Data originated from a cohort of 207 patients (mean age, 47.4 ± 12.8 years, 87 women) with chronic mental illness from a general psychiatric hospital (GGZ Centraal, Amersfoort, the Netherlands). The protocol was approved by the medical ethical committee, and written informed consent was obtained from each patient. Over the course of 4 years (July 2003–May 2007), with a mean (SD) period of observation of 1.1 (0.64), patients had at least two assessments of drug-induced movement disorders, namely tardive dyskinesia, akathisia, tardive dystonia, and parkinsonism. From this cohort, we included all patients with the presence of parkinsonism in the upper and/or lower limb(s) in at least one measurement. Full details of the study design and assessment of movement disorders can be found in a previous publication (18).

Movement disorders (tardive dyskinesia, akathisia, tardive dystonia, and parkinsonism) were assessed at baseline and each follow-up assessment by a trained psychiatrist, using a standard protocol (4). Parkinsonism was assessed on the Unified Parkinson Disease Rating Scale (UPDRS), motor scale (14 items, scored from 0 to 4) (53). Mean values of the right and left scores of the upper and lower limbs (UPDRS items, 20–26: 20–21 tremor, 22 rigidity, 23–26 bradykinesia) were calculated, and then the symmetry index was calculated according to the following formula: symmetry index = |right − left UPDRS score|/(right + left UPDRS score) (54, 55). The symmetry index ranged from 0 to 1, with higher values indicating a higher degree of asymmetry. A cut-off value of ≥0.20 was used to estimate the frequency of asymmetric parkinsonism (56). Severity of psychopathology was evaluated at baseline and each follow-up assessment using the Clinical Global Impression-Schizophrenia scale severity index (CGI-SCH SI), containing five subscales (positive, negative, depressive, cognitive, and global, scored from 1 to 7) (57).

Other variables extracted for the current study were age, sex, diagnosis according to DSM-IV, ethnic group (white or non-white), and duration of hospitalization, collected at baseline from patients’ case notes. At baseline and each follow-up assessment, the use of antipsychotic and anticholinergic medication at the time was collected from the hospital and outpatient pharmacy databases. As the data originated from a study that was conducted before the introduction of DSM-5, DSM-IV codes were used. The diagnosis “schizophrenia” hereafter refers to the DSM-IV codes 295.30, 295.10, 295.20, 295.90, 295.60, 295.70, and other diagnoses of “psychotic disorder” to 295.40, 297.1, 298.8, and 298.9.

We employed multivariate regression analyses for the associations between asymmetry in parkinsonism (symmetry index) and severity of psychopathology (CGI-SCH SI scales) for all measurements of the included patients (with parkinsonism in the upper/lower limb(s) in at least one measurement). Given the multilevel structure with repeated assessments clustered within the subjects, multilevel mixed models were built, using the MIXED routine of the Stata statistical program, with maximum likelihood estimation (58, 59). A random coefficient was included to allow the effect of the symmetry index to vary between subjects. Guided by previous literature, the following covariates were included in the model: age, sex, and diagnosis (schizophrenia or other) as time-independent variables, and dosage of antipsychotics (the defined daily dose), concomitant anticholinergic use (yes or no), and the presence of movement disorders (continuous measures of parkinsonism, dyskinesia, dystonia, and akathisia) as time-dependent variables. An interaction term between the covariates age and asymmetry in parkinsonism was introduced in the model. A quadratic term of the symmetry index was added to allow for a possible non-linear relationship between the symmetry index and severity of psychopathology.

The measures on the CGI-SCH SI scales were checked for normal distribution and allowed to be moderately skewed (60). Independent variables, including the symmetry index, could take almost any distribution, as the multilevel model with maximum likelihood estimation can handle these data well (60). Correction for multiple testing was not applied as this analysis concerned a hypothesis-driven approach, in which reduction of type I errors, with increase of type II errors and reduction of power as a consequence, was not of our main interest (61–63).

Table 1 shows the characteristics of the selected patients from the cohort, who had parkinsonism in the upper and/or lower limb(s) in at least one measurement (N = 190, women N = 79). Of the 190 patients, 181 (95.2%) and 113 (59.5%) patients completed the first and second follow-up assessment, respectively. All patients suffered from a chronic mental illness requiring long-term admission and had a history of cumulative antipsychotic intake of minimally 1 year. At baseline, the mean (SD) age of patients was 48.0 (12.9) years (range, 21.3–85.7 years), with a mean age at first admission of 25.0 (8.6) years. According to the DSM-IV criteria, 70.5% (N = 134) of the patients had a primary diagnosis of schizophrenia. Parkinsonism was the most prevalent persisting drug-induced movement disorder, with a persistence rate of 59.1% (N = 107). Over all measurements at baseline, first and second follow-up (number of measurements N = 477), the first-generation antipsychotic (FGA) and SGA use were 66.9 and 61.2%, respectively. Anticholinergics were concomitantly prescribed in 44.0% of all measurements (Table 2).

Symmetry of parkinsonism was evaluated for all measurements of parkinsonism in the upper and/or lower limb(s) (patients N = 190, number of measurements N = 404), as shown in Table 3. The mean (SD) symmetry index was 0.13 (0.24), which was the highest for the bradykinesia items: 0.16 (0.29). Asymmetric parkinsonism (symmetry index, ≥0.20) was present in 84 (20.8%) of the measurements.

The (multilevel) regression, controlled for predefined covariates, yielded both significant and non-significant coefficients between asymmetry in parkinsonism (symmetry index) and the severity of psychopathology (CGI-SCH SI subscales) with a range of 0.37–3.74 (patients N = 188, number of measurements N = 468), with a significant coefficient for the positive symptom scale (β = 3.74; 95% CI, 0.35–7.31, p = 0.031) (Table 4).

First, parkinsonism was (repeatedly) assessed on the UPDRS motor scale, a validated measurement tool with high test–retest reliability and internal consistency (72). The UPDRS motor scale provides severity ratings of tremor, bradykinesia, and rigidity of the left and right, upper and lower limbs separately, making this scale ideal for assessment of motor symptom asymmetry. Second, the assessments of all movement disorders were performed by a single person, a psychiatrist trained and retrained by an expert on movement disorders (senior author: PH), to prevent confounding of the results by reduced interrater or intrarater reliability. Third, we were able to obtain data from a cohort with a naturalistic and pragmatic design of a representative chronic psychiatric population (18, 73). Thus, the study yields results with a high external validity for clinical practice (73). Fourth, the importance of repeated measures, rather than a single cross-sectional measure, for movement disorders should be noted. Given the continuously fluctuating nature of movement disorders, a prospective study design may measure the phenotype more specifically. Therefore, this study design with repeated measures may result in a more stable phenotype, thus increasing the validity of the associations we found.

First, the approach used to quantify asymmetry in parkinsonism (the UPDRS motor symmetry index) may have led to biased results, as scores on the symmetry index tend to be low. Low index scores may be attributed to the limited range of the UPDRS motor scores (range, 0–4), which can easily produce ceiling effects. Furthermore, the symmetry index is a severity dependent scale, with relatively low scores when overall UPDRS motor scores are high. However, the symmetry index is considered as the most validated measurement for motor asymmetry and is preferred above other methods such as the absolute difference in side to side UPDRS motor scores (54, 55). In the current study population, overall UPDRS motor scores are relatively low (mean item score = 0.64, SD = 0.55), which makes the results unlikely to be biased by severity dependency or ceiling effects. Moreover, we were still able to identify a positive association between UPDRS asymmetry and psychopathology for all CGI-SCH SI subscales.

Second, the CGI-SCH SI is a brief assessment tool that may not be comprehensive or detailed enough to assess the severity of psychopathology properly. However, when compared to the Positive and Negative Symptom Severity (PANSS) scale, interrater reliability, sensitivity to change, and correlation coefficients between the instruments were high and comparable (57). Furthermore, the CGI-SCH SI might be more relevant for clinical practice, as it evaluates both symptom severity and interference with functioning, while the PANSS exclusively evaluates symptom severity (57).

Third, one may argue that Bonferroni correction for multiple testing should be applied. This would have resulted in a corrected critical p value of 0.01, with none of the results reaching significance. However, the Bonferroni correction assumes independency of the performed tests, which is probably not the case for this analysis as symptoms are mutually correlated. Furthermore, multiple testing corrections are designed to reduce type I errors (incorrect rejection of the null hypothesis) and thereby increase type II errors (incorrectly retaining of the null hypothesis) and greatly reduce power at the same time. Therefore, we chose not to correct for multiple testing but to accurately describe the performed tests and to discuss possible interpretations of the results instead (61–63). As this study concerned a hypothesis-driven research exploring a newly generated hypothesis, we considered the increase of type II errors undesirable. Furthermore, all beta coefficients for the CGI-SCH SI scales were estimated to be positive (range, 0.37–3.74), thus making it likely that asymmetry in parkinsonism is indeed positively related to psychopathology.

Finally, some authors may argue about the inclusion criteria for the regression analysis, as we did not use cut-off values for the case definition of parkinsonism. Instead, we included all patients presenting with parkinsonism in the lower and/or upper limb(s), even if symptoms were mild or were observed only once, to ensure a non-restrictive exploration of the association between asymmetry in parkinsonism and symptom severity. Noteworthy, post hoc analyses with more stringent inclusion criteria yielded nearly the same results. Therefore, the use of other inclusion criteria seemed irrelevant for the current analysis.

In conclusion, this study demonstrates a differential diagnostic and a new clinical relevant finding for psychiatric patients with DIP. Asymmetric presentation of DIP is quite common and therefore not useful to differentiate between DIP and other parkinsonian disorders. More important and clinical relevant is that asymmetry in DIP is significantly related to increased severity of psychopathology. If replicated, this could be of prognostic value to identify the more severely affected patient with possibly a poorer prognosis.