Treatment Pathways for Patients With Atrial Fibrillation

J. A. Hodgkinson; C. J. Taylor; F. D. R. Hobbs


Int J Clin Pract. 2012;66(1):44-52. 

In This Article


Study Population

The General Practice Research Database (GPRD) is the world's largest computerised database of anonymised longitudinal patient records from primary care, and currently collects data from over 350 practices on three million patients in the UK. Data are quality assured by checks for consistency and completeness of data recording and adherence to GPRD guidelines, and many epidemiological studies have confirmed the validity and data completeness of the GPRD.[23]

This study drew on data from the whole database, i.e. since 1987, although most of the data are from 1990 onwards. The data extraction was carried out in May 2007 using the most up-to-date data. A series of sensitivity analyses excluding patients with an index date prior to 2002, 1996 and 1990 were also conducted, to ascertain if growing awareness of AF and understanding of the condition had impacted on UK treatment practice. These restrictions were applied, because it was anticipated that there would have been a significant move from antiplatelet to anticoagulant therapy, as doctors have become more aware of AF and had greater availability of ECG for diagnosis, although there is evidence that guidelines were influencing prescribing potentially as far back as the early 1990s.

All practices that collected data and were up to standard (as defined by the database providers, based on internal checks of the completeness and continuity of recording) for GPRD were eligible for inclusion and provided patients for the study. In total, we identified 67,857 AF patients in the GPRD that matched the study population criteria.

Case Definition

All patients aged 18 years and older, with a diagnosis of AF during GPRD data collection, were included in the analysis, except patients with a history of heart valve problems and/or valve replacement surgery. AF diagnosis was based on either free text or Read codes.

The analysis was intended to be stratified by type of AF (chronic and paroxysmal). However, there is no Read term for chronic AF and only two terms for paroxysmal AF in the GPRD, and most codes concern several generic non-specific AF terms (e.g. 'Atrial Fibrillation'). Anonymised key-word searches of entries in the free text field associated with all diagnoses were used to further classify AF cases as either chronic or paroxysmal, but, because of the lack of constant definition and subjectivity involved in labelling and our finding little difference between patients in the any AF, chronic and paroxysmal groups according to our classification, it was decided not to publish any breakdown of this analysis.

Analysis of Treatment Pathways

A series of possible treatment pathways were identified, including no use of AF therapies, antithrombotic therapy (anticoagulation or antiplatelets), rate control and rhythm control, in any combination. For each treatment pathway, we estimated the probability over time of a given treatment pathway using the data from the AF cohort (e.g. the probability of a patient first being treated with an antithrombotic medication and then with rate-limiting medication and then not treated). A finite state Markov process model was used to model the hazard rates for each possible transition using a separate (left truncated) Cox proportional hazard model.[24,25] The transition probabilities were calculated using the base line hazard function.[26,27] The probabilities for each treatment pathway were then used in a simulation model to estimate the distribution of various pathways for a cohort of 10,000.

Using the classified AF diagnoses, a logistic model containing comorbidities in the year before and after diagnosis and AF treatments 6 months before and after diagnoses was fitted. The model included age, gender, BMI, time since first AF diagnosis, the number of AF records within a continuous episode (defined using a 5-day overlap), a record of AF before diagnosis of normal sinus rhythm, ischaemic heart disease (IHD), congestive heart failure (CHF), cardioversion, asthma, diabetes, antiarrythmics, cardiac glycosides, warfarin, other anticoagulants and antidiabetics, and a record after diagnosis of CHF, IHD, hypertension, cardioversion, cardiac glycosides, antiarryhthmics and rate-limiting calcium entry blockers. For all patients, the predicted probabilities for chronic AF and paroxysmal AF were estimated using the results of the logistic regression. Using the sensitivity and specificity of the data in the logistic model, cut-off values for the probability were established. These cohorts were then externally validated using 1000 patients from the Barcelona data set.[28]

The transition probabilities for 1 and 5 years after diagnosis were stratified using CHADS2 score. As CHADS2 scores were not published when some of the data were gathered, the scoring system was applied retrospectively to the database according to published data, rather than being a GP's own assessment at the time of initial therapeutic decision-making.


Comments on Medscape are moderated and should be professional in tone and on topic. You must declare any conflicts of interest related to your comments and responses. Please see our Commenting Guide for further information. We reserve the right to remove posts at our sole discretion.
Post as: