Increase in Suicidal Thinking During COVID-19

There is concern that the COVID-19 pandemic may cause increased risk of suicide. In the current study, we tested whether suicidal thinking has increased during the COVID-19 pandemic and whether such thinking was predicted by increased feelings of social isolation. In a sample of 55 individuals recently hospitalized for suicidal thinking or behaviors and participating in a 6-month intensive longitudinal smartphone monitoring study, we examined suicidal thinking and isolation before and after the COVID-19 pandemic was declared a national emergency in the United States. We found that suicidal thinking increased significantly among adults (odds ratio [OR] = 4.01, 95% confidence interval [CI] = [3.28, 4.90], p < .001) but not adolescents (OR = 0.84, 95% CI = [0.69, 1.01], p = .07) during the onset of the COVID-19 pandemic. Increased feelings of isolation predicted suicidal thinking during the pandemic phase. Given the importance of social distancing policies, these findings support the need for digital outreach and treatment.

daily time spent at home (in hours) per day per participant. Sufficient location data were available for 25 participants. Of the remaining 29 participants, two participants declined to participate in this component of the study at the time of enrolling. Of participants who agreed to this portion of the study, no data were available for five, the app stopped gathering data partway through study participation and before the pandemic phase for 14, and data quality were too low (too few location records were recorded) for imputation of missing data for nine. This level of missingness resulted from the fact that the application was not used for surveys and was running purely in the background; in this situation, the operating system will typically shut it down in order to conserve phone battery, CPU usage, and memory.

Supplemental Results
Full results and model comparisons of all ordinal flexible-threshold mixed models testing the associations between pandemic phase and site with self-reported suicidal thinking are reported in Table S1. In Model 1, pandemic phase was entered as a sole predictor. In Model 2, recruitment site (adolescents vs. adults) was entered as an additional predictor. In Model 3, the interaction between recruitment site and pandemic phase was entered as an additional predictor.
Site did not significantly predict suicidal thinking, whereas the interaction of pandemic phase did, and likelihood ration tests revealed that Model 3 significantly improved model fit.
Full results and model comparisons of ordinal flexible-threshold mixed models testing the association between pandemic phase and social isolation are reported in Table S1. These models tested the effect of pandemic phase alone on social isolation (Model 4), the addition of recruitment site (adolescents vs. adults (Model 5), and the addition of their interaction (Model 6).
Likelihood ratio tests demonstrate that adding parameters for site or the interaction of site and pandemic phase did not significantly improve model fit.
Full results and model comparisons of all ordinal flexible-threshold mixed models testing the association between social isolation and self-reported suicidal thinking are reported in Table   S2. These models tested the effect of social isolation alone on suicidal thinking (Model 7), the addition of pandemic phase (Model 8), and the addition of their interaction term (Model 9).
Likelihood ratio tests indicated that Model 6 was the superior fitting model.
There was substantial variability in how many data points were contributed among participants (average number of surveys per included participant = 346, range = 17-636, overall percentage of surveys completed 43.60%). Each participant was sent six surveys per day for the three months after presentation at the hospital and then one survey per day for the following three months. All participants had missing data and we made the choice a priori to include all possible participant data regardless of whether they had completed their 6 months of participation. Additionally, due to a glitch in the survey-delivery app 1 participant received twice as many survey prompts as intended during a portion of study participation, resulting in a contribution of more data points than any other participant. We ran all analyses with and without this participant included, and the pattern of significant results in all analyses were the same with one exception: the non-linear effect of time and categorical effect of pandemic phase was no longer significant using a GAMM.
Results of linear mixed models testing for effect of pandemic phase on hours spent at home and effect of hours spent at home on suicidal thinking are reported in Table S3.
Additionally, our pattern of primary results was unchanged when using only an item assessing suicide urge rather than the sum of suicide urge with intent.