Portfolio

Problem Definition

Measurement and Correlation. Define, validate measures of attributes and explore relationships

Coagulation abnormality is found in sepsis patients. Bleeding-related complications are also common. Study hypothesis involving whether or not to be aggressive in correcting high INR is a consistent bedside question. Identify ICU patients with sepsis who are at higher risk and quantify a relationship between sepsis coagulopathy, hemorrhage and mortality

‍

Inputs

‍Inputs: Team Lead, 3 person cross-functional project team, Mimic iv ICU dataset ( https://mimic.mit.edu/iv/ ), deidentified longitudinal, cross sectional EHR data: 53,423 critical care admissions, 26 tables, 324 variables Charted Events, laboratory measurements (LOINC), over 2 million rows of unstructured data (provider notes) coded with SNOMED CT, ICD-9, ICD-10 and LOINC codes.

Algorithm

Algorithm: R, SQL, Data extraction, inclusion/exclusion criteria, exploratory analysis, logistic regression, random forest decision tree modeling, correlation matrix

Outputs

Outputs: Relationship between INR levels & bleeding, mortality. Identify high risk ICU patients with sepsis.

Mortality risk score for decision support tool.
Decision tree bedside tool
Regression model: Predict mortality, hemorrhage with INR levels.
Predicts likelihood of sepsis-induced coagulopathy

Sepsis patients categorized by ICD diagnoses code

Odds Ratio = Odds of bleeding is 130% higher for the patients with severely abnormal INR compared to those with normal INR. Risk Ratio = Patients with sepsis who had higher abnormal INR values had 65% higher risk of bleeding compared to patients with normal INR.

Relationship of INR group with Mortality

Pearson's Chi-square provides evidence that INR category is related to mortality Odds Ratio = Estimated odds for mildly abnormal INR is 180% higher for the patients who expired compared to those who lived.

Impact of Mortality

A 20 yr old severely abnormal INR is twice as likely to die compared to a 70 yr old with normal INR. The derived model serves as a bedside aid based on INR grouping and the impact of other factors available to physicians. A 20 yr old severely abnormal INR is twice as likely to die compared to a 70 yr old with normal INR.

Clinical Decision Support tool

Bedside aids to support clinicians in assessing risk of Mortality Prediction: Positive correlation of INR category and risk of bleeding. There`s a 40% chance that the patient will bleed if the patient`s INR ranking is > 1 , platelet count is <= 132 and length of stay is > 26.5 days

GITHUB

Sepsis Coagulopathy

Input:

Algorithm:

Output: