THE EFFECT ON RISK ESTIMATES OF EXCLUDING CASES FROM A CASE-CONTROL STUDY OF ISCHEMIC STROKE

There is limited information about the effect on stroke risk estimates of excluding cases who are unable to respond to interviews. A case-co ntrol study of ischemic stroke between 1988 and 1990 in Shreveport, La ., USA, provided a basis for studying this question. Of 197 consecutiv ely admitted cases of ischemic stroke, 77 were excluded due to dementi a, aphasia or impaired consciousness. Information about risk factors a nd stroke characteristics was obtained from hospital records. Excluded cases had more left hemispheric (52 versus 19%) and fewer vertebrobas ilar (12 versus 29%) and lacunar (5 versus 13%) infarcts than included cases. Excluded cases were also older (p < 0.01), and they had larger infarcts (p < 0.01), multiple strokes (p < 0.01) and congestive heart failure (p < 0.01) more often than included cases. Cases were matched to hospital controls by age, sex, race, and date of admission. Odds r atios (ORs) were higher for excluded cases for 5 of 6 exposures with a significantly higher OR for congestive heart failure (p < 0.01). The ORs changed by as much as 63% when excluded cases were added. These re sults emphasize the importance of acquiring information about excluded cases and considering selection bias when interpreting stroke studies that exclude cases who are unable to respond to interviews.