Longitudinal patterns of alcohol use and psychological symptoms during COVID-19 pandemic and role of alexithymia: A latent transition analysis in the FinnBrain Birth Cohort Study

Background The COVID-19 pandemic has been posing widespread influence on mental well-being. However, research on the dynamic relations between alcohol use and psychological symptoms in the context of the pandemic and the role of alexithymic traits in predicting the development of mental health problems longitudinally remains scarce. Methods Latent profile and transition analyses were conducted to model the longitudinal patterns of transitions in the profiles of alcohol use and psychological symptoms across 10 months during the pandemic (from May 2020 to March 2021) and to investigate the role of alexithymia and its dimensions Difficulty Identifying and Describing Feelings (DIF and DDF), and Externally Oriented Thinking (EOT) in 720 parents from the FinnBrain Birth Cohort Study. Results Three profiles, Risky Drinking, Distressed Non-Risky Drinking, and Non-Distressed, Non-Risky Drinking, and their transitions were identified. The role of alexithymia appeared to be stronger in Risky Drinking than Non-Distressed, Non-Risky Drinking. DIF predicted the development of symptoms in Risky Drinking, whereas DDF predicted Risky Drinking remaining stable over time and showed a trend towards psychological distress in Risky Drinking and Non-Distressed, Non-Risky Drinking. EOT was more likely to be a risk factor for Risky Drinking remaining constant and Non-Distressed, Non-Risky Drinking becoming Risky Drinking. Limitations This study was mainly limited by the generalizability of the findings. Conclusions Our findings add deeper insights into the longitudinal development of alcohol use and psychological symptoms as well as evidence on the role of alexithymia in shaping mental health, providing implications for tailoring clinical preventive and therapeutic measures.


Study participants
This is a sub-study based on the FinnBrain Birth Cohort Study (www.FinnBrain.fi), a prospective cohort study consisting of families (N = 3808) recruited between December 2011 and April 2015 from maternal welfare clinics in the South-Western Hospital District and the Åland Islands (Karlsson et al., 2018). The parents in the initial cohort were invited to this sub-study in May 2020 (T1), at the time when they were raising children around 5 to 8 years old. 856 parents responded to the pandemic sub-study, and they have been followed up at several timepoints across one year until June 2021.
Given the questionnaire assessing alcohol use in past year, an interval of measurements around one year were considered for modeling. However, to keep the sample size in latent profiles and their transitions as large as possible, March 2021, which was 10 months after T1, was selected as T2 due to only 377 respondents available at the June 2021 follow-up. Among all respondents, 726 parents who had completed the questionnaire for alexithymia between 2017 and 2020 were included, of which 6 with missing on all indicators for modeling (see the following sections) at both timepoints were excluded. Thus, the final sample for the current study consists of 720 parents who had no missing data at either timepoint. Of this study sample, 716 parents (99.4%) reported either of the indicator measurements at T1 and 529 (73.5%) at T2 (flowchart in Figure 1).

2.Procedures
Background factors included age, gender, and education collected in the first trimester of pregnancy, pre-pandemic depressive and anxiety symptoms measured at 5 years after delivery, and pandemic stressors covering health events related to self, family members, friends, and acquaintances, free time restrictions, and economic burden assessed at T1 using a questionnaire modified based on the measurement for SARS-related stressors in a prior study (Main et al., 2011). The participants responded to the follow-up questionnaires during the pandemic electronically through the Research Electronic Data Capture (REDCap) platform (Harris et al., 2009).
Regarding the main study variables, alexithymia was measured at 5 years after childbirth (between 2017 and 2020 (Babor et al., 2001;Saunders et al., 1993). It is a 10-item questionnaire assessing three conceptual domains including alcohol consumptions (domain 1: items 1 to 3), dependence (domain 2: items 4 to 6) and alcoholrelated harm (domain 3: items 7 to 10). The AUDIT can be either scored separately on the three domains or as a whole with the scores ranging from 0 to 40, and a proposed cut-off score of 8 was used to identify hazardous or harmful drinking (Babor et al., 2001). In order to obtain more details about alcohol use related problems, the three domains were used as three indicators for modeling. The questionnaire showed sufficiently good internal reliability in our study (Cronbach's α = 0.74 for T1 and 0.71 for T2).

Depressive and anxiety symptoms
The Edinburgh Postnatal Depression Scale (EPDS) (Cox et al., 1987), a valid and sensitive 10item self-report questionnaire assessing depressive symptoms in the past week, has been widely used for screening pre-and postpartum depression for both mothers and fathers (Edmondson et al., 2010).
The total scores range from 0 to 30 with higher scores indicating more symptoms (Cronbach's α = 0.85 for pre-pandemic, 0.86 for T1 and 0.86 for T2). The EPDS was applied in this study given its wide validity in different populations and its longitudinal use in the initial cohort thus enabling the comparisons.
The anxiety subscale of the Symptom Checklist-90 (SCL-90) was used to assess the intensity of anxiety experienced in the past month (Derogatis et al., 1973;Holi et al., 1998). The scores of this 2001; Taylor et al., 2003). The scale consists of three subscales assessing the three main features of alexithymia (DIF, DDF and EOT). Each item is rated on a 5-point Likert scale that ranges from 1 (strongly disagree) to 5 (strongly agree), thus generating total scores ranging from 20 to 100. The Finnish version of the TAS-20 validated by Joukamaa et al. (2001) was used and it showed good internal consistency in the current study (Cronbach's α = 0.83).

2.Analysis strategy
Preliminary analyses including descriptive statistics and comparisons of all variables between T1 and T2 (Chi-square test and Mann-Whitney U test) were conducted using IBM SPSS 28. The main analysis, latent transition analysis (LTA) was performed using Mplus version 8.3 (Muthén andMuthén, 1998-2011). Five indicator variables, that is, alcohol consumption, alcohol dependence, alcohol-related harm, as well as depressive and anxiety symptoms were used for modeling.
The LTA modeling was composed of four steps. First, latent profile analysis (LPA) was used to explore classification based on combination patterns of the indicators at each timepoint. Second, measurement invariance of the models across the two timepoints was tested. Third, LTA, a longitudinal extension of LPA, was performed to investigate the movement of latent profiles over time. Fourth, covariates were included as predictors after the LTA model was specified correctly. The LTA modeling used full information maximum likelihood (FIML) to handle missing data on the indicator variables.
Few missing values on background factors were imputed by mean values ( Table 1). Detailed descriptions of the procedure for LTA modeling were presented in the Supplementary Material. Table 1 presents the descriptive statistics and the proportion of missing data on all variables of the participants at each timepoint. No significant differences were found in the background factors and alexithymia between sample at the two timepoints. The T2 participants showed lower mean values of the AUDIT, domain 1, and the EPDS than the T1 participants. 7.7% of the participants at T1 and 7.4% at T2 reported hazardous drinking (cut-off point of 8). No significant differences in the prevalence of hazardous drinking between T1 and T2 were observed.

3.2. Latent profile analyses (LPA)
J o u r n a l P r e -p r o o f

Journal Pre-proof
The fit indices for evaluating models with increasing number of latent profiles are enumerated in Supplementary Material ( Table S1). The results suggested the 3-profile solution to be the optimal LPA model for both T1 and T2. No significant differences between full invariant model (i.e., all measurement parameters being constrained across T1 and T2) and baseline model (i.e., all parameters being freely estimated) were observed by using the likelihood-ratio test (df = 15, χ 2 diff = 24.65, p = .055). Moreover, Table 2 (upper part) shows that although the item levels slightly changed across T1 and T2, their relative levels within profiles at each timepoint were basically similar, suggesting the general interpretation on the profiles to be equal over time. Hence, the full measurement invariance was reasonably assumed.

LTA model without covariates
As presented in Table 2 (middle part), the first profile representing the individuals with relatively higher levels of alcohol use related problems but low in depressive and anxiety symptoms was named as "Risky Drinking". The second profile representing the individuals with low levels of alcohol use problems but high in psychological symptoms was named as "Distressed Non-Risky Drinking". The third profile in which all the indicator variables were low was named as "Non-

Effects of background factors
There were two ways to detect the role of the predictors as proposed, one for direct effects and another one for group-specific effects (see the Supplementary Material). For the direct effects, male gender and lower education levels were associated with Risky Drinking at T1. More pandemic stressors and pre-pandemic psychological symptoms were related to Risky Drinking and Distressed Non-Risky Drinking at T1. T2 profiles were associated with pre-pandemic depressive symptoms but not anxiety J o u r n a l P r e -p r o o f symptoms. Considering a potential strong collinearity between the two symptoms, only depressive symptoms were included in subsequent modeling (Table 3).
Regarding the group-specific effects, relative to moving to Non-Distressed, Non-Risky Drinking, younger parents as Distressed Non-Risky Drinking at T1 were more likely to remain constant over time

Effects of alexithymic traits
Alexithymic traits were then included in the model controlling for the abovementioned background factors. Higher levels of alexithymia and the dimension DIF and DDF consistently associated with either Risky Drinking or Distressed Non-Risky Drinking at the two timepoints. There was a link between a lower level of EOT and Distressed Non-Risky Drinking at T1 (Table 3).
For the group-specific effects of alexithymia on the transition, as presented in Figure 3, Risky Drinking at T1 with higher overall alexithymia levels were more likely to remain constant or move into  At cross-sectional level, interestingly, simultaneous presence of problematic drinking and high levels of psychological symptoms during the pandemic was not identified in this study, which seems to be a paradox comparing with a number of previous findings that alcohol use disorders tend to be comorbid with major depression and anxiety disorders diagnosed clinically (Brière et al., 2014;Falk et al., 2008). This could be explained by the characteristics of the sample in our study that consists of highly educated parents in a long-term follow-up cohort with few problems in alcohol use and mental health, in which there may be counter-intuitive results. For example, several studies using the sample of general population have shown lower prevalence of alcohol use in individuals with depressive or anxiety symptoms than those without the symptoms (de Sousa et al., 2017;Piwoński et al., 2010;Puddephatt et al., 2022Puddephatt et al., , 2021.  1997;Rolland et al., 2020). The probability of this transition was approximately twofold higher than their reverse change, which to some degree echoes the previous study suggesting that alcohol use more tended to be a consequence rather than a reason for mental health problems (Bell and Britton, 2014).
It is well documented that alexithymia is linked to depression, anxiety, and alcohol use (Devine et al., 1999;Honkalampi et al., 2000;Li et al., 2023;Thorberg et al., 2009). Similarly, the current study indicated alexithymia to be associated with Risky Drinking and Distressed Non-Risky Drinking at both J o u r n a l P r e -p r o o f two time points. One underlying mechanism may be hypothalamic-pituitary-adrenal axis dysfunction in alexithymic individuals thus having problems in the regulation of stress responses (Alkan Härtwig et al., 2013). However, alexithymia was found to be a relatively weak factor per se for the development of Non-Distressed, Non-Risky Drinking in our sample. The role of alexithymia may be clearer under specific conditions such as in clinical samples (Cruise and Becerra, 2018;Honkalampi et al., 2022;Luca et al., 2013). In addition, alexithymia may function as an indirect predictor (Honkalampi et al., 2010). A recent study, for example, demonstrated that alexithymia impacted mental health change during the pandemic via interacting with perceived stress but not directly (Li et al., 2022). Likewise, in the present study, alexithymia predicted Risky Drinking being in the same state or transitioning into Distressed Non-Risky Drinking over time, reflecting the interaction that alexithymia contributed to persistent alcohol use or subsequent psychological symptoms when the individuals had already had alcohol use related problems.
It is noteworthy that specific dimensions of the multifaceted alexithymic traits may play a distinguishing role in the development of mental health problems and alcohol use (Alkan Härtwig et al., 2014;Kajanoja et al., 2017Kajanoja et al., , 2018. Among the individuals with Risky Drinking at T1, difficulty identifying (DIF) and describing feelings (DDF) respectively tended to be risk factors for becoming Distressed Non-Risky Drinking and remaining constant as Risky Drinking. Externally oriented thinking (EOT) was found to be the only one risk factor for both Risky Drinking remaining constant and Non-Distressed, Non-Risky Drinking becoming Risky Drinking. DIF and DDF represent the emotional component of alexithymia, which has been suggested to be associated with maladaptive defense style and social interaction difficulties (Taurino et al., 2021;Vanheule et al., 2007). Whereas EOT refers to a cognitive style with pragmatic thinking and lacking subjective significance of inner feelings (Müller et al., 2003), which may be linked to field dependence (i.e., a tendency to rely on information from outer world) and a lack of social self-confidence in alcoholics (Loas et al., 2000). Additionally, echoing the cognition-related nature of EOT, impaired executive functions may account for the contributing role of EOT in problematic alcohol use (Brion et al., 2017;Riadh et al., 2019).
Interestingly, individuals with Distressed Non-Risky Drinking exhibited a lower level of EOT than Non-Distressed, Non-Risky Drinking at T1. Lower EOT also showed a trend towards contributing to Distressed Non-Risky Drinking remaining over time rather than changing into Non-Distressed, Non-Risky Drinking. It is reported that EOT might play a protective role in mental health problems due to deficiencies of attending to negative emotions (Alkan Härtwig et al., 2014;Davydov, 2017;Wiebe et al., 2017). Nevertheless, EOT was found to act as a significant predictor for the increase of alcohol use J o u r n a l P r e -p r o o f Journal Pre-proof problems. Thus, EOT plausibly has emotional protecting effects but simultaneously renders individuals poor in coping competence such as resorting to substance use when facing stress (Wiebe et al., 2017).

Given that EOT was not observed to predict the other profiles moving into Distressed Non-Risky
Drinking, we speculated that low EOT may be associated with continuing psychological symptoms to some degree, but it did not contribute to the deterioration of the symptoms.
Remaining as Distressed Non-Risky Drinking over time was more likely related to younger age and being mothers, supporting previous findings that these characteristics were associated with poor mental health during the pandemic (Fernández et al., 2020;Joshi et al., 2021). However, except for the aforementioned role of EOT, alexithymia and other dimensions seems unable to influence how Distressed Non-Risky Drinking develops. An explanation may be that the parents with Distressed Non-Risky Drinking at T1 already had high alexithymia levels, and a relatively low level of alexithymia in these individuals can be still high and thus hardly contributing to the decrease of psychological symptoms.
To our knowledge, this is the first study using an exploratory person-centered approach to identify latent profiles of the status of alcohol use and psychological symptoms as well as their development in the context of the COVID-19 pandemic. Moreover, we explore the predictive role of alexithymia in such development in a longitudinal setting, providing sounder understanding of how alexithymia contributes to mental wellbeing and relevant clinical implications such as identifying vulnerable individuals and designing preventive or therapeutic strategies.
However, there are several limitations that should be acknowledged. First, the inconsistent timeframe of the scales assessing alcohol use (for the past year), depressive (for the past week) and anxiety symptoms (for the past month) needs to be considered when interpreting the results.
Nevertheless, the profiles identified based on these indicators were unvarying over time, suggesting that this should not be a major limitation. Second, the interval between the measurement of alexithymia and the indicators for modeling differed across individuals. However, alexithymia is regarded as a highly stable personality feature even over 11 years in adult general population (Hiirola et al., 2017;Karukivi et al., 2014;Tolmunen et al., 2011). Third, the sample consists of parents and is largely mothers, making that the profiles may be more representative of them identified among mothers than fathers and limiting the generalizability of the findings to other types of population. Whereas this did not involve the findings regarding the role of alexithymia in the profile transitions. Fourth, the profiles and transitions may not be applicable to other situations since they were, to a large extent, influenced by social factors such as restrictive policies that are related to alcohol availability and can continuously J o u r n a l P r e -p r o o f vary over time during the pandemic. Fifth, the levels of education as well as alcohol use and psychological symptoms reflected a low-risk and healthy population to some degree, which also limits generalizing the findings to other populations. However, since this appears to underscore the relationships between the profiles or transitions and alexithymia, more significant effects are likely to be found in high-risk populations.

5. Conclusions
The present study determined the transitions in three distinct patterns of alcohol use and     J o u r n a l P r e -p r o o f J o u r n a l P r e -p r o o f Table 3. Direct effects of covariates on the profiles at T1 and T2 Gender: 1 = Women, 2 = Men; Education: 1 Low = high school or lower, 2 Mid = vocational tertiary degree, 3 High = university degree. Pre = pre-pandemic. TAS-20 = 20-item Toronto Alexithymia Scale; DIF = difficulty identifying feelings; DDF = difficulty describing feelings; EOT = externally oriented thinking. OR = odds ratio; CI = confidence interval. *