Hospital visits can be summed up in early illness and its consequences. But medical records tell a different story. Filled with doctor’s notes, patient histories, vital signs and lab results, it can last for weeks. Health studies multiply all that data with hundreds of patients. So, as AI data processing techniques become more and more sophisticated, it’s no wonder doctors are treating health as an AI and big data issue.
In one recent effort, researchers at Northwestern University applied machine learning to electronic medical records to learn more about pneumonia in intensive care units (ICUs) where patients are receiving ventilator respiratory support. I created a daily analysis. The analysis was published on April 27th. clinical research journalThis includes machine learning clustering of patient days, which suggests that the risk of long-term respiratory failure and secondary infections is higher than the cytokine storm that is the subject of early COVID-19 scare. , suggesting that it is much more common in COVID-19 patients.
“Most methods that approach data analysis in the ICU look at the data when the patient is admitted, and then look at outcomes at distant time points,” said study co-author Benjamin D. Singer of Northwestern University. rice field. “Everything in the middle is a black box.”
AI promises to turn daily ICU patient condition data into new clinical discoveries beyond COVID-19 case studies.
A day-by-day approach to the data led the researchers to two related findings. Secondary respiratory infections are a common threat to her ICU patients, including COVID-19 patients. And the strong association between COVID-19 and respiratory failure can be interpreted as an unexpected lack of evidence for a cytokine storm in patients with COVID-19. Ultimately, multiple organ failure would be expected if the patient exhibited an inflammatory cytokine response, which the researchers did not find. Although reported incidences vary, cytokine storms have been considered potentially dangerous in severe cases of COVID-19 since the early days of the pandemic.
Approximately 35% of patients were diagnosed with a secondary infection, also known as ventilator-associated pneumonia (VAP), at some point during their ICU stay. More than 57% of COVID-19 patients developed VAP, compared to more than 25% of non-COVID-19 patients. Multiple VAP episodes were reported in nearly 20 percent of COVID-19 patients.
Catherine Gao, a medical lecturer at Northwestern University and one of the study’s co-authors, said that thanks to the machine-learning algorithms the researchers used, “we saw clear patterns emerge that were clinically meaningful. I was able to do it. The team named the day-centric machine learning approach CarpeDiem, after a Latin phrase that means “seize the day.”
CarpeDiem is built using the Jupyter Notebook platform, and the team has made both code and anonymized data available. The dataset contains 44 different clinical parameters by patient day, and a clustering approach was used to classify 6 types of organ dysfunction (respiratory, ventilator instability, inflammatory, renal, neurological, 14 groups with different characteristics of shock) were returned.
“This field has focused on the idea that we can look at early data and see how it predicts. [patients] We will do it in the next few days, weeks or months,” Singer said. The hope is that studies using daily ICU patient conditions, rather than just a few time points, will help researchers and the AI and machine learning algorithms they use to better understand the efficacy of different treatments and responses to changing patient conditions. He said it would be possible to provide more information about . One direction for future research would be to look at the momentum of the disease, Singer said.
The method the researchers developed—which they call the “patient-day approach”—could capture other changes in clinical conditions in the short time between data points, says the authors, adding that predictive models for clinical practice could be used. said Sayon Dutta, an emergency physician at Massachusetts General Hospital, who helped develop It uses machine learning and was not involved in research. He said hourly data can pose unique problems for clustering approaches and can make it difficult to recognize patterns. “Instead, he thinks that dividing the day into eight-hour chunks would be a good compromise between granularity and dimensionality,” he said.
Even before the COVID-19 pandemic, there was a growing call for new technologies to analyze large volumes of ICU health data. Machine learning and computational approaches have the potential to be widely used in ICU in many ways beyond observational studies. Potential applications might include using real-time data recorded by daily health records or medical devices, or designing responsive machines that incorporate a wide variety of available information.
Overall mortality was approximately 40% in both those who developed secondary infections and those who did not. However, among study patients who had one diagnosed case of VAP, 76.5 percent eventually died or received hospice care if treatment for secondary pneumonia was not successful within her 14 days. was sent to The percentage of people who were considered cured of secondary pneumonia was 17.6%. Both groups included approximately 50 patients.
Singer emphasizes that the risk of secondary pneumonia is usually necessary. “A ventilator is absolutely life-saving in these cases. It’s up to us to find the best way to deal with the complications that arise,” he said. “You have to be alive to experience complications.”
from an article on your site
Related articles on the web
