Study design
In the Netherlands GPs are supposed to treat patients themselves unless referral to a medical specialist or other healthcare provider is needed. About 90% of all health problems presented in GP-practice is treated by the GP self. Referral rates to medical specialists thus are an indicator for what can be handled in general practice and what not. GPs provide community based family medicine and internists hospital based internal medicine.
To answer the research questions we analysed whether the referral rate to internists, ophthalmologists, cardiologists and mental healthcare changed from 2004 to 2006, and whether or not this was different for general practices with and without primary care nurses. We also examined whether the diabetes-related contact rate was different in practices with and without primary care nurses. The diabetes-related contact rate was only analysed in 2006, since no detailed information was available for 2004. A distinction was made between newly diagnosed and known T2DM patients. To convert the treatment of patients to primary care nurses is harder for patients who have been treated by GPs or internists for years, than for newly diagnosed patients.
For the purpose of this study, we used data on healthcare utilization of newly diagnosed and known T2DM patients for a period of 365 days after the first diagnosis of T2DM (newly diagnosed T2DM patients) or after the first consultation or prescription for T2DM (known T2DM patients) in 2004 and 2006. T2DM patients were seen as newly diagnosed when patients had no diabetes record in GPs' electronic medical record (EMR) in the previous years (with minimum of one year). In total, 450 newly diagnosed and 3226 known T2DM patients in 2006 and 301 newly diagnosed and 2124 known T2DM patients in 2004 were included in the analyses.
Subjects
Data were derived from EMRs of general practices that participated in the Netherlands Information Network of General Practice (LINH)[22]. LINH is a representative sample of general practices in the Netherlands. Each year some minor changes in composition of practices occur due to natural turnover. The data hold information about morbidity (international classification of primary care (ICPC codes)[23]), prescriptions, contacts and referrals. Medical ethical approval was not required for this research.
Figure 1 shows the inclusion criteria for general practices and patients in 2004 and 2006 and the number of practices and patients included. In 2004 25 practices and in 2006 29 practices were included. Most practices were excluded from the analyses owing to a poor recording of referrals. The selection of practices forms a representative sample of Dutch general practices with regard to practice type (single handed, duo, group or health centre), degree of urbanisation and province.
T2DM patients were selected on the basis of the ICPC code for diabetes: T90. We were not able to distinguish T2 and T1 diabetes patients on the basis of the ICPC-coding. For the purpose of this study, type I diabetes patients were excluded on the basis of prescription data (ATC-coded[24]). Type I diabetes was characterised by diabetes patients with a prescription of insulin (ATC code A10A), but without oral anti-diabetic medication (ATC code A10B)[25].
Measurements
Referrals
We analysed new referrals to internists, ophthalmologists, cardiologists or mental healthcare. A patient was considered as being referred (1) if a referral had been recorded within 365 days after the first diagnosis or first consultation for diabetes (including this consultation). Referrals to mental healthcare included referrals to psychiatrists, psychologists or ambulatory mental healthcare.
It was unknown whether GPs could perform an eye fundus examination in their own practice and therefore not refer patients to ophthalmologists. Most of these GPs, however, probably perform only retinaphotography and leave the examination of this photo to ophthalmologists.
Diabetes-related contacts with general practice
Diabetes-related contacts were only assessed in 2006 and based on the number of claimed telephone and office consultations and home visits with an ICPC code T90 (diabetes). In 85.8% of all consultations and home visits the diagnosis was known in 2006 and 2007.
Primary care nurses
The presence of a primary care nurse was determined for all general practices on the basis of data from the EMR.
Covariates
We adjusted for factors that could affect the relation between referral rate and presence of a primary care nurse or the relation between diabetes contact and presence of a primary care nurse. These included comorbidity and distance to hospital apart from gender and age (continuous).
Comorbidity
Comorbidity was taken as covariate, since T2DM patients with comorbidity were assumed to be more likely to be referred to a medical specialist than patients without comorbidity[26] and may have more consultations. Using the ICPC codes in the EMR of the practices, we distinguished between diabetes-related comorbidity and unrelated comorbidity. Related comorbidity included heart diseases, stroke, retinopathy, nephropathy and diabetic foot. Non-related comorbidity included depression, lung diseases, musculoskeletal diseases, neurological diseases and cancer. Additional file 2 shows the ICPC codes and descriptions. Patient were regarded having related or unrelated comorbidity (0/1) if s/he had consulted the GP or had a prescription for one of these diseases.
Distance to hospital
Distance to the nearest hospital for a patient might influence the referral behaviour of GPs, since they might be more reluctant to refer patients living further away from a hospital[27]. Road distance to the nearest hospital was based on distance from the centroid of the postal code of the patient's home to the nearest hospital.
Statistical analyses
To analyse the relation of the presence of primary care nurses with contacts with general practice and change in referral in T2DM patients, multilevel logistic regression analyses (referrals) and multilevel linear regression analyses (contacts) were conducted with MLwiN 2.02. Multilevel analysis corrects for the cluster effect of patients within general practices[28].
In analyses of referral rates between 2004 and 2006, time was included as a dummy variable representing 2006, with 2004 as reference category. For all analyses, first a model with only the dependent variables was analysed (model 1). Second, covariates were added to the model (model 2). Last, the interaction term 'primary care nurse in practice*year' was added to the referral analyses (model 3). Covariates in the referral analyses were age, gender, and related and unrelated comorbidity and distance to hospital. Covariates in the contact analyses were age, gender, and related and unrelated comorbidity. In addition, the effect of primary care nurses in practice was analysed separately for 2004 and 2006.
Analyses of referrals were performed separately for referrals to internists, ophthalmologists, cardiologists and mental healthcare. The significance level was set at p < 0.05. For the interaction 'primary care nurse in practice* year', significance level was set at p < 0.10 since this was measured on practice level and the number of practices is much smaller than the number of patients. The models were estimated with multilevel logistic regression analyses with second-order PQL (penalised quasi-likelihood), and multilevel linear regression analyses, both with only a random intercept.