Comparing the CASI-4R and the PGBI-10 M for Differentiating Bipolar Spectrum Disorders from Other Outpatient Diagnoses in Youth

Abstract

We compared 2 rating scales with different manic symptom items on diagnostic accuracy for detecting pediatric bipolar spectrum disorder (BPSDs) in outpatient mental health clinics. Participants were 681 parents/guardians of eligible children (465 male, mean age = 9.34) who completed the Parent General Behavior Inventory-10-item Mania (PGBI-10M) and mania subscale of the Child and Adolescent Symptom Inventory-Revised (CASI-4R). Diagnoses were based on KSADS interviews with parent and youth. Receiver operating characteristic (ROC) analyses and diagnostic likelihood ratios (DLRs) determined discriminative validity and provided clinical utility, respectively. Logistic regressions tested for incremental validity in the CASI-4R mania subscale and PGBI-10M in predicting youth BPSD status above and beyond demographic and common diagnostic comorbidities. Both CASI-4R and PGBI-10M scales significantly distinguished BPSD (N=160) from other disorders (CASI-4R: Area under curve (AUC) = .80, p < .0005; PGBI-10M: AUC = .79, p < .0005) even though scale items differed. Both scales performed equally well in differentiating BPSDs (Venkatraman test p > .05). Diagnostic likelihood ratios indicated low scores on either scale (CASI: 0–5; PGBI-10M: 0–6) cut BPSD odds to 1/5 of those with high scores (CASI DLR− = 0.17; PGBI-10M DLR− = 0.18). High scores on either scale (CASI: 14+; PGBI-10M: 20+) increased BPSD odds about fourfold (CASI DLR+ = 4.53; PGBI-10M DLR+ = 3.97). Logistic regressions indicated the CASI-4R mania subscale and PGBI-10M each provided incremental validity in predicting youth BPSD status. The CASI-4R is at least as valid as the PGBI-10M to help identify BPSDs, and can be considered as part of an assessment battery to screen for pediatric BPSDs.

Diagnostic Efficiency of the Child and Adolescent Symptom Inventory (CASI-4R) Depression Subscale for Identifying Youth Mood Disorders

“…Consistent with prior work (Ong et al, 2016; Van Meter et al, 2016) and commonplace in epidemiological work to get more accurate population parameter estimates (Little, Lewitzky, Heeringa, Lepkowski, & Kessler, 1997), we weighted the data used in these analyses to address the unequal probability of selection and sample response. First, a base weight was created by utilizing the inverse of the probability of selection.…”

“…The resulting final weight was calculated by taking the product of the base weight and non-response weight, rescaled so that the sum was equal to the sample size. This weighting process created a sample that more closely matches the overall presenting clinical population, reducing the risk of misleading results, and has been used in previous studies using similar analytic methods (Ong et al, 2016; Van Meter et al, 2016). …”

Diagnostic accuracy of the Child and Adolescent Symptom Inventory (CASI-4R) substance use subscale in detecting substance use disorders in youth.

“…Earlier investigations of the CASI SU subscale have found adequate to good internal consistency and convergent criterion validity with ASEBA scales (Birungi et al, 2020;Cornellà-Font et al, 2020). Prior investigations of the corresponding CASI scales found good classification accuracy for depression (Salcedo et al, 2018), bipolar disorders (Ong et al, 2017), and psychosis (Rizvi et al, 2019). However, no published study has evaluated classification validity (diagnostic accuracy) of the CASI SU subscale.…”

The Dimensionality of Youth Irritability in a Non-Clinical US Sample

“…A scale with 47 items will have inflated internal consistency due to length (Streiner et al, 2015;E.A. Youngstrom et al, 2019) and may be too long for a clinically useful inventory, so future work may look to identify a well-performing short form from these items(for an example, see Ong et al, 2017). Alternatively, further factoring of these items may separate items more related to depression from those related to anxiety (Chorpita, 2002;Clark & Watson, 1991); however, the fact that these items grouped together on one scale in this sample may lend credence to the idea that these constructs are dimensional.…”