Mapping the timescale of suicidal thinking

Abstract

This study aims to identify the timescale of suicidal thinking, leveraging real-time monitoring data and a number of different analytic approaches. Participants were 105 adults with past week suicidal thoughts who completed a 42-d real-time monitoring study (total number of observations = 20,255). Participants completed two forms of real-time assessments: traditional real-time assessments (spaced hours apart each day) and high-frequency assessments (spaced 10 min apart over 1 h). We found that suicidal thinking changes rapidly. Both descriptive statistics and Markov-switching models indicated that elevated states of suicidal thinking lasted on average 1 to 3 h. Individuals exhibited heterogeneity in how often and for how long they reported elevated suicidal thinking, and our analyses suggest that different aspects of suicidal thinking operated on different timescales. Continuous-time autoregressive models suggest that current suicidal intent is predictive of future intent levels for 2 to 3 h, while current suicidal desire is predictive of future suicidal desire levels for 20 h. Multiple models found that elevated suicidal intent has on average shorter duration than elevated suicidal desire. Finally, inferences about the within-person dynamics of suicidal thinking on the basis of statistical modeling were shown to depend on the frequency at which data was sampled. For example, traditional real-time assessments estimated the duration of severe suicidal states of suicidal desire as 9.5 h, whereas the high-frequency assessments shifted the estimated duration to 1.4 h.

Keywords: ecological momentary assessment; suicidal thinking; suicide.

Fig. 1.

Overview of sampling design. (A) combined EMA and burst measurements; (B) EMA-only dataset; (C) Burst-only (short) dataset; and (D) Questions asked about suicidal thinking during real-time measurements

Fig. 2.

Descriptive statistics across different types of real-time measures of suicidal thinking. (A) The distribution of EMA responses for desire and intent; (B) The distribution of burst responses for desire and intent; (C) The means of desire and intent across response-type; (D) The standard deviations of desire and intent across response-type.

Fig. 3.

In panels (A) and (B), we show a time-series depicting the first two weeks’ worth of responses on the Desire variable for two different participants. In panel (A), the time-series is taken from a participant with low variability, indicated by $p_{\mathit{mode}}= 0.91$ . The red boxes indicate periods of consecutive measurement occasions in which the participant gave a nonzero response on the Desire item, referred to in the main text as episodes of elevated desire. In panel (B), the time-series is taken from a participant with high variability, indicated by $p_{\mathit{mode}} = 0.22$

Fig. 4.

Proportion of consecutive observations that show variation as a function of time between observations for desire (A) and intent (B). Each red dot represents an individual participant, with the light gray lines connecting values of the same participant across timescales. The gray diamond represents the mean proportion in a given timescale (Desire = {0.422, 0.539, 0.631}, Intent = {0.442, 0.535, 0.610})

Fig. 5.

Continuous-time vector autoregression results. Panel (A) depicts the estimated drift matrix fixed effects, with CIs, as a network. Panel (B) shows how the model-implied lagged regression coefficients are dependent on the time interval between measurements $.$ Shaded lines represent 95% credible intervals. Panel (C) shows the model-implied Impulse Response Function, that is, how the model predicts the values of Desire and Intent to change over time given an impulse value of (Desire = 1 and Intent = 0) as indicated by the filled diamond.

Fig. 6.

Transition Probabilities (at a 1-h interval, Left) and Sojurn Times (Right) from the Continuous Time Markov Models. Clocks represent sojurn time duration in blocks and fractions of 12 h. In the networks, transition probabilities point estimates that are smaller than or equal to 0.001 in absolute value are omitted. (A) Desire Markov model estimates; (B) Intent Markov model estimates.

Fig. 7.

Continuous-time vector autoregression results across EMA and short datasets. (A) EMA dynamic network; (B) Short dynamic network; (C) EMA lagged effects; (D) Short lagged effects; (E) EMA impulse response function; (F) Short impulse response function.

Fig. 8.

Transition Probabilities (at a 1-h interval) and sojurn times from continuous time Markov models across EMA and short datasets. (A) Desire EMA Markov model estimates; (B) Desire short Markov model estimates; (C) Intent EMA Markov model estimates; (D) Intent short Markov model estimates. Note: clocks represent sojurn time duration