STUDY DESIGN AND DATA SOURCE
We conducted a retrospective longitudinal observational analysis of medical consultations in PHC and referrals to SC. This sample combines information from 504.940 patients (205.961 men and 298.508 women), 2.414.508 medical consultations, 284.754 referrals to SC, and 837 physicians working non-concurrently in one health district between January 2015 and December 2018. Patients’ consent was not necessary since only anonymized information was used during the study and the RJ-MHD, the actual caretaker of this information, gave the consent to use this dataset for this research. The study was approved by the RJ-MHD research ethics board and it is registered under the number 03795118.0.0000.5279. It was conducted in accordance with the 466/12 resolution from the Brazilian National Health Council26 and the Declaration of Helsinki.
EXPOSURE
Physicians were divided into two categories: (1) Generalists - the reference category aggregating doctors without residency training in FM; and (2) Family physicians (FP) - graduated family physicians, FM preceptors and residents enrolled in the FM residency programs. Residents in FM were included in the same category as FP because they spend two years working 48 hours a week in a community-based primary care clinic under the full supervision of a senior FP (FM preceptor), sharing responsibilities for the same patients in one FHT. Every week they have learning sessions developed by the faculty members23 using active learning methods27,28 to address topics of FM and PHC, such as clinical reasoning, management of the most prevalent health conditions in PHC, communication skills, evidence based-medicine, PHC and health care systems, vulnerable populations, elderly care, multimorbidity, polypharmacy, among others.29 They also have rotations in maternal care, pediatrics, internal medicine, and emergency care. These activities were designed in line with the National Committee for Medical Residencies (CNRM)30 and with the Brazilian Society of Family and Community Medicine (SBMFC).31 Information about other forms of post-graduate training were not available in the database and were not taken into account, nor the number of years in practice for any doctor.
INDEPENDENT VARIABLES
Every patient contributed to the models with individual information – (1) age (linear), (2) sex, and (3) the Charlson Comorbidity Index32 – and contextual information – (4) the Social Development Index (SDI). The SDI is a linear scale combining information about sanitation, schooling, income, and housing conditions from every household in the FHT catchment area, representing the grade of social development of a neighborhood.33 Hence, patients registered in the same FHT have the same SDI. It varies from 0 (least developed) to 1 (most developed).
Charlson Comorbidity Index32 was used to add information about patients’ morbidity burden to the models, assuming that those with more chronic conditions would be more likely to be referred to secondary care and have follow-up consultations after being referred.
Time effects were regarded using dummy variables for months and years in all models. A dummy variable was used to include information identifying if the consultation was a prenatal care visit or not.
All clinics in this sample have the same physical structure, offices equipped with computer, printer, medical equipment, room for small surgical procedures, the same arsenal of laboratory tests and medicines in the pharmacy, and the same type of human resources available: nurses, technicians, dentists, pharmacists, and managers. The availability of medical specialties in SC and diagnostics tests, and the referral procedures are the same for all doctors and clinics in the sample. The distribution of doctors among different clinics and FHT didn’t follow any criteria that could interfere in the relationship between the medical categories, the population assisted, and the study outcomes.
OUTCOMES
Referrals to SC were divided into three groups: (1) outpatient consultations for ambulatory care; (2) surgical evaluation; and (3) diagnostics tests. They were considered as a binary event (referred versus non-referred). To estimate the relative risks (RR) of having a follow-up visit in the PHC clinic after the referral, only patients who had been referred to the specific specialty under analysis were considered. Follow up visits were also considered binary events categorized as the patient having or not having one medical consultation (a) 90 days after the referral or (b) 180 days after the referral. The 32 most commonly requested medical specialties consultations and diagnostics tests in our dataset were used to perform this analysis.
Comparing both the risk of a patient being referred to SC and the risk of having a follow-up visit by doctors with different types of training can bring us evidence about the effect that RTFM has on promoting both a more effective healthcare for the patient in primary care and a better continuity of care and coordination between primary and secondary levels of care. This notion is aligned with the definition of FM from the Brazilian,31 Canadian34 and European35 curricula for FM, i.e., that experts in FM “are skilled clinicians that are capable of managing a full range of health conditions”, “make efficient use of health care resources through coordinating care” and “are responsible for the provision of longitudinal continuity of care as determined by the needs of the patient.”
STATISTICAL ANALYSES
Multilevel multivariate binomial regression models were used to estimate the RRs of patients being referred to SC in one medical consultation and patients having a follow-up visit in three and six months after being referred to SC, according to the medical category of the doctor in charge. A hierarchical data structure was created with consultations from the same patient clustered and ordered per each individual patient, taking into account the correlation among consultations from the same patient.
Each outcome was analyzed individually. Mammography, Gynecology and Gynecologic Surgery entailed just women and high-risk prenatal care (HRPC) entailed only pregnant women. Models were adjusted for first level covariates (consultation), i.e., patient’s age, patient’s Charlson Comorbidity Index, prenatal care consultation, time, and medical category; and for second level covariates – SDI and patient’s sex.
Variance partition coefficients were calculated for all adjusted models in order to explore the proportion of the variance attributed to the second level, i.e., the variance attributed to patients characteristics.36
PAF for each requested service was calculated using the RR from the multivariate regression models to estimate the impact in the number of referrals requested per year in the same health care district if all medical consultations were performed by trained FPs.37,38 Data processing and statistical analysis were performed using R version 3.6.2 and lme4 package.