Academy of Diagnostics & Laboratory Medicine - Scientific Short
Scientific Shorts are brought to you by the
Sepsis is a serious global health issue, causing around 50 million cases and 11 million deaths each year—making up about 20% of all global fatalities1. It’s not only the leading cause of hospital readmissions and deaths2,3, but it also comes with a hefty price tag, costing more than $60 billion annually4. Surprisingly, around 80% of sepsis cases happen outside of hospitals, and 60% of patients visit outpatient facilities within a week before being admitted5,6. This presents a real opportunity to catch the risk of sepsis early, since timely detection and treatment could prevent up to 80% of sepsis-related deaths.
Our research suggests that machine learning models could be key in spotting sepsis risk through routine blood tests. We analyzed a dataset of 25,000 patient records—half of whom had been diagnosed with sepsis, while the other half served as a control group. We used 70% of the data to train our model and the remaining 30% to test it. We employed a gradient boosted model with 300 trees, a maximal depth of 30 layers, and gain ratio as the criterion for attribute selection. The model's performance was evaluated using 10-fold cross-validation, demonstrating optimal results. Input parameters included age, gender, and results of routine blood markers such as complete blood counts, differential counts, comprehensive metabolic panels, and lipid panels recorded up to 1 week before sepsis diagnosis. The results were impressive! Our model was able to identify nearly every case of sepsis and ruled out sepsis in 99% of the control cases. This gave it a 100% negative predictive value and a 99% positive predictive value.
The model pinpointed specific blood markers that indicated sepsis risk, including calcium, albumin, bilirubin, hematocrit, white blood cells, and cholesterol. Research shows how these markers play a role in sepsis. For instance, calcium levels during sepsis can indicate an intensified inflammatory response and potential organ dysfunction. Low albumin is associated with worse outcomes, while high bilirubin levels can signal a greater risk of mortality. Additionally, low hematocrit levels are linked to higher 30-day mortality rates and can serve as independent indicators of prognosis. White blood cells influence inflammatory responses and various immune cells, and cholesterol levels can be protective, with low concentrations often tied to poor outcomes.
Given that every hour of delayed treatment can raise the risk of death by 4% to 9%7, it’s crucial that changes in routine blood markers detected by machine learning models could alert us to potential sepsis risks as much as a week before a diagnosis is made.
