Introduction: The Positive and Negative Syndrome Scale (PANSS) is a widely accepted outcome measure for pediatric schizophrenia trials; however, it has notable limitations. Psychometric investigations have shown a multifactorial structure and some items have limited utility assessing symptom severity in children. To address these issues, we developed and evaluated optimized 10- and 20-item PANSS short-forms (PANSS10 and PANSS20) using patient-level clinical trial data. This study further assesses these optimized forms using independent clinical trial data. Methods: We examined patient-level data from a randomized pediatric schizophrenia trial comparing paliperidone ER to aripiprazole. Data were accessed through the Yale Open Data Access (YODA) secure platform. Analyses included confirmatory factor analyses, graded response models, ω score reliability, internal consistency, sensitivity to change, and criterion validity versus the Clinical Global Impressions of Severity (CGI-S). Bland-Altman analyses examined score calibration versus the 30-item PANSS and inclusion cut scores. Results: Participants (N = 288) were ages 12 to 17 years (M = 15.3, SD = 1.46; 66% male). Total scores for the PANSS10 and PANSS20 showed strong correlations with the 30-item PANSS (0.90 and 0.97, respectively). Average inter-item correlations were 0.10 and 0.14 and ωTotal reliabilities were 0.74 and 0.85. Both PANSS10 and PANSS20 scores showed reliability >0.80-2.3 to 4.5 SD and -3.0 to 6.0 SD about mean symptom severity, respectively. Sensitivity to treatment was also similar (partial eta squared 0.23 and 0.22), as was correlation with CGI-S at baseline (0.45 and 0.48; not significantly different). The mean item-average discrepancy with the 30-item PANSS was 0.095 for PANSS10 and 0.033 for PANSS20. Conclusions: The optimized PANSS forms continue to show impressive reliability, validity, and calibration compared with the 30-item PANSS. Researchers should consider replacing the 30-item PANSS with the PANSS10 as a clinical outcome and screening measure due to its length and psychometric performance.
Read full abstract