Processed food intake assortativity in the personal networks of East European older adults

1. Introduction

The presence of others can significantly alter the quantity and quality of an individual’s eating behaviour. The social facilitation of eating highlights how network structures influence individual actions (Herman, 2015). The influence of social connections on dietary habits represents a complex interplay of individual choices, societal norms, and interpersonal interactions. Specific network configurations lead to particular food choices and, consequentially, to a specific diet quality (Kim, 2016). Also, eating norms correlate with the need for social approval and the avoidance of disapproval. This correlation substantially affects food practices (Higgs, 2015). Essentially, social context influences eating through various mechanisms, including cultural norms and social networks (E. Robinson et al., 2013).

Available evidence substantiates that eating patterns shape and, at the same time, are shaped by social networks (Levy et al., 2021). Current research suggests that specific eating behaviour is the result of social influence. Namely, food habits circulate through social relationships, e.g., strong ties lead to similarities in food preferences (Pachucki et al., 2011), workplace networks impact employees’ dietary choices (Levy et al., 2021), and men’s relationships can influence dietary behaviour and weight loss (Harcourt et al., 2020).

However, food habits are not merely the result of peer influence processes, norms, or social networks (Higgs & Thomas, 2016). Cultural tastes and preferences (eating included) can shape personal networks, too (Lizardo, 2006). In sum, we can identify socialization (peer influence), homophily (social selection), and environmental confounding (VanderWeele & An, 2013) as social mechanisms underpinning human behaviour and, implicitly, dietary patterns. Irrespective of the social processes at work, empirical investigations accounting for the social networks embedding individuals have the virtue of providing new insights into food consumption.

Health behaviours and states can exhibit clustering within social networks (Smith & Christakis, 2008). Namely, individuals’ health correlates with the health of their social connections, such as friends, siblings, spouses, and neighbours. Social networks may be crucial in understanding human practices such as food choices or health-related behaviour (Christakis & Fowler, 2013). Nevertheless, network-oriented research work is still coalescing in this area, being in a nascent state. The few existent network studies are limited in their scope. Their focus is instead on adolescents (Becker et al., 2018; Forney et al., 2019; Zhang et al., 2018) or population samples of friends and family (Conklin et al., 2014; Levy et al., 2021). Thus, generalising findings to other age groups and cultural spaces is challenging.

Furthermore, in the case of older adults, there is mixed evidence concerning the association between social network type and dietary behaviour (Kim, 2016). Also, even if social relationships are generally linked to dietary behaviour, it is unclear which structural aspects of personal networks affect healthful dietary behaviour (Conklin et al., 2014). The investigation of how social networks influence eating habits is at a very early stage of development. Consequently, there is an evident need for more inclusive studies, such as examining food intake in older adults by comparison with other age groups.

Given the vital role that diet plays in the health and well-being of older adults (Volkert et al., 2019), and the increasing global burden of diet-related noncommunicable diseases (Qiao et al., 2022), there is a dire need to expand research into how social networks influence dietary behaviours in the senior people. Older adults often face unique nutritional challenges (Appleton, 2009; Volkert et al., 2020), are more often in situations of increased vulnerability (Langmann, 2023) and may benefit from community and social support in maintaining a healthy diet (Seid & Babbel, 2023). This is particularly true for managing the intake of processed foods high in salt, which are linked to hypertension (Berkman et al., 2000), myocardial infarction (Bibbins-Domingo et al., 2010), stroke (Strazzullo et al., 2009), cancer (Chang et al., 2023) and other health issues. Moreover, the dietary decisions of older adults are often influenced by relationship-oriented factors such as leisure and cognitive activities, network size, marital status, or changes in social roles (Bloom et al., 2016; Conklin et al., 2014). Further research could provide valuable insights into the potential for social network interventions that could promote healthier eating habits among older adults (Smith & Christakis, 2008). Such interventions could target social norms, support behaviour change, and leverage existing relationships to mitigate the risks associated with high salt consumption (A. T. Robinson et al., 2019). Addressing unhealthy food behaviour of older adults from a social network perspective may prove beneficial for gaining insights into prevention.

The global population aged 65-year-old plus is projected to increase from 10% (2022) to 16% (2050) (United Nations Department of Economic and Social Affairs, 2022). Romania (a European Union (EU) Eastern European country) reports approximately 3.7 million people aged 65 or over (19.2% of the population) and expects to reach 27.7% in 2050 (European Union, 2022). In 2022, Romania had the fifth highest number in the EU (37.2%) of older adults at risk of poverty or social exclusion (Eurostat, 2023). The situation of Romania is not distinct from the rest of the Eastern Europe. The ageing population in Eastern Europe presents a significant demographic shift, characterized by an increasing proportion of older adults due to higher life expectancies and declining birth rates. This trend is leading to several challenges: healthcare systems are strained by the growing need for medical care specific to this age group; the economy faces pressures from a shrinking workforce and the rising demand for pensions and social services; and there is a marked urban-rural divide, with rural areas often having a higher percentage of senior residents.

Older adults living in rural areas of Eastern Europe face similar challenges influencing their health, social life, and overall well-being (European Foundation for the Improvement of Living and Working Conditions, 2023). Specifically, the seniors have limited access to healthcare, experience high levels of social isolation, rely on inadequate pensions or savings to meet their needs, have lower levels of formal education than their urban peers, and lack the infrastructure essential for comfortable ageing. On top of that, we underline the health disparities. Rural older adults may have higher rates of certain health problems (e.g., unmet dental examination) due to less access to preventive care, lifestyle factors, and the physical demands of rural living (European Parliament, 2020).

The available literature claims Eastern European countries share not only historical experiences (e.g., the transition from Communism) or trends in population ageing, but also cultural specificities of eating habits. The extant findings show, for instance, that adherence to the traditional Eastern European diet is significantly associated with risks of all-cause and cardiovascular deaths (Stefler et al., 2021).

Despite the acknowledged impact of dietary habits on health outcomes in ageing populations, the intersection of networks and dietary choices in these communities remains underexplored, particularly within rural Eastern Europe’s unique socioeconomic and cultural contexts. In this paper, we investigate the influence of social networks on the consumption of processed food high in salt in community-dwelling older adults in rural Romania, a representative Eastern European country. In the context of our study, we operationalize the influence of social networks as assortativity, i.e., similar people are more likely to be connected or share a tie (Newman, 2002). In this respect, we hypothesize that older adults eating processed food high in salt are more likely to be connected to social contacts with similar eating habits. We advance that food intake preferences are clustered in social networks and not randomly distributed. We test this hypothesis using a personal network research design (McCarty et al., 2019) on cross-sectional real-world observed social data. We collect the data from a sample of individuals living in a small Romanian rural locality. As our data are not longitudinal, we limit the scope of our study to assortativity detection. Therefore, disentangling the causes of assortativity (social selection, influence, and environmental confounding) is not the object of the current analysis. By focusing on older adults from an Eastern European rural area, our research offers valuable insights into the social determinants of dietary behaviour, contributing novel data to a previously scant area of gerontological nutrition literature. Our work may enrich the understanding of dietary patterns among rural Eastern European seniors and provide an empirical foundation for designing targeted interventions to mitigate health risks associated with high-salt diets in older populations.

2. Methods

2.2. Procedures

Our research is part of the 4P-CAN project (Personalized cancer primary prevention research through citizen participation and digitally enable social innovation; HORIZON-MISS-2022- CANCER-01, project ID 101104432, programme HORIZON). This project addresses the major modifiable risk factors for cancer to improve primary prevention activities and reduce inequalities in Eastern European countries. Our paper contributes to the 4P-CAN research objectives. We collected data in one of the 4P-CAN living labs (Leresti). Before initiating our study, we informed the local authorities and community about the 4P-CAN projects’ objectives through press conferences (broadcasted by local media), various meetings with community citizens, authorities and local key actors (video content is available on the project’s web page and social media accounts such as LinkedIn, Facebook and X).

All methods were carried out in accordance with the relevant guidelines and regulations (specifically, those provided by the Romanian Sociologists Society, i.e., the professional association of Romanian sociologists). All the research protocol was approved by a named institutional/licensing committee. Specifically, the Ethics Committee of the Center for Innovation in Medicine (InoMed) reviewed and approved all these study procedures (EC-INOMED Decision No. D001/19-01-2024). Informed consent was obtained from all subjects. The privacy rights of the study participants were observed.

Before each interview, we explicitly informed each participant about the 4P-CAN research objectives. Participants freely signed a consent form and received a paper dossier with informative materials about the project. We granted our participants anonymity and the possibility of opting out at any moment. We informed our respondents that a special web page in Romanian with details about the methodology was specially created for them (4P- CAN, 2023). We also provided the participants with the contact details (phone numbers and email addresses) of the project director and research team in case they have future queries. We kept total transparency about the project and the research activities to facilitate citizenship engagement in the study.

We gathered the data through face-to-face interviews. On average, an interview had a duration of approximately 80 minutes. We recorded the responses using the Network Canvas software package (Complex Data Collective, 2023). Given its flexible architecture, Network Canvas allows for an interactive data collection process wherein participants can see the questions and review their responses in real time.

We did not offer monetary compensation to the study participants. Nevertheless, we rewarded them by providing free access to a local educational program focused on health topics and a hotline number in case they need a secondary opinion on medical problems or healthy nutrition.

We employed a link-tracing sampling methodology to recruit participants, i.e., a respondent-driven sampling variant (Hâncean et al., 2021; Molina et al., 2022). Due to its expected superior response rates, we used this sampling technique, not a probabilistic non- network random sampling strategy. In a face-validity manner, we considered a network- oriented sampling method more appropriate given the questionnaire size (eventually, our response rate was 56%).

Fig. 1 illustrates the sampling procedure. We began with six individuals (the seeds). We invited these people to respond to our questionnaire and recommend other interested peers (we did not restrict the possible number of recommendations). We selected the seeds based on ethnographic fieldwork (June – September 2023) that preceded the face-to-face interviews. (During the fieldwork, we explored the community, developed local connections, presented the project to the authorities, took notes about the local realities, etc.). We were interested in having seeds with a different socio-demographic profile (as possible) located in different geographical points of the community (despite its small size of 140,49 km²).

• Download figure
• Open in new tab

Fig. 1.

Link-tracing sampling. The triangles mark the seeds (initial people invited by the researchers to participate in the study). The circles indicate people referenced by other study participants. The red indicates participation, while the grey marks non-participation for various reasons (refusals, rescheduling, etc.). The arrows show the referee–referral chains.

Notably, our seeds were different in sex (four males and two females), age (Range = 30, Min = 34, Max = 64), personal income (three below and three above the average national net salary), employment sector (one retired, three employed in public sector, one employed in private sector and one self-employed) and education (four with BA studies and two with high-school diploma). Two of the six seeds refused to participate as an interviewee, yet they still made recommendations for participation.

After interviewing the seeds, we continued to collect data from the people recommended by each seed and, then, from the persons recommended by the people recommended by the seeds, and so on. Essentially, we followed referee-referral chains as depicted in Fig. 1. We stopped the data collection process when our sample (n = 83) had individuals of various socio-demographic profiles: sex (males and females), age (young, adults and older adults), education (university, high school, elementary), and personal income (below and above the average national net salary).

2.3. Data

2.3.1. Network data

We asked each participant (ego) to nominate a maximum of 25 social contacts (alters) (Maya Jariego, 2018), i.e., people with whom they interact most frequently (18-year-old plus).

Please nominate 25 people (18 years old plus) you interact with (or meet). You can start with the people you interact with most often. These may be family members, friends, acquaintances, neighbours, work colleagues, etc. This question allowed us to create a list of ego-alter ties for each participant.

Furthermore, each ego evaluated each pair of alters (alter-alter tie): whether the elicited social contacts know each other personally. And if they do, whether they are close friends, simple friends or acquaintances. In this way, we eventually created a personal network for each ego. Fig. 2 presents two personal networks as an illustration. For each ego- alter tie, we asked our respondents information about the frequency of the meetings (How often do you typically meet with each of the people you previously mentioned? Responses: Daily, weekly, every two weeks, once a month, a few times a year, once a year, less than once a year) and their emotional closeness (How emotionally close do you feel to the people you previously mentioned? Responses: Very close, close, not very close, not close at all). We used this information to estimate the intensity of each ego-alter relationship. First, we dichotomized meeting frequency (Daily meeting = 1, the rest of the responses = 0) and emotional closeness (very close = 1, the rest of the responses 0). Second, we multiplied the binary variants of meeting frequency and emotional closeness and obtained the intensity of an ego–alter tie.

• Download figure
• Open in new tab

Fig. 2.

An illustration of two personal networks. The square nodes represent the egos (the study participants), whereas the circles mark the alters (the social contacts elicited by the egos). The colours indicate the alters’ status in relation to the ego: blue (family members), dark red (close friends), green (simple friends), and black (acquaintances). The colours of the ties mark either ego-alter ties (grey) or alter-alter ties (red). The weight of grey ties marks the ego-alter meeting frequency (weight increases as meeting frequency increases). The weight of the red ties marks the type of the alter-alter ties (very thick lines: close friends; simple lines: simple friends; interrupted lines: acquaintances). Images A and B visually illustrate the idea of network density (proportion of materialized alter-alter ties in each personal network). Network A has a lower network density score in comparison to B. Both networks have a size of 25 (this corresponds to the number of alters).

We employed the network information to create two network-level variables for each of the 83 personal networks: the network size (the number of elicited alters) and the network density (the number of observed alter-alter ties divided by the number of theoretically possible alter-alter ties).

2.3.2. Attribute data

Participants provided information about their sex (as assigned at birth: male or female), date of birth, last level of education, occupational status (active or retired), civil status (single or in a relationship), personal income level (above or below the national average net salary) and if they practice any sport (yes/no). The egos gave information about each of their nominated alters: age (in completed years), sex (male or female), education (last level), and whether they were on a diet or had any sport as a hobby (yes/no).

We computed the Body Mass Index (BMI) for each ego based on self-declared weight (in Kg) and height (in cm). We divided the weight by the height squared. Then, we classified the resulting scores into “underweight” (less than 18.5), “normal” (18.5 – 24.9), “overweight” (24.9 – 29.9) and “obesity” (higher than 29.9). We also measured our respondents’ perceptions of their alters’ BMI. Namely, we administered pictorial images of females and males (Harris et al., 2008) to our respondents. We registered the reported corresponding BMI class for each alter (social contact), i.e., ten classes from A to J corresponding to underweight, normal weight, overweight, class I, II and III obesities.

We measured the consumption of possessed food high in salt (PFHS) by administering a pictorial image to our respondents (See Supplementary Material, Fig 1S, available in Hâncean et al. 2024). Each participant reported the frequency of PFHS intake for them and their elicited alters (How often do you (How often does alter) consume foods similar to the ones depicted in this pictorial image? Responses: daily, weekly, every two weeks, once a month, a few times a year, once a year, less than once a year, never). We dichotomized the responses into a binary variable: whether the ego (alter) eats PFHS food frequently (daily or weekly). We built the pictorial image using the guidelines available in the WHO STEPS Surveillance Manual (World Health Organization, 2017). The pictorial image included canned noodle soup, pretzels and chips, burgers, pizza, canned food and a deli meats platter. We adapted the image to the Romanian cultural specificities by including pickled food and grilled minced meat.

2.3.3. Compositional data

Based on the information reported by the ego, for each personal network, we computed the proportion of alters by sex (e.g., proportion of females), age categories (young [18 – 30-year- old], middle [31 – 63 year-old] and older adults [64 year-old and older]), education (e.g., proportion of alters with a university degree), BMI class (e.g., proportion of alters with obesity), diet (proportion of alters on a diet), sport (proportion of alters with a sport as a hobby), PFHS intake (proportion of alters that eat PFHS frequently), closeness (proportion of alters nominated as very emotionally close) and meeting frequency (proportion of alters that ego meets daily).

2.3.4. The assortativity variable

We computed a food intake assortativity variable for each alter embedded in a personal (ego) network. Namely, the difference between the proportion of the alter’s neighbours who eat PFHS frequently and the proportion of alters in the entire ego network that eat PFHS frequently (without the alter of reference). (Eating PFHS is a dichotomized variable, where 1 means frequent processed food intake and 0 does not.) The assortativity variable tests whether alters’ PFHS intake associates with the PFHS intake of those to which they are connected (their social contacts). The assortativity variable tells us whether an alter’s neighbours eat PFHS more or less frequently than all the alters in the ego network. The normalization obtained by subtracting the average PFHS intake overall alters (minus the alter of reference) in a network is necessary. Without this normalization, the effect of this variable would be confounded by the overall ratio of alters frequently eating PFHS in an ego network. (Variation of PFHS ratios over the different ego networks is controlled for by random ego- level effects). Across the 83 ego networks, we dropped the alters without neighbours from the analysis. The assortativity variable was undefined in their case. Fig. 3 visually illustrates the logic behind the assortativity variable. In the displayed personal network, alter k has four neighbours (1, 2, 3, and 4), of which only two eat PFHS frequently (3 and 4). In the entire personal network, there are 24 alters (plus the alter of reference, k, that we exclude from the calculations), and six alters eat PFHS frequently (see the red nodes). The assortativity variable would equal 2/4 – 6/24 = 1/4. This result means that the neighbours of alter k have an average PFHS intake behaviour above the average PFHS intake behaviour in the personal network (they are more likely to eat PFHS frequently than the rest of the alters in the same network). The alter of reference is not included in the computation to avoid circular dependency in the data. Here, we replicate the mathematical formula used to assess assortativity in the case of COVID-19 vaccination (Hâncean et al. 2023).

• Download figure
• Open in new tab

Fig. 3.

The distribution of processed food high in salt (PFHS) intake behaviour in a personal network. The colours indicate eating patterns: red marks people who eat PFHS frequently, whereas blue marks people who eat PFHS less frequently. The geometrical shapes present the ego (the square) and the social contacts (the circles). Alter-alter ties are in blue, and ego-alter ties are in grey. In the personal network, we observe 25 alters and one ego. Alter K has four neighbours (1, 2, 3, and 4).

3. Results

4. Discussion

Our study explores the influence of social networks on the eating habits of older adults dwelling in rural regions of Eastern Europe, namely Romania. The assortativity effect (Model 2, OR 1.17, p 0.01) is statistically significant, highlighting the potential for dietary habits to propagate within social networks, possibly creating clusters of similar eating behaviours. This effect may be more pronounced in older adults, who may rely more heavily on their immediate social circles due to potential limitations in their social interactions. This result is consistent with the concept of homophily in social networks (Hâncean et al., 2022a, 2022b; Lerner & Hâncean, 2023; McPherson et al., 2001) or the social contagion of obesity (Christakis & Fowler, 2013). Also, it underscores the need for interventions that engage with the social environment to promote healthier eating habits (Bahr et al., 2009).

The substantial intra-class correlation coefficients in the multi-level regression models (M1: 0.67 & M2: 0.66) emphasize the impact of network context on PFHS intake behaviour, suggesting that personal networks may significantly shape individuals’ dietary patterns (Pachucki et al., 2011). This insight is of particular relevance when considering the nutritional health of older adults, as it implies that interventions targeting this demographic should extend beyond the individual to encompass their social environment.

The analysis of egos’ social contacts also reveals a distinct negative association between age and the consumption of PFHS, with older adults (alters) demonstrating a lower propensity for such dietary choices (M1: OR 0.57, p 0.00 & M2: OR 0.58, p 0.00). This result aligns with the broader literature indicating a generational divide in dietary preferences, as younger individuals have been shown to have a higher inclination towards processed foods (Monteiro et al., 2018; Popkin, 2006). Also, this finding may be influenced by regional economic and availability factors specific to Eastern Europe that merit further investigation.

Additionally, the alters who adhere to a diet are markedly less likely to eat PFHS (M1 & M2: OR 0.23, p 0.00), indicating potentially a conscientious shift towards healthier eating habits. This finding resonates with previous work (Fiese et al., 2012) that emphasizes the role of individual agency and family rituals in shaping eating practices. However, our study did not explore the nature of these diets. Future research should uncover whether these are formally prescribed diets, self-directed diets, or culturally prevalent eating patterns and how adherence was achieved.

The positive correlation between BMI and PFHS intake (M1: OR 1.44, p 0.00 & M2: OR 1.45, p 0.00) raises concerns about the dietary patterns of individuals with a higher body mass index, perhaps caught in a reinforcing cycle of poor nutritional choices. The complex interplay between dietary choices and metabolic health has been extensively documented (Swinburn et al., 2011). Our study is limited to only underscoring the need to investigate the causal pathways linking dietary behaviours and body weight (Malik et al., 2013).

Turning to the analysis of the egos’ PFHS intake, we suggest that individual characteristics in isolation may not sufficiently explain variations in PFHS consumption. However, introducing network structure measurements improves explanatory power in our logistic regression models. Notably, including the proportion of older adult alters within the network as a significant predictor (M4: OR 0.33, p 0.03) highlights the potential role of this demographic in influencing overall network dietary behaviour. The positive effect of the proportion of alters that regularly eat PFHS (M4: OR 2.42, p 0.01) can suggest a significant likelihood of individuals adapting their eating habits to mirror those of their social contacts who frequently consume PFHS (an assortativity effect). This result is particularly relevant for older adults, who may be more exposed to and influenced by the dietary patterns prevalent within their close-knit social networks.

The robust relationship between physical activity and egos’ dietary behaviour, as evidenced by the significance of sports activity across models (M3: OR 0.33, p 0.04 & M4: OR 0.24, p 0.03), merits attention. This association may reflect a lifestyle pattern within ego networks where healthier physical activity levels correlate with more nutritious dietary choices. Our findings must be contextualized within the broader scientific discourse, particularly considering Eastern Europe’s socio-cultural and economic backdrop (Goryakin et al., 2015). The intersection of physical activity and dietary behaviour necessitates a more nuanced future analysis, potentially informed by the available evidence in the field (Warburton et al., 2006), highlighting the synergistic effects of lifestyle factors on health.

There are several limits that the readers should note while interpreting our results.

First, our study relies on cross-sectional data, meaning separating social selection and influence in the context of assortativity is impossible. For instance, it cannot be ascertained whether being in a network with PFHS intake leads to consuming more processed food (social influence) or if individuals with a preference for PFHS are more likely to form connections with similar dietary habits (social selection). Additionally, cross-sectional designs do not account for periods when processed food intake might be higher or lower (temporal biases). Collecting longitudinal network data in the future can aim to disentangle social selection and influence and improve self-reporting accuracy. Second, the link-tracing sampling of the study participants has inherent selection biases. For example, more health- conscious individuals may be more willing to participate in a study about prevention, which could skew the results. Similar food intake studies focused on East European countries are rare and employ different methodologies (Boylan et al., 2009; Krivonos & Kuhn, 2019), which makes result validation almost impossible. Third, an inherent feature of personal network research designs is that egos report data about themselves and their social contacts. While this type of network research design offers valuable insights into understanding food intake, it may be challenging to control for various biases, such as the false consensus effect (the tendency to see one’s own choices replicated in the others) (Ross et al., 1977) and recall biases (inaccurate reports about PFHS consumption). We tried to manage the false consensus bias by controlling, in our models, the ego-alter tie intensity. Yet, data collected on longitudinal cohorts (balanced panels) can improve the accuracy of self-reported information in the future. Fourth, the generalization of our findings to the entire population is affected by its lack of representativeness.

Our study discusses the impact of assortativity within personal networks on dietary behaviour, specifically among older people in rural Eastern Europe who face various vulnerabilities. Consistent with the social integration theory (Berkman et al., 2000), our findings suggest that personal relationships and networks can profoundly impact health behaviours, including diet. It underscores the complexity of dietary choices, shaped by individual characteristics and personal networks (direct alters). A detailed understanding of these social variables is crucial for older adults, whose eating habits are vital for the promotion of healthy ageing and well-being. This knowledge can be utilised to create specific interventions aimed at encouraging better nutrition and cultivating healthier food habits.

Community-based interventions can draw inspiration from the social-ecological model of health promotion (Stokols, 1996).

Declarations of interest

none.

Data availability

The dataset analysed in the current study is made openly available in the Zenodo data repository as Hâncean, M.-G., Lerner, J., Perc, M., Molina, J.L., Geantă, M., Oană, I., Pintoiu-Mihăilă, B.-E., & Puncioiu, S.-E. (2024). Replication data for: Processed food intake assortativity in the personal networks of East European older adults. Zenodo. https://doi.org/10.5281/zenodo.10534214

Funding

M.-G.H., M.G., B.-E.P.-M., I.O. and S.-E.P. were supported by the 4P-CAN project, HORIZON-MISS-2022-CANCER-01, project ID 101104432, programme HORIZON; J.L. was supported by Deutsche Forschungsgemeinschaft (DFG 321869138); M.P. was supported by the Slovenian Research Agency (Javna agencija za raziskovalno dejavnost RS) (Grant Nos. P1-0403). The funding sources had no involvement in study design, in the collection, analysis and interpretation of data, in writing the paper and in the decision to submit the article for publication.

Author contributions

M.-G.H.: Writing, Visualization, Conceptualization, Methodology, Software, Formal analysis, Investigation, Data curation. J.L.: Writing, Conceptualization, Validation, Methodology, Software, Formal analysis. M.P., J.L.M.: Writing, Conceptualization, Validation. M.G.: Conceptualization, Investigation, Resources, Writing. B.-E.P.-M.: Visualization, Investigation, Data curation, Writing. I.O.: Writing, Investigation, Data curation, Methodology. S.-E.P.: Investigation, Data curation. All authors approved the final version of the paper and accept full responsibility for all aspects of the work described.