Five-Year Suicidal Ideation Trajectories Among Women Receiving or Being Denied an Abortion

Our main outcome variables rely on two validated suicidal ideation items from the Brief Symptom Inventory (BSI) (21) and the Patient Health Questionnaire (PHQ-9) (22). The BSI item asks respondents to indicate whether in the past 7 days they have had thoughts of ending their life, with choices ranging from 0, “not at all,” to 4, “extremely.” The PHQ-9 item asks respondents, “Over the last 2 weeks, how often have you been bothered by any of the following problems? Thoughts that you would be better off dead or of hurting yourself in some way?” Options ranged from 0, “not at all,” to 3, “nearly every day.” The PHQ-9 was included from the second interview wave (6 months after abortion seeking) forward. The two main dichotomous outcomes included any BSI symptoms of suicidal ideation (“extremely,” “quite a bit,” “moderately,” or “a little bit” versus “not at all”) and PHQ-9 symptoms of suicidal ideation (“nearly every day,” “more than half the days,” or “several days” versus “not at all”). Average BSI and PHQ-9 symptoms (continuous) are presented in Table S1 in the online supplement.

Our secondary outcome measure, imminent suicidality, was assessed with an adapted version of the Sheehan Suicidality Tracking Scale (23) among respondents indicating any suicidal ideation symptoms on either the BSI or the PHQ-9. The adapted scale included six questions: “Over the past week, how seriously did you: Think that you would be better off dead or wish you were dead? Want to harm yourself or to hurt or injure yourself? Think about suicide? Plan for a suicide? Take active steps to prepare for a suicide attempt in which you expected or intended to die?” and “Over the past week, did you injure yourself on purpose, in an effort to kill yourself?” As recommended by the scale authors, if the participant raised the possibility of suicide herself or responded “moderately,” “very,” or “extremely” to all of the items and/or “yes” to the last item, she was considered imminently suicidal.

Careful steps were taken to support women identified to be at risk of suicide. During the consent process and at the outset of each interview, participants were informed of the limits to confidentiality, including possible intervention and loss of confidentiality in the event that they expressed a plan to harm themselves. In cases where the participant met the criteria for being imminently suicidal or showed indications that she might inflict self-harm, research assistants immediately implemented a protocol to further evaluate the participant’s risk level and ensure her timely access to social and professional support.

Our primary independent variables of interest were a four-part study group variable (near-limit, turnaway-birth, turnaway-no-birth, and first-trimester), time (in years) since recruitment, and study group-by-time interactions. The near-limit group served as the reference group to allow for simultaneous comparisons with both the turnaway-birth and first-trimester groups. The time variable served to assess whether near-limit outcomes changed significantly over time. The study group-by-time interaction assessed whether a study group’s trajectory differed significantly from that of women in the near-limit group.

We included age, race/ethnicity (white, black, Hispanic or Latina, and other), education level (less than high school, high school or equivalent, associate’s degree or technical school or some college, and college degree or higher), employment (full-time or part-time and not employed), parity (nulliparous, birth in the past year, one child and no birth in the past year, two or more children and no birth in the past year), and marital status (single, married, and divorced or widowed) as demographic covariates. Additional covariates included other factors known to be associated with suicidal ideation during pregnancy (12, 24): ever diagnosed with depression or anxiety (whether a doctor or health professional had ever told them that they had a “depressive disorder such as major depression, depression, dysthymia, or bipolar disorder” or an “anxiety disorder, including panic, obsessive-compulsive, anxiety, and posttraumatic stress”); experiences of intimate partner violence in the past year, including psychological (“frightened for your safety as a result of anger or threats made by another person”) and physical (“pushed, hit, slapped, kicked, choked, or physically hurt in any way”); ever been sexually assaulted or raped; history of child abuse or neglect (physical abuse, neglect, or sexual abuse during childhood); any prepregnancy illicit drug use; and any prepregnancy problem alcohol use (drinking first thing in the morning or inability to remember what happened after drinking). Gestational age at the time of recruitment was not included as a model covariate because it was highly correlated with study group, by study design. All covariates were measured at baseline and were time invariant.

We used mixed-effects regression analyses, accounting for clustering by recruitment site, to assess baseline differences in demographic characteristics and mental health history between women in the near-limit abortion group and the three other study groups. For categorical variables with more than two categories (e.g., race/ethnicity), we used an omnibus postestimation test to accommodate multiple category associations. For longitudinal analyses, we used mixed-effects logistic regression to assess whether trajectories in experience of any suicidal ideation differed by study group. In supplemental analyses, we modeled mean suicidal ideation symptoms using mixed-effects linear regression analyses. Longitudinal analyses accounted for clustering by site and individual. Mixed-effects models produce unbiased estimates even in the context of missing observations. They also accommodate irregular time measurement and clustering by recruitment site and individuals, as in the case of our panel data. We assessed whether trajectories were curvilinear by testing whether including quadratic terms for time improved the model fit. None of the models required quadratic terms for time, as indicated by a nonsignificant (p>0.05) likelihood ratio test. For all longitudinal analyses, we ran unadjusted and adjusted models. The threshold for statistical significance was set at a p value of 0.05.

To determine whether women with suicidal ideation or related risk factors were more likely to discontinue participation in the study, we used mixed-effects regression to test associations between previous reports of suicidal ideation, history of depression or anxiety, history of child abuse or neglect, and past-year history of intimate partner violence with being lost to follow-up. To determine whether women lost to follow-up may have been lost because of suicide, we conducted verbal autopsies when deaths were reported to us by participants’ relatives during data collection. After data collection concluded, we used LexisNexis, a public records search tool, to confirm those deaths and identify possible new deaths. Both protocols were approved by our institutional review board, and verbal autopsies were conducted only with secondary contacts with whom the participant had previously given us permission to speak. Using a secure server, we provided LexisNexis with names, dates of birth, and last known addresses for all participants lost at any point after recruitment. For each death identified, we conducted a Google search for obituaries or other information about the death and sought death certificates from the city or county where the death occurred.

To account for the possibility that some women with suicidal ideation symptoms may have skipped the suicidal ideation items to avoid the research team’s mandatory reporting requirements, we ran an additional set of longitudinal analyses. In these analyses, anyone who skipped a suicidal ideation item and indicated moderate signs of depression (as indicated by a score ≥8 on either the full BSI or PHQ-9 depression scales) was treated as having suicidal ideation symptoms.

The proportion of women in the near-limit group who had any BSI suicidal ideation did not differ significantly from the three other study groups at any point in time (Table 2). However, for women obtaining an abortion (those in the near-limit and first-trimester groups), the proportion with any BSI suicidal ideation declined significantly from 1 week to 5 years after abortion (Table 2 and Figure 1). Neither turnaway group experienced a significant decline in any BSI suicidal ideation (Table 2). The results of the adjusted and unadjusted mixed-effects logistic regression analyses are presented in Tables 3 and 4, respectively. Baseline characteristics associated with having any BSI suicidal ideation (Table 3) included being Latina (adjusted odds ratio=3.59; 95% CI=1.44, 8.95), a history of intimate partner violence (adjusted odds ratio=3.34; 95% CI=1.26, 8.87), a history of depression or anxiety (adjusted odds ratio=2.96; 95% CI=1.37, 6.38), and prior problem alcohol use (adjusted odds ratio=3.38; 95% CI=1.11, 10.23). Five women skipped the BSI suicidal ideation item at one of their interviews, two of whom completed the remainder of the BSI-depression subscale. The results of analyses in which the women who skipped the suicidal ideation item and showed indications of moderate depression were recoded as suicidal were similar to those of the main analyses (data not shown).

The proportion of women in the near-limit group with any PHQ-9 suicidal ideation did not differ significantly from the three other study groups at any time interval (Table 2). The first-trimester group was the only group to experience a significant (p=0.01) decline in any PHQ-9 suicidal ideation, declining from 1.89% at 6 months to 0.35% at 5 years (Table 2). Baseline characteristics associated with any PHQ-9 suicidal ideation (Table 3) included parity, history of depression or anxiety (adjusted odds ratio=4.07; 95% CI=1.96, 8.46), having a prior history of problem alcohol use (adjusted odds ratio=3.00; 95% CI=1.04, 8.68), and education level. Seven women skipped the PHQ-9 suicidal ideation item at one of their interviews, five of whom showed signs of moderate depression. In analyses where these five women were recoded as having suicidal ideation symptoms, results were similar to those of the main analyses (data not shown).

In 109 of 7,247 interviews, women reported any suicidal ideation symptoms, either on the BSI or PHQ-9 suicidal ideation items, over the 5-year study period, and these women completed the Sheehan Suicidality Tracking Scale to assess whether they were imminently suicidal. In four of the 109 observations (two each in the near-limit and first-trimester groups), the women scored as imminently suicidal at one point during the 5-year study period. All four indicated that the abortion was the right decision for them, three had a prepregnancy history of depression or anxiety, and all had a history of child abuse and/or neglect.

Results of mixed-effects regression analyses testing the association between history of depression or anxiety, history of child abuse or neglect, and past-year intimate partner violence with being lost to follow-up indicated that none of these variables were significantly associated with loss to follow-up (data not shown). Results of our death record search indicated that five women died during our study period. These five women (three in the near-limit group and two in the turnaway-birth group) were identified by a verbal autopsy and/or by a LexisNexis death record search. None had a previous history of depression or suicidal ideation symptoms according to previous assessments. Two died from car accidents, one from a heart attack, and two from unknown causes.