CAUSALITY WITH REAL WORLD EVIDENCE: BURDEN OF PROOF, PROOF OF PRINCIPLE

OBJECTIVES: To establish and evaluate whether replicating and reconciling surrogate markers in epidemiological studies can establish causal relationships. METHODS: Multiple longitudinal and panel data sources from different countries quantify the conditional probabilities of surrogate endpoints (e.g., heart rate, treatment(s), etc.) and established biological outcomes (e.g., all-cause mortality, HbAlc and diabetic-complications, etc.). Therapeutic or biological models consist of all-cause mortality and heart rate; diabetic-complications/ HbA1c and treatment; adverse events and pain management; and mortality and HDL-C.Countries include the UK, Germany, U.S., Canada and Australia. Estimation methods include survival analyses and generalized estimating equations (GEE). Pairwise results from those analyses were compared (reconciled) to all previous RCTs and epidemiological studies identified by comprehensive and systematic reviews of published studies where odds-ratios served as measures in common. RESULTS: Heart rate predicted mortality trice with survival and/or GEE analyses, 0.00694 (P< 0.001) in Canada, 0.00683 (P< 0.001) in Copenhagen panel data (1981-1983) and 0.00717 (1991-1993) with the Weibull. With GEE, the coefficients were 0.0268 (P=0.006) in Australia, 0.0249 (P=0.008) in the meta-regression of 16 controlled clinical trials. All three reproduced clinical trials with odd-ratios within 1/100ths. In diabetes, longitudinal analyses of treatment trice reproduced HbA1c reductions by 0.99% (UK), 0.92% (Germany) and 0.89% (US) with complication risks reduced by 0.388%, 0.414% and 0.436% per 1% HbA1c reduction. In pain management, 9 of 10 adverse events are statistically the same in two datasets. GEE analyses from Australia data reproduced 25 epidemiological studies for mortality and HDL-C. CONCLUSIONS: Precise reproductions of conditional probabilities across multiple datasets and their numeric reconciliations with other established clinical and epidemiological evidence establish causality. Real World epidemiological studies support clinicaql findings in larger, more general populations and when replicated and reconciled among one provide proof of principle for casual inferences.