Comparing a Conceptual Framework and Factor Analysis to Achieve Survey Item Reduction in Predicting Medication...

Area of Science:

• Health Services Research
• Psychometrics
• Biostatistics

Background:

• Adherence surveys are often lengthy, prompting item reduction into conceptual domains.
• Item reduction can be guided by conceptual frameworks or statistical methods like factor analysis (FA).
• The impact of different domain creation methods on adherence prediction models was previously unclear.

Purpose of the Study:

• To compare the impact of conceptual versus factor analysis (FA)-guided domain reduction on adherence prediction model performance.
• To evaluate if domain structure affects downstream model utility.

Main Methods:

• Two logistic regression models were developed using the same survey data to predict non-persistence.
• Survey items were reduced into three domains using either a conceptual (WHO framework) or a factor analysis (FA) approach.
• Model predictive performance was assessed using c-statistics.

Main Results:

• Both conceptual and FA approaches identified three domains from 51 survey items.
• FA-guided domains included items that crossed conceptual boundaries, unlike the WHO framework-based domains.
• Both models showed comparable predictive performance: c-statistics of 0.84 (conceptual) and 0.82 (FA), with no significant difference (p=0.060).

Conclusions:

• No trade-off in predictive performance was observed between conceptual and FA-guided domain reduction.
• The conceptual approach provides a more relevant and practice-oriented data structure.
• Domain creation methodology impacts data organization but not necessarily prediction accuracy in this context.