Skip to main content
NCBI home page
UKPMC Funders Author Manuscripts logoLink to UKPMC Funders Author Manuscripts
. Author manuscript; available in PMC: 2022 Jul 31.
Published in final edited form as: Psychiatry Res Neuroimaging. 2020 Jul 15;304:111136. doi: 10.1016/j.pscychresns.2020.111136

Subtyping schizophrenia based on symptomatology and cognition using a data driven approach

Luis FS Castro-de-Araujo a,b,*, Daiane B Machado a,c, Maurício L Barreto a,d, Richard AA Kanaan b
PMCID: PMC7613209  EMSID: EMS145848  PMID: 32707455

Abstract

Schizophrenia is a highly heterogeneous disorder, not only in its phenomenology but in its clinical course. This limits the usefulness of the diagnosis as a basis for both research and clinical management. Methods of reducing this heterogeneity may inform the diagnostic classification. With this in mind, we performed k-means clustering with symptom and cognitive measures to generate groups in a machine-driven way. We found that our data was best organised in three clusters: high cognitive performance, high positive symptomatology, low positive symptomatology. We hypothesized that these clusters represented biological categories, which we tested by comparing these groups in terms of brain volumetric information. We included all the groups in an ANCOVA analysis with post hoc tests, where brain volume areas were modelled as dependent variables, controlling for age and estimated intracranial volume. We found six brain volumes significantly differed between the clusters: left caudate, left cuneus, left lateral occipital, left inferior temporal, right lateral, and right pars opercularis. The k-means clustering provides a way of subtyping schizophrenia which appears to have a biological basis, though one that requires both replication and confirmation of its clinical significance.

Keywords: schizophrenia, positive and negative symptoms, clustering, data-driven subgrouping

1. Introduction

Schizophrenia is a heterogeneous disorder. A broad range of risks has been associated with the condition, from maternal infection (al-Haddad et al., 2019), to urbanicity (Del-Ben et al., 2019) and migration (Fearon et al., 2006). It also has been associated with at least 108 genes (Ripke et al., 2014), life events (Tandon et al., 2008), cognitive performance, and brain alterations (Castro-de-Araujo and Kanaan, 2017). Not surprisingly, subjects with this disorder present with varied clinical phenomenology and course, which leads to calls for endophenotypes that might allow more precise prediction of the clinical course and treatment effect.

Subtyping schizophrenia dates back to Bleuler (Fusar-Poli and Politi, 2008), who divided it using phenomenological features. These subtypes – paranoid, catatonic, hebephrenic and simple - have evolved significantly but are still retained in modern classification manuals. Other phenomenological classifications followed (Jablensky, 2010), notably the ‘deficit’ and ‘non-deficit’ types, which have found significant biological and clinical support (Cohen et al., 2010). For example, assessments of cognitive performance support the “deficit” subtype (Dickinson et al. 2018) with demographic characteristics – typically males with low economic status. And a pattern of brain volumetric alterations has also been associated with it, with changes noted in numerous regions including Medial orbital frontal, Middle temporal and Superior temporal lobes (Weinberg et al., 2016). These findings have proven durable and stable enough that many authors consider the subgroup a discrete category (Kanchanatawan et al., 2018).

Another longstanding phenomenological approach is the schizophrenia spectrum, defining subgroups of severity and disease course, a distinction that has its roots in Bleuler (Fusar-Poli and Politi, 2008) and Kraepelin (Kendler and Engstrom, 2016). The latest variant of this is the definition a group known as the ‘clinical high-risk state for psychosis’ (Fusar-Poli, 2017). This construct captures a pre-psychotic phase, to identify those with higher risk of progressing to full blown psychosis. It has been found that subjects at risk of psychosis presents with altered cognitive performance and altered social cognition prior to the onset of the disorder (Fusar-Poli et al., 2013). These subjects also present with brain volumetric alterations in areas typically altered in schizophrenics, like the hippocampus and the anterior cingulate cortex (Fusar-Poli et al., 2013).

There is more recent history of using statistical methods to define categories. Andreasen and colleagues used factor analysis with varimax rotation based on clinical scales to define subgroups (Andreasen et al., 1995; Andreasen and Olsen, 1982). This approach became popular with other groups using the same approach (Liddle, 2019, 1987), supporting three main subtypes: disorganized, psychotic and negative. Again, biological support for these has come from other research groups, for example in functional brain alterations differing between groups (Hazlett and Buchsbaum, 2001; Schröder et al., 1995). Most impressively, the negative sub-group has been linked to hypofrontality in FDG-PET (Hazlett et al., 2019, 2000).

Another statistical study has yielded the “distress” subgroup. It is characterized by anxious and depressive symptoms (Lönnqvist et al., 2009), including elevated cortisol secretion (Corcoran et al., 2012). This subtype incorporates concepts from the study of personality and implies that subjects can present with strong positive or negative affectivity (Horan et al., 2008).

The approaches listed above have focused on subtyping schizophrenia using either cognitive performance or symptomatology. However, modern machine learning techniques offer clustering methods that are neutral regarding the type of data used in the algorithm. We set out to apply a simple and now well-established clustering method, k-means, in which the algorithm identifies properties of the data points (subjects analysed) that makes them either closer or more distant to cluster prototypes, partitioning the set of subjects. Our approach was to cluster the subjects with schizophrenia based on their cognitive and symptom scores, and then determine whether the clusters showed differences in their regional brain volumes. This aimed to identify a more complex endophenotype, including symptom severity and cognitive performance at the same time, and to seek a neurobiological correlate in their brain volumes, which have been shown to be altered in the previously defined subtypes.

2. Methods

2.1. Data source

Subjects from the NUSDAST data set (NU Schizophrenia Data and Software Tool Federation using BIRN Infrastructure, North-Western University, Illinois) (Wang et al., 2013) were used for this analysis. These are made accessible through the online interface SchizConnect (http://schizconnect.org) (Ambite et al. 2015).

2.2. Data set characteristics

The NUSDAST dataset comprises subjects recruited by advert from the community. Diagnosis was by DSM-IV criteria, by a team comprising a Psychiatrist, using a semi-structured interview (the Structured Clinical Interview for DSM-IV Axis I Disorders (SCID)) (First et al., 1997), in consensus with a research assistant. The team were trained and diagnoses regularly calibrated. Exclusion criteria were subjects who met DSM-IV criteria for substance abuse/dependence, who had a clinically unstable mental state or other severe clinical condition, who had a past or present head injury, or met DSM-IV criteria for mild (or greater) mental retardation (Harms et al., 2007).

This dataset has been previously described (Csernansky et al., 2002; Harms et al., 2007). It comprised two cohorts with identical selection procedures, which were later anonymized and made available online following the stripping of identifiable information in compliance with HIPAA de-identification. A complete explanation of the procedure is presented elsewhere (Wang et al., 2013). After filtering for subjects with schizophrenia who had completed both psychological assessment and neuroimaging we ended up with 151 subjects for this study, all from North American.

2.3. Neuroimaging

The NUSDAST data set includes structural neuroimaging data, available in a processed form. According to Ambite et al. (2015), MRI scans for the NUSDAST data set were performed on a 1.5 T Vision scanner platform (Siemens Medical Systems). FreeSurfer version 3.0.4, (http://surfer.nmr.mgh.harvard.edu/), was used to obtain cortical parcellations. It proceeds by automated neuroanatomical labelling of locations on a cortical surface model (Desikan et al., 2006; Wang et al., 2013). These procedures are detailed in previous papers (Dale et al., 1999; Fischl and Dale, 2000; Fischl et al., 2004). FreeSurfer provides calculated volumes for several cortical areas, including white and grey matter. The SchizConnect interface for the NUSDAST dataset provided the parcellations and volumes in mm3, and scores for numerous cognitive assessments.

2.4. Cognitive assessment and symptom severity

The data set included assessments for numerous cognitive domains, however we used the data for crystallised intelligence, as captured by the scaled score from the Vocabulary subtest of the Wechsler Adult Intelligence Scale (WMS-III) (Wechsler, 1995). The dictionary for this data set was previously published (Wang et al., 2013).

Clinical scales obtained were the Scale for the Assessment of Positive Symptoms (Andreasen, 1984) and the Scale for the Assessment of Negative Symptoms (Andreasen, 1983). Based on these, scores of psychotic, disorganized and negative symptoms were calculated (Ambite et al., 2015). Finally, the data set also provided calculated z-scored values for the clinical domains (psychotic, disorganized and negative symptoms) and for the cognitive assessments above.

2.5. Data analysis

We used R version 3.5.2 (https://cran.r-project.org/) in all further steps of our analyses. The volumetric data information and the scores of the cognitive assessments were provided in separate files, which we linked manually using a data dictionary also provided by the authors of the data set. We accessed only data from the first visit (from baseline) of subjects with a known diagnosis of schizophrenia, and after filtering for these we ended up with 151 observations. We used the r-base package kmeans for clustering and the JVM package for the ANCOVAS (Selker et al., 2019).

Clustering methods partition data values into smaller groups or clusters. K-means is a clustering method that tries to minimise the distance from each data point to the centre of the cluster using the Euclidean distance. It is considered a machine learning method, because the algorithm can decide the position of each observation using properties of these observations and without human interference. We used the z-scored variables to perform the investigation of the optimal number of centres, and the k-means clustering. After this, we labelled all subjects with their respective cluster and performed ANCOVAS with regional brain volumes as the dependent variables and controlled for age and the estimated intracranial volume. See Table 2 for the list of areas analysed.

Table 2.

Brain areas, their mean (SD) volumes (mm3) across groups. P-values calculated with Bonferroni-corrected ANCOVAS. n, number of observations. n2, is a measure of effect power for each comparison.

Overall low-positive high-positive performance F df etaSq pvalue
L Cortex Volume 217678 (30942) 217143 (29292) 207870 (28849) 226581 (32543) 9.32 2 0.04 0.00 *
R Cortex Volume 217756 (31084) 217170 (29717) 208431 (28458) 226294 (33087) 8.40 2 0.03 0.00 *
Cortex Volume 435434 (61923) 434312 (58907) 416300 (57142) 452876 (65565) 9.06 2 0.04 0.00 *
L Cortical White Matter Vol 222032 (32417) 230114 (31517) 209425 (33936) 225333 (29675) 0.68 2 0.00 0.51
R Cortical White Matter Vol 223875 (32647) 232233 (32382) 210945 (33932) 227197 (29336) 0.83 2 0.00 0.44
Cortical White Matter Vol 445907 (64998) 462346 (63824) 420370 (67824) 452530 (58929) 0.76 2 0.00 0.47
Subcortical Gray Vol 57566 (6285) 58566 (5225) 55613 (7245) 58312 (6141) 0.34 2 0.00 0.71
Total Gray Vol 587690 (77072) 589060 (76830) 562206 (70329) 608261 (79015) 9.46 2 0.03 0.00 *
Supratentorial Vol 959915 (122573) 976013 (116049) 912556 (120805) 985560 (122555) 2.49 2 0.01 0.09
L Cerebellum White Matter 13979 (2077) 14453 (2081) 12969 (1626) 14406 (2175) 1.93 2 0.03 0.15
L Cerebellum Cortex 47383 (6330) 48115 (7865) 45270 (5403) 48514 (5128) 1.19 2 0.01 0.31
L Thalamus 6907 (935) 6989 (718) 6589 (1044) 7103 (976) 1.02 2 0.01 0.37
L Caudate 3661 (479) 3694 (426) 3432 (490) 3825 (453) 3.51 2 0.06 0.03 *
L Putamen 6147 (804) 6244 (605) 5999 (937) 6184 (852) 0.02 2 0.00 0.98
L Pallidum 1763 (291) 1811 (234) 1712 (364) 1762 (273) 0.11 2 0.00 0.89
L Hippocampus 3678 (441) 3699 (387) 3604 (445) 3722 (490) 0.15 2 0.00 0.86
L Amygdala 1414 (248) 1431 (265) 1334 (154) 1467 (284) 0.69 2 0.01 0.50
L Accumbens 700 (158) 733 (163) 681 (161) 686 (152) 0.43 2 0.01 0.65
R Cerebellum White Matter 14240 (2170) 14774 (2137) 13140 (1715) 14688 (2266) 2.37 2 0.04 0.10
R Cerebellum Cortex 48118 (6514) 48909 (8419) 45891 (5032) 49292 (5232) 1.22 2 0.01 0.30
R Thalamus 6730 (905) 6812 (798) 6495 (1033) 6854 (873) 0.49 2 0.00 0.62
R Caudate 3888 (518) 3957 (436) 3651 (535) 4026 (520) 2.57 2 0.04 0.08
R Putamen 5950 (837) 6120 (724) 5811 (939) 5912 (845) 0.27 2 0.00 0.76
R Pallidum 1856 (319) 1901 (245) 1836 (402) 1831 (305) 0.34 2 0.01 0.71
R Hippocampus 3783 (433) 3800 (357) 3706 (465) 3832 (475) 0.37 2 0.01 0.69
R Amygdala 1633 (302) 1642 (309) 1573 (213) 1675 (356) 0.22 2 0.00 0.80
R Accumbens 675 (138) 711 (130) 651 (127) 662 (152) 0.77 2 0.01 0.47
L Banks of the superior temporal sulcus 976 (179) 993 (201) 927 (128) 1003 (193) 0.17 2 0.00 0.84
L Caudal anterior cingulate 635 (159) 678 (201) 617 (131) 611 (130) 2.26 2 0.03 0.11
L Caudal middle frontal 2264 (358) 2203 (345) 2223 (350) 2355 (371) 2.90 2 0.04 0.06
L Cuneus 1360 (262) 1291 (269) 1347 (240) 1434 (261) 4.25 2 0.07 0.02 *
L Entorhinal 375 (84) 366 (70) 349 (80) 406 (91) 3.10 2 0.05 0.05
L Fusiform 3157 (510) 3258 (479) 2940 (483) 3247 (518) 1.14 2 0.01 0.33
L Inferior parietal 4498 (746) 4519 (856) 4404 (772) 4558 (624) 0.83 2 0.01 0.44
L Inferior temporal 3125 (570) 3119 (635) 2986 (549) 3249 (513) 0.99 2 0.01 0.38
L Isthmus cingulate 1009 (174) 1051 (186) 941 (169) 1028 (155) 0.71 2 0.01 0.49
L Lateral occipital 4676 (718) 4599 (581) 4485 (704) 4912 (801) 3.19 2 0.04 0.05 *
L Lateral orbital frontal 2501 (374) 2491 (333) 2384 (350) 2610 (410) 1.94 2 0.03 0.15
L Lingual 2922 (477) 2999 (429) 2756 (475) 2993 (499) 0.62 2 0.01 0.54
L Medial orbital frontal 1846 (281) 1917 (234) 1750 (333) 1861 (258) 0.66 2 0.01 0.52
L Middle temporal 2966 (481) 2972 (489) 2839 (488) 3069 (458) 1.14 2 0.01 0.32
L Parahippocampal 690 (110) 717 (122) 636 (96) 712 (96) 2.33 2 0.04 0.10
L Paracentral 1376 (226) 1405 (216) 1370 (192) 1355 (264) 1.25 2 0.02 0.29
L Pars opercularis 1600 (259) 1614 (287) 1541 (193) 1636 (279) 0.27 2 0.01 0.76
L Pars orbitalis 613 (99) 618 (88) 583 (107) 635 (99) 0.83 2 0.01 0.44
L Pars triangularis 1265 (216) 1280 (218) 1206 (165) 1303 (246) 0.55 2 0.01 0.58
L Pericalcarine 1283 (268) 1259 (245) 1236 (254) 1346 (295) 1.18 2 0.02 0.31
L Postcentral gyrus 4036 (563) 4074 (619) 3864 (442) 4150 (585) 0.54 2 0.01 0.59
L Posterior cingulate 1177 (212) 1207 (271) 1141 (169) 1181 (185) 0.38 2 0.01 0.69
L Precentral 4843 (691) 4890 (624) 4545 (570) 5056 (773) 2.29 2 0.03 0.11
L Precuneus 3677 (550) 3747 (531) 3482 (543) 3778 (550) 0.54 2 0.01 0.59
L Rostral anterior cingulate 811 (204) 810 (241) 772 (178) 844 (190) 1.04 2 0.01 0.36
L Rostral middle frontal 5548 (944) 5592 (766) 5175 (947) 5827 (1014) 2.50 2 0.02 0.09
L Superior frontal 7066 (998) 7072 (925) 6719 (874) 7359 (1097) 2.19 2 0.03 0.12
L Superior parietal 5263 (719) 5359 (713) 4973 (580) 5422 (780) 0.86 2 0.01 0.43
L Superior temporal 3726 (556) 3755 (570) 3529 (630) 3868 (433) 1.52 2 0.02 0.22
L Supramarginal 3666 (587) 3762 (656) 3547 (624) 3678 (483) 0.40 2 0.00 0.67
L Frontal pole 214 (37) 211 (38) 207 (31) 222 (41) 1.10 2 0.02 0.34
L Temporal pole 485 (69) 484 (71) 470 (69) 500 (66) 0.86 2 0.02 0.43
L Transverse temporal 454 (83) 454 (90) 428 (87) 475 (68) 1.01 2 0.02 0.37
L Insula 2094 (298) 2088 (257) 1986 (229) 2193 (355) 2.25 2 0.03 0.11
R Banks of the superior temporal sulcus 932 (187) 928 (186) 882 (183) 980 (186) 1.19 2 0.02 0.31
R Caudal anterior cingulate 732 (160) 746 (168) 675 (109) 769 (178) 0.72 2 0.01 0.49
R Caudal middle frontal 2070 (360) 2028 (396) 2022 (333) 2150 (345) 1.43 2 0.02 0.25
R Cuneus 1452 (208) 1462 (207) 1405 (213) 1484 (204) 0.32 2 0.00 0.73
R Entorhinal 322 (90) 302 (73) 310 (69) 350 (114) 2.70 2 0.05 0.07
R Fusiform 3047 (549) 3158 (542) 2901 (523) 3069 (570) 0.13 2 0.00 0.88
R Inferior parietal 5255 (887) 5359 (951) 5052 (833) 5333 (874) 0.05 2 0.00 0.95
R Inferior temporal 2917 (534) 2857 (585) 2886 (490) 3001 (530) 3.57 2 0.04 0.03 *
R Isthmus cingulate 957 (172) 977 (151) 908 (162) 981 (193) 0.05 2 0.00 0.95
R Lateral occipital 4556 (748) 4518 (575) 4264 (727) 4840 (822) 4.09 2 0.05 0.02 *
R Lateral orbital frontal 2520 (353) 2550 (356) 2380 (366) 2612 (312) 1.61 2 0.02 0.21
R Lingual 2937 (421) 2962 (396) 2776 (432) 3052 (404) 1.55 2 0.03 0.22
R Medial orbital frontal 1833 (284) 1842 (287) 1739 (254) 1906 (292) 1.26 2 0.02 0.29
R Middle temporal 3236 (540) 3261 (566) 3110 (506) 3319 (543) 0.86 2 0.01 0.43
R Parahippocampal 661 (121) 692 (117) 625 (110) 663 (128) 0.55 2 0.01 0.58
R Paracentral 1573 (272) 1574 (313) 1544 (226) 1596 (274) 0.46 2 0.01 0.64
R Pars opercularis 1326 (240) 1271 (202) 1272 (220) 1423 (264) 5.18 2 0.08 0.01 *
R Pars orbitalis 761 (120) 783 (131) 714 (101) 782 (118) 0.93 2 0.02 0.40
R Pars triangularis 1440 (255) 1482 (257) 1376 (210) 1456 (284) 0.27 2 0.01 0.76
R Pericalcarine 1425 (278) 1441 (271) 1383 (282) 1447 (288) 0.04 2 0.00 0.96
R Postcentral gyrus 3882 (557) 3933 (612) 3772 (566) 3928 (499) 0.54 2 0.01 0.58
R Posterior cingulate 1154 (196) 1184 (193) 1121 (195) 1153 (201) 0.49 2 0.01 0.61
R Precentral 4832 (677) 4822 (708) 4576 (541) 5062 (694) 3.12 2 0.04 0.05
R Precuneus 3872 (572) 3965 (573) 3663 (575) 3964 (539) 0.37 2 0.00 0.69
R Rostral anterior cingulate 644 (145) 655 (157) 614 (135) 659 (144) 0.08 2 0.00 0.92
R Rostral middle frontal 5792 (870) 5959 (775) 5455 (916) 5925 (860) 0.42 2 0.00 0.66
R Superior frontal 6932 (1015) 6918 (936) 6599 (881) 7230 (1130) 2.86 2 0.03 0.06
R Superior parietal 5256 (665) 5332 (712) 4995 (661) 5408 (576) 0.95 2 0.01 0.39
R Superior temporal 3552 (520) 3612 (570) 3356 (447) 3664 (501) 0.83 2 0.01 0.44
R Supramarginal 3587 (616) 3531 (600) 3468 (604) 3741 (631) 2.72 2 0.03 0.07
R Frontal pole 287 (45) 282 (41) 285 (46) 292 (49) 1.00 2 0.02 0.37
R Temporal pole 425 (54) 421 (62) 419 (42) 434 (56) 0.71 2 0.01 0.49
R transverse temporal 345 (69) 351 (72) 323 (58) 359 (71) 0.44 2 0.01 0.65
R insula 2119 (329) 2144 (360) 2039 (288) 2164 (330) 0.16 2 0.00 0.85
Open in a new tab

3. Results

3.1. Subject characteristics

As we only analysed data from the first visit of patients with a confirmed diagnosis of schizophrenia we had 151 subjects. The mean age of these was 34.68 (sd = 12.72) and most were male (65.6%). Some were reportedly undergoing treatment, however detailed information on this was not made available to us. The sample was mainly Caucasian (51.7%), followed by African American (46.4%). The majority were unemployed at the time of assessment (69.9%).

3.2. Clustering

To decide how many clusters to use in the analysis, we used the silhouette method, which determines how well each observation lies within its cluster. It is a measurement of how closely the observation matches data within the cluster compared to other clusters. This method showed that the optimal number of clusters for the data was three (Figure 1). This first step was performed using the z-scored positive, negative and disorganized symptomatology dimensions from SANS/SAPS, along with the z-scored crystallised intelligence. After deciding the number of centres, we performed clustering in the data set using the same variables to train the algorithm, and 1000 starting random positions in the k-means algorithm. The resulting relationships can be seen in a scatterplot (Figure 2).

Figure 1.

Figure 1

Open in a new tab

The silhouette method determines how well each observation lies within its cluster. It is a measurement of how closely the observation matches data within the cluster compared to other clusters. This method showed that the optimal number of clusters for the data was three.

Figure 2.

Figure 2

Open in a new tab

Scatterplots between the z-scored positive, negative and disorganized domains; and the z-scored crystallised intelligence (IQ) used in the clustering method. The top right scatterplot shows the uniqueness of the three clusters clearly, note that since the z-scored data was used (these are not raw test scores) and values range from negative to positive.

The three groups appear to represent subjects who have either high positive symptomatology (n = 49), low positive symptomatology (n = 47), or high cognitive performance (n = 55), referred to as high positive, low positive and performance hereafter (Figure 2). The groups did not differ on age, sex, ethnicity, number of siblings and father schooling level (Table 1).

Table 1.

Mean (Standard deviation) and differences for demographic variables. P-values were calculated between the three categories (the overall column was not used) with chi-square contingence table test for the categorical variables and one-way ANOVA for the continuous variables.

Overall low-positive high-positive performance p
n 151 47 49 55
age 34.68 (12.72) 34.98 (12.59) 33.22 (12.86) 35.71 (12.82) 0.601
sex = male (%) 99 (65.6) 32 (68.1) 30 (61.2) 37 (67.3) 0.736
handedness (%) 0.492
4 (2.6) 1 (2.1) 0 (0.0) 3 (5.5)
left 13 (8.6) 5 (10.6) 4 (8.2) 4 (7.3)
right 134 (88.7) 41 (87.2) 45 (91.8) 48 (87.3)
race (%) <0.001
Caucasian 78 (51.7) 32 (68.1) 31 (63.3) 15 (27.3)
African American 70 (46.4) 14 (29.8) 17 (34.7) 39 (70.9)
Hispanic 1 (0.7) 0 (0.0) 1 (2.0) 0 (0.0)
Native American 2 (1.3) 1 (2.1) 0 (0.0) 1 (1.8)
ethnicity (%) 0.778
NA 1 (0.7) 0 (0.0) 0 (0.0) 1 (1.8)
Hispanic 3 (2.0) 1 (2.1) 1 (2.0) 1 (1.8)
Non-Hispanic 147 (97.4) 46 (97.9) 48 (98.0) 53 (96.4)
Marital Status (%)
Other 3 (2.0) 0 (0.0) 1 (2.0) 2 (3.7)
Single 114 (76.0) 38 (80.9) 37 (75.5) 39 (72.2)
Married/common law 6 (4.0) 4 (8.5) 1 (2.0) 1 (1.9)
Divorced 21 (14.0) 5 (10.6) 7 (14.3) 9 (16.7)
Separated 2 (1.3) 0 (0.0) 1 (2.0) 1 (1.9)
Widowed 0 (0.0) 0 (0.0) 0 (0.0) 0 (0.0)
Unknown 4 (2.7) 0 (0.0) 2 (4.1) 2 (3.7)
Siblings Number 3.18 (2.52) 3.81 (2.65) 3.13 (2.17) 2.65 (2.60) 0.073
Number Of Children 0.73 (1.53) 1.04 (2.07) 0.79 (1.30) 0.39 (1.02) 0.105
Employment Status (%)
Other 5 (3.5) 2 (4.3) 1 (2.2) 2 (3.9)
Employed full-time 9 (6.3) 4 (8.7) 3 (6.5) 2 (3.9)
Employed part-time 17 (11.9) 4 (8.7) 6 (13.0) 7 (13.7)
Unemployed 100 (69.9) 34 (73.9) 34 (73.9) 32 (62.7)
Homemaker full-time 0 (0.0) 0 (0.0) 0 (0.0) 0 (0.0)
Student full-time 6 (4.2) 0 (0.0) 1 (2.2) 5 (9.8)
Student part-time 6 (4.2) 2 (4.3) 1 (2.2) 3 (5.9)
Years Of Schooling 12.21 (2.40) 11.41 (1.71) 11.43 (2.31) 13.58 (2.43) <0.001
Schooling Level Father 12.74 (3.38) 12.31 (2.29) 12.43 (2.64) 13.20 (4.24) 0.475
Schooling Level Mother 12.52 (3.12) 11.70 (2.96) 12.06 (3.46) 13.31 (2.85) 0.053
Open in a new tab

3.3. ANCOVAS and post hoc tests

ANCOVAS were performed with group as a factor and brain volumes as the dependent variables (Table 2). We found that six regions significantly differed between groups (Tables 3, 4, 5, 6, 7, 8): left caudate (F = 3.51, df = 2, η2 =0.06, p= 0.03), left cuneus (F= 4.25, df= 2, η2 = 0.07, p= 0.02), left lateral occipital (F = 3.19, df = 2, η2 = 0.04, p= 0.05), left inferior temporal (F = 3.57, df= 2, η2 = 0.04, p= 0.03), right lateral occipital (F=4.09, df= 2, η2 =0.05, p = 0.02), and right pars opercularis (F = 5.18, df= 2, η2 = 0.08, p = 0.01).

Table 3. Bonferroni corrected post hoc test for left caudate.

Cluster 1 Cluster 2 md se df t pbonferroni
low-positive - high-positive 114 107 73 1.1 0.86
low-positive - performance -160 101 73 -1.6 0.35
high-positive - performance -274 105 73 -2.6 0.03
Open in a new tab

Table 4. Bonferroni corrected post hoc test for left cuneus.

Cluster 1 Cluster 2 md se df t pbonferroni
low-positive - high-positive -140 62 73 -2.27 0.08
low-positive - performance -157 58 73 -2.71 0.03
high-positive - performance -17 60 73 -0.28 1.00
Open in a new tab

Table 5. Bonferroni corrected post hoc test for left lateral occipital.

Cluster 1 Cluster 2 md se df t pbonferroni
low-positive - high-positive -159 150 73 -1.1 0.88
low-positive - performance -354 141 73 -2.5 0.04
high-positive - performance -195 147 73 -1.3 0.56
Open in a new tab

Table 6. Bonferroni corrected post hoc test for right lateral occipital.

Cluster 1 Cluster 2 md se df t pbonferroni
low-positive - high-positive -259 101 73 -2.57 0.04
low-positive - performance -181 95 73 -1.91 0.18
high-positive - performance 78 99 73 0.79 1.00
Open in a new tab

Table 7. Bonferroni corrected post hoc test for right inferior temporal.

Cluster 1 Cluster 2 md se df t pbonferroni
low-positive - high-positive -34 149 73 -0.23 1.00
low-positive - performance -362 140 73 -2.60 0.03
high-positive - performance -329 146 73 -2.26 0.08
Open in a new tab

Table 8. Bonferroni corrected post hoc test for right pars-opercularis.

Cluster 1 Cluster 2 md se df t pbonferroni
low-positive - high-positive -82 54 73 -1.5 0.40
low-positive - performance -163 51 73 -3.2 0.01
high-positive - performance -81 53 73 -1.5 0.38
Open in a new tab

Bonferroni-corrected post hoc tests were run in order to inspect inter-group relationships. For the left caudate, there was a significant difference between the group with high positive symptomatology and the high performance group (se = 105, df = 73, t = -2.6, p = 0.03, Bonferroni corrected) (Table 3). The high positive symptomatology presented with smaller volumes than the performance group.

The left cuneus volumes had significant difference between groups (se = 58, df = 73, t = -2.71, p = 0.03, Bonferroni corrected) (Table 4). The low positive symptomatology group had smaller volumes than the performance group.

The low positive symptomatology group differed significantly from the high performance group (se = 147, df = 141, t = -2.5, p = 0.04, Bonferroni corrected) in relation to left lateral occipital (Table 5). Again, the low positive symptomatology group presented with smaller volume than the performance group.

For the right lateral occipital there were statistically significant differences between groups (se = 101, df = 73, t = -2.57, p = 0.04, Bonferroni corrected) (Table 6). The low positive symptomatology group had larger volumes than the positive symptomatology group.

In our tests the right inferior temporal region differed significantly between groups. The low positive symptomatology had smaller volumes than the performance group (se = 140, df = 73, t = -2.60, p = 0.03, Bonferroni corrected) (Table 7).

Finally, the right pars-opercularis was a region, which differed significantly between groups. The low positive symptomatology presented with smaller volumes than the performance group (se = 51, df = 73, t = -3.2, p = 0.01, Bonferroni corrected) (Table 8)

All our findings were of small effect sizes for each comparison (η2). The six significant areas were left caudate (F = 3.51, df = 2, η2 =0.06, p= 0.03), left cuneus (F= 4.25, df= 2, η2 = 0.07, p= 0.02), left lateral occipital (F = 3.19, df = 2, η2 = 0.04, p= 0.05), left inferior temporal (F = 3.57, df= 2, η2 = 0.04, p= 0.03), right lateral occipital (F=4.09, df= 2, η2 =0.05, p = 0.02), and right pars opercularis (F = 5.18, df= 2, η2 = 0.08, p = 0.01).

4. Discussion

We used a data-driven approach to classify subjects with schizophrenia in this analysis. The method creates clusters based on the properties of each subject. The properties used in the algorithm were cognitive performance and positive, negative and disorganized symptoms. This optimally divided into three clusters, which we then included in ANCOVAs as factors. Comparisons were run with brain regions volumes as dependent variables, controlling for age and for the estimated intracranial volume. We found six significant associations in the ANCOVAs, and then proceeded to investigate significance between the groups in a second step, Bonferroni-corrected post hoc tests. Effect sizes were calculated for each comparison.

It is likely the clusters identified in this study will overlap with previous subtypologies. Our high-performance group might be expected to align with the ‘good outcome’ groups long identified – those whose symptoms remit, rather then pursue a chronic course. Our other clusters found in this study are not clearly the “deficit” and “distress” subtypes from previous publications (Dickinson et al., 2018). Instead, as opposed to our high-performance group we found our deficit group was optimally separated by the “elbow” method into two distinct clusters, our high and low positive symptom groups. Recent attempts to find subgroups in schizophrenic subjects either used principal component analysis (Clementz et al., 2015) or the 2-step clustering from SPSS (Dickinson et al., 2018), but both of these arrived at a three-cluster solution. Dickinson et al. (2018) addressed affective symptomatology and found a high distress subgroup in their analysis. Our study did not evaluate affective symptoms, however we included clinical severity of psychotic symptoms. Despite differences on the indicator used, our tests suggested a three-cluster scheme as well. The deficit subtype has been the earliest (Ahmed et al., 2015; B Kirkpatrick and R W Buchanan, 1990) and most thoroughly analysed (Cohen et al., 2010), however we show that there is another dimension related to the severity of positive or negative symptomatology.

It now appears certain that multiple elements of small effect contribute to the schizophrenia syndrome (Castro-de-Araujo et al., 2018; Castro-de-Araujo and Kanaan, 2017), and, as expected, all the volumetric effect sizes were small. But the volumetric differences found in this paper corroborate previous findings. These volumes, except for the pars opercularis and lateral occipital gyrus, are reported to be abnormal in schizophrenia by meta-analysis (Honea et al., 2005). Furthermore, they may be critical to subtype differentiation, as Weinberg et al. (2016) found pars opercularis and inferior temporal regions to differ significantly between their three cognitive subtypes (preserved, moderately deteriorated and severely deteriorated). All post hoc Bonferroni corrected tests revealed at least one statistically significant inter-group difference. In all cases, the performance subgroup had larger volumes than the low positive and high positive groups. This is in line with what is known regarding volumetric alterations and cognitive performance (Weinberg et al., 2016). The k-means approach resulted in a high to low positive symptomatology distinction, along with the performance cluster. This should be related to the fact that negative symptomatology is strongly associated with “deficit” subtypes, which may have caused the algorithm not to identify it as unique (Ahmed et al., 2015). We interpret this as negative symptomatology not helping further reduce heterogeneity, as it should match cases with general low cognitive performance.

Traditional factor analytical methods have been used to investigate whether schizophrenic psychopathology sub-grouping presented with corresponding cognitive and cerebral changes. These supported schizophrenic sub-types: disorganized, psychotic and negative (Andreasen and Olsen, 1982; Andreasen et al., 1995; Liddle, 2019) and a “deficit” subtype (Dickinson et al. 2018). The present work presented a method that allowed the incorporation of cognitive performance in addition to positive and negative symptomatology in the clustering of the subjects, which is much less trivial in factor analyses. This approach reduces the dimensionality, while taking a larger group of factors into account. A recent study compared the performance of factor analytical and clustering methods in subjects with obsessive-compulsive disorder, showing that clustering arrived at similar results to the more traditional factor analyses (Cameron et al., 2019).

Some limitations of the study should be noted. Not all volumes had a normal distribution, therefore ANCOVA’s assumptions of normality were partially violated. Since we used a publicly available data set, the demographic information was limited due to privacy. Especially concerning was the absence of information regarding treatment history, since we could not therefore control for the chronicity and duration of the disease, which might have influenced the clustering algorithm. Also, due to the cross-sectional nature of our approach, the stability of these groups over time could not be assessed. However, as we used a publicly available data set, our results should be easy to reproduce.

K-means clustering is an easily available method for clustering, which can potentially facilitate subgrouping disorders in an unsupervised way. The heterogeneity of the symptomatology, course, and associated factors in schizophrenia makes subgrouping the condition a difficult task. There is already evidence that cognitive performance seems to help identify a clinically relevant group, and we show in this paper how grouping can be done using symptom severity in addition. We believe that k-means clustering, which is neutral to the type of indicator used, may help identify subgroups that are biologically plausible.

Footnotes

Contributors

Author CASTRO-DE-ARAUJO designed the study and the statistical analysis. Authors MACHADO and BARRETO helped in writing the manuscript. Author KANAAN provided supervision and helped in writing the manuscript. All authors contributed to and have approved the final manuscript.

Conflict of Interest

Authors report no conflicts of interest.

References

  1. Ahmed AO, Strauss GP, Buchanan RW, Kirkpatrick B, Carpenter WT. Are Negative Symptoms Dimensional or Categorical? Detection and Validation of Deficit Schizophrenia With Taxometric and Latent Variable Mixture Models. Schizophr Bull. 2015;41:879–891. doi: 10.1093/schbul/sbu163. [DOI] [PMC free article] [PubMed] [Google Scholar]
  2. al-Haddad BJS, Jacobsson B, Chabra S, Modzelewska D, Olson EM, Bernier R, Enquobahrie DA, Hagberg H, Östling S, Rajagopal L, Waldorf KMA, et al. Long-term Risk of Neuropsychiatric Disease After Exposure to Infection In Utero. JAMA Psychiatry. 2019 doi: 10.1001/jamapsychiatry.2019.0029. [DOI] [PMC free article] [PubMed] [Google Scholar]
  3. Ambite JL, Tallis M, Alpert K, Keator DB, King M, Landis D, Konstantinidis G, Calhoun VD, Potkin SG, Turner JA, Wang L. SchizConnect: Virtual Data Integration in Neuroimaging. Data Integr Life Sci. 2015;9162:37–51. doi: 10.1007/978-3-319-21843-4_4. [DOI] [PMC free article] [PubMed] [Google Scholar]
  4. Andreasen NC. The Scale for the Assessment of Positive Symptoms (SAPS) The University of Iowa; Iowa city, Iowa: 1984. [Google Scholar]
  5. Andreasen NC. The Scale for the Assessment of Negative Symptoms (SANS) The University of Iowa; Iowa city, Iowa: 1983. [Google Scholar]
  6. Andreasen NC, Arndt S, Alliger R, Miller D, Flaum M. Symptoms of Schizophrenia: Methods, Meanings, and Mechanisms. Arch Gen Psychiatry. 1995;52:341–351. doi: 10.1001/archpsyc.1995.03950170015003. [DOI] [PubMed] [Google Scholar]
  7. Andreasen NC, Olsen S. Negative v Positive Schizophrenia: Definition and Validation. Arch Gen Psychiatry. 1982;39:789–794. doi: 10.1001/archpsyc.1982.04290070025006. [DOI] [PubMed] [Google Scholar]
  8. Braver TS, Cohen JD, Nystrom LE, Jonides J, Smith EE, Noll DC. A Parametric Study of Prefrontal Cortex Involvement in Human Working Memory. NeuroImage. 1997;5:49–62. doi: 10.1006/nimg.1996.0247. [DOI] [PubMed] [Google Scholar]
  9. Cameron DH, Streiner DL, Summerfeldt LJ, Rowa K, McKinnon MC, McCabe RE. A comparison of cluster and factor analytic techniques for identifying symptom-based dimensions of obsessive-compulsive disorder. Psychiatry Research. 2019;278:86–96. doi: 10.1016/j.psychres.2019.05.040. [DOI] [PubMed] [Google Scholar]
  10. Castro-de-Araujo LFS, Allin M, Picchioni MM, Mcdonald C, Pantelis C, Kanaan RAA. Schizophrenia moderates the relationship between white matter integrity and cognition. Schizophrenia Research. 2018 doi: 10.1016/j.schres.2018.03.033. [DOI] [PMC free article] [PubMed] [Google Scholar]
  11. Castro-de-Araujo LFS, Kanaan RA. First episode psychosis moderates the effect of gray matter volume on cognition. Psychiatry Research: Neuroimaging. 2017;266:108–113. doi: 10.1016/j.pscychresns.2017.06.007. [DOI] [PubMed] [Google Scholar]
  12. Clementz BA, Sweeney JA, Hamm JP, Ivleva EI, Ethridge LE, Pearlson GD, Keshavan MS, Tamminga CA. Identification of Distinct Psychosis Biotypes Using Brain-Based Biomarkers. AJP. 2015;173:373–384. doi: 10.1176/appi.ajp.2015.14091200. [DOI] [PMC free article] [PubMed] [Google Scholar]
  13. Cohen AS, Brown LA, Minor KS. The psychiatric symptomatology of deficit schizophrenia: A meta-analysis. Schizophrenia Research. 2010;118:122–127. doi: 10.1016/j.schres.2009.10.010. [DOI] [PubMed] [Google Scholar]
  14. Corcoran CM, Smith C, McLaughlin D, Auther A, Malaspina D, Cornblatt B. HPA axis function and symptoms in adolescents at clinical high risk for schizophrenia. Schizophr Res. 2012;135:170–174. doi: 10.1016/j.schres.2011.11.035. [DOI] [PMC free article] [PubMed] [Google Scholar]
  15. Csernansky JG, Wang L, Jones D, Rastogi-Cruz D, Posener JA, Heydebrand G, Miller JP, Miller MI. Hippocampal Deformities in Schizophrenia Characterized by High Dimensional Brain Mapping. AJP. 2002;159:2000–2006. doi: 10.1176/appi.ajp.159.12.2000. [DOI] [PubMed] [Google Scholar]
  16. Dale AM, Fischl B, Sereno MI. Cortical surface-based analysis. I. Segmentation and surface reconstruction. NeuroImage. 1999;9:179–194. doi: 10.1006/nimg.1998.0395. [DOI] [PubMed] [Google Scholar]
  17. Del-Ben CM, Shuhama R, Loureiro CM, Ragazzi TCC, Zanatta DP, Tenan SHG, Ferreira Santos JL, Louzada-Junior P, Dos Santos AC, Morgan C, Menezes PR. Urbanicity and risk of first-episode psychosis: Incidence study in Brazil. Br J Psychiatry. 2019:1–4. doi: 10.1192/bjp.2019.110. [DOI] [PubMed] [Google Scholar]
  18. Desikan RS, Ségonne F, Fischl B, Quinn BT, Dickerson BC, Blacker D, Buckner RL, Dale AM, Maguire RP, Hyman BT, Albert MS, et al. An automated labeling system for subdividing the human cerebral cortex on MRI scans into gyral based regions of interest. NeuroImage. 2006;31:968–980. doi: 10.1016/j.neuroimage.2006.01.021. [DOI] [PubMed] [Google Scholar]
  19. Dickinson D, Pratt DN, Giangrande EJ, Grunnagle M, Orel J, Weinberger DR, Callicott JH, Berman KF. Attacking Heterogeneity in Schizophrenia by Deriving Clinical Subgroups From Widely Available Symptom Data. Schizophrenia Bulletin. 2018;44:101–113. doi: 10.1093/schbul/sbx039. [DOI] [PMC free article] [PubMed] [Google Scholar]
  20. Fearon P, Kirkbride JB, Morgan C, Dazzan P, Morgan K, Lloyd T, Hutchinson G, Tarrant J, Fung WLA, Holloway J, Mallett R, et al. Incidence of schizophrenia and other psychoses in ethnic minority groups: Results from the MRC AESOP Study. Psychological Medicine. 2006;36:1541–1550. doi: 10.1017/S0033291706008774. [DOI] [PubMed] [Google Scholar]
  21. First MB, Spitzer RL, Gibbon M, Williams J. Structured Clinical Interview for DSM-IV Axis I Disorders, Patient Edition (SCID-P), Version. 2 ed New York State Psychiatric Institute; New York, New York, USA: 1995. [Google Scholar]
  22. First MB, Spitzer RL, Gibbon M, Williams JBW. Structured Clinical Interview for DSM-IV Axis I Disorders, Clinician Version (SCID-CV) 1997 [Google Scholar]
  23. Fischl B, Dale AM. Measuring the thickness of the human cerebral cortex from magnetic resonance images. Proceedings of the National Academy of Sciences of the United States of America. 2000;97:11050–11055. doi: 10.1073/pnas.200033797. [DOI] [PMC free article] [PubMed] [Google Scholar]
  24. Fischl B, Van Der Kouwe A, Destrieux C, Halgren E, Ségonne F, Salat DH, Busa E, Seidman LJ, Goldstein J, Kennedy D, Caviness V, et al. Automatically Parcellating the Human Cerebral Cortex. Cerebral Cortex. 2004;14:11–22. doi: 10.1093/cercor/bhg087. [DOI] [PubMed] [Google Scholar]
  25. Fusar-Poli P, Politi P. Paul Eugen Bleuler and the birth of schizophrenia (1908) American Journal of Psychiatry. 2008;165:1407. doi: 10.1176/appi.ajp.2008.08050714. [DOI] [PubMed] [Google Scholar]
  26. Fusar-Poli P. The Clinical High-Risk State for Psychosis (CHR-P), Version II. Schizophr Bull. 2017;43:44–47. doi: 10.1093/schbul/sbw158. [DOI] [PMC free article] [PubMed] [Google Scholar]
  27. Fusar-Poli P, Borgwardt S, Bechdolf A, Addington J, Riecher-Rössler A, Schultze-Lutter F, Keshavan M, Wood S, Ruhrmann S, Seidman LJ, Valmaggia L, et al. The Psychosis High-Risk State. JAMA Psychiatry. 2013;70:107–120. doi: 10.1001/jamapsychiatry.2013.269. [DOI] [PMC free article] [PubMed] [Google Scholar]
  28. Hazlett EA, Buchsbaum MS, Jeu LA, Nenadic I, Fleischman MB, Shihabuddin L, Haznedar MM, Harvey PD. Hypofrontality in unmedicated schizophrenia patients studied with PET during performance of a serial verbal learning task. Schizophrenia Research. 2000;43:33–46. doi: 10.1016/S0920-9964(99)00178-4. [DOI] [PubMed] [Google Scholar]
  29. Hazlett EA, Vaccaro DH, Haznedar MM, Goldstein KE. Reprint of: F-18Fluorodeoxyglucose positron emission tomography studies of the schizophrenia spectrum: The legacy of Monte S. Buchsbaum, M.D. Psychiatry Res. 2019;277:39–44. doi: 10.1016/j.psychres.2019.06.014. [DOI] [PubMed] [Google Scholar]
  30. Hazlett EA, Buchsbaum MS. Sensorimotor gating deficits and hypofrontality in schizophrenia. Front Biosci. 2001;6:D1069–1072. doi: 10.2741/hazlett. [DOI] [PubMed] [Google Scholar]
  31. Harms MP, Wang L, Mamah D, Barch DM, Thompson PA, Csernansky JG. Thalamic Shape Abnormalities in Individuals with Schizophrenia and Their Nonpsychotic Siblings. J Neurosci. 2007;27:13835–13842. doi: 10.1523/JNEUROSCI.2571-07.2007. [DOI] [PMC free article] [PubMed] [Google Scholar]
  32. Heaton RK, Chelune GJ, Talley JL, Kay GG, Curtiss G. Wisconsin Card Sorting Test Manual. Psychological Assessment Resources; Odessa: 1993. [Google Scholar]
  33. Honea R, Crow TJ, Passingham D, Mackay CE. Regional deficits in brain volume in schizophrenia: A meta-analysis of voxel-based morphometry studies. American Journal of Psychiatry. 2005;162:2233–2245. doi: 10.1176/appi.ajp.162.12.2233. [DOI] [PubMed] [Google Scholar]
  34. Horan WP, Blanchard JJ, Clark LA, Green MF. Affective Traits in Schizophrenia and Schizotypy. Schizophr Bull. 2008;34:856–874. doi: 10.1093/schbul/sbn083. [DOI] [PMC free article] [PubMed] [Google Scholar]
  35. Jablensky A. The diagnostic concept of schizophrenia: its history, evolution, and future prospects. Dialogues Clin Neurosci. 2010;12:271–287. doi: 10.31887/DCNS.2010.12.3/ajablensky. [DOI] [PMC free article] [PubMed] [Google Scholar]
  36. Kanchanatawan B, Sriswasdi S, Thika S, Sirivichayakul S, Carvalho AF, Geffard M, Kubera M, Maes M. Deficit schizophrenia is a discrete diagnostic category defined by neuro-immune and neurocognitive features: Results of supervised machine learning. Metab Brain Dis. 2018;33:1053–1067. doi: 10.1007/s11011-018-0208-4. [DOI] [PubMed] [Google Scholar]
  37. Kendler KS, Engstrom EJ. Kahlbaum, Hecker, and Kraepelin and the Transition From Psychiatric Symptom Complexes to Empirical Disease Forms. American Journal of Psychiatry. 2016 doi: 10.1176/appi.ajp.2016.16030375. appi ajp 2016 1-appi ajp 2016 1. [DOI] [PubMed] [Google Scholar]
  38. Kirkpatrick B, Buchanan RW. Anhedonia and the deficit syndrome of schizophrenia. Psychiatry Research. 1990;31:25–30. doi: 10.1016/0165-1781(90)90105-E. [DOI] [PubMed] [Google Scholar]
  39. Kirkpatrick B, Buchanan RW. The neural basis of the deficit syndrome of schizophrenia. The Journal Of Nervous And Mental Disease. 1990;178:545–555. doi: 10.1097/00005053-199009000-00001. [DOI] [PubMed] [Google Scholar]
  40. Liddle PF. The Core Deficit of Classical Schizophrenia: Implications for Predicting the Functional Outcome of Psychotic Illness and Developing Effective Treatments. Can J Psychiatry. 2019;64:680–685. doi: 10.1177/0706743719870515. [DOI] [PMC free article] [PubMed] [Google Scholar]
  41. Liddle PF. The symptoms of chronic schizophrenia. A re-examination of the positive-negative dichotomy. Br J Psychiatry. 1987;151:145–151. doi: 10.1192/bjp.151.2.145. [DOI] [PubMed] [Google Scholar]
  42. Lönnqvist J-E, Verkasalo M, Haukka J, Nyman K, Tiihonen J, Laaksonen I, Leskinen J, Lönnqvist J, Henriksson M. Premorbid personality factors in schizophrenia and bipolar disorder: Results from a large cohort study of male conscripts. J Abnorm Psychol. 2009;118:418–423. doi: 10.1037/a0015127. [DOI] [PubMed] [Google Scholar]
  43. Ripke S, Neale BM, Corvin A, Walters JTR, Farh K-H, Holmans Pa, Lee P, Bulik-Sullivan B, Collier Da, Huang H, Pers TH. Biological insights from 108 schizophrenia-associated genetic loci. Nature. 2014;511:421–427. doi: 10.1038/nature13595. [DOI] [PMC free article] [PubMed] [Google Scholar]
  44. Selker R, Love J, Dropmann D. Jmv: The ’jamovi’ Analyses. 2019 [Google Scholar]
  45. Schröder J, Buchsbaum MS, Siegel BV, Geider FJ, Niethammer R. Structural and functional correlates of subsyndromes in chronic schizophrenia. Psychopathology. 1995;28:38–45. doi: 10.1159/000284898. [DOI] [PubMed] [Google Scholar]
  46. Tandon R, Keshavan MS, Nasrallah Ha. Schizophrenia, “just the facts” what we know in 2008 2. Epidemiology and etiology. Schizophrenia Research. 2008;102:1–18. doi: 10.1016/j.schres.2008.04.011. [DOI] [PubMed] [Google Scholar]
  47. Wang L, Kogan A, Cobia D, Alpert K, Kolasny A, Miller MI, Marcus D. Northwestern University Schizophrenia Data and Software Tool (NUSDAST) Frontiers in Neuroinformatics. 2013;7:1–13. doi: 10.3389/fninf.2013.00025. [DOI] [PMC free article] [PubMed] [Google Scholar]
  48. Wechsler D. Wechsler memory scale. 3rd ed. WMS-III; San Antonio, TX: 1995. [Google Scholar]
  49. Weinberg D, Lenroot R, Jacomb I, Allen K, Bruggemann J, Wells R, Balzan R, Liu D, Galletly C, Catts SV, Weickert CS, et al. Cognitive Subtypes of Schizophrenia Characterized by Differential Brain Volumetric Reductions and Cognitive Decline. JAMA Psychiatry. 2016;73:1251. doi: 10.1001/jamapsychiatry.2016.2925. [DOI] [PubMed] [Google Scholar]

RESOURCES