Longitudinal and Temporal
Issues for Long-term Studies:
Design, Data, and Analysis
Patrick J. Heagerty PhD Professor and Chair
Department of Biostatistics University of Washington
Longitudinal Studies
• Study design •Outcome ascertainment •Temporal trends •Heterogeneity • Patient forecasting • Statistical re-engineering•Distributed (strictly separated) data
Longitudinal Studies
• BOLD (AHRQ R01 HS019222-1, HS22972-1)
• Backpain Outcomes using Longitudinal Data
• 5,000 subject cohort with PROs and EHR
• Kaiser NoCA, Henry Ford, Harvard Pilgram
• LIRE (NIH UH2/3 AT0007766 )
• Lumbar Image Reporting with Epidemiology
• 4 systems / 100 clinics / 300,000 subjects
Longitudinal Study Design
Longitudinal Study Design
Longitudinal Study Design
• Interrupted time-series methods •Wagner et al. (2002)
•Policy evaluation
• Synthetic controls
•Brodersen et al. (2014)
•A/B testing
• Crossover / Stepped-wedge designs •Hussey and Hughes (2007)
•Woertman et al. (2013)
Cross System Heterogeneity
Cross System Heterogeneity
Longitudinal Study Design
• Q: dynamic / adaptive designs?
• Q: statistical methods for study monitoring?
• Q: designs, methods for simultaneous evaluation of multiple interventions?
Patient Forecasting
Patient Forecasting
• Individualized predictions • Machine learning • “Regular” covariates • ISSUES: • Time origin?• Heterogeneity in timing of history? • Robust distributional summaries? • Comparative predictions?
Statistical Re-engineering
Longitudinal Data
• Individual change in exposure • Individual change in outcomes
• Controlled research designs that address temporal trends
• Robust, dynamic patient forecasting • Privacy
