Abstract
Objective To determine which patient, physician, physician-group, and geographic factors influence the likelihood of health care visits to family physicians (FPs) outside of patients’ enrolling physician groups.
Design Cross-sectional analysis using administrative health data. A multilevel logistic regression model was used to determine the probability of outside use based on a set of observable factors at the patient and physician-group levels, accounting for the variation attributable to the patient, physician, physician-group, and geographic levels.
Setting Ontario.
Participants Patients enrolled in family health organizations (FHOs) during the study period from April 1, 2018, to March 31, 2019.
Main outcome measures Outside use, defined as an encounter with an FP outside of the patient’s enrolling physician group who submitted an Ontario Health Insurance Plan fee code in the FHO core primary care services.
Results Patient-level factors explained 83.7% of the variation in outside use probability. Physician- and physician-group–level variations each explained less than 2% of outside use probability, while 14.2% was explained at the physician-group geographic level. Patient-level factors associated with outside use included age (oldest versus [vs] youngest cohort, odds ratio [OR]=0.44, 95% confidence interval [CI] 0.42 to 0.45); female sex (OR=1.21, 95% CI 1.19 to 1.23); expected relative health care costs (highest vs lowest complexity score quintile, OR=3.65, 95% CI 3.53 to 3.79); distance from enrolled physician group (longest vs shortest travel time, OR=2.10, 95% CI 2.01 to 2.19); and FP to population need ratio (highest vs lowest quintile, OR=1.71, 95% CI 1.62 to 1.80). Statistically significant physician-group–level variables (P<.05) included Rurality Index for Ontario scores, group size, years of existence, proportion of female physicians, average age, and number of weekend and holiday days worked per patient.
Conclusion Physician- or physician-group–level policy options for effectively reducing outside use may include increasing group size to more than 5 and increasing weekend and holiday days worked per patient and after-hours care. However, because outside use is primarily explained by variations at the patient level, other innovative policy options may need to be implemented to improve care continuity.
About one-third of patients in Ontario are enrolled in a family health organization (FHO) for primary care.1 The FHO is a primary care physician payment model in which physicians work in groups of 3 or more and maintain a roster of enrolled patients. Physicians in FHOs are paid through capitation.
Patients are free to seek care from any practice in Ontario. Still, several incentives are in place to improve continuity of care, which is recognized as a key element of high-quality primary health care and which has been associated with better patient outcomes and greater satisfaction.2 One such incentive, the access bonus, is awarded to physician groups whose patients have low levels of outside use, defined as patients receiving core primary care services from physicians outside of the group in which the patients are enrolled. Patients are not offered direct financial incentives to seek care from their enrolling physician group. The underlying assumption is that physicians can influence patient behaviour by increasing accessibility or otherwise encouraging patients to consistently seek care from their enrolling group.
Previous research on the effectiveness of the access bonus has found mixed results. A longitudinal study found that physicians who switched to a payment model with an access bonus had lower patient-level outside use (ie, improved continuity of care) than those who did not switch.3 A cross-sectional study found that in large urban settings, but not in small ones, higher access bonus payments were associated with improved telephone access, after-hours access, timeliness, and wait times.4 Another cross-sectional study found that higher access bonus payments were associated with fewer after-hours visits, greater emergency department (ED) use, and higher ambulatory care costs.5 The authors interpreted these findings as reflecting differences in rural and urban practice patterns.6
In this study, our goal was to address the gap in policy-relevant evidence by using a comprehensive set of patient and physician-group variables to explain outside use. This included multiple measures of physician-group geography, which have been noted in the literature.4,5 We used multilevel modelling to quantify the relative variation in outside use at the patient, physician-group, and geographic levels.6-8
METHODS
Study population
The study population consisted of Ontario residents enrolled with an FHO physician during the 2018-2019 fiscal year (April 1, 2018, to March 31, 2019). We included only those patients who were enrolled with their physician for the entire fiscal year and excluded those with certain data anomalies (eg, age >130 years).
Data sources
We obtained de-identified, linked administrative health care data under a data-sharing agreement between the Ontario Medical Association (OMA) and the Ontario Ministry of Health (MOH), in compliance with Ontario’s Personal Health Information Protection Act. The data sources used in this study are described in Appendix Table 1, available from CFPlus.* Ontario road network data were obtained from publicly available OpenStreetMap services to estimate travel time.9
Main outcomes measures
Our outcome measure, outside use, was defined as an encounter with a family physician (FP) outside of the patient’s enrolling physician group who, as a result, submitted an Ontario Health Insurance Plan fee code in the FHO basket of core services (Appendix Table 2, available from CFPlus*) for services other than those provided by FPs in a focused practice in palliative care, addiction medicine, psychotherapy, or similar areas.
Exposure variables
We included predictors at the patient, physician-group, and geographic levels (see Appendix Table 3, available from CFPlus,* for a description of all the variables). Continuous variables were treated as ordinals to allow for nonlinear relationships and ease of interpretation. These were based on prespecified cut-offs where appropriate and quintiles otherwise.
Patient-level variables included age in years on the last day of the study; sex (male or female); complexity score; patient-physician relationship metrics; and provider availability. The complexity score reflects individuals’ expected relative health care costs. It is assigned using the Canadian Institute for Health Information (CIHI) Population Grouping Methodology, which is applied to diagnostic data from physician encounters, ED visits, and hospital inpatient stays collected over the 4 years before and during the study period.10-12
Patient-physician relationship metrics included the number of years the individual had been enrolled with the current group; the number of past enrolments; and travel time to the enrolling group. Provider availability includes the ratio of FPs to population need and the ratio of emergency services to population need. These ecological variables are defined at the level of patients’ postal codes and reflect the availability of health care providers.
Physician characteristics were defined at the group level because the access bonus is implemented at that level. Physician-group variables included number of years the group had been in existence; group size (the number of physicians in the group throughout the study period); average roster size; physician demographic characteristics (age and sex); access metrics (the number per patient of weekdays worked, weekend and holiday days worked, and weekday after-hours shifts worked); and rurality (based on Rurality Index for Ontario [RIO] scores assigned to the group postal code).
Geographic levels were identified as physician groups’ postal forward sortation areas (FSAs), that is, the first 3 digits of each postal code. There are 521 FSAs in Ontario, with populations from approximately 1000 in remote or highly commercial areas to 110,000 or more in densely populated urban areas.13
Statistical analysis
Multilevel logistic regression analyses were performed on a random 5% sample of the data. This allowed estimating the relative magnitude of variation in outside use at the patient, physician, physician-group, and geographic levels. We used a random sample to reduce the computational demands of the analysis.
Intraclass correlation coefficients (ICCs) were calculated by dividing the variance at each level by the total variation in outside use, indicating the proportion of unexplained variation attributable to each level.
We began with a 4-level intercept-only model, nesting patients within the physician, physician-group, and geographic levels. We excluded from the final model any levels with less than 3% variation and ran subsequent models using the remaining levels and the exposure variables.
We used SAS, version 9.4, for database manipulation and STATA 16.0 for statistical analyses.
Ethics approval
Formal ethics approval and informed consent to participate were not required because we used de-identified administrative health care data obtained from the Ontario MOH under an agreement with the OMA. In addition, the research was initially conducted as part of OMA business operations.
RESULTS
We identified 5,742,406 patients enrolled with 1 of 4903 FHO physicians during the study period. Observations were excluded according to sample selection criteria (Appendix Table 4, available from CFPlus*). We found that 78.8% of the patients had some primary care use; of these, 25.6% had received outside-use primary care (Figure 1).
Family health organization–enrolled patients in Ontario, April 2018 to March 2019: A) With and without primary care visits and B) those with primary care visits within and outside of their enrolling group.
All patient and enrolling physician-group characteristics were significantly different for patients with and without outside use (P<.0001). In unadjusted analyses, outside use was associated with younger patient age, female sex, higher complexity score, longer travel time to the enrolling physician group, shorter time enrolled with the physician, and higher number of past enrolments (Table 1).9-12
Patient and enrolling physician-group characteristics by outside use status
Regarding patient-level factors, outside use was associated with higher ratios of FP to population need and lower ratios of emergency services to population need. Physician-group characteristics associated with outside use included having a practice location in a more urban area, a smaller group size, a larger average roster size, fewer years in existence, a larger proportion of female physicians, older average physician age, fewer weekdays worked per patient, fewer weekend or holiday days worked per patient, and more after-hours blocks during weekdays per patient (Table 1).9-12
Relative importance by level
Based on ICCs in a 4-level intercept-only model, the physician and physician-group levels explained 1.6% and 1.9%, respectively, of variation in outside use. We therefore excluded these levels in the remaining regression analyses. The 2-level intercept-only model, nesting patients within group geography, had an ICC of 16.3% of variation at the geographic level, with the patient level accounting for the remaining 83.7% of variation in outside use. Including the observable factors in the model resulted in the ICC for group geography declining from 16.3% to 9.8%, a 40% decrease.
Regression results
The multilevel regression model, which nested patient within group geography, found all patient-level variables and joint hypothesis tests to be statistically significant (P<.05; Figure 2). All group-level variables except average roster size were also statistically significant predictors at the 5% significance level.
Multilevel multivariable logistic regression of factors explaining outside use: Intraclass correlation coefficient=9.80%; N=287,120.
Outside use was positively associated with female sex (odds ratio [OR]=1.21, 95% confidence interval [CI] 1.19 to 1.23); complexity score (highest versus [vs] lowest quintile; OR=3.65, 95% CI 3.53 to 3.79); travel time to enrolling physician (≥60 vs <10 minutes; OR=2.10, 95% CI 2.01 to 2.19); past enrolments (≥2 vs 0; OR=1.18, 95% CI 1.15 to 1.22); and FP to population need ratio (highest vs lowest quintile; OR=1.71, 95% CI 1.62 to 1.80).
Outside use was negatively associated with patient age (≥65 vs 0 to 25 years; OR=0.44, 95% CI 0.42 to 0.45), physician-group RIO score (RIO ≥45 vs RIO 0; OR=0.40, 95% CI 0.32 to 0.50), and weekend and holiday days worked per patient (highest vs lowest number; OR=0.81, 95% CI 0.76 to 0.87). For other dummy variables where joint hypothesis tests confirmed significance, the ORs had no consistent pattern, increasing in some categories and decreasing in others.
DISCUSSION
We found that 25.6% of patients who used any primary care services during the study period had received care outside of their enrolling physician group. Patient-level variation accounted for approximately 83.7% of the probability of outside use. The remaining variation was primarily at the physician-group geographic level. Physician and physician-group contributions to the variation were small and thus not adjusted for in the final model. All patient- and physician-group–level variables were statistically significant (P<.05) except for roster size.
Our results are consistent with previous research based on patient survey data, suggesting that most of the variation in patient-reported access to primary care in Ontario is associated with patient-level factors.4 However, Premji et al did not directly report variation in outside use.4 Other studies have highlighted the effect of rurality, especially the lack of outside primary care providers in rural areas, as a key factor affecting outside use.5 While our analysis confirms the importance of geographic variation in explaining outside use, the effect of physician- and physician-group–level factors is overshadowed by patient-level factors.
We found that controlling for observable factors can account for a large proportion of group variation, which declined from 16.3% in the model with no observable variables to 9.8% in the model that controlled for all observable variables. This suggests that some geographic differences in outside use can be accounted for by service-need ratios at the postal code level or variations in age, sex, and complexity score distributions. However, patient-level factors were the primary determinants of outside use.
Changes in policy may affect outside use as it relates to some physician-group–level factors. This includes encouraging physician groups to increase to more than 5 and incentivizing increased weekend, holiday, and after-hours care. However, the impact on outside use of these factors as well as the physician-group–level access bonus may be limited as most variation occurs at the patient level. Our results suggest that it may be valuable to research policy innovations that promote continuity of care by, for example, sharing medical records; engaging with patients through multilingual, culturally appropriate public information campaigns; or addressing access barriers. It may also be useful to study other preferences, including how patients balance continuity of care with access across the care continuum, which may vary according to patient needs.14,15
Limitations
There are some limitations to this study. Estimates of variation at a given level do not necessarily indicate cause.16 We excluded the billings of FPs with designated focused practices, although some physicians may practise in this capacity without having formal designations recognized by the Ontario MOH. We also did not account for variations in the impact of after-hours shifts on specific days of the week or the intensity of after-hours work each day in our analysis.
We did not account for groups receiving interdisciplinary funding (ie, family health teams) or locum physicians within FHOs. While ED availability is included in the ED to population need ratio, we did not consider ED use as a measure of outside use in the analysis because it does not affect the access bonus, although this may arguably be seen as analogous.
Conclusion
We used multilevel modelling to quantify the variation in outside use at the patient, physician, physician-group, and geographic levels. We also examined how observable factors accounted for variation at the patient and geographic levels. Although the importance of geographic location has been highlighted in previous studies, we found that patient-level choice has a greater impact than geography, and that at less than 2%, physician- and physician-group–level factors are negligible. Our results emphasize the importance of carefully considering factors and policies that influence patients’ choices in continuity of care within primary care services.
Since completing this study, the number of online physician practices (eg, Maple) and the scope of services offered by pharmacies have increased, which may influence the prevalence of outside use and which warrants further evaluation. Our results suggest that it would be beneficial to consider policies that target patient choice through channels other than enrolling physicians and physician groups.
Footnotes
↵* Appendix Tables 1 to 4 are available from https://www.cfp.ca. Go to the full text of the article online and click on the CFPlus tab.
Acknowledgment
The authors thank Steve Nastos and Yin Li for their data support and Jim Wright for his helpful comments on the manuscript. Population-level health care data were obtained from the Ontario Ministry of Health (MOH) through an agreement with the Ontario Medical Association. The Population Grouping Methodology is owned by the Canadian Institute for Health Information (CIHI) and used under licence. Neither CIHI nor MOH were involved in or had any control over the study design and conduct; the data collection, analysis, and interpretation; the data preparation; the decision to publish; or the manuscript preparation, review, and approval.
Contributors
Drs Shaun Shaikh, Sharada Weir, and Jasmin Kantarevic conceived the study and developed the analysis plan. Dr Shaun Shaikh analyzed the data and drafted the manuscript. All of the authors contributed to interpretation of the data, revised the manuscript for important intellectual content, approved the final version for publication, and agreed to act as guarantors of the work.
Competing interests
Drs Shaun Shaikh, Sharada Weir, Lyn M. Sibley, Silvy Mathew, and Jasmin Kantarevic declare that they have no competing interests. Dr Nadia Alam declares the following competing interests: stipends for medical consulting from Khure Health and Dr Bill; payment for speaking on health policy–related topics from Canadian Women in Medicine, Canadian Federation of University Women, and Welloga; reimbursement for travel and time to attend College of Family Physicians of Canada board meetings; and payment for work done as the primary care lead for Halton Seamless Care for Optimizing the Patient Experience.
This article has been peer reviewed.
- Copyright © 2026 the College of Family Physicians of Canada
References
In this issue
Jump to section
Related Articles
Cited By...
- No citing articles found.