Predicting Suicidal Ideation in Patients with Huntington’s Disease (DSI-SRP)
This DSI-SRP fellowship funded Joseph Sexton to work in the laboratory of David Isaacs, MD, MPH in the Department of Neurology at Vanderbilt University Medical Center during the summer of 2021. Joseph is a junior with majors in Psychology and Medicine, Health, and Society and minors in Quantitative Methods, Mathematics, and Scientific Computing.
Huntington’s disease (HD) is a terminal neurodegenerative disease associated with a number of crippling motor and cognitive impairments. Among those affected by HD, suicide has long been considered a pressing clinical issue; yet, due to limited psychometric data, analysis has been generally limited. Perhaps most notably, the literature has avoided questions of demographic identifiers of risk. This is striking. Cross-culturally, measures of suicidality vary significantly based on categories like age, sex, and race. Here, Joseph utilized the multi-national Enroll-HD dataset to investigate the prevalence of suicidality and the causes of suicidal ideation in patients with HD, paying specific attention to anticipated sex differences. Evaluating the larger dataset (N = 12063, female 54.02%) for death by suicide revealed a significantly heightened risk for male patients. Within the Enroll-HD sample, men died by suicide at over two-and-a-half times the rate of women. This is consistent with the broader epidemiological literature, wherein it is well-established that suicide rates for men eclipse those of women. For subjects reporting psychometric data (N = 4804, female 51.60%), no significant differences were observed between men and women with regards to suicidal ideation (two-week prevalence: 11.91% and 12.71%, respectively). Prior work to compute the prevalence of ideation by sex has suggested variation by demographic factors like sex, age, and region.
Under the premise that the causes of ideation may differ between men and women with HD, Joseph developed four classification models to predict ideation given other patient information. Using 19 well-documented and theoretically valid predictors, a random forest model was first developed. Random forest models are a powerful take on the broader concept of decision trees. These utilize “yes/no” questions to divide data into ideators and non-ideators, producing eventual “branches” that can potentially capture the effects of highly specific, intersectional populations. Random forest modeling develops a “forest” of decision trees, including some component of chance to reduce overfitting, and then makes a prediction based on the forest’s consensus. Ultimately, the model revealed that by-and-large the most significant predictor of ideation was rating of depression. Sex was effectively a nonfactor; taking it into consideration did not significantly reduce classification error. Three models – one sex-inclusive, one male-specific, and one female-specific – were then developed using the least absolute shrinkage and selection operator (LASSO) method and leave-one-out cross-validation. These techniques allow for a parsimonious and robust logistic regression. If sex-specific effects exist, Joseph anticipated seeing limited predictive value in the sex-inclusive model as well as distinct predictor sets in the male- and female-specific models. In each case, rating of depression was the only maintained predictor; all others were removed as nonsignificant. Further, similar performance was observed across the random forest and three LASSO models with regards to recall and precision values. In sum, it seems that depression score corresponds closely enough to odds of ideation that it subsumes all other predictor effects. This holds across sexes.
The simplicity and performance of the LASSO models is promising to clinicians who may in future be able to compute risk manually through assessment of depression (which is a standard baseline measure). In the future, Joseph plans to continue developing these models by employing oversampling methods and making distinctions between pre- and post-manifestation of symptoms.
In addition to receiving support through a DSI-SRP fellowship, this project was supported and facilitated by the DSI Data Science Team through their regular summer workshops and demo sessions.