Our MHRN Suicide Prevention Outreach Trial offers outreach and support interventions based on patients’ response to item 9 of the PHQ9 depression questionnaire.  We reach out to patients who reported thoughts of death or self-harm “more than half the days” or “nearly every day” at a recent visit.  This threshold for outreach is based on our previous research showing that those patients reporting frequent suicidal thoughts have an approximately 2.5% risk of suicide attempt over the following year.  Now, several of our MHRN health systems are using the same threshold (a score of 2 or 3 on item 9 of the PHQ9) to identify patients needing additional assessment and follow-up.

We also know from our previous research that the risk implications of responses to that question differ by race and ethnicity.  Among those scoring 2 or 3 on PHQ9 item 9, the one-year risk of suicide attempt is approximately 3% in non-Hispanic Whites and Native Americans/Alaskan Natives compared to 1.25% in African Americans, Asian-Americans and Hispanics.  This is not surprising given racial and ethnic differences in overall suicide rates (pictured above).  In epidemiologic terms: response to item 9 of the PHQ9 implies approximately the same relative risk of suicidal behavior across racial and ethnic groups.  But it implies a lower absolute risk in racial and ethnic groups with a lower baseline risk.  In practical terms: If we set a 2% risk threshold for active outreach programs, then we would offer those extra services to non-Hispanic White patients who report thoughts of self-harm at least “more than half the days”.  But we would only offer outreach services to African-American, Asian-American, and Hispanic patients who report such thoughts “nearly every day”.

The idea of race-based thresholds for outreach and support to prevent suicide attempt should certainly give us pause.  We’d be especially concerned if the outcome of race-specific thresholds was a higher threshold for offering services to traditionally disadvantaged or underserved groups.  Our comfort with race-based thresholds would probably depend on whether we consider outreach to prevent suicide attempt a clear benefit or a potential burden.  At this point, we cannot say that persistent outreach based on response to the PHQ9 prevents suicide attempts or suicide death.  Whether active outreach actually prevents suicide attempt is the question that motivates our ongoing trial.  We can, however, say that persistent outreach is not completely free of burden.  A small proportion of patients receiving our outreach messages find them an intrusion or invasion of privacy.  We would only implement outreach programs when benefit outweighs that potential harm.  And the benefit of outreach is likely to be less in racial and ethnic groups with lower overall risk of suicide attempt.  Again, using epidemiologic terms:  For any relative risk reduction, number-needed-to-treat for benefit depends on the baseline risk of the outcome.

Thoughtfulness and transparency around this issue will be even more important as we begin to apply more complex algorithms to identify people at risk for suicide attempt and suicide death.  Complex calculations don’t resolve the ethical question, but they can obscure it.  Before we adopt risk measures derived by machine learning, we should decide whether machines should be allowed to learn based on race and ethnicity. 

We certainly accept the use of race and ethnicity in other risk assessments, such as the Gail model for estimating breast cancer risk.  Perhaps we accept race-based cancer prevention because we think of racial differences in cancer risk to be more related to biology or genetic variation.  And we understand that unnecessary testing or treatment for cancer can clearly have harms.  Are racial and ethnic differences in suicide risk a similar story?  Or a very different one?