Identification of transdiagnostic psychiatric disorder subtypes using unsupervised learning

Abstract

Psychiatric disorders show heterogeneous symptoms and trajectories, with current nosology not accurately reflecting their molecular etiology and the variability and symptomatic overlap within and between diagnostic classes. This heterogeneity impedes timely and targeted treatment. Our study aimed to identify psychiatric patient clusters that share clinical and genetic features and may profit from similar therapies. We used high-dimensional data clustering on deep clinical data to identify transdiagnostic groups in a discovery sample (N = 1250) of healthy controls and patients diagnosed with depression, bipolar disorder, schizophrenia, schizoaffective disorder, and other psychiatric disorders. We observed five diagnostically mixed clusters and ordered them based on severity. The least impaired cluster 0, containing most healthy controls, showed general well-being. Clusters 1-3 differed predominantly regarding levels of maltreatment, depression, daily functioning, and parental bonding. Cluster 4 contained most patients diagnosed with psychotic disorders and exhibited the highest severity in many dimensions, including medication load. Depressed patients were present in all clusters, indicating that we captured different disease stages or subtypes. We replicated all but the smallest cluster 1 in an independent sample (N = 622). Next, we analyzed genetic differences between clusters using polygenic scores (PGS) and the psychiatric family history. These genetic variables differed mainly between clusters 0 and 4 (prediction area under the receiver operating characteristic curve (AUC) = 81%; significant PGS: cross-disorder psychiatric risk, schizophrenia, and educational attainment). Our results confirm that psychiatric disorders consist of heterogeneous subtypes sharing molecular factors and symptoms. The identification of transdiagnostic clusters advances our understanding of the heterogeneity of psychiatric disorders and may support the development of personalized treatments.

Fig. 1. Cluster characterization in the discovery sample with clinical and genetic variables not used in the clustering pipeline.

None of the variables shown in this Fig. 1 or Table S3 were included in the clustering pipeline. B–H a horizontal line represents the mean and the error bars indicate the standard deviation. The dot size is proportional to the number of individuals with the given value. Variables that were significant in the one-vs-all comparisons are marked with an asterisk sign. E–H show all PGS significant after Bonferroni correction (adjusted p < 0.05), tested using the Westfall and Young procedure (Methods S6), in either one-vs-all or one-vs-one analyses (Tables S12–S13). PGS were standardized by Z score transformation, the y axis unit is standard deviations. A The distribution of diagnoses within clusters. B The Global Assessment of Functioning (GAF) score, used for sorting clusters. Lower scores imply more severe impairment. C The number of times an individual was hospitalized. D The medication load index [59], reflecting the dose and variety of different medications taken. E Psychiatric cross-disorder PGS, significantly different in two one-vs-all analyses (lower in cluster 0, Bonferroni-corrected p = 0.004; higher in cluster 4, corrected p = 0.01). F MDD PGS, significantly different in two one-vs-all analyses (lower in cluster 0, p = 0.008; higher in cluster 4, corrected p = 0.04). G Schizophrenia PGS, significantly different in two one-vs-all analyses (lower in cluster 0, corrected p = 0.04; higher in cluster 4, corrected p = 0.01). H Educational attainment PGS, significantly different in one one-vs-all analysis (lower in cluster 4, corrected p = 0.004).

Fig. 2. Cluster characterization in the discovery sample with variables used in the clustering pipeline.

A horizontal line represents the mean, and the error bars indicate the standard deviation, whereas the dot size is proportional to the number of individuals with the given value. Variables that were significant in the one-vs-all comparisons are marked with an asterisk sign. A Hamilton Depression Rating Scale (HAMD, 21 items, clinician-administered), range 0–66, scores >7 indicate (mild) depression. B Hamilton Anxiety Rating Scale (HAMA), range 0–56, scores >17 indicate mild to moderate anxiety severity. C Scale for the Assessment of Negative Symptoms (SANS, sum score), range 0–80, a higher score indicates more severe negative symptoms. For subscales, see Table S3. D Scale for the Assessment of Positive Symptoms (SAPS, sum score), range 0–86, a higher score indicates more severe positive symptoms. For subscales, see Table S3. E Beck Depression Inventory (BDI-II, self-reported), range 0–63, scores >9 indicate (mild) depression. F Symptom Checklist–Global Severity Index, an index of overall psychological distress, range 0–4, higher scores reflect higher levels of psychopathological distress as well as a greater severity of self-reported symptoms. G Childhood Trauma Questionnaire sum score, range 25–125, a higher score indicates more experiences of childhood trauma. H SF36–Quality of life measurements–Mental health, range 0–100, high scores define a more favorable health state.

Fig. 3. Cluster characterization in the replication sample with clinical and genetic variables not used in the clustering pipeline.

B–H a horizontal line represents the mean, and the error bars indicate the standard deviation, whereas the dot size is proportional to the number of individuals with the given value. E–H show all PGS that were significant after Bonferroni correction (adjusted p < 0.05) in either one-vs-all or one-vs-one analyses using the Westfall and Young procedure (Methods S6) in the discovery-stage analysis. All p values for the full replication sample are shown in Tables S19 and S20. PGS were standardized by Z score transformation, the y axis unit are standard deviations. A The distribution of diagnoses within clusters. B The Global Assessment of Functioning (GAF) score, used for sorting clusters. Lower scores imply more severe impairment. C The number of times an individual was hospitalized. D Medication load index [59], reflecting dose and variety of different medications taken. E Psychiatric cross-disorder PGS, replicated for the comparison cluster 0-vs-all (corrected p = 0.03). F Major depressive disorder PGS, replicated for the comparison cluster 4-vs-all (corrected p = 0.01). G Schizophrenia PGS, replicated for the comparison cluster 0-vs-all (p = 0.005). H Educational attainment PGS, replicated for the comparison cluster 4-vs-all (corrected p = 0.005).