The DASH model, an early warning score (EWS) tool developed through machine learning, demonstrated superior benefits in identifying deterioration among patients undergoing hematopoietic cell transplantation (HCT), which has great potential to improve patient outcomes, according to data presented by Jeannine M. Brant PhD, APRN, AOCN, FAAN, City of Hope, Duarte, California, at the Oncology Nursing Society (ONS) Annual Congress on April 25, 2024.

Brant and coauthors explained, “[EWS] are widely used in hospital settings to predict patient deterioration but are too sensitive and lack specificity for patients undergoing [HCT], which can lead to false alerts, alarm fatigue, lack of risk identification, and compromised patient outcomes.”

To address this issue, the stated purpose of the study was to develop an EWS for patients undergoing HCT and to implement the model into the clinical setting, the authors explained. To achieve this, a machine learning model – light gradient boosting model (lightGBM) – was used to develop an EWS model called DASH. In this model, the areas under the receiver operating characteristic curve (AUROC) and precision-recall curve (AUPRC) of the model were calculated and compared to those of the former EWS model (Modified-EWS or MEWS).

Once the DASH prototype was developed, the study authors strategically utilized 3 focus groups of nurses, advanced practitioners, physicians, and leaders to address potential challenges and opportunities for successful  implementation and to develop a standardized protocol to escalate care when patient deterioration is detected through DASH alerts. Following those steps, in the study, the DASH model prototype outperformed the current MEWS model; AUROC was 0.86 for DASH compared to 0.71 with MEWS; AUPRC was 0.15 for DASH compared to 0.09 with MEWS.

The data revealed that the strongest predictors of deterioration were low white blood cell count, lack of oxygenation support, high heart rate, high blood urea nitrogen, high respiratory rate, and high blood glucose. The focus group themes also identified issues such as alarm fatigue and ignoring MEWS alerts due to over-alerting the clinician. In these cases, the most beneficial opportunity was to educate the full healthcare team regarding the improved model performance, in order to encourage appropriate response. They also developed care escalation protocols, which included guidance on when to transfer the patient to a higher level of care.

Brant and colleagues concluded, “Machine learning offers opportunities to predict outcomes in patients with cancer and was used in this study to develop a superior model to detect deterioration in patients undergoing HCT. This early detection occurs before clinicians can intuitively or quantitatively distinguish change, which has great potential to improve patient outcomes. This is the first model developed to predict deterioration in patients undergoing HCT. “

“Additionally, while ML models are emerging in cancer care, none recommend implementation strategies. This study engaged frontline clinicians to garner qualitative perspectives about model implementation to explore strategies and barriers to successful implementation,” they added.

Source:

Brant J M, Castro J, Carlin C, et al. Deterioration in patients undergoing hematopoietic cell transplantation: development and implementation of the DASH model. Presented at Oncology Nursing Society Annual Congress. April 24-28, 2024; Washington, DC.