Chronic care provision in general practices and association with patient level outcomes: a nationwide cohort study

Abstract

Background

General practice provides long-term care for most people living with long-term conditions, but the impact of generic chronic care provision on patient outcomes has not been examined on a national level. We aimed to investigate whether the provision of chronic care services in general practice is associated with potentially inappropriate medications (PIMs) and potentially preventable hospitalisations in listed patients.

Methods

Nationwide cohort study using linked health registry data covering 4.42 million patients (18 + years) listed with general practices in Denmark in 2019 (n = 1769). The exposure was the patient’s practice listing. Practices were grouped evenly into low, medium, or high level of service provision (chronic care consultations, chronic care procedures, and daytime consultations) after adjustment for patient case-mix and multimorbidity. Sub-group analyses were based on list size, morbidity load, deprivation score, and urbanisation. The outcomes at patient level were number of patient days with PIMs (modified STOPP-START criteria) and number of potentially preventable hospitalisations.

Results

In practices providing high levels of chronic care consultations, the listed patients had a 1.2% lower risk of PIMs compared to the medium-level group (incidence rate ratio [IRR] 0.988, 95% confidence interval [CI] 0.977 to 0.999, corresponding to 3600 fewer patient years of PIMs per year) and an IRR of 0.964 (95% CI 0.927 to 1.002) for potentially preventable hospitalisations. In practices providing high levels of chronic care procedures, patients had a 1.7% lower risk of PIMs (IRR 0.983, 95% CI 0.972 to 0.993, 5500 fewer patient years of PIMs) and an 8.6% lower risk of potentially preventable hospitalisations (IRR 0.914, 95% CI 0.879 to 0.950, 3700 fewer potentially preventable hospitalisations per year). High levels of daytime consultations were associated with higher risk of PIMs, but not with potentially preventable hospitalisations. We found an inverse dose–response relationship between chronic care provision and adverse outcomes. The findings were stable between different practice characteristics and patient populations.

Conclusions

Patients experienced fewer potentially inappropriate medications and potentially preventable hospitalisations if listed at a general practice with high chronic care provision, regardless of other practice characteristics.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12916-025-04239-z.

Background

Healthcare systems are under significant pressure due to demographic changes, with population ageing and increased prevalence of chronic conditions [1–3]. In Denmark, general practitioners (GPs) provide community-based care for most patients with long-term conditions and act as gatekeepers to the hospital sector. Chronic care consultations are defined as guideline-based GP services offered for chronic conditions. In these consultations, disease status, test results, and medications are reviewed, and a treatment plan is agreed. Routine chronic care check-ups and medication reviews in general practice may reduce the need for specialist treatment in hospitals, reduce ambulatory care sensitive admission, improve appropriateness of prescribing and disease-specific patient outcomes [4–7], and ensure continuity of care, which is known to improve health outcomes [8–11].

Extensive research has explored disparity in health care [12] and for the provision of chronic care services in general practice [13]. Variation may reflect differences in the population served, but it may be unwarranted if not attributed to patient characteristics. Unwarranted variation could indicate suboptimal quality or costly inefficient care, which may lead to potentially preventable hospital admissions or inappropriate medications [14–18]. In primary care, chronic care variation has been linked to health inequalities and the presence of inverse care law mechanisms, i.e. the availability of good medical care tends to vary inversely with the need for it in the population served [19, 20]. However, although unwarranted variation is undesirable from a clinical point of view, its historical presence represents a valuable scientific resource. Because GP listing is considered random in relation to the practice level of chronic care service, the practice variation in chronic care provision can be used as an instrument to investigate patient outcomes of this natural experiment. Such variation analysis may reduce bias from the relationship between treatment indication and patient outcomes [21, 22]. To our knowledge, the impact of practice variation in intensity of chronic care provision on patient outcomes has not been evaluated at a population level.

In this study, we calculated the use of chronic care services for each general practice in Denmark correcting for characteristics of the listed patients; the aim was to investigate whether the practice-specific utilisation of chronic care services was associated with potentially inappropriate medications and potentially preventable hospitalisations in the listed patients. We also aimed to examine whether the magnitude of such association depended on practice and patient characteristics, including the level of multimorbidity.

The overall hypothesis was that a high level of chronic care provision would be associated with positive health outcomes.

Methods

Study population and setting

We performed a population-based cohort study, including all Danish residents above 18 years of age who were listed with a general practice and had been residing in Denmark for at least five consecutive years. The data was linked through the unique identification number assigned to all residents in Denmark to obtain individual-level data from Danish national registers [23, 24]. The study period started on 1 January 2019 and lasted until death, emigration, or end-of-study on 31 December 2019, whichever came first. This period was chosen to avoid the influence of the coronavirus pandemic.

The Danish universal healthcare system is mainly publicly funded, and GPs work as independent primary care contractors in a remuneration system based on a mix of per capita and fee-for-service payments [25]. Nearly all residents (99%) are listed with a practice with one or more GPs. Patients can change to another practice for a small fee (if the list is open), or for free if they move to another area. Citizens have free access to medical services, including acute and chronic care services by GPs, private practice specialists, and the hospital sector [24]. GPs provide ordinary daytime consultations (physical attendance), e-mail consultations, telephone consultations, and home visits. Additionally, GPs provide specific chronic care services, including chronic care consultations (annual consultation dedicated to a specific chronic condition and often including a patient health assessment, medication review, lifestyle talk, and goal setting). These consultations are often preceded by chronic care procedures, e.g. paraclinical measurements of blood-glucose level, electrocardiograms, lung function tests, and at-home measurements of blood pressure. Healthcare performance measures and service provision rates for practices are not publicly available.

Data sources

Danish registers provided complete and validated high-quality data on age, sex, civil status, vital status (Danish Civil Registration System) [23], ethnicity, household disposable income, educational attainment, employment status (Statistics Denmark) [24], prescription medicines (Danish National Prescription Register) [26], primary care service provision (Danish National Health Service Register) [27], outpatient clinic and discharge ICD-10 diagnoses, and hospital admissions (Danish National Patient Register and Danish Psychiatric Central Register) [28].

The Danish Multimorbidity Index algorithm provided information on 39 long-term physical and mental conditions [29]. The Patient List Database was used to create the link between patients and their general practice; GP clinic identifier (provider number) was shared for partnership practices with more than one GP (approx. 70% of clinics). Clinics with less than 500 listed patient years were excluded as they might represent newly started or liquidated practices, or a special administrative unit.

Exposure

The exposure for each patient was the frequency of chronic care services provision in the general practice, in which they were listed. The exposure had two dimensions, chronic care consultations and chronic care procedures, which were studied separately. The service provision was calculated based on remuneration codes (see coding definitions in Additional file 1: Table S1) [13] and practice-specific propensity to provide services was calculated as the ratio between observed number of services and expected number of services according to a prediction model based on register-based information on morbidities and sociodemographic factors for each patient in the practice. The case-mix was defined by baseline patient characteristics (details in Statistical analyses section below). The level of provided daytime consultations was also included as an overall estimate of activity and between-practice variation.

Outcomes

We assessed two outcomes at patient level: the number of days with potentially inappropriate medications (PIMs) and the number of potentially preventable hospital admissions based on ambulatory care sensitive conditions (ACSCs).

PIMs were defined as days of exposure to potentially inappropriate medications, i.e. inappropriate drug-drug and drug-disease combinations that would prompt treatment discontinuation (potential overtreatment) or combinations that would suggest medication initiation (potential undertreatment) based on a modified version of the STOPP/START criteria [11, 15]. These criteria have been adapted to Danish registers by combining data on redeemed prescriptions and diagnoses, which has resulted in an algorithm that identifies periods of time when an individual was subject to PIMs [30] (PIM coding definitions in Additional file 1). Patients could contribute with PIMs time more than once if having multiple concurrent PIMs.

ACSCs were defined as conditions for which hospitalisation might be preventable if optimal first-line treatment is provided in primary care [14]. Potentially preventable hospitalisations included eight ACSC definitions from the Agency for Healthcare Research and Quality and four ACSCs based on previous Danish studies: angina without concomitant cardiovascular procedures, chronic obstructive pulmonary disease exacerbation, chronic heart failure exacerbation, diabetes with short-term complications, diabetes with long-term complications, uncontrolled diabetes, hypertension, appendicitis with perforation, bacterial pneumonia, diabetes-related lower extremity amputations, urinary tract infections, and adult asthma exacerbations [14, 31, 32]. Hospital diagnoses were used to identify ACSC-related hospitalisations.

Statistical analyses

In a preliminary analysis step, we constructed multivariate Poisson models to predict the number of each type of service based on time-at-risk and baseline covariate status (age group, sex, ethnicity, cohabitation status, household disposable income, employment status, education attainment, and indicators for each of the 39 conditions in the multimorbidity index, coding definitions in Additional file 1: Tables S2 and S3) for each patient in the population. Second, we summed the observed and the predicted number of services for all patients listed with each practice. Subsequently, we divided these two aggregate measures to obtain a practice-specific observed-to-expected (O/E) ratio to indicate whether a practice delivered more (O/E > 1) or less (O/E < 1) services than expected given the composition of their patient population. The O/E ratios (one per exposure per practice) were ranked and divided into tertiles (i.e. low, medium, or high level of provided care). The expected and the observed number of services were analysed using a jack-knife type approach, i.e. the contribution of each patient’s service code outcome was subtracted from the practice score.

In the main analysis, we estimated the association between practice provision tertile and patient risk using multivariate Poisson models producing patient-level incidence rate ratios (IRRs) with 95% confidence intervals (CIs) calculated using cluster robust variance estimation with practice as the cluster unit. Patient risk was quantified as the number of days with PIMs (START, STOPP, and total) and the number of potentially preventable hospitalisations, respectively. These analyses were also adjusted for all baseline patient characteristics.

In a pre-planned supplementary sub-group analysis of the relation between practice characteristics and chronic care provision, included practices were divided into three equally sized groups based on their patient list size, overall morbidity load, socioeconomic deprivation score, and degree of urbanisation. The morbidity load for each practice was calculated as the mean number of chronic conditions per patient (captured for each patient by the multimorbidity index) in all the listed patients. The deprivation score for each practice was calculated with the Danish Deprivation Index score (version 2), which included educational level, labour market attachment, ethnicity, income and assets, and cohabitation status of the patient case-mix [33] (see details in Additional file 1: Table S4). Urbanisation degree was based on the number of listed patients living in urban areas.

The estimated absolute number of outcomes associated with being listed in a high-providing practice or a low-providing practice compared to a medium-providing practice was calculated in two steps. First, the observed outcomes for each practice group were divided by its adjusted IRR to produce estimated outcomes if treated as the medium-providing practice group. Second, the difference between the observed outcomes and the estimated outcomes (if treated as medium-providing practice) was expressed as excess days with PIMs and excess potentially preventable hospitalisations, respectively.

All analyses were performed as complete case analysis in Stata 18, College Station, TX. The reporting of this study followed the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) guidelines [34].

Patient and public involvement

The Deep End Denmark organisation, representing GPs in deprived areas, provided input to the analyses. No patients were involved in the design, the conduct, or the interpretation of the study. The published results will be disseminated to patients, GPs, lay people, and policymakers in Denmark through media, academic institutions, and professional bodies.

Results

During the study period, 1769 practices participated with a total of 4.35 million at-risk patient years (see sample characteristics in Table 1). A total of 1.02 million chronic care consultations and 1.76 million chronic care procedures were provided. In the same period, the study population had 32.5 million general daytime consultations.

Table 1.

Study population characteristics

Characteristic		Risk time (years)	%	Unique persons
Age	18–29	805,555	18.52	817,037
	30–39	613,876	14.11	621,221
	40–49	729,275	16.76	736,145
	50–59	775,165	17.82	783,167
	60–69	647,013	14.87	655,700
	70–79	533,781	12.27	545,143
	80–89	206,264	4.74	216,594
	90–99	38,285	0.88	43,187
	+ 100	835	0.02	1100
Sex	Female	2,212,481	50.86	2,245,982
	Male	2,137,568	49.14	2,173,312
Ethnicity	Danish	3,883,247	89.27	3,942,304
	Western	159,277	3.66	163,999
	Non-western	307,525	7.07	312,991
Cohabitation status	Single	1,655,722	38.06	1,691,989
	Married	2,035,850	46.80	2,062,081
	Cohabitating	658,477	15.14	665,224
Household disposable income	1st quintile	602,361	13.85	616,077
	2nd	764,844	17.58	783,411
	3rd	892,587	20.52	906,928
	4th	1,010,786	23.24	1,021,947
	5th quintile	1,079,470	24.82	1,090,931
Employment status	Working	2,690,836	61.86	2,715,941
	Not working	1,659,213	38.14	1,703,353
Educational attainment	≤ 10 years	1,139,981	26.21	1,163,466
	> 10 and ≤ 15 years	2,045,812	47.03	2,073,831
	> 15 years	1,078,649	24.80	1,093,025
	Missing	85,606	1.97	88,972

There was a factor three difference in the unadjusted observed rate of chronic care consultation between high-providing and low-providing practices (Additional file 1: Table S5). The adjusted variation in service provision level is visualised in Fig. 1, with O/E ratio cut-points for the patients listed in high-, medium-, and low-providing practices. In unadjusted numbers, more chronic care consultations were provided per person-year in practices with high morbidity load and in practices with a high number of patients, but not in practices located in deprived areas (Additional file 1: Table S6).

Fig. 1.

General practice variation in service provision. O/E: observed/expected. The x-axis represents the patients in the practices, sorted and listed with the lowest O/E ratio to the left

The practice level of provided chronic care consultations was inversely associated with PIMs in a dose–response relationship. Patients in high-providing practices were subject to a 1.2% lower risk of PIMs (IRR 0.988, 95% CI 0.977–0.999), and patients in low-providing practices were subject to a 3.3% higher risk of PIMs (IRR 1.033, 95% CI 1.022–1.044) compared to patients in medium-providing practices (Fig. 2). We observed no significant association with potentially preventable hospitalisations (high provision: IRR 0.964, 95% CI 0.927–1.002; low provision: IRR 0.991, 95% CI 0.954–1.029).

Fig. 2.

Incidence rate ratios of patient outcomes by general practice service provision

Similarly, the practice level of provided chronic care procedures was inversely associated with PIMs and with potentially preventable hospitalisations in a similar dose–response relationship. Patients in high-providing practices were subject to a 1.7% lower risk of PIMs (IRR 0.983, 95% CI 0.972–0.993) and an 8.6% lower risk of potentially preventable hospitalisations (IRR 0.914, 95% CI 0.879–0.950) compared to patients in medium-providing practices. Conversely, patients in low-providing practices were subject to a 5.1% higher risk of PIMs (IRR 1.051, 95% CI 1.040, 1.062) and a 6.8% higher risk of potentially preventable hospitalisations (IRR 1.068, 95% CI 1.028–1.111) compared to patients in medium-providing practices (Fig. 2).

High provision of general daytime consultations was associated with higher risk of PIMs, but otherwise we observed no associations between provision level and outcomes (Fig. 2).

PIMs comprised both undertreatment and overtreatment. The pattern was similar for both categories, but higher IRR estimates were seen in the overtreatment category with wider confidence intervals (Fig. 3).

Fig. 3.

Incidence rate ratios of potentially inappropriate medications according to service provision in general practice. Undertreatment represents the modified START criteria, and overtreatment represents the modified STOPP criteria of potentially inappropriate medication

Excess outcomes

In absolute terms, practices providing high levels of chronic care consultations accounted for 3600 fewer patient years of PIMs than would have been expected if their listed patients had the same PIMs incidence as patients in medium-providing practices. Correspondingly, the low-providing practices accounted for 10,300 more patient years of PIMs.

Similarly, practices providing high levels of chronic care procedures accounted for 5500 fewer patient years of PIMs and 3700 fewer potentially preventable hospitalisations per year. Practices providing low levels of chronic care procedures accounted for 15,200 more patient years of PIMs and 2700 more potentially preventable hospitalisations per year (Table 2).

Table 2.

Absolute and excess patient outcomes according to service provision in general practice

Outcome by service	Provision level tertile	Observed (× 1000)	Total risk time (1000 years)	Rate	IRR (95% CI)	Estimated if as medium (× 1000)	Excess outcomes*
Potentially inappropriate medication
Chronic care consultations	High	296.2	1449.9	0.204	0.988 (0.977, 0.999)	299.8	− 3600
	Medium	315.9	1471.2	0.215	Ref	Ref	Ref
	Low	325.2	1428.9	0.228	1.033 (1.022, 1.044)	314.9	10,300
Chronic care procedures	High	309.9	1458.7	0.212	0.983 (0.972, 0.993)	315.4	− 5500
	Medium	313.9	1454.4	0.216	Ref	Ref	Ref
	Low	313.6	1436.9	0.218	1.051 (1.040, 1.062)	298.4	15,200
Daytime consultations	High	308.1	1446.9	0.213	1.019 (1.008, 1.030)	302.4	5700
	Medium	310.2	1451.6	0.214	Ref	Ref	Ref
	Low	319.1	1451.5	0.220	1.007 (0.996, 1.019)	316.8	2300
Potentially preventable hospitalisations
Chronic care consultations	High	38.3	1449.9	0.026	0.964 (0.927, 1.002)	39.7	− 1400
	Medium	41.9	1471.2	0.028	Ref	Ref	Ref
	Low	42.1	1428.9	0.029	0.991 (0.954, 1.029)	42.5	− 400
Chronic care procedures	High	38.8	1458.7	0.027	0.914 (0.879, 0.950)	42.5	− 3700
	Medium	41.8	1454.4	0.029	Ref	Ref	Ref
	Low	41.7	1436.9	0.029	1.068 (1.028, 1.111)	39.0	2700
Daytime consultations	High	40.1	1446.9	0.028	1.025 (0.988, 1.063)	39.1	1000
	Medium	40.6	1451.6	0.028	Ref	Ref	Ref
	Low	41.6	1451.5	0.029	0.999 (0.959, 1.040)	41.6	− 100

*Number of excess outcomes for the group if the patients listed with these practices had the same outcome incidence as the medium-level group (negative numbers represent avoided outcomes). Numbers are rounded to the nearest hundreds

IRR incidence rate ratio, CI confidence interval.

Practice characteristics

Stratified analyses investigating practice characteristics (patient list size, morbidity load, deprivation score, and degree of urbanisation) showed a similar overall pattern between practice service provision and patient outcomes as the one seen in the main analysis (p value of differences between strata > 0.05), except that patients in practices located in urban areas benefitted less from high provision of chronic care consultations on the PIMs outcome compared to patients in practices located in rural areas (IRR 0.999 versus 0.953, p value 0.004). Deprivation stratification is shown in Fig. 4. Details on the other outcomes are shown in Additional file 1: Figs. S1–S6.

Fig. 4.

Incidence rate ratios of patient outcomes according to service provision in general practice, stratified by practice deprivation index. DADI: Danish Deprivation Index

Discussion

This nationwide cohort study showed that individuals listed with a general practice providing high levels of chronic care had lower risk of potentially inappropriate medications and potentially preventable hospitalisations. The associations showed a dose–response relationship pattern. The variation in chronic care provision could account for thousands of PIMs years and potentially preventable hospitalisations. Overall, the associations were not correlated with practice characteristics in terms of list size, morbidity load, deprivation score, or degree of urbanisation.

Strengths and limitations

A major methodological strength of this study was the reduced risk of confounding by indication. Patients with chronic conditions are more likely than healthy people to receive chronic care treatment, but patients also have higher risk of hospitalisation due to their conditions, and one might mistakenly attribute any difference solely to their treatment. Utilising the variation in chronic care provision between practices overcomes this challenge by studying a natural experiment and hence mimicking traits from randomised controlled trials [22]. In our study, patients are randomly assigned to a given chronic care provision level based on the assumption that this level does not affect the GP listing, as it is largely an uninformed choice and practically unknown to the patient. Furthermore, our large cohort sample from the entire nation with virtually no-loss to follow-up resulted in high statistical precision, reduced risk of selection bias, and high degree of external generalisability. Owing to the Danish registers, we were able to obtain comprehensive and prospectively collected data on all Danish adults and their general practice, including detailed information on chronic conditions and multimorbidity. The reporting of provided GP services is incentivised through a practice remuneration model, and this ensures high quality and completeness of data.

The study had several limitations. We had no data on the individual GP’s or staff member’s inclination to provide chronic care services as data on remunerated healthcare services were pooled for each practice, i.e. we observed the mean service provision and may, therefore, have underestimated the true variation. However, GPs working together in practices generally plan and share service provision and share support staff; this makes it more likely that there is a practice effect beyond that of individual GPs. PIMs and potentially preventable hospitalisations represent outcomes that are related to quality of care, but they only apply on the systemic level when comparing large number of patients and practices. The need for certain medication combinations may rely on sound individual clinical choices made by the physician and the patient, which we had no way to assess, and hospitalisations even for specific ambulatory care sensitive conditions may not be truly preventable for an individual patient. The algorithm-based measurement of PIMs was based on redeemed prescriptions and may over- or underestimate the true occurrence of potentially problematic medications, but with our time-dependent approach to calculating PIM days calculations we do not expect any systematic bias. We had no information on the resources outside general practice that could influence chronic care provision, including the distance to hospital and the access to outpatient care. Even though we showed that higher levels of chronic care provision were associated with more PIMs and potentially preventable hospitalisations, we cannot infer or propose an overall correct level of chronic care for optimal health regarding other patient level outcomes, e.g. mortality or quality of life. We found statistically significant estimates for most outcomes, and a substantial number of potentially inappropriate medication days and hospitalisations could be associated with the level of chronic care provided. However, the estimated associations were weak, so the risk of residual confounding, which is always present in observational studies, should not be ignored. Causality should therefore be inferred with caution, although the design provides some robustness against the most traditional confounding problems, such as confounding by indication.

Comparison with existing literature

Previous evaluations of different chronic care models have shown positive effects on patient outcomes, including prevention of unplanned admissions, but the interventions have mostly targeted specific chronic conditions [4–6] or have been based on a limited sample size [35]. The Quality and Outcomes Framework in the UK has been subject to various evaluations of its impact on patient outcomes. The evidence suggests that meeting targets for conditions like type 2 diabetes was associated with lower mortality rates, reduced emergency hospital admissions, and improved health outcomes [7, 36, 37]. However, the literature also indicated some limitations, such as decreased quality of care in non-incentivised activities and poor patient experiences [7, 38]. In comparison, our study evaluated a complete delivery system for chronic care management in primary care; this system is unlimited to specific chronic conditions, populations, or areas. We included patients regardless of chronic disease status or reason for encounter, also patients without any registered chronic conditions. Despite these conservative assumptions, we still found an overall positive association between chronic care provision and the investigated patient outcomes.

Variation in healthcare utilisation has been examined in both primary and secondary care settings, e.g. as described in the Dartmouth Atlas [12, 16–18]. We used methodology developed for register-based data which quantifies the variation measure at different levels and accounts for the underlying patient case-mix [13, 30]. Previous studies have shown an association between chronic care provision and the geographical location of general practice clinics, indicating that high deprivation level and low socioeconomic status of listed patients was linked to lower than expected healthcare provision, i.e. evidence of the inverse care law [13, 20, 39]. In this study, we found that more chronic care consultations per patient were provided in clinics with a high comorbidity load, but not in deprived area clinics. However, after adjusting for patient characteristics, we found no systematic evidence that practice characteristics, i.e. composition of patient population, modified the association between the propensity to provide chronic care and patient outcomes. This may be interpreted as the efforts of providing sufficient chronic care being beneficial for patients both in practices with high and low degree of deprivation and multimorbidity. The Danish primary healthcare system includes a deprivation-weighted capitation, but it is minor compared to the fee-for-service payments and could not fully mitigate potential differences in patient needs.

Implications

The number of people with long-term conditions has increased dramatically over the years, and it is becoming more costly to deliver sufficient care in the hospital sector. Therefore, policymakers are now looking towards primary care for a sustainable shift in chronic care with a view to relieve the burden on hospitals, avoid fragmented care, and manage patients at the most cost-effective level of care. The unwarranted variation in the provision of chronic care services found in our study reflected underutilisation, as we accounted comprehensively for patient-related factors. Such variation may stem from practice organisation and staffing, available local resources, GP preferences, or systemic structures such as reimbursement systems. If primary care is to take on this task, it is necessary to prioritise initiatives in general practice and allocate resources to accommodate the increasing number of people with multiple conditions. If more optimal treatment of chronic conditions is achieved in primary care, the need for hospitalisations and specialised care could be reduced. Furthermore, the hospital sector must provide more support for chronic conditions in primary care. We showed a dose–response relationship, which suggests that more chronic care could be preferable. However, little is known about which care elements are most effective in which populations, and how to prioritise efforts sustainably with limited staff and financial resources. More studies are needed to examine the best way to use primary care resources, and how to address variations and ensure equity of care.

Conclusions

Patients experienced fewer PIMs and potentially preventable hospitalisations if listed at a general practice providing a high level of chronic care; this association was seen regardless of other practice characteristics and could account for a substantial number of potentially inappropriate medication days and hospitalisations. Still, the estimated effect sizes were quite small.

Abbreviations

ACSC

Ambulatory care sensitive condition

CI

Confidence interval

DADI

Danish Deprivation Index

GP

General practitioner

IRR

Incidence rate ratio

O/E

Observed/expected

PIM

Potential inappropriate medication

Authors’ contributions

AP, MFG, and MV conceived the study. AP, MFG, and CHV developed the study design and statistical methods. CV did the statistical analysis. AP wrote the first draft of the manuscript. AP, CHV, NBU, PV, SMS, MV and MFG contributed to the interpretation of data and made critical revision of the manuscript. AP and CHV had full access to the data and take responsibility for the integrity of the data and the accuracy of the data analysis. AP is the guarantors. The corresponding author attests that all listed authors meet authorship criteria and that no others meeting the criteria have been omitted. All authors read and approved the final manuscript.

Authors’ Twitter handles

Twitter handles: @Anders_Prior (Anders Prior), @susanmsmith (Susan M. Smith), and @PVedsted (Peter Vedsted).

Funding

Open access funding provided by Copenhagen University This study was supported by unrestricted grants from the Novo Nordisk Foundation (NNF18OC0031194) and from the General Practice Research Foundation of the Central Denmark Region. Deep End Denmark is supported by unrestricted grants from the Novo Nordic Foundation (NNF23SA0082354), the Cancer Society (R330-A20465), and the General Practice Research Foundation. All authors are independent from the funders. The funders had no role in the study design; in the collection, analysis, and interpretation of data; in the writing of the report; or in the decision to submit the article for publication.

Data availability

Dr Anders Prior had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis. The dataset supporting the conclusions of this article are available at Statistics Denmark (https://www.dst.dk/en) and the Danish Health Data Authority (https://sundhedsdatastyrelsen.dk/da/english/). Data access is restricted to authorized research institutions by Danish law. Permission to access the data used in this study can be obtained following approval from the Danish Health Authority.

Declarations

Ethics approval and consent to participate

The Danish Data Protection Agency, the Danish Health Data Authority, and Statistics Denmark approved the study (Project ID 707253). Ethical approval and informed consent were not needed as the study was based on de-identified register data encrypted and stored by Statistics Denmark.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.