An ML model to select patients for early intervention following Blunt Splenic Injury

Dr Sanjay Rathee

8/25/20221 min read

We have developed a machine learning model to predict which haemodynamically stable patients would benefit from early intervention. The current model, which has been validated, performs well across all parameters with sensitivity (1.0), specificity (0.816), PPV (0.588), NPV (1.0), balance accuracy (0.908) and AUC (0.932). *(note: 1.0 =100%)This model has been made into an application (similar to NELA risk calculator) that can be used in emergency setting (Figure 1). It takes less than a minute to input the data and all data all readily available in emergency setting.

Clinical problem:Non-operative management (NOM), including intensive monitoring and splenic artery embolization (SAE), is currently the standard of care for haemodynamically stable patients with blunt splenic injury (BSI).The benefits of NOM include spleen preservation and the avoidance of physiological insult and complications from emergency laparotomy and splenectomy. However, current international guidelines have differing opinions on who should receive SAE.A recent randomised control trial (Arvieux et al. 2020) comparing ‘prophylactic SAE’ versus ‘embolisation as necessary’ for high grade BSI could not provide an answer⁠. Of note, 30% of patients in ‘embolisation as necessary’ group needed secondary SAE.Early embolisation is associated with increased splenic salvage, fewer secondary embolisation and shorter length of hospital stay, highlighting the need for better patient selection.

UHS data: Reviewing data in UHS from 2015-2022, for patients who underwent NOM for BSI (n=174), 18 patients had SAE on arrival and another 13 patients needed subsequent SAE, with time to failure ranging from within 24 hours to 7 days. Of all the patients who had SAE (n=31), five (16%) went on to have splenectomy.

Related Stories