Instability as an early warning sign of severe depression: insight from electronic health records data

Thesis

Introduction: Accurate prediction of transition from mild or moderate to severe depression is critical for secondary prevention and effective healthcare planning. This thesis investigated whether symptom instability (i.e. fluctuations over time) predicts transition to severe depression among individuals diagnosed with Major Depressive Disorder (MDD).

Methods: This retrospective cohort study used electronic health records (EHRs) from NeuroBlu Data, an EHR network of mental health centres in the US (1999–2024). We selected all patients with an ICD diagnosis of MDD who had at least five PHQ-9 measurements within any six-month period and all PHQ-9 scores ≤14 during this phenotyping window. Instability of depressive symptoms was quantified using the time-adjusted root mean square of successive differences (tRMSSD). Using a Cox proportional hazards model adjusted for mean PHQ-9, sex, age, race, and psychiatric comorbidities, we assessed whether symptom instability was associated with the risk of severe depression (PHQ-9 >14) within one year following the phenotyping window.

A second analysis examined whether instability represents a trait (persistent characteristic of an individual) or a state (a transient feature) by comparing its value in the period preceding transition to severe depression to those in a matched baseline window earlier in each patient’s history.

Finally, we tested whether tRMSSD truly captures symptom instability or instead reflects underlying trends in depressive symptom change, using three Cox models assessing the predictive roles of instability, trend, and their interaction for both deterioration and remission.

Results: Symptom instability was found to be predictive of transition to severe depression (HR = 1.25, 95% CI 1.16-1.34, p < 0.0001, n=31 829). This was robust to change in population and outcome specification. Symptom exhibits both trait-like and state-like properties, increasing up to 48.4% compared to baseline (95% CI 35.7-61.1%, p<0.0001, n=1627). Instability remained informative when adjusting for trend and interaction (HR = 1.09, 95 CI 1.08-1.13, p < 0.0001, n=17 561), and was associated to remission (HR = 1.17, 95% CI 1.13-1.22, p < 0.0001, n=12 334).

Conclusions: This thesis showed that symptom instability provides predictive value beyond severity in MDD, independently of trend. It exhibited both traitand state-like properties, identifying at-risk individuals and signalling imminent deterioration. Routine PHQ-9 collection, prospective validation, and integration into predictive tools are required to translate these findings into practice, enabling earlier intervention and more adaptive management of depression.

Authors: De Deyn, T

• DOI: 10.5287/ora-ppjva7yn9
• Type of award: MSc by Research
• Level of award: Masters
• Awarding institution: University of Oxford
• Language: English
• Keywords: Electronic Health Records; EHR data; Major Depressive Disorder; clinical instability
• Subjects: Psychiatry; Depression, Mental health; Medical records--Data processing