Strengths and limitations of the study
A considerable strength in our survey is that the questionnaires, of the consultations studied, mandatorily popped up and were consecutively filled out immediately after each consultation, giving a sample of representative patients. Furthermore, the GP sample was randomly drawn and altogether the referral data was collected through all parts of the year.
However, the response rate of 42% raises the concerns of selection bias. Others have found that non-responder GPs tend to be older and less likely to possess a postgraduate qualification . Our responders were indeed younger than the non-responders and were more often specialists in family medicine (Table 1). Furthermore, more responders worked in private practice and were more often medically educated in Norway. It is difficult to assess the overall impact of non-response bias. But, assuming that the referral practice of the non-responders was similar to that of the responders, increasing the response rate might lower the referral rate related to increasing GP age, especially of those above 50 years. On the other hand, increasing the response rate might also increase the referral rate related to a falling percentage of specialists in family medicine, non-salaried GPs and GPs with medical degree from Norway. In a set of hospital data, which comprised all referrals from all GPs to any hospital in Northern Norway (hospital outpatient clinics only) between 2008 and 2011, our responders referred 23.4% of their list population per year compared to 25.6% among non-responders, (personal communication with Center of Clinical Documentation and Evaluation (SKDE), Northern Norway Regional Health Authority of May 2011). This might indicate that the observed referral rates in our study more probably represent an underestimation than an overestimation of referrals. Additionally, the total secondary care utilization in Northern Norway does not differ substantially from the rest of the country according to a Norwegian report . We therefore believe that our study is fairly representative of the GP referral practice in Norway.
Comparison with existing literature
The referral rate in Norway was reported to be 8.0% in 1983  and 8.2% in 1993 . In the same time period the referral rate in Finland was 4.5% to out- and inpatients services , and the British one was 4.7% . In 2009 the referral rate in Denmark was reported to be 8.4% . There may be problems with comparing referral rates between countries because the consultation practice may differ, i.e. the consultation rate per day may differ. Still, our referral rate of 13.7% to secondary care was higher than anticipated . This increase in referral rate in Norway of 5.5 percentage points from 1993, represents a secondary care work load increase of about 67% and may indicate a less restrictive GP gatekeeper function . The indicated work load increase is supported by the fact that the number of outpatient consultations in Norway has increased by more than 60% between 2002 and 2011 . In the United States the reported probability that an ambulatory visit to a phy sician would result in a referral to another physician, increased by 94%, from 4.8% in 1999 to 9.3% 2009 . Also Moth et al documented a 36% rise in the referral rate to secondary care in Denmark (radiology and laboratory excluded: from 6.2% in 1993 to 8.4% in 2009) .
The GP consultation rate has increased somewhat over time, with a rise of around 5.4% between 2006 and 2009 . The increased referral rate may partly be a result of care becoming more complex, thereby requiring more care by secondary care specialists. In addition, the diagnostic and therapeutic options have expanded, and are applied to patients with either more advanced disease, or who are older. Also, patient demand and consumerism are contributing factors. An increase in the referral rate may have unwanted consequences for the patients. Unnecessarily referred patients may be subjected to unnecessary and dangerous diagnostic and therapeutic procedures . In addition, an increasing referral rate contributes to longer patient waiting lists which may cause poorer health outcomes for seriously ill patients . Increasing referral rates also increase the economic burden of health care on society.
The wide range of referral rates between GPs has been well documented, but our reported seven-fold increase is among the highest ever reported [2, 11]. The variation in referral rates seems to be stable over time and independent of case mix . Sullivan et al reported that morbidity only explained 30.4% of the total variation in referral rates, and patients’ age and sex explained 5.3% of the total variation . Thus, non-medical factors must also contribute to the referral rate range. In our study, the relative proportion of referrals was about the same whether the patient or the GP initiated the issue of referral. We had anticipated that the referral rate would be higher when GPs initiated the issue of referral, and we therefore think our result indicates a stronger responsiveness to the patient’s requests on the part of the GP, resulting in difficulties or reluctance to make rationing decisions. The issue of referral was introduced more frequently in consultations with high referrers, and also resulted more often in referrals. This may be due to higher professional insecurity and/or higher responsiveness to patient demands. The variability in referral rate challenges the basic principle of equal access to health care, as patients with low referrers as their GPs encounter a different health care system than those attending high referring ones.
In the present study the referral rate of female GPs to secondary care was 25% higher than that of male GPs. This is consistent with gender differences found in other studies. Vehvilainen et al found that female GPs referred 22% more than male GPs (female rate 5.48% versus 4.50% among males) . One explanation for this difference may be gender differences in risk tolerance. Indeed, GPs with lower risk tolerance are reported to refer more [25, 26], and a significantly lower risk-taking behavior has been found in females in a meta-analysis . In sub-analyses in our study we found that female GPs referred a significantly larger proportion of patients than males when the issue of referral was introduced, especially when the GPs initiated the issue. This may indicate a female tendency to stronger acquiescence and an attempt to reduce uncertainty when choosing to refer. As the percentage of women is increasing among doctors, the reported gender difference could have major implications for patients and society. The literature, however, reveals conflicting evidence regarding whether women practice medicine differently than men .
We found that specialists in family medicine referred less to secondary care, which has also been reported by others . However, these specialists referred more frequently to radiological examination. May be specialists in family medicine perform more of the diagnostic process themselves instead of referring them to secondary care, or they start investigating the patient while waiting for secondary specialist consultation.
We were surprised to find a significant difference in the probability of referral between salaried male GPs and those in private practice, but no difference among female GPs in these types of practices in multivariable analyses. All private practitioners had significantly more consultations per day than salaried GPs (mean 10.4 per day, 95% CI 9.6-11.3, versus 7.8, 95% CI 6.6-8.9, respectively). It may be that salaried GPs have fewer, but longer consultations per week and refer more frequently per consultation, but not more frequently per list population per year. Our finding is somewhat in line with Fleming, who in 1993 reported that the low referring doctors undertook a high number of consultations compared to high referring ones .
The relatively low number of GPs in our survey and lack of power may give concern of type II errors. We will therefore also display non-significant results. Proximity to specialist is reported to increase referral rates . We found that the probability of being referred were 20% less for patients with the longest travel time (Table 2).
Implications for general practice and future research
We found a high referral rate and a very large range. Although our study does not consider the appropriateness of the referral rates, the lower referral rate of specialists in family medicine indicates that the overall rate may be lowered. Future research should focus on exploring the effectiveness of quality improvement measures and include strategies to both reduce the level and range of referral rates. Our results from contrasting high and low referrers may indicate that intervention on high referrers is a potential area for quality improvement. In a meta-analysis concerning interventions to influence referral patterns, Akbari et al ascertained that few rigorous evaluations have been carried out . Active local educational interventions involving secondary care specialists and structured referral sheets were the only interventions shown to have an impact on referral rates. The effects of ‘in-house’ second opinion and other intermediate primary care-based alternatives to outpatient referral appeared promising.
GP’s gender, specialization in family medicine and type of practice have an important impact on referral rate. The frequency of introducing the issue of referral in the consultation and the corresponding referral rates are different in high and low referrers. This needs to be explored more to shed light on the communication skills of GPs, and to understand how gender, patients’ wishes and other issues relevant to equality and diversity affect the way GPs practice medicine [28, 30, 31].
In Norway a Coordination Reform  began its implementation in January 2012. There is a need to control the demand for secondary health care at the primary health care level, and a system of copayment for secondary care use from the municipality, is introduced. One goal of the reform is to prioritize development of new services in the municipalities and transfer patients from secondary to primary health care when necessary. The high referral rates found in our study call into question the viability of the government’s goal with the reform.