Factors Associated with the Use of Primary Care Centres in Persons with Schizophrenia

Background. Patients with schizophrenia and related disorders have more physical problems than the general population. Primary care professionals play an important role in the care of these patients as they are the main entry point into the healthcare system. We aimed to identify patient, general practitioner, and primary care centre factors associated with the number of visits of patients with schizophrenia and related disorders to general practitioners. Methods. A descriptive, cross-sectional study was conducted in 13 primary care centres belonging to the Clinical Management Unit of Mental Health of the Regional Hospital of Málaga, Spain. The eligible population was composed of patients with schizophrenia and related disorders attending these primary care centres in the study area. Our dependent variable was the total number of general practitioner visits made by patients with schizophrenia and related disorders during the 3.5-year observation period. The independent variables were grouped into three: a) patient variables (sociodemographic and clinical), b) general practitioner variables, and c) primary care centre characteristics. Descriptive, bivariate, and multivariate analyses using the random forest method were performed. Results. A total of 259 patients with schizophrenia and related disorders, 96 general practitioners, and 13 primary care centres were included. The annual mean was 3.9 visits per patient. The results showed that general practitioner age, patient gender, patient marital status, primary care centre to which the patient belonged, taking antipsychotic medication, presenting any cardiovascular risk factor, and frequency of mental health training sessions at the primary care centre were associated with an increased number of visits to general practitioners. Conclusions. The only general practitioner characteristic that was associated with the number of visits was age. There were also patient variables involved in the number of visits. Finally, mental health training for general practitioners was important for these professionals to manage patients with

schizophrenia and related disorders.

Background
Patients with schizophrenia have high levels of medical comorbidity and cardiovascular risk factors [1,2]. Thus, the lifespan of these patients with severe mental illness (SMI) is shorter compared to the general population [3]. This is due not only to their particular lifestyle [4,5,6,7,8] but also to the antipsychotic medication they are taking [8,9,10].
Primary care has multiple benefits for patients in general and for patients with a complex chronic illness specifically, and a relationship has been demonstrated between availability of primary care and survival in patients with schizophrenia and related disorders (SRDs) [11]. However, the attention of these patients to their physical problems by primary care continues to be deficient. In previous work, it was found that patients with SRDs had significantly lower recorded rates of cardiovascular disorders [12,13].
It seems that general practitioners (GPs) stigmatise mental illnesses more than psychiatrists when the attitudes of different categories of health professionals were compared [14,15,16]. This perception could influence their decisions in daily practice [17], which can be detected by patients [18]. In addition, feeling unprepared, having low confidence, and a lack of resources to manage mental illness from GPs [19,20] could encourage GPs to transfer these patients to specialised care [21]. Both factors could explain the low contact of patients with SRDs with primary care services [16,21,22]. On the other hand, mental health professionals often do not feel confident in prescribing physical health medication [23]. However, it has been seen that when there was a collaborative work model between mental health and primary care services, GPs felt more confidence in managing patients with SMI [21], had fewer misperceptions of mental illness, and were less inclined to stigmatise it [24]. Therefore, the excess of mortality of patient with SRDs is due not only to their lifestyle and antipsychotic medication but also to poorer access to and quality of received health care, negative perceptions of people with SMI, and lack of awareness of increased cardiovascular risk in these patients by some health professionals [3,25]. Taking this into account and bearing in mind the important role of GPs as a main entry point into the healthcare system and as specialists in physical health problems, it is essential to study whether the characteristics of these professionals could influence the care of patients with SRDs. In a previous study, we analysed the association between patient factors and primary care centre (PCC) variables with the number of SRD patient visits to GPs [22], but we lacked an important link-the role of GP characteristics in the contact. In this study, we analysed the association of both the patients and PCC variables as well as GP variables with the number of SRD patient visits to GPs.

Methods
An observational, cross-sectional study was conducted.

Study area and temporal scope
The study was carried out in the Community Mental Unit of Mental Health (CMU-MH) of the

Eligible population and sample Patients
The eligible population was composed of the patients with SRDs included in the Málaga Schizophrenia Case Register (RESMA). The RESMA was a case register of patients with SRDs who were attending the CMU-MH Area Regional Hospital of Málaga to improve the care of patients with SMI [26].
The samples consisted of patients with SRDs included in the RESMA (n = 1,663) [27]. The sample size was calculated to be representative of the patients of each PCC. Patients were selected with simple random sampling, stratified by PCC. Inclusion criteria were: a) age over 14 years, b) having a clinical diagnosis of SRD according to the ICD-10, and c) being in contact with a PCC in the coverage area of the CMU-MH of the Regional Hospital of Málaga. Exclusion criteria were: a) being in treatment outside the study area, b) having died during the follow-up, and c) not having a computerized medical history in their reference PCC.
Of the patient sample included in Castillejos et al. [22], only those patients who had information about their GP were included.

General Practitioners
We obtained information from 96 GPs in total, all of whom worked in the study area.

Study variables
Our dependent variable was the total number of GP visits made by patients with SRDs during the 3.5-year observation period.
The independent variables were grouped into three groups: a) patient variables (sociodemographic and clinical), b) GP variables, and c) PCC characteristics (see Table   S1).

Data sources
Data on the total number of GP visits were obtained from digitised primary care records (DIRAYA program). Sociodemographic and clinical information about patients was obtained from RESMA [26,27]. Information about the PCCs was collected from their directors, who were interviewed using a questionnaire designed by the research team of this study.
Finally, GPs were interviewed using another questionnaire also designed by the research team [27].

Data analysis
The categorical variables used in the descriptive analysis were frequency distribution and percentages. Descriptive statistics (mean, standard deviation, and median) were calculated for the continuous variables.
We used a bivariate analysis to examine the relationships between the dependent (number of GP visits) and independent (predictive) variables. For categorical variables with two categories, the Wilcoxon test was used, and for categorical variables with more than two categories, the Kruskal-Wallis test was used. Finally, for continuous explanatory variables, Spearman's correlation coefficient was calculated. Differences were considered significant at p < .05.
The classic multivariate analysis (linear regression) is hindered by the large number of predictor variables in relation to the relatively small sample size of the study. We used the random forest technique to solve this problem.
Random forest (RF) is one of the so-called ensemble methods for predictive regression models, which is based on the methodology of classification and regression trees (CART).
With the RF procedure, a large group of trees is generated in which each tree predicts the result of the dependent variable. In the process of constructing the set of trees, RF uses two types of randomization. First, each tree is constructed using a subsample of the database. The size is approximately two-thirds of the initial sample. In addition, a second level of randomness is added by selecting for each tree a random sample of the predictive variables in each node to choose the best division.
The number of predictors selected in each node and the total number of trees that are built are the two main parameters of the RF algorithm. A third parameter is the minimum size of each node terminal.
As a result of the use of a subsample of training data, about one-third of the samples are omitted when building each tree. These are the so-called out-of-bag samples (OOB), which can be used to evaluate the prediction error and to construct measures of importance of the predictive variables.
We used the permutation index of importance to evaluate the importance of the predictive variables. The importance of a variable was assessed by estimating the change in the prediction error that occurs when, in the OOB data, the variable is randomly swapped while the others remain unchanged. The calculations were made tree-by-tree as the random forest was constructed. If a variable was important in the problem analysed, changing its random values led to larger changes in the prediction adjustment compared to those variables that were not important.
The variables to be included in the model were selected so that it explained the highest percentage of the variability with respect to the result, assuming the least possible prediction error. To do this selection, the VSURF package in the R software was used, which followed a two-step strategy. First, the variables of the initial model (all) were classified according to their importance, and those that did not exceed a threshold were eliminated, which was determined based on the standard deviation of that importance. In a second step, with the variables selected in the previous step, a series of nested RF models were built, and the one that generated a lower OOB error was selected. The variables selected will be important and will have a strong relationship with the dependent variable.
For missing data that did not exceed 15% of the total data of the variable, the miss forest package in R was used, which can be used to impute continuous and/or categorical data including complex interactions and nonlinear relationships. Variables presenting more than 15% of missing data were eliminated.
We used the Random Forest R package and its default parameters, and then these hyper parameters were tuned to obtain the least possible error.

Results
Of the total sample of 494 patients included in the study carried out by Castillejos et al. [21], 259 were included. Therefore, the total sample in the present study consisted of 259 patients and 96 GPs. Regarding the patients included, the sociodemographic profile was as follows. The mean age at the beginning of the study was 43.88 years (range: 22-80); 63.71% were male; 70.66% were single; 43.63% completed only primary school; 55.21% were living with their original family or other relatives; 37.84% were unable to work; 88.03% were living in urban areas; and 89.19% were living in non-socioeconomicallydeprived areas. Most patients had a clinical diagnosis of schizophrenia (69.5%), 82.24% of whom had a low or medium global level of severity, 60.23% had at least one cardiovascular risk factor, and 69.11% were not taking antipsychotic medication (see Table 1).  Tables 2 and 3). agreed with the good communication between GPs and social workers (see Table 4). The mean number of patient contacts with their GPs during the study period was 13.53 (median = 11; range: 0-69). The annual mean was 3.9 visits per patient.
The bivariate analysis showed that the patient factors associated with more contact with GPs were: being a woman (p ≤ .001), being married or with a partner (p = .0038), having no formal education (p = .011), having a schizoaffective disorder diagnosis (p = .0213), having any cardiovascular risk factor (p ≤ .001), and taking antipsychotic medication (p = .0011; see Tables 5 and 6).  Table 6 Bivariate analysis between categorical variables with more than two categories and number of contact with general practitioners. In addition, the GP factors that were clearly associated with more contact with GPs were as follows: GP age was negatively associated, as older GPs were associated with less contact (p ≤ .001); and the training variables were also positively associated, such that GPs who had a perception of higher level of training in mental health received more contact (p = .0219; see Table 8). Table 8 Bivariate analysis between continuous variables and number of contact with general practitioners. and Carlinda (15) were the PCCs that presented more visits to the GPs (see Table 6).
There were others PCC factors that almost reached statistical significance: GP and nurse roles in managing patients' mental health, such that higher punctuation in GP role was positively associated with amount of contact (p = .053), but higher punctuation in nurse role was negatively associated (p = .0589); and communication between GPs and social workers was positively associated with amount of contact with GPs (p = .0508; see Table 6).
Regarding the multivariate analysis, based on the method explained previously in the data analysis, we achieved the model with the lowest OOB error with seven variables (see Fig. 1). The selected variables were: GP age, patient gender, patient marital status, PCC to which the patient belonged, taking antipsychotic medication, presenting any cardiovascular risk factor, and frequency of mental health training visits to the PCC (see Fig. 2). Therefore, these variables were strongly associated with an increase in the number of visits to GPs in the multivariate analysis.

Discussion
In a previous study carried out by our research group, we analysed the relationship between patients with SRD/PCC characteristics and the number of visits to GPs [22]. The novelty of the present study is in the association of characteristics of the GPs who treated these patients with the amount of contact with GPs, which led to important findings. GP age, patient gender, patient marital status, PCC to which the patient belonged, taking antipsychotic medication, presenting any cardiovascular risk factor, and frequency of mental health training visits to the PCC were associated with an increased number of visits to GPs. In addition, a novel statistical analysis technique based on the random forest method was used to perform a multivariate analysis.
With the subsample analysed, the mean annual number of visits to GPs during the observation period was similar to that found with the entire sample analysed and discussed in the previous study [22].
The variable that was most associated with the number of visits was GP age, such that older GPs had less contact with patients. This could be related to older GPs having a more negative attitude toward patients with schizophrenia than their younger counterparts [28], although not all studies have found this result [24,29].
Women visited GPs more often, which is in line with other studies [11,30,31].
Patient marital status was also related to the number of visits to GPs, so that patients who were married or living with a partner visited GPs more than single patients. This could be because family support causes these patients to receive continued care. We also found that patients who were separated, divorced, or widowed made more visits to their GPs than single patients. Copeland et al. [11] found that the number of visits to GPs by married patients or those with a partner was greater, but no greater number of these visits was found in patients who were separated, divorced, or widowed.
The fourth variable that was more related to contact with GPs was the PCC to which patients belonged. In a previous study performed by our group, we found that GPs working at PCCs in the catchment area of Central CMHU had less stigma about mental illness, coinciding with the fact that this area had been working collaboratively with mental health services for 15 years [24]. However, in this study we found that the PCCs that received the highest number of visits to GPs belonged to both Central and Northern CMHU. These two facts seem to indicate that the stigma presented by GPs toward mental illness was not related to the number of visits they received from patients with SRDs. In addition, working in a collaborative model was not related to the increase in contact with GPs, which is in line with the results found in a previous study in which good communication between primary care and mental health services did not translate into increased contact, and this might be due to mental health professionals assuming some of the functions that otherwise would have been performed by the GPs [24].
Moreover, taking antipsychotic medication and presenting cardiovascular risk factors were also associated with a higher number of visits to GPs. On one hand, the fact that GPs are the health professionals in charge of the follow-up with diseases such as cardiovascular risk factors mean that patients with SRDs who present these problems visit their GP more than those who do not present them. On the other hand, GPs are also in charge of the follow-up in prescribing medication, hence both patients taking antipsychotic medication and those with cardiovascular risk factors (who usually also take medication for this disease) visit their GPs more frequently.
Finally, higher frequency of mental health training visits to the PCC resulted in more visits to the GP. Related to this is that, in the bivariate analysis, GP perception of higher level of training in mental health received more contact. These results are in line with the study carried out by Bagayogo et al. [20], who found that GP felt low confidence in treating complex mental health problems and the need to integrate mental health into primary care. Moreover, it was also found that when there was a collaborative work model between mental health and primary care services, GPs felt more confidence in managing patients with severe mental illness [21]. Therefore, it seems that a collaborative work model between mental health and primary care services at the training level can make GPs feel more capable of managing these patient problems and refer less to specialised services.

Strengths and limitations
This study analysed multiple associations between patient, PCC, and GP characteristics and the number of visits by SRD patients to GPs.
The study was carried out in a wide catchment area, including the participation of 13 PCCs.
We analysed a large, homogeneous sample of patients diagnosed with SRDs living in the community.
We performed a multivariate analysis using a novel statistical analysis technique based on the random forest method.
For unknown reasons, not all GPs in the study area participated, and in some cases, the number of GPs per centre was very low. However, in general, the participation rate was high.
This was a cross-sectional study from which we cannot infer causality but only association.
Clinical diagnoses were not made through clinical interviews but rather by the patients' long-term psychiatrists and updated in the RESMA database.
Information about patients was collected from medical records instead of specific health screening tests of individual patients.

Conclusions
This study analysed multiple associations between patient, GP, and PCC characteristics and the number of visits by SRD patients to GPs. It was surprising that the only GP characteristic that was clearly associated with the number of visits was GP age. It was confirmed that there were patient variables involved in the number of visits, such as gender and marital status. Finally, it was shown that the training of GPs is important for these professionals to manage patients with SRDs. Therefore, a collaborative model between mental health and primary care services should be promoted at the training level. In accordance with the Official Law on Data Protection (LOPD), patient identification information was separated from clinical information. Therefore, informed consent was not required for participants, neither from a parent or guardian for participants under 16 years old.

Consent for publication
Not required.

Availability of data and material
The data that support the findings of this study are available from Berta Moreno Küstner but restrictions apply to the availability of these data, which were used under license for the current study, and so are not publicly available. Data are however available from the authors upon reasonable request and with permission of Berta Moreno Küstner.

Supplementary Files
This is a list of supplementary files associated with this preprint. Click to download.