We conducted a population-based longitudinal study in order to examine the impact of transitioning from a traditional fee for service (tFFS) model to the enhanced fee for service (eFFS) model. This study looked at outcome measures on a yearly basis before and after practices adopted the eFFS model using health administrative data housed at the Institute for Clinical Evaluative Sciences (ICES) from April 1st, 2000 to March 31st, 2013. This study was approved by the institutional review board at Sunnybrook Health Sciences Centre, Toronto, Canada and the Queen’s University Health Sciences and Affiliated Teaching Hospitals Research Ethics Board, Kingston, Canada (6015466).
From 2002 to 2006, the Ontario government introduced various new primary care models with differing physician payment and organizational structures to encourage physicians to shift away from the traditional FFS model (Table 1).
In an attempt to promote patient rostering, the Ontario government introduced two eFFS models – The Family Health Group and the Comprehensive Care Model. In the eFFS model, physicians are strongly encouraged to roster patients, but are not required to offer enrolment to all patients . Physicians in this model receive the majority of their payment through traditional fee-for-service billing, although additional premiums can be obtained for delivering specific preventive care (e.g., pap smear, mammograms, flu shots, colorectal screening) and chronic disease management services (diabetes and chronic heart failure) for rostered patients only . Physicians receive a fee for each patient they roster in their first year ($5 per patient) and a more substantial payment (i.e., $110 - $180 depending on patient age) for enrolling new patients that do not have a family physician (i.e., ‘orphaned’ patients) . Furthermore, physicians receive a small monthly comprehensive care fee per rostered patient and a 10% increase in the amount paid for included core comprehensive FFS codes . Physicians are contractually obligated to provide after-hours care for rostered patients, and those in Family Health Groups must work in a group of three or more physicians.
All health administrative databases required to carry out the analyses in this study were stored at the Institute for Clinical Evaluative Sciences (ICES). These datasets were linked using unique encoded identifiers and analyzed at the Institute for Clinical Evaluative Sciences (ICES) at Queen’s University. Databases at ICES have the advantage of near-complete population coverage (the lowest is OHIP with approximately 94% of visits) .
Family physicians that transitioned to eFFS were identified along with their profile using the ICES Corporate Provider Database (CPDB), which captures physician socio-demographic information, their practice model, and location. The Ontario Health Insurance Program database captures all provider billing claims for the provision of care to residents of Ontario who are eligible for insurance coverage. The OHIP database was used to obtain information on referrals to medical specialists.
The Registered Person’s Database captures patient demographic information, including age, sex and postal code for those that are eligible for health insurance coverage in Ontario. The National Ambulatory Care Reporting System (NACRS) provides information on all emergency room encounters.
An open cohort of family physicians who transitioned from tFFS to eFFS was created. Physicians in the cohort were followed longitudinally, with their exposure and outcome (see Study outcomes) data being tracked on an annual basis before and after they transitioned to an eFFS model.
Specifically, we identified family physicians who transitioned from tFFS to an eFFS model at any point between April 1st, 2003 and March 31st, 2013 using the ICES Corporate Physician Database, which contains information about their practice model, location, and sociodemographic characteristics. We excluded physicians who were not providing comprehensive family medicine during a given study year (i.e., identified as a specialist in the Corporate Provider Database or billed OHIP for fewer than 8 of the 18 standard primary care fee schedule codes within a given year), had fewer than 100 patients under their care, or had a prolonged absence during a given study year (8 weeks or greater). Furthermore, using the primary practice location for each physician using the Corporate Provider Database, we limited the study to urban physicians as 78% of physicians that transitioned to eFFS were practicing in urban centres and since there are significant contextual differences related to access based on rurality. Also, as mentioned above, many physicians that transitioned to an eFFS model subsequently switched to a capitated model.
For each study year, we identified the group of patients that were under the care of individual study physicians. Patients were included in the study if they had a valid Ontario Health Insurance Plan number and were alive and attributed to a study physician as of March 31st of the fiscal year being examined. Patients were attributed to the physician that billed the largest dollar amount of primary care services for their care over a 2 year period (‘virtual’ attribution method) using the Ontario Health Insurance Program database, which captures all provider billing claims for the provision of care to residents of Ontario who are eligible for insurance coverage . Since the Client Agency Program Enrolment (CAPE) dataset only identifies rostered patients for the eFFS practices in this study and not those in the tFFS model, we used the ‘virtual’ attribution method to create provider rosters both pre- and post-transition, despite the fact that the CAPE database tracks official patient rostering for eFFS practices. This method has been used in previous studies and is the accepted reporting method of the Ministry of Health and Long Term Care of Ontario . This was done to avoid differential misclassification that would have resulted from using a different attribution method for patients before and after the transition to eFFS. Previous work done by our group has shown that the percentage agreement between the virtual rostering method and the CAPE database is greater than 85% (see limitations for further discussion) .
In addition, patients were excluded if they did not have a primary care visit to their family physician for two consecutive years (i.e., during the year of interest and the year prior). Eligible patients were subsequently linked to the Registered Person’s Database, which captures patient demographic information, including age, sex and postal code for those that are eligible for health insurance coverage in Ontario.
To ensure the pre-transition phase was adequately captured, we then excluded physicians with less than 4 years of pre-transition data. Similarly, physicians with less than 2 years of post-transition data were excluded. This occurred because the physician transitioned within that period from eFFS to another model, or because they moved from the province. Follow up of physicians was discontinued if they transferred out of the eFFS model into another model (eg, capitation model) type (i.e, if a physician subsequently switched to capitation, only data collected during years that they were in tFFS and eFFS were used in the analysis).
We assessed measures of continuity of care, coordination of specialist care, and primary care access.
Relational continuity of care was assessed using the Usual Provider of Care Index (UPC). UPC is a patient level outcome that looks at the percentage of primary care visits to the main provider relative to all primary care visits (i.e., high UPC = better continuity) over a 2 year period (i.e., fiscal year of interest and the year prior) . Patients with less than three visits over the 2 year span were excluded from the analysis, as data for these patients tend to cluster around 0, 50, and 100%, which has been shown to impact the reliability of this measure . The UPC index is a validated measure that is commonly used to assess continuity .
In order to assess coordination of specialist care, we developed a referral index (RI). RI is a physician level measure that represents the percentage of total primary care referrals for a physician’s roster made by the main provider (i.e, as opposed to referrals made by walk-in physicians or other family physicians). Since diagnostic radiology makes up a large percentage of all referrals and does not represent a traditional referral per se, they were excluded from this metric. Also, referrals to allied health professionals was not assessed in this outcome.
Lastly, access was assessed using non-urgent emergency department (ED) visits. Non-urgent ED visits is a commonly used proxy for primary care access [25, 36, 37]. Specifically, this study looked at the number of ED visits (Source: NACRS) for family practice sensitive conditions (FPSCs) on the patient level. These ED visits are for health conditions that are less urgent and have less than a 1% chance of an inpatient visit, and thus, represent conditions that would more appropriately be handled in a primary care setting . Examples of FPSCs include conditions such as conjunctivitis, otitis media, acute pharyngitis, sinusitis, and acute upper respiratory tract infection. This measure was established by the Health Quality Council of Alberta, and has been used as a proxy measure for primary care access by organizations such as the Canadian Institute of Health Information. Since there was a coding change in NACRS in 2002 that would have impacted this outcome, we only looked at data for this outcome from 2003 to 2013. Since the percentage of individuals across Ontario that have a FPSC ED visit is quite low, and the majority that do, only have a single visit, this outcome measure was treated as a dichotomous outcome.
We used mixed-effects segmented linear and logistic regression models to examine changes in outcomes while controlling for patient and provider contextual factors. This approach divides the data into pre- and post-intervention periods, determining separate intercepts and slopes for each time period . Statistical tests were used to compare the intercepts and slopes of each line to see if the transition to eFFS resulted in a change in outcome measures that was significantly greater than any underlying secular trend. All models accounted for the clustering of patients to providers using a generalized mixed effects model. The intercept, time (measured as a continuous variable in years), type of care model, and time after transition (measured as a continuous variable in years) were all assigned as random effects in all models to deal with the heterogeneity within the data across physicians.
A multivariate logistic regression model was used to assess FPSC ED visits, while a multivariate linear regression was used for continuous measures (i.e., UPC, RI). All models adjusted for both patient (age, sex, socioeconomic status via neighbourhood income quintile, urban/rural residence, case mix) and provider (sex, years since graduation, foreign medical training, and total number of patients under the care of each physician (i.e., panel size)) level contextual factors as they have all been shown to impact access, continuity, and specialist referrals in previous studies [19, 40,41,42,43]. Since the above models assume linear trends over time, descriptive linear plots were constructed (for the overall population and for cohorts belonging to each individual transition year) in order to ensure the data followed a linear trend.
Furthermore, previous studies have shown that a portion of patients (approximately 15%) opt not to formally roster with their physicians after they transition to enrollment models [32, 44]. Since the intent of the Ontario Government was to have all patients rostered, these patients were kept in the main analysis and assessed in the eFFS group even though they were not formally rostered.
A secondary analysis was done to compare the impact of the transition to eFFS on early versus late adopters of the new model, as previous studies have demonstrated differences between both groups . An interaction term for early adoption (i.e, early_adoption = 1 for physicians that transitioned between 2004 to 2006, early_adoption = 0 for those that transitioned from 2007 to 2011) was added to each model to examine if there was a differential impact on early versus late adopters of the eFFS model. Plots of these analyses were created by setting patient and provider factors to their mean and mode values. All analyses were conducted using SAS, Version 9.3, SAS Institute Inc.