Last year for World Suicide Prevention Day, I wrote about how we have little evidence — for or against—differences in the functioning of the brains of those who experience suicidality. The reason for this was that existing studies that used functional magnetic resonance imaging (fMRI) and/or electroencephalography (EEG) were severely underpowered and used tasks that result in relatively poor reliability. Since then, using EEG I have continued to look into whether there are neural differences in people experiencing suicidal thoughts and who engage in suicidal behavior. From this work, a clearer, although still messy, picture is starting to emerge. What I have found so far has led me to seriously doubt that we will find a neural function that reliably differs in all people who experience suicidal thoughts and behaviors. To help you understand how I have gotten to this point, I’m going to tell you about some of my work over the last couple of years, and where we can go from here.
Event-Related Potentials and Early Unpublished Work
I started studying neural differences in suicide in Spring 2018 when I took a course on event-related potentials (ERPs). ERPs are waveforms in the EEG that appear during specific events. For example, imagine a task where you are presented with X’s and O’s, one at a time in random order on a computer screen while your EEG is recorded. Your job is to count the number of O’s, which are more rare than the X’s. Every time an X or an O appears on the screen, the computer marks in your EEG recording when it appeared. If we averaged all of the waveforms of your EEG together after an O appeared, you’d see a waveform that looks a lot like the one below, where time zero is when the O first appeared on the screen. Each of the labeled waveforms (e.g., P1, N1), are ERPs that reflect a neural process. If we averaged your EEG to when the X’s appeared and compared that waveform to the averaged waveform when O’s appeared, you would see something interesting. You’d see that the ERP that we call the P3 would be much bigger to when O’s appeared compared to when X’s appeared. What the neural function the P3 specifically represents is debated, but some think it is your brain’s response to a stimulus that is significant or salient to you in a given context [1]. In this case, the O’s are more significant or salient because of the nature of the task.
Getting ERPs is a little more complex than averaging the single-trial EEG waveform together though. We have to deal with other things that we pick up in the EEG that we aren’t interested in, like eye-blinks or electrical noise from lights in the room. And it was these more complicated bits that I was learning how to do in this ERP course. At the end of the course, we got to take a large data set and try and answer a research question using ERPs. I, of course, was interested in correlating ERPs with measures of suicidal thoughts and behaviors. The problem was, I was not finding any correlations. Because I knew that journals almost never publish papers where there isn’t a significant difference or correlation, I tried just about every combination of ERP and suicide measure in the data set. Still, I found no correlation. Because I was aware of tons of studies that found significant correlations between suicide and the ERPs I was looking at, I figured it was just my data set and never published my findings.
Meta-Analysis Study
As I learned more about the replication crisis in neuroscience and psychology and the problem with scientific journals only publishing significant results (along with the p-hacking I had unsuccessfully implemented), I began to wonder if my experience in my ERP course was actually representative of the field. Fast forward to 2020, and we posted a preprint (i.e., a scientific paper that has not yet completed peer review that is freely available) that is a meta-analysis of all of the studies that have looked at whether those who experience suicidal thoughts or who engage in suicidal behavior show differences in any ERP compared to controls [2]. In simple terms, a meta-analysis is a process where we review the entire literature, take the numbers showing the size of the effect (in this case, usually the mean difference in the amplitude of an ERP), and then average together these effects to get a better idea what the “true” effect is. In our original analyses, we found small-to-moderate differences in ERPs between those who experience suicidality and those who don’t. I was excited about these findings because they showed that we might be able to find a neural difference in people experiencing suicidality, we would just need larger sample sizes to find these effects. But, reviewers suggested that we change our analyses to improve the paper. At first we disagreed, but upon reflection and further feedback from our reviewers, agreed that these new analyses were much better than our original approach. With this new approach, we found that almost every single kind of ERP was not different in those experiencing suicidality, and the one case that we did had an infinitesimally small effect that was almost surely explained by depression rather than suicidality. Our meta-analysis suggests that based on the current ERP and suicide literature, there are no differences in those experiencing suicidality.
Bayesian Statistics Study
Since last World Suicide Prevention Day, some of my colleagues and I have been working on a paper building on these results. We wanted to conduct a study where we could compare the relative evidence between there being a difference in any ERP in those experiencing suicidality, and there not being an effect. You see, the traditional analyses scientists use, including the kind of analyses in our meta-analysis, can’t provide evidence for no effect because of their underlying assumptions. Instead, we can only get evidence for if there is an effect. If our results are not “statistically significant” (i.e., below that magical threshold of p = .05), then we cannot technically say there is no effect, just that we failed to reject the hypothesis that there is no effect. To get around this, we used a different kind of analysis that uses Bayesian statistics.
Bayesian statistics are powerful for many reasons, but two of them are: (1) they can quantify which hypothesis (that there is an effect vs. there is no effect) is more likely and (2) they can take into account your prior beliefs about whether there is an effect. How it does this is complex, but think of it this way using the figure below. First, we have our prior beliefs. Then, we get the data. Combining our prior beliefs and the evidence from the data, we get a posterior belief, which quantifies the relative evidence between our two hypotheses, telling us which is more likely. We used this approach in a new sample of adults who all had depression. Some also had current suicidal ideation or had previously attempted suicide, while the rest had not. For our prior distribution, we used the effect sizes from our first version of our meta-analysis that found small-to-moderate effects. This essentially biased our analyses to find that the there was more evidence for an effect versus that there isn’t an effect. We then compared people who had suicidal ideation or a previous suicide attempt with controls who were also experiencing depression, but without suicidality, across four different kinds of ERPs. We found that the hypothesis that there is an effect was about 2/5’s to 1/2 as likely as that there is no effect. In other words, we found more evidence that there is no difference than that there is a difference across all four of the ERPs we measured.
The Future of Studying ERPs in Suicidality and Suicide Prevention
Now, what does this all mean for the study of suicidal thoughts and behaviors? Though the work I’ve described in no way confirms this for sure, I have serious doubts that there are any reliable neural correlates of suicidal thoughts and behaviors that will be found across all people. There are avenues still yet to be explored, including many that I discussed in my first post, such as increased sample sizes through collaboration, or examining whether there are “phenotypes” or groups of suicidality that have distinct neural correlates. These are questions that many would argue for before concluding there are no neural differences in suicidality, and I agree. However, I think it is time that the scientific community starts seriously considering the possibility that there are no neural differences in suicidal behavior and what this might mean about how we understand suicide. For example, our field is largely based on a medical model that places psychiatric disorders at the center of how we understand suicide. While I think mental illness often plays a significant role, this does not mean that this is our best approach to understanding suicide. Instead, suicide may be a learned behavior that, even in the most dire of life circumstances, is not seriously considered by most people due to things like fear of death (a theory that I will probably be chasing my entire career). This does not mean we should stop trying to prevent suicide, but it may mean that in addition to better and more accessible mental health care, we should be striving for other societal changes that can help reduce suicide. This includes decreased access to lethal means, better access to healthcare and housing, and helping with a host of other social problems. In short, while we should not abandon neuroscience research, we should seriously question our current paradigm, and also make sure we are advocating for policies that will help people feel like there is a reason to live.
References
- Hajcak, G., & Foti, D. (2020). Significance?… Significance! Empirical, methodological, and theoretical connections between the late positive potential and P300 as neural responses to stimulus significance: An integrative review. Psychophysiology, e13570. https://doi.org/10.1111/psyp.13570
- Gallyer, A. J., Dougherty, S. P., Burani, K., Albanese, B. J., Joiner, T. E., & Hajcak, G. (2020). Suicidal thoughts, behaviors, and event-related potentials: A systematic review and meta-analysis. BioRxiv. https://doi.org/10.1101/2020.04.29.069005
Austin J. Gallyer (@AGallyer) is a doctoral candidate in Neuroscience at Florida State University, United States. Email: agallyer@neuro.fsu.edu.
netecr.org 