Endocrine Abstracts

Introduction: Diabetes mellitus is a risk factor for micro-macrovascular complications and increases the risk of hospitalization, and adverse events. Early detection of impending deterioration can prompt interventions and prevent adverse events. Early warning systems (EWS) combined with rapid response teams can mitigate adverse events in the hospital setting. Machine learning-based EWS can use electronic health record data and automatically alert the care team with enough time for an early intervention. However, implementation of these systems can be impacted by their reliability, biases, and poor clinical adoption. Here we present our findings evaluating the performance of an EWS integrated into the EHR in patients with diabetes mellitus compared with those without diabetes.

Methods: We collected machine learning-based EWS scores for adult patients (≥ 18 y-o) hospitalized on general wards. The EWS estimated the probability of an adverse event (mortality, cardiac arrest, intensive care transfer, or evaluation by rapid response team) on a scale of 0 to 100. We used three metrics to characterize and compare the distributions of the scores among patients with diabetes and without diabetes: the First Score 3 hours after admission; the Highest Score at any time during the hospitalization; and the Last Score just before an adverse event or before dismissal for those without an adverse event. Additional data was collected including age, sex, and length of stay.

Results: Among 61,151 admissions (female 51%, mean age 62(SD 18.5) years, length of stay 4.7(6.1) days) there were 15,727 (26%) patients with diabetes. Patients with diabetes had similar rate of adverse events compared to non-diabetics (11.92% vs 11.63%, P=0.323). However, the distributions of the First, Highest and Last Scores were higher in patients with diabetes and significantly different (P<0.005) when compared with non-diabetics (mean 32.5 vs 27.4, 46.3 vs 40.1 and 31.5 vs 27.2 respectively).

Conclusion: When using a machine learning-based EWS, patients with diabetes seem to have higher risk of deterioration when compared with nondiabetic. However, they seem to have the same proportion of adverse events compared to nondiabetics. These findings are reassuring of the hospital care received by patients with diabetes; despite their higher risk they are not suffering additional complications. Our study was performed in a single institution and did not investigate the specific causes or clinical interventions responsible for the findings. Additional assessment of these factors could prove to be important when planning best hospital practices for diabetic patients admitted in general wards.