Strong association was also detected when the alcoholics with known (and reported) alcoholic liver disease, cirrhosis, or pancreatitis were excluded (P = 7×10−46 and OR = 0.23 (0.19, 0.28) under the random effects model). Furthermore, the association was still strong after all the three studies containing alcohol abuse were excluded (P = 2×10−43) or when the Mexican American subjects were excluded (P = 1×10−45); or both (P = 5×10−43). Additionally, when the study of heroin dependence and abuse was included, the between-studies heterogeneity was more significant but the association was still strong (P = 6×10−43).
