Frequent attendance in primary care: comparison and implications of different definitions

Participants

A stratified multistage probability sample without replacement was drawn in the DASMAP study. Replacement was prohibited to ensure that every individual had a known probability of selection. The sampling frame was the seven health regions of Catalonia.

Stage 1 was selection of the primary care centres within each health region (there were 350 primary care centres in Catalonia in 2005). The number of primary care centres to be selected in each region was proportional to the population of the region. However, in order to have a minimum set of interviews even in the smaller regions, at least six primary care centres were chosen per region. The probability of selection of each primary care centre was related to the population of the catchment area covered by the centre. Eighty primary care centres were selected to participate and two refused (97.5% acceptance rate).

In stage 2, all GPs (n = 913) of the primary care centres were invited to participate. A total of 618 GPs (67.7%) participated in the study.

In stage 3, patients were selected from the daily list of all patients with an appointment with each of the participating GPs, using a systematic sampling strategy. A total of 5402 patients were randomly selected. Among these, 654 (12.1%) did not attend the visit with the GP, so they were not invited to participate in the DASMAP study. A total of 4748 were invited to participate. Among these, 764 (16.1%) did not accept the offer to participate and 164 (3.5%) were excluded because they showed cognitive impairment severe enough to preclude an adequate interview, leaving a study sample of 3820 participants from 78 primary care centres. One of the primary care centres was excluded from the statistical analyses because of data loss. Therefore, the analysed sample comprised 3815 patients (80.3% of the patients initially invited) with a mean age of 54.3 years (standard deviation [SD] = 17.31; range 18–97 years). Females represented 63% of the sample.

Definitions of frequent attendance

Three different definitions of frequent attendance were considered, each divided into two according to a 1-year number of visits cut-off: (1) frequent attenders as the top 25% and the top 10% consulting patients during a 1-year period; (2) frequent attenders as the top 25% and the top 10% consulting patients during a 1-year period stratified by age and sex; and (3) frequent attenders as the top 25% and the top 10% consulting patients during a 1-year period stratified by the presence of physical/mental conditions: patients with only mental disorders, with only chronic physical conditions, with comorbid physical and mental conditions, and those that did not have any of the assessed mental/physical conditions. Table 1 shows the different definitions.

Survey instruments

The following battery of instruments was administered to the DASMAP participants:

• Sociodemographic/clinical questionnaire: this questionnaire collected information about the following variables: sex, age, marital status, employment status, educational level, place of birth, and main reason for consulting with the GP in the last visit (physical, mental/emotional, other).

• The Structured Clinical Interview for DSM-IV Axis I Disorders (SCID-I research version: major depression episode, dysthymia, and anxiety disorder modules, excluding obsessive-compulsive disorder).15

• The Mini Neuropsychiatric Diagnostic Interview (manic/hypomanic episodes, obsessive-compulsive disorder, substance and alcohol use disorders, anorexia nervosa, and bulimia nervosa) (MINI).16^,17

• Chronic physical conditions ‘yes-or-no’ checklist: this included asthma, bronchitis, ulcer, constipation, high blood pressure, heart disease, heart attack, stroke, migraines, allergies, arthritis, back pain, neck pain, and diabetes. Responders were asked whether they had ever in their life experienced each of the symptom-based conditions in the checklist. In the case of conditions typically identified by medical diagnosis, responders were asked whether a doctor or other health professional had ever told them they had the condition.

• An adapted version of the Client Socio-Demographic and Service Receipt Inventory (CSRI).18 The CSRI version used in this study was designed to collect data on health services utilisation during the previous 12 months. For the purposes of the present work, only face-to-face contacts between patients and GPs from the public sector were taken into account.

Statistical analysis

Multilevel logistic regression was used, with GP and primary care centre as random factors. Firstly, it was tested whether multilevel logistic regression was more appropriate than usual logistic regressions. When the null models including primary care centre as random factor were compared, likelihood ratio tests were very significant for all models. Secondly, a comparison was carried out between null models that included primary care centre and GP as random factors versus those with only primary care centre; likelihood ratio tests were also significant for the three definitions with the cut-off at the 90th percentile; however, they were not significant for the three definitions with the cut-off at the 75th percentile. That is, for the definitions of frequent attenders at the 90th centile, primary care centre and GP were used as random factors, and for the definitions at the 75th centile, only primary care centre was used as random factor. Intracluster coefficients (ICCs) of random components were calculated as follows:19 σ²/(σ² + π²/3). The ICCs of the primary care centres in the null models were: model 1 (cut-off 75% = 0.0436; cut-off 90% = 0.0518), model 2 (cut-off 75% = 0.0380; cut-off 90% = 0.0573), model 3 (cut-off 75% = 0.0464; cut-off 90% = 0.0436). The ICCs of the GPs were: model 1 (cut-off 90% = 0.1011), model 2 (cut-off 90% = 0.0758), model 3 (cut-off 90% = 0.0886). Then, univariable multilevel logistic regression was performed. The final multivariable models included those variables that had been significant (P≤0.20) in univariable analyses.20 ‘Region’ was retained because of an a priori assumption of clustering within region, although it had few categories (n = 7) that could be considered as random factors.21 The area under the receiver operating characteristic (ROC) curve was calculated to estimate the discriminative power of each model. The analyses were conducted using STATA (version 10).

Summary of main findings

The way frequent attendance is defined affects which variables are associated with this phenomenon. Definitions without stratification or stratified by sex and age provided similar results, whereas a definition taking into account health conditions as well as the cut-off used (75% or 90%) had an impact on the random components of the model. It is of relevance that those models using the 90% cut-off point provided better discrimination between frequent and non-frequent attenders than those using the 75% cut-off point.

Beyond the definition, being on sick leave (highest odds ratio), being born outside of Spain, and reporting mental health problems as the main reason for consulting were associated in all models with increased odds of being considered a frequent attender. The only chronic physical conditions that were systematically associated with high consultation were arthritis/rheumatism and bronchitis.

Strengths and limitations of the study

The study design allowed selection of a sample that was representative of Catalan primary care patients and led to results that could be generalised to this wide population. On the other hand, the data were analysed considering three definitions of frequent attendance (two of them very close to previously reported definitions, and a new one considering wide categories of health conditions). This allows discussion of the concept of frequent attendance, its components, and its implications. In an effort to clarify the impact of mental disorders, well-known psychiatric instruments were administered.

The results of the study should be interpreted with caution because of the following limitations: first, there was no differentiation between whether patients or doctors initiated the contact, that is, it is not known which visits were spontaneously booked by patients and which by GPs (for instance, consultations for checking the evolution of patients with sickness leave). Second, data about health services utilisation were collected by means of a self-report measure; such measures have been criticised for their inaccuracy.22 Third, the study focused on the sociodemographic and clinical characteristics of patients, not taking into account certain GPs' characteristics that might be associated with frequent attendance (age, sex, experience, type of training, style of practice, values, orientations, and attitudes).23^,24

Comparison with existing literature

The variables that were found to be systematically associated with frequent attendance were in general consistent with those reported in other studies.25^,26 The individuals on sick leave (compared to those working) usually have the highest mean number of visits to the GP as well as to other healthcare providers.25 In line with the present results, Scaife et al found that South Asian and African–Caribbean immigrants were more likely to be frequent attenders than British patients.26 A lot of people from underdeveloped countries, characterised by high rates of unemployment, have migrated to Catalonia in the last decade to find employment, mainly in the industry and service sector. A considerable proportion of these immigrants belong to low social classes and suffer social isolation, factors that undermine their physical and mental health, and might explain their pronounced need to consult the primary care services.

Concerning chronic physical conditions, the present results are in line with those obtained in other Spanish regions and other countries.10^,27^,28 One study conducted in the city of Granada, Spain, found that chronic bronchitis was a very common disease among frequent attenders.27 Al-Windi, by means of questionnaire surveys and computerised medical records, demonstrated that musculoskeletal, cardiopulmonary, and head symptom groups were significantly related to high past and prospective primary care consultation rates.28

The main difference between the present study and previous ones is related to definitions of frequent attendance. Definitions 1 and 2 (that do not consider patients' health conditions) are very close to previously published definitions, and are, in general, concordant to previous findings. However, when considering health conditions, results were different, especially regarding mental disorders. The review of Vedsted and Christensen concluded that mental health problems are associated with frequent attendance.1 This finding was not supported by the present definition considering health conditions. According to available clinical guidelines,29 mental disorders need frequent follow-ups. Not taking this into account could promote an over-estimation of the impact of these conditions on frequent attendance.

Mental illness is still stigmatised by both society and health professionals, who show adverse attitudes toward it.30 When defining frequent attendance adjusted by health conditions (definition 3), no mental disorder was associated with frequent attender status. However, it is possible that this could be related to some classification bias or to the way in which mental disorders were assessed in the present study. The SCID-I was used, which is more restrictive than criteria used in other studies (for example, assessment of depressive symptoms with general questionnaires). The fact that mental health problems as the main reason for consulting were systematically related to increased odds of being a frequent attender is in line with this. It is possible that depressive symptoms or emotional distress could explain frequent attendance, but it is important to notice that some depressive symptoms, like sadness or anhedonia, or a certain degree of emotional distress are not mental disorders.

Implications for future research

Two main implications could be derived from the present study. First, future studies should take into account the intracluster correlation of GP and primary care centre. Second, the way in which frequent attendance is defined has an impact on the factors associated with it and their discriminative power. The use of the top decile cut-off seems to be more recommended than the top quartile. Finally, it is important not only to take into consideration the patients' type of pathology, but also to make a subsequent explicit reflection about the possible reasons for the high rate of consultations of each frequent attender.31