all88study10unit27outcomes43rob8

rob8 fields

robins_confounding	LowModerateSeriousCritical	ROBINS-I Domain 1: bias from confounding in the intervention-vs-comparator effect estimate. LOW: design adds RCT-grade control (target-trial emulation with active comparator, negative-control falsification, or a quasi-experimental instrument) AND analysis covers important confounders. MODERATE: analysis adjusts for important measured confounders via propensity matching/weighting, regression, or stratification, but residual unmeasured confounding remains plausible. SERIOUS: the adjustment set omits a clinically-recognized driver of treatment choice or strong prognostic factor for the outcome, OR no quantitative adjustment was performed. CRITICAL: confounding renders the result uninterpretable.
robins_selection	LowModerateSeriousCritical	ROBINS-I Domain 2: bias from selection of participants into the analytic cohort. LOW: selection based on pre-intervention characteristics; analytic cohort defined at start of follow-up; no post-baseline exclusion correlated with intervention or outcome. MODERATE: some post-baseline exclusion or analytic-cohort filtering present but unlikely to depend on both intervention and outcome. SERIOUS: selection depends on both intervention and outcome (collider stratification, immortal-time bias, post-baseline cohort definition correlated with outcome). CRITICAL: selection virtually guarantees biased results.
robins_classification	LowModerateSeriousCritical	ROBINS-I Domain 3: bias from misclassification of intervention status. LOW: intervention status documented from the procedural record at the start of intervention. MODERATE: intervention status inferred from a proxy variable (administrative category, time-threshold cut-off, billing or ICD code) or from records recorded after intervention started; classification not influenced by outcome knowledge. SERIOUS: classification influenced by outcome knowledge (recall, retrospective recoding) or differential misclassification across groups. CRITICAL: classification errors invalidate the comparison.
robins_deviations	LowModerateSeriousCritical	ROBINS-I Domain 4: bias from deviations from intended interventions. LOW: no substantial deviations from intended intervention, OR the analysis appropriately estimates the targeted effect (ITT-style for assignment effect; g-methods for per-protocol effect). MODERATE: minor deviations not handled analytically. SERIOUS: substantial unbalanced deviations affecting the outcome, OR per-protocol analysis uses standard regression on time-varying confounders affected by prior intervention. CRITICAL: deviations render the comparison meaningless.
robins_missing	LowModerateSeriousCritical	ROBINS-I Domain 5: bias from missing data on intervention, outcome, or important confounders. LOW: complete data for at least 95% of participants with proportions similar across groups, OR appropriate multiple imputation, OR sensitivity analyses show results robust to missingness. MODERATE: some missing data with reasonable handling but incomplete documentation; missingness unlikely to depend on the true outcome value. SERIOUS: substantial missingness depending on the true outcome value or differing between groups, OR only LOCF / single imputation used for substantial missingness, OR no information on extent of missing data. CRITICAL: missing data so extensive and selective that the result cannot be interpreted.
robins_measurement	LowModerateSeriousCritical	ROBINS-I Domain 6: bias from outcome measurement (differential or non-differential measurement error). LOW: outcome is hard and objective (mortality, lab values, registry-linked death) AND ascertainment methods are identical across groups AND surveillance frequency does not differ by intervention. MODERATE: assessment methods comparable across groups; blinding not documented for a judgment-dependent outcome; minor ascertainment timing differences without documented detection differential. SERIOUS: differential ascertainment across groups (active vs passive surveillance, different diagnostic thresholds, different follow-up intensity), OR subjective outcome assessed by unblinded assessors with strong prior beliefs about the intervention. CRITICAL: measurement so unsuitable that the intervention-outcome relationship is uninterpretable.
robins_reporting	LowModerateSeriousCritical	ROBINS-I Domain 7: bias from selective reporting of the result. LOW: pre-registered protocol or statistical analysis plan available and results match the pre-specified analysis. MODERATE: no protocol available but no apparent selective outcome or analysis reporting. SERIOUS: evidence of selective outcome reporting, analysis switching, or subgroup-cherry-picking. CRITICAL: reporting selection virtually guarantees a biased result.
robins_overall	LowModerateSeriousCritical	ROBINS-I overall risk-of-bias judgment, computed as the worst per-domain judgment. LOW only when every domain is LOW (the result is comparable to a well-conducted RCT). MODERATE when at least one domain is MODERATE and no domain is SERIOUS or CRITICAL. SERIOUS when at least one domain is SERIOUS and no domain is CRITICAL. CRITICAL when at least one domain is CRITICAL.