjusted R² values remained low or negative (ranging from − 55.83% to 33.09%), indicating limited explanatory power. Substantial residual heterogeneity persisted across models (I²_res = 39.62%–79.18%; τ²= 0.018–0.176).

Fig. 3.

Association between standardized mean differences (SMD) in Boston Naming Test (BNT) performance and potential moderator variables. A Relationship between BNT SMD and age SMD; B Relationship between BNT SMD and education SMD; C Relationship between BNT SMD and log relative risk (logRR) for gender; D Relationship between BNT SMD and MMSE SMD

Sensitivity analysis and publication bias

The NOS assessment indicated that 17 studies were classified as low risk of bias, while 3 studies were rated as moderate risk. No studies were identified as having a high risk of bias. (Fig. 4) Funnel plot inspection suggested that the distribution of pooled SMD estimates was approximately symmetrical within the range of -2.5 to 0, although slight asymmetry was observed in regions of lower precision (SE > 1.5). (Fig. 5) Egger’s regression test yielded a p-value of 0.145, providing no significant evidence of publication bias. (Fig. 6) Leave-one-out sensitivity analyses demonstrated the robustness of the results, with pooled SMD estimates remaining stable between − 1.01 and − 0.63 across all iterations. These findings indicate that no single study exerted a disproportionate influence on the overall meta-analytic outcome.

Fig. 4.

Risk of bias assessment across domains (Sampling, Confounders, Outcomes). Total scores were categorized as Low (7–9), Moderate (4–6), or High (0–3) risk of bias

Fig. 5.

Funnel plot for global function. In the case of no bias, the figure is symmetrical inverted funnel; When there is publication bias, the funnel plot is asymmetric, and it is skewed

Fig. 6.

Egger’s of global function. Egger’s test showed p = 0.145, p > 0.05. The result showed there was no significant publication bias

Discussion

This systematic review and meta-analysis sought to clarify the presence of naming impairment and its potential contributing factors in individuals with MCI, addressing persistent inconsistencies in the existing literature. Data were synthesized from 20 studies employing the BNT to assess naming ability. The meta-analysis revealed that individuals with MCI exhibited significantly poorer naming performance than cognitively healthy controls, with consistent deficits observed across all BNT versions (15-item, 30-item, and 60-item). These findings provide robust evidence that naming impairment is already present during the MCI stage. This represents a measurable linguistic difficulty associated with cognitive decline and supports its relevance for characterizing language dysfunction in MCI. Compared with more complex language measures, confrontation naming tasks such as the BNT are brief, widely used in clinical settings, and relatively easy to administer, which may enhance their feasibility as early screening tools in MCI populations.

Although naming impairment was observed in most studies, moderate to high heterogeneity was still evident across studies, suggesting that the results may be influenced by multiple factors. Notably, substantial heterogeneity remained even after moderator analyses, indicating that a considerable proportion of between-study variability could not be fully explained. Meta-regression analyses indicated that age, sex, and years of education were not identified as significant moderators in the present analyses. In contrast, global cognitive function, as indexed by the Mini-Mental State Examination (MMSE) and the Montreal Cognitive Assessment (MoCA), was significantly associated with naming performance. Because these global cognitive measures includes language-related items, its association with BNT performance may partly reflect shared linguistic components rather than an entirely independent cognitive effect. These findings should be interpreted cautiously, as the absence of significant effects for age, sex, and education does not preclude potential associations, particularly given the limited number of studies and the exploratory nature of meta-regression.

Beyond demographic variables, differences in the naming assessment tool may also contribute to variability. Although the BNT is a widely used standardized measure, its versions (15-item, 30-item, and 60-item) differ in item count and cognitive load, potentially affecting sensitivity and limiting comparability across studies [11, 19, 21, 23, 26, 28, 32–35]. In addition, cross-linguistic and cultural differences—such as familiarity with test items, vocabulary, and phonological structure—may further influence performance on naming tasks [36, 37]. Methodological inconsistencies, including variations in scoring criteria, cueing strategies, and examiner training, may also introduce bias and contribute to heterogeneity [38, 39].

In addition, unmeasured or unreported moderator variables may further contribute to inconsistencies across studies. For example, most included studies did not explicitly distinguish between amnestic MCI (aMCI) and non-amnestic MCI (naMCI), despite well-established differences in their neuropsychological profiles and pathological trajectories. Notably, aMCI is more likely to progress to Alzheimer’s disease and is frequently associated with medial temporal lobe atrophy and semantic network dysfunction, which may underlie more pronounced naming impairments [40]. Existing evidence indicates that naming ability is supported by a left-lateralized semantic-lexical network, particularly involving the left anterior temporal lobe, medial temporal structures, angular gyrus, and inferior frontal gyrus [41–43]. Early atrophy or functional disruption in these regions, especially in individuals with aMCI, has been observed and may be closely linked to naming impairments [44, 45]. Furthermore, disruption of this semantic-lexical network may precede overt cognitive decline, underscoring the potential of naming tasks as early markers of neurodegenerative processes [46].

Taken together, variations in overall cognitive function, BNT versions, linguistic and cultural context, methodological practices, and clinical characteristics such as MCI subtype are likely to account for a substantial proportion of the residual heterogeneity observed in the present meta-analysis. These findings highlight the need for more standardized assessment procedures and more detailed reporting of clinical and methodological variables in future research. Future studies adopting harmonized naming protocols and clearly defined MCI subtypes, preferably in longitudinal designs, and exploring digital or alternative naming assessments may help to clarify the temporal dynamics and clinical significance of naming impairment while improving sensitivity, standardization, and clinical utility in MCI populations.

Several limitations of the present meta-analysis should also be acknowledged. First, restricting inclusion to studies published within the past decade may have led to the omission of earlier relevant evidence. In addition, most included studies employed cross-sectional designs, limiting conclusions regarding longitudinal changes in naming ability and its prognostic significance. Second, critical variables such as MCI subtypes and language background were inconsistently reported, precluding subtype-specific analyses and limiting the interpretability of the pooled results. Additionally, factors such as participants’ native language, cultural familiarity with test items, and emotional or motivational states were rarely addressed, all of which may introduce potential bias. Nevertheless, the consistent observation of naming impairments across studies suggests that these difficulties are not isolated language impairments, but may instead reflect early manifestations of underlying neurodegenerative pathology [47].

Conclusions

This meta-analysis provides updated and comprehensive evidence that naming ability is significantly impaired in individuals with MCI compared to cognitively healthy controls. These findings support the notion that lexical retrieval deficits may emerge during the MCI stage. Future research should prioritize the standardization of assessment protocols and the detailed reporting of key variables, such as language background and clinical subtypes, to clarify the underlying mechanisms of naming impairment and enhance its clinical utility in the early detection and management of MCI.

Authors’ contributions

**Shuang Zuo: ** Conceptualization, Data curation, Formal analysis, Methodology, Writing – original draft, Writing – review & editing. **Lixin Wu: ** Conceptualization, Data curation, Formal analysis, Methodology, Writing – original draft, Writing – review & editing. **Yunshi Huang: ** Investigation, Data curation, Validation, Writing – original draft. **Linsong Chai: ** Data curation, Validation. **Bingbing Lin: ** Conceptualization, Formal analysis, Methodology, Supervision, Writing – review & editing. **Jia Huang: ** Conceptualization, Resources, Supervision, Methodology, Writing – review & editing.

Funding

This work was supported by the Ministry of Science and Technology of the People’s Republic of China from the Key Research and Development project (2023YFC3503701), Science Fund for Distinguished Young Scholars of Fujian Province (2024J010033), and Technology Planning Project-social Development Guidance (key) Project (No.2023Y0035).

Data availability

The data supporting the findings of this study are available within the article. Further inquiries can be directed to the corresponding author.

Declarations

Ethics approval and consent to participate

As all analyses were based on existing published research, patient consent and ethical approval were deemed unnecessary.

Consent to publication

Not applicable.

Competing interests

The authors declare no competing interests.