Concomitant factor analyses on PANSS and SAPS/SANS items revealed a series of factors with shared positively loaded items across scales, separately in the positive and negative domains. Three positive factors correspond to primary criteria for the diagnostic of schizophrenia in the DSM-5 [21], namely hallucinations, delusions and disorganization. Two negative factors relate to the description of diminished emotional expression and amotivation/avolition as the two most prominent negative symptoms in this prevailing diagnostic manual. A last so-called cognition factor was isolated in the cross-scale factor structure of negative symptoms. However, as discussed hereafter, the validity of this factor is however much questionable and the conversion between scales for this factor was inaccurate and unreliable. This is in sharp contrast with the five other positive or negative factors for which between-scale symptom rating conversion could be performed accurately and reliably.
Subdomains of positive and negative symptomatology
Our cross-scale decomposition of the PANSS and SAPS/SANS into distinct positive and negative symptom subdomains can be comprehended in light of qualitative comparison of the two scales [1] as well as numerous past factor analyses conducted on either scale [22, 23]. In contrast to prior studies, it should be noted that our factor analyses investigated both scales concomitantly rather than separately, considered positive and negative symptoms separately, relied on global SAPS or SANS scores rather than on individual items, and excluded items from the PANSS general psychopathology subscale.
The distinction between reality distortion (combining hallucinations and delusions) and disorganization within positive symptoms has received much support from the triadic syndrome model [10, 11, 24], for instance underlying the SAPS/SANS. This dissociation is also found in rating scales with a more comprehensive list of symptoms such as the PANSS, which is however often described through a 5-factor model that includes excitement/mania and depression dimensions in addition to reality distortion, disorganization and negative dimensions when including items from the general psychopathology subscale [22]. Of note, the hostility item loads highly on the excitement/mania factor in PANSS 5-factor models, rather than being included in a delusion (reality distortion) factor as reported in the present findings. While factor analysis on either the SAPS/SANS or PANSS does not typically distinguish hallucinations and delusions within reality distortion as found in the present work, it is worth noting that such has been the case using the Schedule for Clinical Assessment in Neuropsychiatry (SCAN) [25, 26]. This symptom dimension separation is also supported by the DSM-5 dimensional model, allowing better correspondence with categorical diagnoses and clinical treatment targets [27].
Consistent with our results, ample evidence supports the delineation of diminished expressivity (poverty of speech and affective flattening) and amotivation (avolition/apathy and asociality) as two key underlying subdomains of negative symptoms [12, 13, 28]. While the DSM-5 dimensional model only includes a single negative symptom item for pragmatic reasons, it does describe the negative symptom dimension as somewhat of a hybrid, referring to deficits in either of the two subdomains [27].
The third factor we identified based on negative symptoms alone included cognition-related items, namely attention in the SANS and abstract thinking in the PANSS. Cognitive impairment is now widely recognized as a core feature of schizophrenia [29, 30], as acknowledged in the DSM-5 dimensional model [27]. Yet, past factor analysis on either the SAPS/SANS or PANSS did not uncover a purely cognitive dimension. Rather, poor attention and difficulty in abstract thinking are often part of a disorganization factor [10, 22]. Accordingly, the inclusion of attentional impairment as a negative symptom in the SANS has been questioned [31]. Besides, it should be noted that the attention item in the PANSS is part of the general psychopathology subscale, and it was therefore not included in our factor analysis. More generally, cognitive assessment based on symptom rating scales such as the SAPS/SANS or PANSS is likely imperfect, and instead requires formal neuropsychological testing [29]. Moreover, attention and abstract thinking point to only two of the many nonsocial and social cognitive domains with compromised functioning in schizophrenia [29, 30]. For all these reasons, we conclude that the so-called cognition factor evidenced here has poor face validity across scales and should therefore be considered with caution.
Accuracy and reliability of conversion equations
Regression-based equations to quantitatively convert symptom ratings between distinct clinical scales have been successfully derived not only in schizophrenia [7, 16] but also in other mental illnesses such as major depression [32, 33]. Simple linear regression models adequately described the relationship between scales, with only trivial gains induced by more complex functions [32]. In keeping with these findings, we assessed conversion reliability by contrasting original and predicted scores for a subset of data independent from data used to derive regression equations. We further evaluated how variable reliability measures were by conducting similar regression analyses multiple times using different permutations of training and test data, thereby allowing to report representative conversion equations.
Using a slightly modified approach, we first replicated previous results showing that overall indices of positive and negative symptomatology can be accurately and reliably converted between the PANSS and SAPS/SANS using the same dataset [7]. Our main contribution was to show that the severity of various subdomains of positive and negative symptoms can be converted between those rating scales as accurately and reliably as overall indices. One notable exception relates to the inaccurate and unreliable conversion of the so-called cognition factor, the validity of which we have questioned. We thus strongly advice against using conversion equations for this specific factor. In order to facilitate the use of all other regression-based conversion equations by the clinical research community, an interactive version is made available through an R Shiny app. A first note of caution is however warranted: the conversion equation we provide are based on classical scoring of PANSS items from 1 to 7. In case PANSS items are originally scored from 0 to 6 [17], individual item scores should be rescaled prior to using our conversion equations. Moreover, we underscore that the good accuracy and reliability of conversion equations applied to averages does not imply good translation at the idiographic level [33]. Hence, care should be exercised when using our conversion equations in a clinical context or when predicting individual cases using machine learning approaches in research.
Limitations
Inherent to the dataset are a number of limitations which we only briefly address here, as they have been previously discussed [7, 16]. First, assessment of symptom severity by a unique rater in a single session likely leads to overestimate the accuracy of score conversion one would observe if ratings on the two scales were obtained by different raters [6]. Second, while the dataset rests on a community-based sample, all schizophrenia patients were psychiatrically stable with only mild to moderate symptom severity and no substance abuse. Hence, it is unknown how well the proposed conversion equations would perform for more severely ill or acute schizophrenia patients, given that our regression approach depends highly on the dispersion of symptom ratings. In the same line, positive and negative symptoms are present in other non-affective and affective psychotic disorders. Similar to schizophrenia, symptom ratings on the PANSS and SAPS/SANS appear highly correlated in bipolar disorder [34], however, it is unknown to what extent our specific conversion equations for subdomains of positive and negative symptomatology would generalize well to psychotic disorders other than prototypical stable schizophrenia. Finally, future work should investigate conversion equations with the Brief Psychiatric Rating Scale (BPRS, [35]), as this other symptom rating scale widely used to assess symptom severity in schizophrenia shows good correspondence with both the PANSS and SAPS/SANS [1].