General practice variation when initiating long-term prescribing of proton pump inhibitors: a nationwide cohort study

In the Danish National Prescription Register [20] we identified all individuals over 18 years redeeming a prescription for PPI in 2011. The date of the first redemption of PPI in 2011 was defined the index date. In order to include only incident PPI users, patients having redeemed PPIs within six years prior to the index date were excluded. The cohort was followed for six months, and similar to other studies [21, 22], patients redeeming >60 DDDs within six months were defined incident long-term users (Fig. 1). Patient factors in terms of gastrointestinal morbidity [23], use of ulcerogenic medication [23], general comorbidity [10] and socioeconomic status [10, 11] have in previous studies been associated with use of PPIs. Therefore, we retrieved additional data on diagnoses from the Danish National Patient Register [24], the Danish National Prescription Register and data on income, cohabitation status and highest attained education from sociodemographic registers contained in Statistics Denmark [25, 26]. Drugs were defined as comedication, if the patient had redeemed a quantity covering the date of redeeming the PPI. Morbidity was analysed as gastrointestinal morbidity (diagnoses of peptic ulcer, gastrooesophageal reflux disease (including oesophagitis), functional dyspepsia) and as general comorbidity measured by the Charlson Comorbidity Index [27].

The prescriptions for PPI were linked to the prescriber through the Danish National Health Service Provider Register. We extracted data on the general practices prescribing PPIs to the cohort of first-time users. The average list size of a typical Danish GP is 1600 patients [18], and practices with atypical small list sizes (<500 patients) were excluded, because they were thought not to be representative (Fig. 2). Practices with missing data in 2010 or 2011 were omitted as well, as this indicates that these practices were established or closed in that period (Fig. 2).

We identified the number of doctors registered at each practice and catagorised them as established GPs or temporary doctors. Doctors not registered in the entire period were defined temporary doctors. Practices were defined single-handed practices, if only one established GP was registered, and partnership practice if two or more established GPs were registered. The majority of temporary doctors in general practice are junior doctors having a six- to twelve-month residency, and practices with these doctors listed were defined training practices. We calculated the number of patients per GP by dividing the practice’s patient list size by the number of established GPs within the practice. In single-handed practices the age and gender of the GP were retrieved, and in partnership practices we calculated the mean age of the established group of GPs and assessed, whether the group of established GPs comprised exclusively men or women, predominantly men or women or equally mixed.

The postcode of all practices was extracted. Practices within the greater Copenhagen postcode area were categorised as capital area practices. Practices outside the capital area, but in a postcode comprising a city with more than 10 000 inhabitants, were labelled provincial city practices. All other practice locations were categorised as rural [28]. According to Danish legislation neither the individual GP nor the individual patient should be identifiable for the researchers. Therefore, a few practices (N = 4) with postcode comprising less than three other practices had missing postcode and could not be characterised according to degree of urbanisation.

The analyses were conducted both with the entire cohort of general practices and with stratification into single-handed and partnership practices. This was done because the variables age and gender were exact values in single-handed practices, but average values in partnership practices. Mixed effects logistic regression models with patients nested within practice were used to calculate odds ratios (ORs) with 95 % confidence intervals (CI) for associations between long-term prescribing of PPIs and practice characteristics. Two regression models were used. Model one estimated the crude OR for the association of each practice characteristic and prescribing of long-term treatment with PPIs. Model two estimated the OR for each practice characteristic, adjusted for both patient characteristics and other practice characteristics included in the analyses. Patient characteristics adjusted for were age, gender, gastrointestinal morbidity, socioeconomic status (income, highest attained education and cohabitation status), comedication with non-steroidal anti-inflammatory drugs (NSAIDs), antiplatelet drugs, anticoagulants, selective serotonine reuptake inhibitors (SSRIs) and comorbidity [7].

A sensitivity analysis with a definition of long-term use increased to 90 DDDs within six months was performed in order to explore the consistency of the associations when changing threshold for long-term use.

According to “The Act on Research Ethics Review of Health Research Projects in Denmark” this register-based study did not require approval by the Research Ethics Committee. The study was approved by the Danish Data Protection Agency, J. nr. 2012-41-0280. Neither patients nor GPs were identifiable in the dataset.

Male gender and increasing age of GPs are significantly associated with initiating long-term treatment with PPIs in a cohort of first-time users treated in a primary care setting. However, general practice characteristics do not seem to be strong predictors of initiating long-term treatment with PPIs as the magnitude of the associations is modest. The apparent variation with regard to rurality and partnership status observed in the crude analyses was no longer statistically significant when adjusting for practice- and patient-related factors. Given that the patient population in rural and single-handed practices is older and has a higher degree of morbidity [29, 30], this might reflect a different composition of patient populations in different practice types and locations.

The major strength of this study is the register-based design enabling us to include information on all GPs issuing first-time prescriptions of PPI to a nationwide cohort of incident PPI users. The validity of the register data is considered high [20], and the accurate linkage between the patient-related factors and the practice characteristics makes it possible to adjust for numerous predisposing patient factors and thereby isolate and analyse the independent associations between practice characteristics and initiation of long-term treatment with PPIs. Only 3 % of the total sale of PPI in Denmark is over-the-counter [1], and therefore registration of PPI use is considered almost complete.

The fact that we extracted practice data at the time of initiating long-term treatment with PPIs provides the study with the strength to accurately identify factors of the physician responsible for initiation of long-term prescribing and to characterise practice features as predictors for incident long-term prescribing. In case we had analysed practice characteristics associated with prevalent long-term prescribing of PPIs, we would not have been able to be sure of the validity of the predictors, because prevalent long-term prescribing might have been established several years earlier by another physician.

However, influence of unmeasured factors cannot be ruled out. The linkage between the prescription and the physician factors can only be made at practice level, impeding identification of the GP within partnership practices primarily providing care to the patient. This limitation makes it difficult to assess the effects of doctors’ age and gender in partnership practices. Exact information on age and gender would have been more precise than the mean age of doctors within the practice and the relative gender distribution. Therefore, the associations between long-term prescribing of PPIs and GP age and gender can only be certain for single-handed practices, though the associations seem similar for partnership practices.

Other potential influential variables could have been interesting to include in the analyses had they been available. Practice characteristics such as distance to outpatient clinics (i.e. endoscopy, gastroenterological expertise etc.) could influence prescribing of PPIs. However, data on practice location are limited to postcodes in order to keep the individual prescriber unidentifiable, and the exact address is not available to calculate distance to outpatient clinics.

To the best of our knowledge no other study has investigated the associations between practice characteristics and prescribing of PPIs. Our study does not clarify why older GPs to a greater extent prescribe long-term treatment with PPIs in first-time users, but it is important to keep in mind that few single-handed practices are run by younger GPs (Table 1). These practices could have a smaller long-term user proportion due to random variation, decreasing the comparability between younger and older GPs in single-handed practices. Nevertheless, in line with our findings it has been demonstrated that older GPs have higher prescribing rates [14, 31, 32] and lower rates of non-pharmacological treatments [14]. Similar to our analyses, other studies have found that GPs of male gender have higher prescribing rates [31, 33]. This could be due to the traditional thought of female GPs being more psychosocially orientated and more patient-centred compared to male GPs [34, 35].

In the multivariate analyses we found no association between single-handed practices and high rate of initiating long-term prescribing of PPIs. In some studies partnership practices have been associated with higher scores for quality of care in chronic diseases [12, 36], although the opposite has been shown as well [37, 38]. Moreover, patient satisfaction seems to be higher in single-handed practices [39]. However, in the present study we cannot determine either quality of care or patient satisfaction.

Being a training practice has also been shown to influence management of patients and quality of care [12, 40]. In our study we saw no association between training practice status and initiating long-term treatment with PPIs except for partnership practices, where training practices had slightly higher odds of initiating long-term treatment (OR 1.08, CI 1.02-1.14). Why training practice status influences partnership practices, but not single-handed practices, is unknown. The association found is, however, fairly weak and just barely statistically significant.

Rurality has been associated with higher rates of inappropriate prescribing [41] and different prescribing patterns [42]. That we do not find similar associations could be due to the fact that Denmark is a quite homogenous country with small distances and less difference between rural and urban areas compared to other countries.