Analysing spatial accessibility to health care: a case study of access by different immigrant groups to primary care physicians in Toronto

Abstract

This article analyses the spatial accessibility of a number of immigrant groups to linguistically diverse primary care (family) physicians in the Toronto Census Metropolitan Area (CMA). The two-step floating catchment area (2SFCA) method, a special type of gravity model, is employed to measure spatial accessibility using Network Analyst in ArcGIS 9.3. The context of this study is the predominantly publicly funded Canadian health-care system and a multicultural urban setting where both the population and the physicians are culturally and linguistically diverse. This article focuses on a total of eight ethnicities: six groups of recent immigrants – from Hong Kong, Iran, Mainland China, Pakistan, Russia and Sri Lanka; and two groups of long-standing immigrants – from Italy and Portugal. It examines the spatial (mis)match between the residential distribution of immigrant populations and the distribution of linguistically appropriate family physicians. The quantitative data analysed in this article include the physician data set from the College of Physicians and Surgeons of Ontario and geo-referenced 2006 Canadian Census data. This article highlights areas of poor accessibility and provides a comparison of the different ethnic groups. It demonstrates the use of the geographical information system (GIS) in public health research and yields important policy implications for public health planning.

Keywords:

• spatial accessibility
• GIS
• health care
• family physician
• immigrants
• Canada

1. Introduction

This article examines the geographic accessibility of various immigrant populations to health services in a multi-ethnic urban setting. Specifically, it analyses the spatial accessibility of eight immigrant groups residing in the Toronto Census Metropolitan Area (CMA) in Canada to linguistically diverse family physicians offering primary care. According to Hoernig and Walton-Roberts (2006, p. 409), Toronto is ‘the most ethnically diverse city in the world’. Since the early 1990s, more than 40% of all immigrants to Canada have chosen to settle in the Toronto CMA. It is essential to the adaptation and integration of both newly arrived and long-standing immigrants that their health needs be met. Although Canada has a publicly financed and administered health insurance system, research has found that immigrants underutilize health services, face significant barriers to health-care access and exhibit declining health status (Leduc and Proulx 2004, Newbold 2005).

Some of the most noticeable barriers to health-care access for immigrants are those related to geography and geographic distribution, language and culture. The availability and distribution of family physicians or general practitioners has a direct impact on access to and utilization of physician services. The traditional points of entry for immigrants are located in the city core (e.g. Chinatown), where hospitals, medical services and the majority of a CMA's family physicians are concentrated. However, new immigrants tend to settle in the suburbs upon landing and long-term immigrants tend to relocate to the suburbs, where housing is more affordable (Siemiatycki and Isin 1997, Murdie and Teixeira 2000). This suburbanization of immigrants suggests a degree of patient–physician location discordance or ‘spatial (mis)match’. For immigrants arriving from non-English-speaking countries, language can be a strong barrier to the access and utilization of health services because of the difficulty in understanding English medical terms and the poor distribution of same-language physicians (Wang et al. 2008). Linguistic barriers and cultural differences between care providers and users can contribute to patient–physician miscommunication and misinterpretation of symptoms by physicians (Zhang and Verhoef 2002, Field and Blakemore 2007, Arorian et al. 2008).

Economic or financial costs are generally regarded as the main barrier to the consumption of health services (Khan and Bhardwaj 1994). However, in Canada, where the health insurance system provides medically necessary hospital and physician services for all members of the society on a prepaid basis, without direct charges at the point of service, access to health care depends instead on the local availability of physicians as well as the individual preferences based on factors such as ethnicity and language, especially for newcomers who face linguistic and cultural barriers to integration.

Spatial accessibility is a central concept in this article. It refers to the ease with which individuals in one location can reach other locations (see review of accessibility in Pirie (1979) and Kwan and Weber (2003)). Accessibility measures range from simple cumulative opportunity measures (Handy and Niemeier 1997, Casas 2008) to more complex gravity-based (Wang and Luo 2005) and space-time (Kwan 1998, Weber 2003) measures. In health research, much has been written about geographical access to health services, mostly through the use of spatial models within a geographical information system (GIS) (Joseph and Bantock 1982, Martin and Williams 1992, Parker and Campbell 1998, Cromley and McLafferty 2002, Lin et al. 2002, Lovett et al. 2002, McLafferty and Grady 2005), including access to physicians (Luo and Wang 2003, Guagliardo et al. 2004, Laditka 2004). Among the various types of accessibility measures, gravity-based measures are the most widely used. The class of gravity-based measures conceptualizes accessibility as a trade-off between opportunity attractiveness and spatial separation (Guy 1983), and, in the context of health-service utilization, accessibility is generally seen as influenced by the spatial structures of health-care supply and demand, neither of which is distributed uniformly in space.

Kwan and Weber (2003) argue that traditional accessibility measures may not be sufficient for modelling individual experiences, due to the changes in urban form and other complexities in human spatial behaviour. It has been suggested that a cultural lens be used when examining immigrant accessibility to health services, because of an observed preference, whether by choice or due to constraint, for physicians who can communicate in one's mother tongue and who are at least somewhat acquainted with health beliefs that may differ significantly from those prevalent in the host country (Betancourt et al. 2003, Wang 2007). In an effort to model geographic accessibility of Mainland Chinese immigrants to same-language physicians, Wang (2007) modifies the traditional gravity-type accessibility measures, a special type of which is the so-called two-step floating catchment area (2SFCA) method, by taking into consideration the sociocultural context, in particular the linguistic diversity among the physicians and the population. The modified model calculates the accessibility of a subgroup in a population (i.e. immigrants) to a subgroup in the opportunity set (i.e. linguistically matched physicians) while recognizing the complex interrelationships among sub-opportunity sets and subgroups (see Section 3). It is employed in this article as the main analytical tool for understanding the spatial mismatch in immigrants' access to linguistically diverse health-care providers.

This article first examines the geographic distribution of eight selected recent and long-standing immigrant groups in Toronto in relation to the location patterns of linguistically matched family physicians, who, in Canada, are gatekeepers to specialists. It then employs the 2SFCA method to analyse the spatial accessibility among various immigrant populations to same-language family physicians (SLFPs). The 2SFCA model is implemented within a GIS environment primarily through the use of Network Analyst in ArcGIS 9.3. Areas that are underserviced with regard to SLFP accessibility are identified. Lastly, this article compares the spatial access to SLFPs among different immigrant groups and the general population and discusses the complex relationship among geographic access, language, ethnicity and utilization of health care. Two important data sets – the 2006 Canadian Census (Statistics Canada 2001, 2006) and the 2005 College of Physicians and Surgeons in Ontario (CPSO) data set – are used to provide information on immigrant settlement patterns and the locational and linguistic characteristics of Toronto's physician supply. This article focuses on family physicians as, in general, one cannot use specialist services in Canada without an initial referral by a family physician. This article yields an important policy for identifying underserviced areas with respect to health professionals, developing health-care programmes, addressing issues related to foreign-trained physicians, formulating health promotion strategies and enhancing delivery of primary care that is relevant for immigrants.

2. Data, study area and study groups

The study area is the Toronto CMA, which receives 43% of all recent arrivals to Canada, compared with 16% for Vancouver and 14% for Montreal (Citizenship and Immigration Canada (CIC) 2005). According to the 2006 Canadian Census, approximately 46% of the Toronto CMA's population is foreign-born. Quality primary health care is critical for newcomers, as it directly affects one's well-being. Data collected by the CPSO (2005) are critical to this article, as they include attributes of each family physician, such as language spoken, location of practice and university of graduation as well as whether the physician accepts new patients. The data were geocoded to points using six-digit postal-code information. In some cases, multiple physicians shared the same point location, indicating that they were likely located in the same medical building. Approximately 29% of the CMA's family physicians self-reported speaking at least one non-official language, and a total of 100 non-official languages were reported.

Toronto's immigrant population is extremely heterogeneous in terms of socio-economic status. Instead of analysing all immigrant groups, this study focuses on eight ethnicities: six groups of recent immigrants, from Hong Kong, Iran, Mainland China, Pakistan, Russia and Sri Lanka, and two groups of long-standing immigrants, from Italy and Portugal. These particular groups were selected on the basis of their immigration history, size and sociocultural and economic characteristics. With the exception of those from Hong Kong and Russia, the groups of recent immigrants were among the top 12 source countries for immigrants to Canada between 1996 and 2006 (CIC 2005, 2006, 2007). The size and recency of the six groups point to the appeal of studying their adaptation experiences in Canada (see Table 1). Canada's primary and secondary source countries for immigrants have been Mainland China and Pakistan, respectively, since the late 1990s; Hong Kong was the top source country until 1997, when the former British colony was returned to China. While the Cantonese-speaking Hong Kong Chinese represent an established immigrant community that has achieved ‘institutional completeness’ (Lo and Wang 2004), the Mandarin-speaking Mainland Chinese and other recent immigrant groups are emergent. Canada's permanent residents come under three categories: economic immigrants (e.g. skilled workers, entrepreneurs); family class (e.g. spouses and children); and refugees. Most immigrants from Hong Kong, Mainland China, Pakistan and Russia have come as skilled workers, entrepreneurs or investors (Lo et al. 2007), whereas many of those from Sri Lanka have been admitted on a humanitarian basis. In terms of socio-economic status, apparently those in the six selected groups face various degrees of settlement challenge, as evidenced by their higher-than-average unemployment rates, their lower-than-average household income and the fact that large proportions live under the poverty line – despite their very high educational credentials. This reflects the well-documented difficulty experienced by new Canadian immigrants with regard to economic integration (Preston et al. 2003, Green and Green 2004).

Table 1. Socio-economic characteristics of selected recent immigrant populations

		Iran	Mainland China	Pakistan	Russia	Sri Lanka	Hong Kong	Toronto CMA
Population
(2006) ^a		45,205	190,515	85,195	29,875	45,205	102,675	5,064,690
Primary language(s)		Farsi (Persian)	Mandarin	Urdu, Hindi	Russian	Tamil, Sinhalese	Cantonese	English, French
Newcomers 1999–2001 ^b	Count of newcomers 1999–2001	9,143	53,739	28,855	6,841	12,104	1,536	119,488
	% Newcomers	22.5	47.9	44.3	48.2	17.5	1.5	20.04
	% Economic	52.8	77	58.3	75.6	11.4	30.4	31^3
	% Family	29.9	19.8	35.1	17.1	39.8	69.5	54
	% Humanitarian	17.3	3.2	6.6	7.3	48.7	0.2	15
	% Bachelor's and above	35.4	52.7	40.5	49.9	7.7	16.7	19.0
	% No proficiency in English or French	46.1	70.2	49.1	56	42.2	40.8%	54.9
Immigrants by place of birth ^c	% Recent immigrants	37.7	34.5	50.9	52	21.8	3.0	16.6
	% Unemployed	10.6	10.2	12.4	11.2	10.2	NA	6.7
	% Mean Household income ($)	51,363	56,406	54,893	54,214	53,503	NA	76,454
	% Low income	35.7	34.1	37.7	33	32.4	NA	22.9
Notes: CMA, Census Metropolitan Area.
^aLo et al. (2007).
^bCanadian Census 2006, Lo et al. (2007).
^cWang and Lo (2000).

The two long-standing immigrant groups selected, those from Italy and Portugal, are used to contrast the experience of the recent arrivals. Before 1986, the Italians were the most populous and the Portuguese the third-most populous immigrant group in Canada (CIC 2001). In 2006, there were 130,325 Italian-born and 76,025 Portuguese-born people living in the Toronto CMA. Although the United Kingdom ranks second as an immigrant source country pre-1986, migrants from the United Kingdom are eliminated because they speak one of Canada's official languages and are less likely to face linguistic barriers to health-care access.

The 2006 Canadian Census provides information on the settlement patterns of the study populations and the general population at a census tract level. The variable Place of Birth is used to identify immigrants from the eight countries and the variable Recent Immigrants is used to identify those who arrived in Canada in the past 5 years. The 2006 Census was a detailed enumeration of the Canadian population and the census data used were derived from the 20% sample who completed the long questionnaire. Data of a finer geographical level – dissemination area (DA) – were not used due to Statistics Canada's data-suppression strategy of rounding small numbers in a DA to zero in order to achieve confidentiality.

The terms ‘recent immigrants’ and ‘newcomers’ are used in this article. The Statistics Canada Census uses the term recent immigrants for those who have immigrated in the past 5 years. CIC uses the term newcomers for new immigrants who are within 3 years of landing in the country and are eligible for federally funded settlement services such as Language Instruction for Newcomers to Canada. Since this article employs both Census data and CIC data, the two terms are used interchangeably.

3. Methods

This article employs the modified 2SFCA accessibility measure to model the geographic accessibility of selected immigrant groups to SLFPs. The assumption of the modelling exercise is the preference among immigrants for linguistically and culturally appropriate physicians, as revealed in recent literature examining the linguistic barriers facing foreign-born immigrants, such as South Asians in the United Kingdom, Hispanics and African-Americans in the United States and Chinese in Canada and the United States (Komaromy et al. 1996, Derose and Baker 2000, Field and Blakemore 2007, Wang 2007, Arorian et al. 2008, Loue 2008). The origin of the 2SFCA accessibility measure is the traditional simple gravity-type accessibility measure put forward by Hansen (1959) – see Equation (1)(1); this is a type of spatial interaction model of travel behaviour that assesses the interaction between destination attractiveness and spatial separation represented by some form of cost, such as travel time or travel distance:

(1)

where A_j refers to the geographical accessibility at location i. represents the attractiveness of destination j measured by variables such as facility size or number of service types. is the impedance function representing the disincentives to reach a destination and is usually measured by the distance and driving time between the origin and destination (d_ij ). While the original gravity model is useful, a limitation is that it focuses on the supply side (i.e. the opportunity set), overlooking how trip distribution is determined by demand from departure point, and in particular competition among users. As an enhancement to Equation (1)(1), the modified gravity model (2) takes into account the competition on the demand side for a limited number of opportunities (Joseph and Bantock 1982, Shen 1998, Luo and Wang 2003):

(2)

where, in the context of this article, is the gravity-based accessibility at census tract i. P_k is the population in census tract k competing for physicians. S_j is the number of family physicians at location j. The 2SFCA – see Equation (3)(3) – is a simpler version of the modified gravity model (2) and is easier to implement in a GIS environment (Peng 1997, Radke and Mu 2000, Wang and Luo 2005). It replaces the continuous distance impedance function with a dichotomous travel impedance that is predefined by patient travel threshold or physician catchment area:

(3)

where is the accessibility at census tract i to family physicians; d_ij is the travel time between i and j; and is the physician-to-population ratio at physician location j, which is within the travel threshold d ₀ from i. As with Equation (2)(2), the calculation of the 2SFCA accessibility involves two steps (see Wang and Luo (2005) for a detailed illustration of the 2SFCA model). In step one, for each physician location, compute physician-to-population ratio for centroids within the physician catchment area. The physician-to-population ratio reflects the degree of competitiveness for physicians. In step two, for each census tract centroid, sum up all the physician-to-physician ratios associated with all physicians within the travel threshold.

The modified gravity model (2) also involves two steps, similar to those in Equation (3)(3). However, it uses distance impedance function as a scale and assigns various weights to destination attractiveness (W_j ) and demand (P_k ), based on spatial separation. Unlike in Equation (2)(2), the 2SFCA measure does not have to scan all the demand points (e.g. centroids) in step one and all physician locations in step two, and there is no need to specify a distance impedance function required by the gravity model. The relative computational simplicity of the 2SFCA model is one of its attractive features, since there is no need to specify a distance impedance function, as required in the gravity model. However, the 2SFCA model is arbitrary in that the predefined travel threshold draws an artificial line between accessible and inaccessible opportunities. Attempts have been made to address this weakness in the 2SFCA model by dividing the study area into various travel zones associated with different travel time thresholds and by incorporating a distance friction coefficient applied to predetermined thresholds (Luo and Qi 2009, Wang and Roisman 2011). However, survey data on actual travel behaviour are sometimes required to determine the travel time threshold for different zones (Wang and Roisman 2011), and such data are not always readily available. This article uses the original 2SFCA method.

Wang (2007) improves on the 2SFCA model (3) to calculate the accessibility of a subgroup of users on the demand side to a sub-opportunity set by adding two location-quotient-type terms measuring the relative abundance of opportunities in the sub-opportunity set and competition from users in a sub-population group:

(4)

where in the context of this article is the 2SFCA accessibility measure of immigrants to SLFPs for census tract i. is the number of SLFPs in physician location j and is co-ethnic immigrant population in census tract k. The rationale for Equation (4)(4) is that immigrants may compete with each other for same-language physicians, and therefore their accessibility to same-language physicians is influenced by the relative abundance of such physicians and the size of the co-ethnic population in the area. Note that the original gravity model (2) and the 2SFCA model (3) cannot be applied directly to measure accessibility to SLFP, because one cannot assume that family physicians serve linguistically matched populations only and exclude patients of different ethnicities and different linguistic backgrounds. Use of the sub-measure Equation (4)(4), as in this article, is necessary in order to measure accessibility of a sub-population group to a sub-opportunity set.

Both population data and physician data are imported into ArcGIS 9.3 for analysis. Using the 2008 Street Network Dataset provided by the DMTI Spatial Inc., Network Analyst 9.3 is used to compute the shortest network travel time by car based on road speed limit between any possible pair of population centroid and physician point location and create an origin–destination matrix. In this article, the travel threshold time is set at 11 minutes. In the literature, various travel threshold times, ranging from 9 to 30 minutes, are used in the FCA and other accessibility models (Luo and Wang 2003, Roisman 2007), including 11.3 minutes for accessing employment in Cleveland, Ohio (Wang 2003) and 11 minutes for accessing SLFPs by suburban Chinese immigrants in Toronto and 9 minutes for urban Chinese immigrants (Wang and Roisman 2011). Note that the 11 minutes is a theoretical travel time, as it represents the calculated unimpeded car-based travel time. It is appropriate to use a relatively small travel threshold in this article as the actual travel time is always larger than the calculated trip time. For example, the reported travel time in Toronto can be up to 228% of the calculated unimpeded travel time due to the impact of factors such as ‘time-of-day traffic’, weather conditions and selection of routes on actual travel time (Phibbs and Luft 1995, Roisman 2007). According to Wang (2003), the relationship between theoretical and reported travel time is unlikely to be linear and the difference is mainly due to an intercept related to activities at trip origin and trip destination such as getting into a car, finding parking and walking to office. The selected travel time is appropriate also because this article focuses on new immigrants who may have a small activity space due to unfamiliarity with the city. The selected travel time in this article is also theoretical in nature. As suggested in Wang and Roisman (2011), surveys on actual physician-seeking behaviour among various immigrant groups would be able to provide more insight on identifying and adjusting group-specific and location-specific travel threshold in accessing same-language physicians.

4. Analysis

This section first discusses the aggregated statistics on population and family physicians at the CMA level (Table 2). It then presents two scenarios of spatial analysis where care by a family physician is sought by immigrants. In the first scenario this article examines the spatial accessibility of each of the eight study populations to SLFPs. The second scenario computes accessibility of newcomers – from among the six groups who arrived in the past 5 years – to SLFPs who are accepting new patients, as newcomers face the more daunting settlement challenge. Comparisons are made among various groups in relation to the general population.

Table 2. Aggregated statistics on population and family physicians (FPs) in Toronto CMA

			Same-language family physicians (SLFPs)			By recent arrivals (in past 5 years)		SLFPs accepting new patients
Study populations by source country	Count	%	Count	%	SLFPs/1000 immigrants	Count	%	Count	%	SLFPs accepting new patients/1000 recent immigrants
Iran	45,205	0.9	37	0.6	0.82	13,965	0.03	8	1.0	0.57
Hong Kong	102,675	2.0	404^a	6.6	3.93	3,060	0.01	104	12.5	33.99^b
Mainland China	190,515	3.8	360^a	5.9	1.89	63,370	0.14	88	10.5	1.39
Pakistan	85,195	1.7	212	3.5	2.49	36,975	0.08	61	7.3	1.65
Russia	29,875	0.6	63	1.0	2.11	9,895	0.02	23	2.8	2.32
Sri Lanka	83,835	1.7	42	0.7	0.52	17,135	0.04	18	2.2	1.05
Italy	130,325	2.6	144	2.4	1.12	NA	NA	43	5.1	NA
Portugal	76,025	1.5	58	0.9	0.75	NA	NA	13	1.6	NA
Toronto CMA	Population		FPs		FPs/1,000persons	Recent immigrants		FPs accepting new patients	FPs accepting new patients/all FPs (%)	FPs accepting new patients/1,000 recent immigrants
	5,064,690		6,078		1.20	446,710		835		1.87
Notes: CMA, Census Metropolitan Area.
^aSome physicians self-reported speaking both Mandarin and Cantonese, the two Chinese languages most commonly spoken in Canada; in the table, there is double counting for these two statistics only.
^bThe high ratio results from the much-reduced number of immigrants from Hong Kong after 1997.

As shown in Table 2, the number of SLFPs for different ethnic groups varies. The two Chinese subgroups have the highest number of SLFPs. The physician-to-population ratio in Table 2 serves as a broad indicator of the relative abundance of SLFPs. Low physician-to-population ratios are observed for Iranian, Portuguese and Sri Lankan immigrants. For arrivals in the past 5 years, Iranians, Mainland Chinese, Pakistanis and Sri Lankans have lower physician-to-population ratios than the CMA average. However, the ratios do not take into account the spatial distribution of physicians and population. A high ratio does not necessarily ensure adequate geographical access to physicians at the neighbourhood level if the physicians are poorly distributed.

The 2SFCA method is used to compute the geographical accessibility of the general population to family physicians, regardless of ethnicity and language (Figure 1). The accessibility scores decrease as one moves from downtown to the outskirts. Approximately 76% of the CMA's family physicians practise in the City of Toronto (see Table 3), where there is a high concentration of hospitals, whereas only 43% of the CMA population reside in that area. A spatial mismatch between family physician location and population is clear. The edge census tracts not adjacent to Lake Ontario were removed, because people living near the edge of the study area may use resources outside the study area and people residing outside the study area may compete for physicians within the study area (Luo and Wang 2003, Yang et al. 2006).

Figure 1. Spatial accessibility of general population to family physicians.

Table 3. Accessibility, immigrant population and physician distribution in urban and suburban Toronto

	Immigrants		Recent immigrants		SLFPs		Accessibility to SLFPs (mean)		Newcomers' accessibility to SLFPs accepting new patients (mean)
Country of origin	Urban	Suburb	Urban	Suburb	Urban	Suburb	Urban	Suburb	Urban	Suburb
Total of Toronto
CMA	1,082,375	1,234,245	238,290	208,420	4,652^a	1,430^a	2.107^b	0.623^b	4.151^b	1.264^b
	47%	53%	53%	47%	76%	24%	NA	NA	NA	NA
Iran	22,120	23,085	8,515	5,450	19	18	0.696	0.624	0.017	0.039
	49%	51%	61%	39%	51%	49%	NA	NA	NA	NA
Hong Kong	36,965	65,710	43,165	20,205	245	159	7.42	3.56	1.171	0.636
	36%	64%	68%	32%	61%	40%	NA	NA	NA	NA
Mainland China	112,265	78,250	43,165	20,205	220	140	2.13	2.042	0.169	0.115
	59%	41%	68%	32%	61%	39%	NA	NA	NA	NA
Pakistan	32,890	52,305	16,500	20,475	111	101	2.404	1.598	0.151	0.129
	39%	61%	45%	55%	52%	48%	NA	NA	NA	NA
Russia	12,865	17,010	238,290	208,420	50	12	2.133	0.91	0.163	0.071
	43%	57%	53%	47%	81%	19%	NA	NA	NA	NA
Sri Lanka	46,435	37,400	10,990	6,145	28	14	0.394	0.269	0.068	0.047
	55%	45%	64%	33%	67%	33%	NA	NA	NA	NA
Italy	58,560	71,765	NA	NA	115	29	2.206	0.644	NA	NA
	45%	55%	NA	NA	80%	20%	NA	NA	NA	NA
Portugal	41,705	34,320	NA	NA	48	10	0.864	0.244	NA	NA
	55%	45%	NA	NA	83%	17%	NA	NA	NA	NA
Notes: SLFPs, same-language family physicians; CMA, Census Metropolitan Area.
^aThey are the numbers of family physicians in urban and suburban Toronto.
^bThey are calculated accessibility to SLFPs for all the immigrants and recent immigrants in the Toronto CMA.

The first scenario of spatial analysis concerns SLFP accessibility for each of the eight study populations. Those who reside in the city in general enjoy better spatial access to SLFPs (see Table 3). Using the example of Italians, Mainland Chinese, Hong Kong Chinese and Pakistanis, Figure 2 shows spatial accessibility scores in relation to immigrant settlement patterns. The dots, each representing 100 immigrants, in Figure 2 and the following figures represent the approximate location of individuals in selected immigrant groups based on the country of origin variables in the 2006 Census. In the maps the dots are randomly distributed within specific census tract where immigrants reside and do not represent the accurate residential location of the population. In the cases of Hong Kong Chinese and Italians, census tracts associated with a high level of SLFP accessibility do not necessarily have large co-ethnic populations – a clear spatial discordance between SLFP location and population settlement. Hong Kong Chinese cluster in Markham and north of Scarborough and North York, and the main Italian settlement area extends from Little Italy in the city to Woodbridge in suburban Vaughn. For all the study groups, census tracts in the city core are generally associated with good SLFP accessibility, largely due to the concentration of physicians in the area. For Pakistanis, some suburban census subdivisions, such as Mississauga, are associated with good SLFP access due to the relatively large number of SLFPs in the area. However, the high level of SLFP accessibility for Hong Kong Chinese in Mississauga is due to the much lower degree of competition from co-ethnics there than in Markham. The Russian community is found along Bathurst Street extending north from North York. Similar to the downtown area, suburban areas such as Brampton are associated with good SLFP accessibility, due primarily to the location of a few Russian-speaking family physicians in those areas where the Russian population is small. Iranians who cluster in North York and Richmond Hill do not necessarily have good SLFP access, due to competition within the group.

Figure 2. Spatial accessibility of immigrants from Pakistan, Hong Kong and Italy to same-language family physicians.

The second scenario is the spatial accessibility of newcomers to SLFPs who accept new patients. There are no census data available on recent Italians and Portuguese immigrants, as the number of recent arrivals from these two countries is insignificant. The CPSO data reveal a family physician's capacity to accept new patients. Based on Equation (4)(4), spatial accessibility is calculated by multiplying a new ratio (between the share of newcomers and the share of SLFPs who accept new patients within the travel time threshold) by the general accessibility scores.

The scores for accessibility of recent immigrants to SLFPs who accept new patients are in general much lower than the scores relating to general accessibility and the scores for immigrants in the first scenario (Figure 3). Both Chinese groups tend to settle in the northern suburbs, such as Markham, where the competition for SLFPs is heavy, thus producing a spatial mismatch on SLFP supply and demand. For Iranian, Pakistani, Russian and Sri Lankan newcomers, although the calculated accessibility scores are extremely low, there seems to be a degree of spatial match between accessibility and settlement pattern. For example, recent Sri Lankan immigrants concentrate in the west and east of the city, where the calculated accessibility is higher than in other areas (although much lower than in scenario 1). Recent Sri Lankans who settle in the northern suburbs face extremely poor accessibility.

Figure 3. Accessibility of newcomers from Mainland China, Sri Lanka and Russia to SLFPs accepting new patients.

Statistical analyses of accessibility scores reveal much intergroup variability among immigrant groups in the two scenarios. The median general accessibility scores calculated in scenario 1 are higher for immigrants from Mainland China, Hong Kong, Italy and Pakistan than for those from Iran, Portugal, Russia and Sri Lanka. Calculated accessibility in scenario 2 for newcomers seeking SLFPs who accept new patients is much poorer than the general accessibility and accessibility for both new and long-standing immigrants in scenario 1. The relatively large numbers of family physicians speaking Chinese, Hindi, Italian and Urdu partly explain the high degree of SLFP accessibility of these immigrant populations at a group level. The relatively small numbers of family physicians speaking Farsi, Persian, Portuguese, Russian, Sinhalese and Tamil contribute to the poor SLFP accessibility of Iranians, Portuguese, Russians and Sri Lankans. Recent immigrants are the most disadvantaged in terms of accessing SLFPs who accept new patients. The zero median statistics for all the recent immigrant groups, with the exception of Mainland Chinese, are due to the absence of SLFPs accepting new patients within the travel time threshold of a large number of centroids.

Table 5 shows the correlations between calculated accessibility indices and study populations. General accessibility has a predominantly negative, and in most cases significant, relationship with total population in the CMA, immigrant population and recent immigrant population at a census tract level. This indicates that areas with good access to family physicians, regardless of ethnicity and language, have relatively small populations, including small immigrant and newcomer populations except for recent Mainland Chinese immigrants and Portuguese. For example, downtown Toronto, which has the highest concentration of physicians (52%), does not have a high concentration of general population (14%) or immigrant population, suggesting a location discordance between physician location and population settlement.

In scenario 1, among those significant relationships, the SLFP accessibility of immigrants from Mainland China, Hong Kong and Italy relates negatively to the size of the community, while the accessibility of those from Pakistan, Portugal and Sri Lanka relates positively with the population, which is consistent with Figure 2. In scenario 2, the accessibility of all six recent immigrant groups is positively and significantly associated with the population of the respective study group, in conformity with the patterns revealed in Figure 3. Note that correlation coefficients should be analysed in combination with Table 4 when assessing the spatial equity between physician supply and demand. The correlation coefficients show the relationship between calculated access scores and population of census tracts, whereas the descriptive statistics indicate which group has better accessibility. For example, although a significant relationship between spatial accessibility and group size is observed for all six recent immigrant groups, the calculated SLFP accessibility for these groups is extremely poor, with a large number of tracts showing zero accessibility (see Table 5). Therefore, one cannot say that these groups are well serviced by linguistically appropriate physicians, who are few in number but whose spatial location somewhat matches the settlement patterns of newcomers.

Table 4. Descriptive statistics for calculated accessibility in scenarios 1 and 2

	Minimum	Maximum	Mean	Median	SD
General accessibility	0	4.0	1.4	0.000.8	1.251
Scenario 1: accessibility of immigrants to SLFPs
Mainland China	0	12.5	2.2	2.1	1.479
Hong Kong	0	23.9	5.7	3.6	6.258
Pakistan	0	11.5	2.1	2.0	1.932
Sri Lanka	0	2.6	0.3	0.2	0.385
Iran	0	11.7	0.7	0.4	1.046
Russia	0	13.3	1.6	0.0	2.119
Italy	0	6.7	1.5	1.0	1.464
Portugal	0	6.8	0.6	0.3	0.803
Scenario 2: accessibility of newcomers to SLFPs accepting new patients
Mainland China	0	0.8	0.1	0.2	0.149
Hong Kong	0	36.2	0.9	0.0	3.106
Pakistan	0	1.1	0.1	0.0	0.188
Sri Lanka	0	0.9	0.1	0.0	0.113
Iran	0	2.1	0.0	0.0	0.125
Russia	0	1.5	0.1	0.0	0.222
Notes: SLFPs, same-language family physicians.
Accessibility scores in this table are inflated by 1000.

Table 5. Correlations between calculated accessibility and immigrant population (N = 944 census tracts)

			Scenario 1: accessibility of immigrants to SLFPs								Scenario 2: accessibility of newcomers to SLFPs accepting new patients
		Ai ^1	MainlandChina	HongKong	Pakistan	Sri Lanka	Italy	Portugal	Iran	Russia	Mainland China	HongKong	Pakistan	Sri Lanka	Iran	Russia
	Ai	1
Immigrants	Mainland China	−0.015	−0.165**
	Hong Kong	−0.156**		−0.135**
	Pakistan	−0.130**			0.078**
	Sri Lanka	−0.156**				0.205*
	Italy	−0.155**					−0.152**
	Portugal	0.233**						0.117**
	Iran	−0.084**							0.017
	Russia	−0.064**								0.060
Recent Immigrants	Mainland China	0.009									0.184**
	Hong Kong	−0.105**										0.416**
	Pakistan	−0.064*											0.258**
	Sri Lanka	−0.125**												0.465**
	Iran	−0.039													0.135**
	Russia	−0.018														0.334**
CMA	Total population	−0.183**
Notes: SLFPs, same-language family physicians. is the general accessibility to family physicians regardless of ethnicity and language.
*Correlation is significant at the 0.05 level two-tailed.
**Correlation is significant at the 0.01 level two-tailed.

5. Conclusion and discussion

This article employs the 2SFCA method in a GIS environment to examine the geographical accessibility of eight selected recent and long-standing immigrant groups to linguistically matched family physicians in the Toronto CMA. The context of this article is the widely reported linguistic, cultural and geographical barriers faced by immigrants in accessing health care and the location and language discordance between the physicians and patients in a multi-ethnic urban setting.

In addition to using the physician-to-population ratio to understand the supply of and demand for SLFPs among immigrant populations, this article employed the 2SFCA model to examine census-tract-level geographical accessibility of the general population to family physicians and accessibility of immigrants to SLFPs. The 2SFCA accessibility measure, originating from the modified gravity-based accessibility model, is a useful model that can be implemented in GIS in computing spatial accessibility. It takes into account the competition among users on the demand side, which is neglected in the original gravity model.

This article has shown that the 2SFCA-based general accessibility to family physicians, regardless of language and ethnicity, has a clear centrifugal spatial pattern with a cluster of high scores in the city core, which has the highest concentration of physicians and hospitals (See Figure 1, Table 3). Accessibility decreases with the increase in distance from the downtown. In scenario 1, immigrants from Mainland China, Hong Kong, Italy and Pakistan, regardless of their length of residence in Canada, enjoy greater SLFP accessibility overall than the general population in accessing family physicians, as evidenced by the median accessibility score. However, for immigrants from Mainland China, Hong Kong and Italy, accessibility is negatively associated with population (Table 4), indicating a degree of spatial mismatch between the physician location and settlement pattern. Accessibility for those from Pakistan, Portugal and Sri Lanka is positively related to immigrant population at a census tract level, indicating a spatial convergence between physician location and immigrant residential pattern, even though the median accessibility score for Portuguese and Sri Lankans is low. The relationship between accessibility and settlement pattern for Iranians and Russians is statistically insignificant, although Figure 2 shows that in many areas SLFPs seem to follow the residential patterns of co-ethnic immigrants. In scenario 2, which considers only newcomers and SLFPs who accept new patients, a significant correlation between spatial accessibility and population is observed for all six groups (Table 5), thus confirming a spatial converging trend similar to that shown in Figure 3. However, the calculated accessibility scores are extremely low for all of these groups, which suggests that newcomers are the most disadvantaged in their settlement process, in terms of locating SLFPs who accept new patients, compared with both their long-standing counterparts and the general population.

The varying patterns of spatial accessibility to SLFPs among the selected immigrant populations reflect their different settlement and adaptation experiences. Among the recent groups, Hong Kong Chinese are the most entrepreneurial. They have immigrated to Canada en masse since the 1970s, and in Toronto they have created one of the largest and most viable Chinese ethnic economies outside Asia. It has been observed that many Cantonese-speaking physicians are now offering Mandarin services in order to capture the growing Mainland Chinese community. Although the large number of Chinese-speaking family physicians explains the good SLFP accessibility among Chinese immigrants on average, many census tracts populated by Chinese immigrants are associated with poor accessibility due to a spatial mismatch of physician location and immigrant settlement.

The two South Asian subgroups – Pakistanis and Sri Lankans – are similar in size yet differ greatly in terms of experience. While most Pakistanis have arrived in Canada as skilled workers, most Sri Lankans have been admitted on a humanitarian basis. And while the majority of the Sri Lankans in Toronto speak Tamil, a unique South Asian language that is not easily understood by speakers of other South Asian languages, Urdu-speaking Pakistanis generally understand Hindi, the national language of India, and thus have a wider selection when it comes to choosing SLFPs. Therefore, Pakistani immigrants have relatively good spatial accessibility to SLFPs and Sri Lankans have the poorest accessibility among the eight groups.

Russians are the smallest community among the eight, yet the number of Russian-speaking family physicians is higher than that of physicians speaking Farsi, Portuguese, Sinhalese and Tamil.

There is no consistent pattern between the recent and long-standing groups. While the Italians fare better in terms of median accessibility, the Portuguese have the second-poorest access to SLFPs. Given the acculturation process, perhaps it is less important for long-standing immigrants to use SLFPs, as their English may have improved or they may have English-speaking children to accompany them to physician visits.

6. Limitations

Although the 2SFCA method has many advantages in computing spatial accessibility to health care at a point location, the model is not without limitations. Unlike the gravity-type accessibility models that use a distance impedance function to assign weights to destination attractiveness and demand, the 2SFCA model does not have a distance friction function. Instead, it relies on a predefined travel threshold (i.e. physician catchment area or service area), which draws an artificial line between accessible and inaccessible opportunities. Opportunities (i.e. physicians) that fall within the travel threshold are treated as equally accessible, whereas opportunities outside the threshold are all inaccessible, regardless of the distance between the origin (i.e. census tract centroid) and destination (i.e. physician location). The 2SFCA model has been enhanced to address this weakness – by dividing the study area into different travel zones based on empirically calibrated travel time thresholds or adding hypothetical distance decay to the 2SFCA model (Luo and Qi 2009, Wang and Roisman 2011). However, this article does not use the enhanced 2SFCA model, as there is no information available on the travel behaviour of the eight selected immigrant groups and the aim of this article is to provide a comparison among the eight groups at an aggregate level in accessing SLFPs, rather than to simulate their travel behaviour.

The modelling exercise in this article concerns travel behaviour based on the private automobile, and the 2SFCA model is run based on the least amount of travel time calculated by road length and speed limit using Network Analyst in ArcGIS 9.3. It is, however, acknowledged that individual travel patterns may vary due to the possible choice of another transportation mode, such as public transit. Commuting to work by public transit is especially prevalent among recent immigrants in Toronto. This means of commuting is usually chosen by approximately 36% of recent immigrants, compared with 21% of the Canadian-born population (Heisz and Schellenberg 2004). There is limited information about the mode of transport used for visiting family physicians. However, Wang and Roisman (2011) report that 74% of the Chinese immigrants in their survey said they drove to their family physician, whereas only 5% went by public transit. Within the Toronto CMA, the City of Toronto has an extensive network of buses, subway trains and streetcars, while the predominant mode of transport in the suburbs is the automobile (Kennedy 2002). With more immigrants settling in the suburbs, it is important to examine the use of public transit in both the city and the outskirts, particularly for newcomers who may not have access to a car during their settlement period and who have to wait at least 6 months before taking a road test to obtain a driver's licence in the province of Ontario (Heisz and Schellenberg 2004, Lo et al. 2007).

The physician data from the CPSO also have limitations. The language used is self-reported by physicians in the CPSO's annual survey and the information may not always be accurate and complete. For example, in the data set some physicians self-reported speaking Chinese, which is an umbrella term, without indicating the specific Chinese language or dialect spoken (e.g. Mandarin, Cantonese). Even when a physician did report speaking a non-official language, there is no information about proficiency in that language.

The 2SFCA accessibility model is based on the assumption that immigrants in the selected eight groups favour linguistically appropriate physicians. Although a number of recent studies suggest linguistic barriers facing foreign-born immigrants and a preference for a same-language health-care practitioner (Komaromy et al. 1996, Derose and Baker 2000, Wang 2007, Loue 2008, Wang et al. 2008), it is acknowledged that there may be individual differences in physician-seeking behaviour due to proficiency in English, personal choice and other factors. Some sort of survey data on individual travel behaviour for all the groups would permit calibration of the model for each individual study group by applying travel thresholds specific to group, location or transportation mode. Trip or demand origins would be represented by residential addresses rather than census tract centroids, allowing for more accuracy in modelling. Given the unavailability of such comprehensive survey data, with the exception of data from a survey with a small sample of Mainland Chinese immigrants as reported in Wang (2007), this article serves as a departure point for understanding group-based patterns in the accessibility of family physicians. This approach yields important implications for identifying underserviced areas at an urban level in terms of access to same-language physicians and the extent of spatial mismatch.

The sub-accessibility measure (4) is derived from multiplying the general 2SFCA measure (3) by adding two location-quotient-type terms measuring the relative abundance of opportunities in the sub-opportunity set and competition from users in a sub-population group. The 2SFCA model (3) cannot be directly applied in calculating accessibility of a subgroup in the demand side to a subgroup in the supply side as there is no evidence that physicians serve same-language patients only and immigrants visit same-language physicians only. Future research can collect data from physicians on the composition of their patients in terms of ethnicity and language in order to enhance the sub-accessibility measure by estimating a ratio of demand that is matched with a sub-opportunity set.

7. Implications and future research

The results have important policy and practical implications for identifying shortage areas in terms of health professionals serving culturally diverse populations, developing health-care programmes, addressing issues related to foreign-trained physicians, formulating health promotion strategies and enhancing the delivery of primary care relevant for immigrants. The spatial inequality in accessing family physicians for the general population as well as for selected immigrant populations, as revealed in this article, challenges the Canadian health-care system with its mandate to provide universal coverage on the basis of need rather than the ability to pay. Poor access to health services may be reflected in delayed care-seeking, absence of preventive care and low levels of patient satisfaction. In Canada, physicians generally see proximity to hospitals as a crucial location factor (Rosenberg 1984). There are too few incentives for physicians to relocate within an urban area so that spatial equity in health-care provision might be achieved. Some community health centres in Toronto have been striving to promote health and well-being and to improve access to services for immigrants and refugees, through such initiatives as the Access Alliance Multicultural Health and Community Centre. However, they have had limited success, with only two locations currently in the city core.

The findings may be useful for primary care physicians and clinicians in increasing their cultural understanding of immigrant patients, which is a critical factor in addressing the health-care needs of immigrant populations. As many newcomers face linguistic barriers to health-care access, it would be helpful if interpreter services were available at family physicians' offices. Like many other highly qualified immigrants in Canada, foreign-trained physicians are often forced to enter other occupations. Only 21% of foreign-trained medical graduates pass the Canadian qualifying examinations, compared with 95% of Canadian-trained graduates (Gray 1999), despite their potential to provide culturally sensitive health care to their communities.

The modelling of spatial accessibility could be enhanced in future research by collecting and using primary survey data on the actual travel behaviour of various social groups in accessing primary health care. For example, a detailed travel diary or survey would permit calibration of a more accurate travel threshold or distant decay function, which should be dependent on residential location, transportation mode and other personal socio-economic characteristics, and thereby improve the 2SFCA accessibility modelling of physician-seeking behaviour. Accommodating variations in travel time threshold associated with different socio-economic groups living in different neighbourhoods would greatly further current accessibility research that generally uses a uniform theoretic travel time threshold. Also, it would be interesting to explore the relationship between accessibility to health services and changes in immigrant health status and, in particular, whether spatial accessibility contributes to the diminishing ‘healthy immigrant effect’ (Newbold 2005).

Acknowledgements

I would like to thank the two anonymous reviewers and the journal editor for their constructive comments that greatly improved the paper. Financial assistance from SSHRC (Social Science and Humanities Research Council of Canada) (Grant # 410-2008-2575) and CERIS, The Ontario Metropolis Centre is gratefully acknowledged. My research assistant Deborah Roisman did an excellent job in this research project. Thank you, Deborah!