Developing Equivalent Nicotine Dependence Measures to Facilitate Analysis of Integrated Data Sets

Introduction: There has been an increased call for secondary data analysis to capitalize on existing data. However, researchers often find that existing individual data sets are inadequate for testing secondary hypotheses due to small sample sizes, lack of heterogeneity on key characteristics, and differences construct measurement. For example, development of nicotine dependence (ND) in young Hispanic smokers who are often recent onset, light smokers remains understudied, primarily due to small sample sizes and lack of smoking exposure heterogeneity in independent studies. Integrated data analysis (IDA) allows us to pool independent data sets to create larger, more heterogeneous data sets to test novel hypotheses. The key to IDA is developing measures that are psychometrically equivalent across data sets. Moderated nonlinear factor analysis (MNLFA) is a particularly versatile method to identify differences in psychometric properties of variables across studies and adjust for these differences to facilitate IDA. We used MNLFA to identify differential item functioning (DIF) in DSM ND symptoms across multiple studies and by gender and ethnicity.

Methods: We used integrated DSM ND symptom data for current smokers aged 12-25 (N=20,328) from 3 nationally representative surveys (1999, 2000 National Surveys on Drug Use and Health (NSDUH) and Wave 1 National Epidemiologic Survey on Alcohol and Related Conditions (NESARC). MNLFA tested symptom DIF in the pooled sample containing 2,193 Hispanic smokers and variability in smoking exposure.

Results: There was study DIF for 4 of 7 symptoms. NESARC participants were more likely to report tolerance, using larger amounts for longer periods, inability to cut down/quit, and more time spent smoking at higher levels of ND severity. Hispanic smokers were more likely to endorse tolerance, giving up activities, and physical/emotional symptoms at lower levels of ND severity compared to White smokers, but less likely than Black smokers to endorse tolerance and larger/longer symptoms at lower levels of ND severity. The process of developing and testing an MNLFA model for integrating ND symptom data revealed both the promise and pitfalls of MNLFA from harmonizing measures across studies through model specification, estimation, and interpretation.

Conclusions: MLFA is a feasible and promising approach for developing equivalent measures to facilitate IDA for testing novel hypotheses not possible in independent studies. However, it needs wider dissemination in the prevention research community. Online learning platforms and project-based learning initiatives, which provide increased opportunity for prevention researchers to apply MNLFA for IDA in an accessible and supportive environment, may accelerate dissemination.