Replication of Fraser & Frazer (2020): GP Prescribing and Deprivation in Northern Ireland

Using May–October 2025 Prescribing Data (6-month period)

Analysis date: March 2026 | Data sources: OpenDataNI GP Prescribing, NISRA Postcode Directory, NIMDM 2017

1. Introduction

This report replicates the methodology of Fraser & Frazer (2020), which explored the relationship between socioeconomic deprivation and GP prescribing patterns in Northern Ireland using open-source data. The original study used prescribing data from May–October 2019 (6 months); this replication uses the same months from 2025 (May–October) to examine whether the same patterns persist six years later.

The original study found statistically significant correlations between ward-level deprivation (measured by the Northern Ireland Multiple Deprivation Measure 2017) and prescribing rates across all 12 prespecified BNF sections, with higher prescribing in more deprived areas.

2. Methods

2.1 Data Sources

Prescribing data: GP prescribing data for May–October 2025 (6 months) from OpenDataNI (22,695,322 total items across 305 practices).

Practice data: GP Practice Reference File (January 2026) providing practice postcodes, LCG membership and registered patient numbers (2,075,632 total registered patients).

Deprivation data: Northern Ireland Multiple Deprivation Measure 2017 (NIMDM 2017), providing deprivation ranks for 4,537 Small Areas.

Geography: NISRA Central Postcode Directory mapping practice postcodes to Small Areas (SA2011) and Electoral Wards (WARD1992).

2.2 Linkage and Aggregation

Following the original methodology, practice postcodes were mapped to electoral wards (1992 boundaries) via the NISRA Central Postcode Directory. Ward-level deprivation was calculated by averaging the NIMDM 2017 Multiple Deprivation Measure ranks of all Small Areas within each ward, then ranking wards from 1 (most deprived) to 178 (least deprived). Prescribing data was aggregated to ward level by summing items across all practices within each ward, and rates were expressed as items per 1,000 Potential Registered Patients (PRPs).

2.3 Statistical Analysis

Kendall's tau rank correlation was used to assess the relationship between ward deprivation rank and prescribing rate for each BNF section and individual drug. A negative tau indicates higher prescribing in more deprived wards. Following Fraser & Frazer (2020), statistical significance was assessed at p < 0.0005 (Bonferroni-corrected). Twelve BNF sections were prespecified based on clinical relevance and the original study.

3. Results

3.1 Overview

Key Finding: Of the 12 prespecified BNF sections, 9 showed statistically significant correlations between prescribing rate and ward deprivation (p < 0.0005). All 12 showed negative Kendall's tau values, indicating consistently higher prescribing in more deprived wards.

Analysis summary: 22,695,322 prescription items | 178 wards | 305 practices | 2,075,632 registered patients | Total actual cost: £239,395,149

3.2 Prespecified BNF Sections

Table 1 shows Kendall's tau correlations for the 12 prespecified BNF sections. D1 and D10 refer to the mean prescribing rate (items per 1,000 PRP) in the most and least deprived deciles respectively.

BNF	Section Name	N	Total Items	Kendall's τ	P-value	Sig.	D1 Mean	D10 Mean
2.12	Lipid-Regulating Drugs	178	1,370,684	-0.2450	1.21e-06	***	752.7	541.1
2.2	Diuretics	178	433,524	-0.1739	5.70e-04		220.4	173.8
2.5	Hypertension & Heart Failure	178	1,121,891	-0.1878	1.97e-04	***	579.9	440.1
3.1	Bronchodilators	178	448,906	-0.3021	2.14e-09	***	266.7	153.9
3.2	Corticosteroids (Respiratory)	178	508,833	-0.3128	5.71e-10	***	295.3	181.5
4.1	Hypnotics & Anxiolytics	178	728,826	-0.1773	4.42e-04	***	406.2	268.2
4.10	Substance Dependence	178	27,885	-0.2625	1.97e-07	***	15.0	10.7
4.2	Antipsychotics	178	329,855	-0.2715	7.43e-08	***	194.7	114.5
4.3	Antidepressants	178	2,091,529	-0.2819	2.31e-08	***	1216.7	792.6
5.1	Antibacterials	178	645,642	-0.1000	4.75e-02		325.7	277.1
6.1	Diabetes	178	1,034,110	-0.2390	2.18e-06	***	546.2	399.1
7.3	Contraceptives	178	114,846	-0.1310	9.45e-03		57.3	51.9

*** = significant at p < 0.0005 (Bonferroni-corrected). Negative τ indicates higher prescribing in more deprived wards.

Figure 1. Kendall's tau correlations between ward deprivation rank and prescribing rates for the 12 prespecified BNF sections. Bubble size proportional to total items prescribed. Red = statistically significant (p < 0.0005).

Figure 2. Mean prescribing rates (items per 1,000 PRP) comparing the most deprived decile (D1, red) with the least deprived decile (D10, blue) for each prespecified BNF section.

3.3 Scatter Plots: Selected BNF Sections

Figure 3. Ward-level prescribing rates vs deprivation rank for six key BNF sections. Each point represents one ward. Red line = linear trend. *** = p < 0.0005.

3.4 All BNF Sections

Table 2 shows correlations for the top 40 BNF sections ranked by Kendall's tau (most negative first). Highlighted rows are the 12 prespecified sections.

BNF	Section Name	Prespec.	Total Items	τ	P-value	Sig.
3.2	Corticosteroids (Respiratory)	Yes	508,833	-0.3128	5.71e-10	***
1.3	Antisecretory & Mucosal		1,435,427	-0.3035	1.80e-09	***
3.1	Bronchodilators	Yes	448,906	-0.3021	2.14e-09	***
4.3	Antidepressants	Yes	2,091,529	-0.2819	2.31e-08	***
4.2	Antipsychotics	Yes	329,855	-0.2715	7.43e-08	***
10.1	NSAIDs & DMARDs		533,261	-0.2669	1.22e-07	***
4.10	Substance Dependence	Yes	27,885	-0.2625	1.97e-07	***
4.8	Antiepileptics		526,776	-0.2522	5.79e-07	***
2.12	Lipid-Regulating Drugs	Yes	1,370,684	-0.2450	1.21e-06	***
3.7	Mucolytics		69,303	-0.2438	1.35e-06	***
1.2	Antispasmodics		126,430	-0.2419	1.63e-06	***
6.1	Diabetes	Yes	1,034,110	-0.2390	2.18e-06	***
3.3	Cromoglycate & Leukotriene		80,498	-0.2384	2.31e-06	***
4.6	Nausea & Vertigo		199,994	-0.2262	7.38e-06	***
9.6	Vitamins		590,359	-0.2258	7.65e-06	***
21.1	BNF 21.1		193,019	-0.2239	9.12e-06	***
2.4	Beta-Blockers		805,549	-0.2170	1.70e-05	***
1.1	Antacids & Simeticone		98,316	-0.2068	4.18e-05	***
2.9	Antiplatelet Drugs		534,565	-0.1990	8.02e-05	***
10.2	Corticosteroids (Musculoskeletal)		85,552	-0.1922	1.40e-04	***
2.6	Nitrates & Calcium Channel Blockers		829,429	-0.1919	1.43e-04	***
4.5	Appetite Suppressants		8,827	-0.1904	1.76e-04	***
2.5	Hypertension & Heart Failure	Yes	1,121,891	-0.1878	1.97e-04	***
4.7	Analgesics		1,479,569	-0.1820	3.10e-04	***
4.1	Hypnotics & Anxiolytics	Yes	728,826	-0.1773	4.42e-04	***
2.2	Diuretics	Yes	433,524	-0.1739	5.70e-04
1.5	Chronic Bowel Disorders		53,932	-0.1568	1.89e-03
3.4	Antihistamines & Allergic Emergenci		371,669	-0.1565	1.93e-03
13.5	Psoriasis & Eczema		44,642	-0.1515	2.67e-03
0.0	BNF 0.0		114	-0.1512	1.64e-01
13.9	Shampoos		47,688	-0.1476	3.45e-03
6.2	Thyroid Disorders		587,547	-0.1472	3.53e-03
20.5	BNF 20.5		1,955	-0.1428	1.30e-02
13.12	Antiperspirants		2,062	-0.1425	6.78e-03
1.9	Drugs affecting intestinal secretio		21,490	-0.1415	5.05e-03
21.24	BNF 21.24		2,789	-0.1370	1.04e-02
12.2	Nasal		203,007	-0.1364	6.86e-03
21.32	BNF 21.32		2,761	-0.1331	1.40e-02
7.3	Contraceptives	Yes	114,846	-0.1310	9.45e-03
7.4	Urinary Disorders		345,758	-0.1296	1.02e-02

3.5 Individual Drug Analysis

Table 3 shows Kendall's tau correlations for 28 commonly prescribed drugs, grouped by therapeutic category.

Drug	Category	N	Total Items	τ	P-value	Sig.	Mean/1000
Metformin	6.1 Diabetes	178	321,484	-0.3134	5.27e-10	***	153.3
Mirtazapine	4.3 Antidepressants	178	258,246	-0.3093	8.77e-10	***	118.0
Salbutamol	3.1 Bronchodilators	178	343,923	-0.2711	7.74e-08	***	158.2
Sertraline	4.3 Antidepressants	178	541,263	-0.2639	1.70e-07	***	249.6
Amitriptyline	4.3 Antidepressants	178	351,620	-0.2631	1.84e-07	***	161.0
Fluoxetine	4.3 Antidepressants	178	204,405	-0.2607	2.38e-07	***	95.5
Atorvastatin	2.12 Lipid-Regulating Drugs	178	932,795	-0.2490	8.01e-07	***	444.5
Ramipril	2.5 Hypertension & Heart Failure	178	402,485	-0.1963	9.98e-05	***	194.7
Venlafaxine	4.3 Antidepressants	178	233,207	-0.1919	1.43e-04	***	107.1
Gliclazide	6.1 Diabetes	178	55,889	-0.1812	3.29e-04	***	27.7
Zopiclone	4.1 Hypnotics & Anxiolytics	178	173,552	-0.1811	3.32e-04	***	85.8
Sitagliptin	6.1 Diabetes	178	39,067	-0.1682	8.61e-04		18.9
Amlodipine	2.5 Hypertension & Heart Failure	178	434,507	-0.1661	9.94e-04		213.9
Amoxicillin	5.1 Antibacterials	178	136,762	-0.1577	1.77e-03		68.4
Trazodone	4.3 Antidepressants	169	11,692	-0.1551	2.76e-03		5.8
Bendroflumethiazide	2.2 Diuretics	178	80,257	-0.1270	1.18e-02		39.5
Flucloxacillin	5.1 Antibacterials	178	66,905	-0.1242	1.38e-02		33.7
Diazepam	4.1 Hypnotics & Anxiolytics	178	279,426	-0.1237	1.42e-02		135.6
Citalopram	4.3 Antidepressants	178	300,131	-0.1024	4.24e-02		141.6
Duloxetine	4.3 Antidepressants	178	127,102	-0.0912	7.06e-02		58.9
Rosuvastatin	2.12 Lipid-Regulating Drugs	178	128,269	-0.0893	7.67e-02		62.0
Simvastatin	2.12 Lipid-Regulating Drugs	178	172,157	-0.0695	1.68e-01		83.3
Co-Amoxiclav	5.1 Antibacterials	178	33,504	-0.0614	2.24e-01		16.7
Candesartan	2.5 Hypertension & Heart Failure	178	90,188	-0.0583	2.48e-01		43.9
Nitrofurantoin	5.1 Antibacterials	178	57,318	-0.0536	2.88e-01		28.3
Furosemide	2.2 Diuretics	178	165,474	-0.0317	5.30e-01		81.7
Losartan	2.5 Hypertension & Heart Failure	178	99,793	0.0349	4.90e-01		49.1

Figure 4. Individual antidepressant drugs: Kendall's tau correlation with ward deprivation rank. Red bars indicate statistically significant associations (p < 0.0005).

4. Comparison with Fraser & Frazer (2020)

Table 4 directly compares Kendall’s tau correlation coefficients from the original study (May–October 2019, 6 months, 174 wards) with this replication (May–October 2025, 6 months, 178 wards) for all 12 prespecified BNF sections.

BNF	Section	τ (2019)	Sig.	τ (2025)	Sig.	Change
2.12	Lipid-Regulating Drugs	-0.3054	***	-0.2450	***	↑ weaker
2.2	Diuretics	n/r	n/r	-0.1739		—
2.5	Hypertension & Heart Failure	-0.2318	***	-0.1878	***	↑ weaker
3.1	Bronchodilators	-0.4459	***	-0.3021	***	↑ weaker
3.2	Corticosteroids (Respiratory)	-0.3806	***	-0.3128	***	↑ weaker
4.1	Hypnotics & Anxiolytics	-0.1733	***	-0.1773	***	↔
4.10	Substance Dependence	-0.2373	***	-0.2625	***	↓ stronger
4.2	Antipsychotics	-0.3858	***	-0.2715	***	↑ weaker
4.3	Antidepressants	-0.3785	***	-0.2819	***	↑ weaker
5.1	Antibacterials	-0.1177		-0.1000		↔
6.1	Diabetes	-0.3004	***	-0.2390	***	↑ weaker
7.3	Contraceptives	-0.0705		-0.1310		↓ stronger

*** = significant at Bonferroni-corrected threshold (p < 0.0005). “n/r” = not reported in original paper. “Weaker” = correlation closer to zero (less negative); “Stronger” = further from zero (more negative).

Figure 5. Side-by-side comparison of Kendall’s tau coefficients from the original Fraser & Frazer (2020) study and this replication, for each prespecified BNF section. More negative values indicate stronger association between deprivation and prescribing.

4.1 Key Differences

The most striking finding is that while the direction of all associations is preserved (negative τ throughout), the magnitude of correlations is consistently weaker in 2025 compared to 2019. The original study found 8 of 12 prespecified sections significant at the Bonferroni-corrected threshold; this replication finds 9 of 12. The overall pattern of attenuation is notable.

Several specific differences stand out:

Bronchodilators (3.1) showed the strongest correlation in both analyses, but the magnitude has reduced from τ = −0.446 to τ = -0.302. This remains highly significant but may reflect changes in respiratory prescribing practice, including increased use of combination inhalers (which may be classified elsewhere) and the impact of COVID-19 on respiratory healthcare patterns.

Antipsychotics (4.2) showed a marked reduction from τ = −0.386 to τ = -0.272. This may partly reflect changes in antipsychotic prescribing guidance and greater movement toward shared care between primary and secondary services for serious mental illness.

Antibacterials (5.1) were weakly correlated in the original study (p = 0.021, not significant at corrected threshold) and remain non-significant (τ = -0.100). The further attenuation may reflect the impact of sustained antimicrobial stewardship programmes across NI, which aim to reduce unnecessary antibiotic prescribing regardless of area deprivation.

Contraceptives (7.3) were not significant in either analysis. In the original paper, the authors noted the difficulty of interpreting this in the context of Northern Ireland’s demographic and social factors.

4.2 Possible Explanations for Attenuation

The generally weaker correlations in 2025 could reflect several factors. First, COVID-19 and its aftermath may have disrupted established prescribing patterns. Second, targeted public health interventions and prescribing guidelines implemented since 2019 may have begun to reduce deprivation-related disparities in some areas. Third, changes in the practice landscape (mergers, closures, boundary changes) may have altered the relationship between practice location and catchment deprivation.

4.3 Individual Drug Patterns

Among individual antidepressants, mirtazapine (τ = -0.309) and amitriptyline (τ = -0.263) showed the strongest deprivation gradients, while citalopram and duloxetine did not reach significance. The original paper found metformin to have the strongest individual drug correlation of those plotted (τ = −0.372); this replication confirms that pattern (τ = -0.313), consistent with the well-documented socioeconomic gradient in type 2 diabetes prevalence.

Notably, the original paper found zopiclone did not correlate with deprivation (τ = −0.076); this replication also finds a non-significant result (τ = -0.181), which is interesting given that the broader hypnotic/anxiolytic class (4.1) does show a trend. This may reflect that zopiclone prescribing is driven more by individual patient factors (e.g. insomnia patterns) than by area-level deprivation.

4.4 Implications

These findings demonstrate that the socioeconomic gradient in GP prescribing in Northern Ireland has persisted from 2019 to 2025, though with some attenuation. The consistency of direction across all 12 prespecified sections suggests that structural factors related to deprivation continue to drive prescribing patterns. Areas of highest deprivation still have approximately 1.5–3 times the prescribing rates of the least deprived areas across most therapeutic categories.

5. Limitations

Primary limitation — practice postcode as a proxy for patient deprivation:

The most important limitation of both this analysis and the original study is the use of the GP practice postcode as a proxy for the deprivation level of the registered population. Deprivation is assigned based on where the practice building is located, not where its patients actually live. In reality, patients often register with practices outside their own ward — particularly in urban and peri-urban areas where practice catchments extend across ward boundaries. A practice located in a deprived ward may draw a substantial proportion of its patients from less deprived neighbouring areas, and vice versa. This means the assigned deprivation rank may not accurately reflect the socioeconomic profile of the practice’s patient list. This limitation will tend to attenuate the true relationship (diluting the signal), meaning the actual deprivation gradient in prescribing may be stronger than what we observe. Ideally, patient-level postcode data would allow individual deprivation scores to be assigned, but this is not available in the open-source prescribing data.

Other limitations:

• Deprivation measure: NIMDM 2017 was used for both the original study and this replication, now nearly nine years old. An updated deprivation measure might show different spatial patterns, particularly given the economic impacts of Brexit and COVID-19 on NI communities.

• Temporal scope: This replication uses a 6-month period (May–Oct 2025), matching the original’s 6-month period (May–Oct 2019), providing comparable prescribing volumes for analysis.

• Ward boundaries: 1992 ward boundaries were used for comparability with the original study. These historical boundaries may no longer reflect meaningful community units.

• Multi-practice wards: Where multiple practices share a ward, registered patient totals are summed. However, some wards contain practices whose combined lists substantially exceed the resident population, indicating patients travelling in from other areas. This further compounds the practice-postcode limitation above.

• Items vs Defined Daily Doses: Prescribing volume is measured in items rather than DDDs, which does not account for variation in dosage, formulation or duration of treatment.

• Ecological study design: As with the original, this is an ecological analysis and cannot infer individual-level causal relationships. Higher prescribing in deprived areas reflects a combination of genuine disease burden, health-seeking behaviour, prescribing culture, and access to services.

• GP prescribing only: Hospital prescriptions, dispensing by community pharmacists under Patient Group Directions, and private prescriptions are excluded, which may differentially affect deprived and affluent populations.