Childhood Trauma and Youth Suicide Rates

Childhood Trauma and Youth Suicide Rates

Note: this is part of the Youth Suicide Rise project.

Does exposure to traumatic events during childhood substantially increase the risk of suicide several years later?

If such strong long-term effects are true, then the correlation of child suicide with overall suicide could increase if we instead average the overall rate across several previous years.  After all, the suicide rate in any given year indicates the likelihood that children have been exposed to suicide within family or community that year, a highly traumatic event.

Teenage Child Suicide 1997-2018

Let us compare visually how well the concurrent and past rates correspond with teenage child rates:

Note: the scales were modified to move the curves closer to each other

Here we see that the Child curve (blue) follows considerably closer the Past 7 Years curve (green) than it does the Concurrent Rates curve (red).

It is interesting to note that the greatest deviation of the blue curve from the green occurs in 2004 and 2005, precisely the years when child suicide is believed to have been affected by a youth-specific event (the black-box warnings controversy).

Visual inspections can mislead; let us include calculations:

Here the visual observation is confirmed by calculation: the R squared -- an measure of how closely the points fit with the regression line -- is much lower for the red Concurrent rates than it is for the green Past rates.

It is important to keep in mind that knowing the average rate over the past 7 years says nothing about the recent trend -- it does not tell us if rates were falling and increasing.  Furthermore, we are using past values of the general suicide rate, not child rate. And finally, we are not including the current year in the past rate -- the past rate is calculated strictly from preceding years.

In view of the above the correlation is remarkable both due to the close fit and due to the huge improvement over concurrent rates.

Youth Suicide 1990-2018

Will this result hold if we look further back?  The 1990s were a time of child suicide rates falling down as rapidly as they have climbed recently -- will the close relationship with past all-age rates break during this era?

Since the CDC WISQAR tool does not allow Custom Age Range selection before 1990, we have to switch to the predefined Age Group 15-19 (Youth) and see if our finding extends not only to the 1990s but also to a slightly older group of teens:

Once again we see that the blue curve (youth aged 15-19) is closer to the green curve (previous 7 years) than to the red curve (concurrent rates).

Let us confirm the visual observation with calculation:

Once again we see a very high correlation with the past and a huge improvement over the correlation with concurrent rates (R squared 0.92 versus 0.42).

Scale

It is important to note the scale manipulation used to position the trend curves closer to each other in the charts above: the all-ages scale on the left covers half the distance of the youth scale on the right.

What this means is that every change in all-ages rates is associated with a proportionally double change in youth rates:  when the all-ages past rate increases by 1 percentage point, the expected change in youth rates would be an increase of 2 percentage points.

This fact will be important once we discuss causality and impact.

Robustness

We need to keep in mind the possibility that the correlation was so high due to chance -- after all this is not a result based on a large number of data points. To test robustness, we should look further and deeper, e.g. longer into the past, or at international data, or at state data, or at subgroups such as girls.  If chance played a strong role here than we should quickly encounter far smaller correlations.

Implications

Even if the relationship we discovered is highly robust, great care must be taken regarding causality.  We will discuss potential implications of the high correlation in the next post.

Technical notes:

All rates are age-adjusted; this has a (fairly small) effect only on the all-ages rates.

Stretching the scales in the trend charts to position the curves close to each other is a geometric form of calculating linear regression.  To better understand this, note that the equations of both lines in the Correlation of Youth chart was roughly y = 0.4 x + 8, which in turn is close to the relationship between the left and right scales on the preceding Youth trends chart.

The scales on the correlation charts were not stretched -- they were merely moved in order to separate the two groups of rate points (thus the distances of the points from the regression line were not affected).