Probability Assessment of Pulmonary Embolism (PE) Using Laboratory Values in Hospitalized Patients: A Single Center, Retrospective Observational Study and a Review of the Literature

Abstract

Introduction: Venous thromboembolism, including pulmonary embolism (PE) and deep venous thrombosis (DVT), is a common and potentially life-threatening condition in hospitals. Currently, the assessment of PE probability relies on subjective scoring systems, such as Well's and Geneva's scores, which have limitations in certain patient populations (demented, sedated, or intubated patients). This study aims to investigate the use of laboratory values for PE probability assessment, which has not been explored before. Methods: This retrospective observational study included 855 patients meeting the inclusion criteria for suspected PE (age 18 or older and the presence of signs and symptoms of suspected PE such as shortness of breath, chest pain, hemoptysis, tachypnea, tachycardia, and signs of DVT). Sixteen variables were examined, including laboratory values and clinical factors (Table 1). Logistic regression analysis, adjusting for confounding factors, was performed, and significant variables were used to develop a scoring method. Continuous variables that showed statistical significance were transformed into categorical variables using their respective maximum Youden index as the cut-off score. The receiver operating characteristic (ROC) curve was used to detect PE using the developed scoring system and determine the optimal cut-off value. A p-value of less than 0.05 was considered statistically significant. Results: The logistic regression model, incorporating several variables, achieved an 84.6% accuracy in predicting PE after accounting for confounding factors. Among the variables, hypocapnia, fever, alkaline phosphatase (ALP), D-dimer, and lactate had significant predictive values for PE (Table 1). ALP, D-dimer, and lactate were continuous variables with respective maximum Youden indices of 83.89, 952.50, and 1.614. Hypocapnia, ALP, and lactate showed negative slopes (B values), while fever and D-dimer had positive slopes in the logistic regression model. A scoring method was developed based on these variables. Fever, with an adjusted odds ratio (OR) of 1.995, received a score of 2 for values above the cut-off, while other variables with adjusted OR values at or below 1 were assigned a score of 1 above the cut-off. The mean ± SD score among all study participants was 1.84 ± 1.10, with a median (IQR) 2.0 (1.0 - 2.0). Patients without PE had a mean ± SD score of 1.80 ± 1.13, with a median (IQR) of 2 (1 - 2), while patients with PE had a mean ± SD score of 2.07 ± 0.91, with a median (IQR) of 2 (1 - 3). Patients with PE exhibited significantly higher mean ± SD scores than those without PE (P = 0.001; t = -5.813) (Table 2). The area under the ROC curve was 0.585 (95% CI: 0.563 - 0.606), indicating a statistically significant ability to detect PE (p-value = 0.001). Using a cut-off score of 1.5 based on the maximum Youden index, the scoring system achieved a sensitivity of 73.1% and specificity of 43.4% in detecting PE among the study participants. The Well's score, calculated for the study participants, demonstrated a sensitivity of 51.13% and specificity of 75.10% in detecting PE. Conclusion: This study presents a novel approach using laboratory values for PE probability assessment. The developed scoring method showed promising results, providing an alternative to the current accepted scoring systems. Further research and validation are needed to confirm the utility of laboratory-based assessment in clinical practice.