Impact of definition on the study of avoidable mortality: Geographical trends in British deaths 1981–1998 using Charlton and Holland's definitions
Introduction
Avoidable mortality is a concept that has been used to evaluate the efficiency and efficacy of health care systems. The idea was originally developed by Rutstein et al. (1976) who created a list of conditions that they considered either treatable or preventable given current medical knowledge and technology. In an ideal situation these conditions would not result in ‘unnecessary, untimely death’ (Rutstein et al., 1976). The concept was developed further by Charlton, Hartley, Silver, and Holland (1983), Charlton and Velez (1986) and Charlton, Lakhani, and Aristidou (1986) who used a selection of causes of death from the Rutstein et al. (1976) list and first coined the phrase ‘avoidable mortality’. Holland (1988) developed the ideas further by carrying out the main study on avoidable mortality in the European Community and creating a different list of causes of avoidable death that is the commonly used alternative to Charlton et al.'s (1983) definition.
To date there have been few studies of avoidable mortality in Great Britain (Bauer & Charlton, 1986; Carr-Hill, Hardman, & Russell, 1987; Charlton, Hartley, Silver, & Holland (1983), Charlton, Lakhani, & Aristidou (1986); Holland, Fitzgerald, Hildrey, & Phillips, 1994) and indeed there has been little work that has looked at avoidable mortality in Great Britain since the early 1980s (Andreev, Nolte, Shkolnikov, Varavikova, & McKee, 2003; Nolte & McKee, 2003). Moreover, while these studies examined temporal trends, relatively little attention has been given to geographical variations. There have been studies comparing the amount of avoidable mortality between countries (Andreev et al., 2003; Boys, Forster, & Jozan, 1991; Charlton & Velez, 1986; Gaizauskiene & Westerling, 1995; Mackenbach, Kunst, Looman, Habbema, & van der Maas, 1988; Manuel & Mao, 2002; Nolte & McKee, 2003; Poikolainen & Eskola, 1988; Simonato, Ballard, Bellini, & Winkelmann, 1998; Treurniet, Boshuizen, & Harteloh, 2004; Velkova, Wolleswinkel-van den Bosch, & Mackenbach, 1997), but to date there have been no studies specifically examining how avoidable mortality varies geographically within an entire country. Leveque, Humblet and Lagasse (2001) examined the geographical trends in the impact of road traffic accidents on total avoidable mortality in Belgium, and Andreev et al. (2003) produced maps of age-standardised avoidable mortality rates across regions in European Russia for 1999–2000, but did not undertake any analysis of these geographical variations.
The original list of ‘sentinel health events’ devised by Rutstein et al. (1976) has formed the basis behind most studies on avoidable mortality since 1976. The diseases chosen from this list have varied mainly between two different approaches devised by Charlton et al. (1983) and Holland (1988). As Lagasse, Humblet, Lenaerts, Godin, and Moens (1990) observe, the choice of causes designated as ‘avoidable’ will have an important influence on the relationships observed. There has, however, been little debate over which causes of death should be considered amenable to medical intervention, despite the observation that the “sensitivity of the method may vary with the selection of causes of death” (Westerling, 1992, p. 492). There is no ‘gold standard’ of causes that are used when analysing this type of population health indicator. This has marked implications and causes problems for comparisons between studies. Only one study has been carried out that investigates the impact of the use of different causes on the results seen (Albert, Bayo, Alfonso, Cortina, & Corella, 1996). We aim to evaluate the impact of definition by comparing the two most common approaches to avoidable mortality (that of Charlton et al., 1983; Holland, 1988) in a detailed geographical analysis of avoidable mortality in Britain.
Most studies on avoidable mortality since 1976 have based their definition of what is avoidable on the list devised by Rutstein et al. (1976) that was divided into those causes that were deemed to be treatable and those that were considered preventable. The diseases chosen from this list have varied mainly between two different approaches devised by Charlton et al. (1983) and Holland (1988). In their analysis of avoidable mortality in England and Wales, Charlton et al. (1983) excluded conditions whose control depended mainly on primary prevention, that were seen as not under the influence of health services, and they aimed for the list to be “sufficiently common to allow comparisons to be made between relatively small populations” (Gil & Rathwell, 1989, p. 652). Holland (1988) followed the Rutstein method of division in his choice of causes by dividing them into 16 medical care indicators and 13 health policy indicators. The medical care indicators were a selection of causes from the Rutstein list and the three causes taken as health policy indicators were lung cancer, cirrhosis of the liver and motor vehicle accidents (Westerling & Smedby, 1992). These causes of death were seen as measures of the efficacy of public health warnings about the dangers of drinking large quantities of alcohol, of smoking and of reckless driving. We compare Charlton et al.'s (1983) and Holland's (1988) definitions firstly because they constitute the main approaches to avoidable mortality, and secondly to enable comparison with the findings of Albert et al. (1996). In practice deaths were classified avoidable on the basis of the Charlton et al. (1983) and Holland (1988) lists of ICD codes and the ages at which each cause was considered avoidable by each definition. For a list of these causes, their ICD codes and the ages at which each cause is considered avoidable see Table 1.
The analysis is undertaken using a multilevel modelling approach that “despite obvious applications in health services research” (Rice & Leyland, 1996, p155) has not been used before in the field of avoidable mortality. Multilevel modelling allows the evaluation of the “nature of heterogeneity at different levels of analysis” (Gould, Jones, & Moon, 1997) and thereby overcomes the atomistic and ecological fallacies that beleaguer individual and aggregate studies respectively (Alker, 1969). There has been considerable debate over the importance of contextual effects, or those related to the characteristics of an area, and compositional effects, or those related to the characteristics of the individuals who reside in an area, to health in areas (Ecob & Jones, 1998; Sloggett & Joshi, 1994). However, multilevel analysis with complex heterogeneity allows analysis of differential geographies for different groups of people and at different spatial scales, and thereby does not attempt to separate contextual and compositional factors. In this analysis we will derive quantitative measures of the variations in both definitions of avoidable mortality at three scales (local authority district, electoral ward and individual) in Great Britain between 1981 and 1998. Adopting an explicit modelling approach allows us not only to quantify the size of the variations, but also to estimate confidence intervals and significance levels.
Section snippets
Data and methods
Data from the Office for National Statistics (ONS) on all deaths in England and Wales that occurred between 1981 and 1998, and the equivalent from the Registrar General for Scotland were used in this analysis (we acknowledge the help of Danny Dorling in accessing these data). These data contain information on each individual who died during the period 1981–1998 (approximately 11.8 million people) including the year of death, their age at death, gender, cause of death (recorded as an
Results
Detailed model estimates are available from the first author. Here we present the key results as a set of diagrams and maps.
The main difference between these two definitions of avoidable mortality is the proportion of total deaths that each determined avoidable. Holland's definition is wider than Charlton's and therefore includes a greater proportion of deaths. Of all 2,102,681 premature (i.e. aged 5–64) deaths in Britain between 1981 and 1998, Charlton's definition found 62,976 deaths
Discussion
The odds of both types of avoidable mortality have, on average, decreased between 1981 and 1998 in Great Britain in line with the trends in premature mortality noted in previous research (Doll, 1983; Dorling, 1997). The only other studies to have examined the temporal trends in avoidable mortality in Britain, albeit using different definitions, during the 1980s and 1990s also found a decline in avoidable mortality (Andreev et al., 2003; Nolte & McKee, 2003). In addition, Nolte and McKee (2003)
Acknowledgements
This research was funded by ESRC/NERC Grant R00429934147.
References (49)
- et al.
Variations in avoidable mortality and variations in health care resources
Lancet
(1987) - et al.
Geographical variation in mortality from conditions amenable to medical intervention in England and Wales
Lancet
(1983) - et al.
Mortality variations in England and Wales between types of place: An analysis of the ONS longitudinal study
Social Science & Medicine
(1998) - et al.
Analyzing perceived limiting long-term illness using UK census microdata
Social Science & Medicine
(1996) - et al.
Health and social inequities in Belgium
Social Science and Medicine
(1990) - et al.
Regional variations in mortality rates in England and Wales: An analysis using multi-level modelling
Social Science and Medicine
(1996) - et al.
The effectiveness of health systems in influencing avoidable mortality: A study in Valencia, Spain, 1975–1990
Journal of Epidemiology and Community Health
(1996) A typology of ecological fallacies
- et al.
The evolving pattern of avoidable mortality in Russia
International Journal of Epidemiology
(2003) - et al.
Area variation in mortality from diseases amenable to medical intervention: The contribution of differences in morbidity
International Journal of Epidemiology
(1986)