For each subject, we determined the lifetime presence of the 12 combined DSM-IV and DSM-5 criteria for each of four substances: alcohol, cocaine, opioids, and cannabis. We applied the DSM-IV and DSM-5 algorithms to classify subjects in terms of DSM-IV abuse (without dependence) and dependence diagnoses and DSM-5 mild, moderate or severe SUD. For each substance, our primary comparisons were between binary classifications of 1) neither abuse nor dependence versus either abuse or dependence under DSM-IV, and 2) no SUD versus any SUD (mild, moderate, or severe) under DSM-5. We used the kappa statistic (Agresti, 2002) to measure chance-corrected levels of agreement between the two classifications and conditional odds ratios (Agresti, 2002) to indicate the direction and magnitude of discordance among the participants whose diagnosis changed between the two systems. We also compared the three-level DSM-IV diagnosis to three- and four-level DSM-5 diagnoses.
