Baneen Alhmoud1, Amitava Banerjee1, Daniel Melly2

1University College London, London, United Kingdom, 2Barts Health Trust, London, United Kingdom

Background:
Individuals with cardiovascular disease (CVD) have critical healthcare needs that often have related risks of deterioration when the disease is in an acute phase. The short-term outcomes resulting from acute deterioration, such as cardiac arrest and transfer to critical care, are common while morbidity and mortality are major concerns in these individuals. Risk models for the early detection of short-term deterioration have not been employed for this disease subgroup. Early Warning Score (EWS) are designed tools for the systemic monitoring of patients’ physiological parameters to provide early detection of deteriorating events. EWS’ performance accuracy in cardiac patients’ subgroup and cardiac care settings remains vague. This study demonstrates an evaluation of an early warning score, the non-digital format of NEWS2 prior EHR integration,  in accurately predicting deterioration in patients admitted to a cardiac-specialised hospital in the United Kingdom.

Method(s):
We conducted a retrospective cohort study in one of the leading cardiovascular disease centres in the UK, St Bartholomew’s Hospital, London. Included participants were patients admitted to the hospital with a primary diagnosis of CVD between January and December 2019. NEWS2 score is recorded within 48 hours before a critical event. The primary outcome was cardiac arrest and mortality within 48 hours, and a secondary outcome was a 2-day composite outcome (cardiac arrest and mortality). The predictive accuracy was measured by plotting the sensitivity against 1-specificity (ROC) and calculating the area under the curve (AUC). Positive predictive value (PPV), negative predictive value (NPV), sensitivity and specificity for various NEWS2 cut-offs points were calculated.

Result(s):
The AUC of NEWS2 in predicting cardiac arrest is 0.64 (95% CI 0.59–0.69), predicting mortality is 0.67 (95% CI 0.58–0.75), and the composite outcome is 0.75 (95% CI 0.70–0.80). When comparing the different examined cut-off points, a marked improvement is noted in sensitivity when predicting the three outcomes at a lower cut-off of 3. Cardiac arrest is the least accurately predicted outcome by NEWS2 from the outcomes examined.

Conclusion(s):
In patients with cardiovascular diseases, NEWS2 seems to have lower than expected ability to provide accurate scoring when escalation of care is needed. Referring to NEWS2 seemed to provide complementary rather than necessary information. Although NEWS2’s accuracy has been reported to be statistically significant, yet it lacks the discriminative power to be clinically useful for patients admitted with CVD. The poor accuracy detected in the study may indicate a need for tool threshold-calibration and parameters-validation when implementing NEWS in a cardiac setting. EHRs integration with NEWS2 is likely to contribute to facilitated documentation and more accurate scoring. A revaluation of the EWS score format and factors contributing to its success in the cardiac settings seems necessary.