Appendix 3 – Reporting information for selected analytical methods | NICE real-world evidence framework | Guidance

Corporate document

Appendix 3 – Reporting information for selected analytical methods

**Guide to reporting on selected analytical methods**
Method	Description	Reporting information
Direct or indirect standardisation	Methods to increase comparability of exposure groups in terms of selected covariates	Standard reference population (description) Covariates used for standardisation
Stratification	Dividing the data into subsets, or strata for analysis	Covariate definition of strata Number of observations in each stratum Descriptive statistics and results within each stratum
Matching	Matching individuals with the same or similar characteristics	Variables used for matching Matching algorithm Matching caliper (if relevant) Matching ratio Number matched and number excluded
Propensity score (general)	Estimate of probability of receiving a particular intervention; range of methods available (below)	Model used to estimate propensity scores (such as logistic or multinomial) Covariates used and how they were included in the model Propensity score distribution before and after adjustments (for example, pre- and post-matching) N/% contributing to matched, trimmed, truncated or weighted analyses Diagnostic checks for any statistical analysis done See Tazare et al. 2022 for reporting of high-dimensional propensity score models
Propensity score (stratification)	Patients grouped into strata (for example, deciles) based on propensity score and stratum-specific effects aggregated	How strata are defined Trimming and whether applied before or after strata defined Tables for stratified population characteristics
Propensity score (weighting)	Weights attached to individuals based on inverse of propensity scores	How weights are calculated Whether and how weights are trimmed, truncated or stabilised Tables for unweighted and weighted population characteristics Mean and distribution of weights
Propensity score (matching)	Matches individuals with similar propensity scores	Matching algorithm used including caliper and scale Matching ratio (such as fixed 1:1 or variable 1:5) Tables for unmatched and matched population characteristics
Multivariable regression adjustment (includes using propensity scores)	Statistical models comparing outcomes as a function of the intervention and covariates	Type of model (such as linear regression or Poisson) Covariates used and how they were included Diagnostic checks
Instrumental variable analysis	Exploits external variation in exposure across people or over time using an 'instrument'. An instrumental variable is associated with the intervention but is otherwise unrelated to the outcome.	Type of model (such as 2-stage least squares) and diagnostic checks Strength of association between instrument and intervention (for example, odds ratio, risk difference) Theoretical justification that the instrument does not affect the outcome except through the intervention and that the instrument does not share any causes with the outcome Tables with distribution of population characteristics across levels of the instrument and intervention For binary outcomes, exposures and instruments, table of the frequencies of each combination of instrument, treatment, and outcome See Swanson and Hernán 2013 for reporting by specific causal effects in instrumental variable analysis and their dependent assumptions (for example, monotonicity) The results of falsification tests: see Labrecque and Swanson 2018 for specific examples
Interrupted time series	Individuals or groups are used as their own controls and observed over multiple time points. Effects are observed by comparing outcome trends in the time period before and after intervention.	Type of model (such as segmented linear regression with ordinary least squares regression) Study time period and time intervals Pre-specification of point of intervention effect (for example, explanation needed if point of analysis is not point of intervention delivery) Number of pre-intervention, post-intervention, and between-intervention data (time) points, and the data points contributing to forecasting Table comparing participant characteristics and missing data across each group analysed (for example, before and after intervention and for defined subgroups) Table and graph showing outcomes across time (that is, pre- and post-intervention trend) Results of diagnostic checks (for example, for autocorrelation, stationarity, seasonality, model specification checks) and any adjustments made Results of falsification tests (for example, the use of pseudo start periods before intervention delivery)