James P. Scanlan Attorney at Law Washington, DC, USA jps@jpscanlan

British Society for Population Studies, 2008 Annual Conference University of Manchester10-12 September 2008An Approach to Measuring Differences between Rates that is not Affected by the Overall Prevalence of an Outcome James P. Scanlan Attorney at Law Washington, DC, USA jps@jpscanlan.com

Subjects 1. The problem with standard binary measures of differences between rates (relative differences, absolute differences, odds ratios): that all exhibit patterns of correlation with overall prevalence (i.e., among other things, they tend to change as overall prevalence changes) (BSPS 2006, 2007) 2. A plausible alternative approach that avoids the problem with standard measures: a measure that does not change as overall prevalence changes

References • Measuring Health Disparitiespage (especially the Solutionstab) and Scanlan’s Rulepage on jpscanlan.com • Can We Actually Measure Health Disparities? (Chance 2006) (A12) • Race and Mortality (Society 2000) (A10) • The Misinterpretation of Health Inequalities in the United Kingdom (BSPS 2006) (B6) • BSPS 2007 (B11)

Patterns by Which Binary Measures Tend to Change as the Overall Prevalence of an Outcome Changes – Scanlan’s Rules • SR1: The rarer an outcome, the greater tends to be the relative difference in rates of experiencing it and the smaller tends to be the relative difference in rates of avoiding it (IR1 in BSPS 2006; see Bauld, Day, Judge, 2008) • SR2: As an outcome changes in overall prevalence, the odds ratio tends to change in the same direction of the larger of the two relative differences and the absolute difference tends to change in the same direction as the smaller of the two relative differences ─ where numerators are reversed on the two risk ratios (see Semantic Issues tab on Scanlan’s Rule page)

Fig 1. Ratio of (1) Advantaged Group (AG) Success Rate to Disadvantaged Group (DG) Success Rate at Various Cutoffs Defined by AG Success Rate

Fig 2. Ratios of (1) AG Success Rate to DG Success Rate and (2) DG Fail Rate to AG Fail Rate

Fig 3. Ratios of (1) AG Success Rate to DG Success Rate, (2) DG Fail Rate to AG Fail Rate, and (3) DG Fail Odds to AG Fails Odds

Fig 4. Ratios of (1) AG Success Rate to DG Success Rate, (2) DG Fail Rate to AG Fail Rate, and (3) DG Fail Odds to AG Fails Odds; and Absolute Diff Between Rates

Table 1 Illustration of the Problem and Intimation of the Solution (in terms of a favorable outcome increasing in overall prevalence) Period Yr 1 Yr 2 Yr 3 Yr 4 AG Rate 40% 58% 76% 94% DG Rate 23% 39% 58% 85% Measures of Difference (Blue=decrease; Red=increase) Ratio 1 1.74 1.45 1.31 1.11 Ratio 2 1.28 1.49 1.75 2.50 Odds Ratio 2.23 2.16 2.292.76 Absol Diff .17 .19.18.09 EES (z) .50 .50 .50 .50

Estimated Effect Size (EES) Difference between means of hypothesized underlying distributions of risks of experiencing an outcome, in terms of percentage of a standard deviation, assuming normality of the distributions

Table 2 Simplified Illustration of the Solution (in terms of a favorable outcome increasing in overall prevalence) Period Yr 1 Yr 2 AG Rate 40% 58% DG Rate 23% 40% Measures of Difference Change Direction Ratio 1 1.74 1.43 Increase Ratio 2 1.28 1.45 Decrease Odds Ratio 2.23 2.07 Decrease Absol Diff .17 .18 Increase EES (z).50.47 Decrease

Table 3 Illustration of Meaning of Various Ratios at Different Prevalence Levels

Caveat – Compositional Differences • Differences in the proportions groups being analyzed comprise of the total population in settings being compared undermine otherwise valid measurement approaches • See Heller et al. (BMJ 2002), Boström and Rosén (SJPH 2003), D43 • Solution seems always to compare same proportions (e.g., top vs. bottom quintile), which may often limit analyses to income • Less of an issue for US racial comparisons

Table 4 Illustration Based on Morita et. al. (Pediatrics 2008) Data on Black and White Hepatitis B Vaccination Rates Pre and Post School-Entry Vaccination Requirement (see D52)

Table 5 Illustration of UK Changes Over Time from Table 4.13 of The Widening Gap (rates per 100,000)

Table 6 Illustration of UK Differences across Age Groups from Table 4.13 of The Widening Gap

Table 7 Illustration of Comparisons as to Different Conditions from Lawlor (AJPH 2006)(Aberdeen 1950 birth cohort) (rates are per 10,000) (see D28)

Table 8 Illustration of Age Group Comparisons in Whitehall Studies from Marang-van de Mheen (JECH 2001) (rates are per 1,000)

Table 9 Illustration Based on Boström and Rosén (SJPH 2003) Data on Mortality by Occupation in Seven European Countries (see D43 caveat)

Problems with the Solution • Always practical issues (we do not really know the shape of the underlying distributions) • Sometimes fundamental issues (e.g., where we know distributions are not normal because they are truncated portions of larger distributions, see D43); cf. BSPS 2007, Fig. 6 • Absolute minimum issue (D43,B6)

Conclusion • If we are mindful of the problems, the approach provides a framework for cautiously appraising the size of differences between rates. • Regardless of problems, the approach is superior to reliance on standard binary measures of differences between rates without regard to the way those measures tend to be correlated with the overall prevalence of an outcome.

Formula to derive EES?

Example of 8 of 76,960 rows in table downloadable from jpscanlan.com

Supp Table 1 Illustrations Based on Escarce and McGuire (AJPH 2004) Data on White and Black Coronary Procedure Rates, 1986 and 1997 (see D48)

Supp Table 2: Illustration Based on Hetemaa et al. (JECH 2003) Data on Finnish Revascularization Rates, 1988 and 1996, by Income Group (see D21, D58)

Supp Table 3: Illustration Based on Laaksonen et al. (JECH 2008) Data on Mortality Rates of Finnish Men by Owner or Renter Status (see follow-up on D43)

Supp Table 4 Illustration from Valkonen et al. (JEPH 2000) Based on All Cause Mortality in Finland for Three Time Periods f

James P. Scanlan Attorney at Law Washington, DC, USA jps@jpscanlan