Comparison of psychometric methods for examining the factor structure of the Supportive Care Needs Survey- Short Form (SCNS-SF34) in Colombian cancer patients.

Abstract

Background: to ensure optimal oncology care a structured assessment of different needs is highly desirable. The Supportive Care Needs Survey in its short form (SCNS-SF34) is frequently used to such objective. The survey has been validated in several contexts; accordingly, some reports suggest alternative structures. However, there are differences in validation outcomes attributable to study heterogeneity. Thus, we aimed to test the factor structure of the SCNS-SF34 comparing different psychometric methods.

Methods: an instrumental study was conducted in Bogotá-Colombia. A sample of 200 adult patients diagnosed and already treated for any type of cancer was estimated. Patients were randomly selected at a referral center; data was collected by trained personnel. To test the factor invariance, we used a parallel analysis for the exploratory factor analysis (EFA), the exploratory structural equation modelling (ESEM), and the exploratory graph analysis (EGA).

Results: overall, 245 patients were recruited; 64.0% women; 83.7% lived in urban areas; 34.7% had elementary education, and all were affiliated to a health insurance company in the Colombian health system. The parallel analysis yielded 5 factors explaining 55% of variance with low goodness of fit (CFI = 0.687, TLI = 0.602, RMSEA = 0.139, SRMR = 0.040). The ESEM adjusted for 5 factors with good fit (CFI = 0.971, TLI = 0.960, RMSEA = 0.023, SRMR = 0.04), however, some items presented a Heywood effect and loaded to different domains. The EGA showed 5 node communities, but some patient care and support items were found integrated within the health systems-and-information domain. The theoretical domains showed adequate reliability.

Conclusions: the 5 domains of the SCNS-SF34 showed structural validity for its application in Colombia. Further analysis in other Spanish-speaking Latin American countries are anticipated but our results suggest that ESEM and EGA approaches may be better to understand the structure of the survey.