Below are descriptions of projects we have ongoing. Prof. Lorenzo-Luaces is the PI or co-PI for the projects below, unless otherwise noted. For all projects we take special interest in individual differences

Heterogeneity of depression

Depression is known to be heterogeneous in its presenting symptoms (Buss et al., 2023; Fried et al., 2020; Lorenzo-Luaces, Buss, et al., 2021). This knowledge comes from studies in which researchers have counted combinations of symptom profiles (i.e., the specific ways a person meets criteria for depression). While this is a sensible and face-valid approach, it treats depression symptoms between individuals as if they were categorical (i.e., people either have the same symptoms or they do not). We have introduced methods from information theory to quantify diagnostic heterogeneity in depression. One advantage of these methods is that they allow us to answer questions about current diagnostic practices. For example, we can ask whether subtypes like the DSM’s melancholic and atypical specifiers reduce heterogeneity. Whether analyzing heterogeneity with the simple symptom-profile approach or the somewhat more complex information theory metrics, we do not find evidence that proposed subtypes reduce symptom heterogeneity.

We have also used machine learning methods to understand if different symptom combinations predict differential treatment outcomes in CBT vs. medications (DeRubeis et al., 2014; Lorenzo-Luaces, Rodriguez-Quintana, et al., 2021), positive psychotherapy (Lopez-Gomez et al., 2019), interpersonal psychotherapy (Van Bronswijk et al., 2021), or control conditions (Bronswijk et al., 2022; Lorenzo-Luaces et al., 2017; Lorenzo-Luaces, Rodriguez-Quintana, et al., 2021). However, with the exceptions of positive emotionality (Lopez-Gomez et al., 2019; Lorenzo-Luaces et al., 2017) and insomnia (Lorenzo-Luaces et al., 2017; Lorenzo-Luaces, Rodriguez-Quintana, et al., 2021), we have found little support for the idea that symptoms differentially predicts treatment outcomes. Instead, we find more consistent support for the idea that individuals with more severe symptoms overall benefit more from high-intensity treatments (e.g., the combination of CBT and medications) than those with less severe symptoms (Bronswijk et al., 2022; Lorenzo-Luaces et al., 2017; Lorenzo-Luaces, Rodriguez-Quintana, et al., 2021)

References

Personalized medicine

I have argued for the importance of studying predictors of the course and prognosis of depression (Lorenzo-Luaces, 2015; 2018; Lorenzo-Luaces et al., 2017a; 2021b). In epidemiological studies, roughly half of the individuals with depression have relatively brief courses (e.g., 3-6 months). In contrast, roughly 50% have a chronic, non-remitting course or a course characterized by recovery followed by relapse (Lorenzo-Luaces, 2015). The application of the same diagnosis to individuals with extremely heterogeneous prognoses can be traced back to a historical shift in the Diagnostic and Statistical Manual for Mental Disorders (DSM-III), in which different conceptualizations of depression were lumped into a single category (Horwitz et al., 2017; Lorenzo-Luaces, 2015; Wakefield et al., 2017a, 2017b). One implication is that the diagnosis does not aid in treatment allocation. Even if providers follow official guidelines for the treatment of depression, many individuals will be “overtreated” (e.g., prescribed antidepressants though their depression would improve on its own), and many individuals will not receive the adequate level of care they need (Lorenzo-Luaces et al., 2015).

Our work suggests that individual differences in prognosis can be predicted using machine learning (ML) methods and that these predictions seem useful in allocating individuals to higher vs. lower intensity treatments (Lorenzo-Luaces et al., 2017a; Lorenzo-Luaces et al., 2021b). For example, in one study, I used an ML approach to analyze data from an RCT (N=622) in which no statistically significant differences were found between CBT, a brief non-specific therapy, and a treatment-as-usual control. We created a treatment rule which predicted response to treatment from overall symptom severity, unemployment, insomnia, hostility, and positive emotionality. Most (75%) patients performed equally well in all treatments (see Figure 1), including the brief and non-specific brief therapy condition, which is more scalable than a full course of CBT. In the remaining 25% of patients with the most “difficult” prognoses, CBT was superior to both control groups, with effect sizes larger than typically reported in the literature (see also Lorenzo-Luaces et al., 2021a). My collaborators and I have also studied predictions of response to CBTs vs. other treatments like antidepressants (DeRubeis et al., 2014), interpersonal therapy (van Bronswijk et al., 2020), and positive psychology intervention (Lopez-Gomez et al., 2019). A barrier to this work is that large samples of individuals are needed to identify statistically significant multivariable treatment algorithms and almost all studies that have explored machine learning predictors of response (Lorenzo-Luaces et al., 2021c). These studies challenge current models of the treatment of depression that focus on averages without attempting to personalize treatment delivery.

Low-intensity treatments

Psychological interventions are effective treatments for depression, anxiety, stress, insomnia, and other common mental health concerns. Despite this, it is very difficult for most people to access treatments because they are expensive, time-consuming, and difficult to find. We have several projects studying interventions that are not as expensive as face-to-face psychotherapy with trained therapists, including internet apps and books. The projects include:

  • Leveraging computational social sciences and natural language processing to optimize engagement and response to low-intensity CBT for depression and anxiety
  • Feasibility of stepped care with single-session interventions and guided self-help CBT
  • Psychologists’ use of waiting lists and willingness to use low-intensity treatments (PI: Peipert)
  • Barriers to internet-based CBT use (PI: Peipert)
  • Racial-ethnic diversity in trials of internet-based cognitive-behavioral therapy (PI: De Jesús-Romero)
  • Willingness of individuals to use different low-intensity treatments (PI: De Jesús-Romero)
  • Predictors of engagement with guided self-help (PI: Starvaggi)

Emotion regulation

One commonality of depression, anxiety, stress, insomnia, and other common mental health concerns may be that people have a hard time regulating their emotions, especially negative emotions. When we conduct studies, we usually include measures of emotion regulation, usually the Emotion Regulation Questionnaire (Gross and John, 2003). The ERQ measures the habitual or regular use of two emotion regulation strategies: cognitive reappraisal and expressive suppression. Projects that specifically focus on emotion regulation include:

  • Cognitive reappraisal of LGB-identity vs. non-LGB-identity stress in LGBT young adults (PI: De Jesús-Romero)
  • Mechanisms of change in transdiagnostic guided self-help (PI: De Jesús-Romero)

Social media

Social media is a relatively recent development. As of 2021, over 75% of adults in the United States are on a social media platform. That alone makes social media an interesting topic to study.

Most relevant to our work, there are reported correlations between social media use and poorer mental health with some worrying that social media use causes poorer mental health, at least in some people. While we do not know if this is true, social media is also interesting from a research perspective because people openly talk about their mental health and some social media behaviors can clue you in to people’s mental health (e.g., when individuals discuss feeling sad). Moreover, we can make inferences about people mental health and emotions based on their behavior. For example, in one study, we looked at the timing of activity on Twitter as an index of a person’s sleep/wake cycle. We found differences between Twitter users who reported being depressed and a random sample suggesting that people who were depressed were more active into the night and less active early in the morning. Watch me talk about this study below:


We have several ongoing studies under the umbrella of the Surveys of Online Cohorts for Internalizing symptoms And Language (SOCIAL). In SOCIAL-I, we surveyed a nationally representative sample of 1121 U.S. adults and administered a transdiagnostic battery of symptom assessments. We also obtained their consent to access their social media accounts. The aim of the project is to triangulate self-reported mental health data and social media activity. SOCIAL-II is an ongoing study that has overlapping measures with SOCIAL-I but expands the assessments to include temperament and eating pathology. We’ve recruited at least 2,015 college students and continue to collect each semester.

Bronswijk, S. C. van, Howard, J., & Lorenzo-Luaces, L. (2022). Data-driven personalized medicine approaches to cognitive-behavioral therapy allocation in a large sample: A reanalysis of the ENRICHED study. https://psyarxiv.com/ex3gh/
Buss, J. F., Watts, A. L., & Lorenzo-Luaces, L. (2023). Information theory methods for quantifying diagnostic heterogeneity in psychopathology. https://www.researchsquare.com/article/rs-2472751/latest.pdf
DeRubeis, R. J., Cohen, Z. D., Forand, N. R., Fournier, J. C., Gelfand, L. A., & Lorenzo-Luaces, L. (2014). The Personalized Advantage Index: Translating research on prediction into individualized treatment recommendations. A demonstration. PLOS ONE, 9(1), e83875. https://journals.plos.org/plosone/article?id=10.1371/journal.pone.0083875
Fried, E. I., Coomans, F., & Lorenzo-Luaces, L. (2020). The 341,737 ways of qualifying for the melancholic specifier. The Lancet Psychiatry, 7(6), 479–480. https://scholarlypublications.universiteitleiden.nl/access/item%3A3069973/view
Lopez-Gomez, I., Lorenzo-Luaces, L., Chaves, C., Hervas, G., DeRubeis, R. J., & Vazquez, C. (2019). Predicting optimal interventions for clinical depression: Moderators of outcomes in a positive psychological intervention vs. Cognitive-behavioral therapy. General Hospital Psychiatry, 61, 104–110. https://www.sciencedirect.com/science/article/pii/S0163834319300532?casa_token=8EOcO0nes88AAAAA:1Vr7zk30JBheghLkOfNKHqbHxieT4xlvo8OQpXYYLRN-un-x0SI5npMgKIoE_X4OOlxPvIZQ8Q
Lorenzo-Luaces, L., Buss, J. F., & Fried, E. I. (2021). Heterogeneity in major depression and its melancholic and atypical specifiers: A secondary analysis of STAR* d. BMC Psychiatry, 21(1), 1–11. https://bmcpsychiatry.biomedcentral.com/articles/10.1186/s12888-021-03444-3
Lorenzo-Luaces, L., DeRubeis, R. J., Straten, A. van, & Tiemens, B. (2017). A prognostic index (PI as a moderator of outcomes in the treatment of depression: A proof of concept combining multiple variables to inform risk-stratified stepped care models. Journal of Affective Disorders, 213, 78–85. https://bpb-us-w2.wpmucdn.com/web.sas.upenn.edu/dist/d/257/files/2016/10/lorenzo-luaces-Prognostic-Index-paper-10f0t61.pdf
Lorenzo-Luaces, L., Rodriguez-Quintana, N., Riley, T. N., & Weisz, J. R. (2021). A placebo prognostic index (PI) as a moderator of outcomes in the treatment of adolescent depression: Could it inform risk-stratification in treatment with cognitive-behavioral therapy, fluoxetine, or their combination? Psychotherapy Research, 31(1), 5–18. https://weiszlab.fas.harvard.edu/files/jweisz/files/lorenzo-luaces_et_al_2020.pdf
Van Bronswijk, S. C., Bruijniks, S. J., Lorenzo-Luaces, L., Derubeis, R. J., Lemmens, L. H., Peeters, F. P., & Huibers, M. J. (2021). Cross-trial prediction in psychotherapy: External validation of the Personalized Advantage Index using machine learning in two Dutch randomized trials comparing CBT versus IPT for depression. Psychotherapy Research, 31(1), 78–91. https://www.tandfonline.com/doi/pdf/10.1080/10503307.2020.1823029?needAccess=true&role=button