Technical annex Download

3 Sampling

3.1 Sample overview

The 2025 survey adopted the same sampling approach as the 2024 survey. The sample was designed to ensure that, as far as possible, a minimum of 100 responses per practice and 200 responses per Primary Care Network (PCN) would be received, providing confidence in the overall results and allowing subgroup analysis where appropriate. This resulted in an issued sample size of around 2.72 million patients who were invited to take part.

Patient information was obtained for each practice using registration records held on the Personal Demographics Service (PDS) database maintained by NHS England. The sampling procedure involved two distinct stages. Firstly, NHS England provided an anonymised list of patients for sample size determination and individual patient selection. The selected anonymous records were returned to NHS England, and a second file containing the contact details of the selected patients was provided.

3.2 NHS England population extraction procedure

As in previous years, NHS England provided a file of anonymous patient data for all eligible patients who reside in England or Wales and were registered with a practice in England. The file contained patient NHS numbers, practice code, patient gender, patient age band, and patient postcode.

A sample of patients was then drawn at practice level, as detailed below.

3.3 Patients eligible for the survey

Patients were eligible for inclusion in the survey if they had a valid NHS number, had been registered with a GP practice continuously for at least six months at the point of selection, and were 16 years of age or over. Note that prior to 2018 the survey had only been open to those aged 18 or over.

3.4 Practices included in the survey

The list of practices to be included was taken from the NHS England system and comprised all practices that had eligible patients, as defined above, where the practice had not previously opted out of the survey as they felt it was inappropriate to their patient population. The number of practices with eligible patients for the 2025 survey was 6,244, but between sample selection and mail-out, some practices became ineligible for the survey. In total, patients in 6,230 different practices were sent questionnaires, and at least one completed questionnaire was received from patients in 6,215 practices.

3.5 Sample size calculation

Prior to 2022, the practice sample size was based on aiming for a set confidence interval for each practice. Since 2022, the sample size has been determined to deliver at least 100 responses in each practice and 200 responses in each Primary Care Network (PCN), where possible, and 720,000 responses overall.

The sample design involved a proportionately stratified, unclustered sample, which was drawn for each practice. The sample for each practice is stratified by age, gender, and postcode. The number of patients initially selected for inclusion in the sample for each practice (the ‘issued sample’) was determined by the following components:

• the number of cases required to deliver 100 responses per practice and 200 responses per PCN; and

• the proportion of patients included in the issued sample predicted to respond to the survey.

These components were combined to determine the issued sample size in each practice, PCN and nationally, as follows:

The above figure shows the calculation for the issued sample: the number of responses aimed for at practice, PCN and national level is divided by the proportion of the issued sample predicted to respond.

The proportion of issued sample predicted to respond was estimated using the actual response rate for those practices who took part in the 2024 GPPS and was set at 28% for practices new to the survey or to whom fewer than 100 surveys were issued in the 2024 GPPS. This is to prevent unrealistically high or low response rates being used for new and very small practices.

3.6 Adjustments to response rate estimates

To prevent issuing very large numbers of questionnaires in practices which had very low response rates in 2024, a minimum response rate of 12.6% was assumed. To ensure that a reasonable number of questionnaires were sent to practices with very high response rates, on the other hand, a maximum assumed response rate of 50.4% was set.

3.7 Patient sample selection

The pseudonymous patient data sent by NHS England was sorted within each practice by gender then age band. The required number of patients per practice was selected on a ‘1 in n’ basis, and the unique survey reference numbers returned to NHS England.

3.8 Personal data extractions

On receipt of the selected records, NHS England extracted the contact details for each of the sampled patients. The extracted file contained, for each patient, their NHS number, patient name, practice code, address, month and year of birth, gender, and mobile number (if available). Where a selected patient had become ineligible since the provision of the anonymous data, an “exclusion” record was sent instead, containing the unique survey serial number and no other details.

3.9 Sample cleaning and exclusions

Checks were made on the supplied names, mobile numbers and addresses to remove inappropriate records. These checks included:

• duplicates (identified by patient NHS number).

• incorrect mobile numbers (not starting with 07 and/or not 11 digits long).

• non-address details or other inappropriate information contained in the address. This could include:

  • key safe numbers, telephone numbers and other numerics not related to the address.
  • unexpected words or phrases in the name or address (including “unknown”, “homeless”, “deceased”, “test”, etc.).
  • incomplete addresses.

The final selected sample was also checked against the Ipsos “Do Not Contact” and GPPS opt-out lists, to ensure that those who have previously expressed a wish not to be contacted by Ipsos, or specifically requested to be removed from future waves of GPPS, are not included.

For the SMS reminders, there were around 23,500 duplicate mobile numbers (where more than one person had the same mobile number listed). Of these, more than 22,500 were included twice and analysis suggested that they were mostly young people and an adult. As young people tend to be underrepresented in GPPS, where a text message included the patient’s name, duplicate messages were sent as it would be possible for the recipient to determine which link corresponded with which survey.

All sampled patients from all practices were then randomly sorted before being allocated a 12-character alphanumeric ID (to ensure there was no link between reference numbers and practices).

3.10 Final mailed sample after cleaning and exclusions

The final number of patients to whom questionnaires were sent after all sample cleaning had been finished was 2,721,415.

3.11 Movers checks

During fieldwork, in order to optimise the delivery of the final mailing packs, the GPPS sample was checked against a commercial database that identifies people who have moved.

A total of around 2,220,000 records were checked (removing all responders, opt outs and mail returned to sender from the initial sample). This identified around 260,000 (c11.5%) potential movers, i.e., people who were very likely to no longer be living at their registered address. In order to save on printing and postage costs, these people were removed from the final physical mailing but continued to be included in subsequent SMS contacts.