Abstract
Background:
People with high levels of neuroticism are greater users of health services. Similarly, people with dementia have a higher risk of hospitalization and medical visits. As a result, dementia and a high level of neuroticism increase healthcare use (HCU). However, how these joint factors impact the HCU at the population level is unknown. Similarly, no previous study has assessed the degree of generalization of such impacts, considering relevant variables including age, gender, socioeconomic, and country-level variability.
Objective:
To examine how neuroticism and dementia interact in the HCU.
Methods:
A cross-sectional study was performed on a sample of 76,561 people (2.4% with dementia) from 27 European countries and Israel. Data were analyzed with six steps multilevel non-binomial regression modeling, a statistical method that accounts for correlation in the data taken within the same participant.
Results:
Both dementia (Incidence Rate Ratio (IRR): 1.537; α= 0.000) and neuroticism (IRR: 1.122; α= 0.000) increased the HCU. The effect of having dementia and the level of neuroticism increased the HCU: around 53.67% for the case of having dementia, and 12.05% for each increment in the level of neuroticism. Conversely, high levels of neuroticism in dementia decreased HCU (IRR: 0.962; α= 0.073). These results remained robust when controlling for age, gender, socioeconomic, and country-levels effects.
Conclusion:
Contrary to previous findings, neuroticism trait in people with dementia decreases the HCU across sociodemographic, socioeconomic, and country heterogeneity. These results, which take into account this personality trait among people with dementia, are relevant for the planning of health and social services.
INTRODUCTION
Sustained high levels of neuroticism lead to brain damage [1, 2] in the long run. Additionally, it has been proven as a risk factor for different types of dementias [3–5], and other conditions associated with dementia [6–9]. Neuroticism is a robust predictor of high prevalent dementias such as Alzheimer’s disease (AD) and vascular dementia [10]. People with high levels of neuroticism use more primary care mental health services [11, 12], and some authors have stated that neuroticism increases health care use (HCU) [13] linked to behaviors that are not conducive to a healthy lifestyle (such as smoking and alcohol consumption), and to less adherence to treatment like in the case of diabetes [14]. Additionally, dementia influences the use of health services such as doctor visits, and periods and frequency of hospitalization [15]. Patients with AD are more likely to have all-cause hospitalizations [16, 17], including potentially avoidable hospitalizations [18]. AD increases the risk to be hospitalized [19], and for longer periods of time, that can be reduced if the person received a tailored treatment [20].
One consequence of the increased use of healthcare services among people with dementia is the increase in direct and indirect costs associated with them, as evidenced by a recent systematic review [21] and a meta-analysis [22]. Similarly, some authors [23] have demonstrated that the costs per person related to neuroticism exceed significantly those associated with mental disorders. Specifically, individuals who score in the top 25% for neuroticism resulted in a total excess cost of $1.393 billion per 1 million inhabitants, which is roughly 2.5 times higher than the excess costs of common mental disorders, amounting to $585 million. Additionally, the consumption of alcohol and tobacco is more frequent among people with high levels of neuroticism, which again results in higher costs for the use of the healthcare system [24].
To our knowledge, despite the different studies that have highlighted the association between neuroticism and HCU, and dementia and UCU, the simultaneous assessment of these factors has not been yet investigated. Similarly, most of the previous studies did not assess population-level data, which are crucial for public policy decision making processes. Moreover, no previous study has assessed the degree of generalization of such impacts, considering relevant variables including age, gender, socioeconomic and country-level variability. The degree of generalization for HCU recommendations from communities to countries to regions is crucially dependent on such potential drivers of heterogeneity and representativeness.
This paper aims to examine how neuroticism and dementia interact in the HCU, while controlling for relevant factors including sociodemographic characteristics such as age and gender, socioeconomic characteristics (proxied by educational level), and the combined influence of these factors, using a large sample obtained from the Survey of Health, Ageing and Retirement in Europe (SHARE) [25]. The SHARE, in its seventh wave, introduced a section dedicated to assessing personality traits including neuroticism and based on the Big Five model [26–28]. The SHARE comprises data regarding the health status (including diagnosis of dementia by a physician), the number of visits to doctors, the main driver for the HCU, and others variable related to sociodemographic (e.g., age and gender) and socioeconomic factors (e.g., educational level), in a large sample of people from Europe and Israel.
We hypothesized that high levels of neuroticism and the presence of dementia will associate with higher HCU. However, the effect of a long history of neuroticism frequently implying the participation of support systems provided by informal caregivers, is unknown in these cases. Although there are many forms of HCU, enquiring about visits to the doctor is a good indicator [29–31]. We focused on these visits as a proxy to HCU. Our hypothesis and result are relevant for the HCU recommendations, such as the incorporation of personality characteristics into models to understand variations in service utilization.
METHODS
Participants
Participants were part of the Wave 7 of SHARE (a survey of aging and retirement conducted across European countries of individuals aged 50 years and older) [25]. This wave collected information on personality traits using the Big Five model [26–28]. The SHARE is a longitudinal study that has followed individuals for 14 years. Respondents reported their use of healthcare over the last 12 months [25]. All participants provided informed consent according to the Declaration of Helsinki, and the SHARE holds Institutional Review Board (IRB) approval for data publication.
The sample of the 7th wave of the SHARE study included a total of 76,718 people (2.4% with diagnosis of dementia) from 26 European countries (i.e., Austria, Belgium, Bulgaria, Cyprus, Croatia, Czech Republic, Denmark, Estonia, Finland, France, Germany, Greece, Hungary, Italy, Latvia, Lithuania, Luxemburg, Malta, Poland, Portugal, Romania, Slovakia, Slovenia, Spain, Sweden, and Switzerland; collapsed into four categories: Northern Europe, Central Europe, Eastern Europe, and Mediterranean Europe) and Israel. The decision to include participants from Israel within a cohort of participants from European countries was based on shared cultural and historical links. Additionally, Israel participates in many European-based research collaborations. Therefore, including Israeli participants within the European cohort allows for a broader representation of the population and increases the generalizability of the study findings.
Measures
Dementia
SHARE [25] includes a variable (ph006d16, defined in SHARE [25] as “Alzheimer’s disease, dementia, senility: ever diagnosed/currently having”) that allows differentiating people who declared had Alzheimer’s disease, or dementia (“having dementia” from now on). This variable was recoded as a dummy variable (yes/no).
Dependent variable: HCU
The dependent variable (HCU) was the number of visits to a doctor in the past 12 months. These variable values range from 0 to 96.
Independent variable: Neuroticism
Neuroticism trait is measured in SHARE [25] by a categorical variable with five values, from one to five (one = lowest trait; five = highest trait), increasing by.5 at each of the nine sublevels of the variable.
Confounders
Socio-demographic variables
Age was measured in years of life. Although the SHARE data includes a variable titled ‘gender, ’ it is important to note that gender refers to socially constructed roles, behaviors, expressions, and identities of men, women, and gender-diverse people [32]. The characteristic included in the SHARE is ‘gender,’ which is a binary variable coded as 1 for female and 0 for male. However, to maintain consistency with the SHARE variable’s title and past research, we refer to this variable in the paper as ‘gender’.
Socioeconomic variables
Years of education was used as a proxy variable to measure the socio-economic status. It was coded by the International Standard Classification of Education 1997 (ISCED) [33]. This variable takes values from 0 to 6:0 (preliminary education), 1 (primary education or first stage of basic education), 2 (lower secondary or second stage of basic education), 3 (upper secondary education), 4 (post-secondary non-tertiary education), 5 (first stage of tertiary education), 6 (second stage of tertiary education). The justification for this is that years of education reported to correlate with income, occupation, and other measures of socio-economic status, especially in contexts where income or occupation data are not easily comparable between countries [34]. Increasing the number of controlling variables may result in the penalization of the significant requirements. In application of the parsimony principle, we opted to include only ISCED.
Descriptive statistics by European region and for Israel for the general sample
HCU, healthcare use; SD, standard deviation. Northern Europe: Sweden, Denmark and Finland; Central Europe: Austria, Germany, Switzerland, Belgium, Luxembourg; Eastern Europe: Czech Republic, Poland, Hungary, Slovenia, Estonia, Lithuania, Bulgaria, Latvia, Romania, Slovakia; Mediterranean Europe: Spain, Italy, France, Greece, Portugal, Croatia, Cyprus, Malta.
Multilevel non-binomial regression modelling
Mixed-effects negative binomial modelling is a flexible and powerful tool for analyzing count data with overdispersion, especially in situations where the data structure is complex or hierarchical [34]. Multi-level models are usually the model of choice when working with “nested” data structures. The independence assumption is a requirement for the application of traditional statistical analyses like ANOVA and ordinary least-squares (OLS) multiple regression. Nested data structures violate this assumption [35]. Multi-level models extend the scope of ordinary regression modeling by accounting for multiple levels of information within a model [36]. Multi-level models are frequently use for this type of analysis in similar works with similar variables [37–39].
Results are expressed as an Incidence Rate Ratio (IRR), standard deviations (SD), 95% confidence intervals (CI), Log. likelihood, and a measure of the goodness of fit of an estimated statistical model: the Akaike and Bayesian Information Criterion (AIC) and the Bayesian Information Criteria (BIC). AIC is good for making asymptotically equivalent to cross-validation. On the contrary, the BIC is good for consistent estimation [35].
Dealing with missing data
We considered for the case of incomplete data that it was not a function of its own value after controlling for other variables, including the possibility of death as an explanatory cause. Consequently, incomplete data were considered as missing data at random [36]. They were discharged and not included in the data analysis. The sample of complete cases available for analysis was 76,561. More information on the assessment, sampling and how to access to SHARE database is provided in the following link: http://www.share-project.org Statistical analyses were performed using Stata software version 14.1 (StataCorp LP).
RESULTS
Main characteristics of the cohort, expressed as mean and standard deviation (SD) for continuous data and frequencies (n) and percent (%) for categorical data are reported in Table 1 for the whole cohort, and in Table 2 for the cohort of people with a diagnosis of dementia. The mean HCU and SD was 5.82±7.21 for the general sample and 9.64±10.11 for people with dementia.
Mixed-effects negative binomial modelling
Mixed-effects negative binomial modelling
To analyze data that exhibit over-dispersion and account for both within-subject and between-subject variations, we built mixed-effects negative binomial models. In order to measure the robustness of the model, a six-step analysis was performed to control by sociodemographic and socioeconomic characteristics, and by the level of neuroticism at each country level.
Descriptive statistics by country for the general sample of people with dementia
HCU, healthcare use; SD, standard deviation.
Model 1. Results of the regression analysis of the main model
*** p < 0.010, **p < 0.050, *p < 0.1 LR test versus nbinomial model: chibar2(01) = 3127.59 Prob > = chibar2 = 0.0000
Mixed-effects negative binomial model is a flexible and powerful tool for analyzing count data with overdispersion, especially in situations where the data structure is complex or hierarchical [37]. Multi-level models are usually the model of choice when working with “nested” data structures. The independence assumption is a requirement for the application of traditional statistical analyses like ANOVA and OLS multiple regression. Nested data structures violate this assumption [38]. Multi-level models extend the scope of ordinary regression modeling by accounting for multiple levels of information within a model [39]. Multi-level models are frequently use for this type of analysis in similar works with similar variables [40–42].
Model 1: Main model
Due to the nested nature of the data, the multilevel approach was used to account for it. The main model (model 1) examines how high levels of neuroticism in people with dementia are associated with changes in HCU. The model controls for having a diagnosis of dementia, level of neuroticism, and the interaction of these two variables on HCU. The main results for model 1 can be seen in Table 3.
In a first step, we applied a mixed-effects negative binomial model to evaluate only the effect of the main independent variables: neuroticism, having a diagnosis of dementia, and their interaction. As can be seen in Table 3, having a dementia diagnosis is a relevant driver for the HCU as well as the level of neuroticism. Both variables are highly significant (α= 0.000) and both incremented the HCU. In contrast, the interaction between them both reduced the HCU (α= 0.030) (Table 4).
Sensitivity analyses
The robustness of the association found in the interaction of dementia and levels of neuroticism on healthcare use were examined by running a series of additional regression models that control different covariance and examining how these alter our findings. Main results of the sensitivity analyses can be seen in Table 4, including: Incidence Rate Ratio (IRR), standard deviations (SD), 95% confidence intervals (CI), Log. likelihood, and a measure of the goodness of fit of an estimated statistical model: the Akaike and Bayesian Information Criterion (AIC) and finally a type of model selection among a class of parametric models with different numbers of parameters: the Bayesian Information Criteria (BIC). AIC is good for making asymptotically equivalent to cross-validation. On the contrary, the BIC is good for consistent estimation [35]. It is out of the scope of the analysis to do comparisons among countries.
Model 2: Controlling for sociodemographic characteristics
Model 2 controlled by the effect of demographic characteristics (age and sex). As can be seen in Table 4. Age and sex remained highly significant variables (α= 0.000). In both cases, there was an increment in the HCU that was greater if female, and if older. The effect of having dementia and the level of neuroticism did not change after controlling for sociodemographic characteristics.
Results of sensitivity analysis. Multilevel regression outputs. Model 1 to model 6
***p < 0.010, ** p < 0.050, * p < 0.1. IR, Incidence Rate Ratio; Isced-97, International Standard Classification of Education 1997: Level 0 – Pre-primary education., Level 1 – Primary education or first stage of basic education, Level 2 – Lower secondary or second stage of basic education, Level 3 – (Upper) secondary education, Level 4 – Post-secondary non-tertiary education, Level 5 – First stage of tertiary education, Level 6 – Second stage of tertiary education.
Model 3: Controlling by socioeconomic characteristics
With model 3 we controlled by the effect of the proxy variable used to control socioeconomic characteristics, the educational level as coded by the ISCED [33]. For this variable, with the exception of the lowest level of education, in which the tendency was to increase the level of the HCU, in the rest of the educational level, the tendency was to decrease the HCU. This decrement was higher the higher the educational level was, with a high level of statistical significance (α= 0.000). There was a very small increment between levels three (IRR: 0.829) and four (IRR: 0.856).
As seen in Table 3, dementia and the level of neuroticism remained on being highly significant (α= 0.000) and in the same direction (they increase the HCU). The interaction of the level of neuroticism and dementia decreased the HCU, this time at a significant level (α= 0.049).
Model 4: Controlling for the country level by sociodemographic characteristics (age and sex)
Further analysis controlling at the country level by age and sex was performed. When controlled by age and gender, the IRR for HCU was lower in model 2 versus model 4 for people with dementia (3.61%) (Table 4). The effect of the interaction of having dementia and the level of neuroticism was very similar to the one shown in the other five models but this time was not significant (α= 0.070).
Model 5: Controlling for the country level by level of neuroticism
Model 5 controls the effect on the interaction of neuroticism and dementia by the level of neuroticism at the country level. After controlling for this variable, the effect of having dementia increased the HCU. The HCU also increased by level of neuroticism. In both cases, the tendencies were highly significant (α= 0.000). The interaction between dementia and the level of neuroticism tended to reduce the HCU, this decrement was significant (α= 0.047) (Table 4).
Model 6: Controlling for the country level by socioeconomic characteristics
Additionally, the effect of socioeconomic by the country level was assessed by using the proxy variable educational level. As can be seen in Table 4, the effect of having a diagnosis of dementia increased the HCU. It also can be seen by each increment in the level of neuroticism. The effect of the interaction of dementia and the level of neuroticism decrease the HCU. The decrease was significant (α= 0.034).
To summarize, after reviewing the main model and the five models used to control by different variables (sociodemographic factors, such as age and gender, a socioeconomic proxy variable, educational level, and controlling by this latter variable and by the level of neuroticism at country level) the effect of having dementia or a higher level of neuroticism, independently, increased the HCU. On average it means around 53.67% for the case of having dementia, and 12.05% for each increment (i.e., 0.5) in the level of neuroticism. Age had a very relevant effect as can be seen in models 2 and 4. In contrast, the effect of neuroticism is very similar in the six dementia models, as can be seen in Table 4. In all the models having dementia and a higher level of neuroticism concurrently, on average, reduced the HCU in all the cases by around 4.29%.
DISCUSSION
This paper confirms previous findings suggesting that HCU increased when the level of neuroticism also increased [11, 13], and the same happened for people with dementia [16, 20].
In contrast, our results contradict some expected results. Among people with dementia, increasing the level of neuroticism was associated with reducing HCU. These results are novel, as reported at the population level and after controlling for multiple confounders. These results may lead to considering new health and social policies for people with dementia and high level of neuroticism, such as specific follow-up measures, including strategies for encouraging healthcare utilization, psychological support, improving communication and collaboration between healthcare professionals and caregivers to help identify and address potential barriers to healthcare use in individuals with dementia who are high in neuroticism, respite resources addressing potential anxiety and fear related to healthcare use, and social support programs people with dementia and high levels of neuroticism. This is extremely important for primary prevention, as people with neuroticism trait are more vulnerable to mental disorders [43], including depression, anxiety disorders, schizophrenia, eating disorders, and personality disorders, known risk factors for dementia and excess HCU [44, 45].
As expected, age and being a female increased the HCU. Additionally, in the case of age our analysis not only confirmed the relationship between age and the HCU, they additionally allowed to quantify how is the IRR of the HCU. It increased with each year of age. As seen in models 2 and 4, when controlled by age and sex they kept being highly significant but, the IRRs decreased. Increasing the number of controlling variables may result in the penalization of the significant requirements. In this sense, it can be considered more relevant the tendency in all the models, than the results in some of them. These results can provide useful insights for policymakers in order to plan health services in communities where old people represent an important proportion of the population.
Additionally, education also impacted the HCU. Results showed that in general, the higher the educational level, the lower the HCU. The HCU was higher for people at the lowest educational level. It is known that people with a low level of education usually have the worst health levels [46, 47] due to their living conditions, healthy habits, and even due to their risk behaviors [48]. Increasing the educational level may improve living conditions and reduces the HCU.
Some of the strengths and limitations of this study had to do with the type of data. Using a population sample such as the SHARE is a great advantage. It allows getting to representative and solid conclusions. In contrast, there is certain information that is not included in the dataset or is included without being checked with an official source. For instance, contact with a doctor in the SHARE is self-reported, which may result in inaccuracies. Additionally, doctor contacts are for the past 12 months, which may result in recall bias. Regarding dementia measurements, the main concerns include dementia diagnosis based on self-report, no indication about time since diagnosis, and no information about the type of dementia, nor about severity stage. Likewise, it is important to acknowledge that years of education as a measure of socio-economic status may not fully capture the complex nature of social and economic factors that contribute to healthcare seeking behavior. In addition, the relationship between education and socio-economic status can vary across different cultural contexts, and may be influenced by factors such as race, ethnicity, and gender, which have not been captured in this research. It was not within the scope of this study to compare different European countries and Israel, which is a limitation in terms of drawing conclusions. However, this highlights the potential for future research to explore cultural and country-level differences in more detail within this line of work, and it could provide valuable insights into the complex interplay between cultural contexts and health utilization outcomes.
Likewise, studying the incidence of educational and work trajectories (e.g., manual workers versus intellectuals) and their relationship with personality types in the development of dementia and hospitalization was beyond the scope of this research. However, it is a line of work that we find interesting for future projects. While we were interested in analyzing the reasons for healthcare utilization, but the dataset does not include information on patients’ reasons for seeking care. Finally, our study is cross-sectional, which did not allow temporal sequence to be investigated nor did it give us the ability to address risk factors. These limitations could be bear in mind and also be overcome in further longitudinal studies specifically designed to answer questions that advance knowledge of psychological traits as a determinant for healthcare utilization in people with dementia.
This study, based on a large international and multicentric sample, contributed to typify the effect of neuroticism and dementia in the HCU, not being affected by country level, nor by any other sociodemographic and socioeconomic variables. However, it should be considered that regardless of the fact that the results are not influenced at the country level, in future studies it would be advisable to include in the analysis the influence of the different healthcare systems, which in our analysis can be considered a source of bias.
Conclusion
The fact that HCU decreases as the level of neuroticism increases among people with dementia is contrary to what it would initially expect, given that each of these conditions separately increases the use of these services. It is possible that these profiles have generated a certain exhaustion among caregivers, or even that the environment, due to a long history of healthcare, has developed attention capabilities that reduce the demand for these services when dementia is present. However, because neuroticism is a personality trait characterized by emotional instability, anxiety, and a tendency to experience negative emotions such as fear, worry, and sadness, in people with dementia, high level of neuroticism may be related to avoidance behavior in seeking healthcare services due to fear of negative outcomes, such as receiving bad news about progression of dementia. It is also possible that people with dementia who are high in neuroticism may have difficulty recognizing the severity of their condition, leading to a decreased use of health services, or they may have a smaller social network and fewer social connections, which can result in a lack of encouragement or support to seek out healthcare services. While we were not in the position to investigate these hypotheses, future qualitative research should include people with dementia and their caregivers, to investigate how neuroticism and other personality traits, such as extraversion, openness, and agreeableness, impact connectedness between people with dementia and their social support system, and the healthcare needs and use.
As far as we know, this is the first study in which the result of the combination of this personality factor and dementia on HCU has been examined. We understand that this study encourages the conduct of more specific studies in order to provide a solid answer to this question.
In contrast with other previous works [19, 20] that did not assess the combined effect of neuroticism and dementia, this research has shown that these combined conditions induce a reduction in the HCU that probably was due to the role of caregivers which is out of the scope of this research. The extra burden and social costs for these people should be assessed in further studies in order to design tailored evaluations and treatment for these profiles of patients and their follow-up.
Footnotes
ACKNOWLEDGMENTS
The authors would like to thank the Public Economics and Globalization Research Group (SEJ393) at the University of Granada, and the Research Aid Program of the Faculty of Social and Legal Sciences at the University of Granada, Melilla Campus (Spain), for their assistance in making this research possible.
FUNDING
No funding was received for conducting this study.
CONFLICT OF INTEREST
The authors have no conflicts of interest to declare that are relevant to the content of this article.
Agustín Ibañez is supported by NIH NIA R01 AG057234, NIH-NIA R01 AG075775, CARDS-NIH, Alzheimer’s Association SG-20-725707, Rainwater Charitable Foundation/Tau consortium, Takeda CW2680521 R-202203-2023090, FONCYT-PICT 2017-1818 and 2017-1820, ANID/FONDAP 15150012, ANID/FONDECYT Regular 1210195 and 1210176 and 1220995, ANID/PIA/ANILLOS ACT210096, and ANID/FONDEF ID20I10152, ID22I10029.
Tatyana Mollayeva is supported by Canada Research Chair in Neurological Disorders and Brain Health CRC-2021-00074 and the GBHI ALK UK-23-971123 Pilot Awards for Global Brain Health Leaders.
The content is solely the responsibility of the authors and does not represent the official views of the institutions.
