Unsupervised Machine Learning to Identify Depressive Subtypes

OBJECTIVES

This study evaluated an unsupervised machine learning method, latent Dirichlet allocation (LDA), as a method for identifying subtypes of depression within symptom data.

METHODS

Data from 18,314 depressed patients were used to create LDA models. The outcomes included future emergency presentations, crisis events, and behavioral problems. One model was chosen for further analysis based upon its potential as a clinically meaningful construct. The associations between patient groups created with the final LDA model and outcomes were tested. These steps were repeated with a commonly-used latent variable model to provide additional context to the LDA results.

RESULTS

Five subtypes were identified using the final LDA model. Prior to the outcome analysis, the subtypes were labeled based upon the symptom distributions they produced: psychotic, severe, mild, agitated, and anergic-apathetic. The patient groups largely aligned with the outcome data. For example, the psychotic and severe subgroups were more likely to have emergency presentations (odds ratio [OR] = 1.29; 95% confidence interval [CI], 1.17-1.43 and OR = 1.16; 95% CI, 1.05-1.29, respectively), whereas these outcomes were less likely in the mild subgroup (OR = 0.86; 95% CI, 0.78-0.94). We found that the LDA subtypes were characterized by clusters of unique symptoms. This contrasted with the latent variable model subtypes, which were largely stratified by severity.

CONCLUSIONS

This study suggests that LDA can surface clinically meaningful, qualitative subtypes. Future work could be incorporated into studies concerning the biological bases of depression, thereby contributing to the development of new psychiatric therapeutics.

Identifying subtypes of depression in clinician-annotated text: a retrospective cohort study

Current criteria for depression are imprecise and do not accurately characterize its distinct clinical presentations. As a result, its diagnosis lacks clinical utility in both treatment and research settings. Data-driven efforts to refine criteria have typically focused on a limited set of symptoms that do not reflect the disorder's heterogeneity. By contrast, clinicians often write about patients in depth, creating descriptions that may better characterize depression. However, clinical text is not commonly used to this end. Here we show that clinically relevant depressive subtypes can be derived from unstructured electronic health records. Five subtypes were identified amongst 18,314 patients with depression treated at a large mental healthcare provider by using unsupervised machine learning: severe-typical, psychotic, mild-typical, agitated, and anergic-apathetic. Subtypes were used to place patients in groups for validation; groups were found to be associated with future outcomes and characteristics that were consistent with the subtypes. These associations suggest that these categorizations are actionable due to their validity with respect to disease prognosis. Moreover, they were derived with automated techniques that might theoretically be widely implemented, allowing for future analyses in more varied populations and settings. Additional research, especially with respect to treatment response, may prove useful in further evaluation.

Prospective delivery of reliably dated term infants of diabetic mothers without determination of fetal lung maturity: comparison to historical control

OBJECTIVE

To determine whether omitting fetal lung maturity (FLM) testing prior to delivery in term pregnancies complicated by gestational (GDM) and pregestational diabetes mellitus would increase the risk of neonatal respiratory distress syndrome (RDS).

METHODS

In a 2-year study (1990-91), 1,457 pregnant women with accurately dated pregnancies were enrolled after 37 completed weeks and prospectively followed through delivery without FLM testing (study group). The prevalence of RDS and other neonatal outcomes was compared with a historical control group (n = 713, 1988-89) who had undergone determination of lecithin/sphingomyclin ratio prior to delivery at term. Logistic regression analysis was performed to determine independent predictors of RDS.

RESULTS

The study group compared to the control group had less severe diabetes: diet-controlled GDM, 35% vs. 18%, respectively; insulin-requiring GDM, 42% vs. 42%, respectively; undiagnosed type-2 diabetes, 14% vs. 31%, respectively; and pre-existing diabetes, 9.6% vs. 8.8%, respectively, p < 0.001. RDS rates in the study group (0.8%) and control group (1.0%) were not significantly different, nor were rates of resuscitation at delivery, neonatal intensive care admission or hospitalization days. Logistic regression analysis found only Cesarean delivery to be independently predictive (adjusted OR 2.21, 95% CI 2.04-2.27) of RDS. Non-predictive variables included FLM testing, diabetic classification, insulin use, poor third-trimester glycemic control, chronic hypertension, pre-eclampsia, labor, neonatal gender, gestational age or large-for-gestational-age fetuses.

CONCLUSIONS

Routine FLM testing did not change the RDS prevalence in reliably dated, term infants of diabetic mothers and should be abandoned. Delivery by Cesarean section was associated with increased RDS.

Design of peptides with high affinities for heparin and endothelial cell proteoglycans

Proteoglycan-binding peptides were designed based on consensus sequences in heparin-binding proteins: XBBXBX and XBBBXXBX, where X and B are hydropathic and basic residues, respectively. Initial peptide constructs included (AKKARA)(n) and (ARKKAAKA)(n) (n = 1-6). Affinity coelectrophoresis revealed that low M(r) peptides (600-1,300) had no affinities for low M(r) heparin, but higher M(r) peptides (2,000-3,500) exhibited significant affinities (K(d) congruent with 50-150 nM), which increased with peptide M(r). Affinity was strongest when sequence arrays were contiguous and alanines and arginines occupied hydropathic and basic positions, but inclusion of prolines was disruptive. A peptide including a single consensus sequence of the serglycin proteoglycan core protein bound heparin strongly (K(d) congruent with 200 nM), likely owing to dimerization through cysteine-cysteine linkages. Circular dichroism showed that high affinity heparin-binding peptides converted from a charged coil to an alpha-helix upon heparin addition, whereas weak heparin-binding peptides did not. Higher M(r) peptides exhibited high affinities for total endothelial cell proteoglycans (K(d) congruent with 300 nM), and approximately 4-fold weaker affinities for their free glycosaminoglycan chains. Thus, peptides including concatamers of heparin-binding consensus sequences may exhibit strong affinities for heparin and proteoglycans. Such peptides may be applicable in promoting cell-substratum adhesion or in the design of drugs targeted to proteoglycan-containing cell surfaces and extracellular matrices.