Reliability of smoking history in medical charts in relation to interview data

We previously reported on a series of studies exploring relationships between cigarette smoking and mutations of tumor suppressor genes; the studies relying on medical charts for data on smoking. To assess the accuracy of these data we compared them with data obtained using a conventional epidemiologic interview. As part of an ongoing case-control study of tobacco-related cancers, a trained interviewer questioned 144 patients about their smoking habits using a structured interview instrument; the medical charts were then reviewed. Using the Spearman test to compare the data obtained from the two sources, the correlation coefficient was 0.85 for cigarettes per day; 0.88 for years of smoking and 0.95 for pack-years, suggesting that the data in the charts were reliable. The kappa coefficient was 0.96 for current smoking status, indicating a very strong agreement between the two sources of data. The sensitivity and specificity for chart review were 96.3% and 100% respectively, compared with the 'gold standard' of a structured interview. These results show that when patients cannot be interviewed, for example when performing retrospective studies using existing collections of archived tumors or other biologic materials, then data on smoking abstracted from medical charts may be a reliable substitute.