Deterioration Index Model Modestly Predicts Patient Outcomes

A new study found that although the use of the Deterioration Index (DTI) within the hospital setting provided modest results, it inconsistently s handled various demographic groups. Researchers concluded that further validation is needed for this model.

Deterioration refers to mechanical ventilation, intensive care unit transfer, or death within the hospital. According to the study, about 15 percent of avoidable deaths in hospitals result from neglecting clinical decline.

Given the prevalence of this issue, methods of measuring clinical deterioration exist. One tool is the DTI, a machine-learning model that was developed in 2017. Although hundreds of hospitals engage with this model, it remains unvalidated externally. Lack of validation creates a gray area surrounding its ability to operate equitably.

Thus, researchers aimed to validate this model and determine its potential for bias. In this study, researchers included eight heterogeneous Midwestern United States hospitals with a population of 13,737 patients. This patient population produced 5,143,513 DTI predictions, 14,834 hospitalizations, and 13,918 encounters.

According to the study, deterioration describes instances related to mechanical ventilation, intensive care unit transfer, or death in the hospital. The total prevalence of deterioration was 10.3 percent. There was not one consistent result of bias measures among all subgroups. For those who identified as American Indian or Alaska Native, bias measures were 14 percent worse. Among patients who did not disclose ethnicity, this measure was 19 percent.

This result led researchers to conclude that DTI is modestly able to foresee patient deterioration. However, inconsistent outcomes at the observation and encounter levels across various demographic groups prompted researchers to call for further action. This involved the need to integrate transparency in model training data and further validate models.

The use of machine learning to predict patient deterioration is common and values transparency.

In April, the Nationwide Children’s Hospital created a machine-learning model that considered the deterioration risk index (DRI) to predict risk for hospitalized children w. In doing so, researchers aimed to execute this process faster than traditional programs, as early detection is valuable in preventing adverse events. Along with the DRI, researchers also considered EHRs. This allowed for access to extensive data.

Using information from cardiac, malignancy, and general diagnostic groups, researchers trained three predictive models. These models assisted them in creating the algorithms for the tool.

Following the research, they found that the DRI reached a sensitivity level significantly higher than the existing situational awareness program. Precise alerting was also a feature that the model displayed.

Compared to the situational awareness program, the model led to a 77 percent reduction in deterioration events during the initial 18 months. Along with this, the model was transparent.

"This is not a black box. We show clinicians what goes in and how the algorithm evaluates the data to trigger alarms," said Tyler Gorham, data scientist in IT research and innovation at Nationwide Children's and co-author of the publication, in a press release. "The tool helps support clinical decision-making because the clinical team is able to see why an alarm was triggered."  "The tool helps support clinical decision making because the clinical team is able to see why an alarm was triggered."

Furthermore, machine learning can predict patient deterioration. However, researchers must consider transparency when using this resource type to predict deterioration.