Methods
This study was approved by the Public Health Institute's Institutional Review Board as an exempt study of non-identifiable secondary data (IRB #I04–015).
Data
Yearly liver cirrhosis mortality rates were taken from the Vital Statistics of the USA (National Office of Vital Statistics, 1954–1960; National Center for Health Statistics, 1961–1969) from 1950 to 1967 and from the National Center for Health Statistics Compressed Mortality File series (National Center for Health Statistics, 2000, 2003, 2004) for the years 1968–2002. These sources report deaths according to underlying cause, rather than using multiple cause coding, so that each death is attributed to only one cause. The rates were age-standardized to the 2000 US population using the number of deaths in each specific age group along with population estimates for that group. State- and age-specific population estimates came from the same sources as the mortality data from 1968 to 2002. The 1968 population estimates were used with data from the 1950 and 1960 US Decennial Census to calculate population estimates for all remaining years, with interpolations assuming a constant rate of increase (or decrease).
Natural logs of yearly age-standardized cirrhosis mortality rates were calculated for each state. Liver cirrhosis was defined based on several revisions of the International Classification of Diseases (ICD) using code 581 for ICD-6 and ICD-7, code 571 for ICD-8 and ICD-9 and codes K70, K73 and K74 for ICD-10. The data include White and non-White male and female mortality from 1950 to 1967, because more specific race/ethnicity categories were not available; the age-standardized mortality rates for White and African-American men and women are used from 1968 to 2002. The limited classification of race/ethnicity in the first 17 years of the series should not pose a serious problem for these analyses, as the states included in this study had total populations that were overwhelmingly classified as either White or Black by the US Decennial Census (U.S. Census Bureau, 2011), as shown in Table 1. For example, in 1950, from 99.2 to 100% of the total population was included in these two groups; in 1970, the range was from 98.7 to 99.8%.
Data for per capita apparent consumption of ethanol (in liters) from beer, wine, spirits and combined for total alcohol are derived from volume sales by beverage type using state- and year-specific estimates of the mean alcohol content. Alcohol sales figures came from US beer, wine and spirits industry statistics for the years 1950–1969 and from the Alcohol Epidemiologic Data System (Nephew et al., 2004) for 1970–2002. Estimates of alcohol concentration for each beverage type and year were developed from a variety of industry and government alcohol monopoly sources (Kerr et al., 2004; Kerr et al., 2006a, b). We used a 0.7 geometric distributed lag truncated after 5 years for the alcohol variables, because previous consumption is theoretically relevant to cirrhosis risk (Ye and Kerr, 2011) and because the lag relationship fit the data better than current year data. As several years of data are lost in the creation of distributed lags, models start in 1955.
Analysis Strategy
States with large numbers and percentages of African-American residents were the focus of the analyses; we included those states where 90–100% of the population was either African American or White. As noted above, figures from the US Decennial Census were used to determine the number of African-American residents and the proportion of non-White residents in each state. States with large Hispanic and Native American non-White populations (i.e. those where >10% of the total population was of a race/ethnicity other than White or African American), particularly in the west, were not included. Graphical analyses of White and African-American cirrhosis mortality trends for males identified two groups of states corresponding to the regions identified by Herd (1985). The 13 states with a large differential between race groups were Connecticut, Delaware, District of Columbia, Illinois, Maryland, Michigan, Missouri, New Jersey, New York, North Carolina, Ohio, Pennsylvania and West Virginia. The 11 states where such a difference is not seen for men were mainly southern states (Alabama, Arkansas, Georgia, Kentucky, Louisiana, Mississippi, South Carolina, Tennessee, Texas and Virginia) and Massachusetts. States with a large cirrhosis mortality rate difference for men also showed a similar difference for women. In the group of states with a small men's mortality rate difference, some states showed a similar pattern for women, but others had a large differential in women's mortality rates, with substantially higher cirrhosis mortality for African-American women compared with White women.
Panel models were estimated using generalized least squares (GLS). GLS is a generalized method specifying the variance-covariance matrix of the error structure, allowing the modeling of differences in variances across panels (heteroscedasticity) as well as panel-specific, first-order AR error terms. Panel models were fit using STATA version 10 (Stata Corp, 2007). The first difference of both the log mortality rate and alcohol consumption series was used to remove unit roots present in the original series, which is a conservative approach to address state-specific time trends. The panel models estimated the effects of total alcohol consumption volume on cirrhosis mortality rates and then separate models estimated the effects of beer, wine and spirits consumption volumes (Kerr and Ye, 2011; Ye and Kerr, 2011). Each model was estimated separately for each sex- and race-defined group, first using data from all 24 states in the subsample of interest, and then for the two subgroups of states with large and small African-American/White mortality rate differentials. A final set of models for African Americans adjust for the White cirrhosis mortality rate. These models examine the contribution of alcohol to the mortality differential by documenting alcohol's association with mortality rate changes for African Americans beyond those experienced by Whites. All panel models controlled for the years corresponding to an ICD version change. Coefficients represent the percentage change in cirrhosis mortality in response to a 1 l change in ethanol consumption.
The geographic patterning of mortality rate changes over time among African-American males gives an indication of the spread of detrimental drinking patterns across states. Granger causality tests (Granger, 1989) used lagged values of cirrhosis mortality rates from one state to predict the current value of cirrhosis mortality rates in another state. All pair-wise comparisons of states were tested. These tests incorporated two-lagged values and a 0.05 significance level was used for rejecting the null hypothesis of no relationship. Granger causality is determined when a significant relationship is found in one direction but not in the other, in this case, indicating a mortality rise (or decline) in one state was followed by a rise (or decline) in another.