Mental health burden of persons living in Ukraine and Ukrainians displaced to Switzerland: the mental health assessment of the Ukrainian population (MAP) studies

Study design, participants and setting

The MAP studies are prospective, population-based, observational cohort studies that include people from the general population of Ukraine (MAP-U), people from Ukraine with refugee status residing in the canton of Zurich, Switzerland (MAP-Z (UA)) and people from the Zurich general population (MAP-Z (ZH)). A detailed study protocol has been published,15 the study is preregistered (ISRCTN17240415), and we followed the Strengthening the Reporting of Observational Studies in Epidemiology Statement for reporting.

Eligibility criteria for MAP-U were people aged 18 years of age or older, residing in 10 oblasts of Ukraine (Zhytomyr, Rivne, Sumy, Kharkiv, Mykolaiv, Kherson, Lviv, Chernivtsi, Kyiv and Dnipropetrovsk), providing informed consent and the ability to complete the survey; for MAP-Z (UA): being Ukrainian, aged 18 or older, registered with refugee status S in the canton of Zurich, providing informed consent and the ability to complete the survey online; for MAP-Z (ZH): being aged 18 or older, residing in the Canton of Zurich with either Swiss citizenship or residence permit B or C, providing informed consent and the ability to complete the survey online.

Ukrainians living in Ukraine were randomly selected by the Center for Social Research of Sumy State University. Ukrainians living in the canton of Zurich with protection status S as well as the people living permanently in the canton of Zurich were randomly selected by the Swiss Federal Statistical Office. Random selection was age-stratified (18–24 years, 25–44 years, 45–64 years and ≥65 years). Randomly selected persons were invited for study participation between 25 March and 9 August 2024.

Randomly selected individuals in Ukraine (MAP-U) received an email with a personalised link to the assessment via a secure digital platform (REDCap),16 17 including two reminders. If there was no response, the interviewers from our local teams in Ukraine approached them via Telegram (until May 2024 when Ukrainians were recommended to stop using it), Viber (from May 2024) or phone calls. Follow-up assessments are planned every 3 months for at least 2 years, with invitations sent by email, Telegram or Viber.15 For MAP-Z, initial contact was made by postal mail, which has proven effective in other population-based studies in Switzerland. This design mirrors that of the digital cohort in Corona Immunitas, the COVID-19 Social Monitor and the Swiss MS Registry.18 19 The recruitment of the MAP-Z participants was conducted between 3 April and 8 August 2024.

To compare the state of mental health of the population within Ukraine, we focused on three regional clusters formed from 10 selected oblasts across different parts of the country, which are affected differently by the war: The North-West cluster (Rivne, Lviv and Chernivtsi oblasts) is distant from the war zone and has comparatively low security risks, the Central cluster (Zhytomyr, Kyiv and Dnipropetrovsk oblasts) is situated closer to the border with the aggressor and has medium security risks, while the South-East cluster (Sumy, Kharkiv, Mykolaiv and Kherson oblasts) is near the war zone and has high security risks due to its proximity to the border with Russia.

Data sources and outcome measurement

All data were collected directly from participants. We collected data through the electronic database REDCap using electronic questionnaires. At baseline, we collected data on sociodemographic information along with the primary endpoints which included the prevalence of symptoms of PTSD (measured by the Posttraumatic Stress Disorder Checklist (PCL-5)), anxiety (Generalised Anxiety Disorder 7-item scale (GAD-7)), depression (Patient Health Questionnaire-9 (PHQ-9), as well as secondary endpoints such as experiences of somatic symptom disorder (Patient Health Questionnaire-15 (PHQ-15) and participants’ health-related quality of life (Visual Analogue Scale of EuroQol (EQ-VAS)).

Bias

To minimise selection bias, we used stratified random sampling based on population registries and made a great effort to increase participation rates despite the very challenging situation in Ukraine. Because the gold standard approach of expert administered semistructured interviews cannot feasibly be employed in this context, we rely on commonly used, standardised questionnaires which we chose in a systematic selection process. For the three mental health conditions of interest, we identified potential instruments and compared them against our selection criteria: (a) strong psychometric properties; (b) applicable estimates of sensitivity and specificity for statistical correction; (c) digitally implementable; (d) available in Ukrainian; (e) appropriate in the context of war; (f) established cut-offs for likely presence of disorder; (g) established use in population studies; (h) short duration or low completion burden; (i) previously used in Ukraine. Through this careful selection process, we made sure that the data from the MAP studies could inform the Global Burden of Disease studies of the Institute for Health Metrics and Evaluation (Seattle, Washington, USA). The final decisions were made at an international working group meeting held in Zurich in Spring 2023.

Study size

For MAP-U, we aimed for a precision of ±5% for 95% CIs around prevalence estimates. We considered a specific age group in a specific regional cluster (eg, participants aged 25–44 years in the central cluster) as the unit for which we needed the ±5% for 95% CIs. Based on the available prevalence data in Ukraine and data from other conflict areas, we expected a prevalence of 25% for anxiety and depression and 40% for PTSD in the Ukrainian population. Sample size requirements are highest for a prevalence of 40%, on which we thus based sample size calculations.20 21 To provide a precision of ±5% in estimating prevalence with 95% CIs for a specific age group in a specific cluster, the required sample size was 369 survey respondents per combination of age group and cluster. Assuming a potential attrition of 15% among those who consent to participate, the required sample size for enrolment was 435 per age and cluster stratum. The size of the target sample for MAP-U was thus 5220 persons (net sample; given four age and three clusters of oblasts strata). Expecting a participation rate of 6%–7% for MAP-U, reflecting the difficult situation in Ukraine (ie, inability to reach randomly selected persons, concerns to participate, power outages and other reasons), the final number randomly drawn from the population registry in Ukraine was 6715 per stratum and 80 579 for the total sample.

For the MAP-Z study, we assumed a participation rate of 25%, based on experiences with previous population-based studies in the canton of Zurich (eg, Corona Immunitas studies).19 For the age groups 25–44 years and 45–64 years among people living with protection status S, as well as for all four age groups in the general population of the canton of Zurich, we initially aimed for 435 participants per age group, totalling 1740 people to be invited. However, there were fewer than 1740 people with protection status S in the youngest (18–24 years) and oldest (65+ years) age groups with protection status S in Zurich. Therefore, we randomly selected 684 people in those two age groups to achieve a sample of 171 (25%). In these age groups, the precision was expected to be lower, with 95% CI expected at ±8%. The total sample sizes for people living with protection status S and for the general population living in the canton of Zurich were 1220 and 1740, respectively. Sample size calculations were performed using the Scalex SP calculator.22

Quantitative variables

Scores for the questionnaires were summed according to guidance from the developers of the respective instruments. If less than 20% of items were missing for any single survey, we used mean imputation to calculate the sum scores. If more than 20% of items were missing, we excluded these surveys. A total of 661 out of 6678 participants (9.9%) had any missing data imputed. This proportion was similar across the three study arms, ranging from 8.8% to 10.1%. The observed internal consistency assessed using Cronbach’s Alpha of the employed instruments was either good or excellent (α_PHQ=0.87, α_GAD-7=0.91, α_PCL-5=0.95), PCL-5 criterion subscales showed good consistency as well (range=0.87–0.89).

Statistical methods

We used means with SD, medians and IQRs or absolute numbers and percentages for the descriptive analyses of the study population as well as war-related experiences. Symptom sum scores, reflecting symptom burden, were also shown descriptively for PTSD, anxiety and depression, as well as for somatic symptom disorder.

We conducted prevalence estimations for PTSD, anxiety and depression for subgroups defined by the cluster of oblasts (see above), age group and sex. Because the questionnaires used were primarily designed as screening tools, they prioritise sensitivity, which creates a serious risk of overestimation if prevalence is calculated based on the established cut-offs. A recent systematic review indicated that prevalence estimates based solely on standardised instruments are approximately twice as high as those prevalence estimates based on standardised diagnostic interviews.23 We addressed this challenge by estimating prevalence and 95% credible intervals using a Bayesian logistic regression model stratified by age group, sex and cluster of oblasts and adjusted for sensitivity and specificity of the instruments based on available literature.24 To obtain the most accurate available estimates of sensitivity and specificity for the involved instruments and our situation, when available, source studies included a population-based sample, used structured diagnostic interviews as reference standard, had a sample size of >100 and >50 positive cases. Additionally, we preferred studies with mean symptom scores that were comparable to those in our population (based on participants included up to 30 June 2024), and where we deemed the study population similar to our study population. We favoured primary studies over meta-analyses because we were better able to assess similarity between our population and a single well-characterised sample than to compare our population with the heterogeneous aggregate of samples that underlie a meta-analysis. We conducted sensitivity analyses to ensure that our results were robust by using alternative studies as sources of diagnostic accuracy metrics for adjustment of prevalence estimates.

For the PCL-5, we chose a cut-point of ≥31 with a sensitivity of 0.94 and a specificity of 0.94.25 The cut-point for the GAD-7 was ≥10, the assumed sensitivity 0.89 and specificity 0.82.26 For the PHQ-9, we chose a cut-point of ≥10 and assumed a sensitivity of 0.88 and specificity of 0.88.27

We conducted all analyses in R V.4.4.2 https://www.r-project.org/

Role of the funding source

The funders of the study had no role in study design, data collection, data analysis, data interpretation or writing of the report.

Patient and public involvement

Together with the support of nearly three dozen stakeholders from Switzerland and Ukraine, including over five Swiss universities and organisations, Ukrainian universities, members of the International Red Cross, representatives from the Psychiatric Hospital of the University of Zurich, the Pediatrics Children’s Hospital of Zurich and public health officials from both countries, we hosted a full-day working session in March 2023 to finalise a Ukrainian-specific mental health digital surveillance protocol. The experts included specialists in public and mental health, political science and health policy, as well as representatives from governmental (Ukrainian and Swiss) and non-governmental organisations (eg, International Red Cross, UNESCO), social media specialists and Ukrainian citizens.