Topic

Causal inference and real-world evidence

When patients were not assigned to treatment at random, effects have to be estimated in ways that allow for confounding. Methods include propensity score matching, instrumental variable analysis, difference-in-differences and interrupted time series. Real-world data from routine sources, such as administrative data, supply the evidence these methods work on.

Concepts in this topic

• Administrative DataAdministrative data are records made to run or pay for care, such as claims, hospital admissions and prescriptions, reused to measure costs and outcomes.
• Average Treatment EffectThe average treatment effect (ATE) is the mean change in costs or health outcomes if a whole population received a treatment rather than its comparator.
• Causal InferenceCausal inference uses data, study design, and explicit assumptions to estimate what would happen to an outcome if an exposure, intervention, or policy were changed.
• Decision Curve AnalysisDecision curve analysis (DCA) plots the net benefit of a prediction model or test across risk thresholds against treating all or no patients.
• Difference-in-DifferencesDifference-in-differences (DiD) estimates a policy's causal effect by comparing outcome changes over time in an exposed group and a comparison group.
• EffectivenessThe benefits and harms an intervention produces under care conditions relevant to a specified population, comparator, outcome and time horizon.
• Heterogeneous Treatment EffectVariation in the size or direction of a treatment's effect across individuals or subgroups, so the overall average may not fit any one patient.
• Instrumental VariableAn instrumental variable is a variable that changes treatment or exposure uptake while, under specified causal assumptions, being independent of potential outcomes and affecting the outcome only through that treatment or exposure.
• Instrumental Variable AnalysisInstrumental variable analysis is a causal estimation approach that uses variation in treatment or exposure induced by a qualifying instrument to estimate an effect under explicit identification and modelling assumptions.
• Interrupted Time SeriesA quasi-experimental design analysing observations before and after an intervention, testing for a significant change in level or trend at that point.
• Matching-Adjusted Indirect ComparisonA method comparing treatments from separate trials with different populations by reweighting one trial's patient data to match the other trial's reported characteristics.
• Net BenefitNet benefit places health gains and costs on one scale by valuing health at a cost-effectiveness threshold, or directly in money, and subtracting costs.
• Propensity ScoreThe estimated probability that an individual would receive a particular treatment given their observed characteristics, used to balance confounders in observational studies.
• Propensity Score MatchingPropensity score matching is an observational study design method that pairs treated and untreated individuals with similar probabilities of treatment given measured baseline characteristics to estimate a treatment contrast in the population retained by the matches.
• Real-World DataRoutinely collected information about patient health or healthcare delivery that may support research and decision making when fit for a defined purpose.
• Real-World EvidenceClinical evidence about a medical product's use, benefits, or risks derived from analysing real-world data, unlike evidence from a controlled trial.
• Simulated Treatment ComparisonA method comparing treatments from separate trials by fitting a regression model to one trial's data to predict outcomes for the other's population.
• Treatment EffectThe change in an outcome attributable to a specific treatment, distinguished from changes that would have occurred regardless due to other factors.