4. Chapter Four: Improving Quality of Ethnic Codes in HES
4.3 Previous Research Using HES Ethnicity Data
Although the overall data quality is not very satisfactory, the HES data haven’t been neglected from the research on ethnic inequalities in health and healthcare. Lowdell et al. (2000) attempted to look at the distribution of causes of admission of people aged 65 and over across ethnic groups using Hospital Episode Statistics (1997/1998), acknowledging that about 31% of all the admissions didn’t have valid ethnicity codes. By ignoring the uncoded cases, Bardsley et al. (2000) have examined the proportional hospital admission rates for heart operation by ethnic groups in London, which has been rarely undertaken before on a London wide basis. Including ethnicity information of new born children, Hospital Episode Statistics was identified and recommended as a potential data source for analyzing generational differences in fertility among ethnic groups (Haskey and Huxstep, 2002). For example, it was used to analyse the ethnic difference in fertility in London by calculating Age-Specific Fertility Rate structures (ASFRs) and Total Period Fertility Rate (TPFR) for different ethnic groups, which has developed a more robust fertility projection methodology in London (Klodawski, 2003). In order to compare the proportion of all hospital admissions in each ethnic group, proportional admission ratios have been calculated for coronary heart disease, revascularization, diabetes and cataracts for ethnic groups using Hospital Episode Statistics across English regions (Fitzpatrick et al., 2005). Mindell et al. (2007) examined the ethnic difference in coronary revascularization procedures in London to further measure ethnic inequalities in access to health services using the proportional ratios method, concluding that even if the data was not perfect, the analysis can identify inequalities that warrant further investigation. In addition, Mann et al. (2008) have investigated the difference in age-standardised morbidity and mortality ratios and morbidity and mortality odds ratios of hepatitis C-related end-stage liver disease in ethnic minorities in England using Hospital Episode Statistics from 1997/98 to 2004/05 and the 2001 Census. These previous studies using the ethnicity data from Hospital Episode Statistics did provide importantly further analysis about health and healthcare among ethnic groups,
indicating that although the Hospital Episode Statistics is imperfect in ethnicity coding, as the data quality improves over time, the Hospital Episode Statistics with valuable information about ethnic groups is attracting more and more public health interest
In most previous research using the HES ethnicity data, it is the proportional mortality ratio (PMR) method (or proportional morbidity ratios, proportional admission ratios) that make it possible to derive ethnic disparities in health from Hospital Episode Statistics with incomplete ethnicity codes (Lowdell et al., 2000, Bardsley et al., 2000, Mindell et al., 2007). A proportional mortality ratio (PMR) measures whether the proportional mortality of the study population from a certain cause is higher or lower than the proportional mortality due to that cause in the standard or general population, where the proportional mortality or morbidity (PM) can be calculated as the ratio between the number of cases due to that cause and the total number of cases (Fitzpatrick et al., 2005). In addition, it is possible to make the denominator (total number of causes) cause specific in the PMR method. For example, the deaths from a certain kind of disease could be examined as a proportion of deaths from similar causes rather than all the causes (Aspinall and Jacobson, 2007). The PM and PMR can be expressed as follows (Aspinall and Jacobson, 2006):
Number of Cases due to Cause X Proportional Mortality or Morbidity (PM) =
Total Number of Cases
(4.1)
PM in Population A (the Study Population) Proportional Mortality or Morbidity Ratio (PMR) =
PM in Population B (the Standard or Reference Population )
(4.2)
The proportional mortality (morbidity) ratio method is particularly useful when the underlying population at risk can’t be accurately measured, since this method doesn’t require appropriate population denominators as in the age-standardised rates method (Mindell et al., 2007). The ethnicity information is poorly recorded in the HES data
and it is difficult to measure the true corresponding population for the cases with valid ethnicity codes. So the proportional mortality ratio (PMR) method was widely used in previous research conducted using the incomplete HES data. The use of the PMR method is also strengthened in the following situations. For some NHS hospital trusts, it is hard to estimate the population for a hospital catchment area, because their catchments are not discrete geographical areas and usually overlap with neighbourhood trusts (Aspinall and Jacobson, 2006). In addition, given three or five years of data are frequently used in epidemiological analysis, the ethnic group population data, which is mainly derived from the decennial census could only be accurate several years beyond the census. PMR was recommended to be more widely used by NHS organizations to monitor the health of the population (Aveyard, 1998).
Although it is argued that the bias of PMR is small and of no practical importance (Aveyard, 1998), the accuracy of the PMR method is doubtful. Firstly, this ratio depends not only on the number of cases from the disease under study but also on the number of cases of the reference disease (Bhopal, 2002). The fundamental assumption is that the distribution of cases from other causes rather than the one of interest is the same in the population of interest and reference population, which is unlikely to hold (Bhopal, 2002).So the proportional mortality (morbidity) ratio is more likely to be an overestimate when overall mortality rate is low and underestimate mortality (morbidity) if the overall rate of the comparison group is high. Secondly, another underlying assumption of using the PMR method to analyze the HES ethnicity data is that the records with no ethnicity codes that have been ignored should have a similar ethnic mix to those records that have a valid ethnicity code at the geographical or aggregated level of study, which is inappropriate and might not be true for most of the cases. Otherwise, ignoring the records with missing ethnicity codes will introduce bias and uncertainties. So users of the PMR method should be aware of the flaws and the results need to be interpreted with caution (Aspinall and Jacobson, 2007).
is flawed by the underlying assumptions, rather than simply ignoring the records with missing ethnicity codes, two methods have been developed by me, including the record linking method and the coding rate method. The aim of the first one is to improve the data quality of ethnicity codes in the HES. And the latter one is to adjust the total number of cardiovascular cases within regions based on the observed number of cases with valid ethnicity codes and the estimated coding rates across regions.