Skip to main content

Permitted Methods Policy

1: Purpose

This policy outlines the requirement for all research conducted within the OpenSAFELY Service to comply with the permitted analytical methods of NHS England. It contains the following sections for reference:

  • Purpose
  • Scope
  • Policy requirements 
  • Updates to this policy

2: Scope

This policy outlines the analytic methods that are currently supported and not supported within OpenSAFELY due to resources required for compute and output checking. The policy is approved by NHS England, and managed in collaboration with the Bennett Institute for Applied Data Science.

3: Policy requirements

Factors informing which methods are permitted:

  • Availability and cost of compute resource;
  • Complexity of output-checking requirements for each method;
  • Privacy risks;
  • Stakeholder support (including patients and GPs).

Statistical methods:

The OpenSAFELY service permits most established analytical methods, including those listed below:

  • Descriptive statistics (counts, proportions, means, quantiles, etc)
  • Basic statistical tests (e.g., t-test, chi-squared tests)
  • Generalised linear models (GLMs), including linear, logistic and Poisson regression
  • Survival analyses (e.g., Kaplan-Meier estimates, Cox regression, parametric survival modelling)
  • Traditional time series analysis (e.g., ARIMA, Exponential Smoothing, STL)
  • Visualisation of non-patient-level summary statistics, including data smoothing (e.g., histograms, time-series, forest plots)
  • Structural equation models

The following table lists the analytic methods not currently supported by the OpenSAFELY service. OpenSAFELY studies must not use any libraries, scripts, or run any code that uses these analytic methods:

  • Neural networks (including deep NNs, CNNs, RNNs, ANNs)
  • Support vector machines (SVM)
  • Random forests
  • Gradient boosting machines
  • Unsupervised clustering algorithms (e.g., k-means clustering)
  • Advanced time series models (e.g., LSTM, GRU, TCN)
  • Natural language processing models (e.g., BERT, GPT)
  • Reinforcement learning models (RLMs)
  • Adversarial learning models (ALMs)
  • Large language models (LLMs)
  • Generative artificial intelligence (AI) models

Project Leads must familiarise themself with this policy and ensure that only outputs from supported methods are requested for release.

Exceptions:

Users are not permitted to use methods on the not-supported list, even if they are not planning to release the outputs, without exceptional prior permission.

If a user judges that one of these methods is necessary for their project – for example, using such models for propensity matching, or inverse propensity weighting – with no requirement to release the outputs, or performance statistics, then an exception to this policy may be granted by the OpenSAFELY team. 

Users wishing to use one of these methods should discuss this with their co-pilot in the first instance, who will be able to offer advice before a formal request is submitted. It is essential that no work using these methods is started prior to the user receiving formal written approval (via email) from the OpenSAFELY team.

User import and linking of data:

Users are only permitted to process and analyse data made available through the OpenSAFELY service. Users must not attempt to link these datasets with any individually identifiable data, either from other NHSE or non-NHSE datasets.

Users are not permitted to import data into the service, either for linkage or for analysis within their wider project (i.e., no .csv or other files may be uploaded into an OpenSAFELY GitHub repo). 

4: Updates to this policy

As OpenSAFELY’s features and resources change over time, it is possible that it will become possible to support additional methods.

Any changes to this policy, including the addition of new methods, will be decided by NHS England, in discussion with the Bennett Institute and the OpenSAFELY Steering Group, and communicated to users through the OpenSAFELY website. 

Last updated: 17 October 2025