Even if self-rated health was the strongest predictor for use of health care services, we have shown important inequalities in the utilisation according to SES in a country where relative health differences are considered large between SES groups [3, 4], and where equitable access to health care services is a political objective  and even a statutory right .
Five key points stand out. First, GP services utilisation was significantly higher in lower SES groups, where the greatest needs are likely to be found. Smaller sample studies from Norway report similar non-significant trends [8, 20, 21], or no trends . Our findings were the most consistent for men, and apply both to the initial and subsequent visits. For lower income women this applies to frequency but not to the initial visit, possibly due to regular gynaecological and preventive GP consultations leading to levelling between SES groups.
Second, GP services utilisation seemed highest in the low middle income group, a tendency found in the study by Jensen as well . We made similar findings for education, but only in men. Since health status follows a continuous gradient from the better-off to the worse-off, this break of the curve is not likely to reflect differences in need. In Sweden, more people in the lowest income group reported needing but not seeking care . In fact, looking at the lower income and educational groups separately, the utilisation profile may resemble middle- and low-income countries [10, 11], indicating barriers due to absolute rather than relative deficiencies. Financial (user fees, transport costs), organisational (inflexible service delivery) and cultural barriers (language, communication, stigma) might have an impact [6, 23]. Difficulty in paying for health care has been reported for 7% of patients in Norway . For men in the lowest income group the probability of an initial visit was lower, not only to a GP, but also to a somatic specialist (Table 3). In a gate-keeper system the initial GP visit is crucial, making up the basis for access to additional care.
Third, the probability of an initial visit to a somatic specialist was higher among the richer and better educated, first and foremost for women. This is noteworthy since health is worse in lower SES groups. Our findings are consistent with previous research [8, 20, 22] and the inverse care law . Former studies from Norway have shown the same tendency for private but not consistently for public specialist visits [21, 26]. Supplementary private insurance could explain some, but not all of this pro high SES bias . Like others, we found no evidence that frequency of somatic specialist visits was influenced by SES [8, 20–22].
Fourth, the probability of an initial psychiatric specialist visit was higher for the better educated, although significant only in women. This contrasts with the general suggestion of higher utilisation in deprived groups [27, 28], but corresponds to findings from Norway  and the US . A possible explanation could be that higher educated groups and women might recognise and accept psychiatric needs more than lower educated groups and men . The opposite trend for income might be a sign of minimal financial barriers, but possibly also an expression of negative influence by psychiatric disease on income (social selection) .
Fifth, since lower SES groups were more frequent users of GP services and less likely to make an initial specialist visit, it may be that more additional somatic care tends to take place in primary services for disadvantaged groups and in specialist services for the better-off. This applies particularly to women (Tables 1 and 2). GP referrals might thus be biased according to SES and gender. This interpretation is in accordance with former findings of higher GP referral rates for the better or more educated [22, 31], and later referral of women . On the other hand, disadvantaged groups may miss out on attending additional somatic care. Also, affluent groups may bypass the GP and achieve specialist care directly. Our cross-sectional data cannot indicate the reasons for these differences in the utilisation of GP and specialist care, neither infer whether follow-up in primary or specialist care is preferable, and for whom.
Our study had several limitations. Despite a high response rate, our sample may not be entirely representative of the general population, as it is well known that women, married/cohabitants, higher socio-economic groups and healthier persons are more likely to participate in population surveys . In Tromsø 6, attendees were older, and the proportions of married/cohabitants and women were higher than in non-attendees [15, 16]. Also, as reflected in our sample the level of education in Tromsø is higher than for Norway as a whole (Table 5) , and thus generalisation must be made with caution. Further, the validity of self-reported data may be questioned, but agreement between self-reported and registered utilisation of health care is generally high . One should also bear in mind the potential for recall bias and underreporting, which will probably be largest for psychiatric services utilisation. We were not able to adjust household income for number in household, though it was adjusted for marital status. A cross-sectional study like the present cannot establish causal relationships, nor explore the quality of care. Finally, we cannot exclude the possibility of unmeasured confounders of the associations between SES and health care utilisation.
Particular strengths of this study were the large sample size, which allowed us to study somatic and psychiatric outpatient visits, and gender, separately. Furthermore, the comprehensive coverage of relevant issues in the questionnaires made it possible to validate variables against each other and to some extent compensate for recall bias and underreporting. For instance, validations of self-rated health made us consider it robust and preferable to adjust for self-rated health in all models, even if this is not an indisputable measure of need . Also, education was a robust variable in our sample since most of the participants had completed their education. The geographic location and availability of health services made Tromsø particularly suitable for this study.