Population mental health, help-seeking and associated barriers following the COVID-19 pandemic: Analysis of repeated nationally representative cross-sectional surveys in Czechia

This study investigated the Czech adults ’ mental health following the COVID-19 pandemic and the potential influence of data collection methodology on prevalence estimates. Separately, it investigated changes in help-seeking and associated barriers. Data from representative surveys on Czech adults, conducted in November 2017 ( n = 3,306), in May ( n = 3,021) and November 2020 ( n = 3,000), and in November and December 2022 ( n = 7,311), were used. Current mental disorders were assessed by the Mini International Neuropsychiatric Interview, and the treatment gap was established in individuals scoring positively. In help-seeking individuals, encountering barriers was investigated. In 2017 and 2022, 20.02 % and 27.22 % of individuals had at least one mental disorder, respectively. The 2022 panel sampling and online and telephone interviewing estimates (34.29 % and 26.7 %) were substantially higher than those from household sampling and personal interviewing (19.9 %). Prevalence rates based on household sampling and personal interviewing were broadly consistent in 2017 and 2022. The treatment gap was around 80 % from 2017 to 2022. More than 50 % of individuals encountered structural barriers in help-seeking in 2022. This study showed that prevalence rates were still elevated in 2022, but suggests that data collection methodology influenced the estimates. Separately, the treatment gap remained consistently very high, and encountering structural barriers in help-seeking was common.


Introduction
The COVID-19 pandemic and the associated mitigation strategies, consisting, among others, of closures of non-essential services and businesses and stay-at-home-orders, exposed large parts of the Czech population to levels of known risk factors of mental disorders (e.g., social isolation, stress, financial hardship, and uncertainty about the future) that were unprecedented in the last decades.
To provide timely evidence on the state of the population mental health, we conducted two nationally representative surveys during the pandemic.The data collection for the first survey aligned with the first wave of the pandemic (May 2020), when both the cumulative number of COVID-19-related hospitalizations and deaths was very low (Kasal et al., 2023).On the other hand, the second data collection was realized during November 2020, coinciding with the second wave of the pandemic, during which Czechia was hit particularly hard, leaving the healthcare system overloaded for months (Kasal et al., 2023).Data for both surveys was collected using a mixed computer-assisted web interviewing (CAWI) and computer-assisted telephone interviewing approach (CATI), with individuals randomly drawn from a panel.When compared with pre-pandemic data collection from 2017 that utilized household sampling and paper and pencil interviewing (PAPI), we found that the prevalence of mental disorders increased from 20.02 % to 29.63 % and 32.94 % during the first and second data collection, respectively (Formánek et al., 2019;Winkler et al., 2020Winkler et al., , 2021)).While these results suggest that the mental health of the Czech population worsened during the first year of pandemic, the potential longer-term impacts of the pandemic on population mental health are not known yet.
In addition, the pandemic surveys utilized a different sampling strategy and mode of data collection than the pre-pandemic one; thus, we cannot rule out that part of the previously observed trends might be related to methodological choices (Kasal et al., 2023).In particular, individuals drawn from a panel are effectively "volunteers" who might be systematically different from individuals who are not present, both in terms of underlying socio-demographic, behavioral, and clinical characteristics and interest in the topic, and who have a telephone or access to the internet (Pierce et al., 2020).Moreover, the use of differing modes of data collection (personal vs. online and telephone) might have also influenced the individuals' responses (Davies et al., 2020;Norman et al., 2010), particularly in terms of willingness to disclose the true underlying mental states.
Separately, the implementation of COVID-19 mitigation strategies led to changes in mental health service provision, with a temporary move from in-person to virtual appointments.Considering that in our 2017 data collection, we found that 61 % of individuals with mood disorders and 69 % of individuals with anxiety disorders did not seek professional help for their mental health conditions (Kagstrom et al., 2019), information on help-seeking behavior and perceived barriers in help-seeking is of paramount importance for the evaluation of mental health service provision in Czechia but largely lacking.
In the present study, our objective was to investigate the state of the population mental health approximately two and half years since the start of COVID-19 (November and December 2022), and to compare it with data collections realized in 2017 and during 2020.By using data obtained by (1) household sampling and personal interviewing and (2) sampling from a panel and online and telephone interviewing, we aimed to assess the consistency of results across different data collection methodologies.Moreover, we aimed to investigate the self-reported utilization and barriers in service utilization approximately two and half years since the beginning of the pandemic.

Methods
We pre-specified the research questions and the analysis plan for this study at Open Science Framework before the analysis started (Potočár et al., 2023).Any deviations from the analytical plan are described in Supplementary Methods.

Data
We analyzed data from four nationally representative cross-sectional surveys of Czech community-dwelling adults (18+).The construction of the 2017 pre-pandemic dataset and the two 2020 pandemic datasets, including the employed sampling methodologies, is described in-depth elsewhere (Formánek et al., 2019;Winkler et al., 2018Winkler et al., , 2020Winkler et al., , 2021) ) and in Supplementary Methods.Briefly, the pre-pandemic survey was conducted by a professional agency in November 2017, using the PAPI method.Two-stage sampling, consisting of randomly choosing individuals inside households in randomly chosen voting districts, was used.The dataset consisted of 3306 respondents (response rate (RR) = 59.77%).
For both 2020 data collections, social distancing requirements effectively prohibited the use of personal interviewing techniques, resulting in the use of mixed CAWI and CATI interviewing.The procedure consisted of randomly emailing or telephoning individuals present in a panel of a professional data collection agency (see details in Supplementary Methods).The May 2020 dataset consisted of 3021 participants (2114 CAWI; RR = 35.35% and 907 CATI; RR = 12.31 %) and the November 2020 dataset consisted of 3000 participants (2100 CAWI; RR = 36.32% and 900 CATI; RR = 11.82%).
We note that in our former studies (Formánek et al., 2019;Winkler et al., 2020Winkler et al., , 2021)), we calculated the RRs by considering only those who responded to the surveys (Winkler et al., 2018); however, in the present study, we considered all contacts that were made (including no response), resulting in lower RRs for each data collection than we previously reported.See flowcharts in Supplementary Fig. 1 for details.
The 2022 survey was realized by the same professional data collection agency that conducted the previous pre-pandemic and pandemic surveys.After piloting the questionnaire, the main data collection was conducted in November and December 2022.The same two-stage sampling method was employed for CAPI in 2022 as in 2017.For CAWI and CATI, in-line with the 2020 data collections, the procedure consisted of randomly emailing or telephoning individuals present in a panel of the data collection agency.See details in Supplementary Methods.
In each data collection, samples were representative of the Czech community-dwelling adult population (aged 18 or more years) in terms of age, sex, education, and region of residence.In line with existing definitions, we considered a sample to be representative if the estimates obtained in that sample are generalisable to the target population, and generalizability is achieved when the distributions of key covariates in the sample mimic the distributions in the target population (Jacqueline et al., 2023).Some key vulnerable groups such as people who were homeless, in hospitals or health facilities, and correctional facilities were not included (Wright et al., 2023).All respondents provided informed consent.All study protocols were approved by the Ethics Committee of the National Institute of Mental Health (registration numbers 97/18, 127/20, 198/20, 173/21).

Presence of mental disorders
We used the fifth version of Mini International Neuropsychiatric Interview (M.I.N.I.), a brief, fully structured interview for the assessment of common mental disorders per diagnostic criteria in the fourth edition of the Diagnostic and Statistical Manual of Mental Disorders (DSM-IV) and the 10th version of the International Classification of Diseases (ICD-10), to establish the presence of mental disorders (Sheehan et al., 1998).M.I.N.I. can be both used in clinical settings and administered by lay-interviewers (Sheehan et al., 1998).
We collected data on (1) a major depressive episode, (2) anxiety disorders (panic disorder, generalized anxiety disorder, agoraphobia, social phobia, posttraumatic stress disorder), (3) alcohol use disorders (alcohol dependence, alcohol abuse), and (4) suicidal thoughts and behaviors.The time frames are (1) the past two weeks for a major depressive episode, (2) the past month for panic disorder, posttraumatic stress disorder, social phobia, and suicidal thoughts and behaviors, (3) the past six months for generalized anxiety disorder, and (4) the past twelve months for alcohol use disorders (Sheehan et al., 1998).Agoraphobia has no specified time frame (Sheehan et al., 1998).The presence of any current mental disorder was understood as the occurrence of at least one of the above mental disorders.See details in Supplementary Methods.

Help-seeking
In each of the four data collections, participants who indicated that they sought medical or other professional help due to their mental health during their lifetime were asked whether they sought such help in the past 12 months.In the 2017 and 2020 data collections, the list of medical or other professionals consisted of (1) psychiatrists, (2) psychologists, and (3) general practitioners, while the 2022 data collection included, in addition to these, (4) crisis psychological interventions and (5) online therapists or online therapeutic platforms.
To ensure comparability between the data collections, we considered contact with (1-3) as being indicative of seeking help (i.e., legacy definition).Then, for the 2022 data collection, we also calculated an extended help-seeking variable by considering all medical or other professionals (1-5).

Barriers in help-seeking in 2022
Questions on barriers in help-seeking were only included in the 2022 data collection.We investigated barriers in help-seeking by using two approaches.First, individuals who indicated that they sought medical or other professional help due to their mental health in the past 12 months were asked what the barriers in service utilization were that they encountered in that time period.The list of barriers consisted of (1) unavailability of services in participants' proximity, (2) long waiting lists, (3) absence of in-person examination or consultation, (4) absence of online examination or consultation, (5) the services being financially burdensome, and (6) worries, fears and shame related to help-seeking.We considered a positive answer to (1-5) to be indicative of structural barriers, while a positive answer to (6) to be indicative of individuallevel barriers.
Second, individuals who never sought help during their lifetime or who did not seek help in the past 12 months were asked why they did not seek help due to their mental health.The answer options included (1) absence of a problem that would require help, (2) inability to get treatment due to long waiting lists, financial constraints, and other factors, and (3) worries, fears and shame related to help-seeking.Here, we considered a positive answer to (2) to be indicative of structural barriers, while a positive answer to (3) to be indicative of individuallevel barriers.

Socio-demographic and behavioral characteristics in 2022
We collected information on a number of socio-demographic and behavioral characteristics, including age, sex (registered at birth), gender, region of residence, size of place of residence, marital status, work status, level of education, income level, alcohol consumption (frequency, quantity, and binge drinking).Further, we used the 5-item Self-Identification as Mentally Ill (SELFI) scale to assess to what degree people regard present complaints as evidence for an underlying mental disorder (i.e., self-identification with having a mental disorder) (Schomerus et al., 2012).Higher scores translate to a higher degree of self-identification with having a mental disorder.See details in Supplementary Methods.

Statistical analysis
We provided descriptive statistics for samples across each of the four data collections based on key socio-demographic characteristics, including sex, age, level of education, marital status, work status, and size of region of residence.In the context of comparison across data collections, marital status was coded as married or lives with a partner in the same household, divorced, widowed, single and other, while work status was coded as employed, self-employed, unemployed, retired, student, on parental leave and other.
We computed the prevalence of mental disorders across each of the four data collections, separately for each studied diagnosis and diagnostic group.To examine the consistency of results across different data collection methodologies, we computed the prevalence rates also according to sampling strategy and mode of data collection (i.e., household samples and personal interviewing vs panel samples and online or telephone interviewing).Then, we computed the prevalence of treatment gap across the four data collections using the legacy definition of treatment gap in individuals who scored positively for (1) any mental disorder, (2) major depressive episode, (3) anxiety disorders, (4) alcohol use disorders, and (5) suicidal thoughts and behaviors, separately.For the 2022 data collection, we calculated the prevalence also using the extended definition of treatment gap.Further, in individuals who sought help due to their mental health in the past 12 months, we established the prevalence of barriers in help-seeking, looking both into specific barriers and categorized structural and individual-level factors.In individuals who never sought help during their lifetime or who did not seek help in the past 12 months, we established the prevalence of barriers in helpseeking, separately for structural and individual-level factors.
For the prevalence of mental disorders, treatment gap, and barriers in help-seeking, we computed the 95 % confidence intervals (95 % CIs) using the delta method.
Next, we combined data from each data collection and used binary logistic regression to assess the risk of having (1) any mental disorder, (2) major depressive episode, (3) anxiety disorders, (4) alcohol use disorders, and (5) suicidal thoughts and behaviors in the years 2020 and 2022, when compared with the reference year 2017, and while controlling for age, sex, level of education, marital status, work status, and size of the region of residence.We adjusted for these characteristics to account for potential changes in the underlying population that could, in part, influence the results.
However, while we adjusted for these socio-demographic characteristics in binary logistic regression models, it is possible that simple adjustment for these characteristics would still lead to residual confounding due to covariate imbalance.To further reduce residual confounding due to socio-demographic characteristics, we performed propensity score weighting.We calculated the propensity score based on age, sex, level of education, marital status, work status, and size of region of residence.The target estimand was average treatment effect (ATE), meaning that we weighted each distinct data collection to resemble the combined data from each data collection.We assessed covariate imbalance pre-and post-weighting using standardized mean differences (cut-off = 0.1).To ensure that the presence of extreme weights does not substantially influence results, we performed also an analysis where all weights larger than at the 99 th quantile were set to the weight at the quantile (i.e., weight trimming) (Greifer, 2023).
In the subset of participants who scored positively for mental disorders, we used binary logistic regression to assess which sociodemographic and behavioral characteristics were associated with helpseeking behavior in 2022, using the extended help-seeking variable.We established these associations in the subsets of participants having (1) any mental disorder, (2) major depressive episode, (3) anxiety disorders, (4) alcohol use disorders, and (5) suicidal thoughts and behaviors, separately.Since our intention was not to establish a particular causal relationship, we used bivariable models, assessing the association of each characteristic with help-seeking behavior separately.The considered socio-demographic and behavioral characteristics included age, sex (registered at birth), gender, region of residence, size of the region of residence, marital and work status, level of education, income level, alcohol consumption (frequency, quantity, binge drinking), and self-identification with mental illness as measured by the SELFI scale.Throughout the study, we z-score transformed the continuous exposures before their inclusion in the regression model, and we did not report the results for groups with 10 or less individuals to avoid extremely uncertain estimates.
Finally, using binary logistic regression, we assessed the associations between socio-demographic and behavioral characteristics and barriers in help-seeking, both in individuals who received professional help in the past 12 months and individuals who never sought help during their lifetime or who did not seek help in the past 12 months.We examined the associations between demographic and behavioral characteristics and structural barriers and individual-level barriers separately.We fitted bivariable models, with the list of demographic and behavioral characteristics being consistent with the analysis of help-seeking behavior.
In each analysis involving the entire sample, we applied post-stratification weights to correct for small sampling imperfections and to align with the most recent population characteristics using the R library survey (version 4.1.1).Reflecting the statement from the American Statistical Association on p-values (Wasserstein and Lazar, 2016), we refrained from conducting null hypothesis significance tests.All analyses were conducted in R statistical programming language (version 4.2.2).

Description of samples
The mean age in 2022 data collection was 48.82 years (standard deviation (SD) = 16.57), and the sample included 53.17 % women, 12.9 % individuals with elementary education, 62.23 % individuals married or living with a partner, 2.76 % unemployed individuals, and 21.87 % individuals living in a city with a population of more than 100,000 inhabitants.For detailed sample characteristics stratified by period of data collection and data collection methodology see Table 1 and Supplementary Table 1, respectively.

Prevalence of mental disorders stratified by data collection methodology
In 2022, the proportion of individuals fulfilling the criteria for at least one current mental disorder was 34.29 % (95 % CI = 32.65-35.94) in the panel sample interviewed online, 26.7 % (23.87-29.53) in the panel sample interviewed by telephone, and 19.9 % (18.45-21.34) in the household sample interviewed in-person.The prevalence rates based on panel samples and online and telephone data collection modes were consistently elevated across all data collections when compared with the pre-pandemic survey.The prevalence rates per household sampling and personal interviewing in 2022 were similar to those we observed in the 2017 survey, with 19.9 % (18.45-21.34)and 20.02 % (18.64-21.4) of individuals scoring positively for at least one mental disorder in 2022 and 2017, respectively.The only notable exception was suicidal thoughts and behaviors, which increased from 3.88 % (3.22-4.55) in 2017 to 6.3 % (5.43-7.17) in 2022.The detailed results, including results for specific mental disorders are provided in Fig. 1 and Supplementary Table 3.

Logistic regression on risk of mental disorders in time
After adjusting for age, sex, level of education, marital status, work status, and region of residence, individuals were more likely to report symptoms of at least one mental disorder in 2022 (odds ratio = 1.57; 95 % CI = 1.41, 1.75) than in the 2017 pre-pandemic survey.Similarly, we found elevated risks for anxiety disorders (1.70; 1.46-1.98),a major depressive episode (2.48; 2.04-3.04),and suicidal thoughts and behaviors (3.1; 2.56-3.79).For alcohol use disorders, the estimates were consistent with a decreased to an increased risk (1.1; 0.96-1.27).The results following propensity score weighting were in line with the results based on simple adjustment for socio-demographic characteristics.See Fig. 2 and Supplementary Tables 4-6 for detailed results.

Treatment gap
When compared with the 2017 pre-pandemic estimates (82.22 %; 95 % CI = 79.29-85.15), the treatment gap in people who had at least one of the studied mental disorders was largely unchanged in both 2020 pandemic (78.7 %; 76.06-81.35 and 76.02; 73.34-78.69) and 2022 (77.49 %; 75.66-79.31)data collections.We found similar results for individuals fulfilling the criteria for alcohol use disorders, major depressive episode, anxiety disorders, and suicidal thoughts and  behaviors.
When comparing the legacy and the extended definition of treatment gap that considered also contact with crisis intervention teams and online therapists, we found that the treatment gaps were narrower only residually for each of the studied mental disorders.The detailed results, including results for specific mental disorders, are provided in Table 2.

Barriers in help-seeking
Almost 60 % of individuals who sought professional help in the last 12 months reported that they were confronted with structural barriers such as long waiting lists and unavailability of services (56.97 %; 95 % CI = 53.37-60.56).Individual barriers, such as worries and feelings of fear and shame, were present in 20.36 % (17.43-23.28) of individuals who sought professional help.
Then, out of those individuals who never sought help during their lifetime or who did not seek help in the past 12 months, 8.53 % (7.85-9.20)and 4.59 % (4.08-5.1)reported individual and structural barriers in their potential help-seeking, respectively.See Table 3 for detailed results.

Bivariate associations of socio-demographic and behavioral characteristics and help-seeking
We found that females and transgender/non-binary people were more likely to seek professional help when they scored positively for at least one of the studied mental disorders (odds ratio = 2.38; 95 % CI = 1.91-2.97 and 8.21; 3.1-22.95).Individuals on sick leave and disability pension had increased odds of seeking help (3.92; 1.72-8.95 and 8.33; 5.37-13.18),when compared with employed people.Next, relative to the lowest income level (0-9,999CZK), individuals with 10,000-19,999CZK income had an increased risk (1.47; 1.06-2.05) of seeking help, whereas individuals with 30,000-39,999CZK and 40,000-49,999 income had a decreased risk (0.57; 0.37-0.88 and 0.49; 0.25-0.9).When compared with married individuals, divorced people were more likely to seek professional help for their mental health (1.67; 1.21-2.3).One SD increase on the SELFI scale, indicative of selfidentification with mental illness, was associated with odds of 4. 35 (3.73-5.11)for seeking help.In contrast, one SD increase in binge drinking was associated with decreased odds for help-seeking (0.64; 0.56-0.71).The detailed results, including for specific mental disorders, are shown in Supplementary Table 7.

Bivariate associations of socio-demographic and behavioral characteristics and barriers to help-seeking
In those who sought professional help in the past 12 months, having lower secondary education (i.e., vocational school) was associated with decreased reporting of structural barriers relative to having primary education (odds ratio = 0.56; 95 % CI 0.35-0.91).Similarly, retired people were less likely to report any structural barriers (0.47; 0.27-0.8)than employed individuals.Then, when compared with married individuals, widowed individuals were more likely to perceive structural barriers (3.4; 1.56-8.07).Next, one SD increase in age and binge drinking was associated with lower and higher odds of reporting structural barriers, respectively (0.59; 0.5-0.69 and 1.35; 1.14-1.6).
When compared with married individuals, single and divorced people were more likely to report individual-level barriers, respectively (1.97; 1.09-3.65 and 2.62; 1.15-5.89).Finally, having a residence in a city with a population between 20 and 99,999 and over 100,000 inhabitants was associated with lower odds of reporting individual-level barriers in help-seeking (0.57; 0.34-0.94and 0.56; 0.34-0.91).The detailed results, including for specific mental disorders, are shown in Supplementary Table 5.

Discussion
Using data from four nationally representative cross-sectional surveys of the Czech adult population, we found that the prevalence rate of individuals scoring positively for at least one current mental disorder was still substantially elevated in November and December 2022, when compared with the pre-pandemic baseline.However, importantly, prevalence rates based on household sampling and personal interviewing in 2022 were broadly consistent with the 2017 pre-pandemic rates obtained by the same methodology.
Separately, we found consistently high rates of treatment gap for all studied diagnostic groups of mental disorders before and during the COVID-19 pandemic as well as in 2022.Including previously unconsidered help from online therapists or online therapeutic platforms translated only to residual decreases in the treatment gap.Then, worryingly, more than 50 % of individuals who sought help for their mental health reported encountering structural barriers.

International evidence on changes in population mental health
A substantial body of evidence on population mental health was produced since the beginning of the COVID-19 pandemic, with an umbrella review of the evidence showing small increases in depression, anxiety, and/or general mental health symptoms in the general population, in people with pre-existing physical health conditions, and in children (Witteveen et al., 2023).However, most of the existing cross-sectional studies did not use household probability sampling or diagnostic instruments or had no comparable pre-pandemic baseline (Santomauro et al., 2021).
In a Dutch study that utilized probability sampling and conducted most of the interviews using personal interviewing, with only a small proportion of participants completing a video call, the authors found no The legacy definition contained (1) psychiatrists, (2) psychologists, and (3) general practitioners, while the extended definition, in addition to these, (4) crisis interventions and (5) online therapists or online therapeutic platforms.(worries, fear, shame) 20.36 (17.43, 23.28)We note that data on barriers in help-seeking were collected only in 2022.
L. Potočár et al. differences in prevalence rates immediately before and during the pandemic using a diagnostic interview (ten Have et al., 2023).Assessing for the presence of mental disorders using a diagnostic interview in probability samples from the general population in Trondheim (Norway), the authors found a decrease in prevalence in the first period of the pandemic, with no difference between the interim and second period relative to the pre-pandemic baseline (Knudsen et al., 2021).Then, albeit most included studies did not use probability sampling and diagnostic instruments, a meta-analysis of 134 longitudinal cohorts showed no or minimal changes in symptoms of general mental health, anxiety, and depression (Sun et al., 2023).Similarly, a meta-analysis of 64 longitudinal cohorts showed minor increases in mental health symptoms immediately following the beginning of the pandemic; however, by mid-2020, these returned to pre-pandemic levels in most population subgroups (Robinson et al., 2022).

Effect of data collection methodology
Due to legal restrictions implemented during the ongoing COVID-19 pandemic, we were unable to conduct household surveys using personal interviewing in 2020.Instead, we had to rely on telephone and online interviewing of individuals present in a data collection agency's panel.We demonstrated that using telephone, and, in particular, online interviewing on a non-household sample led to systematically higher prevalence estimates, when compared with household probability sampling and personal interviewing.In fact, prevalence rates based on household samples and personal interviewing in 2022 were broadly consistent with the pre-pandemic rates using the same methodology.We can think of two, most likely mutually inclusive, primary explanations for these discrepancies between data collection methodologies.
First, online, and, to a lesser degree telephone interviewing, might have mitigated the effect of social desirability by increasing the sense of anonymity, thus enhancing an individual's willingness to provide a more negative answer and potentially disclose the true underlying mental state.However, a study that performed a diagnostic assessment of mental disorders by means of telephone and personal interviewing in the same sample found an excellent agreement between the methods for anxiety disorders and very good for major depressive disorder and alcohol and substance use disorders (Rohde et al., 1997).Moreover, whereas the prevalence rate for suicidal thoughts and behaviors increased in 2022 personal interviewing compared with the 2017 baseline, we found broadly consistent estimates for the remaining mental disorders.It seems unlikely that willingness to report would increase for these, but would remain unchanged for other mental disorders, particularly affective disorders.Thus, social desirability or changes in it likely had a modest effect on prevalence estimates at most.
Second, different sampling strategies might be potentially associated with specific sources of bias or the same biases to different degrees, in particular non-response and selection bias.In the household probability sample and the panel samples interviewed online and by telephone, the non-response rate, indicative of non-response bias, was approximately 40 %, 90 %, and 70 %, respectively.Multiple studies showed that mental disorders are associated with the risk of non-response (Knudsen et al., 2010;Lundberg et al., 2005;Torvik et al., 2012), with non-response bias increasing as response rates decrease (Maclennan et al., 2012).This would suggest that the prevalence rates of mental disorders should be underestimated more in our panel samples since they had substantially higher non-response rates compared with our household probability sample.However, this is not supported by the fact that estimates in these samples were markedly higher than in our household probability sample, with the highest prevalence rates detected in the panel sample interviewed online that had the highest non-response rate.On the other hand, to be present in the panel of a professional data collection agency, an individual must effectively volunteer, and such "volunteers" might be systematically different from individuals who are not present, thus leading to selection bias (Pierce et al., 2020).These differences are most likely above and beyond the key socio-demographic characteristics since respondents drawn from a panel had similar distributions on these as participants in the household sample, and adjusting for this information in regression models did not nullify the associations.We cannot reconstruct the precise reasons, but it is possible that individuals in the panel samples were more likely to be present in the panel because of pre-existing mental health conditions or other health conditions and life situations that would be associated with worsened mental health outcomes.
Overall, while we cannot be certain about the precise mechanisms, these suggest that the dominant factor responsible for differing results between data collection methodologies might be substantial selection bias among respondents drawn from a panel.
These observations underline the importance of longitudinal cohorts focused on mental health outcomes.Considering that the cohorts are established using probability sampling before an exceptional event (e.g., a pandemic) that would not allow personal interviewing, changes in the mode of data collection would potentially have a smaller impact on the estimates of population mental health.Additionally, the detailed characteristics of non-responders could be evaluated and non-response weights could be prepared to adjust for these.However, longitudinal cohorts with mental health outcomes are lacking in Czechia as well as in the vast majority of countries around the globe.

Treatment gap and barriers in help-seeking
There were concerns about the widening treatment gap in mental health services during the pandemic (The Lancet Regional Health -Western, 2021); however, we found that the treatment gap was largely consistent in the first year of the pandemic and in November and December 2022, when compared with the pre-pandemic estimates in Czechia.On the other hand, the rates were consistently very high, ranging from approximately 60 % in people with major depressive disorder to around 90 % in people with alcohol use disorders.Using an extended definition of the treatment gap that included online therapists or online therapeutic platforms translated only to residual decreases in the treatment gap, suggesting that these services did not replace the traditional ones.
Then, among those seeking help for their mental health, more than half of individuals reported encountering structural barriers, most commonly long waiting lists and unavailability of services.Not seeking help due to worries, fear or shame was also common, reported by around 20 % of help-seeking individuals.

Limitations
First, our design did not allow us to precisely decompose the potential effect of data collection methodology on estimates to the potential effect of the mode of data collection and the potential effect of sampling methodology.Second, given the cross-sectional character of our data, we could not establish the intra-individual changes in the occurrence of mental disorders.Third, we had no information on nonresponders in any of the modes of data collection; thus, we were unable to investigate whether there were systematic differences between them and the responders.Last, while we administered the same instrument at the same time-period in each of the modes of data collection, we cannot rule out that the instrument led to different false positive and/or false negative rates per mode of data collection.In the future, a twostage design (Tong et al., 2014), consisting of, first, assessing the presence of mental disorders using a screening instrument, and, subsequently, in a subset of these individuals, assessing the presence by a medical professional, would help to answer this question.

Conclusions
The analysis of the entire 2022 dataset showed that the prevalence L. Potočár et al. rates of mental disorders were still elevated relative to the pre-pandemic levels.However, the comparison of results based on household samples and personal interviewing in 2017 and 2022 demonstrated mostly unchanged prevalence rates, with the exception of suicidal thoughts and behaviors, which showed an increase in time.While we cannot be certain about the specific impact of sampling strategy and mode of data collection, these results suggest that data collection methodology influenced the prevalence estimates.Since household surveys using personal interviewing might not be always feasible, as the COVID-19 pandemic illustrated, these results highlight the importance of establishing longitudinal cohorts based on probability sampling in Czechia and other countries that are lacking them.
Separately, we found that the levels of treatment gap are broadly consistent with the 2017 levels, remaining still very high.In addition, worryingly, more than 50 % of individuals reported encountering structural barriers when actually seeking help.

Table 1
Unweighted descriptive statistics of the samples across data collections.
*The marital status category "other" was present only in the 2022 data collection.L.Potočár et al.

Table 2
Prevalence of treatment gap in mental healthcare between 2017 and 2022.

Table 3
Prevalence of barriers in help-seeking in 2022.