Longitudinal evolution of the transdiagnostic prodrome to severe mental disorders: a dynamic temporal network analysis informed by natural language processing and electronic health records
============================================================================================================================================================================================

* Maite Arribas
* Joseph M. Barnby
* Rashmi Patel
* Robert A. McCutcheon
* Daisy Kornblum
* Hitesh Shetty
* Kamil Krakowski
* Daniel Stahl
* Nikolaos Koutsouleris
* Philip McGuire
* Paolo Fusar-Poli
* Dominic Oliver

## ABSTRACT

**Background** Modelling the prodrome to severe mental disorders (SMD), including unipolar mood disorders (UMD), bipolar mood disorders (BMD) and psychotic disorders (PSY), should consider both the evolution and interactions of symptoms and substance use (prodromal features) over time. Temporal network analysis can address this by representing prodromal features as nodes, with their connections (edges) indicating the likelihood of one feature preceding the other. Node centrality could reveal insights into important prodromal features and potential intervention targets. We developed a SMD network and compared sub-networks specific to UMD, BMD and PSY.

**Methods** We analysed 7,049 individuals with an SMD diagnosis (UMD:2,306; BMD:817; PSY:3,926) from the South London and Maudsley NHS Foundation Trust electronic health records. Using validated natural language processing algorithms, we extracted the occurrence of 61 prodromal features every three months from two years to six months prior to SMD onset. To construct temporal networks of prodromal features, we employed generalized vector autoregression panel analysis, adjusting for covariates. We computed edge weights (correlation coefficients, *z*) in autocorrelative, unidirectional and bidirectional relationships. Centrality was calculated as the sum of connections leaving (out-centrality, *cout*) or entering (in-centrality, *cin*) a node. We compared the three sub-networks (UMD, BMD, PSY) using permutation analysis.

**Findings** The strongest autocorrelation in the SMD network was tearfulness (*z*=·10). Unidirectional positive relationships were observed for irritability-agitation (*z12*=·03), mood instability-tearfulness (*z12*=·03) and irritability-aggression (*z12*=·03). Aggression-hostility (*z12*=·04, *z21*=·03), delusions-hallucinations (*z12*=·04, *z21*=·03) and aggression-agitation (*z12*=·03, *z21*=·03) were the strongest bidirectional relationships. The most central features included aggression (*cout*=·082) and tearfulness (*cin*=·124). The PSY sub-network showed few significant differences compared to UMD (3·9%) and BMD (1·6%), and UMD-BMD showed even fewer (0·4%).

**Interpretations** This study represents the most extensive temporal network analysis conducted on the longitudinal interplay of SMD prodromal features. These findings provide further evidence to support early detection services across SMD.

**Evidence before this study** Preventive approaches for severe mental disorders (SMD) can improve outcomes, however, their effectiveness relies on accurate knowledge of the prodromal symptoms and substance use preceding their onset and how they evolve over time. We searched PubMed from database inception to 26th January 2024 for studies investigating the dynamic prodromes for unipolar mood disorders (UMD), bipolar mood disorders (BMD) or psychotic disorders (PSY) published in English. The search terms were prodrom* AND (depression OR bipolar OR psychosis) AND (timecourse OR dynamic OR “network analysis” OR longitudinal). First, while many studies have investigated the prodromal phases of SMD, particularly for PSY, the majority of studies have taken a cross-sectional rather than longitudinal approach which are unable to detect causal dependence between and within prodromal symptoms and substance use. Second, there are no studies focusing on the evolution of features during the prodromal period. Finally, studies have focused on diagnosis-specific analyses, considering UMD, BMD or PSY alone, limiting the possibility for comparison between them.

**Added value of this study** We have used a temporal network analysis approach, in combination with a large electronic health record database (n=7,049) and natural language processing, to examine the dynamic evolution of symptoms and substance use in the prodrome to an SMD diagnosis in secondary mental healthcare. This is the largest network analysis investigating prodromal features in SMD, the first assessing longitudinal changes and the first to directly compare the prodromes to UMD, BMD and PSY. Our results add to the growing evidence for a transdiagnostic prodrome to SMD, by showing small differences between UMD, BMD and PSY in how symptoms and substance use evolve over the course of the prodrome.

**Implications of all the available evidence** Our study explores the patterns of evolution of symptom and substance use events across and within SMD diagnostic groups. We highlight the importance of understanding the dynamic progression of these prodromal features to fully characterise the prodrome to SMD. These findings, together with a growing literature base, also support the potential for broader transdiagnostic early detection services that provide preventive psychiatric care to individuals at risk for SMD.

Keywords
*   psychosis
*   bipolar
*   depression
*   network analysis
*   electronic health record
*   artificial intelligence
*   natural language processing
*   early detection
*   severe mental disorder
*   temporal network analysis

## 1. BACKGROUND

Severe mental disorders (SMD) include non-psychotic unipolar mood disorders (UMD), non-psychotic bipolar mood disorders (BMD) and psychotic disorders (PSY), and are characterised by high clinical, societal, familial and personal burden.1–3 Electronic health records (EHRs) can provide an opportunity to examine prodromal symptoms contemporaneously, reducing recall bias and enriching our insight into symptom presentation during the prodrome.4 This knowledge can help enhance specialised preventive care for young people at-risk of emerging SMD.

Temporal network analysis allows us to statistically model the relationships between nodes (prodromal features) as edges within a network (prodrome) over time. Weak, sparse networks are more modifiable, while strong, dense networks resist change5, needing intensive interventions to alter them6 (e.g. preventing SMD onset). Edge estimates in temporal nodes could suggest directed causality between features, potentially enhancing our understanding of SMD development.7 Node centrality, representing connection strength in and out of a node,8 may highlight the significance of a prodromal feature in the progression of the disorder and its potential as an intervention target.9–12

Firstly, we aimed to develop a global transdiagnostic SMD network to quantify the temporal relationships between prodromal features. Secondly, we aimed to examine within-group differences by computing and comparing sub-networks specific to UMD, BMD and PSY.

## 2. METHODS

### 2.1. Data Source

Data were from the South London and Maudsley National Health Service Foundation Trust (SLaM). SLaM provides secondary mental healthcare across four socioeconomically diverse South London boroughs (Lambeth, Southwark, Lewisham and Croydon, 1.3 million people, eMethods 1). A Clinical Record Interactive Search (CRIS) tool was implemented in the EHR to facilitate research with full but anonymised clinical information.13 CRIS has already been extensively validated in previous research studies.14–16 CRIS received ethical approval as an anonymised dataset for secondary analyses from Oxfordshire REC C (Ref: 23/SC/0257).

### 2.2. Study Design

Retrospective (2-year), real-world, EHR cohort study (Figure 1). The 2-year period was chosen to mirror the typical duration of care in clinical services for primary indicated prevention of SMD (72.4% provide care for 24 months or less).17 The index date reflected the date of the first diagnosis within an individual’s SMD group recorded in the EHR (index diagnosis, T-0mo, Figure 1). The antecedent date was defined by a data cut-off at 6 months before the index date (T-6mo), defining the antecedent period, to avoid overlap with the actual onset of SMD. The lookback period (Figure 1) was defined as the 1.5-year period prior to the antecedent date (T-6mo). To minimise the time invariance imposed by network analyses,18 we split the 1.5yr lookback period into six three-month follow-up intervals.

![Figure 1.](http://medrxiv.org/https://www.medrxiv.org/content/medrxiv/early/2024/03/09/2024.03.08.24303965/F1.medium.gif)

[Figure 1.](http://medrxiv.org/content/early/2024/03/09/2024.03.08.24303965/F1)

Figure 1. 
Study design. The look-back period was split into six three-month follow-up intervals (FU 1-6) relative to the index date (T-0mo) of SMD diagnosis. This pipeline (steps 1-4) was followed for both the primary analysis (SMD model) and secondary analysis (sub-networks).

### 2.3. Study Population

All individuals accessing SLaM services between 1st January 2008 and 10th August 2021 and receiving a primary (i.e. not comorbid) ICD-10 index diagnosis of any SMD were eligible. SMD was defined as either UMD, BMD or PSY (operationalised as in eTable 1). Individuals with multiple SMD diagnoses were assigned the diagnosis of greatest severity (i.e. UMD<BMD<PSY). Therefore, if an individual receives a diagnosis of UMD and BMD simultaneously, they would be included in the BMD group due to the higher associated severity.

Individuals with EHR entries (e.g. clinical notes and letters recorded in each month) recorded exclusively after the index date or exclusively in the antecedent period were excluded as they had no detectable prodrome. Individuals who only had empty EHR entries within the lookback period were also excluded, as well as those with EHR entries recorded within four or fewer follow-up intervals within the lookback period, as they did not have sufficient data to contribute to the fitted networks.

### 2.4. Variables

At index date, data were extracted from structured text on age, gender, self-assigned ethnicity (UK Office of National Statistics, eTable 2), ICD-10 diagnoses and prescription of antipsychotics, antidepressants, mood stabilisers and anxiolytics (see eTable 3 for medication classification details).

During the lookback period, data were extracted as binary variables on the occurrence (yes:1/no:0) of 61 natural language processing (NLP)-based prodromal features across each follow-up interval (FU 1-6; Figure 1). These NLP algorithms convert unstructured EHR information (i.e. free text) into structured quantifiable data.19 NLP algorithms with precision ≥80% (mean=90%) were included. Precision was defined as the ratio of the number of relevant (true positive) instances retrieved out of the total NLP-labelled positive instances (including irrelevant [false positive] and relevant [true positive] instances) in human-annotated EHR (see eMethods 2 for further details on NLP algorithm development and validation, and eTable 4 for the final list of NLP algorithms employed). Within each follow-up interval, the EHR entry frequency (number of entries) and length (total number of words recorded across all entries) were computed.

### 2.5. Statistical analysis

All analyses were conducted in R (version 4.2.3) on a virtual machine (AMD EPYC 7763 64-Core Processor) in Ubuntu 22.04.1 operating system. All analysis code is publicly available on GitHub: [https://github.com/m-arribas/network_analysis.git](https://github.com/m-arribas/network_analysis.git).

#### 2.5.1. Sociodemographic and Clinical Characteristics

We computed descriptive analyses for sociodemographic variables at index date (age, gender, self-assigned ethnicity) as well as the proportion (N [%]) of individuals with specific ICD-10 diagnoses and prescription of antipsychotics, antidepressants, mood stabilisers and anxiolytics at index in UMD, BMD and PSY.

In a sensitivity analysis, to test for any sampling bias in the final population, we compared excluded individuals (with four or fewer follow-up intervals) to those included (with five or more follow-up intervals) on sociodemographic variables (age, gender, self-assigned ethnicity), clinical variables (proportion of individuals belonging to each SMD group and medication prescriptions at index), as well as the severity of presenting features (frequency of prodromal clusters within antecedent period).

#### 2.5.2. Network analysis

As a primary analysis, we quantified a set of local network metrics in a transdiagnostic SMD network (hereby called “SMD network”) on the entire study population. In a secondary analysis, we repeated this on each SMD sub-sample separately (UMD, BMD, PSY), to compute three diagnosis-specific sub-networks (hereby called “sub-networks”). For each network (SMD network and three sub-networks), the following steps (pre-processing, network development and stability assessment) were repeated separately in each relevant dataset using a similar step-wise procedure to prior work modelling temporal features in psychopathology.20

Pre-processing and network development methods are detailed in the Supplement (eMethods 3 and 4, respectively). For each network (SMD network and three sub-networks), we extracted the temporal (within individuals), contemporaneous (relationships between nodes averaged over time and averaged across the sample), and between-individuals subject matrices. From each matrix, the strength of connections between features (edge weights) were estimated as correlation coefficients (*z*), and categorised into 3 types: autocorrelative (node predicts itself in the next time point), unidirectional (node predicts another, without reciprocation) and bidirectional (mutual prediction between two nodes). Degree centrality were extracted from each graph. For temporal networks, centrality was defined as the sum of absolute (directed) edge weights in (in-centrality*, cin*) and out (out-centrality*, cout*) of a node (including autocorrelative edges). For contemporaneous and between-subject networks, centrality was defined as the sum of absolute (undirected) edge weights for a node (autocorrelative edges do not exist).

To evaluate robustness of the edge weight estimates and to avoid overfitting in our networks, we computed the stability of edges within each full fitted network using bootstrapping procedures: over 250 iterations, 25% of the sample was randomly held out and the full model refitted on the remaining 75% of participants (following standard methods).21 Within each iteration the selected data was pre-processed in the same manner as in the full model to control for errors and variance within the data cleaning and scaling process. The averaged edge weights and 95%CIs over all 250 iterations were retained and reported. All edges with 95%CIs crossing zero were forced to 0.

#### 2.5.3. Permutation analysis

To test for statistically significant differences in the temporal, contemporaneous, and between-subject relationships across the three sub-networks (UMD, BMD, PSY) we conducted permutation analyses22.

To generate networks with the same topology required for valid comparisons, we re-fitted the three original sub-networks restricted to common features only, after pre-processing the data for each sub-sample.

In each permutation iteration, raw data in each sub-sample population (UMD, BMD, PSY) was randomly re-sampled for each individual node. The permuted dataset for each sub-sample was then pre-processed in the same manner as in the original sub-networks, and permuted sub-networks were fitted. From each permuted sub-network, temporal, contemporaneous and between-subjects matrix estimates were obtained, as in the main analysis. For each edge, the difference in permuted edge weights was calculated (UMD-BMD, UMD-PSY and BMD-PSY). Each edge weight comparison was visually inspected using histograms to assess normality of the permuted data.

For each edge weight comparison, the number of iterations where the permuted difference was equal to or greater than the absolute observed difference (from the actual dataset) was divided by the total number of permutations (250) to obtain the p-value. We corrected the resulting p-values for multiple comparisons using the false discovery rate method set at 5% to ensure the robustness of our findings and used p*permuted*<0·05 as the threshold for statistical significance.

Observed differences in edge weights (from the actual dataset) and corrected p-values (from the permutation analysis) were reported, and visualised with heatmaps.

## 3. RESULTS

### 3.1. Study Population

The final study population (n=7,049, Table 1; eTable 5-6) had EHR entries with more than four follow-up intervals (mean number of intervals [SD]=5·69 [0·46]) within the lookback period (Figure 2). Included participants were similar to excluded participants in terms of sociodemographics and clinical characteristics (eResults 1). Eight participants were excluded as the imputation method was unable to converge on stable approximations.

View this table:
[Table 1.](http://medrxiv.org/content/early/2024/03/09/2024.03.08.24303965/T1)

Table 1. Demographics and medication variables at index of final study population (N=7,049). Continuous variables are represented by mean (SD), categorical variables are represented by count (frequency).

![Figure 2.](http://medrxiv.org/https://www.medrxiv.org/content/medrxiv/early/2024/03/09/2024.03.08.24303965/F2.medium.gif)

[Figure 2.](http://medrxiv.org/content/early/2024/03/09/2024.03.08.24303965/F2)

Figure 2. 
Flow-chart of the study population.

### 3.2. Primary Analysis (SMD network)

Out of the 61 NLP-derived prodromal features, 38 displayed near-zero variance and were excluded, leaving 23 features for the analyses (eTable 7): aggression, agitation, anxiety, cannabis use, cocaine use, cognitive impairment, delusional thinking, disturbed sleep, emotional withdrawal, feeling hopeless, guilt, hallucinations (all), hostility, irritability, mood instability, paranoia, poor concentration, poor insight, poor motivation, suicidality, tearfulness, tobacco use, and weight loss (eFigure 1).

A saturated model (a densely connected network with all available edges) was fitted with the 23 features at 6 follow-up intervals (Figure 3A). This network demonstrated excellent fit (RMSEA=·0091 [95%CI: ·0088, ·0094]; *X*2(8625)=13650, p<·0001; CFI=·97; TLI=·97) and had better fit than a sparse network (pruned edges) (ϕ1*X*2(733)=2734·10, p<·0001). The model showed high recoverability (eResults 2) and robustness (see Figure 3B and eTables 8-9 for actual model and bootstrapped estimates).

![Figure 3.](http://medrxiv.org/https://www.medrxiv.org/content/medrxiv/early/2024/03/09/2024.03.08.24303965/F3.medium.gif)

[Figure 3.](http://medrxiv.org/content/early/2024/03/09/2024.03.08.24303965/F3)

Figure 3. 
Temporal relationships between nodes in SMD network **A.** Temporal network graph displaying positive (blue) and negative (red) relationships between nodes from actual model estimates. Edges are displayed as lines, with the thickness representing the strength of the edge weight estimate (correlation coefficient, *z*). Edges are thresholded (|*z*| > ·022) and labelled (|*z*| > ·03). For visualisation purposes, nodes are clustered into six categories (depressive, manic, negative, positive, substance use and other) according to the type of prodromal feature.

**B.** Bootstrapped (250 repetitions; black) vs actual model (n=7,140; red) edge weight estimates (|*z*| > ·022). Edges are directed such that “node1 – node2” represent the edge from node1 to edge2. All edges were positive except the one marked with an asterisk (HOST-INS).

**C.** Centrality measures for all nodes

AGGR: aggression, AGIT: agitation, ANX: anxiety, CANN: cannabis use, COC: cocaine use, COGN: cognitive impairment, CONC: poor concentration, DEL: delusional thinking, EMOT: emotional withdrawal, GUIL: guilt, HALL: hallucinations (all), HOPE: feeling hopeless, HOST: hostility, INS: poor insight, IRR: irritability, MOOD: mood instability, MOTIV: poor motivation, PAR: paranoia, SLEEP: disturbed sleep, SUIC: suicidality, TEAR: tearfulness, TOB: tobacco use, WGHT: weight loss

The strongest autocorrelation was observed for tearfulness (correlation coefficient, *z*=·10), with all the other autocorrelations between 0·05-0·10 (Figure 3A). The most prominent unidirectional relationships were positive: irritability-aggression (*z*12=·03), irritability-agitation (*z*12=·03), hallucinations (all)-disturbed sleep (*z*12=·03) and mood instability-tearfulness (*z*12=·03). All other unidirectional relationships were | *z*12| <·03.

With respect to bidirectional relationships, positively recurring pairs were observed between aggression-hostility (*z*12=·04, *z*21=·03), delusional thinking-hallucinations (all) (*z*12=·04, *z*21=·03), aggression-agitation (*z*12=·03, *z*21=·03) and delusional thinking-hostility (*z*12=·02, *z*21=·03).

Considering centrality (Figure 3C), aggression (*cout*=·098), hostility (*cout*=·082), and hallucinations (all) (*cout*=·081) had the strongest out-centrality, whereas tearfulness (*cin*=·124), aggression (*cin*=·09) and delusional thinking (*cin*=·085) had the strongest in-centrality (eTable 10).

Results and visualisations for the contemporaneous and between-subject relationships of nodes are presented in eResults 3 and eFigure 2. See eTable 12 for actual model and bootstrapped estimates.

### 3.3 Secondary Analysis (sub-networks)

Out of the 61 NLP-derived prodromal features, after applying the relevant exclusions within each sub-sample, 21 features were included for the UMD network, 19 for BMD and 24 for PSY (eMethods 5).

A saturated model was fitted with the relevant features at 6 follow-up intervals in each sub-sample (UMD, BMD, PSY). Similarly to the primary analysis, saturated networks showed excellent fit and better fit than sparse models for the three networks (UMD: Δ*X*2(687)=1737, p<·0001; BMD: Δ*X*2(606)=1547, p<·0001; PSY: Δ*X*2(856)=2961, p<·0001). Further model fit results, including recoverability (eResults 4), and bootstrapping estimates (eTable 11, eFigure 3) can be found in the Supplement.

#### a. UMD

The strongest autocorrelations were observed for cannabis use (*z*=·12), feeling lonely (*z*=·12) and hallucinations (all) (*z*=·11) with all the other autocorrelations between 0·03-0·10 (Figure 4A).

![Figure 4](http://medrxiv.org/https://www.medrxiv.org/content/medrxiv/early/2024/03/09/2024.03.08.24303965/F4.medium.gif)

[Figure 4](http://medrxiv.org/content/early/2024/03/09/2024.03.08.24303965/F4)

Figure 4 
Temporal relationships between nodes in sub-networks

Temporal network graphs displaying positive (blue) and negative (red) relationships between nodes from actual model estimates for sub-networks (**A.** UMD, **B.** BMD, **C.** PSY). Edges are displayed as lines, with the thickness representing the strength of the edge weight estimate (correlation coefficient, *z*). Edges are thresholded (UMD: |*z*| > ·026, BMD: |*z*| > ·045, PSY: |*z*| > ·03) and labelled (UMD: |*z*| > ·04, UMD: |*z*| > ·06, UMD: |*z*| > ·05). For visualisation purposes, nodes are clustered into six categories (depressive, manic, negative, positive, substance use and other) according to the type of prodromal feature. D. Centrality measures for all nodes in sub-networks (green: UMD, blue: BMD, red: PSY)

AGGR: aggression, AGIT: agitation, ANX: anxiety, AROUS: arousal, CANN: cannabis use, COC: cocaine use, COGN: cognitive impairment, CONC: poor concentration, DEL: delusional thinking, ELAT: elation, EMOT: emotional withdrawal, GUIL: guilt, HALL: hallucinations (all), HOPE: feeling hopeless, HOST: hostility, INS: poor insight, IRR: irritability, LONE: feeling lonely, LOW: low energy, MOOD: mood instability, MOTIV: poor motivation, NIGHT: nightmares, PAR: paranoia, SLEEP: disturbed sleep, SUIC: suicidality, TEAR: tearfulness, TOB: tobacco use, WGHT: weight loss

The most prominent unidirectional relationships were all positive: poor motivation-low energy (*z*12=·06), tobacco use-weight loss (*z*12=·04), paranoia-nightmares (*z*12=·04) and mood instability-weight loss (*z*12=·04). All other unidirectional relationships were |*z*12|<·04. With respect to bidirectional relationships, positively recurring pairs were observed between guilt and tearfulness (*z*12=·03, *z*21=·03).

Considering centrality, weight loss (*cin*=·140), aggression (*cin*=·128) and suicidality (*cin*=·106) had the strongest in-centrality, whereas tobacco use (*cout*=·105), mood instability (*cout*=·103) and poor motivation (*cout*=·092), had the strongest out-centrality (eTable 13A).

#### b. BMD

The strongest autocorrelation was observed for hallucinations (all) (*z*=·13), with all the other autocorrelations between 0·03-0·10 (Figure 4B). The most prominent unidirectional relationships were mixed, with some positive: guilt-feeling hopeless (z12=·07), aggression-elation (*z*12=·06) and hallucination-suicidality (*z*12=·06); and others negative: guilt-paranoia (*z*12=-·07), irritability-tobacco use (*z*12=-·06), and feeling hopeless-elation (*z*12=-·06). All other unidirectional relationships were |*z*12|<·06. With respect to bidirectional relationships, positively recurring pairs were observed between elation-irritability (*z*12=·06, *z*21=·06).

Considering centrality, elation (*cin*=·176), irritability (*cin*=·157) and tobacco use (*cin*=·152), had the strongest in-centrality, whereas elation (*cout*=·165), irritability (*cout*=·163) and guilt (*cout*=·143) had the strongest out-centrality (eTable 13B).

#### c. PSY

The strongest autocorrelation was observed for feeling hopeless (*z*=·11) and tearfulness (*z*=·11), with all the other autocorrelations between 0·04-0·10 (Figure 4C). The most prominent unidirectional relationships were all positive: hallucinations (all)-disturbed sleep (*z*12=·04), hostility-arousal (*z*12=·04), irritability-agitation (*z*12=·04). All other unidirectional relationships were |*z*12|<·04. With respect to bidirectional relationships, positively recurring pairs were observed between aggression-hostility (*z*12=·04, *z*21=·03), delusional thinking-hallucinations (all) (*z*12=·04, *z*21=·03), aggression-agitation (*z*12=·04, *z*21=·04), arousal-elation (*z*12=·04, *z*21=·04).

Considering centrality, agitation (*cin*=.115), aggression (*cin*=·078) and arousal (*cin*=·072), had the strongest in-centrality, whereas aggression (*cout*=·143), hostility (*cout*=·135) and hallucinations (all) (*cout*=·104) had the strongest out-centrality (eTable 13C).

Results and visualisations for the contemporaneous and between-subject relationships of nodes for all sub-networks are presented in eResults 5. See eTable 12 for actual model and bootstrapped estimates.

### 3.4 Permutation Analysis

The final nodes for permutation analysis are found in eMethods 6, with the actual model estimates in eTable 14. The histograms of permuted edge weights exhibited a normal (bell-shaped), zero-centred curve, indicating that further iterations are unlikely to affect the distribution or the results of these analyses. Out of all possible edge weight comparisons in the permutation analysis, few of them were significantly different: UMD-PSY (3·9%), followed by BMD-PSY (1·6%) and then UMD-BMD (0·4%).

UMD showed a significantly stronger edge weight from irritability to tobacco use compared to BMD (*z*UMD-BMD=·059, p*permuted*<·001). In addition, the following edge weights were significantly stronger compared to PSY: cannabis use (autocorrelation) (*z*UMD-PSY=·051, p*permuted*<·001), agitation-suicidality (*z*UMD-PSY=·048, p*permuted*<·001), tobacco use-suicidality (*z*UMD-PSY=·038, p*permuted*<·001), mood instability-aggression (*z*UMD-PSY=·035, p*permuted*<·001), suicidality-hallucinations (all) (*z*UMD-PSY=·034, p*permuted*<·001) and cannabis-suicidality (*z*UMD-PSY=·033, p*permuted*<·001) (Figure 5).

![Figure 5.](http://medrxiv.org/https://www.medrxiv.org/content/medrxiv/early/2024/03/09/2024.03.08.24303965/F5.medium.gif)

[Figure 5.](http://medrxiv.org/content/early/2024/03/09/2024.03.08.24303965/F5)

Figure 5. 
Heat-maps for pairwise edge comparisons (UMD-BMD, BMD-PSY, UMD-PSY) in temporal sub-networks in permutation analysis. Magnitude and direction of effect size is colour-coded such that for the pairwise comparison Group1-Group2, yellow indicates the edge estimate is more positive in Group1>Group2 and blue indicates the opposite Group1<Group2. Significant pairwise comparisons (corrected p<0·05) are marked with an asterisk (*).

BMD did not show any edge weights that were significantly stronger than UMD, and only one edge weight was stronger compared to PSY (from tearfulness-cannabis use) (*z*BMD-PSY =·049, p*permuted*<·001) (Figure 5).

PSY showed significantly stronger edge weights compared to UMD for the following edges: hallucinations (all)-paranoia (*z*PSY-UMD=·048, p*permuted*<·001), paranoia (autocorrelation) (*z*PSY- UMD=·048, p*permuted*<·001), tobacco use (autocorrelation) (*z*PSY-UMD=·045, p*permuted*<·001) and paranoia-disturbed sleep (*z*PSY-UMD=·033, p*permuted*<·001). PSY also showed significantly stronger edge weights compared to BMD for the edges: hallucinations (all)-disturbed sleep (*z*PSY- BMD=·072, p*permuted*<·001), irritability-tobacco use (*z*PSY-BMD=·070, p*permuted*<·001) and disturbed sleep-paranoia (*z*PSY-BMD=·053, p*permuted*<·001) (Figure 5).

Histograms showing null distributions for significant temporal comparisons are presented in eFigure 4, with the results from the contemporaneous and between-subject matrices presented in eFigures 5-6.

## 4. ​DISCUSSION

This study represents the most extensive temporal network analysis modelling the evolution of prodromal features in SMD, with respect to both the breadth of features and the large sample. We found a dynamic and densely interconnected prodromal phase in the lead-up to SMD onset, mainly with autocorrelations and unidirectional relationships. Notably, this network structure shows consistencies across the SMD diagnostic groups, highlighting a transdiagnostic overlap in the prodromal stages of these conditions.

First, our analysis elucidates the dynamic progression of the SMD prodrome. The best fitting networks were those where all nodes were heavily interconnected. The nature of the associations between prodromal features were predominantly positive; presenting with one feature typically predicts the emergence, rather than absence, of the same or another feature in the future. This reflects the sequential build-up in severity or complexity of the SMD prodrome nearing disorder onset and supports early detection efforts. The strongest positive associations were autocorrelations, meaning that once an individual experiences a feature, it tends to persist during the prodrome. Moreover, there was a higher prevalence of unidirectional (feature A leads to feature B at the next time point but not vice versa) rather than bidirectional (either feature leads to the other emerging at the next time point) relationships. Understanding these dynamics can help map how prodromal features evolve and the impact of targeted interventions.

Second, our analysis revealed that denser networks, with higher saturation, fit better than sparse networks for both the SMD network and the sub-networks specific to SMD diagnostic groups. This highlights the complexity of the interrelationships among prodromal features and holds implications for early detection strategies. Effective prevention of SMD or addressing existing prodromal symptoms requires high-intensity interventions targeting the most influential network features to reduce the risk of further prodromal features emerging.

Third, our findings provide additional evidence to the concept of transdiagnostic features within the prodromal phase of SMD (at least in the context of secondary mental health care).23–28 The minimal edge weight differences among UMD, BMD and PSY sub-networks in the permutation analysis suggest that there are only few relationships specific to diagnostic groups.

Echoing our earlier findings,28 PSY exhibited the most distinctive pattern of relationships between prodromal features, with stronger connections compared to both UMD and BMD in symptom pairs relating to positive symptoms, tobacco use and disturbed sleep. The prominence of positive symptoms in PSY affirms the relevance of psychometric tools, like the CAARMS29 and SIPS,30 which primarily assess positive symptoms to identify psychosis risk with excellent population-level prognostic accuracy.31 However, individuals who test positive on these tools are less likely to develop psychosis than not.31 Our results could inform refined versions of these tools, which are more sensitive to psychosis risk, or individualised prediction models. Interestingly, these findings underscore the disruptive impact of positive psychotic symptoms on sleep. Given that clinical high risk for psychosis individuals (CHR-P)32,33 and people with psychosis34,35 often experience sleep issues, our findings reinforce the potential for interventions targeting sleep disturbances.36

Moreover, central features, indicating highest network influence, differed across sub-networks. In UMD, suicidality was central, which has been shown to be more common in UMD than other mental disorders37 (with suicidal ideation and attempt rates at 53% and 31%, respectively38,39), but less prominent in early prodromal stages.40 For BMD, centrality of elation and irritability aligns with hypomanic symptoms as diagnostic risk factors41, supporting psychometric instruments42–44 for bipolar at-risk45 focusing on these symptoms. In PSY, aggression and agitation were central, but this finding requires careful interpretation as our NLP algorithm for aggression does not distinguish between forms of violence directed to others or oneself, and individuals with PSY are more likely to be victims of violence than the general population.2,46

Our results show evidence for an existing prodrome in UMD, BMD and PSY which is similar but not completely overlapping across the SMD diagnostic groups. This finding supports the potential to broaden preventive services which target SMD. While transdiagnostic early detection services are emerging, they can encompass a range of at-risk states beyond those studied here, including eating disorders, anxiety and personality disorders47–49, and their effectiveness is yet to be determined50. Like with the CHR-P state, effective recruitment strategies are crucial for risk enrichment31 and optimizing preventive intervention potential.

This study, while comprehensive, is subject to several limitations. First, the features which comprise the networks are prodromal in the sense that they are the symptoms that are detectable in secondary care prior to these diagnoses. However, despite our extensive range of prodromal, sociodemographic and treatment variables, there may still be unaccounted factors that influence the temporal evolution of SMD prodromes, such as functioning.51 Future work should focus on mapping specific symptom trajectories to identify confounding factors affecting symptom presence and absence in extensive networks such as ours. Second, to reduce the missingness in the dataset, we used a relatively short look-back period. However, this two-year period before disorder onset aligns with the typical duration of clinical care for at-risk individuals.52 Third, the final population presents a selection bias towards those receiving more frequent secondary care, limiting generalizability. Similarly, specific features, such as disorganized symptoms, may be underrepresented due to the need for consistent clinical visits. However, there were minimal differences between included and excluded individuals, in terms of demographics, clinical variables and presenting symptoms. EHR and NLP-related limitations are discussed in eLimitations 1.

Overall, our study highlights the presence of a detectable transdiagnostic SMD prodrome by modelling the evolution of symptoms and substance use over time. Our findings illustrate the need to understand dynamic symptom progression to fully characterise the prodrome to SMD. These findings also support the potential for broader transdiagnostic early detection services for SMD that provide preventive care to individuals at-risk and a research platform for investigating putative interventions.

## Supporting information

Supplementary Material [[supplements/303965_file02.docx]](pending:yes)

## Data Availability

The data accessed by CRIS remain within an NHS firewall and governance is provided by a patient-led oversight committee. Subject to these conditions, data access is encouraged and those interested should contact Robert Stewart (robert.stewart{at}kcl.ac.uk), CRIS academic lead. There is no permission for data sharing. Covariance matrices to estimate networks and all analysis code are available on GitHub: [https://github.com/m-arribas/network_analysis.git](https://github.com/m-arribas/network_analysis.git).

[https://github.com/m-arribas/network\_analysis.git](https://github.com/m-arribas/network_analysis.git) 

## Funding

MA is supported by the UK Medical Research Council (MR/N013700/1) and King’s College London member of the MRC Doctoral Training Partnership in Biomedical Sciences. JMB has received funding from the Wellcome Trust (WT228268/Z/23/Z). RP has received funding from an NIHR Advanced Fellowship (NIHR301690) and a Medical Research Council (MRC) Health Data Research UK Fellowship (MR/S003118/1). PFP is supported by #NEXTGENERATIONEU (NGEU), funded by the Ministry of University and Research (MUR), National Recovery and Resilience Plan (NRRP), project MNESYS (PE0000006) – A Multiscale integrated approach to the study of the nervous system in health and disease (DN. 1553 11.10.2022).

## Data sharing

The data accessed by CRIS remain within an NHS firewall and governance is provided by a patient-led oversight committee. Subject to these conditions, data access is encouraged and those interested should contact Robert Stewart (robert.stewart{at}kcl.ac.uk), CRIS academic lead. Further details regarding the CRIS platform can be found elsewhere13. There is no permission for data sharing. Covariance matrices to estimate networks and all analysis code are available on GitHub: [https://github.com/m-arribas/network_analysis.git](https://github.com/m-arribas/network_analysis.git).

## Ethics committee approval

Permissions for the study were granted by the Oxfordshire Research Ethics Committee C; because the data set comprised deidentified data, informed consent was not required13.

## Authors’ contribution

MA, JMB and DO designed the study under PFP’s supervision. MA and DO ran the statistical analyses. All authors drafted, edited, and approved the final version of the manuscript.

## Conflict of interest

MA has been employed by F. Hoffmann-La Roche AG outside of the current study. RP has received grant funding from Janssen, and consulting fees from Holmusk, Akrivia Health, Columbia Data Analytics, Boehringer Ingelheim and Otsuka. PFP has received research funds or personal fees from Lundbeck, Angelini, Menarini, Sunovion, Boehringer Ingelheim, Mindstrong, Proxymm Science, outside the current study.

## Footnotes

*   * Joint senior authorship

*   Received March 8, 2024.
*   Revision received March 8, 2024.
*   Accepted March 9, 2024.


*   © 2024, Posted by Cold Spring Harbor Laboratory

This pre-print is available under a Creative Commons License (Attribution 4.0 International), CC BY 4.0, as described at [http://creativecommons.org/licenses/by/4.0/](http://creativecommons.org/licenses/by/4.0/)

## References

1.  1.Estradé A, Onwumere J, Venables J, Gilardi L, Cabrera A, Rico J, et al. The Lived Experiences of Family Members and Carers of People with Psychosis: A Bottom-Up Review Co-Written by Experts by Experience and Academics. Psychopathology. 2023 Jan 23;1–12.
    
    

2.  2.Fusar-Poli P, Estradé A, Stanghellini G, Venables J, Onwumere J, Messas G, et al. The lived experience of psychosis: a bottom-up review co-written by experts by experience and academics. World Psychiatry. 2022 Jun;21(2):168–88.
    
    

3.  3.Fusar-Poli P, Estradé A, Stanghellini G, Esposito CM, Rosfort R, Mancini M, et al. The lived experience of depression: a bottom-up review co-written by experts by experience and academics. World Psychiatry. 2023 Oct;22(3):352–65.
    
    

4.  4.Oliver D, Arribas M, Perry BI, Whiting D, Blackman G, Krakowski K, et al. USING ELECTRONIC HEALTH RECORDS TO FACILITATE PRECISION PSYCHIATRY. Biol Psychiatry. 2024 Feb 24;S0006-3223(24)01107–7.
    
    

5.  5.Wichers M, Wigman JTW, Myin-Germeys I. Micro-level affect dynamics in psychopathology viewed from complex dynamical system theory. Emotion Review. 2015;7:362–7.
    
    [CrossRef](http://medrxiv.org/lookup/external-ref?access_num=10.1177/1754073915590623&link_type=DOI) 

6.  6.Kuppens P, Allen NB, Sheeber LB. Emotional inertia and psychological maladjustment. Psychol Sci. 2010 Jul;21(7):984–91.
    
    [CrossRef](http://medrxiv.org/lookup/external-ref?access_num=10.1177/0956797610372634&link_type=DOI) 
    
    [PubMed](http://medrxiv.org/lookup/external-ref?access_num=20501521&link_type=MED&atom=%2Fmedrxiv%2Fearly%2F2024%2F03%2F09%2F2024.03.08.24303965.atom) 

7.  7.Epskamp S, van Borkulo CD, van der Veen DC, Servaas MN, Isvoranu AM, Riese H, et al. Personalized Network Modeling in Psychopathology: The Importance of Contemporaneous and Temporal Connections. Clinical Psychological Science. 2018 May 1;6(3):416–27.
    
    

8.  8.Levine SZ, Leucht S. Identifying a system of predominant negative symptoms: Network analysis of three randomized clinical trials. Schizophr Res. 2016 Dec;178(1–3):17–22.
    
    

9.  9.McNally RJ. Can network analysis transform psychopathology? Behav Res Ther. 2016 Nov;86:95–104.
    
    

10. 10.Tomba E, Bech P. Clinimetrics and clinical psychometrics: macro- and micro-analysis. Psychother Psychosom. 2012;81(6):333–43.
    
    [CrossRef](http://medrxiv.org/lookup/external-ref?access_num=10.1159/000341757&link_type=DOI) 
    
    [PubMed](http://medrxiv.org/lookup/external-ref?access_num=22964522&link_type=MED&atom=%2Fmedrxiv%2Fearly%2F2024%2F03%2F09%2F2024.03.08.24303965.atom) 

11. 11.van Borkulo C, Boschloo L, Borsboom D, Penninx BWJH, Waldorp LJ, Schoevers RA. Association of Symptom Network Structure With the Course of [corrected] Depression. JAMA Psychiatry. 2015 Dec;72(12):1219–26.
    
    

12. 12.Madhoo M, Levine SZ. Network analysis of the Quick Inventory of Depressive Symptomatology: Reanalysis of the STAR*D clinical trial. Eur Neuropsychopharmacol. 2016 Nov;26(11):1768–74.
    
    

13. 13.Stewart R, Soremekun M, Perera G, Broadbent M, Callard F, Denis M, et al. The South London and Maudsley NHS Foundation Trust Biomedical Research Centre (SLAM BRC) case register: development and descriptive data. BMC Psychiatry. 2009 Aug 12;9(1):51.
    
    [CrossRef](http://medrxiv.org/lookup/external-ref?access_num=10.1186/1471-244X-9-51&link_type=DOI) 
    
    [PubMed](http://medrxiv.org/lookup/external-ref?access_num=19674459&link_type=MED&atom=%2Fmedrxiv%2Fearly%2F2024%2F03%2F09%2F2024.03.08.24303965.atom) 

14. 14.Roberts E, Wessely S, Chalder T, Chang CK, Hotopf M. Mortality of people with chronic fatigue syndrome: a retrospective cohort study in England and Wales from the South London and Maudsley NHS Foundation Trust Biomedical Research Centre (SLaM BRC) Clinical Record Interactive Search (CRIS) Register. Lancet. 2016 Apr 16;387(10028):1638–43.
    
    [PubMed](http://medrxiv.org/lookup/external-ref?access_num=http://www.n&link_type=MED&atom=%2Fmedrxiv%2Fearly%2F2024%2F03%2F09%2F2024.03.08.24303965.atom) 

15. 15.Oram S, Khondoker M, Abas M, Broadbent M, Howard LM. Characteristics of trafficked adults and children with severe mental illness: a historical cohort study. Lancet Psychiatry. 2015 Dec;2(12):1084–91.
    
    

16. 16.Fusar-Poli P, Rutigliano G, Stahl D, Schmidt A, Ramella-Cravaro V, Hitesh S, et al. Deconstructing Pretest Risk Enrichment to Optimize Prediction of Psychosis in Individuals at Clinical High Risk. JAMA Psychiatry. 2016 Dec 1;73(12):1260–7.
    
    

17. 17.Salazar de Pablo G, Estradé A, Cutroni M, Andlauer O, Fusar-Poli P. Establishing a clinical service to prevent psychosis: What, how and when? Systematic review. Transl Psychiatry. 2021 Jan 13;11(1):43.
    
    

18. 18.Epskamp S, Waldorp LJ, Mõttus R, Borsboom D. The Gaussian Graphical Model in Cross-Sectional and Time-Series Data. Multivariate Behav Res. 2018;53(4):453–80.
    
    [PubMed](http://medrxiv.org/lookup/external-ref?access_num=http://www.n&link_type=MED&atom=%2Fmedrxiv%2Fearly%2F2024%2F03%2F09%2F2024.03.08.24303965.atom) 

19. 19.Jackson RG, Patel R, Jayatilleke N, Kolliakou A, Ball M, Gorrell G, et al. Natural language processing to extract symptoms of severe mental illness from clinical text: the Clinical Record Interactive Search Comprehensive Data Extraction (CRIS-CODE) project. BMJ Open. 2017 Jan 1;7(1):e012012.
    
    [Abstract/FREE Full Text](http://medrxiv.org/lookup/ijlink/YTozOntzOjQ6InBhdGgiO3M6MTQ6Ii9sb29rdXAvaWpsaW5rIjtzOjU6InF1ZXJ5IjthOjQ6e3M6ODoibGlua1R5cGUiO3M6NDoiQUJTVCI7czoxMToiam91cm5hbENvZGUiO3M6NzoiYm1qb3BlbiI7czo1OiJyZXNpZCI7czoxMToiNy8xL2UwMTIwMTIiO3M6NDoiYXRvbSI7czo1MDoiL21lZHJ4aXYvZWFybHkvMjAyNC8wMy8wOS8yMDI0LjAzLjA4LjI0MzAzOTY1LmF0b20iO31zOjg6ImZyYWdtZW50IjtzOjA6IiI7fQ==) 

20. 20.Barnby JM, Haslbeck JMB, Rosen C, Harrow M. Modelling the Longitudinal Dynamics of Paranoia in Psychosis: A Temporal Network Analysis Over 20 Years [Internet]. medRxiv; 2023 [cited 2023 Jan 17]. p. 2023.01.06.23284268. Available from: [https://www.medrxiv.org/content/10.1101/2023.01.06.23284268v2](https://www.medrxiv.org/content/10.1101/2023.01.06.23284268v2)
    
    

21. 21.Epskamp S, Borsboom D, Fried EI. Estimating psychological networks and their accuracy: A tutorial paper. Behav Res. 2018 Feb 1;50(1):195–212.
    
    

22. 22.Fredrickson MM, Chen Y. Permutation and randomization tests for network analysis. Social Networks. 2019 Oct 1;59:171–83.
    
    

23. 23.Uhlhaas PJ, Davey CG, Mehta UM, Shah J, Torous J, Allen NB, et al. Towards a youth mental health paradigm: a perspective and roadmap. Mol Psychiatry. 2023 Aug 14;1–11.
    
    

24. 24.Shah JL, Scott J, McGorry PD, Cross SPM, Keshavan MS, Nelson B, et al. Transdiagnostic clinical staging in youth mental health: a first international consensus statement. World Psychiatry. 2020;19(2):233–42.
    
    [CrossRef](http://medrxiv.org/lookup/external-ref?access_num=10.1002/wps.20745&link_type=DOI) 
    
    [PubMed](http://medrxiv.org/lookup/external-ref?access_num=http://www.n&link_type=MED&atom=%2Fmedrxiv%2Fearly%2F2024%2F03%2F09%2F2024.03.08.24303965.atom) 

25. 25.Shah JL. Bringing Clinical Staging to Youth Mental Health: From Concept to Operationalization (and Back Again). JAMA Psychiatry. 2019 Nov 1;76(11):1121–3.
    
    

26. 26.Shah JL, Jones N, van Os J, McGorry PD, Gülöksüz S. Early intervention service systems for youth mental health: integrating pluripotentiality, clinical staging, and transdiagnostic lessons from early psychosis. Lancet Psychiatry. 2022 May;9(5):413–22.
    
    

27. 27.McGorry PD, Hartmann JA, Spooner R, Nelson B. Beyond the “at risk mental state” concept: transitioning to transdiagnostic psychiatry. World Psychiatry. 2018 Jun;17(2):133–42.
    
    [CrossRef](http://medrxiv.org/lookup/external-ref?access_num=10.1002/wps.20514&link_type=DOI) 
    
    [PubMed](http://medrxiv.org/lookup/external-ref?access_num=http://www.n&link_type=MED&atom=%2Fmedrxiv%2Fearly%2F2024%2F03%2F09%2F2024.03.08.24303965.atom) 

28. 28.Arribas M, Oliver D, Patel R, Kornblum D, Shetty H, Damiani S, et al. A TRANSDIAGNOSTIC PRODROME FOR SEVERE MENTAL DISORDERS: AN
    
    

29. 29.ELECTRONIC HEALTH RECORD STUDY. (in press in Molecular Psychiatry).
    
    

30. 29.Yung AR, Yung AR, Pan Yuen H, Mcgorry PD, Phillips LJ, Kelly D, et al. Mapping the Onset of Psychosis: The Comprehensive Assessment of At-Risk Mental States. Aust N Z J Psychiatry. 2005 Nov 1;39(11–12):964–71.
    
    [CrossRef](http://medrxiv.org/lookup/external-ref?access_num=10.1111/j.1440-1614.2005.01714.x&link_type=DOI) 
    
    [PubMed](http://medrxiv.org/lookup/external-ref?access_num=16343296&link_type=MED&atom=%2Fmedrxiv%2Fearly%2F2024%2F03%2F09%2F2024.03.08.24303965.atom) 
    
    [Web of Science](http://medrxiv.org/lookup/external-ref?access_num=000233205400003&link_type=ISI) 

31. 30.McGlashan T, Walsh B, Woods S. The Psychosis-Risk Syndrome: Handbook for Diagnosis and Follow-Up. Oxford University Press, USA; 2010. 256 p.
    
    

32. 31.Oliver D, Arribas M, Radua J, Salazar de Pablo G, De Micheli A, Spada G, et al. Prognostic accuracy and clinical utility of psychometric instruments for individuals at clinical high-risk of psychosis: a systematic review and meta-analysis. Mol Psychiatry. 2022 Sep;27(9):3670–8.
    
    

33. 32.Poe SL, Brucato G, Bruno N, Arndt LY, Ben-David S, Gill KE, et al. Sleep disturbances in individuals at clinical high risk for psychosis. Psychiatry Res. 2017 Mar;249:240–3.
    
    [PubMed](http://medrxiv.org/lookup/external-ref?access_num=http://www.n&link_type=MED&atom=%2Fmedrxiv%2Fearly%2F2024%2F03%2F09%2F2024.03.08.24303965.atom) 

34. 33.Tan HY, Ang YG. First-episode psychosis in the military: a comparative study of prodromal symptoms. Aust N Z J Psychiatry. 2001 Aug;35(4):512–9.
    
    [CrossRef](http://medrxiv.org/lookup/external-ref?access_num=10.1046/j.1440-1614.2001.00912.x&link_type=DOI) 
    
    [PubMed](http://medrxiv.org/lookup/external-ref?access_num=11531734&link_type=MED&atom=%2Fmedrxiv%2Fearly%2F2024%2F03%2F09%2F2024.03.08.24303965.atom) 
    
    [Web of Science](http://medrxiv.org/lookup/external-ref?access_num=000171409200137&link_type=ISI) 

35. 34.Cohrs S. Sleep disturbances in patients with schizophrenia : impact and effect of antipsychotics. CNS Drugs. 2008;22(11):939–62.
    
    [CrossRef](http://medrxiv.org/lookup/external-ref?access_num=10.2165/00023210-200822110-00004&link_type=DOI) 
    
    [PubMed](http://medrxiv.org/lookup/external-ref?access_num=18840034&link_type=MED&atom=%2Fmedrxiv%2Fearly%2F2024%2F03%2F09%2F2024.03.08.24303965.atom) 
    
    [Web of Science](http://medrxiv.org/lookup/external-ref?access_num=000260569800004&link_type=ISI) 

36. 35.Afonso P, Brissos S, Cañas F, Bobes J, Bernardo-Fernandez I. Treatment adherence and quality of sleep in schizophrenia outpatients. Int J Psychiatry Clin Pract. 2014 Jan;18(1):70–6.
    
    

37. 36.Waite F, Kabir T, Johns L, Mollison J, Tsiachristas A, Petit A, et al. Treating sleep problems in young people at ultra-high-risk of psychosis: study protocol for a single-blind parallel group randomised controlled feasibility trial (SleepWell). BMJ Open. 2020 Nov 1;10(11):e045235.
    
    [Abstract/FREE Full Text](http://medrxiv.org/lookup/ijlink/YTozOntzOjQ6InBhdGgiO3M6MTQ6Ii9sb29rdXAvaWpsaW5rIjtzOjU6InF1ZXJ5IjthOjQ6e3M6ODoibGlua1R5cGUiO3M6NDoiQUJTVCI7czoxMToiam91cm5hbENvZGUiO3M6NzoiYm1qb3BlbiI7czo1OiJyZXNpZCI7czoxMzoiMTAvMTEvZTA0NTIzNSI7czo0OiJhdG9tIjtzOjUwOiIvbWVkcnhpdi9lYXJseS8yMDI0LzAzLzA5LzIwMjQuMDMuMDguMjQzMDM5NjUuYXRvbSI7fXM6ODoiZnJhZ21lbnQiO3M6MDoiIjt9) 

38. 37.Cai H, Xie XM, Zhang Q, Cui X, Lin JX, Sim K, et al. Prevalence of Suicidality in Major Depressive Disorder: A Systematic Review and Meta-Analysis of Comparative Studies. Front Psychiatry. 2021 Sep 16;12:690130.
    
    

39. 38.Dong M, Wang SB, Li Y, Xu DD, Ungvari GS, Ng CH, et al. Prevalence of suicidal behaviors in patients with major depressive disorder in China: A comprehensive meta-analysis. J Affect Disord. 2018 Jan 1;225:32–9.
    
    

40. 39.Dong M, Zeng LN, Lu L, Li XH, Ungvari GS, Ng CH, et al. Prevalence of suicide attempt in individuals with major depressive disorder: a meta-analysis of observational surveys. Psychol Med. 2019 Jul;49(10):1691–704.
    
    [CrossRef](http://medrxiv.org/lookup/external-ref?access_num=10.1017/S0033291718002301&link_type=DOI) 

41. 40.Benasi G, Fava GA, Guidi J. Prodromal Symptoms in Depression: A Systematic Review. PPS. 2021;90(6):365–72.
    
    

42. 41.Axelson D, Goldstein B, Goldstein T, Monk K, Yu H, Hickey MB, et al. Diagnostic Precursors to Bipolar Disorder in Offspring of Parents With Bipolar Disorder: A Longitudinal Study. AJP. 2015 Jul;172(7):638–46.
    
    

43. 42.Correll CU, Olvet DM, Auther AM, Hauser M, Kishimoto T, Carrión RE, et al. The Bipolar Prodrome Symptom Interview and Scale–Prospective (BPSS-P): description and validation in a psychiatric sample and healthy controls. Bipolar Disorders. 2014;16(5):505–22.
    
    

44. 43.Fusar-Poli P, De Micheli A, Rocchetti M, Cappucciati M, Ramella-Cravaro V, Rutigliano G, et al. Semistructured Interview for Bipolar At Risk States (SIBARS). Psychiatry Res. 2018 Jun;264:302–9.
    
    [PubMed](http://medrxiv.org/lookup/external-ref?access_num=http://www.n&link_type=MED&atom=%2Fmedrxiv%2Fearly%2F2024%2F03%2F09%2F2024.03.08.24303965.atom) 

45. 44.Leopold K, Ritter P, Correll CU, Marx C, Özgürdal S, Juckel G, et al. Risk constellations prior to the development of bipolar disorders: rationale of a new risk assessment tool. J Affect Disord. 2012 Feb;136(3):1000–10.
    
    [CrossRef](http://medrxiv.org/lookup/external-ref?access_num=10.1016/j.jad.2011.06.043&link_type=DOI) 
    
    [PubMed](http://medrxiv.org/lookup/external-ref?access_num=21802741&link_type=MED&atom=%2Fmedrxiv%2Fearly%2F2024%2F03%2F09%2F2024.03.08.24303965.atom) 

46. 45.Bechdolf A, Nelson B, Cotton SM, Chanen A, Thompson A, Kettle J, et al. A preliminary evaluation of the validity of at-risk criteria for bipolar disorders in help-seeking adolescents and young adults. J Affect Disord. 2010 Dec;127(1–3):316–20.
    
    [CrossRef](http://medrxiv.org/lookup/external-ref?access_num=10.1016/j.jad.2010.06.016&link_type=DOI) 
    
    [PubMed](http://medrxiv.org/lookup/external-ref?access_num=20619465&link_type=MED&atom=%2Fmedrxiv%2Fearly%2F2024%2F03%2F09%2F2024.03.08.24303965.atom) 

47. 46.Radua J, Ramella-Cravaro V, Ioannidis JPA, Reichenberg A, Phiphopthatsanee N, Amir T, et al. What causes psychosis? An umbrella review of risk and protective factors. World psychiatry: official journal of the World Psychiatric Association (WPA). 2018 Feb;17(1):49–66.
    
    

48. 47.Iorfino F, Scott EM, Carpenter JS, Cross SP, Hermens DF, Killedar M, et al. Clinical Stage Transitions in Persons Aged 12 to 25 Years Presenting to Early Intervention Mental Health Services With Anxiety, Mood, and Psychotic Disorders. JAMA Psychiatry. 2019 Nov 1;76(11):1167–75.
    
    

49. 48.Carpenter JS, Scott J, Iorfino F, Crouse JJ, Ho N, Hermens DF, et al. Predicting the emergence of full-threshold bipolar I, bipolar II and psychotic disorders in young people presenting to early intervention mental health services. Psychological Medicine. 2022 Jul;52(10):1990–2000.
    
    

50. 49.Hartmann JA, Nelson B, Spooner R, Paul Amminger G, Chanen A, Davey CG, et al. Broad clinical high-risk mental state (CHARMS): Methodology of a cohort study validating criteria for pluripotent risk. Early Interv Psychiatry. 2019 Jun;13(3):379–86.
    
    [CrossRef](http://medrxiv.org/lookup/external-ref?access_num=10.1111/eip.12483&link_type=DOI) 

51. 50.Hartmann JA, McGorry PD, Destree L, Amminger GP, Chanen AM, Davey CG, et al. Pluripotential Risk and Clinical Staging: Theoretical Considerations and Preliminary Data From a Transdiagnostic Risk Identification Approach. Front Psychiatry. 2020;11:553578.
    
    [CrossRef](http://medrxiv.org/lookup/external-ref?access_num=10.3389/fpsyt.2020.553578&link_type=DOI) 

52. 51.Paquin V, Malla AK, Iyer SN, Lepage M, Joober R, Shah JL. Combinations and Temporal Associations Among Precursor Symptoms Before a First Episode of Psychosis. Schizophrenia Bulletin. 2023 Oct 20;sbad152.
    
    

53. 52.Salazar de Pablo G, Estradé A, Cutroni M, Andlauer O, Fusar-Poli P. Establishing a clinical service to prevent psychosis: What, how and when? Systematic review. Transl Psychiatry. 2021 Jan 13;11(1):1–14.