Differences in morbidity measures and risk factor identification using multiple data sources: the case of stroke

Epidemiologic studies utilise medical information from a variety of sources. These include subject or proxy interviews, medical records, death certificates and administrative records. Since the choice of data source may affect the validity of study results, it is important to understand the effect of different case-ascertainment methodologies on estimates of risk.

The NHANES I Epidemiologic Follow-up Study (NHEFS) contains several sources of information that can be used to define case status. In this report we investigate whether the use of seven different algorithms for case ascertainment, each based on different combinations of data sources, results in substantive differences in the estimates of incidence rates and relative risks associated with selected, documented, risk factors for stroke.

The seven different models of case identification gave very different estimates of stroke incidence. However, the characteristics of the cases defined by the models, except for cases identified by death certificate only, were remarkably similar. There was also remarkable similarity in relative risks obtained from six of the seven models. The model using only death certificate information generally produced higher relative risk estimates.

Despite wide variations in the estimates of incidence, characteristics of the cases using different case definition were remarkably similar, as were the risks associated with stroke incidence. The main difference occurred when cases were identified from the death certificate only. These results furnish some evidence that analyses based on self report can provide valid, useful information.