Abstract: Psychometric Validation of the Computer Adaptive Multidimensional Scale (CAMS) (Society for Prevention Research 27th Annual Meeting)

74 Psychometric Validation of the Computer Adaptive Multidimensional Scale (CAMS)

Tuesday, May 28, 2019
Pacific D/L (Hyatt Regency San Francisco)
* noted as presenting author
Wenqiu Cao, BS, Graduate student, University of Rhode Island, Narragansett, RI, China
Theodore Walls, PhD, Associate professor, University of Rhode Island, Kingston, RI
Lyn Stein, PhD, Professor, University of Rhode Island, Kingston, RI
Devon Bush, BSE, Chief Product Officer, Mirah, Inc., Somerville, MA
Introduction: In inpatient and outpatient psychotherapy treatments, therapists increasingly record clinical judgments in well-structured on-line measurement systems. One such database was considered in depth in terms of constructs implicated in a specific constellation of substance use behavior. Specifically, one privately-housed clinical database employed a measure entitled Computer Adaptive Multidimensional Scale (CAMS), which was developed in Norway and recently translated for use in US based system. Patients are asked to finish the measurement before every session. In this poster, we report findings of psychometric validation efforts of three related subconstructs implicated in substance use. This work is crucial both with respect to temporal assessments over the course of treatment and to cross-validation efforts between US and Norway clinical samples. Psychometric validation of the dimensional space involving susbstance use behavior is urgently needed to monitor misuse risks, particularly in inpatient settings. The author conducted a secondary data analysis, approved by University of Rhode Island IRB, HU1718-124.

Methods: In this study, we incorporated 17 Likert-scale questions reflecting three subsconstructs: substance use, pressure from negative affect, and perfectionism. Analyses were conducted using a nested model strategy in a latent variable modeling framework. The data was collected from several clinical sites in West coast of US. The data of adult patients’ first visit were selected as sample, reflecting 260 patient reports on 17 variables. We deployed an unconstrained 1-factor model, an unconstrained 3-factor model, and a constrained 3-factor model. The three latent factors in the later models were allowed to covary. We assessed each model for absolute fit and considered the relative fits of the constrained model against both unconstrained models.

Results: For 1-factor model, the standardized loadings of the indicators ranged from 0.33 to 0.69, but the absolute fit of the model was not adequate. The results of 3-factor unconstrained model reflected a good fit, χ2(88,N=260)=141.648, p<0.001, RMSEA=0.055 with 90%CI [0.039,0.069]. The constrained 3-factor model was conducted with factor variances fixed at 1.0 and it showed a good absolute fit with χ2(116,N=260)=275.407, p<0.000, CFI=0.920, RMSEA=0.068 with 90%CI [0.057,0.080]. Moreover, RMSEA of constrained 3-factor model fell within the 90%CI for RMSEA of 3-factor unconstrained, which indicated that the 3-factor model fit relatively well with the data. Coefficient alphas for substance use subscale was 0.92, 0.63 for pressure from negative affect, and 0.74 for perfectionism.

Conclusions: These results establish baseline psychometric properties of the three subscales in at once occasion in a US sample. Our broader goal is to ensure that our measures can be employed effectively in a structured data-tracking system over time and across cultures. Hence, this baseline psychometric report is pivotal to our next steps in both longitudinal analysis of factorial invariance and cross-cultural validation study. Longer term goals involve development of adaptive real-time algorithmic prediction systems based on this construct constellation.

Devon Bush
Mirah, Inc: Employment with a For-profit organization