Integrating analytical methods to improve comparative effectiveness research / John B. Wong, Elizabeth R. DeLong, Thomas A. Trikalinos.

Format:

Book

Author/Creator:

Wong, John B., author.

Contributor:

DeLong, Elizabeth R., contributor.

Trikalinos, Thomas A., contributor.

Language:

English

Subjects (All):

Medical care.

Medical policy.

Physical Description:

1 online resource (98 pages)

Place of Publication:

Washington, DC : Patient-Centered Outcomes Research Institute (PCORI), 2019.

Summary:

BACKGROUND: PCORI has identified 3 methodological gaps: (1) "Comparison of the validity of different methods for reducing confounding and bias using randomized controlled trials (RCT) and observational registry studies"; (2) "Research on methods for improving the validity of systematic reviews of comparative effectiveness research"; and (3) "Development of methods to improve and/or compare research prioritization methods, including value of information approaches." OBJECTIVE: To develop a framework for comparative effectiveness research by integrating causal inference methods (comparing regression modeling, inverse probability of treatment weighting, and optimal full propensity matching) for observational data; evidence synthesis methods for observational, RCT, individual patient- and study-level data; and research prioritization through value of information analysis. METHODS: For subsets (by the number of diseased vessels) of patients with chronic stable angina who would or would not have met inclusion criteria from 6 RCTs, we compared survival with percutaneous coronary intervention (PCI), coronary artery bypass grafting (CABG), or medical therapy (MED) in a large observational cardiovascular registry using causal inference methods to adjust for nonrandomization. Second, we combined the individual-level registry regression results with published study-level RCT results to compare the mortality hazard ratios (HRs) for the 3 treatments in a cumulative network meta-analysis. Last, based on the regression survival model from the registry and the mortality HRs from the cumulative network meta-analysis, we assessed the value of performing future research by estimating the net benefits of reducing residual uncertainty over time. RESULTS: In the registry, the minority of patients would have met RCT inclusion criteria. RCT-eligible patients differed significantly from ineligible ones and had higher unadjusted 5-year survivals than ineligible patients. Although some subgroups had evidence of better survival among RCT-eligible patients, when adjusting for covariate differences and accounting for RCT eligibility, the causal inference methods (comparing regression modeling, inverse probability of treatment weighting, and optimal full propensity matching) produced essentially equivalent survival plots, suggesting similar comparative effectiveness conclusions about treatment benefit. To compare treatment HRs from a regression analysis in the registry with RCTs, we performed a cumulative network meta-analysis and found similar mean treatment effect across studies, with CABG and PCI favored over MED, and CABG favored over PCI. Adding the above-described individual-level registry regression analysis to the RCT network meta-analysis increased the precision of the estimates and the 95% CI. Increasing the amount of information "borrowed" (using Bayesian methods) from other observational studies increased precision but did not change the comparative effectiveness point estimates appreciably. Using the above-described registry-based regression and cumulative network meta-analysis comparative mortality HRs for treatment, we retrospectively determined the value of obtaining additional information through research to reduce uncertainty about the optimal treatment after the publication of each RCT. The value of additional research was greatest in the 1980s because of uncertainty about the comparative effects of PCI but fell substantially in the 1990s as PCI studies emerged and as accumulating evidence from direct comparisons to MED and CABG substantially reduced uncertainty on the optimal revascularization decision. Currently, CABG vs PCI has the largest decisional uncertainty about their comparative survival benefits. CONCLUSIONS: These analyses demonstrate the feasibility and potential utility of integrating causal inference methods, evidence synthesis, and value of information analysis as a framework for comparative effectiveness research to inform health policy and research priorities. This integrated approach can improve the validity (through advanced statistical causal inference methods), credibility (through evidence synthesis), and efficiency (through value of information analysis) of comparative effectiveness research. LIMITATIONS AND SUBPOPULATION: Our analysis provides the application of our integrated comparative effectiveness approach for 1 disease area, chronic stable angina.

Contents:

Background

Participation of Patients and Other Stakeholders and Dissemination of Research

Methods

Results

Discussion

Conclusions

References

Publications

Acknowledgment

Appendix.

Notes:

Description based on publisher supplied metadata and other sources.