Insomnia and depression are distinct yet highly comorbid mental disorders. Up to 79% of patients with depression experience clinically significant insomnia, and insomnia is among the most consistent longitudinal predictor of depression episodes and suicidality. ^{Reference Toynbee, Zygmunt, Levett, Moghaddacy and Pappa1} Yet despite decades of epidemiological literature, insomnia and depression continue to be treated with separate manualised protocols such as cognitive–behavioural therapy for insomnia (CBT-I) and CBT for depression (CBT-D), delivered in a fixed order and standardised across patients, regardless of which specific mechanism actually drives an individual’s symptoms. Although CBT-I and CBT-D are highly effective in the short term, treatment effects may decay by 60–70% in the year following treatment end. ^{Reference Cuijpers, Miguel, Harrer, Plessen, Ciharova and Ebert2–Reference Fiori, Lombardo, Baglioni and Ballesio4} This requires a call for clinical innovation in the field. Here, a possible shift towards idiographic, network-informed research is proposed to assess person-specific treatment ingredients that may endure efficacy.

Insomnia has long been conceptualised as a symptom of depression, absorbed into a single latent construct. It was then recognised as a distinct disorder, correlated with but distinguishable from depression. That paradigm shift did not change the underlying logic. Comorbidity is still explained as a correlation between two latent entities, insomnia and depression. The ‘correlation’ framing leaves unanswered the question of how a specific symptom in one domain (for instance, night-time awakenings) might drive a specific symptom in the other (such as depressive rumination) in a given patient over time.

Network approaches to psychopathology model symptoms as components of a dynamic system in which central nodes (e.g. core elements in the network) activate and sustain a cascade of symptoms. ^{Reference Borsboom and Cramer5} Notably, the structures of these networks can be estimated individually rather than assumed to be the same across a sample. Idiographic network-informed research seeks to overcome the limitations of nomothetic ‘latent construct’ approaches to psychopathology by estimating the strength of associations between symptoms for each patient. From a partial correlation matrix of intensive longitudinal data, nodes can be identified as symptoms that are most strongly correlated with the others (i.e. those that have higher centrality in the network). These are proposed to act as ‘propagators’, which activate the entire network of symptoms and maintain psychopathology ^{Reference Borsboom and Cramer5} and should be therefore primarily targeted in personalised interventions. ^{Reference Bringmann6} Applied to insomnia and depression, this approach reframes comorbidity as a single system in which sleep and mood-related symptoms are causally entangled, with a structure that differs from one patient to another. For one patient, pre-sleep cognitive arousal may be the node sustaining the entire network, driving both insomnia severity and daytime rumination. For another, the same clinical picture may be sustained instead by circadian misalignment, or by stress system dysregulation and fatigue.

No studies to date have used idiographic networks to guide the treatment of comorbid insomnia and depression. Group-level networks exist, ^{Reference Blanken, Van Der Zweerde, Van Straten, Van Someren, Borsboom and Lancee7} and observational studies have begun tracking sleep and mood dynamics longitudinally, but none of the existing studies has combined idiographic network estimation with treatment matched to an individual’s own structure – nor asked whether the structure itself is subject to change. If a patient’s psychopathology is structured as a dynamic network, clinical success should be measured by whether that network reorganises during and after treatment (i.e. whether the node driving the clinical picture loses its centrality), not only by whether average symptoms decrease. Group-level pre–post comparisons cannot answer that question.

Person-specific symptom network dynamics are built on data collected from intensive longitudinal assessment (i.e. ecological momentary assessment or experience sampling), a methodology that has massively expanded in the past few years. Person-specific time-series models (graphical vector autoregression and Bayesian dynamic structural equation models) and wearable physiological sampling now make it possible to estimate an individual’s symptom network, including biomarkers, with enough precision to act on it clinically.

Network reconfiguration, and not symptom-sum reduction alone in itself, may become a primary, mechanistic outcome in trials testing personalised treatment for comorbid insomnia and depression. Concretely, this would mean designing trials that assign patients to treatment sequences matched to their own network structure and that track (session by session or week by week) whether the driving node loses influence over the system, alongside not instead of conventional clinical end-points.

Two further elements would strengthen any such effort. First, a consensus should be reached on minimal reporting standards for the ecological momentary assessment protocols used to build networks. These standards may include sampling density, duration and timing across the day, so that idiographic networks become comparable across studies. Second, explicit attention should be given to how these networks are used in the clinical setting. An estimated network is not a diagnosis to be handed to the patient but a hypothesis on the patients’ functioning to be openly discussed. Historically, CBT approaches have used subjective diaries to formulate case conceptualisation. Person-specific symptom networks may provide a robust evidence base for case conceptualisation and consequent treatment trajectories. ^{Reference Qu, Che, Chen, Zhu, Cai and Wu8} Their real value lies in shared case conceptualisation, helping the patient and clinician to jointly identify which processes seem to sustain the clinical picture; this is precisely where the model needs to remain open to revision. When an estimated network points to a driving node that the patient’s own account does not support, this discrepancy should not simply be resolved in favour of the model. It warrants closer clinical scrutiny of the network estimate and of what the patient is reporting.

In conclusion, although guideline-recommended treatments for insomnia and depression are available, they do not address the high comorbidity between the two conditions. Clinicians have no precise recommendations on how to handle complex comorbid cases. Whereas personalised psychological interventions are associated with superior outcomes compared with standardised psychotherapies, ^{Reference Nye, Delgadillo and Barkham9} the precise approaches to use in personalisation remain controversial. Idiographic network approaches may provide a data-driven framework for formulation of case conceptualisation and selection of person-specific treatment ingredients. Trials need to be designed around this logic from the start, with network reconfiguration as a primary outcome.