Notes from an aspiring academic

Search

SearchSearch

McNally2021 - Network Analysis of Psychopathology, Controversies and Challenges

Aug 31, 2023, 2 min read

McNally2021 - Network Analysis of Psychopathology, Controversies and Challenges

Example citation

Key takeaways

  • The current version of the DSM-5 has its roots in an anti-Freudian movement where there was a need to establish psychiatry as a legitimate branch of medicine, hence signs/symptoms/course etc. This enabled epidemiologic research, RCTs and basic research on the biology/behavior/cognitive aspects of discrete conditions.
  • two
  • three

Challenges and controversies

Comorbidity

Bridge symptoms can account for the spread of activation in two separate symptom clusters, i.e. two disorders.

Replicability

Difference between criticizing network analyses themselves and identifying that network coefficients may not replicate across samples. Forbes has done a couple of studies comparing networks between studies and finding low overlap. But Borsboom and Jones have retorted that this not invalidate the method per se.

He references the bootnet package to bootstrap results.

Ontology

Latent variables in psychiatry are very different from “true” latent variables/underlying causes like a malignant tumor. Borsboom 2008: The cause must be distinguishable from its effects.

Centrality

Central symptoms hypothesised to be important. You can also check nodepredictability, how much of a symptom can be predicted by others. “A highly predicted symptom may constitute low-hanging fruit suitable for intervention”.

Lower variance pre-treatment may lead to findings that improvement -> increased network density. Ceiling effect pre-treatment.

The seemingly paradoxical finding of density and symptom severity moving in opposite directions may have arisen as a function of increased symptom variance. Patients in the program described by Beard et al. (2016) tend to score uniformly high on symptom measures upon admission to the unit, producing a ceiling effect that constrains variance, thereby attenuating edge weights and thus network density. A diversity of responses to treatment, plus regression to the mean, would increase variance and increase the likelihood of network density increasing at posttreatment.

Temporality

Maybe central symptoms at baseline can predict therapy outcomes? Some mixed findings here. Also, network density at baseline has mixed findings too.

Temporal time-series networks

Cross-sectional networks have limitations because you can’t see how interactions unfold over time, and within-individual changes cannot be observed.

Time-series data can identify “tipping points” where individuals transition into other network states.

Graph View

Backlinks