Objectives: To illustrate the use of machine learning methods to search for heterogeneous effects of a target modifiable risk factor on suicide in observational studies. The illustration focuses on secondary analysis of a matched case-control study of vitamin D deficiency predicting subsequent suicide. Methods:We describe a variety of machine learning methods to search for prescriptive predictors; that is, predictors of significant variation in the association between a target risk factor and subsequent suicide. In each case, the purpose is to evaluate the potential value of selective intervention on the target risk factor to prevent the outcome based on the provisional assumption that the target risk factor is causal. The approaches illustrated include risk modeling based on the super learner ensemble machine learning method, Least Absolute Shrinkage and Selection Operator (Lasso) penalized regression, and the causal forest algorithm. Results:The logic of estimating heterogeneous intervention effects is exposited along with the illustration of some widely used methods for implementing this logic.

Objective Psychiatric hospitalizations and emergency department (ED) visits are costly, stigmatizing, and often ineffective. Given the immune and kynurenine activation in bipolar disorder (BD) and schizophrenia, as well as the immune-modulatory effects of statins, we aimed to compare the relative risk (RRs) of psychiatric hospitalizations and ED visits between individuals prescribed lipophilic vs. hydrophilic statins vs. no statins. We hypothesized (a) reduced rates of hospitalization and ER utilization with statins versus no statins and (b) differences in outcomes between statins, as lipophilia increases the capability to penetrate the blood–brain barrier with potentially beneficial neuroimmune, antioxidant, neuroprotective, neurotrophic, and endothelial stabilizing effects, and, in contrast, potentially detrimental decreases in brain cholesterol concentrations leading to serotoninergic dysfunction, changes in membrane lipid composition, thus affecting ion channels and receptors. Methods We used VA service utilization data from October 1, 2010 to September 30, 2015. The RRs for psychiatric hospitalization and ED visits, were estimated using robust Poisson regression analyses. The number of individuals analyzed was 683,129. Results Individuals with schizophrenia and BD who received prescriptions for either lipophilic or hydrophilic statins had a lower RR of psychiatric hospitalization or ED visits relative to nonstatin controls. Hydrophilic statins were significantly associated with lower RRs of psychiatric hospitalization but not of ED visits, compared to lipophilic statins. Conclusion The reduction in psychiatric hospitalizations in statin users (vs. nonusers) should be interpreted cautiously, as it carries a high risk of confounding by indication. While the lower RR of psychiatric hospitalizations in hydrophilic statins relative to the lipophilic statins is relatively bias free, the finding bears replication in a specifically designed study. If replicated, important clinical implications for personalizing statin treatment in patients with mental illness, investigating add-on statins for improved therapeutic control, and mechanistic exploration for identifying new treatment targets are natural next steps.

Background: Immune activation or high levels of stress may lead to increased metabolism of tryptophan during pregnancy. Porphyromonas gingivalis (Pg), the “keystone” periodontal pathogen, induces immune and indoleamine 2,3-dioxygenase (IDO) activation. Thus, we hypothesized that larger gestational decreases in tryptophan and elevations in neopterin and kynurenine would occur in pregnant women with elevated Ig G antibodies to Pg capsular (K) serotypes. Methods: Venous blood of 52 Hispanic pregnant women with a mean age (SD) of 31.8 (5.9) years was sampled once per trimester of pregnancy (V1, V2, V3), and plasma was obtained and stored. ELISAs were used to measure Pg capsular (K) serotype IgG serointensity (V1 only) and neopterin levels (V1-V3). Tryptophan and kynurenine (V1-V3) were measured with high-performance liquid chromatography. The participants having IgG serointensity for any of the seven Pg K serotypes in the highest quartile were defined as the “High PgK_IgG” group and those having Ig G serointensity for all K serotypes in the lowest three quartiles were defined as the “Low PgK_IgG” group . Statistics included multivariable linear and nonparametric methods. Results: Significant decreases in plasma tryptophan levels and increases in neopterin during gestation were found in “High PgK_IgG” women but not in “Low PgK_IgG”. Kynurenine changes were not significantly different between the two groups. Conclusions: If replicated in larger studies and further characterized clinically, radiologically, and microbiologically, our results may potentially lead to novel interventional targets as well as developing more complete prognostic and predictive interactive biomarkers for adverse obstetrical outcomes and peripartum depression, and their prevention.