INTRODUCTION: Medical chart-review and self-reported questionnaire are two common methods of determining cancer screening and symptoms. We investigate the validity of these methods and therefore of a class of clinical/epidemiological studies. We compare variables on prostate cancer, any prostate-specific antigen (PSA) test, asymptomatic screening PSA, any digital rectal exam (DRE), and urinary symptoms. We used data from a 2005 case control study of PSA and metastatic prostate cancer (253 cases and 496 controls). Data were collected from 1999 to 2002.
METHODS: We calculated kappa, percent agreement (PPA) and prevalence adjusted bias adjusted kappa (PABAK). We compared percentage positive response (PPR) and sensitivities/specificities of questionnaire against chart and vice versa. We measured the degree of differential agreement between cases and controls using odds ratios.
RESULTS: We found almost perfect agreement on prostate cancer, moderate agreement on any PSA and DRE, and slight agreement on asymptomatic screening PSA and urinary symptoms. PABAK ranged from 0.134 (urinary symptoms) to 0.879 (prostate cancer). Differences between cases/controls in PPR are similar according to chart or questionnaire, though PPR itself is usually higher on the questionnaire. Only for any PSA (including diagnostic), cases had better recall than controls. We found no evidence of differential agreement that might lead to bias in a case control study.
CONCLUSIONS: Some variables are more reliable than others comparing medical chart review and self-report. Diagnosis of prostate cancer has near perfect agreement, but for less catastrophic events such as PSA (especially asymptomatic screening tests), DRE or urinary symptoms, agreement ranges from slight to moderate.