Implementation of a national AI technology program on cardiovascular outcomes and the health system

Fig. 1. Flow diagram of the 3-year coronary computed tomography angiography (CCTA) data. Data provided by the 27 NHS England hospitals with subsequent identification of patient numbers, repeat tests and patients excluded due to withdrawal of consent (national ‘Opt out of research’ database).

Fig. 2. Kaplan–Meier charts of the cumulative incidence of the individual primary objectives over 2 years after index CCTA. a–c, The incidence of all-cause death (a), cardiovascular death (b) and MI (c) rates. d, The incidence of ICA without subsequent revascularization. The shaded areas indicate the 95% CIs.

Fig. 3. The number of cardiac diagnostics performed as a second-line test within 2 years of the index CCTA scan (rate per 1,000 patients). Subcategorized into noninvasive tests (cardiovascular MRI, cardiac CT, nuclear medicine (positron emission tomography and radionuclide imaging) and echocardiography (excluding transthoracic echocardiography)) and invasive tests (total ICA and intracoronary imaging (optical coherence tomography (OCT), intravascular ultrasound (IVUS) and invasive FFR (FFR))). The error bars indicate the 95% CIs.

Extended Data Fig. 1. English hospital site map. A map of NHS England with the 25 different NHS Trusts plotted, their corresponding Integrated Care Board (ICB) and number of patients (n) contributed to the study.

Extended Data Fig. 2. Patient recruitment and site onboarding for the AI technology. a Patient recruitment and drop out for 6 months, 12 months and 24 month time points: Categorized as pre (blue line) or post (red line) FFR-CT availability. All 27 sites provided patient data from April 2017 – April 2020 with 2 sites (UHD and BRI) providing data from April 2017-December 2020. b Site FFR-CT availability: Introduction of FFR-CT at a site level was used to define whether the CCTA was performed before or after FFR-CT was made available at their site. The bubbles represent the time that FFR-CT was made available for clinical use and the number of CCTA scans performed at each site over the study recruitment time period (April 2017-December 2020).

Extended Data Fig. 3. Event curves for myocardial infarction and percutaneous intervention at different time points. a Myocardial Infarction (MI) events and B. Percutaneous Intervention (PCI) rates at 90 days, 1 year and 2 years. Cox proportional hazards univariate analysis (unadjusted) p values. The shaded areas indicate the 95% CIs.

Extended Data Fig. 4. Propensity Score Matching balancing. Propensity Score Matching covariate balancing pre and post matching using a standard mean difference of <0.05.

Extended Data Fig. 5. Event curves for the primary outcomes of the Propensity Matched Scoring population. a. All-cause death, b. Cardiovascular death, c. Myocardial infarction, d. ICA with no revascularization. Kaplan-Meier (KM) charts of the cumulative incidence of the individual primary objectives over 2 years post index CCTA on the Propensity Matched population (n = 30,665 in each group). Blue represents pre FFR-CT availability. Red line represents post FFR-CT availability. The shaded areas indicate the 95% CIs.

Extended Data Fig. 6. Referral to treatment time. Time (in years) from CCTA date (Time 0) to the date of revascularisation (PCI or CABG) for the 6,784 patients who had coronary revascularisation within the 2-year follow-up. The mean wait was 0.41 years for both FFR-CT unavailable and FFR-CT available groups. The group who received FFR-CT had a insignificant shorter wait (0.4 years). The box represents the interquartile range with the median as the middle segment, the mean as the black dot and the whiskers as minimum and maximum waits.

Extended Data Fig. 7. Autoregressive Integrated Moving Average curve for the incidence of myocardial infarction over teh study time period. Autoregressive Integrated Moving Average (ARIMA) method were performed. Weekly observed rates of all primary outcomes were assessed before and after FFR-CT introduction, then modelled for observed versus expected changes in outcomes post-health intervention over time. The dashed black line represents the time that FFR-CT was made available to the hospital sites.

Extended Data Fig. 8. FFR-CT report example. A positive FFR-CT (centre panel). The LAD shows a gradual reduction from proximal to distal vessel with a stenosis specific value of 0.72 and distal vessel value of 0.65. The circumflex was ‘negative’ at 0.89 and the RCA was occluded. The scale on the left panel shows the degree of flow limitation and likelihood of functional significance as a continuum with the margins of error on the right panel to help the physician in their decision process.