83 Utility of the national echocardiography database of the United Kingdom (NED-UK) pilot in predicting time to cardiac surgery

<h3>Background</h3> The National Echocardiography Database of the United Kingdom (NED-UK) pilot study demonstrates the feasibility of extracting echocardiographic (echo) ‘big data’ from NHS databases in a standardised way. Our aim was to demonstrate utility of NED-UK pilot data by showing associations of echo reported variables with patient outcomes, specifically the time from echo to cardiac surgery (bypass grafting and/or valve replacement or repair). <h3>Methods</h3> Data were collected under the ‘EchoVision’ project (UK HRA approval 251473). Data from consecutive echos, undertaken at an NHS Trust from 2017–2023 were extracted using Philips Advanced Analytics software. The data were pre-processed in RStudio (v2023.09.1) using R (v4.2.1) to remove scans with age &lt;18 years or missing age, sex, height or weight. Echo report data (measurements, standardised interpretive phrases and free-text comments) were crossmatched with an existing internal cardiac surgery dataset (including date of surgery) to identify the earliest echo prior to patients receiving their first cardiac surgery. Patients with prior cardiac intervention, congenital heart disease or an implantable cardiac electronic device were excluded. The time from echo to cardiac surgery was calculated (time-to-surgery (TTS)). Variables or patients with &gt;30% data incompleteness were removed. Residual missing values were imputed using the modal class (categorical variables) or trimmed scores regression in Matlab (numerical variables). Echo variables had normality determined using the Shapiro-Wilk test. Categorical variables were binary coded as 0/1. A multiple linear regression model was built using R in RStudio to predict TTS. The echo variables that significantly correlated with TTS were entered with forward-stepwise selection. Alpha was set at p&lt;0.01 to account for multiple hypothesis testing. Visualisations were made using the ‘ggplot2’ R library. <h3>Results</h3> 367 patients (mean age 64±11 years, 295 (80.4%) male) receiving their first cardiac surgery with a suitable echo were obtained. TTS ranged from 1–2248 days (median 57, interquartile range 459). 7.3% of echo values were imputed. Of 39 possible predictor variables, 14 had significant correlations with the TTS (table 1). The final regression model contained 7 of these with the following equation; Predicted TTS = (21.58×EF) + (125.24×LVIDd) - (7.98×LAVi) + (294.72 if comment of LA dilation) + (147.22 if ≥moderate MR) + (460.08 if comment of hyperdynamic LV) + (178.78 if ≥moderate AS) - 1373. The predicted-TTS correlated with the actual-TTS (Spearman r=0.395, p&lt;0.0001) with a near-linear locally weighted scatterplot smoothed trendline (figure 1). <h3>Conclusions</h3> This study demonstrates proof of concept that echo report data can be linked to patient outcomes. Routine echo variables are associated with, and could be used to predict, the time from echo to cardiac surgery. Multi-site application of NED-UK methods could be valuable in predicting cardiovascular outcomes in a diverse UK patient population. <h3>Conflict of Interest</h3> None