2008 HSR&D National Meeting Abstract
3057 — A Test of Dual-Process Cognitive Theories in Modeling Variation in A Meta-Analysis of Computerized Interventions
Weir CR (SLC TREP)
Recent reviews of healthcare Information Technology literature have found large and unexplained variation across studies. One explanation is that the literature lacks sufficient theoretical specification of contextual and moderating variables to produce generalizable results. Current dual-process cognitive models could clarify our understanding of this literature. These models propose two memory systems, one experiential and emotional, where decisions are effortless, quick and less conscious. The other system is linguistic, symbolic and decisions are more rational, slower, effortful and take more awareness. One important implication of this perspective is that differences in the complexity of a decision-support intervention would require different implementation strategies. The purpose of this study was to test this prediction using meta-analytical techniques on a systematic review of: 1) infectious disease protocols, 2) antibiotic prescribing support and 3) vaccination reminders. The specific prediction is that high social influence implementation strategies will improve low complexity decision-support but not for highly complex decision-support. Similiarly, low social influence implementation strategies would be expected to increase the effectiveness of highly complex decision-support, but but not for simple decision-support interventions.
Standard procedures were used to conduct the search and relevance coding. Medline, CINAHL, Cochrane, PsychInfo, DARE, INSPEC, and EMBASE were searched from 1976 through the end of 2005. Articles were judged relevant if they reported a computerized intervention that occurred in a real clinical environment targeted at clinicians.
The search yielded 2,073 citations with 38 articles meeting relevance and quality criteria. Inter-rater reliability for coding intensity of social influence and decision-support complexity was good (kappa = 0.72). The Q test for heterogeneity was highly significant (X2 = 104.7; df = 27). Meta-regression analysis modeling sources of heterogeneity found that the proposed interaction (creating interaction coding terms) showed a significant effect such that social influence factors had more impact in studies with low complex tasks and explained significant heterogeneity (p < .02).
The interaction between social influence implementation variables and decision-support complexity explained more variance in study outcomes than each of those variables alone.
Using dual process cognitive models may improve the effectiveness of decision support design and implementation.