Assessing the impact of social prescribing on health service utilisation: Evidence from the UK

Drawing on longitudinal data from 4,547 participants in the UK, this study utilizes Bayesian growth curve modelling to demonstrate that social prescribing significantly reduces health service utilization across primary and secondary care settings, with substantial decreases observed in GP visits, A&E attendances, and hospital admissions.

The Gist

The Big Picture: A New Kind of Prescription

Imagine you go to your doctor because you're feeling down, lonely, or stressed. In the old days, the doctor might just hand you a pill or tell you to "take it easy."

But in the UK, doctors are now trying something different called Social Prescribing (SP). Instead of just medicine, they write a "prescription" for things like joining a gardening club, taking an art class, volunteering, or going for a group walk. The idea is that fixing your social life and hobbies can actually fix your health, too.

This study asked a very practical question: Does this "social prescription" actually stop people from needing to visit the doctor, the emergency room, or the hospital as much?

The Experiment: Tracking the Journey

The researchers didn't just guess; they looked at the digital records of 4,547 people across the UK who had been "socially prescribed" between 2017 and 2025.

Think of this like tracking a group of hikers. Before they started their new hiking path (the social activities), the researchers counted how many times they had to stop at the "help station" (the doctor/hospital). Then, they waited a few months and counted again after they had been hiking for a while.

The Results: A Major Drop in Visits

The findings were surprisingly positive. After people started their social activities, they stopped visiting the healthcare system as much. Here is the breakdown:

• The Doctor's Office (GP): People visited their family doctor 53% less often.
  • The Analogy: Imagine you used to call your mechanic every week because your car was making weird noises. After you joined a car club and learned how to fix small things yourself, you only call the mechanic once every two months. That's what happened here.
• The Emergency Room (A&E): Visits dropped by 62%.
  • The Analogy: It's like a fire department seeing fewer emergency calls because the neighborhood learned how to put out small kitchen fires before they became big blazes.
• Hospital Stays: Admissions dropped by 61%.
  • The Analogy: Fewer people needed to be "admitted" to the hospital, meaning their health was stable enough to stay at home.

Why Did This Happen?

The paper suggests a few reasons, which act like different gears in a machine:

1. The "Social Glue": Loneliness is bad for health. By connecting people to their communities, SP acts like a strong glue that holds their mental and physical health together, preventing small problems from becoming big emergencies.
2. The "Self-Manager" Effect: When people feel supported, they become better at managing their own health. They might exercise more, eat better, or just feel less anxious, so they don't need to run to the doctor for every little worry.
3. The "Right Place, Right Time" Effect: Sometimes, a lonely person calls the doctor just to have a chat. With SP, they get that chat from a "Link Worker" or a community group, so they don't clog up the doctor's schedule.

Who Benefits?

The study found that almost everyone benefited, regardless of whether they were rich or poor, young or old, or living in a city or the countryside.

• The Exception: The only group that didn't see a huge drop in hospital visits was the very elderly (70+). The researchers think this is because older adults often have complex, long-term health issues that are harder to fix just by joining a club. It's like trying to fix a very old, rusted engine with a new paint job; the paint helps, but the engine still needs major work.

The Catch (Limitations)

The authors are honest about what they didn't do:

• No Control Group: They didn't have a "control group" of people who didn't get social prescribing to compare against. It's like testing a new fertilizer on a garden but not having a second garden to see if the plants would have grown anyway. However, they used very smart math to account for this.
• Self-Reporting: The data came from people telling the researchers how many times they visited the doctor, not from official hospital logs. People might forget or want to sound "better" than they are.
• Short Term: They only looked at a 3-to-12-month window. We don't know if these benefits last for years.

The Bottom Line

This study is like a strong signal flare in the fog. It suggests that Social Prescribing works. By connecting people to their communities, we aren't just making them happier; we are actually giving the healthcare system a much-needed break.

It's a win-win: People feel less lonely and healthier, and the doctors and hospitals have more time to focus on the people who truly need urgent medical care. The researchers are now calling for more studies to prove this long-term, but the initial results look very promising.

Technical Summary

1. Problem Statement and Context

The UK National Health Service (NHS) faces significant strain due to demographic shifts, population aging, and the aftermath of the COVID-19 pandemic. A substantial portion of General Practitioner (GP) time (estimated at 19%) is dedicated to non-medical issues rooted in social determinants of health. To address this, Social Prescribing (SP) was introduced in the NHS Long Term Plan (2019). SP connects patients to non-clinical community support (e.g., arts, exercise, volunteering) via Link Workers.

While SP aims to improve holistic well-being, there is a critical evidence gap regarding its impact on health service utilisation. Theoretical models suggest SP could either:

• Reduce utilisation by improving health behaviors and self-management.
• Increase utilisation by enhancing "patient activation," leading to more proactive engagement with preventative services.
• Reflect increasing need (where referrals are a symptom of deteriorating health rather than a cause of change).

Previous studies have been limited by small sample sizes, lack of peer review, weak methodological designs, and a focus on pre-2019 pilot projects that may not reflect current system-wide implementation. This study aims to provide robust, large-scale evidence on whether SP reduces primary and secondary care utilisation.

2. Methodology

Data Source and Sample:

• Platform: Data was drawn from Access Elemental, the most widely adopted digital SP platform in the UK.
• Cohort: The analytical sample consisted of 4,547 patients from over 120 SP sites across the UK.
• Inclusion Criteria: Patients with repeated measures of health service utilisation at two time points (pre- and post-intervention) within a 3–12 month window (2017–2025). The mean follow-up was 5.3 months.
• Exclusion: Patients with multiple referrals used their first record; ethnicity was excluded due to >83% missing data.

Variables:

• Outcomes (Self-reported, last 3 months):
  1. GP visits (count data).
  2. Accident & Emergency (A&E) attendances (count data with excess zeros).
  3. Hospital admissions (count data with excess zeros).
• Covariates: Gender, Age, Area Deprivation (IMD quintiles), Urbanicity, and Referral Route (Medical vs. Non-medical).

Statistical Analysis:

• Approach: Bayesian growth curve modelling was employed to analyze person-specific and average changes between time points.
• Models:
  • GP Visits: Fitted using Poisson models (unconditional and conditional).
  • A&E and Hospital Admissions: Fitted using Hurdle Lognormal models. These models handle excess zeros by separating the process into two parts:
    1. Hurdle component: Predicting the probability of zero vs. non-zero usage.
    2. Lognormal component: Predicting the magnitude of usage given it is non-zero.
• Implementation: Models used non-informative priors, 4 Markov chains, 4,000 iterations (1,000 burn-in), and were executed in R 4.4.1 using the brms package.

3. Key Results

Overall Reduction in Utilisation:
The study found consistent, significant reductions across all three outcome measures following SP intervention:

• GP Visits: Decreased by an average of 1.0 visit per person over 3 months (95% CI: -1.07 to -0.95). This represents a 53.1% reduction (from a baseline of 1.89 to 0.89 visits).
• A&E Attendance: Decreased by 0.04 admissions per person (95% CI: -0.06 to -0.03), a 62.6% reduction.
  • The probability of any A&E attendance dropped significantly (from 94.4% to 97.8% probability of no attendance).
  • Among those with non-zero attendance, frequency decreased by 5.1%.
• Hospital Admissions: Decreased by 0.03 admissions per person (95% CI: -0.03 to -0.02), a 61.7% reduction.
  • The probability of no admission increased from 96.7% to 98.6%.
  • Among those with non-zero admissions, frequency decreased by 8.2%.

Heterogeneity and Subgroup Analysis:

• Demographics: There was limited evidence that the magnitude of reduction varied significantly across sociodemographic groups (gender, age, deprivation), suggesting broad applicability.
• Urbanicity: Rural residents showed slightly larger reductions in GP visits compared to urban residents, though the absolute difference was negligible.
• Referral Route: Medical referrals showed a significant increase in the probability of no A&E attendance, whereas non-medical referrals showed little change (likely a "floor effect" due to lower baseline usage).
• Age:
  • <30 years: Showed reductions in the frequency of hospital admissions (preventative effect).
  • 70+ years: Showed little change in the probability of admission, likely due to the complexity of long-term conditions in older adults.
• Baseline Correlation:
  • GP: Positive correlation between baseline visits and reduction (higher baseline = smaller reduction).
  • Secondary Care (A&E/Hospital): Strong negative correlation (higher baseline = greater reduction), suggesting SP is particularly effective for high-frequency secondary care users.

4. Key Contributions

1. Scale and Scope: This is one of the first large-scale national analyses of SP using data from over 120 sites and 4,500+ patients, moving beyond small, localized pilot studies.
2. Methodological Rigor: The use of Bayesian growth curve modelling with hurdle lognormal distributions allows for robust handling of zero-inflated count data and accounts for individual heterogeneity and random effects, addressing limitations in previous literature.
3. Differentiated Outcomes: The study distinguishes between the probability of using a service and the frequency of use, providing a nuanced view of how SP impacts different patient subgroups (e.g., those who never use A&E vs. those who do).
4. Evidence for Policy: The findings support the inclusion of SP in the 2025 NHS 10-Year Plan by demonstrating potential cost-saving mechanisms through reduced secondary care utilisation.

5. Significance and Limitations

Significance:
The study provides robust empirical evidence that Social Prescribing is associated with substantial reductions in health service utilisation. While the absolute reductions in secondary care (A&E/Hospital) appear small in raw numbers, the authors argue that given the high unit costs of these services and the large volume of annual referrals, the aggregate economic impact could be substantial. The findings suggest SP effectively addresses the social determinants driving unnecessary healthcare use, particularly for patients with complex needs.

Limitations:

• Study Design: The lack of a control group means the study cannot definitively prove causality. The pre-post design is vulnerable to regression to the mean, although the Bayesian models attempted to mitigate this by modeling baseline-dependent growth.
• Data Source: Reliance on self-reported data introduces potential recall bias or social desirability bias.
• Selection Bias: The analytical sample had overrepresentation of rural areas, Scotland, and Northern Ireland compared to the full Access Elemental cohort.
• Timeframe: The analysis focuses on a 3-month window post-intervention, lacking long-term cumulative data on utilisation trends beyond one year.

Conclusion:
The authors conclude that SP is a viable intervention for reducing pressure on primary and secondary care systems. However, they emphasize the need for future high-quality observational or controlled studies to confirm causality, explore mechanisms of action, and optimize SP components for specific populations.