The Risk of Selection Bias in a Clinical Multi-Center Cohort Study. Results from the Norwegian Cognitive Impairment After Stroke (Nor-COAST) Study.

Purpose: The Norwegian Cognitive Impairment After Stroke (Nor-COAST) study aimed to estimate the prevalence and incidence of neurocognitive disorder in an unselected stroke cohort. The aim of the present study was to investigate whether selection bias occurred by comparing baseline characteristics from participants with non-participants in Nor-COAST.

Patients and Methods: Nor-COAST is a prospective cohort multi-center study, recruiting participants from five Norwegian hospitals. Patients with the diagnosis of acute stroke were screened for inclusion. Baseline data from the participants recruited between May 2015 and March 2017 were compared to corresponding data from those not participating in Nor-COAST but registered in the Norwegian Stroke Registry. Regression analysis was used to assess whether age, stroke severity, sex and stroke subtype were independently associated with inclusion in the study.

Results: Out of 2505 available patients, 815 (32.5%) were included in Nor-COAST. There were no differences between participants and non-participants with respect to age (mean (SD) age 73.5 (11.7) versus 74.2 (14.5) years) or sex (44.8% versus 46.9% women). A significantly larger proportion of the participants were independent prior to stroke (87% versus 78%), had mild strokes (69% versus 55%) and suffered from cerebral infarction (90% versus 84%). The regression analysis showed decreased odds ratio (OR) of being included for those with higher degree of pre-stroke dependency (OR 0.895, 95% CI 0.825 to 0.971, p=0.007) and a more severe stroke (OR 0.952, 95% CI 0.939 to 0.966, p<0.001).

Conclusion: The participants in Nor-COAST had a better pre-stroke health condition and milder strokes compared to non-participants. However, the participants should be regarded as representative of the majority of the stroke population which suffers from mild strokes. Nevertheless, baseline information for non-participants should be available also in future clinical studies to make it easier to identify which part of the stroke population the results can be generalized to.