Running ehrQL

You can run ehrQL in two places:

• on your own computer, where you can try out ehrQL against dummy data, and test that your analysis code runs correctly
• on an OpenSAFELY backend database, to use ehrQL with real data

Running ehrQL on your own computer against dummy data

There are three ways to run ehrQL on your own computer against dummy data:

1. with the sandbox mode, to try out ehrQL in an interactive Python console
2. as a standalone action, to test your dataset definition, via opensafely exec
3. as the first step in an OpenSAFELY pipeline, to test the whole pipeline, via opensafely run

1. Running ehrQL interactively via the ehrQL sandbox mode

The ehrQL sandbox lets you try out ehrQL queries against dummy data in an interactive Python console.

The ehrQL sandbox can be useful to:

• become familiar with how ehrQL works
• develop more complicated ehrQL queries against dummy data

The ehrQL sandbox can help minimise constant re-editing and re-running of your dataset definitions by allowing you to interactively query some dummy data.

You need to have a directory containing CSV files of dummy data. If you followed the steps in Installation and setup, you will have a suitable directory of CSV files at learning-ehrql/example-data.

To start the sandbox, from the learning-ehrql directory, run:

opensafely exec ehrql:v0 sandbox example-data

You will now be in a session with an interactive Python console, and you should see something like this:

Python 3.11.3 (main, Apr  5 2023, 14:15:06) [GCC 9.4.0] on linux
Type "help", "copyright", "credits" or "license" for more information.
(InteractiveConsole)
>>>

The >>> is the Python prompt for user input. When you see this, you can input Python statements, press the return key, and if the statement returns a value, it will be displayed below your input. For example if you type 1 + 1 and press the return key, you should see:

>>> 1 + 1
2
>>>

To use ehrQL, you'll first need to import the tables that you want to interact with:

>>> from ehrql.tables.beta.core import patients, medications

Now, you can inspect the contents of these tables, by entering the names of the tables:

>>> patients
patient_id        | date_of_birth     | sex               | date_of_death
------------------+-------------------+-------------------+------------------
0                 | 1973-07-01        | female            | 2015-09-14
1                 | 1948-03-01        | male              | None
2                 | 2003-04-01        | male              | None
3                 | 2007-06-01        | female            | None
4                 | 1938-10-01        | male              | 2018-05-23
5                 | 1994-04-01        | female            | None
6                 | 1953-05-01        | male              | None
7                 | 1992-08-01        | female            | None
8                 | 1931-10-01        | female            | 2017-11-10
9                 | 1979-04-01        | male              | None
>>> medications
patient_id        | row_id            | date              | dmd_code
------------------+-------------------+-------------------+------------------
0                 | 0                 | 2014-01-11        | 39113611000001102
1                 | 1                 | 2015-08-06        | 39113611000001102
1                 | 2                 | 2018-09-21        | 39113311000001107
1                 | 3                 | 2020-05-17        | 22777311000001105
3                 | 4                 | 2022-11-09        | 22777311000001105
4                 | 5                 | 2017-05-11        | 39113611000001102
5                 | 6                 | 2017-07-11        | 3484711000001105
5                 | 7                 | 2019-07-06        | 39113611000001102
7                 | 8                 | 2021-01-27        | 3484711000001105
9                 | 9                 | 2015-03-14        | 3484711000001105

If you see an error when trying to access these tables, check that you have the dummy data files in the correct location.

And you can enter ehrQL to perform queries, such as this one:

>>> patients.date_of_birth.year
0 | 1973
1 | 1948
2 | 2003
3 | 2007
4 | 1938
5 | 1994
6 | 1953
7 | 1992
8 | 1931
9 | 1979

Or this one:

>>> patients.date_of_birth.is_on_or_before("1999-12-31")
0 | True
1 | True
2 | False
3 | False
4 | True
5 | True
6 | True
7 | True
8 | True
9 | True

Or this one:

>>> medications.where(medications.dmd_code == "39113611000001102").sort_by(medications.date).first_for_patient()
patient_id        | date              | dmd_code
------------------+-------------------+------------------
0                 | 2014-01-11        | 39113611000001102
1                 | 2015-08-06        | 39113611000001102
4                 | 2017-05-11        | 39113611000001102
5                 | 2019-07-06        | 39113611000001102

Can you work out what these do?

When things go wrong

If you enter some invalid ehrQL, you will see an error message:

>>> medications.where(medications.date >= "2016-01-01").sort_by(medications.dat).first_for_patient()
Traceback (most recent call last):
  File "<console>", line 1, in <module>
AttributeError: 'medications' object has no attribute 'dat'

Can you work out what this is telling us?

Refer to the catalogue of errors for details of common error messages and what they mean.

Exiting the sandbox

To exit the sandbox, type exit() and then press the return key

2. Running ehrQL as a standalone action via opensafely exec

To actually run your ehrQL queries against real patient data, you need to write a dataset definition and save it in a file.

But first, while you are developing an ehrQL query, you can run your dataset definition against dummy data to produce an output file that you can inspect.

Copy and paste the following dataset definition into a new file called dataset_definition.py and save it in your learning-ehrql directory:

from ehrql import Dataset
from ehrql.tables.beta.core import patients, medications

dataset = Dataset()

dataset.define_population(patients.date_of_birth.is_on_or_before("1999-12-31"))

asthma_codes = ["39113311000001107", "39113611000001102"]
latest_asthma_med = (
    medications.where(medications.dmd_code.is_in(asthma_codes))
    .sort_by(medications.date)
    .last_for_patient()
)

dataset.med_date = latest_asthma_med.date
dataset.med_code = latest_asthma_med.dmd_code

Can you work out what the dataset definition will generate?

Make sure you save the file!

From the learning-ehrql directory, use the command below to run your dataset definition with ehrQL.

opensafely exec ehrql:v0 generate-dataset dataset_definition.py --dummy-tables example-data --output output/dataset.csv

ehrQL dataset definitions are written in Python. But, unlike typical Python code, we instead run the dataset definition via the OpenSAFELY CLI. The OpenSAFELY CLI internally uses a correctly configured version of Python to run the dataset definition.

What each part of this command does

• opensafely exec ehrql:v0 uses the OpenSAFELY CLI to run ehrQL. The v0 after the : refers to the version of ehrQL being used.
• generate-dataset instructs ehrQL to generate a dataset from the dataset definition.
• dataset_definition.py specifies the filename of the dataset definition to use.
  • The dataset definition file is in the directory that we are running opensafely exec so we do not need to specify the full path to the file in this case.
• --dummy-tables example-data specifies that the dummy CSV input data is in the example-data directory.
  • If the --dummy-tables option is omitted, randomly generated data will be used instead.
• --output output/dataset.csv specifies the path to the output CSV file.

What you should see when you run the command

You should see output displayed similar to this:

2023-04-19 08:53:41 [info     ] Compiling dataset definition from dataset_definition.py [ehrql.main]
2023-04-19 08:53:41 [info     ] Generating dummy dataset       [ehrql.main]
2023-04-19 08:53:41 [info     ] Reading CSV data from example-data [ehrql.main]
2023-04-19 08:53:41 [info     ] Building dataset and writing results [ehrql.main]

The date and time you see will differ from that here.

The output file

The output will be stored in a file called dataset.csv in the output directory.

The file will contain the following CSV data:

patient_id,med_date,med_code
0,2014-01-11,39113611000001102
1,2018-09-21,39113311000001107
4,2017-05-11,39113611000001102
5,2019-07-06,39113611000001102
6,,
7,,
8,,
9,,

The bottom 4 rows in the generated dataset show that there are 4 patients in the defined population that do not have any record for the medications specified in the dataset definition.

Try running the ehrQL dataset definition again, modifying the command to remove the --dummy-tables example-data. This gives you a random data output, instead of one based on the sample dummy data that you downloaded previously.

When things go wrong

If your dataset definition contains some invalid ehrQL, an error message will be displayed on the screen.

This is one example:

$ opensafely exec ehrql:v0 generate-dataset dataset_definition.py --dummy-tables example-data --output output/dataset.csv
2023-04-21 17:53:42 [info     ] Compiling dataset definition from dataset_definition.py [ehrql.main]
Failed to import 'dataset_definition.py':

Traceback (most recent call last):
  File "/workspace/dataset_definition.py", line 10, in <module>
    dataset.med_date = latest_asthma_med.dat
                       ^^^^^^^^^^^^^^^^^^^
AttributeError: 'medications' object has no attribute 'dat'

Refer to the catalogue of errors for help with interpreting error messages.

3. Running ehrQL in an OpenSAFELY pipeline via opensafely run

To run your ehrQL queries as part of an OpenSAFELY pipeline with opensafely run, you need to have a file called project.yaml.

There is considerably more technical detail on the project pipeline in the OpenSAFELY documentation.

Copy the following into a file called project.yaml in your learning-ehrql directory:

version: '3.0'

expectations:
  population_size: 1000

actions:
  generate_dataset:
    run: ehrql:v0 generate-dataset dataset_definition.py --dummy-tables example-data --output output/dataset.csv.gz
    outputs:
      highly_sensitive:
        cohort: output/dataset.csv.gz

  summarise_dataset:
    run: python:latest summarise_dataset.py
    needs: [generate_dataset]
    outputs:
     moderately_sensitive:
        cohort: output/summary.txt

Users already familiar with the OpenSAFELY research template may notice that the research template already includes a basic project.yaml file that can be edited. Here, for the purposes of this tutorial, to skip setting up the template, we create this file entirely by hand.

The project.yaml file defines two actions: generate_dataset and summarise_dataset. Each of these actions specifies one or more outputs.

The definitions of "highly sensitive" and "moderately sensitive" are explained in the project.yaml documentation.

The generate_dataset action's run: command should look familiar from the previous section. However, note that the --output path is now to a compressed CSV file (dataset.csv.gz).

We recommend the use of compressed CSV files when generating a dataset definition as part of an OpenSAFELY pipeline.

summarise_dataset uses a Python script called summarise_dataset.py. Copy the following into a file called summarise_dataset.py in your learning-ehrql directory.

import pandas as pd

dataframe = pd.read_csv("output/dataset.csv.gz")
num_rows = len(dataframe)

with open("output/summary.txt", "w") as f:
    f.write(f"There are {num_rows} patients in the population\n")

Even if you don't know how to use pandas, can you guess at what this code might do before you run the OpenSAFELY project?

From the learning-ehrql directory, use the command below to run all of the actions in project.yaml:

opensafely run run_all

If is this is the first time you have used opensafely exec, the OpenSAFELY CLI may fetch some other Docker images (python and busybox) needed to run the action.

What you should see when you run the command

You should see in the logs output displayed similar to this:

$ opensafely run run_all

Running actions: generate_dataset, summarise_dataset

jobrunner.run loop started
generate_dataset: Preparing your code and workspace files
...
summarise_dataset: Extracting output file: output/summary.txt
summarise_dataset: Finished recording results
summarise_dataset: Completed successfully
summarise_dataset: Cleaning up container and volume

=> generate_dataset
   Completed successfully

   log file: metadata/generate_dataset.log
   outputs:
     output/dataset.csv.gz  - highly_sensitive

=> summarise_dataset
   Completed successfully

   log file: metadata/summarise_dataset.log
   outputs:
     output/summary.txt  - moderately_sensitive

Some of the middle lines of this log have been omitted.

The output files

The generate_dataset action will generate a compressed CSV file called dataset.csv.gz in the output directory. If you unzip this, you should see the same output as the previous example.

The summarise_dataset action will generate a small text file called summary.txt in the output directory. This will tell you how many patients are in your population.

Running ehrQL on an OpenSAFELY backend database

Once you are happy with your ehrQL queries and any analysis code, you can submit your project to run against real data in an OpenSAFELY backend database.

To submit your project to run against real data, refer to the existing documentation on using the OpenSAFELY jobs site.

You will require approval for an OpenSAFELY project, before you can submit your project to the jobs site.

Questions

• Why would you use the ehrQL sandbox?
• Which opensafely command would you use to run just a dataset definition as a single action?
• Which opensafely command would you use to run an entire OpenSAFELY project consisting of multiple actions?