We found a positive GP assessment to be associated with increasing patient age and increasing frequency of attendance. Patients reporting a chronic condition were more positive, whereas a poor self-rated health was strongly associated with less positive scores also after adjustment. The association between patient gender and assessment was weak and inconsistent and depended on the focus. We found no association either with the patients' educational level or with the duration of listing with the GP even after adjusting for patient characteristics.
This project was part of a larger national patient evaluation project, which may have introduced some sources of bias. Thus, all GPs in the involved counties were invited and those who signed in may not necessarily be a representative sample. The method for patient inclusion would ideally secure a random sample of the doctor-seeking part of the listed patients where frequently attending patients, evidently, would be overrepresented. We do not know to what extent GPs forgot to hand out questionnaires or even if they more systematically let some patients out. However, in this study we focused on adjusted associations between assessments and patient characteristics. Selection bias would therefore seem to have a smaller impact than if we had studied actual levels of assessment.
Duration of further education is only a rough indicator of education. We chose this simple indicator because it lends itself better to use in a self-administered questionnaire than more complex indicators. A certain recall bias may have affected indications of frequency of attendance and duration of listing with the GP. Such information bias may be differentiated which would tend to overestimate the magnitude, but not the direction of the associations found [24].
We learned from our pilot-study that many patients did not understand the expression "chronic illness", so we added "or a serious disease lasting more than three months". This probably enhanced the sensitivity of the question but lowered its specificity, resulting in overrepresentation of more trivial conditions. In order to be able to compare evaluations by patients with a genuine chronic disease with that of those with no chronic conditions, we divided the respondents into three groups: those reporting no chronic condition, those reporting a cardiovascular, respiratory, endocrine or cancer diagnosis (ICPC-2-categories K, R and T and the ad hoc category C), and those reporting other or multiple diagnoses. This may have resulted in the exclusion of some very ill patients with for instance cardiovascular disease in addition to diabetes thus tending to underestimate the significance of suffering from a chronic condition.
This study enjoyed a very high statistical power with over 27,000 cases included. We were therefore able to detect quite small, statistically significant associations. Some statistically significant associations were so small that their clinical relevance could be questioned. However, the considerable power of our analyses is accompanied by an almost negligent risk of overlooking associations (type II-error).
While some earlier studies have presented diverging results on the association between patient gender and assessment [6, 13, 28], we found only small and inconsistent associations, which is in concordance with a meta-analysis performed by Hall et al. [11]. This finding may be rooted in the absence of any gender influence on the way patients experience health care or in the GPs' possible intuitive adjustment of their care to the different needs of different patients [9].
The adjusted analyses showed a strong positive association between patient age and assessment level, which is also a consistent finding in other studies [6, 11]. This association may be rooted not only in the long-standing relationship with their GP and a higher age-related morbidity, but also in a more realistic view on health, health care expectancies and doctors' skills due to the patients' life experience. The finding may also be due to a general positivity of – some may say more mellow way of judging by – older people [29]. However, we could also be facing to a cohort-effect which, though, is less probable considering the linearity of the association.
Crude analysis showed an expected negative association between educational level and assessment scores, which is in concordance with earlier findings [11]. In our study, however, the association was eliminated after adjustment except for the heterogenous group of patients undergoing education and patients who were unable to report the length of their education, which indicates that the association may have been confounded by other characteristics. This difference may be due to the use of different methods for measuring educational level, but it may also reflect that associations found in one cultural setting may not necessarily be valid in another.
Frequency of attendance is a multifaceted variable. For example, we do not know if a high number of encounters is the result of the patient's or the GP's initiative (ex. half-yearly control-appointments for chronic disease). Still, it is an indicator of the intensity of the doctor-patient relationship, just as the duration of listing with the GP is an indicator of the relational continuity between the GP and his patient [30]. Patients' age, time on the GP's list, frequency of attendance and health are closely interconnected. Our adjusted PRs therefore capture a more "clean" effect on the assessment of being listed for years with the same GP and of the frequency of attending the GP. Adjustment of the latter for health ensured that the positive assessment was not an expression of the ill patient relying on the quality of the GP care [28].
Continuity is one of the core qualities of the doctor-patient-relationship in a health care system where the GP is the patient's primary contact with the health care system [31]. The possible migration of dissatisfied patients from the GPs' lists favours a positive association between assessment and time on the GPs list. However, unlike Hjortdahl and Laerum [32] we found no association with the duration of the relationship but a positive association with the intensity. This may indicate that the positive association between continuity and assessment demonstrated in earlier studies [10, 33, 34] may be correlated with other characteristics which we adjusted for in the present study.
We found diverging results regarding the association between health and assessment of the care depending on whether we looked at self rated health or diagnosed chronic illness. In a paper by Rahmqvist [14] this was very well illustrated. The health indicator used in this study was a mix of self and physician ratings and the study found no association between the patients' health and their rating of the care. Hall et al. [35] also used a mixed health indicator with a seeming emphasis on self rated health indicators and found that poor health was associated with dissatisfaction.
Both Hall et al. [36, 37] and Wensing et al. [38] found an association between less positive assessments and poor self-rated health. We also found this strong negative association between self-rated health and assessment after adjusting for confounders, but we also found that patients who reported a chronic condition assessed more positively in all dimensions except accessibility with patients suffering from cardiovascular, respiratory, endocrine and cancer diseases giving the most positive assessments. This is an assessment paradox because GPs received more negative assessments from patients with a poor self-rated health, but more positive assessments from patients with chronic illness. This was also found by Zapka et al. [39]. This may be due to our adjusting for self-rated health which may be somewhat risky when dealing with chronic illness. On the other hand, a possible explanation may be that the GPs were more capable at handling patients with exact diagnoses and maybe even capable of improving their self-rated health, than at handling patients who rated their health as poor did not fit into a specific disease category – e.g. patients with somatization disorders [40]. We may have been demonstrating an effect of the clinical recommendations on the handling of different chronic diseases that have been implemented in Danish general practice through the past few years. All in all, these results illustrates how the use of different health indicators may affect the association between health and assessment and that it is crucial to specify how health is measured whenever this parameter is being used.
We only included a limited number of patient variables. Inclusion of more and specific variables reflecting psycho-socio-cultural aspects might have added value to the study; in particular, it might have helped explain the oppositely directed associations between self-rated health, chronic conditions and assessments.
If patients' demands, expectations and experience, which are determinants of satisfaction in most models [6, 7], were always in balance we would probably see no assessment variation between patients with different characteristics. But, as we have demonstrated, assessments do vary with patient characteristics. In this paper we chose to publish the crude as well as the adjusted results of the analyses of possible associations between patient characteristics and evaluations. This kind of results serves to point out possible quality deficits in general, and to serve this purpose the results need to be adjusted for possible confounders. However, our study offers no possibility for deciding whether the source for evaluation differences between groups of patients is embedded with the patient or with the care and hence with the GP.
Whether or not the results from patient evaluations of care providers should be adjusted for uneven distribution of patient characteristics also depends on the purpose for which they are produced. Adjustment for patient differences may produce a more fair comparison between GPs, and when patient evaluations are used for accreditation purposes it may also seem fair to adjust for differences in the evaluated GPs' patient populations. Yet, adjustment may also blur the assessment of GPs' ability to meet the needs of the populations actually served [9, 11] and thus render quality improvement at a GP level difficult.