In this study we estimated the prevalence and annual healthcare costs of DR in a diabetic population. Our results show that the prevalence of any DR in the total diabetic population is 29.2%. DR was more frequent among patients with type 1 diabetes than among those with type 2 diabetes, with DR prevalence of 41.8% and 27.9% respectively. The prevalence of DR was also significantly higher for men (30.9%) than for women (27.4%). Sight-threatening retinopathy such as PDR and maculopathy was present in 5.7% of all individuals with diabetes. The study also showed that treatment of DR costs €72 per year per patient. The costs varied considerably, however, between different severity levels of the disease and were estimated at €26, €257, €216 and €433 per patient per year for BR, PDR, maculopathy, and PDR combined with maculopathy respectively. Combined, the results for prevalence and average costs yielded an annual total healthcare cost of DR of €106,000 per 100 000 inhabitants.
The main strengths of this study are the large size of the study population and the fact that, being based on register data, it covers an entire population. However, even though the register-based method eliminates the risk of recall bias, it includes a risk of misclassification, such as missed or erroneous registrations. This risk of misclassification of DR was reduced in the present study by using information from two different registers. For approximately 40% of the diabetic population, we were able to compare the DR diagnoses in the CDWÖ with registrations of DR in the NDR. This comparison showed that 96.8% (1993/2058) of the total cases of DR in our population were identified in the CDWÖ. Assuming that this sensitivity is also valid for the remaining 60%, this gives a maximum prevalence of any DR of 29.9%, which is very close to our original figure of 29.2%. In total, the level of agreement between the two registers was approximately 86%. The 14% on which the registers disagreed were mainly cases in which individuals had received a DR diagnosis in the CDWÖ but not in the NDR. A possible explanation for this could be that patients with diabetes are registered in the NDR voluntarily at any time during the year, which could be before the patient has been screened for DR. New cases of DR that are detected after the registration form has been sent to the NDR would consequently not be registered until the following year. Of course, there may still be a few cases of DR that have been missed in both the registers we have used, which could have led to an underestimation of the prevalence of DR. However, our estimates may be further validated by comparison with the results of a previous study  from a region included in the catchment area of the present study. All patients <70 years old with type 2 diabetes living in this region had been screened for DR, with the results showing a prevalence of BR of 19.8% and a prevalence of severe DR of 6.7% (22.5% and 5.0% in the present study). Our estimates for the prevalence of diabetes (types 1 and 2), which were 4.5% among women and 5.0% among men, were based only on registrations in the CDWÖ but correspond well with the self-reported prevalence published in a report by the Swedish National Institute of Public Health  (4% for women and 5% for men among individuals 16–84 years old in Östergötland).
A further strength of the current study is that we had good information on the local practice behind the data in the registers we used. This information made us aware of the differences in practice between the county’s two eye clinics, which resulted in a decision to limit our study population to the catchment area of the eye clinic at Linköping University Hospital (M. Abrahamsson, personal communication).
It also made us aware of the potential risk that not all cases of maculopathy had been registered in the last months of the period for which we searched the CDWÖ, and that individuals diagnosed with maculopathy for the first time in the last months may thus mistakenly have been included in the BR or PDR groups. Since this would have concerned only the registration of a few individuals, we do not believe that it would have substantially affected the prevalence of the different severity levels.
A limitation of the study is that the DR patients identified solely through the NDR could not be classified into a specific severity level, which may have caused a slight underestimation of the prevalence of BR, PDR and maculopathy. As these patients (1.8% of our DR population) were not registered with a diagnosis of DR in the CDWÖ during the period of data extraction (2000–2007), they are not likely to have had DR for a long period, and we may therefore suspect that the majority of these belong to the BR group. If we include these patients in the BR group, the prevalence of BR increases from 23.2% to 23.6%. A second limitation is that we could not identify patients who were blind, since there is no specific ICD 10-code for blindness. As a result, patients with DR who became blind before the year 2000 have been classified as not having DR, which might have affected the estimates of prevalence as well as costs. However, only a negligible number of individuals (approximately 20) in the catchment area of the eye clinic of Linköping University Hospital are known to be blind as a consequence of DR (unpublished data, patient register at the Vision Central, Linköping University Hospital). Further, most of these individuals are no longer receiving treatment and are therefore not expected to generate large healthcare costs. That blindness increases costs in other areas of society is beyond the perspective of this study. It could also be considered a limitation that we have used costs from a regional database. However, there are various reasons for us to believe that these costs can be extrapolated to a national level. First, the demographics of the patients included in the database are similar to those of the national population . Second, there are national guidelines concerning treatment of diabetes and diabetic retinopathy [26, 27]. Third, healthcare costs in Östergötland have been seen to be similar to national healthcare costs . Furthermore, previous studies within other disease areas have used cost data from the CPP database to approximate costs on an aggregate national level .
Our prevalence estimates for any DR, for type 1 as well as type 2 diabetes, are substantially lower than those reported in two previous studies from Sweden [30, 31], in which the authors have reported a DR prevalence of more than 60% for younger-onset diabetes and over 35% for older-onset diabetes. However, both of these clinic-based studies were conducted almost two decades ago, which might explain some of the differences from our study. Over recent decades, diabetes care has improved in terms of decreased mean HbA1c and mean systolic blood pressure , changes that have been shown to have a preventive effect on diabetic complications [32, 33]. Previous research [34, 35] has shown a similar effect on the incidence of DR in Sweden. However, for type 2 diabetes, this explanation is contradicted by two Swedish population-based studies from the same time period [25, 36]. Lundman et al.  have reported a DR prevalence of 18.7% in non-insulin-dependent adults living in a county in central Sweden, and Falkenberg et al.  have reported a prevalence of 26.5% among type 2 diabetic patients <70 years old living in a rural area within the catchment area of the eye clinic of Linköping University Hospital. Our results for any DR are in line with the results from an American study from 2007 , in which the authors have reported a prevalence of 27.4% among individuals aged 40 years or older with diabetes. For the same age interval the prevalence of DR in our study was estimated to be 29.4%.
Comparing our results on the healthcare costs of any DR with the results of other studies is difficult because many cost studies lack information on the prevalence of PDR and maculopathy. The costs of PDR (€257), maculopathy (€216) and of both complications combined (€433) reported here could, however, be compared with the results of a German study  and a Canadian study . Including only costs for ophthalmologist fees, medication and hospitalisation—corresponding to our definition of healthcare costs—the estimations from the German study give an average cost of €468 for patients with PDR and €681 for macular oedema. In the Canadian study, the costs for PDR, macular oedema and both complications combined were estimated to be €284, €254 and €333 respectively in the first year of these conditions. The annual subsequent cost was €26 for all these health states. Differences between these results and those of the present study could perhaps be explained by dissimilarities in how DR is treated in different countries. In Sweden, PDR is treated with photocoagulation, and in severe cases by vitrectomy, while maculopathy is primarily treated with photocoagulation. At present, pharmaceuticals are usually not used to treat DR. Other possible explanations are differences in the definitions used for classification of the different severity levels, and different proportions of type 1 and type 2 diabetes in the study populations. Regarding the German study, another explanation for the higher costs could be that only patients who had received treatment during the year for which their costs were calculated were included in the study.
In summary, the present study offers not only information on prevalence and costs of DR in Sweden but also information on the prevalence of diabetes. Approximately 5% of the population have diabetes, and approximately one-third of these patients have some form of DR. Of all patients with diabetes, approximately 6% have sight-threatening DR, such as PDR or maculopathy. By estimating the average additional healthcare costs for different severity levels of DR, our study shows the potential cost savings per patient per year of preventing development of DR. These average annual healthcare costs vary considerably with DR severity, indicating that prevention of DR progression may reduce healthcare costs. However, in order to investigate further whether it is worth spending resources on interventions directed to DR prevention, the costs of DR must be set in relation to the effect of a specific treatment in a formal cost-effectiveness analysis. In such an analysis, we believe that the average costs of DR that have been presented in this study would constitute a good input.