A comparison of depressive symptoms in stroke and primary care: applying Rasch models to evaluate the center for epidemiologic studies-depression scale.

OBJECTIVES: Clinical trials and community-based studies often include the Center for Epidemiologic Studies-Depression scale (CES-D) as a measure of depression outcome. We compared responses to symptom-related items on the CES-D by depressed stroke and primary-care patients for several purposes: 1) to illustrate the use of Item Response Theory (IRT)-based (Rasch) models for comparing scale functioning across different patient subgroups; and 2) to inform clinicians and outcome researchers about scale functioning and depressive symptomatology in stroke- compared with primary care-based depression.
METHODS: Two data sources were analyzed, including 32 depressed patients who were 3 months poststroke, and 366 depressed primary-care patients. Presence of depression was based on a CES-D score 16 or higher. Rasch models were used to assess item fit and compare item hierarchies between depressed primary-care and stroke patients.
RESULTS: Item hierarchies were similar for poststroke depression and primary care-based depression. Interpersonal disruption items were the most difficult to endorse for both groups. No items misfit the scale in primary-care depression. Items relating to restless sleep, unfriendliness, and crying slightly misfit the scale in stroke patients, that is, may measure a different trait. Differential item functioning (DIF) between the groups was identified for items relating to appetite, restless sleep, crying, and feeling disliked.
CONCLUSIONS: Results generally supported the use of the CES-D as measure of depression outcome, particularly in primary care-based depression. DIF may imply that slightly different clusters of depressive symptoms are reported by depressed stroke patients compared with primary care, but this is conjectural given the small stroke sample size and the same items have been previously associated with bias in studies of large nonstroke samples. This study found Rasch models to be useful tools to investigate scale performance for different clinical applications.