fi=3. UTUCris-artikkelit|en=3. UTUCris-articles|

Dose–response relationship between obstructive sleep apnoea severity and C-reactive protein levels: data from the European Sleep Apnoea Database

Dose–response relationship between obstructive sleep apnoea severity and C-reactive protein levels: data from the European Sleep Apnoea Database Grote, Ludger; Gouveris, Haralampos; Lethuillier, Lea; Verbraecken, Johan; Basoglu, Ozen K.; Schiza, Sophia; Ludka, Ondrej; Ryan, Silke; Joppa, Pavol; Fanfulla, Francesco; Mihaicuta, Stefan; Saaresranta, Tarja; Sliwinski, Pawel; Hedner, Jan; Pepin, Jean Louis; Bailly, Sebastien; ESADA Study Group

Introduction

Obstructive sleep apnoea (OSA) characterised by intermittent hypoxia promotes systemic inflammation. This study evaluated the association between OSA severity and circulating C-reactive protein (CRP) levels as marker of systemic inflammation in a pan-European patient cohort.

Methods

This cross-sectional analysis of the multicentre European Sleep Apnoea Database (ESADA) cohort used inverse probability weighted regression adjustment for multiple covariates within a linear mixed-effects model (LMEM) to test the independent association between OSA severity and CRP levels. Covariates included anthropometrics and comorbidities. Study centre and year of analysis accounted for methodological variability in CRP analysis.

Results

18 445 subjects (71% male, median age 53 years (interquartile range 44–62), median apnoea–hypopnoea index (AHI) 22.1 events per h (9–44.9)) were included. CRP (median 3.0 mg·L⁻¹ (1.2–5.1)) increased in a dose–response fashion across OSA severity categories (2.0 (1.0–4.0) for AHI <5 events per h; 2.5 (1.0–5.0) for AHI 5–<15 events per h); 2.9 (1.2–5.0) for AHI 15–<30 events per h; and 3.7 mg·L⁻¹ (1.8–6.4) for AHI ≥30 events per h; p<0.001, respectively). In the final LMEM model, AHI remained an independent predictor of CRP concentration (p<0.001). Other significant predictors of CRP were age and female sex. Obesity (body mass index ≥35 kg·m⁻²) had, among other comorbidities, the strongest independent effect on CRP levels with 2.7 mg·L⁻¹ (95% CI 2.45–2.90).

Conclusions

Our results showed a consistent and robust dose–response relationship between OSA severity and systemic inflammation independent of usual confounders. The combination of OSA and obesity amplified the association. Future studies should address whether elevated CRP could serve as a prognostic marker for subsequent cardiovascular events in OSA.

Register Always Matters: Analysis of LLM Pretraining Data Through the Lens of Language Variation

Register Always Matters: Analysis of LLM Pretraining Data Through the Lens of Language Variation Myntti, Amanda; Henriksson, Erik; Laippala,Veronika; Pyysalo, Sampo

Pretraining data curation is a cornerstone in Large Language Model (LLM) development, leading to growing research on quality filtering of large web corpora. From statistical quality flags to LLM-based labelling systems, datasets are divided into categories, frequently reducing to a binary: those passing the filters are deemed as valuable examples, others are discarded as useless or detrimental. However, a more detailed understanding of the contribution of different kinds of texts to model performance is still largely lacking. In this article, we present the first study utilising registers or genres—a widely used standard in corpus linguistics to model linguistic variation—to curate pretraining datasets and investigate the effect of register on the performance of LLMs. We train small generative models with register classified data and evaluate them using standard benchmarks, and show that the register of pretraining data substantially affects model performance. We uncover surprising relationships between the pretraining material and the resulting models: using the News register results in subpar performance, and on the contrary, including the Opinion class, covering texts such as reviews and opinion blogs, is highly beneficial. While a model trained on the entire unfiltered dataset outperforms those trained on datasets limited to a single register, combining well-performing registers such as How-to-Instructions, Informational Description, and Opinion leads to major improvements. Furthermore, analysis of individual benchmark results reveals key differences in the strengths and drawbacks of specific register classes as pretraining data: How-to-Instructions excels at physical reasoning and sentence completion while barely crossing random baselines on world-knowledge benchmarks, while Narrative boosts performance on social interaction tasks but struggles with scientific questions. These findings show that register is an important explainer of model variation and can facilitate more deliberate and detailed future data selection practices.

Finlandiya joins NATO: Patterns of evaluation in Turkish media registers

Finlandiya joins NATO: Patterns of evaluation in Turkish media registers Erten-Johansson, Selcen

Following Russia’s invasion of Ukraine, Finland reconsidered its longstanding policy of military non-alignment, ultimately deciding to apply for NATO membership. As a NATO member since 1952, Türkiye played a key role in this process, which sparked extensive discussion in Turkish media. This study investigates how Finland and its NATO accession are evaluated across two Turkish media registers – Sözcü news reports and Ekşi Sözlük interactive discussions – through a corpus-based approach grounded in register theory. The research explores how the situational characteristics of each register shape the expression of evaluative language, both overt and covert. Employing keyword and concordance analyses combined with qualitative interpretation, the study examines evaluative and non-evaluative uses of language, with particular focus on the term country and its surrounding context on a positive-negative axis. The findings reveal that news reports tend to express evaluation subtly, often embedding it in covert forms that align with societal norms and values. In contrast, interactive discussions rely on more direct and explicit evaluative language. The analysis highlights how patterns of evaluation are shaped by the communicative functions of each register. Portrayals of Finland also diverge: news media frame Finland’s foreign policy in a largely positive light while expressing criticism of its domestic leadership, whereas online discussions emphasize Finland’s military strength and quality of life, yet raise scepticism about its NATO membership. Overall, this study illustrates how differing media registers contribute to distinct constructions of a country’s portrayal.

Keywords: Turkish media registers, Finland’s NATO accession, keyword analysis, concordance analysis, evaluation patterns, media representation.

Evaluation in social media discourse: A corpus-assisted discourse study of evaluative images of the Covid-19 pandemic on the Finnish Twitter-sphere

Evaluation in social media discourse: A corpus-assisted discourse study of evaluative images of the Covid-19 pandemic on the Finnish Twitter-sphere Saarni, Jenna; Tarkka, Otto; Laippala, Veronika