Efforts to appraise changes in inequalities in poor health over the life course must consider the implications of general increases in poor health as the population ages
James Scanlan, James P. Scanlan, Attorney at Law, Washington, DC
12 January 2012
Benzeval et al.[1] have endeavored to address some complex issues concerning whether socioeconomic inequalities in health increase or decrease with age. But, as with other research on the topic, the effort suffers from a failure to recognize the way that, for reasons related to the shapes of normal distributions of factors associated with experiencing an outcome, standard measures of differences between rates tend to be affected by the overall prevalence of the outcome.
Most notably, the rarer an outcome the greater tends to be the relative difference in experiencing it and the smaller tends to be the relative difference in avoiding it.[2-6] Absolute differences between rates tend also to be systematically affected by the overall prevalence of an outcome, though in a more complicated way. With normal distributions of risk, as uncommon outcomes (less than 50% for both groups) increase in overall prevalence absolute differences between rates tend to increase; as common outcomes (greater than 50% for both groups) further increase in overall prevalence absolute differences tend to decrease. When the outcome is neither uncommon nor common the patterns are influenced by the size of the difference between the means of the distributions.[5]
The distributional forces are only part of the picture, of course. Observed patterns are functions both of these forces and the size of differences between the distributions in the settings being compared. The latter is what should be the subject of any inquiry. But it is only possible to draw useful conclusions about such differences by taking the distributional forces into account or by employing a measure of difference that is not affected by the overall prevalence of an outcome (such as that discussed in reference 7).
A great deal of the prior work in this area has focused on relative differences in mortality and has typically concluded that health inequalities are greater among the young than the old because relative differences in mortality are greater among the young. But such work failed to consider the distributionally-based reasons to expect larger relative differences among the young or the fact that relative differences in survival tended to be greater among the old. As it happens, mortality comparisons across age ranges provide particularly useful illustrations of the distributionally-driven patterns because the prevalence of mortality differs so much by age range (as discussed and illustrated in reference 6). Prior work on inequalities in mortality that has discussed absolute differences in mortality, while noting larger absolute differences among the old than the young, has similarly failed to consider the extent to which observed patterns are influenced by differences in overall mortality at different ages.
Benzeval et al. analyzed absolute differences between the rates at which manual and non-manual workers self-rate their health (SRH) as fair/poor (rather than good/excellent), which SRH is termed ¿poor¿ in the illustrative figures. Poor self-rated health, like mortality, tends to increase with age. Thus, conclusions about the comparative size of differences in SRH that are based on relative differences, like conclusions about differences in mortality, commonly will turn on whether one analyzes relative differences in the adverse outcome (fair/poor health) or the favorable outcome (good/excellent health). (See reference 8 regarding the way whether one focuses on the favorable of the adverse outcome can affect the SRH analyses in Benzeval¿s references 36, 38, 39, and 41.). From Benzeval¿s figures 1a, 2a, 3a, and 4a one could roughly divine the extent to which the patterns of relative differences in the favorable and adverse SRH rates conform to the distributionally-drive patterns described above. But, particularly given that the patterns are only part of the picture and not the focus of the authors, it is unlikely to be worthwhile to do that here.
The authors¿ reliance on absolute difference raises a different issue. The absolute differences between rates based on dichotomized self-rated health (Figures 1a, 1c, 2a, 2c) or dichotomized self-rated health including mortality (Figures 3a, 3c, 4a, 4c) seem generally to conform to the distributionally-driven patterns. That is, the differences tend to increase as the adverse outcome increases in overall prevalence until the point where the adverse outcome predominates; then, at roughly the point where the distributional forces would tend to cause absolute differences to decline, the absolute difference do decline. It should be recognized, however, that if one dichotomized the matter differently ¿ i.e., by including in the adverse SRH category only poor health or including in the favorable SRH category only excellent health ¿ one would reach the prevalence level at which absolute differences tend to decline at different ages from those reflected in the Benzeval tables.
But, however one dichotomizes SRH, a sounder analysis would be based on the approach in reference 7, though it would still be subject to the cohort issues and other complexities that can render uncertain even conclusions based on a sound measure of difference between outcome rates.
Finally, I note that authors also translated SRH into a continuous measure, which is also what the approach in reference 7 seeks to do. Were health rated on a continuous scale of 1 to 10 or 1 to 100, such approach might be entirely sound (though I am inclined to think that the difference between the means so derived should framed in terms of percentages of a standard deviation, if such can be divined, rather than absolute differences [9]). I am unsure, however, that these ratings can be effectively translated to a continuous scale and uncertain of the implications of the treatment of death. It nevertheless would be interesting to compare the results of that approach to the various results derived from the approach in reference 7 with regard to answering the question of whether inequalities increase or decrease with age.
References:
1. Benzeval M, Green MJ, Leyland AH. Do social inequalities in health widen or converge with age. Longitudinal evidence from three cohorts in the West of Scotland. BMC Public Health 2011, 11:947: doi:10.1186/1471-2458-11-947
2. Scanlan JP. Can we actually measure health disparities? Chance 2006:19(2):47-51: http://www.jpscanlan.com/images/Can_We_Actually_Measure_Health_Disparities.pdf
3. Scanlan JP. Race and mortality. Society 2000;37(2):19-35: http://www.jpscanlan.com/images/Race_and_Mortality.pdf
4. The Misinterpretation of Health Inequalities in the United Kingdom, presented at the British Society for Populations Studies Conference 2006, Southampton, England, Sept. 18-20, 2006:
http://www.jpscanlan.com/images/BSPS_2006_Complete_Paper.pdf
5. Scanlan¿s Rule page of jpscanlan.com: http://jpscanlan.com/scanlansrule.html
6. Life Tables Illustrations sub-page of Scanlan¿s Rule page of jpscanlan.com: http://jpscanlan.com/scanlansrule/lifetableillustrations.html
7. Scanlan JP. Measuring Health Inequalities by an Approach Unaffected by the Overall Prevalence of the Outcomes at Issue, presented at the Royal Statistical Society Conference 2009, Edinburgh, Scotland, Sept. 7-11, 2009: http://www.jpscanlan.com/images/Scanlan_RSS_2009_Presentation.ppt
8. Reporting Heterogeneity sub-page of Measuring Health Disparities page of jpscanlan.com: http://jpscanlan.com/measuringhealthdisp/reportingheterogeneity.html
9. Scanlan JP. Article on disparities in control of cardiovascular disease and diabetes raises several measurement issues. Ann Int Med Nov. 30, 2009 (responding to McWilliams JM, Meara E., Zaslavsky AM, Ayanian JZ. Differences in control of cardiovascular disease and diabetes by race, ethnicity, and education, U.S. trends from 1999 to 2006 and effects of Medicare coverage. Ann Int Med 2009;150:505-515): http://www.annals.org/content/150/8/505/reply#annintmed_el_120732
Efforts to appraise changes in inequalities in poor health over the life course must consider the implications of general increases in poor health as the population ages
12 January 2012
Benzeval et al.[1] have endeavored to address some complex issues concerning whether socioeconomic inequalities in health increase or decrease with age. But, as with other research on the topic, the effort suffers from a failure to recognize the way that, for reasons related to the shapes of normal distributions of factors associated with experiencing an outcome, standard measures of differences between rates tend to be affected by the overall prevalence of the outcome.
Most notably, the rarer an outcome the greater tends to be the relative difference in experiencing it and the smaller tends to be the relative difference in avoiding it.[2-6] Absolute differences between rates tend also to be systematically affected by the overall prevalence of an outcome, though in a more complicated way. With normal distributions of risk, as uncommon outcomes (less than 50% for both groups) increase in overall prevalence absolute differences between rates tend to increase; as common outcomes (greater than 50% for both groups) further increase in overall prevalence absolute differences tend to decrease. When the outcome is neither uncommon nor common the patterns are influenced by the size of the difference between the means of the distributions.[5]
The distributional forces are only part of the picture, of course. Observed patterns are functions both of these forces and the size of differences between the distributions in the settings being compared. The latter is what should be the subject of any inquiry. But it is only possible to draw useful conclusions about such differences by taking the distributional forces into account or by employing a measure of difference that is not affected by the overall prevalence of an outcome (such as that discussed in reference 7).
A great deal of the prior work in this area has focused on relative differences in mortality and has typically concluded that health inequalities are greater among the young than the old because relative differences in mortality are greater among the young. But such work failed to consider the distributionally-based reasons to expect larger relative differences among the young or the fact that relative differences in survival tended to be greater among the old. As it happens, mortality comparisons across age ranges provide particularly useful illustrations of the distributionally-driven patterns because the prevalence of mortality differs so much by age range (as discussed and illustrated in reference 6). Prior work on inequalities in mortality that has discussed absolute differences in mortality, while noting larger absolute differences among the old than the young, has similarly failed to consider the extent to which observed patterns are influenced by differences in overall mortality at different ages.
Benzeval et al. analyzed absolute differences between the rates at which manual and non-manual workers self-rate their health (SRH) as fair/poor (rather than good/excellent), which SRH is termed ¿poor¿ in the illustrative figures. Poor self-rated health, like mortality, tends to increase with age. Thus, conclusions about the comparative size of differences in SRH that are based on relative differences, like conclusions about differences in mortality, commonly will turn on whether one analyzes relative differences in the adverse outcome (fair/poor health) or the favorable outcome (good/excellent health). (See reference 8 regarding the way whether one focuses on the favorable of the adverse outcome can affect the SRH analyses in Benzeval¿s references 36, 38, 39, and 41.). From Benzeval¿s figures 1a, 2a, 3a, and 4a one could roughly divine the extent to which the patterns of relative differences in the favorable and adverse SRH rates conform to the distributionally-drive patterns described above. But, particularly given that the patterns are only part of the picture and not the focus of the authors, it is unlikely to be worthwhile to do that here.
The authors¿ reliance on absolute difference raises a different issue. The absolute differences between rates based on dichotomized self-rated health (Figures 1a, 1c, 2a, 2c) or dichotomized self-rated health including mortality (Figures 3a, 3c, 4a, 4c) seem generally to conform to the distributionally-driven patterns. That is, the differences tend to increase as the adverse outcome increases in overall prevalence until the point where the adverse outcome predominates; then, at roughly the point where the distributional forces would tend to cause absolute differences to decline, the absolute difference do decline. It should be recognized, however, that if one dichotomized the matter differently ¿ i.e., by including in the adverse SRH category only poor health or including in the favorable SRH category only excellent health ¿ one would reach the prevalence level at which absolute differences tend to decline at different ages from those reflected in the Benzeval tables.
But, however one dichotomizes SRH, a sounder analysis would be based on the approach in reference 7, though it would still be subject to the cohort issues and other complexities that can render uncertain even conclusions based on a sound measure of difference between outcome rates.
Finally, I note that authors also translated SRH into a continuous measure, which is also what the approach in reference 7 seeks to do. Were health rated on a continuous scale of 1 to 10 or 1 to 100, such approach might be entirely sound (though I am inclined to think that the difference between the means so derived should framed in terms of percentages of a standard deviation, if such can be divined, rather than absolute differences [9]). I am unsure, however, that these ratings can be effectively translated to a continuous scale and uncertain of the implications of the treatment of death. It nevertheless would be interesting to compare the results of that approach to the various results derived from the approach in reference 7 with regard to answering the question of whether inequalities increase or decrease with age.
References:
1. Benzeval M, Green MJ, Leyland AH. Do social inequalities in health widen or converge with age. Longitudinal evidence from three cohorts in the West of Scotland. BMC Public Health 2011, 11:947: doi:10.1186/1471-2458-11-947
2. Scanlan JP. Can we actually measure health disparities? Chance 2006:19(2):47-51: http://www.jpscanlan.com/images/Can_We_Actually_Measure_Health_Disparities.pdf
3. Scanlan JP. Race and mortality. Society 2000;37(2):19-35: http://www.jpscanlan.com/images/Race_and_Mortality.pdf
4. The Misinterpretation of Health Inequalities in the United Kingdom, presented at the British Society for Populations Studies Conference 2006, Southampton, England, Sept. 18-20, 2006:
http://www.jpscanlan.com/images/BSPS_2006_Complete_Paper.pdf
5. Scanlan¿s Rule page of jpscanlan.com: http://jpscanlan.com/scanlansrule.html
6. Life Tables Illustrations sub-page of Scanlan¿s Rule page of jpscanlan.com: http://jpscanlan.com/scanlansrule/lifetableillustrations.html
7. Scanlan JP. Measuring Health Inequalities by an Approach Unaffected by the Overall Prevalence of the Outcomes at Issue, presented at the Royal Statistical Society Conference 2009, Edinburgh, Scotland, Sept. 7-11, 2009: http://www.jpscanlan.com/images/Scanlan_RSS_2009_Presentation.ppt
8. Reporting Heterogeneity sub-page of Measuring Health Disparities page of jpscanlan.com: http://jpscanlan.com/measuringhealthdisp/reportingheterogeneity.html
9. Scanlan JP. Article on disparities in control of cardiovascular disease and diabetes raises several measurement issues. Ann Int Med Nov. 30, 2009 (responding to McWilliams JM, Meara E., Zaslavsky AM, Ayanian JZ. Differences in control of cardiovascular disease and diabetes by race, ethnicity, and education, U.S. trends from 1999 to 2006 and effects of Medicare coverage. Ann Int Med 2009;150:505-515): http://www.annals.org/content/150/8/505/reply#annintmed_el_120732
Competing interests
None