{"columns":["prior_id","parameter","model_scale","applies_to","prior","sensitivity_prior","rationale"],"rows":[["PR01","Trial baseline (reference arm) mu_j","MD and SMD","All continuous outcomes","Normal(0; (5 s_y)^2)","Normal(0; (100 s_y)^2)","Weakly informative on the outcome scale; s_y is the reference SD in reference_sd.csv (s_y = 1 for SMD outcomes)"],["PR02","Trial baseline mu_j","log_odds; log_hazard_cloglog; log_rate","All binary and count outcomes; the mean of the reference arm meta-analysis of @sec-absolute on the same scales; a single usable arm's theta_a, and each usable arm's theta_a for the scenario risks of @sec-absolute","Normal(0; 5^2)","Normal(0; 10^2)","Covers event risks from about 1 in 10000 to near 1 without favoring 50 percent strongly. In the network meta-analysis complementary log log model the offset is log follow-up in years, so the trial baseline is a log cumulative hazard per year; in the reference arm meta-analysis of a hazard based outcome without a competing event the same prior is placed on the mean log cumulative hazard at the analyzed horizon; fixed time reference arm models use the logit scale"],["PR03","Relative effect d","MD","All MD outcomes","Normal(0; (2.5 s_y)^2)","Normal(0; (10 s_y)^2)","Weakly informative; plausible drug effects lie well within 2.5 reference SDs"],["PR04","Relative effect d","SMD","O24B;O24C","Normal(0; 2.5^2)","Normal(0; 10^2)","As PR03 on the SMD scale"],["PR05","Relative effect d","log_odds; log_hazard; log_hazard_cloglog; log_rate","Outcomes with rare_outcome = no in outcomes.csv (fixed before extraction)","Normal(0; 2.5^2)","Normal(0; 10^2)","Weakly informative; ratio of 134 at 1.96 SD leaves room for large responder effects"],["PR06","Relative effect d","log_odds; log_hazard; log_hazard_cloglog; log_rate","Outcomes with rare_outcome = yes in outcomes.csv (fixed before extraction)","Normal(0; 1.5^2)","Normal(0; 2.5^2)","Regularizes sparse rare event networks; the same a priori rare_outcome flag sets the relative threshold, so prior and threshold choices never depend on observed data"],["PR07","Heterogeneity tau","MD","All MD outcomes","Half-normal(0; (0.5 s_y)^2)","Uniform(0; 5 s_y)","Weakly informative; consistent with empirical heterogeneity of continuous outcomes (Rhodes 2015) once expressed in SD units"],["PR08","Heterogeneity tau","SMD","O24B;O24C","Half-normal(0; 0.5^2)","Uniform(0; 5)","As PR07"],["PR09","Heterogeneity tau^2","log_odds; log_hazard; log_hazard_cloglog; log_rate","All binary; time to event and rate outcomes","Log-normal predictive distribution of Turner 2015 Table IV for the mapped outcome type (turner_map table) and the comparison type rule","Half-normal(0; 0.5^2) on tau and Half-normal(0; 1^2) on tau","Empirical predictive distributions; applied to log hazard and log rate ratios as an approximation"],["PR10","Class effect SD","All scales","Class effects model","Half-normal(0; (0.5 s_y)^2) for MD; Half-normal(0; 0.5^2) for log scales","Half-normal with double the scale","Allows moderate differences between agents within a class"],["PR11","Meta-regression interaction beta","All scales","Network meta-regression","Normal(0; 1^2) per 1 SD of the centered covariate (on the outcome scale divided by s_y for MD)","Normal(0; 2.5^2)","Grok round 1 comment 23"],["PR12","Emax dose response","All scales","Dose response models","Emax ~ Normal(0; (2.5 s_y)^2) or Normal(0; 2.5^2) on log scales; log ED50 ~ Normal(log of the median studied maintenance dose; 1^2)","Emax scale doubled; log ED50 SD 2","Weakly informative within the studied dose range"],["PR13","Bias adjustment","All scales","Exploratory bias adjusted synthesis of nonrandomized evidence","Published before use","Not applicable","Prespecified separately"],["PR14","Relative effect d","log_rate","O04B (reported log rate ratios only)","Normal(0; 2.5^2)","Normal(0; 10^2)","As PR05; no count model is fitted, so no dispersion prior is needed"],["PR15","Reference log hazard mu (per year)","log_hazard","Piecewise integration and exponential approximation of reference cause specific hazards (competing risks)","Normal(0; 5^2) per event and interval","Normal(0; 10^2)","Weakly informative on the log hazard per year; covers rates from about 1 in 10000 to more than 100 per year"],["PR16","Arm level heterogeneity tau_k of reference hazards","log_hazard","Piecewise integration and exponential approximation (one tau per event)","Half-normal(0; 1^2)","Half-normal(0; 0.5^2) and Half-normal(0; 2^2)","Weakly informative for proportional deviations of reference arm hazard functions"],["PR17","Heterogeneity tau of reference arm risks","log_odds; log_hazard_cloglog; log_rate","Random effects meta-analysis of reference arm risks for absolute effects (@sec-absolute)","Half-normal(0; 1^2)","Half-normal(0; 0.5^2) and Half-normal(0; 2^2)","Weakly informative on the logit or log scale; settles the reference arm prior deferred in round 12"]]}