Medication persistence trajectories among individuals prescribed lipid‐lowering therapy in primary care settings: A retrospective cohort study

Abstract

Aims

This study aimed to identify distinct trajectories of lipid‐modifying medication (LMM) persistence over time and to explore characteristics associated with each pattern.

Methods

Using primary care data from IQVIA, we conducted a retrospective cohort study of adults prescribed LMMs between January 2015 and December 2017, with 5 years of follow‐up. Persistence was defined as ≥1 prescription every 6 months. Group‐based trajectory modelling identified medication persistence patterns; characteristics were compared using 1‐way ANOVA and chi‐squared tests.

Results

Among 51 504 individuals (mean age 62 years, 53% male), 4 distinct trajectories were identified: persistent use (PU, 20%), gradual decline (10%), rapid decline (29%) and early discontinuation (41%). Compared to those who discontinued early, individuals in the rapid decline, gradual decline and PU groups were older by 1.18 (95% confidence interval: 0.82–1.55), 2.61 (2.08–3.13) and 3.71 (3.30–4.12) years, respectively. Persistent users were more likely to have cardiovascular risk factors: a higher proportion of smokers (44.4 vs. 39.6%), elevated systolic blood pressure (≥140 mmHg: 36.9 vs. 32.9%) and reduced renal function (estimated glomerular filtration rate >45 mL/min/m²: 14.4 vs. 11.7%) compared to those who discontinued LMMs early. In contrast, the early discontinuation group had a greater proportion of metropolitan residents (76.1 vs. 69.2%) and individuals with elevated total cholesterol ratios (>4: 60.8 vs. 53.8%) than the PU group.

Conclusions

Distinct profiles across trajectories highlight the need for tailored interventions to improve long‐term medication use, particularly among younger, healthier individuals and those residing in metropolitan areas.

What is already known about this subject

• Use of lipid‐modifying medications declines substantially within the first 3 years following initiation. However, there is limited evidence on long‐term persistence in primary prevention cohorts managed in primary care.

• Persistence with lipid‐modifying therapy is often reported as a single overall percentage, which does not capture the diversity of medication‐taking behaviours within the population.

• While some patient characteristics have been linked to lipid‐modifying medication use, evidence on their long‐term associations across distinct patterns of medication‐taking behaviour remains scarce.

What this study adds

• This study identified 4 distinct patterns of lipid‐modifying medication use over 5 years: persistent use (20%), gradual decline (10%), rapid decline (29%) and early discontinuation (41%).

• Younger age, metropolitan residence and nonsmoking status were associated with early or gradual discontinuation, while individuals with cardiovascular risk factors were more likely to persist with treatment.

1. INTRODUCTION

Lipid‐modifying medications (LMMs), primarily statins, are recommended in clinical guidelines for individuals at high risk of cardiovascular diseases (CVD) or those with a history of CVD. ¹ , ² , ³ , ⁴ , ⁵ Despite their established safety and efficacy, statin use for primary prevention of CVD is suboptimal. ⁶ There is evidence that only 35% of patients in Canada and 21% in Germany remain persistent 3 years after initiating statin therapy. ⁷ , ⁸ In Australia, persistence with statins stands at 29% 3 years after initiation. ⁹ , ¹⁰ This limits the benefits of LMMs, aggravating the burden of CVD and necessitating actions to improve their use.

Evidence on long‐term persistence with LMM in primary prevention remains limited. A systematic review and meta‐analysis of 26 studies on statin persistence among older adults (age ≥65 years) included only 5 studies focused on primary prevention and found a 76% persistence rate at 1 year, with no studies reporting outcomes over 5 years. ¹¹ Moreover, traditional persistence measures such as time‐to‐discontinuation or proportion of days covered summarizes persistence into a single overall percentage, which fails to capture the variation in medication‐taking behaviour over time. In contrast, group‐based trajectory analysis (GBTA) accounts for this heterogeneity by identifying distinct, data‐driven subgroups with similar longitudinal patterns. ¹²

GBTA enables the detection of dynamic changes, the identification of associated patient characteristics and the generation of clinically meaningful, interpretable trajectories. ¹³ In a recent systematic review of 21 studies on medication adherence trajectories, only 6 studies focused on statins, with a maximum length of 2 years ¹⁴ Therefore, this study aimed to identify trajectories of medication persistence with LMMs over 5 years and examined demographic and clinical characteristics among new users of LMM in primary care settings using the GBTA approach.

2. METHODS

2.1. Study design, data source and cohort

This retrospective cohort study used data from the General Practice Electronic Medical Records dataset, representing approximately 12% of primary care practices in Australia. ¹⁵ De‐identified data were extracted by IQVIA, a health data analytics company, either as part of a benchmarking and quality improvement initiative (54%) or for research purposes (46%). The dataset included 235 306 adults (aged ≥18 years) with 1 505 453 LMM prescription records between 1 January 2013 and 31 March 2023. Incident users of LMMs were defined as individuals with no LMM prescriptions in the 2 years preceding their first recorded prescription. To ensure a consistent follow‐up duration for all incident users and facilitate valid comparisons, the follow‐up period for identifying persistence trajectories to LMMs was limited to 5 years from the date of the first LMM prescription. Consequently, the final cohort comprised individuals with at least 1 LMM prescription between January 2015 and December 2018. Further details are provided in Figure 1.

FIGURE 1.

Examples of 3 individuals (icons) initiating a lipid‐modifying medication as indicated by a round dot. A dashed line indicates a 2‐year lookback period for identification of incident users, while a solid line with an arrow indicates a 5‐year follow‐up period for identification of persistence trajectories to lipid‐modifying medications.

2.2. Outcome and covariates

The primary outcome was persistence with LMMs, defined as at least 1 primary care prescription record of an LMM every 6 months starting from the first prescription. This definition was chosen because prescriptions in Australian primary care settings typically include 5 repeats, and the available data lacked information on prescription refills, dosages or package sizes. LMMs were identified from the available medication records (analgesics, antimetabolites, and drugs used for cardiovascular diseases and diabetes) and included agents coded as C10 according to the Anatomical Therapeutic Chemical Classification System of the World Health Organization. ¹⁶

Individual characteristics were selected based on availability and existing evidence on predictors of discontinuation of LMM. ¹⁷ , ¹⁸ Sociodemographic covariates included age, sex and living area, categorized as metropolitan and regional/remote based on primary care practice location according to the Modified Monash Model classification. ¹⁹ Clinical covariates included smoking status, systolic blood pressure (SBP), relevant pathology test results for CVD, and medical history of CVD, hypertension, diabetes and hyperlipidaemia. Both SBP and history of hypertension were included, as patients may have elevated SBP without a recorded hypertension diagnosis at treatment initiation or, conversely, may have controlled SBP despite a history of hypertension.

Medical conditions were identified using diagnosis records (free text) and codes from the Australian Modification of the International Statistical Classification of Diseases and Related Health Problems, Tenth Revision. ²⁰ CVD conditions included angina pectoris, atrial fibrillation, coronary artery bypass graft, heart failure, myocardial infarction, percutaneous coronary intervention, peripheral artery disease and stroke.

SBP and pathology tests were dichotomized based on the Australian guidelines ²¹ , ²² , ²³ :

• SBP: ≥140 mmHg;

• Total cholesterol ratio: >4;

• Estimated glomerular filtration rate (eGFR): >45 mL/min/1.73 m²;

• Gamma‐glutamyl transferase (GGT): >45 U/L;

• Serum albumin at >45 g/L;

• Serum bilirubin at >20 μmol/L.

SBP, total cholesterol ratio, ²⁴ and GGT ²⁵ were included as established CVD risk factors, while eGFR, ²⁶ serum albumin ²⁷ and bilirubin ²⁸ were included as clinically relevant biomarkers also linked to CVD.

2.3. Statistical analysis

Cohort characteristics were summarized using descriptive statistics, and GBTA was used to identify medication persistence trajectories. In particular, clinical characteristics of the cohort were identified using records from 2 years before the first LMM prescription. GBTA is a data‐driven method that identifies subgroups of individuals based on distinct, statistically derived patterns of behaviour or outcomes within a cohort, rather than assuming a single average trajectory applies to all patients. ¹³ Moreover, sensitivity analyses were conducted, defining persistence as at least 1 prescription of an LMM within 3 months or identifying incident users as individuals with no records of an LMM prescription for 12 months before their first record of an LMM prescription. These additional sensitivity analyses provided insights into the robustness of GBTA findings under assumptions based on certain definitions used in this study. Since our dependent variable was binary, GBTA was conducted using logistic models ¹² , ²⁹ and comprised 2 stages.

In the first stage, a 2‐group quadratic model was employed, increasing the number of groups in the quadratic model to 7. The best model was selected based on: (i) distinct trajectories assessed by graphical presentations; (ii) the Bayesian information criterion (the closer to zero, the better); and (iii) a group membership >10% for each trajectory. Additionally, the entropy (the lower, the better) and statistical significance (P value) of each trajectory in the model were considered in the model selection. A second‐best model was also selected based on the above‐mentioned criteria, accepting a group membership of >5%, which was presented as supplementary material.

In the second stage, we explored varying polynomial orders across trajectory groups to improve model fit. In particular, the quadratic form of the selected model, which had P values >.05, was replaced with either zero‐order, linear, cubic, or quartic form. The characteristics of the selected model, including average posterior probabilities (APPs), odds of correct classification (OCC) and odds of correct classification based on posterior probability (OCCPP), were assessed to ensure model adequacy. Recommended thresholds were set at >0.7 for APP and >5 for OCC and OCCPP. Additionally, a mismatch between the proportions based on the assignments for the maximum posterior probability and the proportions based on the sums of the posterior probabilities was characterized.

Individual characteristics were compared across persistence trajectory groups using chi‐squared tests for categorical variables and 1‐way ANOVA (ANalysis Of VAriance) for age, followed by posthoc Tukey's Honestly Significant Difference test. Results were presented as mean difference with 95% confidence intervals (CIs), using the poorest persistence group as the reference. Two‐sided P‐values ≤.05 were considered statistically significant. All analyses and graphical presentations were performed using Stata Statistical Software, Release 18 (StataCorp LLC, College Station, TX, USA).

3. RESULTS

3.1. Cohort characteristics

In total, 51 504 incident users with 243 949 LMM prescriptions were included in the main analyses (Figure 2). The cohort had an average age of 62 years, with over half being male and nearly 3/4 residing in metropolitan areas (Table 1). Additionally, 2/5 were smokers, 1/10 had hypertension, over half had total cholesterol ratio >4 and 1/3 had SBP ≥ 140 mmHg. Although the cohort included incident users of LMMs, fewer than 5% had a documented history of hyperlipidaemia. Statins dominated prescribing patterns, comprising over 88% of all LMM use, whereas fibrates (3.6%) and other nonstatin LMMs (3.1%) were used much less frequently. Further details on pathology tests and CVD‐related diagnoses are provided in Table 1.

FIGURE 2.

The flowchart of individuals included in the analyses.

TABLE 1.

Demographic and clinical characteristics by 4 trajectory groups.

	Total		Early discontinuation		Rapid decline		Gradual decline		Persistent‐use		P value
Number of patients, n (%)	51 504	(100)	20 946	(40.6)	14 832	(28.8)	5344	(10.4)	10 382	(20.2)	‐
Age (years), mean [SD]	62	[13.4]	60.6	[14.3]	61.8	[13.3]	63.2	[12.8]	64.3	[11.3]	<.01
Male, n (%) †	27 348	(53.6)	11 077	(53.4)	7905	(53.7)	2818	(53)	5548	(54)	.64
Residency in metro, n (%)	37 782	(73.4)	15 949	(76.1)	10 885	(73.4)	3758	(70.3)	7190	(69.2)	<.01
Smoker, n (%) ‡	11 501	(41.8)	3949	(39.6)	3650	(42.3)	1406	(43.1)	2496	(44.4)	<.01
SBP ≥ 140 mmHg, n (%) §	4 078	(34.2)	1 824	(32.9)	1385	(34.8)	459	(35)	410	(36.9)	.03
Total cholesterol ratio>4, n (%) ¶	12 338	(58.6)	4292	(60.8)	3845	(59.7)	1471	(59.3)	2730	(53.8)	<.01
GGT > 45 U/L, n (%) #	1075	(24.9)	438	(25.1)	339	(25.8)	118	(24.9)	180	(22.6)	.43
Serum albumin >45 g/L, n (%) **	1931	(11.3)	804	(11.3)	551	(1.7)	192	(10.9)	384	(12.6)	.06
Serum bilirubin >20 μmol/L, n (%) ††	1156	(4.1)	398	(3.8)	345	(4.1)	146	(4.6)	267	(4.3)	.16
eGFR > 45 mL/min/m2, n (%) ‡‡	2840	(12.7)	966	(11.7)	830	(12.4)	347	(13.9)	697	(14.4)	<.01
History of:
Hypertension, n (%)	6350	(12.3)	2007	(9.6)	2010	(13.5)	859	(16.1)	1474	(14.2)	<.01
Diabetes, n (%)	3497	(6.8)	1002	(4.8)	1081	(7.3)	500	(9.4)	914	(8.8)	<.01
Hyperlipidaemia, n (%)	2523	(4.9)	716	(3.4)	791	(5.3)	368	(6.9)	648	(6.2)	<.01
CVD, n (%)	2137	(4.1)	622	3	691	(4.7)	320	(6.0)	504	(4.8)	<.01

Abbreviations: CVD, cardiovascular diseases incl. Angina pectoris, atrial fibrillation, coronary artery bypass graft, heart failure, myocardial infarction, percutaneous coronary intervention, peripheral artery disease and stroke; eGFR, estimated glomerular filtration rate; GGT, γ‐glutamyl transferase; metro, metropolitan area; SBP, systolic blood pressure; SD, standard deviation.

^* n (%) unless otherwise specified.

^† Excluding missing data for 446 individuals.

^‡ Excluding missing data for 24 008 individuals.

^§ Excluding missing data for 39 567 individuals.

^¶ Excluding missing data for 30 450 individuals.

^# Excluding missing data for 47 179 individuals.

^**Excluding missing data for 34 445 individuals.

^†† Excluding missing data for 23 134 individuals.

^‡‡ Excluding missing data for 29 218 individuals.

3.2. Long‐term persistence trajectories

Out of 7 quadratic models tested, a model with 4 distinct trajectories was chosen according to our selection criteria (Table 2). Replacing the nonsignificant quadratic form of the selected 4‐group model with linear, cubic or quartic forms did not improve the Bayesian information criterion. Four distinct trajectories of the selected model included early discontinuation (41%), rapid decline (29%), gradual decline (10%) and persistent‐use (20%), as shown in Figure 3 (95% CIs were too narrow to show, thus omitted). These proportions based on the assignments for the maximum posterior probability were similar to those based on the sums of the posterior probabilities. All other model characteristics were above the recommended thresholds (>0.7 for APP and >5 for OCC and OCCPP), as shown in Table 3. In summary, only 1/5 of the cohort persistently used LMMs for 5 years, nearly 2/5 discontinued immediately and another 2/5 stopped following their LMM prescription between 24 and 48 months.

TABLE 2.

Characteristics of group‐based trajectory models tested in a 2‐stage selection process.

Stage	Models	Bayesian information criterion	All P values <.05	Smallest group membership (%)	Entropy
1st	2 2	−195 751	Yes	35	0.922
	2 2 2	−181 207	No	22	0.877
	2 2 2 2	−179 013	No	11	0.802
	2 2 2 2 2	−177 935	No	5	0.801
	2 2 2 2 2 2	−177 745	No	0.8	0.806
	2 2 2 2 2 2 2	−177 917	No	0	0.828
2nd	2 0 2 2	−180 666	No	4	0.861
	2 1 2 2	−181 221	No	0	0.913
	2 3 2 2*	−181 103	‐	‐	0.904
	2 4 2 2	−179 024	No	11	0.802
	2 2 2 2	−179 013	No	11	0.802

^*Variance matrix was nonsymmetric or highly singular.

FIGURE 3.

Trajectories of lipid‐modifying medication use based on primary care prescriptions.

TABLE 3.

Characteristics of the selected group‐based trajectory model with 4 groups.

	Average posterior probabilities	Odds of correct classification	Odds of correct classification based on posterior probability	Proportions based on the assignments for the maximum posterior probability	Proportions based on the sums of the posterior probabilities
Early discontinuation	0.95	28.8	26.6	41%	43%
Rapid decline	0.79	9.2	10.6	29%	26%
Gradual decline	0.82	40.6	35.9	10%	12%
Persistent use	0.94	62.4	63.8	20%	20%

Additionally, trajectories identified for individuals who had no LMM prescription 24 months (n = 51 504) or 12 months (n = 77 187) before their first recorded LMM prescription were similar (Figure 3 and Figure S1). When medication persistence was defined as at least 1 prescription within 3 months instead of 6 months, trajectories showed lower probabilities of persistence and slightly different proportions (Figure S2). However, the number and shapes of trajectories remained similar when medication persistence was defined as at least 1 prescription within 3 months compared to 6 months. Furthermore, a 5‐group model, chosen as the second most meaningful, revealed a fifth trajectory labelled as an initial decline followed by an increase (Table S1 and Figure S3).

3.3. Individual characteristics across medication persistence trajectories

Compared to the early discontinuation group, individuals in the rapid decline, gradual decline and persistent use groups were older by 1.18 (95% CI: 0.82–1.55), 2.61 (2.08–3.13) and 3.71 (3.30–4.12) years, respectively (Table 1). The lower the persistence, the greater the proportion of the cohort residing in metropolitan areas or having total cholesterol ratio of >4. In contrast, the higher the persistence, the greater the proportion of smokers and of the cohort having SBP ≥ 140 mmHg or eGFR > 45 mL/min/m². There was no statistically significant difference across trajectories in sex, GGT > 45 U/L, serum albumin or bilirubin. History of hypertension, diabetes, hyperlipidaemia and CVD were more prevalent in the persistent use and gradual decline groups, compared to rapid decline and early discontinuation groups.

4. DISCUSSION

We provide novel evidence on 5‐year medication persistence trajectories to LMMs and their associated factors in a large cohort of LMM initiators treated in primary care. Four distinct and clinically meaningful trajectories included persistent use, gradual decline, rapid decline and early discontinuation. Only 1/5 of the cohort persistently used LMMs over 5 years, while 2/5 discontinued LMMs immediately and the remainder stopped following their LMM prescription within 48 months. Furthermore, younger age, having a total cholesterol ratio above the recommended level, residing in metropolitan area and nonsmoking were associated with discontinuation or decline in LMM use. In contrast, having SBP ≥ 140 mmHg, eGFR > 45 mL/min/m², and a history of diabetes, hypertension, hyperlipidaemia and CVD were more prevalent in the gradual decline and persistent use groups. This warrants further investigations to understand the reasons behind different persistence behaviours, guiding future interventions to improve LMM use.

Our finding that only 20% of LMM incident users maintained persistent use over 5 years aligns with previous research. A systematic review of 19 studies reported that the proportion of patients following their lipid‐lowering therapy ranged from 18 to 79%. ⁶ This low level of persistence is concerning, as discontinuation of LMM has been associated with adverse health outcomes. For example, statin use has been linked to a significantly reduced risk of all‐cause mortality (hazard ratio: 0.72, 95% CI: 0.66–0.76). ³⁰ Further research is warranted to explore the underlying reasons for LMM discontinuation and to inform strategies aimed at improving long‐term medication‐taking behaviour.

Although our findings regarding the shape and number of trajectories were consistent with previous research, ¹³ , ³¹ , ³² , ³³ , ³⁴ , ³⁵ , ³⁶ , ³⁷ , ³⁸ , ³⁹ the percentage of group membership across trajectories differed from earlier publications. Specifically, our findings on the percentage of individuals who persistently used LMMs were similar to some studies but contrasted with others. For example, the proportion of participants following LMM therapy ranged between 40 and 80% in several previous studies, ³³ , ³⁴ , ³⁵ , ³⁶ , ³⁷ , ³⁸ , ³⁹ but was similar to ours in a few other studies. ¹³ , ³¹ , ³² Contrary to our study, the percentage of those who discontinued immediately did not exceed 25% in previous studies. ¹³ , ³¹ , ³² , ³³ , ³⁴ , ³⁵ , ³⁶ , ³⁷ , ³⁸ , ³⁹ This variation may stem from differences in the types of statin users included in previous GBTA studies, which often focused on prevalent or incident users at specific ages. ¹³ , ³¹ , ³² , ³³ , ³⁴ , ³⁵ , ³⁶ , ³⁷ , ³⁸ , ³⁹ Additionally, prior studies were constrained by their 1–2‐year duration and broad participant exclusion criteria. Our findings make a distinctive contribution to the literature on long‐term use of LMMs.

Evidence on the association between patient characteristics and LMM persistence trajectories in primary prevention remains limited. Our finding that LMM persistence increased with age is consistent with results reported over a 3‐year period among 34 582 statin users using Australian national dispensing claims, ⁹ as well as other GBTA studies with follow‐up periods of up to 15 months, ¹³ , ³⁹ although these studies included both primary and secondary prevention cohorts. Conversely, several studies with shorter follow‐ups ≤1 year reported poor LMM use among older adults (≥80 years) with acute coronary syndrome or stroke, ³⁸ chronic heart disease ³⁴ and among new statin users with prior antihypertensive use. ³⁵ One study of 9430 individuals found poor use in both younger (<60 years) and older (≥80 years) individuals, ³⁷ and a systematic review of 9 reviews echoed these findings across all age groups. ⁴⁰ These mixed findings may be explained by variations in patient characteristics, sample sizes, medication types and follow‐up periods across GBTA studies in high‐income countries.

There is also limited evidence on how biomarkers are associated with long‐term persistence patterns of LMMs in individuals at risk of CVD. Some studies suggest that hypertension, diabetes and smoking are not associated with persistence or adherence trajectories over 9 months in people with coronary heart diseases, although these differ in setting and follow‐up. ³⁴ In contrast, a systematic review of 9 reviews found these factors associated with good statin use. ⁶ Our study contributes to the literature by identifying unique patient and clinical characteristics, including nearly all CVD risk factors, forming distinct risk profiles for long‐term persistence behaviours among LMM initiators.

Compared to traditional measures of medication persistence, such as time‐to‐discontinuation, proportion of days covered or binary persistent/nonpersistent classification, GBTA offers several advantages. It effectively captures the heterogeneity of medication‐taking behaviours by identifying distinct, data‐driven subgroups with statistically similar longitudinal patterns, rather than assuming a single average or imposing arbitrary thresholds. GBTA models the full trajectory of use over time, making it particularly well‐suited to detect dynamic changes such as early discontinuation, gradual decline or reinitiation. This approach also allows for the identification of individual and clinical characteristics associated with specific patterns of persistence, which is valuable for informing targeted interventions. Moreover, the resulting trajectories are visually interpretable and offer greater clinical insight than summary metrics alone, making GBTA a powerful method for understanding real‐world medication use.

4.1. Strengths and limitations

Compared to existing research, our study has several noteworthy strengths. Firstly, our data represented 12% of all primary care practices in Australia, providing a comprehensive representation of the adult population. Our study sample comprised all adults prescribed LMMs without restrictions based on specific patient characteristics, ensuring inclusivity and enhancing the generalisability of our findings. Secondly, the observational period spanned 7 years, significantly longer than the typical 1‐ or 2‐year follow‐ups in existing research related to LMMs. This extended timeframe allowed for a more comprehensive assessment of long‐term persistence trajectories. Thirdly, our analyses included clinical biomarker data, commonly excluded in medication persistence and adherence studies based on administrative linked datasets. Fourthly, our study included sensitivity analyses testing different definitions of medication persistence and incident users of LMMs. These strengths not only distinguish our study from existing research but also contribute to a more robust understanding of patient behaviour towards following LMM prescriptions. By encompassing a diverse sample and employing a longer observational period, our study offers valuable insights that can inform more effective interventions and strategies for improving long‐term medication persistence behaviours among statin initiators.

Our study had several limitations. The data did not allow us to identify individuals who changed their primary care practices. However, we assumed that medication cessation was more likely than a practice change, given that 72% of Australians attend only 1 primary care practice. ⁴¹ Additionally, the dataset lacked information on deaths, preventing us from distinguishing between individuals who discontinued medication due to mortality and those who stopped for other reasons. We assumed that primary care prescription records indicated actual medication use. However, sensitivity analyses using a 3‐month definition yielded similar trajectories.

In this study, we identified first‐time users of LMMs using a 2‐year look‐back period, under the assumption that these individuals would not have a prior history of hyperlipidaemia. However, despite this approach, approximately 5% of the cohort had a documented history of hyperlipidaemia, suggesting that a 2‐year look‐back period may be insufficient to capture all prior cases. This probably reflects individuals who were previously treated with LMMs before the look‐back window and subsequently restarted therapy. In addition, the dataset lacked variables potentially associated with LMM use such as income, medication dosage, side effects and costs. Despite these limitations, our study offers valuable insights and highlights the need for future research employing more precise persistence measures and improved data collection in primary care settings to enhance reliability and validity.

5. CONCLUSION

This study revealed substantial variation in long‐term persistence with LMMs, with the majority of individuals exhibiting early or rapid discontinuation. Distinct demographic and clinical profiles were associated with each trajectory, highlighting the need for tailored interventions to improve long‐term medication use, particularly among younger individuals living in metropolitan areas and those with relatively fewer CVD risk factors, such as smoking, elevated blood pressure and diabetes.

AUTHOR CONTRIBUTIONS

Concept, design, data analysis and interpretation: Orman, Ademi and Talic. Acquisition of data: Talic, Liew, Reid, Ademi, Bell, Zomer, Berkovic, Ilomaki, Lybrand and Schoeninger. Manuscript drafting: Orman, Ademi, Bell and Talic. Obtaining funding: Talic, Liew, Reid, Ademi, Bell, Zomer, Berkovic and Ilomaki. All authors provided a revision of the manuscript for critically important intellectual content. All authors read and approved the final version of the manuscript.

CONFLICT OF INTEREST STATEMENT

All authors report no potential conflicts of interest.

CODE AVAILABILITY

The Stata codes used for the data analyses in this study will be made available upon reasonable request to the corresponding author.

Supporting information

FIGURE S1 Four trajectories of lipid‐modifying medication use based on primary care prescriptions with a 12‐month lookback period.

FIGURE S2 Four trajectories of lipid‐modifying medication use defined as at least 1 primary care prescription within 3 months.

FIGURE S3 Five trajectories of lipid‐modifying medication use based on primary care prescriptions.

TABLE S1 Characteristics of the group‐based trajectory model with 5 groups.

Orman Z, Koh JW, Trin C, et al. Medication persistence trajectories among individuals prescribed lipid‐lowering therapy in primary care settings: A retrospective cohort study. Br J Clin Pharmacol. 2025;91(11):3257‐3265. doi: 10.1002/bcp.70170

DATA AVAILABILITY STATEMENT

The data underlying this article will be shared on reasonable request to the corresponding author.