A retrospective study took place at a 350-bed community teaching hospital in Poughkeepsie, NY. Patients diagnosed with intermediate PEs between February 2020 and February 2024 underwent a chart review of clinical, laboratory, computed tomography pulmonary angiography (CTPA), and transthoracic echocardiography (TTE) findings from the time of PE diagnosis and for 72 h thereafter. The outcome of interest was short-term clinical deterioration, defined as either respiratory failure or hemodynamic instability, within 72 h of PE diagnosis.

This study was approved by the Institutional Review Board (IRB), with a waiver of informed consent. We adhered to the Transparent Reporting of a Multivariable Prediction Model for Individual Prognosis or Diagnosis (TRIPOD) reporting guideline for reporting multivariate prediction model development and validation [12].

Subjects were identified through the inpatient discharge database (DAD) using the ICD-10 code for PE. The electronic medical record (EMR) was reviewed to confirm diagnosis of PE by CTPA. All confirmed PE cases were screened for markers of RVD within the first 24 h of admission. Thresholds for RVD included: high-sensitivity cardiac troponin (hs-cTnT) > 30 ng/L, N-terminal pro-brain natriuretic peptide (NT-proBNP) > 300 pg/mL, CTPA findings (RV: LV ratio > 1.0, septal deviation, or retrograde contrast reflux into the vena cava), or TTE findings (RV dilation (end-diastolic diameter > 30 mm), right ventricular systolic pressure (RVSP) > 40 mm Hg, septal deviation, positive McConnell’s sign, or positive “D” sign.

Eligible subjects were ≥ 18 years old with confirmed acute PE and at least one marker of RVD. Low-risk PE patients and pregnant patients were excluded. Subjects were divided into two cohorts. The derivation cohort consisted of all available data at the time of retrieval on December 18, 2023. This cohort was used to develop, evaluate, and internally validate the predictive model and clinical scoring system. The validation cohort consisted of subjects whose data was uploaded after December 19, 2023. This cohort was used to evaluate the model’s performance on an independent sample. The data for both the derivation and validation cohorts was obtained from the same hospital. Subjects missing any predictor variables were excluded from the final derivation and validation models. Subjects were used in a post-hoc extrapolation cohort if missing only one predictor variable.

The outcome of interest was whether the subject had clinical deterioration, defined as worsening respiratory failure or hemodynamic instability, within 72 h of PE diagnosis. Respiratory failure was defined as the need for high flow nasal cannula (HFNC), non-invasive positive pressure ventilation (NIPPV), or mechanical ventilation (MV) to maintain oxygen saturation (SpO2) above 90%. Hemodynamic instability was defined as at least one of the following criteria: systolic blood pressure (SBP) less than 90 mm Hg or a drop of greater than 40 mm Hg from baseline for greater than 15 min, the need for catecholamine administration to maintain SBP above 90 mm Hg, or cardiac arrest.

Subjects were collected from a central list, de-identified, and uploaded into a secure Redcap database. Data entry occurred periodically between October 2022 and February 2024. Data was collected from the time of PE diagnosis and for 72 h afterwards. The data collected at the time of PE diagnosis was used for the current study and included the following:

Patient characteristics: Baseline demographics (age, sex, race, BMI); chronic medical conditions (including preexisting heart failure, chronic lung disease, active or past malignancy, and prior venous thromboembolism); additional risk factors for PE (recent COVID-19 or other systemic infection, travel, surgery, or hospitalization).

Clinical data: Symptomatology (syncope, chest pain, dyspnea, leg pain or swelling, altered mental status); highest and lowest heart rate (HR), SBP, and supplemental oxygen (O2) requirements (above baseline) prior to admission. Additional clinical data was collected for 72 h from the time of PE diagnosis, including trends in vital signs, supplemental oxygen requirements, vasopressor requirements (if any), and medical documentation pertaining to escalation of care or clinical deterioration.

Laboratory and imaging results: Venous plasma samples were collected on arrival by laboratory personnel and included hs-cTnT, NT-proBNP, and lactic acid. Radiologists independently reviewed CTPA images and reported the most proximal and distal anatomical thrombus distribution. Radiologists reported the presence or absence of RVD, and the RV: LV ratio as either less than or equal to 1.0, or greater than 1.0. A TTE was performed within 24 h by a certified ultrasound technician and read by in-house cardiologists per the Intersocietal Accreditation Commission (IAC) Standards and Guidelines for Echocardiography Accreditation [13]. The parameters collected included the presence or absence of: RV dilation (end-diastolic diameter greater than or equal to 30 mm from the parasternal view), interventricular septum deviation, clot in transit, McConnell’s sign, and valvular dysfunction. Numerical values were collected for left ventricular ejection fraction (LVEF [%]), RVSP [mm Hg], and TAPSE [mm]. Bilateral lower limb complete duplex ultrasound (CDUS) testing was performed within 24 h by certified ultrasound technicians using two-dimensional imaging, graded compression, and Doppler analysis. Images were reviewed by in-house radiologists and reported as: proximal deep vein thrombosis (DVT), distal DVT, both, or absent.

All baseline characteristics and potential predictor variables were compared by study group (Adverse outcome, Y/N). Descriptive data was reported as absolute numbers, percentages, or means +/- standard deviation (means +/- SD) or medians (interquartile range). Subject baseline characteristics and clinical factors, laboratory results, CT, and echocardiogram results were compared by study group [Clinical Deterioration (Yes/No)]. Some continuous variables (TAPSE, Lactate, supplemental O2, hs-cTnT) were converted to dichotomous variables using predefined cut-offs based on clinical significance and prior research studies. Normally distributed continuous variables were compared using the student’s t test, whereas non-normally distributed variables were compared using the Wilcoxon rank-sum tests. Categorical variables were compared using chi-square analysis or Fisher’s exact test. Variables with a significant relationship to the outcome in the univariate analysis were considered as covariates in the multivariable model. A stepwise logistic regression was performed to screen for potential variables for inclusion in the final model. In addition to variables selected by the stepwise procedure, variables deemed clinically relevant by the investigators were included in the selection procedure. Subjects with missing data for significant predictor variables were excluded from analyses by the regression procedure. Five variables associated with an increased risk of short-term clinical deterioration were incorporated into the final regression model. The goodness-of-fit was assessed using Akaike Information Criterion (AIC) and the Hosmer-Lemeshow test. A clinical risk score (PEP Score), ranging from 0 to 11 points, was created based on the results of the final logistic regression model. For each of the five included variables, a point value was assigned based on the rounded square root of the odds ratio. A receiver’s operator curve (ROC) was created to assess the accuracy of the model. For each score cutoff, the probability, sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV) were calculated. Youden’s index was calculated to identify the point with the highest sensitivity and specificity. Accuracy was measured using the brier score, measuring the difference between predicted vs. observed outcomes. Using a second dataset, PEP scores were calculated for each subject, and the process was repeated– an ROC curve was created, and the sensitivity, specificity, PPV, NPV were calculated for validation. A post-hoc analysis was performed using the subjects excluded from the derivation and validation cohorts due to missing variables. This modified PEP score (mPEP) included all subjects missing one of the five predictor variables. The missing variable was assumed to be normal if not preset, making the modified PEP score with a reduced potential range of 0 to 9 points. Statistical analysis was performed using the SAS statistical software version 9.4 (SAS Institute, Cary, NC, USA).