Studies of effects of health conditions on self-rated health must consider the ways standard measures of health disparities tend to be affected by the prevalence of an outcome
James Scanlan, James P. Scanlan, Attorney at Law
14 February 2012
Like virtually all other efforts to examine variations in the ways that persons of similar objective health status in different socioeconomic groups perceive their health, commonly termed reporting heterogeneity, the study by Delpierre et al.[l] suffers from a failure to recognize the patterns by which, for reasons inherent in the shapes of normal risk distributions, standard measures of differences between outcome rates (proportions) tend to be affected by the overall prevalence of an outcome. The most notable such pattern is that whereby 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.[2-4] The failure to recognize this pattern is responsible for the perception, cited by Delpierre et al., that socioeconomic differences in mortality are greater than socioeconomic differences in self-rated health (SRH) and hence that measuring health disparities in terms of SRH may underestimate the magnitude of such disparities. As it happens, SRH data are particularly useful for illustrating the above-referenced statistical pattern, because such data show the contrasting effects of the manner in which a variable is dichotomized on the relative differences in adverse and favorable health outcomes. That is, in terms of the five categories of self-rated health on which the Delpierre study is based ((1) very good, (2) good, (3) fair, (4) poor, or (5) very poor), typically the fewer categories grouped together as the adverse health outcome, the larger will be the relative difference in rates of experiencing that outcome and the smaller will be the relative difference in rates of experiencing the opposite, favorable outcome. The dichotomization choice also affects the way in which SRH differences might be compared with other differences in health outcomes. Table A of reference 5, for example, shows that when all categories of SRH less than very good are grouped as adverse the SRH outcome, the relative difference in mortality is greater than the relative difference in adverse SRH (though the relative difference in survival is smaller than the relative difference in favorable SRH). But the relative difference in the lowest category of SRH is greater than the relative difference in mortality (though the relative difference avoiding that SRH category is smaller than the relative difference in survival). Thus, it is a mistake to attach significance to a perception that SRH disparities, as commonly measured, are smaller (or larger) than other health disparities. More pertinent to the precise focus of the study by Delpierre et al., a corollary to the pattern whereby 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 is a pattern whereby a factor that increases an adverse outcome rate tends to increase it proportionately more in groups with lower baseline rates while reducing the opposite outcome proportionately more in other groups.[3-7] Thus, there is reason to expect that, solely as a result of the distributionally-driven forces described above, factors that increase rates of adverse health, like the functional limitations and chronic low back pain studied by Delpierre et al., will tend to increase rates of adverse SRH proportionately more in groups with the lower baseline rates of adverse health (higher socioeconomic groups) while reducing the favorable outcome rates proportionately more in other groups. Tables 1 and 2 to this comment[8] show that, comparing the highest and lowest socioeconomic categories in Delpierre Table 2 and 3, such are in fact the consistent patterns of relative effects of functional limitations and chronic low back pain on the rates of reporting adverse SRH or favorable SRH. The final columns of Tables 1 and 2 to this comment also show that, using the procedure discussed in references 6,7, and 9 (a measure theoretically unaffected by the prevalence of an outcome), to the extent that such things can be measured, the effect of the functional limitations appear just as likely to be larger in the lowest SES group as in the highest SES group. Three other matters warrant mention. First, a premise of the Delpierre study is that absent SES-related differences in expectations (or SES-related differences in ability to precisely characterize one’s health) a factor that increases perceptions of adverse health will result in equal proportionate changes in adverse SRH rates for higher and lower SES group. In clinical epidemiology there exists a similar premise that, absent a meaningful subgroup effect, a factor that increases or decreases an adverse outcome rate will cause equal proportionate changes in different baseline rates. But, even apart from the distributional considerations discussed above, in either context such premise must be deemed illogical for the simple reason that it is mathematically impossible for a factor to cause equal proportionate changes in different adverse outcome baseline rates while at the same time causing equal proportionate changes in the opposite, favorable outcome rates.[10] Second, because the authors adjusted for other factors, they used odds ratios as the relative measure. Unlike relative differences, differences measured by odds ratios are the same whether one examines the favorable or the adverse outcome. But differences measured by odds ratios tend also to be affected by the overall prevalence of an outcome, though in a more complicated way than relative differences.[2,11] Hence, like relative differences, odds ratios cannot effectively identify differences in the strength of an effect across different baseline rates. Third, Delpierre also sought to compare the effects of the functional limitations and chronic low back pain on the SF-36 score, which seems to be a continuous variable. To the extent that a variable is in fact continuous, as, for example, where individuals rate their health on a scale of 1 to 10, or 1 to 100, the above-described patterns by which binary measures of differences tend to be affected by the overall prevalence of an outcome would not necessarily undermine an analysis based on such measure. Indeed, the approach to quantifying differences reflected by differing outcome rates described in references 6, 7, and 9 is based on deriving from outcome rates differences in a continuous variable. But the SF-36 score is a composite based on responses by categories. Such fact raises issues similar to issues raised by standard binary measures. Thus, without analyzing the SF-36 with regard to the way changes in the general prevalence of an outcome affects the score, it is not possible to know how useful the measure is for comparing the effects of a factor on groups with different baseline scores. References: 1. Delpierre C, Kelly-Irving M, Munch-Petersen M., et al. SRH and HRQ: does social position impact differently on their with health status. BMC Public Health 2012, 12,19 (doi:10.1186/1471-2458-12-19): http://www.biomedcentral.com/1471-2458/12/19
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. Scanlan JP. Divining difference. Chance 1994;7(4):38-9,48: http://jpscanlan.com/images/Divining_Difference.pdf 5. Reporting Heterogeneity sub-page of Scanlan’s Rule page of jpscanlan.com: http://jpscanlan.com/measuringhealthdisp/reportingheterogeneity.html 6. Subgroup Effects sub-page of Scanlan’s Rule page of jpscanlan.com:http://www.jpscanlan.com/scanlansrule/subgroupeffects.html 7. Interpreting Differential Effects in Light of Fundamental Statistical Tendencies, presented at 2009 Joint Statistical Meetings of the American Statistical Association, International Biometric Society, Institute for Mathematical Statistics, and Canadian Statistical Society, Washington, DC, Aug. 1-6, 2009: http://www.jpscanlan.com/images/Scanlan_JSM_2009.ppt 8. http://jpscanlan.com/images/Tables_to_Comment_on_Delpierre.pdf 9. Solutions sub-page of Measuring Health Disparities page of jpscanlan.com: http://www.jpscanlan.com/measuringhealthdisp/solutions.html 10. Illogical Premises sub-page of the Scanlan’s Rule page of jpscanlan.com: http://jpscanlan.com/scanlansrule/illogicalpremises.html 11. Scanlan’s Rule page of jpscanlan.com: http://jpscanlan.com/scanlansrule.html
Studies of effects of health conditions on self-rated health must consider the ways standard measures of health disparities tend to be affected by the prevalence of an outcome
14 February 2012
Like virtually all other efforts to examine variations in the ways that persons of similar objective health status in different socioeconomic groups perceive their health, commonly termed reporting heterogeneity, the study by Delpierre et al.[l] suffers from a failure to recognize the patterns by which, for reasons inherent in the shapes of normal risk distributions, standard measures of differences between outcome rates (proportions) tend to be affected by the overall prevalence of an outcome. The most notable such pattern is that whereby 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.[2-4]
The failure to recognize this pattern is responsible for the perception, cited by Delpierre et al., that socioeconomic differences in mortality are greater than socioeconomic differences in self-rated health (SRH) and hence that measuring health disparities in terms of SRH may underestimate the magnitude of such disparities. As it happens, SRH data are particularly useful for illustrating the above-referenced statistical pattern, because such data show the contrasting effects of the manner in which a variable is dichotomized on the relative differences in adverse and favorable health outcomes. That is, in terms of the five categories of self-rated health on which the Delpierre study is based ((1) very good, (2) good, (3) fair, (4) poor, or (5) very poor), typically the fewer categories grouped together as the adverse health outcome, the larger will be the relative difference in rates of experiencing that outcome and the smaller will be the relative difference in rates of experiencing the opposite, favorable outcome.
The dichotomization choice also affects the way in which SRH differences might be compared with other differences in health outcomes. Table A of reference 5, for example, shows that when all categories of SRH less than very good are grouped as adverse the SRH outcome, the relative difference in mortality is greater than the relative difference in adverse SRH (though the relative difference in survival is smaller than the relative difference in favorable SRH). But the relative difference in the lowest category of SRH is greater than the relative difference in mortality (though the relative difference avoiding that SRH category is smaller than the relative difference in survival). Thus, it is a mistake to attach significance to a perception that SRH disparities, as commonly measured, are smaller (or larger) than other health disparities.
More pertinent to the precise focus of the study by Delpierre et al., a corollary to the pattern whereby 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 is a pattern whereby a factor that increases an adverse outcome rate tends to increase it proportionately more in groups with lower baseline rates while reducing the opposite outcome proportionately more in other groups.[3-7] Thus, there is reason to expect that, solely as a result of the distributionally-driven forces described above, factors that increase rates of adverse health, like the functional limitations and chronic low back pain studied by Delpierre et al., will tend to increase rates of adverse SRH proportionately more in groups with the lower baseline rates of adverse health (higher socioeconomic groups) while reducing the favorable outcome rates proportionately more in other groups.
Tables 1 and 2 to this comment[8] show that, comparing the highest and lowest socioeconomic categories in Delpierre Table 2 and 3, such are in fact the consistent patterns of relative effects of functional limitations and chronic low back pain on the rates of reporting adverse SRH or favorable SRH. The final columns of Tables 1 and 2 to this comment also show that, using the procedure discussed in references 6,7, and 9 (a measure theoretically unaffected by the prevalence of an outcome), to the extent that such things can be measured, the effect of the functional limitations appear just as likely to be larger in the lowest SES group as in the highest SES group.
Three other matters warrant mention. First, a premise of the Delpierre study is that absent SES-related differences in expectations (or SES-related differences in ability to precisely characterize one’s health) a factor that increases perceptions of adverse health will result in equal proportionate changes in adverse SRH rates for higher and lower SES group. In clinical epidemiology there exists a similar premise that, absent a meaningful subgroup effect, a factor that increases or decreases an adverse outcome rate will cause equal proportionate changes in different baseline rates. But, even apart from the distributional considerations discussed above, in either context such premise must be deemed illogical for the simple reason that it is mathematically impossible for a factor to cause equal proportionate changes in different adverse outcome baseline rates while at the same time causing equal proportionate changes in the opposite, favorable outcome rates.[10]
Second, because the authors adjusted for other factors, they used odds ratios as the relative measure. Unlike relative differences, differences measured by odds ratios are the same whether one examines the favorable or the adverse outcome. But differences measured by odds ratios tend also to be affected by the overall prevalence of an outcome, though in a more complicated way than relative differences.[2,11] Hence, like relative differences, odds ratios cannot effectively identify differences in the strength of an effect across different baseline rates.
Third, Delpierre also sought to compare the effects of the functional limitations and chronic low back pain on the SF-36 score, which seems to be a continuous variable. To the extent that a variable is in fact continuous, as, for example, where individuals rate their health on a scale of 1 to 10, or 1 to 100, the above-described patterns by which binary measures of differences tend to be affected by the overall prevalence of an outcome would not necessarily undermine an analysis based on such measure. Indeed, the approach to quantifying differences reflected by differing outcome rates described in references 6, 7, and 9 is based on deriving from outcome rates differences in a continuous variable. But the SF-36 score is a composite based on responses by categories. Such fact raises issues similar to issues raised by standard binary measures. Thus, without analyzing the SF-36 with regard to the way changes in the general prevalence of an outcome affects the score, it is not possible to know how useful the measure is for comparing the effects of a factor on groups with different baseline scores.
References:
1. Delpierre C, Kelly-Irving M, Munch-Petersen M., et al. SRH and HRQ: does social position impact differently on their with health status. BMC Public Health 2012, 12,19 (doi:10.1186/1471-2458-12-19):
http://www.biomedcentral.com/1471-2458/12/19
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. Scanlan JP. Divining difference. Chance 1994;7(4):38-9,48: http://jpscanlan.com/images/Divining_Difference.pdf
5. Reporting Heterogeneity sub-page of Scanlan’s Rule page of jpscanlan.com: http://jpscanlan.com/measuringhealthdisp/reportingheterogeneity.html
6. Subgroup Effects sub-page of Scanlan’s Rule page of jpscanlan.com:http://www.jpscanlan.com/scanlansrule/subgroupeffects.html
7. Interpreting Differential Effects in Light of Fundamental Statistical Tendencies, presented at 2009 Joint Statistical Meetings of the American Statistical Association, International Biometric Society, Institute for Mathematical Statistics, and Canadian Statistical Society, Washington, DC, Aug. 1-6, 2009: http://www.jpscanlan.com/images/Scanlan_JSM_2009.ppt
8. http://jpscanlan.com/images/Tables_to_Comment_on_Delpierre.pdf
9. Solutions sub-page of Measuring Health Disparities page of jpscanlan.com: http://www.jpscanlan.com/measuringhealthdisp/solutions.html
10. Illogical Premises sub-page of the Scanlan’s Rule page of jpscanlan.com: http://jpscanlan.com/scanlansrule/illogicalpremises.html
11. Scanlan’s Rule page of jpscanlan.com: http://jpscanlan.com/scanlansrule.html
Competing interests
na