Nursing assistants: 'He seems to be ill' - a reason for nurses to take action: validation of the Early Detection Scale of Infection (EDIS)

Background: Signs and symptoms of infection in frail elderly are atypical, causing delay in diagnosis and treatment. To improve communication between healthcare staff of signs and symptoms of infection we developed an instrument, using qualitative data from observations by nursing assistants when they suspected infection. The aim of this study was to assess the validity of nursing assistants observations by developing and testing the instrument for early detection of infection in elderly nursing home residents. Methods: The early detection of infection (EDIS) instrument was based on data from focus interviews with nursing assistants. Over one year the nursing assistants used EDIS to document episodes of suspected early signs and symptoms of infection in 204 nursing home residents. Two physicians classified documented episodes as "no infection", "possible infection", and "infection". The content validity of the 13 items of the EDIS was established to explore the relationships between the items. The construct validity was used to explore the relationship between the items and the presence or absence of infection. The predictive value of the developed model was evaluated by the percentage of correct classifications of the observed cases. Generalized linear model (ordinal multinomial distribution and logit link) was used. Results: Of the 388 events of suspected infection, 20 % were assessed as no infection, 31 % as possible infection and 49 % as infection. Content validity analysis showed that 12/13 of the items correlated significantly with at least one other statement. The range in number of significant inter-correlations was from 0 ("pain") to 8 ("general signs and symptoms of illness"). The construct validity showed that the items "temperature", "respiratory symptoms" and "general signs and symptoms of illness" were significantly related to "infection", and these were also selected in the model-building. These items predicted correct alternative responses in 61 % of
Funding Agencies|Medical Research Council of Southeast Sweden; Futurum/The Academy for Health and Care, Jonkoping County, Sweden