Abstract

Background: The UK Renal Registry (UKRR) reports on equity and quality of renal replacement therapy (RRT). Ethnic origin is a key variable, but it is only recorded for 76% patients overall in the UKRR and there is wide variation in the degree of its completeness between renal centres. Most South Asians have distinctive names.

Aim: To test the relative performance of a computerized name recognition algorithm (SANGRA) in identifying South Asian names using the UKRR database.

Design: Cross-sectional study of patients (n = 27 832) starting RRT in 50 renal centres in England and Wales from 1997 to 2005.

Methods: Kappa statistics were used to assess the degree of agreement of SANGRA coding with existing ethnicity information in UKRR centres.

Results: In 12 centres outside London (number of patients = 7555) with 11% (n = 747) self-ascribed South Asian ethnicity, the level of agreement between SANGRA and self-ascribed ethnicity was high (κ=0.91, 95% CI 0.90–0.93). In two London centres (n = 779) with 21% (n = 165) self-ascribed South Asian ethnicity, SANGRA's agreement with self-ascribed ethnicity was lower (κ=0.60, 95% CI 0.54–0.67), primarily due to difficulties in distinguishing between South Asian ethnicity and other non-White ethnic minorities. Use of SANGRA increased numbers defined as South Asian from 1650 to 2076 with no overall change in percentage of South Asians. Kappa values showed no obvious association with degree of missing data returns to the UKRR.

Conclusion: SANGRA's use, taking into account its lower validity in London, allows increased power and generalizability for both ethnic specific analyses and for analyses where adjustment for ethnic origin is important.

Introduction

Over 2 million people of South Asian descent (South Asians: persons whose families originated from India, Pakistan, Bangladesh or Sri Lanka) live in the UK (2001 UK census). Although they make up only 4% of the UK population, South Asian patients are over-represented amongst those starting renal replacement therapy (RRT).1–4 Data on those starting RRT in the UK are held in the Renal Association UK Renal Registry (UKRR). Within the UKRR, the ascertainment of ethnicity is based on mixed methods. Some renal centres provide near-complete data by cross-linking their renal data system with data on self-ascribed ethnicity from the hospital Patient Administration Systems (PASs). In other renal centres, ethnicity is manually entered on the renal data system by clinical staff, but these entries are not always complete, leading to incomplete ethnicity coding across the whole UKRR dataset. Assessment of health inequalities is hampered by incomplete data on ethnic-specific acceptance rates, quality of care and outcomes of RRT. To circumvent this problem, analyses have usually been restricted to data from renal centres with high degree of ethnicity recording. However, this limits the power and generalizability of the analyses.

One way to address these issues is to use the indirect information on ethnicity that is captured in patient names. South Asian names in particular are distinctive and allow identification. South Asian Names and Group Recognition Algorithm (SANGRA) is a validated computer algorithm that identifies South Asian names.5 SANGRA's name directory is based on a national collection of South Asian names. Its validation was carried out on London and Midlands hospital in-patient admission data in the mid- to late 1990s. SANGRA had a sensitivity of 89–96% in all data sets, but specificity was 94% in London vs. 98% in the Midlands.5 This algorithm has been used with success on other datasets6 and appeared to be of potential value to identify South Asian patients in the UKRR data.

The aims of the present study were 3-fold: (i) to assess the relative performance of SANGRA as a tool in identifying South Asian names using the data from dialysis centres with known self-ascribed ethnicity as reference centres, (ii) to assess the utility of using SANGRA for the UKRR data to identify centres with potential bias in the way self-ascribed data are being collected and (iii) to check if SANGRA identifies a different type of South Asian patients by examination of baseline characteristics when compared to South Asians known to the UKRR.

Patients and methods

The UK Renal Association Registry

The data used in the present study were collected by the UKRR. Details of data collection methods are described elsewhere.3,7 The master data set contains 27 832 incident patients starting RRT between 1 January 1997 and 31 December 2005 in 50 renal centres in England and Wales, all equipped with electronic data transfer facilities (by 2005, 50 out of total 58 centres in the UK were participating in the UKRR). For these patients, information was available on socio-demographic variables and on diagnostic and clinical history including start date of RRT, dates of changes in treatment modality (e.g. transplantation), co-morbidity and survival experience. The UKRR has approval from the Patient Information Advisory Group (PIAG) to retain name information for data quality checks. The LSHTM ethics committee approved the current study.

SANGRA

SANGRA software, including the reference name directories to identify people of South Asian origin, was run by the UKRR staff to assign ethnicity based on patients’ names as a new variable. The dataset was then anonymized by deleting all the names, leaving the registry ID numbers, existing registry ethnicity codes and SANGRA ethnicity codes. Only SANGRA's output variables, as well as new ID numbers and relevant clinical variables, were subsequently provided to the researchers. Names were not disclosed to the researchers.

Statistical analyses

Centres which have a high degree of completeness in their ethnicity data (⩾95%), and where self-reporting of ethnicity alone was used as the method of collecting this information, were selected as reference centres. Overall agreement between SANGRA and the existing UKRR ethnicity coding was assessed using 2 × 2 kappa statistics (and their 95% CI) for both reference and other UKRR centres (without self-ascribed ethnicity coding) combined and separately by degree of completeness of their UKRR ethnicity data. Standard diagnostic accuracy statistics were calculated (sensitivity, specificity, PPV and NPV) with their 95% CIs for two reasons (i) in order to assess whether these estimates for the London centres differed from those for the centres outside London and (ii) to compare the efficiency of SANGRA in predicting South Asian ethnicity of the UKRR data to the published validation results.5 We also assessed whether prediction of ethnicity by SANGRA led to a change in South Asian patients’ baseline characteristics by testing for differences between patients who were only identified by SANGRA as South Asians (but not by the UKRR), and patients who were South Asians known to the UKRR (but not identified by SANGRA).

Chi-squared and Fisher's exact tests were used as appropriate to test associations with binary and categorical variables. The Mann–Whitney test was used for skewed continuous variables. The significance level for the statistical tests was set to 5%. The analysis was carried out using Stata 10.

Results

Analysis was based on 27 832 incident patients from 50 centres accepted onto RRT between 1 January 1997 and 31 December 2005. UKRR-recorded ethnicity data were available for 21 134 (76%) patients with 1650 of these coded as South Asians (see top two rows of Figure 1). Figure 2 shows the distribution of UKRR-recorded ethnicity data by renal centres (red bars for South Asians and yellow bars for non-South Asians). The percentage of missing data on ethnicity ranged from 0 to 91% (blue bars): the highest proportions of missing ethnicity data (>20%) occurred generally in centres with few South Asian patients (Figure 3, <10% South Asian ethnicity).

Figure 1.

Incident cohort of patients and definition of cohorts used in this analysis.

Figure 1.

Incident cohort of patients and definition of cohorts used in this analysis.

Figure 2.

Centre-wise percentages of South Asians known to the Registry (red), percentage of missing data on ethnicity (blue), and percentage non-South Asians (yellow) in centres participating in the UKRR. Black horizontal lines show the percentage of South Asians that was predicted by SANGRA in each centre on the basis of patients’ names.

Figure 2.

Centre-wise percentages of South Asians known to the Registry (red), percentage of missing data on ethnicity (blue), and percentage non-South Asians (yellow) in centres participating in the UKRR. Black horizontal lines show the percentage of South Asians that was predicted by SANGRA in each centre on the basis of patients’ names.

Figure 3.

Association between the proportion of South Asian patients (as ascertained in the UKRR records) and the proportion of patients with missing ethnicity information for centres with South Asian attendance (dots represent centres, n = 40).

Figure 3.

Association between the proportion of South Asian patients (as ascertained in the UKRR records) and the proportion of patients with missing ethnicity information for centres with South Asian attendance (dots represent centres, n = 40).

Because the SANGRA ethnicity code is based on patient names, the total number of South Asians identified by SANGRA was more than that in the UKRR records (2076 = 7.5% of all records, vs. 1650 = 7.8% of the subset of UKRR records with available ethnicity data). Of the 1650 South Asians known to the UKRR, SANGRA identified 89% (n = 1468) as South Asians, of the Whites known to the UKRR 93 (0.5%) were mis-classified by SANGRA to be South Asians (Figure 1). Centre-wise proportions of South Asians identified using SANGRA on the basis of their name is depicted by black horizontal lines in Figure 2.

Agreement between SANGRA and self-ascribed ethnicity data in a subset of centres

To assess the level of agreement between the SANGRA algorithm and self-ascribed ethnicity, only the 14 centres with a high degree of data completeness (⩾95%) and self-ascribed ethnicity coding were included in the analysis. There were 7555 patients in 12 centres from outside London and 779 incident RRT patients in two London centres. The proportion of South Asians known to the UKRR ranged from 1 to 16% in centres located outside London and was 12 and 23%, respectively, for the two London centres. Figure 2 shows how SANGRA-predicted South Asian ethnicity (black horizontal lines) differs by 5% to UKRR-recorded South Asian ethnicity (red bars) in the two London centres, whilst for the centres located outside London, the top bars coincide. In line with the visual impression from Figure 2, in 12 reference centres outside London, the overall kappa value was 0.91 (95% CI 0.90–0.93) which was much higher than the kappa value of 0.60 (95% CI 0.54–0.67) for centres within London. In the reference centres, SANGRA classified a high percentage of people to be of ‘other’ ethnic minority in South Asian (Table 1). Those of ‘other’ ethnic origin as coded by the UKRR included people of Middle-Eastern and Arabic origin. There was no evidence for difference in diagnostic accuracy between men and women (data not shown).

In order to compare these results with the results of previous validation studies of SANGRA which use standard diagnostic statistics, we calculated measures of sensitivity and specificity for SANGRA in the UKRR database. These calculations assume that the self-ascribed PAS hospital data are the gold standard as has been done in previous validation studies of SANGRA. For centres outside London, sensitivity was 93.0% (95% CI 91.0–94.8), specificity 99% (98.7–99.2), with positive predictive value 90.8% (88.6–92.8) and negative predictive value 99.2% (99.0–99.4). In the London centres the performance of SANGRA was significantly poorer for both sensitivity [78.8% (71.8–84.3)] and specificity [87.3% (84.4–89.8)], with positive predictive value 62.5% (55.8–68.8) and negative predictive value 93.9% (91.6–95.7) (see Supplementary Table 1).

Table 1

Numbers of those identified as South Asians by SANGRA from the UKRR ethnicity codes in reference centres located in or outside London

Registry ethnicity code Outside London
 
London
 
% (n) identified by SANGRA as South Asian in all centres Total number of patients in all 14 reference centres 
 % (n) identified by SANGRA as South Asian Total number of patients in 12 reference centres % (n) identified by SANGRA as South Asian Total number of patients in two reference centres   

 
South Asians 93.0 (695) 747 78.8 (130) 165 90.5 (825) 912 
Afro-Caribbeans 6.6 (15) 227 20.7 (17) 99 9.8 (32) 326 
Chinese 3.0 (1) 33 0 (0) 2.7 (1) 37 
Others 44.4 (32) 72 44.3 (58) 131 44.3 (90) 203 
Whites 0.3 (22) 6476 0.8 (3) 380 0.4 (25) 6856 
Registry ethnicity code Outside London
 
London
 
% (n) identified by SANGRA as South Asian in all centres Total number of patients in all 14 reference centres 
 % (n) identified by SANGRA as South Asian Total number of patients in 12 reference centres % (n) identified by SANGRA as South Asian Total number of patients in two reference centres   

 
South Asians 93.0 (695) 747 78.8 (130) 165 90.5 (825) 912 
Afro-Caribbeans 6.6 (15) 227 20.7 (17) 99 9.8 (32) 326 
Chinese 3.0 (1) 33 0 (0) 2.7 (1) 37 
Others 44.4 (32) 72 44.3 (58) 131 44.3 (90) 203 
Whites 0.3 (22) 6476 0.8 (3) 380 0.4 (25) 6856 

Note: The centres included for the analysis are Dorset, Nottingham, Basildon, Leicester, Reading, Sheffield, Wolverhampton, Birmingham-Heartlands, Birmingham-QEH, Preston, Dudley, New Castle, London H and Charing Cross and London Royal Free.

Agreement between SANGRA and Registry ethnicity coding in all centres in the UK

The above analysis is based on only 14 centres with high degree of completeness of UKRR ethnicity data and PAS hospital systems which collect self-ascribed ethnicity data. UKRR relies largely on the ethnicity data entered manually by the renal staff across the UK rather than using self-reported ethnicity from hospital systems. We were interested in checking if the quality of UKRR's ethnicity coding is dependent on how complete these data were for each of the centres. This would be reflected in an association between kappa values for the centres with South Asian patients and the extent of missing ethnicity data in those centres. A plot of kappa against the proportion of missing ethnicity data did not reveal any association (Figure 4). For example, SANGRA performed better in Chelmsford (κ=1.0) which had 77% of its ethnicity missing than in Liverpool (κ=0.42) with only 22% of its ethnicity missing.

Figure 4.

Association between proportion of missing ethnicity data and level of agreement between SANGRA and UKRR ethnicity data (as estimated by kappa statistics) for all 40 centres with South Asian patients. The arrows indicate the centres with the highest and lowest kappa statistics (1 and 0.42, respectively).

Figure 4.

Association between proportion of missing ethnicity data and level of agreement between SANGRA and UKRR ethnicity data (as estimated by kappa statistics) for all 40 centres with South Asian patients. The arrows indicate the centres with the highest and lowest kappa statistics (1 and 0.42, respectively).

Impact of using SANGRA on baseline characteristics at start of RRT

Of the 24% (6698/27 832) observations with missing information on ethnicity, SANGRA identified 5% (n = 319) to be South Asians (Figure 1). Hence, the overall increase of South Asians was from 1650 to 2076. Of the 1650 patients coded as South Asians by UKRR, 89% (n = 1468) were correctly identified as South Asians by SANGRA leaving 11% (n = 182) mis-classified as non-South Asians by SANGRA. Comparison of baseline characteristics of South Asians as classified by UKRR alone (n = 182) with those identified by SANGRA alone (n = 608) is displayed in Table 2. We did not perform the comparison on the full dataset in order to avoid the results being driven by the relatively large number of South Asians (n = 1468) who were common to both the UKRR and SANGRA ethnicity variables. Across all groups, there was no evidence for a difference (Table 2).

Table 2

Characteristics of South Asians known to UKRR alone and of those predicted by SANGRA alone. If not indicated otherwise, numbers in brackets represent percentage breakdowns of column totals

 UKRR-only South Asians SANGRA-only South Asians COMPARISON South Asians P (Chi-square) 
 N = 182 N = 608  

 
Age at start of RRT    
Median years of age (IQR) 58.1 (46.8–67.2) 58.4 (46.5– 67.4) 0.96a 
    
 n (%) n (%)  
Distribution of age groups    
    18–54 years 78 (43) 254 (42) 0.86 
    55–64 years 40 (22) 127 (21)  
    >65 years 64 (35) 227 (37)  
Gender    
    Male 118 (65) 355 (58) 0.12 
    Female 64 (35) 253 (42)  
Primary renal disease    
    Diabetes 71 (39) 186 (31) 0.28 
    Glomerulonephritis 13 (7) 47 (8)  
    Polycystic kidney disease 4 (2) 12 (2)  
    Pyelonephritis 8 (4) 44 (7)  
    Renovascular disease 13 (7) 56 (9)  
    Other 13 (7) 66 (11)  
Uncertain + missing 60 (33) 197 (32)  
Treatment modality    
At start of RRT    
    Haemodialysis 131 (72) 436 (72) 0.41 
    Peritoneal dialysis 49 (27) 155 (25)  
    Transplant 2 (1) 17 (3)  
    Recovered  
    No RRT  
 UKRR-only South Asians SANGRA-only South Asians COMPARISON South Asians P (Chi-square) 
 N = 182 N = 608  

 
Age at start of RRT    
Median years of age (IQR) 58.1 (46.8–67.2) 58.4 (46.5– 67.4) 0.96a 
    
 n (%) n (%)  
Distribution of age groups    
    18–54 years 78 (43) 254 (42) 0.86 
    55–64 years 40 (22) 127 (21)  
    >65 years 64 (35) 227 (37)  
Gender    
    Male 118 (65) 355 (58) 0.12 
    Female 64 (35) 253 (42)  
Primary renal disease    
    Diabetes 71 (39) 186 (31) 0.28 
    Glomerulonephritis 13 (7) 47 (8)  
    Polycystic kidney disease 4 (2) 12 (2)  
    Pyelonephritis 8 (4) 44 (7)  
    Renovascular disease 13 (7) 56 (9)  
    Other 13 (7) 66 (11)  
Uncertain + missing 60 (33) 197 (32)  
Treatment modality    
At start of RRT    
    Haemodialysis 131 (72) 436 (72) 0.41 
    Peritoneal dialysis 49 (27) 155 (25)  
    Transplant 2 (1) 17 (3)  
    Recovered  
    No RRT  

aMann–Whitney test.

Discussion

The present study validates SANGRA, a name-recognition algorithm for South Asians, in a major national dataset, the UKRR database. Similar to previous studies where SANGRA was used,5,6 it identifies South Asians with high accuracy in data from centres located outside London. However, SANGRA did not perform as well in data from London centres. The baseline characteristics of the South Asians coded by UKRR alone was comparable to those of the South Asians identified by SANGRA alone except diabetic nephropathy as cause of renal disease.

SANGRA is not the only automatic name-recognition software in use for UK data. The other algorithm, Nam Pehchan (NP), uses a much smaller name directory of South Asian names compiled mainly in Bradford, where it has been shown to have a sensitivity of 100%.8,9 However, on validation with national reference datasets, NP's sensitivity was only 61%. By comparison, the sensitivity of SANGRA in the UKRR dataset is 93%. This difference in performance between SANGRA and NP may have been due to the size and nature of the name directories used in the two algorithms. NP uses stem-matching of South Asian names that are not nationally representative. SANGRA uses full names, with added variations due to spelling differences, and its name-directory is based on listings with national coverage. Given the higher validity of SANGRA we opted to use the latter. A direct comparison between SANGRA and Nam Pechan was not possible for logistical reasons.

Both SANGRA and NP performed better with ethnicity data from centres outside London than those from the capital.9 The difference in performance of name-recognition algorithms for centres in London when compared to those outside London may be due to several reasons.

Inter-marriage between South Asians and non-South Asians may result in South Asian women having non-South Asian surnames and vice versa. However, there were no obvious gender differences within London (data not shown), suggesting that this is not the main mechanism for misclassification. South Asia has one of the largest Muslim populations in the world but their names are not exclusive to South Asia.5,9 From un-published 1991 census data up to 7% of people that could potentially be classified as South Asian by SANGRA were Muslims originating from non-South Asian countries.5 It is most likely that the lower performance for SANGRA in the London centres is partly due to the much higher prevalence of non-South Asian Muslims relative to the rest of the UK. The UKRR registry may comprise patients of mixed ethnicity with a non-indicative surname coded as South Asians in their records. This would increase the apparent number of false negatives produced by SANGRA artificially. Amongst those South Asians recorded only by the UKRR and not identified by SANGRA the proportion of diabetic nephropathy is similar to the total proportion of the South Asians of the UKRR records which is in line with the hypothesis that SANGRA misses those with mixed South Asian ethnicity. Finally, people who regard themselves as ‘East African-Asians’ or ‘Indo-Caribbean’ are often classified as ‘others’ by hospital systems. This may also be another source of mis-classification of South Asian ethnicity in the UKRR dataset, giving rise to an apparent increase in the number of false-positives generated by SANGRA.10

Strengths/limitations

The strength of the present analysis is the use of a national registry with multiple centres to assess the relative performance of SANGRA. However, the subset with data on self-ascribed ethnicity was limited to data from only 14 centres, albeit from different parts of England. Hospital systems, although based on self-reported ethnicity, classify people who regard themselves as ‘East African-Asians’ or ‘Indo-Caribbean’ as ‘others’.10 Hospital data also lack information on religion, which would have enabled the systematic examination of more precisely defined population subgroups. Unfortunately, we were restricted to using hospital data due to the nature of our ethical approval. Visual inspection of names would have helped to identify the exact source of disagreement between UKRR ascertainment and SANGRA-prediction of South Asian ethnicity in the London centres. Visual inspection of the 182 patients who were not identified as South Asians by SANGRA would have been particularly useful in assessing the diagnostic accuracy of SANGRA in predicting South Asian ethnicity from missing UKRR ethnicity records.

We were unable to extend the evaluation of SANGRA's relative performance to all renal centres for two reasons. First, the negative association that was observed between the centre-wise proportions of South Asians and the proportion of missing ethnicity recordings (Figure 3) indicated that the proportion of recorded ethnicity within the UKRR data for a given dialysis centre is an indirect marker of that centre's prevalence of South Asians on RRT.

Outside the 14 renal centres used for the purpose of SANGRA's relative performance, the proportion of missing ethnicity recordings varied across the centres (5.3–91%). Such a variation would be unhelpful in judging the performance of SANGRA since the missing data would introduce instability in the kappa estimates.

Secondly, there was variation in how the ethnicity records were generated for the UKRR dataset. The methods ranged from ascertainment by self-ascribed ethnicity to clinical guessing by hospital staff, the latter being of unknown accuracy. Any inaccuracies in ascertainment of ethnicity may contribute to the observed variations in kappa (measure of agreement) across the centres (Figure 4). However, it was reassuring that this variation in agreement was not associated with the centre's completeness in ethnicity data. The low specificity of SANGRA in the London centres might have been partly overcome, if UKRR had asked the centres to collect information on the country of birth of their patients at an earlier stage of the data collection process. However, given previous experience with low data returns it is unlikely that UKRR will be able to get full data on this variable.

Implications of the present analysis

All previous analyses using the UKRR database were restricted to data from those centres with a high percentage of ethnicity ascertainment due to concerns over bias.

The present analysis suggests that this approach is not necessary with the availability of name-recognition algorithms such as SANGRA, if one accepts a small degree of potential misclassification. Future analyses will include data from all renal centres to increase the power of the study as well as the generalisability of the results. For example, inclusion of patients from all the dialysis centres in conjunction with ethnicity prediction by SANGRA will result in more representative estimates for baseline characteristics and survival outcomes of South Asians on RRT.

Conclusions

SANGRA is a very useful tool in identifying South Asian patients of dialysis centres with high accuracy outside London. Accurate prediction of South Asian ethnicity helps in a more realistic analysis on the equity and effectiveness of RRT care for people of South Asian ethnicity. However, in the long run, the importance of automatic collection of accurate ethnicity data through the hospital PASs should not be underestimated. Such a mode of ethnicity ascertainment is recommended particularly for the renal centres in London with high prevalence of Muslim communities, where SANGRA has less accuracy to identify South Asians.

Supplementary Data

Supplementary Data is available at QJMED Online.

Conflict of interest: None declared.

References

1
Roderick
PJ
Jones
I
Raleigh
VS
McGeown
M
Mallick
N
Population need for renal replacement therapy in Thames regions: ethnic dimension
Br Med J
 , 
1994
, vol. 
309
 (pg. 
1111
-
4
)
2
Roderick
PJ
Raleigh
VS
Hallam
L
Mallick
NP
The need and demand for renal replacement therapy in ethnic minorities in England
J Epidemiol Community Health
 , 
1996
, vol. 
50
 (pg. 
334
-
9
)
3
Feest
TG
Rajamahesh
J
Byrne
C
Ahmad
A
Ansell
D
Burden
R
, et al.  . 
Trends in adult renal replacement therapy in the UK: 1982–2002
QJM
 , 
2005
, vol. 
98
 (pg. 
21
-
8
)
4
Trehan
A
Winterbottom
J
Lane
B
Foley
R
Venning
M
Coward
R
, et al.  . 
End-stage renal disease in Indo-Asians in the North-West of England
QJM
 , 
2003
, vol. 
96
 (pg. 
499
-
504
)
5
Nanchahal
K
Mangtani
P
Alston
M
dos Santos Silva
I
Development and validation of a computerized South Asian Names and Group Recognition Algorithm (SANGRA) for use in British health-related studies
J Public Health Med
 , 
2001
, vol. 
23
 (pg. 
278
-
85
)
6
dos Santos Silva
I
Mangtani
P
De Stavola
BL
Bell
J
Quinn
M
Mayer
D
Survival from breast cancer among South Asian and non-South Asian women resident in South East England
Br J Cancer
 , 
2003
, vol. 
89
 (pg. 
508
-
12
)
7
Ansell
D
Feest
T
Wallis
J
British renal registry is fully electronic
Br Med J
 , 
1998
, vol. 
316
 (pg. 
471
-
2
)
8
Harding
S
Dews
H
Simpson
SL
The potential to identify South Asians using a computerised algorithm to classify names
Popul Trends
 , 
1999
, vol. 
97
 (pg. 
46
-
9
)
9
Cummins
C
Winter
H
Cheng
KK
Maric
R
Silcocks
P
Varghese
C
An assessment of the Nam Pehchan computer program for the identification of names of south Asian ethnic origin
J Public Health Med
 , 
1999
, vol. 
21
 (pg. 
401
-
6
)
10
Aspinall
P
Department of Health's requirement of mandatory collection of data on ethnic group inpatients
Br Med J
 , 
1995
, vol. 
311
 (pg. 
1006
-
9
)