Surname lists to identify South Asian and Chinese ethnicity from secondary data in Ontario, Canada: a validation study

Because many secondary data sources used in health research do not include information on race or ethnicity, surnames are often used as a proxy when studying health care for ethnic populations. In the United States, the Bureau of the Census compiled a list of "Spanish surnames" in the 1950s which has since been updated every decade [1]. These lists have been used extensively to study Hispanic populations using secondary data sources.

Canada's two largest minority populations are those with South Asian (from India, Pakistan, Bangladesh and Sri Lanka) and Chinese origins [2]. Surname lists to identify both of these population groups have been developed. To identify South Asian populations, two proprietary computerized algorithms using first and surnames were developed in the United Kingdom [3, 4]. The SANGRA algorithm yielded sensitivity from 89% to 96% versus self-reported ethnicity from various data sources across England, with positive predictive value between 80% and 89% [3]. In contrast, the Nam Pehchan program had a sensitivity of 61% and a positive predictive value of 97% against self-reported ethnicity in a national sample [4]. In addition an American list of South Asian surnames was developed by Lauderdale and Kestenbaum [5]. However, it had only 38% sensitivity and 77% positive predictive value against census records. In addition, the US Census derived an Asian-Indian surname list from the 2000 Census [6], but the test characteristics of this list have not been evaluated. Sheth et al. [7] derived a list of South Asian surnames using Canadian death certificate data, by selecting people whose place of birth was South Asia or countries with large South Asian populations. They performed a telephone survey of 100 people who had surnames on the list and 100 who did not, and found that the list had a sensitivity of 96% versus self-reported ethnicity. However, this list did not specifically exclude surnames that are common to both South Asian and other populations, such as Khan, Ahmed, DeSouza or Fernandes. To identify Chinese populations, several surnames lists have been developed. In 1990, Hage et al. [8] developed a list of 145 Chinese surnames from the Melbourne, Australia telephone directory. In 1993, Choi et al. [9] published a list of over 200 Chinese surnames that was derived by finding the surnames of all people whose place of birth recorded in the Ontario vital statistics registry was China, Taiwan or Hong Kong. This list was further expanded and modified by Tjam [10] and by Quan et al. [11]. However, a limitation of most of these surname lists for both minority groups is that they are either unvalidated, or they are validated only in small settings rather than in a broad population, and so their generalizability to other contexts is uncertain.

The objective of this study was to develop and validate surname lists that could be used to identify cohorts of people with South Asian and Chinese origins from secondary data sources. The study was conducted in Ontario, Canada's largest province and also one of its most ethnically diverse. To identify such cohorts, we want to be confident that individuals identified using these surname lists truly come from the target ethnic population. Ensuring that such cohorts are representative of the entire ethnic population is secondary. Therefore, our goal was to maximize positive predictive value of the surname lists, potentially sacrificing sensitivity.

The initial list of potentially South Asian surnames included 13,949 names to be screened. Of them, 9,950 were selected by consensus for the final list of uniquely South Asian surnames. From the initial 1,185 Chinese surnames, 1,133 were selected for the final list. Each surname list was linked with the nominal RPDB, and all 16,688,384 current and former residents of Ontario were assigned to South Asian ethnicity (500,807), Chinese ethnicity (808,670) or the General Population (15,378,907). The 200 most common surnames from each group are presented in Table 1.

There were 69,859 CCHS respondents who were included in this validation study, of whom 1,400 self-identified as South Asian (5.4% of the weighted population), and 1,129 self-identified as Chinese (4.0% of the weighted population). The baseline characteristics of the population are shown in Table 2. The sensitivity, specificity and positive and negative predictive values of the South Asian and Chinese surnames lists are shown in Table 3. Both lists perform extremely well, with approximately 90% positive predictive value. The sensitivity of the South Asian list was low, as expected, since the many surnames common for but not unique to South Asians were excluded from the final list. Compared to the previously published Chinese list, our modified Chinese surname list had slightly better positive predictive value with the sacrifice of a small amount of sensitivity.

Table 4 shows the positive predictive value of the surname lists among patients stratified by sex, age and immigration status. The surname list was slightly less likely to accurately predict self-identified ethnicity among South Asian women, but the difference between sexes was not marked among people with Chinese origin. As expected, longer time since immigration was associated with worse positive predictive value in both ethnic groups. Both lists performed poorly among those individuals who were Canadian-born.

We have developed surname lists that can accurately identify people with self-reported South Asian or Chinese origins through secondary data sources. Despite the relatively low prevalence of minorities in the total population, both lists had excellent positive predictive value [15], so they can be used to confidently assemble ethnic minority cohorts for epidemiologic or health services research studies of people with South Asian or Chinese origins. However, the South Asian list had, as expected, low sensitivity. Therefore, it cannot be used, for example, to determine the proportion of a disease cohort that has South Asian origins. The Chinese list had a slightly better positive predictive value than the previously published list on which is was based [11], and also retained good sensitivity.

The surname lists showed markedly poorer positive predictive value for people who were Canadian-born. This most likely reflects people from other backgrounds who have adopted an ethnic surname through marriage, or to people whose parents were from different ethnic backgrounds. (However, it is noteworthy that people of South Asians and Chinese origins are the least likely in Canada to report being married to someone from outside their ethnic group [2].) It may also indicate that Canadian-born minorities may not self-identify with their ethnic group. Hence, although the surname lists are useful in the overall population, they may be less valid for use in this specific subgroup. In the future, as more people adopt surnames from other ethnic groups and as more people with multiple ethnic origins are born, the usefulness of surname lists to identify ethnic populations from secondary data may decline.

Identification of Chinese surnames is relatively straightforward, as about 95% of the Chinese population is Han Chinese. There are only a few hundred different common surnames in Chinese script [16], and a limited number of Westernized spellings of each name. In contrast, the South Asian population includes a large number of language, religious and cultural groups, each with a multitude of common surnames. Hence, our original list of South Asian surnames to screen was more than 10 times longer than the list of Chinese surnames. Because we sought to maximize positive predictive value by limiting our final list to include only those names that were uniquely South Asian, the final list includes mostly Hindu, Sikh and Sri Lankan surnames. Names used by South Asian Muslims or Christians are shared with people from those faiths in other parts of the world, and indeed may be more common in people with other ethnic backgrounds (e.g., Arab, Persian or Portuguese). Although including these names could have improved the sensitivity of our surname list, their inclusion would have led to an unacceptably dramatic drop in positive predictive value, and would have rendered the surname list invalid for our purposes. Other South Asian name algorithms have included such names to maximize sensitivity, but this came at a cost of very poor positive predictive value [17].

There are limitations to this methodology. As noted above, the included surnames on the South Asian list are mostly Hindu, Sikh and Sri Lankan names while surnames from other religious or cultural groups are excluded. Since there is evidence of biological and behavioural heterogeneity between South Asian populations [18‐21], any cohorts built using the surname list may therefore systematically exclude people with different biological or clinical risks. The same is true for the Chinese list, although the Han population does make up an overwhelming majority of the Chinese population. In addition, there are limitations to the gold standard. We restricted our gold-standard definition to questions about ancestry and cultural/racial background. We did not use survey questions about place of birth or first language as these may not be specific nor sensitive for ethnicity. There may have been confusion about the meaning or intent of the survey questions, which might have led to incorrect responses. In particular, terms like "South Asian" (versus "Southeast Asian") or "Indian" (versus "Aboriginal") may have been confused by both respondents and interviewers. Finally, generalizing surname lists to populations outside of the areas in which they were derived is problematic, because of differing immigration patterns to different parts of the world. This phenomenon was demonstrated with the Nam Pehchan algorithm in the United Kingdom, which had a sensitivity of over 94% when evaluated for individuals in Bradford where it was developed, but only 61% when evaluated for individuals from across England [4]. Therefore, our surname list would likely not have the same performance in other jurisdictions.

In conclusion, these surname lists can be used to accurately identify cohorts of people with South Asian and Chinese origins from secondary data sources, although these cohorts would, by necessity, not always be comprehensive or representative of the entire South Asian or Chinese population.