A cross-sectional online survey was administered in English, the official language in Singapore, to adult residents (age > 21) who are members of a national web panel curated by Kantar Profiles Division. This cross-sectional survey was part of a broader exercise to quantify the prevalence and economic burden of depression and anxiety in both adults and children. Participants were recruited between April 20, 2022 and June 1, 2022 through email invitations and asked to fill out a screener questionnaire to identify symptoms consistent with depression or anxiety disorder among all household members as indicated by the PHQ-4 [16] (Supplementary Appendix A). Participants were asked to fill out the screener for themselves and for all members living in their household. If the main respondent did not have symptoms consistent with depression or anxiety and at least one youth between the ages of 4 and 21 in their household did, then the main respondent filled out a detailed questionnaire on behalf of the oldest child. This manuscript focuses exclusively on results for youth. IRB approval was obtained prior to fielding the survey (IRB# 2021 − 836).

The first case of COVID-19 in Singapore was detected in January 2020 [17, 18]. Community cases swelled in March 2020 and a quarantine, termed locally a “circuit breaker”, was imposed on April 7th 2020 for two months where public movement was limited to essential activities [19]. Pre-schools, schools, and universities introduced home-based learning at this time [20, 21]. In the following months schools reopened but social gatherings were limited to control the spread of subsequent outbreaks and mask mandates were in effect in public spaces. In January 2021, vaccination programmes began and by December 2021, 80% of the population was fully vaccinated. By March 2022, international travel restrictions were lifted and by May 2022, almost all pandemic-related infection control measures were lifted. However, indoor mask mandates remained in effect to February 2023, when Singapore considered the virus endemic [22].

The PHQ-4 is the combination of the PHQ-2 and GAD-2, which are validated two-item, two-week recall period, ultra-brief screeners shown to have high sensitivity and specificity for identifying depression and anxiety disorder [16, 23,24,−25]. The PHQ-4 includes four questions about the frequency of which respondents feel: 1 - Nervous, anxious or on edge; 2 - Not being able to stop or control worrying; 3 - Down, depressed, or hopeless; and 4 - Little interest or pleasure in doing things. Responses were recorded on a 4-point Likert scale, including: 1 - Not at all, 2 - Several days, 3 - More than half the days, 4 - Nearly every day. Although the PHQ-4 is not a diagnostic tool for depression or anxiety, its brevity allows for rapid identification of individuals who have symptoms consistent with these conditions. Moreover, the PHQ-2 and GAD-2 are validated against the broader PHQ-9 and GAD-7 across multiple countries in adolescents [26,27,28,29]. At the cutoff scores of 3 or greater, PHQ-2 had a sensitivity of 74% and specificity of 75% for detecting youth who met Diagnostic and Statistical Manual of Mental Disorders [26]. GAD-2, at cutoff scores of 3 or greater, had specificity and sensitivity of 0.84–0.87 and 0.93–0.95 respectively against the GAD-7 [30].

Parent proxies were asked to complete the full survey for their oldest eligible child only if they themselves did not have symptoms of depression or anxiety. This approach limited survey fatigue as parents had to fill the survey out only one time regardless of how many individuals qualified in the household. Parent proxy-reported psychosocial measures have a high degree of reliability and validity to patient-reported outcomes in pediatric populations [31,32,33]. The full survey (available in Supplementary Appendix A and B) included the following domains: Mental Healthcare Utilization, Productivity Losses from Absenteeism and Presenteeism, Quality of Life Measures, and Preferences for Peer Support. The full survey also included socioeconomic questions about the youth (e.g. age, gender, education level, and about the primary caregivers (e.g. employment status and monthly income of caregivers).

To estimate prevalence among youth and to determine eligibility for the full survey instrument, panelists were asked to report the ages of all youth in the household and complete the PHQ-4 for each child. A child was assumed to have symptoms consistent with anxiety if they scored 3 or higher on the anxiety sub-scale (sum of items 1 and 2) [34]. Similarly, a child was assumed to have symptoms consistent with depression if they scored 3 or higher on the depression sub-scale (sum of items 3 and 4). Previous research has established that a score of 3-or-greater on the Depression and Anxiety subscales represent a reasonable cut-point for identifying potential cases of major depression and generalized anxiety respectively [34]. Prevalence rates were calculated by dividing the number of children who have symptoms consistent with anxiety or depression by the total number of reported children across all households in the screener. We present overall estimates for each condition and report the percentage ‘never being told by a healthcare professional that the child has depression or anxiety’, and thus who likely remain untreated. All results are based on the parent proxy responses.

Before conducting any analyses, data was cleaned and suspicious responses were re-coded as missing. We recoded 17 (1.7%) and 38 (3.8%) responses from healthcare resource utilization and school absenteeism, respectively, to missing. This is due to respondents reporting some form of healthcare visits or missed school hours despite indicating “No healthcare utilization” and “No absence from school”. Details on data cleaning are available in Supplementary Appendix C. Children who met the threshold for PHQ-4 and had full survey data were included in the healthcare resource utilization analysis. To quantify healthcare utilization attributable to depression or anxiety, participants were asked about the frequency of physician and outpatient visits (including tele-visits), medications, and alternative therapies (e.g., acupuncture, reflexology) for their eligible child. For these questions, the recall period was three months. Other questions focused on diagnostic tests, emergency department visits, and number and duration of hospitalizations. For these we used a recall period of twelve months as these episodes are less frequent and easier to remember [35]. To monetize healthcare utilization, unit costs were applied to each type of service based on unsubsidized costs collected through publicly available sources. Full breakdown of unit costs and assumptions are available in Supplementary Table 1. Per capita healthcare cost estimates were taken by averaging across responses. Total cost estimates were generated by multiplying the total youth population (between ages 4 and 21) from the Singapore Department of Statistics, by our estimated prevalence rates and then by the per capita cost estimates.

School absenteeism was determined by multiplying the estimates of weekly hours missed from school due to depression and anxiety by the number of weeks in a school year (i.e., 40 weeks). School performance was captured by a question asking parent proxies the degree to which the youth’s depression or anxiety symptoms affected performance at school on a scale of 0–10 with 0 being “no symptoms or symptoms had no effect on my child’s school performance” and 10 being “symptoms completely prevented my child from attending school”. Similarly, performance of regular daily activities was captured by a question asking parent proxies the degree to which the youth’s depression or anxiety symptoms affected their ability to do his or her regular daily activities (other than school performance) on a scale of 0–10 with 0 being “no symptoms or symptoms had no effect on my child’s regular daily activities” and 10 being “symptoms completely prevented my child from his or her regular daily activities”.

All costs are reported in 2022 Singapore dollars (S$) and also reported in 2022 USDFootnote 1. 95% confidence intervals (CI) for prevalence and mean results are estimated using standard approaches. To obtain the 95% confidence interval for the total cost estimate, which is a function of both prevalence and unit costs, we ran 1,000 Monte Carlo simulations. In each run, we drew a prevalence proportion (pi) from a beta distribution centered at the sample prevalence proportion and standard deviation based on the standard error of the prevalence estimate. We then calculated total number of youth with symptoms consistent with depression or anxiety (ki) as the product of pi and the Singapore population aged 4 to 21. We then drew ki observations with replacement from the distribution of costs in our sample and summed to generate the total cost for each iteration. 95% CI interval was calculated based on the 2.5th and 97.5th percentile of the total costs obtained from the 1,000 simulations. Given results showing that depression and anxiety symptoms co-occur in the vast majority of cases, we present burden estimates for depression and anxiety combined. All analyses were conducted in Stata/SE 17.0 (College Station, Texas, United States).