Clinical practice variation has been of scientific interest the last forty years [1], [2], [3], [4] and has been documented throughout OECD countries [5]. Variation has been problematized as a symptom of suboptimal care and inefficient resource spending [6], [7], [8], [9] related to a risk of underuse, overuse or misuse of services [5]. Unwanted variation is a contested term and, given patients’ varying preferences and the uncertainty involved in managing many clinical conditions, an acceptable level of variation is difficult to determine [7]. Policy goals for clinical practice are therefore often formulated as an ambition of obtaining ‘consistency in the delivery of care’, i.e. “patients with identical clinical conditions should receive identical care, irrespective of the healthcare professional, the healthcare institution, or the (socio-geographic) practice setting” [7].

Two common strategies to foster consistency in healthcare delivery are the development of clinical practice guidelines (CPGs) and the centralization of healthcare delivery in high-volume units. The CPGs strategy assumes that clinical practice variation stems from lacking or variable knowledge translation, such as clinicians’ difficulties in keeping up-to-date with rapidly developing scientific knowledge [7,8]. The idea that centralization should foster quality and consistency is supported by a long list of theoretical mechanisms that can be broadly grouped into scale and scope effects [10]. Scale effects refer to the idea that larger units have more resources for establishing well-functioning infrastructures, for example more resources to improve care processes and quality control. Scope effects cover for example learning-by-doing and knowledge spillovers from teams working together [10]. Centralization may contribute to reaching consistency in care by reducing the number of departments delivering specialist services with low-skilled staff or in potentially unsafe circumstances [11,12].

This paper explores possible explanations for inconsistencies in healthcare delivery. Based on a longitudinal case study of anti-vascular endothelial growth factor (anti-VEGF) therapy in Danish ophthalmology specialist care, we investigate whether and how patient outcomes and treatment costs varied across geographical regions over time, and whether patient, professional or structural determinants can have contributed to such inconsistencies.

In 2007, the groundbreaking and costly anti-VEGF therapy was introduced for the treatment of wet age-related macular degeneration (wAMD) in Denmark, decreasing the incidence of legal blindness attributable to wAMD substantially [13]. The introduction of anti-VEGF therapy in ophthalmology specialist care is a particular interesting case to look at in terms of inconsistencies in health care delivery as the therapy is very expensive and is given to a large and increasing share of the population. The treatment of wAMD is one of many treatments that challenge the future delivery of universal health care coverage. Studies of inconsistencies in healthcare delivery of expensive treatments may help health authorities to identify scope for efficiency improvements and inform debates about value for money. Moreover, the introduction of anti-VEGF therapy can be considered a ‘most likely’ case [14,15] for the obtainment of consistency in healthcare delivery as it was introduced with two research-based strategies to ensure consistency. The therapy came with a national clinical guideline representing consensus about best practice among leading ophthalmologists, and care delivery was centralized to a few high-volume treatment units. Despite the consensus on best clinical practice and care centralization, this study shows a lack of consistency: regional variation in patient outcomes and treatment costs were pronounced and increased over time. We find that the lack of consistency was largely unrelated to patient-specific characteristics, instead it seems to reflect regional differences in the preparedness and strategies for scaling up treatment activity to cope with substantial patient accumulation. We found that the regions that managed to delegate and outsource treatment activity – rather than centralize – obtained higher treatment effects; presumably because this provided for higher treatment intensity.

We make empirical contributions to two strands of literature: 1) the literature on the drivers of clinical practice variation, and 2) the literature on the impact of research-based strategies to ensure consistency in care. To the first strand of literature, our main contribution is to employ a mixed-methods design, as previous studies examine drivers of clinical practice variation with either quantitative or qualitative methods. Previous quantitative methods range from purely descriptive to more advanced regression analyses [16,17], while qualitative studies include interview-based analyses, scoping reviews and theoretical discussions [[7], [8], [9],[18], [19], [20]]. Quantitative studies often result in general decompositions with broad categories of determinants (e.g. patient and structural), while qualitative studies may point to case-specific explanations without being able to clearly isolate the contribution of one determinant from another. In this study, our mixed-methods design enables us to quantitatively assess to what extent patient-level data can explain regional variation in patient outcomes and treatment costs, while we qualitatively explore specific structural and professional-level determinants after having adjusted for differences in patient case mix.

The second strand of literature on the impact of strategies to ensure consistency in care is sparse. Although studies suggest that centralization may contribute to reaching consistency in care [11,12], the empirical relationship between the number of units and the degree of consistency in care across units is largely undocumented. A few studies highlight that centralization of care implies more inconsistency in the delivery of care as patients living in rural areas may be exposed to increased travel distance to acute care necessary for survival [21,22]. For non-acute care, the empirical relation between centralization and consistency remains an open question. Only a few studies empirically evaluate the impact of and adherence to CPGs as a means of ensuring consistency in healthcare delivery [23], [24], [25]. To contribute to the literature on the impact of research-based strategies to provide consistency in care delivery, this study analyzes individual-level longitudinal data across eight years, and explores a most-likely case where we would expect to find consistency in care. The longitudinal data structure enables us to study the development between healthcare regions over time, and the individual-level data allows us to exploit the variation across patients within regions and avoid using averaged patient data that often mask patient heterogeneity and lead to incorrect conclusions about relative performance [26,27]. Moreover, previous studies are often based on survey data [25], while this study includes both clinical and administrative data.