Predicting patient deterioration by nurse intuition: The development and validation of the nurse intuition patient...

Early warning scores based on vital signs are used in hospitals to estimate patient deterioration and to initiate an adequate and timely response when necessary. These scores show acceptable performance in predicting patient outcomes. However, they tend to generate many false positives leading to an increased workload in clinical practice. Additionally, nurses feel a tension between the application of an early warning score and their own clinical judgement. Nurse intuition is often included as an extra call criterion next to an early warning score. It is therefore important to investigate its predictive value. The aim of this study was to develop and validate a Nurse Intuition Patient Deterioration Scale (NIPDS).

The NIPDS was developed using the latest evidence after which relevant items were selected by an expert panel. The scale was tested in a prospective observational study in 2 surgical and 2 medical wards in a Belgian hospital. Data were collected from December 1st, 2019 until March 31st, 2020. A NIPDS registration was done at each patient admission and whenever the attending nurse felt worried. The studied outcomes were urgent physician assistance call, resuscitation team call, patient death, and unplanned transfer to intensive/medium care. Psychometric analyses and scale optimisation were carried out using Rasch modelling techniques. Finally, the scale's accuracy and an optimal threshold were determined.

The scale item content validity index is 0.88 indicating that the selected items should be included in the instrument according to the expert panel. Item-total score correlation coefficients range between 0.573 (item 9 - pain) and 0.874 (item 6 - behaviour). The Person Separation Index is 0.814 indicating satisfactory discriminatory power. An overall fit of the NIPDS data to the Rasch model was confirmed. Rasch modelling showed that the item 'pain' signalled misfit. Furthermore, the person-item map showed disordered items which were corrected in the final model. The AUROC to predict an event within 24 h after registration was 0.957 (95% CI 0.932-0.982; p < 0.001) indicating excellent model performance.

The results showed that the NIPDS is a valid and accurate instrument to predict events in surgical and medical patients. It showed better performance compared to an existing score estimating nurse intuition. In practice, the NIPDS could be used by nurses to estimate clinical deterioration in addition to an early warning score. It remains unclear if the combination of NIPDS with an early warning score could reduce workload without losing accuracy and this should be explored in future research.

Newly developed nurse intuition scale, which uses clinical cues to estimate deterioration in hospitalised patients, is brief and performs well in predicting physician assistance, resuscitation team calls, patient death and unplanned transfer to intensive or medium care.