Tuesday, October 25, 2016
Bayshore Ballroom ABC, Lobby Level (Westin Bayshore Vancouver)
Poster Board # PS 3-1

Tara Lavelle, PhD1, David M. Kent, MD, MSc2, Christine Lundquist, MPH2, Teja Thorat, M.Sc., MPH2, Joshua T. Cohen, PhD2, John B. Wong, MD2, Natalia Olchanski, MS2 and Peter J. Neumann, Sc.D.2, (1)Center for the Evaluation of Value and Risk in Health, Institute for Clinical Research and Health Policy Studies, Tufts Medical Center, Boston, MA, (2)Tufts Medical Center, Boston, MA
Purpose: Previous research has shown that heterogeneity in patient characteristics can lead to important variation in cost-effectiveness analyses (CEAs), but how patient variability has been incorporated into published CEAs is unknown.

Method: Using the Tufts Medical Center Cost-Effectiveness Analysis Registry, which contains information from 5,000 original CEAs published in peer-reviewed journal articles through 2014, we randomly selected 100 out of 642 articles published in 2014 for review.  Registry data included the type of intervention under study, the disease classification, country of study, author affiliation, and funding source. Two trained readers reviewed each article and determined through consensus whether it incorporated patient-level heterogeneity into its analysis, which observable patient characteristics represented heterogeneity, and what related patient attributes, including differences in baseline event rates and treatment effectiveness, potentially influenced heterogeneity.  We described the data and compared studies that did and did not incorporate heterogeneity using the chi-squared test.

Result: Among the 100 CEAs reviewed, 19 (19%) incorporated patient-level heterogeneity.  Within this subset, the most common characteristics representing patient-level heterogeneity were patient age (68%, 13 studies), a test or procedure result (32%, 6), and the history or presence of a co-morbid disease (26%, 5).  Three studies (16%) used a risk score to incorporate multiple patient characteristics into calculations of predicted risk for cardiovascular disease (using the Framingham Risk Score; 2 studies) and stroke (using the CHADS2 Score; 1 study). Articles most frequently reported variability in CEA results to be due to patient differences in baseline event rates (63%, 12 studies), treatment effectiveness (21%, 4 studies), and health state costs (16%, 3).  Studies that incorporated heterogeneity were more likely than those that did not to focus on pharmaceuticals (53% versus 43%, p-value=0.61) and screening interventions, (21% versus 11%, p-value=0.26), and more likely to target primary prevention (26% versus 12%, p-value=0.15).  A greater proportion of studies incorporating heterogeneity received government funding (58% versus 31%, p-value=0.04). 

Conclusion: Most CEAs do not incorporate patient-level heterogeneity.  Among those studies that do, the most common characteristics representing patient-level heterogeneity are age, a test or procedure result, and the history or presence of co-morbid disease.  Future studies should endeavor to incorporate patient-level variability into CEAs.