Abstract
Background:
Coronary artery calcium (CAC) can be identified on nongated chest computed tomography (CT) scans, but this finding is not consistently incorporated into care. A deep learning algorithm enables opportunistic CAC screening of nongated chest CT scans. Our objective was to evaluate the effect of notifying clinicians and patients of incidental CAC on statin initiation.
Methods:
NOTIFY-1 (Incidental Coronary Calcification Quality Improvement Project) was a randomized quality improvement project in the Stanford Health Care System. Patients without known atherosclerotic cardiovascular disease or a previous statin prescription were screened for CAC on a previous nongated chest CT scan from 2014 to 2019 using a validated deep learning algorithm with radiologist confirmation. Patients with incidental CAC were randomly assigned to notification of the primary care clinician and patient versus usual care. Notification included a patient-specific image of CAC and guideline recommendations regarding statin use. The primary outcome was statin prescription within 6 months.
Results:
Among 2113 patients who met initial clinical inclusion criteria, CAC was identified by the algorithm in 424 patients. After chart review and additional exclusions were made, a radiologist confirmed CAC among 173 of 194 patients (89.2%) who were randomly assigned to notification or usual care. At 6 months, the statin prescription rate was 51.2% (44/86) in the notification arm versus 6.9% (6/87) with usual care (P<0.001). There was also more coronary artery disease testing in the notification arm (15.1% [13/86] versus 2.3% [2/87]; P=0.008).
Conclusions:
Opportunistic CAC screening of previous nongated chest CT scans followed by clinician and patient notification led to a significant increase in statin prescriptions. Further research is needed to determine whether this approach can reduce atherosclerotic cardiovascular disease events.
Registration:
Clinical Perspective
What Is New?
•
In this randomized quality improvement project, statin-naive patients with coronary artery calcium on previous nongated computed tomography scans were identified through a deep learning algorithm.
•
Notification of patients and their clinicians regarding incidental coronary artery calcium with a personalized image of the coronary artery calcium increased statin prescription rates.
What Are the Clinical Implications?
•
Opportunistic coronary artery calcium screening on previous chest computed tomography scans is a potential approach to identify patients at high risk of atherosclerotic cardiovascular disease events who would benefit from preventive interventions.
•
The impact of such a screening strategy on clinical events is unknown and should be tested in a prospective randomized clinical trial.
Editorial, see p 715
More than 700 000 Americans have a first acute myocardial infarction or die of coronary artery disease annually.1 Many events could be avoided with greater application of preventive interventions among individuals at increased risk, including statin therapy. Multiple analyses have demonstrated suboptimal statin therapy rates compared with guideline recommendations.2–4 Strategies to promote shared decision-making discussions between patients and clinicians are needed to address patient-level factors that may influence decisions to initiate statin therapy.5
Coronary artery calcium (CAC) testing is a promising approach to identify high-risk individuals and motivate adoption of preventive interventions. The presence of CAC is a strong predictor of atherosclerotic cardiovascular disease (ASCVD) events.6 The 2018 American College of Cardiology/American Heart Association Guidelines on Management of Blood Cholesterol include a IIa recommendation for initiating statin treatment among patients with a CAC score ≥100, a CAC score ≥75th percentile for age and sex, or a CAC score >0 with age ≥55 years.7 However, traditional CAC testing, which is performed on ECG-gated, noncontrast computed tomography (CT) scans, is rarely performed; and >19 million nongated, noncontrast chest CT scans are performed annually for reasons other than to measure CAC.8 We previously developed a deep learning algorithm to estimate CAC score (DL-CAC) on nongated, noncontrast chest CT scans to efficiently implement opportunistic screening for incidental CAC.9 For detecting CAC, the DL-CAC algorithm demonstrated a sensitivity of 82% to 94%, a specificity of 79% to 100%, and a positive predictive value of 87% to 100% across 4 external data sets.
With no previous randomized studies using incidental CAC, the effect of notifying clinicians and patients regarding the presence of incidental CAC is unknown. We designed a randomized quality improvement project, NOTIFY-1 (Incidental Coronary Calcification Quality Improvement Project), to understand whether notifying primary care clinicians and patients of incidental CAC would affect statin prescription rates among statin-naive patients without known ASCVD.
Methods
This was a prospective, randomized quality improvement project comparing the notification of incidental CAC on previous nongated chest CT scans with usual care among statin-naive patients without known ASCVD in the Stanford Health Care System. This project was motivated by institutional efforts to increase primary prevention statin therapy among high-risk patients. The quality improvement project was deemed exempt from human subject research requirements by the Stanford institutional review board. The project was registered on clinicaltrials.gov (NCT04789278).10 To protect confidentiality, deidentified data are available from the corresponding author only on reasonable request.
Study Population
Patients <85 years of age with any amount of CAC on a noncontrast, nongated chest CT scan between 2014 and 2019 were considered for inclusion. Patients required a Stanford primary care clinician or a Stanford endocrinologist with whom they had a previous encounter (in-person or telehealth) in the previous 2 years. Patients with an ASCVD diagnosis, previous statin prescription, or previous invasive coronary angiography or coronary CT angiography were excluded. Patients with dementia, hospice enrollment, metastatic cancer or nonmetastatic cancer receiving treatment, or other advanced diseases with limited life expectancy were also excluded. Exclusions were identified in 2 stages: first, through structured electronic health record (EHR) data review, and second, through manual chart review (see Figure 1 for Study Flow Diagram). Inclusion and exclusion criteria are listed in Table S1.
Identification of CAC
After the initial exclusions using structured EHR data, the DL-CAC algorithm identified potentially eligible patients by using the most recent nongated chest CT scan between 2014 and 2019. The algorithm calculated the DL-CAC score in Agatston units; patients with a DL-CAC score of 0 were excluded. After manual chart review of exclusion criteria, a radiologist reviewed each chest CT scan to confirm the presence of CAC.
Notification
Eligible patients with confirmed CAC were randomly assigned 1:1 to notification or usual care through permuted block randomization using an online randomization module by study personnel. Patients randomly assigned to notification first had an EHR message sent to their primary care clinician or endocrinologist. The clinician message (Figure S1) notified them regarding the presence of CAC on the previous nongated chest CT. The letter included an axial image of the patient’s chest CT scan with a circle around the CAC and a reference to the 2018 American College of Cardiology/American Heart Association Guidelines on Management of Blood Cholesterol, which included a IIa recommendation for statin treatment of patients with CAC.7 The letter to clinicians stated that their patient would be contacted in 2 weeks with the same information unless they thought patient notification would be clinically inappropriate and recommended against it. Before the start of the project, we held educational sessions regarding CAC with primary care leadership and a subset of primary care clinicians in the Stanford Health Care System.
Two weeks after clinician notification, messages were sent to patients through the EHR patient portal. These messages (Figure S2) noted the increased cardiovascular risk associated with the presence of CAC, included the same CT image with CAC encircled as in the clinician notification, and recommended discussion of risk-reducing interventions with their clinician, including statin use. The notification did not include the DL-CAC score. If patients did not open their message within 2 weeks, we sent the same letter by postal mail.
The EHR was reviewed for documentation of any discussions regarding incidental CAC or statins. For patients without any discussion with their care team or prescription of a statin within 3 months, we resent messages once to both the clinician and patient.
This project used an adaptive 2-stage design. In the first stage, 50 patients were randomly assigned to notification versus usual care. In the second stage, the remainder of the cohort was randomly assigned. This allowed potential modification to the protocol on the basis of feedback from patients and clinicians. There were no modifications to the intervention after the first stage. Therefore, both stages were pooled for the analysis.
If the intervention increased statin rates at 6 months and was deemed acceptable by primary care at the end of this project, we planned to also send notifications to the usual care arm after the 6-month follow-up.
Outcomes
The primary outcome was the statin prescription rate within 6 months of randomization. We evaluated statin prescription rates at 6 weeks, 3 months, and 6 months after randomization. As secondary outcomes, we captured discussions of statin therapy, aspirin rates, antihypertensive therapy rates, individual biological cardiovascular risk factors, and cardiovascular resource use. The biological cardiovascular risk factors extracted from the EHR were last systolic blood pressure, lipid levels (total cholesterol, low-density lipoprotein cholesterol, high-density lipoprotein cholesterol, and triglycerides), hemoglobin A1c, and body mass index during follow-up. Using lipid measurements, systolic blood pressure, and antihypertensive therapy rates during follow-up, we calculated the 10-year risk of ASCVD events using the American College of Cardiology/American Heart Association Pooled Cohort Equations. For resource use, we captured the number of primary care encounters (in-person or telehealth), total cardiology encounters, new cardiology encounters, and coronary artery disease (CAD) testing. CAD testing included coronary CT angiograms, CAC scans, stress tests, and invasive coronary angiograms. We evaluated diagnostic testing as a pooled composite outcome and as individual outcomes. All outcomes were assessed through the EHR by a nonblinded investigator (S.N.).
Statistical Analysis
The power analysis found that 150 patients would provide 87% power for a 20% absolute difference in statin prescription rates. All analyses were conducted with the intention-to-treat principle. The primary analysis compared statin prescription rates at 6 months across arms using the Fisher exact test. The primary outcome was also compared across prespecified subgroups: above and below median age, sex, ethnicity and race, low-density lipoprotein cholesterol level >100 mg/dL or ≤100 mg/dL, time since last primary care/endocrinology visit stratified at the median, antihypertensive medication use, and the DL-CAC score stratified at the study median. Heterogeneity in the effect across subgroups was tested using the Tarone test.
For secondary outcomes, we compared continuous variables (laboratory values, systolic blood pressure, and 10-year ASCVD risk) using analysis of covariance with adjustment for baseline value and age. For binary outcomes (laboratory testing, CAD testing, and new cardiology encounters), we used logistic regression with adjustment for age. For the ordinal resource use outcomes (number of primary care encounters, cardiology encounters, and blood pressure medications), we used negative binomial regression with adjustment for age and baseline frequency (eg, number of cardiology encounters in the previous year). Adjustment variables were identified a priori. We repeated all analyses for secondary outcomes without adjustment.
There were missing values for baseline and follow-up laboratory values, baseline and follow-up 10-year ASCVD risk, and follow-up body mass index. For missing data, we performed a complete case analysis and 2 imputation analyses: last observation carried forward and multiple imputation with chained equations. With multiple imputation with chained equations, we imputed 100 data sets using missing variables, randomization allocation, baseline values, and comorbidities. Further statistical analysis details are included in the Supplemental Material.
Statistical significance was evaluated at a 2-sided P value threshold of 0.05. All statistical analyses were performed in STATA v16 (StataCorp LLC).
Results
There were 2113 patients with nongated, noncontrast chest CT scans between 2014 and 2019 after excluding patients with previous diagnoses of ASCVD or dementia, statin prescription, or coronary angiography (Figure 1). In this cohort, 424 (20.1%) patients were identified as having CAC on nongated chest CT scan by the DL-CAC algorithm. Of these, an additional 230 patients were excluded on the basis of manual chart review (non-Stanford primary care [n=58], previous statin [n=48], metastatic cancer or nonmetastatic cancer receiving treatment [n=38], death [n=25], hospice/advanced disease [n=19], non-English speaking [n=18], ASCVD/previous coronary testing [n=14], and limited Stanford follow-up [n=10]). Of the remaining 194 patients, a radiologist confirmed CAC among 173 (89.2%) patients, and these patients were randomly assigned between March 30, 2021, and June 7, 2021.
Baseline Characteristics
Patient characteristics were balanced across both arms (Table 1). The median patient age was 70.8 years (interquartile range, 64.0–75.8). The median 10-year pooled cohort equation risk for an ASCVD event was 14.6% (interquartile range, 8.7%–27.7%) among the 149 patients for whom the risk could be calculated. The 10-year risk was ≥7.5% for 94.0% of patients (140/149). The median time from chest CT scan to notification was 857 days (interquartile range, 641–1069 days). The most common indication for the CT scan was to evaluate a pulmonary nodule (63/173; 36.4%), followed by lung cancer screening (22/173; 12.7%). Most radiology reports noted CAC in the body of the report (146/173; 84.4%), but only 3 reports (1.7%) noted CAC in the final impression.
| Characteristics | Notification arm (n=86) | Usual care arm (n=87) |
|---|---|---|
| Age | 70.9 (64.0–75.8) | 70.8 (64.0–76.0) |
| Female sex, n (%) | 44 (51) | 47 (54) |
| Race, n (%) | ||
| Asian | 9 (10) | 11 (13) |
| Black | 4 (5) | 3 (3) |
| White | 65 (76) | 67 (77) |
| Other | 7 (8) | 5 (6) |
| Unknown | 1 (1) | 1 (1) |
| Ethnicity, n (%) | ||
| Hispanic | 5 (6) | 6 (7) |
| Non-Hispanic | 80 (93) | 78 (90) |
| Unknown | 1 (1) | 3 (3) |
| Clinician, n (%) | ||
| Stanford primary care | 84 (98) | 84 (97) |
| Endocrinology | 2 (2) | 3 (3) |
| Time from primary care/endocrinology visit, days | 115 (44–304) | 170 (43–506) |
| Previous cardiology encounter, n (%) | 6 (7) | 7 (8) |
| Chest computed tomography scan indication, n (%) | ||
| Pulmonary nodule evaluation | 23 (27) | 40 (46) |
| Lung cancer screening | 12 (14) | 10 (11) |
| Parenchymal lung disease evaluation | 4 (5) | 2 (2) |
| Malignant disease evaluation | 7 (8) | 4 (5) |
| Pleural effusion evaluation | 0 (0) | 2 (2) |
| Nonspecific symptom evaluation | 8 (9) | 9 (10) |
| Other indications | 32 (37) | 20 (23) |
| Time from chest computed tomography scan, days | 848 (623–1056) | 882 (646–1105) |
| DL-CAC score | 18 (7–112) | 20 (3–70) |
| DL-CAC >0 to <100, n (%) | 63 (73) | 69 (79) |
| DL-CAC ≥100, n (%) | 23 (27) | 18 (21) |
| CAC description in radiology report, n (%) | ||
| No CAC noted | 6 (7) | 11 (13) |
| Mild | 53 (62) | 52 (60) |
| Moderate | 16 (19) | 15 (17) |
| Severe | 6 (7) | 4 (5) |
| Noted, no qualitative description | 5 (6) | 5 (6) |
| CAC mentioned in report impression, n (%) | 1 (1) | 2 (2) |
| Vital signs | ||
| Body mass index, kg/m2 | 25.3 (22.5–29.4) | 25.9 (22.8–29.2) |
| Systolic blood pressure, mm Hg | 128 (120–142) | 129 (119–146) |
| Diastolic blood pressure, mm Hg | 78 (72–84) | 76 (72–84) |
| Atherosclerotic cardiovascular disease risk, 10-y | 16.1% (8.1%–29.7%) | 13.6% (9.0%–26.5%) |
| Medications, n (%) | ||
| Antihypertensive | 33 (38) | 39 (45) |
| Aspirin | 12 (14) | 12 (14) |
| Comorbid conditions, n (%) | ||
| Atrial fibrillation | 9 (10) | 6 (7) |
| Cancer | 35 (41) | 34 (39) |
| Chronic kidney disease | 14 (16) | 7 (8) |
| Chronic obstructive pulmonary disease | 16 (19) | 20 (23) |
| Connective tissue disease | 12 (14) | 5 (6) |
| Depression | 28 (33) | 19 (22) |
| Diabetes | 14 (16) | 6 (7) |
| Hypertension | 43 (50) | 44 (51) |
| Hypothyroidism | 28 (33) | 17 (20) |
| Liver disease | 14 (16) | 13 (15) |
| Other lung disease | 17 (20) | 16 (18) |
| Smoking, current | 9 (11) | 9 (10) |
| Laboratory values | ||
| Creatinine, mg/dL | 0.9 (0.8–1.0) | 0.8 (0.7–0.9) |
| Missing creatinine, n (%) | 0 (0) | 1 (1) |
| Hemoglobin A1c, % | 5.6 (5.4–5.8) | 5.6 (5.4–5.8) |
| Missing hemoglobin A1c, n (%) | 32 (37) | 32 (37) |
| High-density lipoprotein, mg/dL | 60 (49–77) | 60 (44–79) |
| Missing high-density lipoprotein, n (%) | 9 (10) | 15 (17) |
| LDL, mg/dL | 115 (98–133) | 116 (100–128) |
| LDL <70mg/dL, n (%) | 5 (6) | 4 (5) |
| LDL 70–99 mg/dL, n (%) | 17 (20) | 13 (15) |
| LDL 100–129 mg/dL, n (%) | 30 (35) | 38 (44) |
| LDL ≥130 mg/dL, n (%) | 25 (29) | 17 (20) |
| Missing LDL, n (%) | 9 (10) | 15 (17) |
| Triglyceride, mg/dL | 94 (69–133) | 94 (63–145) |
| Missing triglyceride, n (%) | 9 (10) | 13 (15) |
Continuous variables are displayed as median (interquartile range), and categorical variables are displayed as frequency (percentage).
CAC indicates coronary artery calcium; DL, deep learning; and LDL, low-density lipoprotein.
Notification was sent to all 86 patients in the notification arm after notification of their clinicians. No clinicians objected to notifying their patients. Two endocrinologists requested the clinician notification be sent to the patient’s primary care clinician outside the Stanford Health Care System.
Primary Outcome
The notification arm was more likely to receive a statin prescription than the usual care arm (Table 2). Six months after randomization, statins were prescribed to 51.2% (44/86) of the notification arm versus 6.9% (6/87) of the usual care arm (P<0.001; Figure 2). Among patients prescribed statins in the notification arm, 72.7% (32/44) received a moderate-intensity statin prescription, and 18.2% (8/44) received high-intensity statin therapy.
| Outcomes | Notification arm (n=86) | Usual care (n=87) | P value |
|---|---|---|---|
| Primary outcomes, n (%) | |||
| Statin discussion/prescription, 3 mo | 53 (64.6) | 7 (8.4) | <0.001 |
| Statin discussion/prescription, 6 mo | 67 (77.9) | 10 (12.0) | <0.001 |
| Statin prescription, 6 wk | 7 (35.0) | 0 (0.0) | 0.008 |
| Statin prescription, 3 mo | 32 (39.0) | 4 (4.8) | <0.001 |
| Statin prescription, 6 mo | 44 (51.2) | 6 (6.9) | <0.001 |
| Statin intensity, n (%) | <0.001 | ||
| High intensity | 8 (9.3) | 3 (3.4) | |
| Moderate intensity | 32 (37.2) | 3 (3.4) | |
| Low intensity | 4 (4.7) | 0 (0.0) | |
| No statin | 42 (48.8) | 81 (93.1) | |
| Secondary outcomes*,† | |||
| Aspirin treatment, 6 mo, n (%) | 16 (18.6) | 17 (19.5) | 0.848 |
| New aspirin treatment, 6 mo,‡ n (%) | 7/74 (9.5) | 7/75 (9.3) | 0.879 |
| Number of antihypertensives, 6 mo | 0 (0–1) | 0 (0–2) | 0.985 |
| Hemoglobin A1c measured, follow-up, n (%) | 29 (33.7) | 23 (26.4) | 0.211 |
| Hemoglobin A1c, % (nonmissing), follow-up | 5.7 (0.7) | 5.5 (0.5) | 0.107 |
| Lipids measured, follow-up,§ n (%) | 50 (58.1) | 29 (33.3) | 0.002 |
| LDL, mg/dL (nonmissing), follow-up | 97.2 (30.3) | 115.3 (29.4) | 0.005 |
| LDL <70 mg/dL, n (%) | 9 (10.5) | 2 (2.3) | |
| LDL 70–99 mg/dL, n (%) | 20 (23.3) | 6 (6.9) | |
| LDL 100–129 mg/dL, n (%) | 12 (14.0) | 12 (13.8) | |
| LDL ≥130 mg/dL, n (%) | 9 (10.5) | 9 (10.3) | |
| Missing LDL, n (%) | 36 (41.9) | 58 (66.7%) | |
| High-density lipoprotein, mg/dL (nonmissing), follow-up | 64.2 (21.6) | 61.7 (22.5) | 0.872 |
| Triglycerides, mg/dL (nonmissing), follow-up | 87.1 (40.7) | 123.4 (70.8) | 0.009 |
| Systolic blood pressure measured, follow-up,§, n (%) | 69 (80.2) | 64 (73.6) | 0.287 |
| Systolic blood pressure, mm Hg (nonmissing), follow-up | 131.3 (17.4) | 128.9 (15.0) | 0.374 |
| Body mass index measured, follow-up,§ n (%) | 66 (76.7) | 63 (72.4) | 0.486 |
| Body mass index, kg/m2 (nonmissing), follow-up | 25.5 (5.1) | 26.7 (5.6) | 0.630 |
Continuous variables are displayed as median (interquartile range), and categorical variables are displayed as frequency (percentage).
LDL indicates low-density lipoprotein.
*
For secondary outcomes of laboratory values, vital signs, antihypertensives, and aspirin treatment, statistical testing is adjusted for baseline value and age. For laboratory testing, statistical testing is adjusted for age. Results with imputation through last observation carried forward and multiple imputation available in Table S2.
†
Aspirin and number of antihypertensives are on the basis of assessment 6 months after notification. All other outcomes are based on last assessment during the 6-month follow-up period.
‡
New aspirin treatment among those not on treatment at baseline.
§
Defined as a measurement during the 6-month follow-up period.
Figure 3 displays statin rates stratified by patient characteristics. There was no significant effect modification across the prespecified subgroups (Table S2). Statins were prescribed to 59.1% of women in the notification group compared with 42.9% of men (P=0.108 for difference in the effect of notification across sex).
The notification arm was more likely to have discussions about statin therapy with clinicians. Among patients in the notification arm, 77.9% (67/86) had a documented statin discussion or new statin prescription compared with 12.0% (10/87) in the usual care arm (P<0.001).
Secondary Clinical Outcomes
At 6 months after randomization, the aspirin treatment rate was similar across notification and usual care arms (18.6% versus 19.5%, P=0.848, respectively; Table 2). There was also no significant difference in the change in number of antihypertensive medications at 6 months.
Lipid levels were checked during follow-up among 58.1% (50/86) of the notification arm compared with 33.3% (29/87) of the usual care arm (P=0.002). Among those with repeat lipid testing, low-density lipoprotein cholesterol levels were lower in the notification arm versus usual care (97.2 mg/dL [SD, 30.3] versus 115.3 mg/dL [SD, 29.4], P=0.005, respectively). These results were consistent after imputing missing values (Table S3). There was no significant difference between groups in the change in hemoglobin A1c levels or systolic blood pressure during follow-up. The results were similar without adjustment for age (Table S4).
Health Care Utilization
During the 6-month follow-up period, there were more primary care encounters per patient in the notification arm than in the usual care arm (2.2 [SD, 2.2] versus 1.4 [SD, 1.6]; P=0.011). There was no significant difference in the number of cardiology encounters per patient across arms, but there were more patients with new cardiology encounters in the notification arm (16.3% versus 4.6%; P=0.015).
There was an increase in CAD testing in the notification arm (Table 3). In the notification arm, 15.1% (13/86) underwent testing for coronary artery disease compared with 2.3% (2/87) in the usual care arm (P=0.008). The largest absolute difference was in noninvasive stress testing (11.6% versus 2.3%; P=0.018). No patients underwent invasive coronary angiography during the 6-month follow-up. The results were similar without adjustment (Table S5).
| Health care use | Notification arm (n=86) | Usual care (n=87) | P value |
|---|---|---|---|
| Primary care/endocrinology encounters, count per patient† | 2.2 (2.2) | 1.4 (1.6) | 0.011 |
| Cardiology encounters, count per patient | 0.4 (0.8) | 0.2 (1.0) | 0.143 |
| New cardiology encounters, No. of patients (%) | 14 (16.3) | 4 (4.6) | 0.015 |
| Coronary artery disease testing, No. of patients (%)‡ | 13 (15.1) | 2 (2.3) | 0.008 |
| ECG-gated coronary artery calcium scans, No. of patients (%) | 3 (3.5) | 0 (0.0) | 0.121 |
| Coronary computed tomography angiography, No. of patients (%) | 1 (1.2) | 0 (0.0) | 0.497 |
| Invasive coronary angiography, No. of patients (%) | 0 (0.0) | 0 (0.0) | – |
| Stress tests, No. of patients (%) | 10 (11.6) | 2 (2.3) | 0.018 |
| Resting echocardiograms, patients (%) | 5 (5.8) | 7 (8.0) | 0.766 |
Continuous variables are displayed as median (interquartile range), and categorical variables are displayed as frequency (percentage).
*
The number of primary care/endocrinology visits and number of cardiology encounters were adjusted for baseline frequency and age. The new cardiology encounter and coronary artery disease testing outcomes were adjusted for baseline age. Unadjusted analyses are available in Table S4.
†
Primary care encounters for patients with Stanford primary care clinicians; primary care and endocrinology encounters for patients without Stanford primary care clinicians.
‡
Coronary artery disease testing includes ECG-gated coronary artery calcium scans, coronary computed tomography angiography, invasive coronary angiography, and stress tests (eg, echocardiographic, nuclear, or treadmill stress tests).
Discussion
In this randomized quality improvement project, notifying clinicians and patients about the presence of incidental CAC on previously performed nongated chest CT scans led to a significant increase in statin prescription rates. The effect of notification was consistent across patient subgroups. The novel DL-CAC algorithm allowed rapid screening of thousands of previous CT scans for the presence of CAC. This demonstrates the potential for using a deep learning algorithm to perform opportunistic screening for incidental CAC in the millions of nongated chest CT scans performed annually for other reasons.8
A general concern with diagnostic imaging is incidental findings.11,12 CAC is a unique incidental finding because it can be leveraged to improve population health. The presence or absence of CAC improves risk assessment for ASCVD,8,13–17 a high-prevalence, preventable condition. At present, most nongated chest CT reports do not focus on the presence or severity of CAC.18–20 CAC was mentioned in the initial CT report for 90% of our cohort, but CAC was included in the final impression in only 2% of reports, and none of the patients was taking a statin. This suggests that identifying CAC in the radiology report may not be as effective at promoting statin therapy as targeted clinician and patient notifications tied to actionable recommendations. Extracting and disseminating this valuable, often unused finding, with actionable information about its significance could promote implementation of lifestyle and pharmacological interventions that substantially reduce cardiovascular morbidity.21
There is increasing interest in leveraging imaging data that is incidental to the clinical indication of the test to better understand a patient’s risk of future events.22 This has been commonly termed opportunistic screening, which contrasts with systematic screening of a segment of the population with a test such as mammography. Given that the imaging has already been performed, the cost of extracting these data is low. However, there is still an impetus to demonstrate the potential value of the extracted data. This project is an example of how opportunistic screening with previously performed imaging can be leveraged to improve quality of care.
Notifying patients regarding CAC may have multiple potential benefits. Not only does CAC improve risk estimation above existing risk equations, but the image of calcification also has unique motivating effects. The power of visualizing one’s atherosclerosis has also been demonstrated for carotid disease.23 We leveraged that power by providing a personalized image of each patient’s CAC. Although 94% of our cohort met guideline criteria for taking a statin on the basis of their 10-year ASCVD risk, none of these patients were on statins at baseline. The notification and visual demonstration of patient-specific CAC led to a substantial increase in statin rates compared with no notification. These results are consistent, overall, with the EISNER trial (Early Identification of Subclinical Atherosclerosis Using Non-Invasive Imaging Research), in which patients with visualized CAC had improvements in risk factor control and statin adherence compared with those without CAC.24 The DANCAVAS trial (the Danish Cardiovascular Screening) further strengthened the evidence for CAC testing.25 Among male participants randomly assigned to a cardiovascular screening invitation, the most frequent positive finding was an elevated CAC score. In this trial, screening led to an increase in lipid-lowering drugs and antiplatelets and, among those between 65 and 69 years of age, a reduction in all-cause mortality. Our opportunistic screening approach is distinct from EISNER and DANCAVAS by leveraging imaging data that already exist.
Our project does not separate out the effect of notification alone versus notification with the CAC image. Notifying patients of their elevated risk, on the basis of their 10-year ASCVD risk alone, may have led to an increase in statin rates. However, previous efforts to notify clinicians of elevated cardiovascular risk on the basis of risk equations have traditionally demonstrated smaller effect sizes.26–30 Therefore, we tested a novel approach of notifying patients regarding incidental CAC with personalized images of their CAC.
Although opportunistic CAC screening may have multiple potential clinical benefits, there are important considerations before a health system implements such a program. First, there is a fine balance between motivating action and excessively increasing patient anxiety. To ensure access to treatment and counseling, all patients in our project had an active relationship with a health system clinician, and messages were sent to both patients and clinicians. Additional qualitative research with patients and their caregivers regarding appropriate wording of notifications and the appropriateness of direct patient notification will be important. Second, clinicians need to feel responsible for the patient’s cardiovascular health and feel capable of acting on the results. Previous surveys estimated that more than half of referring clinicians are unaware of the significance of CAC.31 We conducted limited educational sessions before launching the project for a subset of clinicians who might be notified. More extensive health system education may have augmented the effectiveness of our intervention. Primary care clinicians raised no concerns about the notifications to their patients, possibly because of the education about the intervention before its launch. Ensuring that clinicians are aligned with the intervention efforts is likely an important element to implementation success.
Our intervention integrated automated processes (the DL-CAC algorithm and the automated EHR review) with manual processes (manual chart review and radiologist review for the presence of CAC). Our project used strict screening to evaluate the efficacy of our approach; future studies can increase automation to facilitate large-scale implementation of opportunistic screening programs. Manual chart review could be largely replaced with expansion of the automated EHR review and simplifying the exclusion criteria. There is a critical balance between considerations of patient autonomy to know their data and avoiding unnecessary anxiety by not screening for CAC in patients unlikely to benefit from preventive therapies (eg, patients in hospice). In addition, manual radiologist review of images may be less critical at higher thresholds in which a small absolute error in the DL-CAC estimate would have a minimal effect on the implications of notification.
Our project was restricted to patients with CAC who had no known ASCVD and were statin naive. We therefore excluded patients with ASCVD who already had a class I guideline recommendation for statin therapy.7 We also applied additional exclusion criteria, such as requiring a Stanford primary care clinician or endocrinologist, for this proof-of-concept study. However, this limited the total eligible population for notification substantially. In future studies, the population eligible for opportunistic screening and notification will be expanded. Opportunistic screening could be broadened to include patients with known ASCVD, given that nearly 50% are not on statin therapy.32 Even for patients on a statin, identification of CAC may lead to dose intensification, improved medication adherence, or lifestyle changes. Increasing the eligible population may reduce the per-patient effect of notification on statin rates; for example, patients without a primary care clinician in the health care system may be less likely to receive statin therapy after notification. However, even if the per-patient effect is smaller, expanding the intervention would still increase the population-level benefit at a low incremental cost.
Our approach used the DL-CAC algorithm retrospectively and had a radiologist confirm the binary presence of CAC. The algorithm could also be applied prospectively to assist radiologists in consistently identifying and quantifying CAC. Quantification of CAC severity may further improve the effect of notification, especially among those at highest risk. Future work will include expanding the DL-CAC algorithm to apply to low-dose chest CT scans used for lung cancer screening and contrast chest CT scans. Applying this approach to broader populations and more CT scans can potentiate the potential effect of opportunistic CAC screening but will require further evaluation.
On average, the DL-CAC scores in our cohort were low. However, the presence of any CAC increases risk compared with patients without CAC; for this reason, the American College of Cardiology/American Heart Association 2018 Guidelines on Blood Cholesterol have a IIa recommendation for considering statin therapy among patients ≥55 years of age and for patients with a CAC score ≥75th percentile for their age, sex, and race. For a 54-year-old woman, a CAC score of 1 would be ≥75th percentile.33 We did not observe differences in the effect of notification on statin rates between those with higher versus lower CAC scores, but our CAC range was relatively narrow. Notifying patients with higher CAC scores and more pronounced calcification images may lead to larger increases in statin prescription rates.
Notifications also increased clinical encounters and testing. Although the increase in primary care encounters and lipid testing are expected and potentially desirable, the benefit of increased cardiology encounters and CAD testing is unclear. There may also be a distinction between dedicated CAC scans, which could be used to validate the results of the DL-CAC algorithm, with noninvasive stress testing and coronary CT angiograms used to detect obstructive coronary disease but not recommended by guidelines in an asymptomatic cohort. Our results contrast with the EISNER trial, which found no significant increase in testing among the cohort randomly assigned to a CAC scan.24 These differences may be explained by the differences between unexpected opportunistic screening and patients being actively enrolled and educated in advance of their gated CAC scan as part of a systematic screening program. Additional education for clinicians and patients may mitigate the observed increase in testing rates. We did not collect data regarding the patient experience and any potential anxiety of those notified. Further research regarding the effect on both health care resource utilization and the patient experience will be important steps before scaling this intervention.
Limitations
There are important limitations to our intervention. First, although we demonstrated a significant increase in statin rates from notifying patients about their subclinical atherosclerosis, the effect on clinical outcomes among this select cohort is unknown. Future studies to demonstrate whether notification reduces ASCVD events would be valuable. Second, this study was performed at a single center; the generalizability to other health systems or to patients who did not meet our inclusion criteria is uncertain. Third, data on persistence of statin therapy beyond 6 months is critical to better understanding the effect of the intervention. We are continuing to follow patients and will plan to subsequently report outcomes at 12 months, including statin treatment and downstream testing rates. Fourth, our intervention notified both clinicians and patients to facilitate shared decision-making; we did not investigate the effect of notifying only the clinician or only the patient. Last, the investigators were not blinded to allocation when evaluating outcomes.
Conclusion
Notifying clinicians and their patients about the presence of incidental CAC detected on previously acquired nongated chest CT scans led to a significant increase in statin prescription rates compared with no notification. These findings illustrate the opportunity to use an automated method to screen nongated chest CT scans to identify patients with subclinical coronary atherosclerosis and the power of notification to motivate them and their clinicians to initiate preventive interventions.
Article Information
Supplemental Material
Supplemental Methods
Tables S1–S5
Figures S1 and S2
Acknowledgments
The authors would like to thank the patients participating in the NOTIFY-1 project and Stanford primary care clinicians.
Footnote
Nonstandard Abbreviations and Acronyms
- ASCVD
- atherosclerotic cardiovascular disease
- CAC
- coronary artery calcium
- CAD
- coronary artery disease
- CT
- computed tomography
- DL-CAC
- deep learning based coronary artery calcium
- EHR
- electronic health record
Supplemental Material
References
1.
Tsao CW, Aday AW, Almarzooq ZI, Alonso A, Beaton AZ, Bittencourt MS, Boehme AK, Buxton AE, Carson AP, Commodore-Mensah Y, et al. Heart disease and stroke statistics-2022 update: a report from the American Heart Association. Circulation. 2022;145:e153–e639. doi: 10.1161/CIR.0000000000001052
2.
Metser G, Bradley C, Moise N, Liyanage-Don N, Kronish I, Ye S. Gaps and disparities in primary prevention statin prescription during outpatient care. Am J Cardiol. 2021;161:36–41. doi: 10.1016/j.amjcard.2021.08.070
3.
Nanna MG, Navar AM, Wang TY, Mi X, Virani SS, Louie MJ, Lee LV, Goldberg AC, Roger VL, Robinson J, et al. Statin use and adverse effects among adults >75 years of age: insights from the Patient and Provider Assessment of Lipid Management (PALM) registry. J Am Heart Assoc. 2018;7:e008546. doi: 10.1161/JAHA.118.008546
4.
Saeed A, Zhu J, Thoma F, Marroquin O, Aiyer A, Agarwala A, Virani SS, Gulati M, Lee JS, Reis S, et al. Cardiovascular disease risk-based statin utilization and associated outcomes in a primary prevention cohort: insights from a large health care network. Circ Cardiovasc Qual Outcomes. 2021;14:e007485. doi: 10.1161/JAHA.118.008546
5.
Clough JD, Martin SS, Navar AM, Lin L, Hardy NC, Rogers U, Curtis LH. Association of primary care providers’ beliefs of statins for primary prevention and statin prescription. J Am Heart Assoc. 2019;8:e010241. doi: 10.1161/JAHA.118.010241
6.
Detrano R, Guerci AD, Carr JJ, Bild DE, Burke G, Folsom AR, Liu K, Shea S, Szklo M, Bluemke DA, et al. Coronary calcium as a predictor of coronary events in four racial or ethnic groups. N Engl J Med. 2008;358:1336–1345. doi: 10.1056/nejmoa072100
7.
Grundy SM, Stone NJ, Bailey AL, Beam C, Birtcher KK, Blumenthal RS, Braun LT, de Ferranti S, Faiella-Tommasino J, Forman DE, et al. 2018 AHA/ACC/AACVPR/AAPA/ABC/ACPM/ADA/AGS/APhA/ASPC/NLA/PCNA Guideline on the management of blood cholesterol: a report of the American College of Cardiology/American Heart Association Task Force on Clinical Practice Guidelines. J Am Coll Cardiol. 2019;73:e285–e350. doi: 10.1016/j.jacc.2018.11.003
8.
Budoff MJ, Lutz SM, Kinney GL, Young KA, Hokanson JE, Barr RG, Steiner R, Nath H, Lopez-Garcia C, Duca LM, et al. Coronary artery calcium on noncontrast thoracic computerized tomography scans and all-cause mortality. Circulation. 2018;138:2437–2438. doi: 10.1161/CIRCULATIONAHA.118.036835
9.
Eng D, Chute C, Khandwala N, Rajpurkar P, Long J, Shleifer S, Khalaf MH, Sandhu AT, Rodriguez F, Maron DJ, et al. Automated coronary calcium scoring using deep learning with multicenter external validation. NPJ Digit Med. 2021;4:88. doi: 10.1038/s41746-021-00460-1
11.
O’Sullivan JW, Muntinga T, Grigg S, Ioannidis JPA. Prevalence and outcomes of incidental imaging findings: umbrella review. BMJ. 2018;361:k2387. doi: 10.1136/bmj.k2387
12.
Ganguli I, Simpkin AL, Lupo C, Weissman A, Mainor AJ, Orav EJ, Rosenthal MB, Colla CH, Sequist TD. Cascades of care after incidental findings in a US National Survey of Physicians. JAMA Netw Open. 2019;2:e1913325. doi: 10.1001/jamanetworkopen.2019.13325
13.
Chiles C, Duan F, Gladish GW, Ravenel JG, Baginski SG, Snyder BS, DeMello S, Desjardins SS, Munden RF, NLST Study TeamAssociation of coronary artery calcification and mortality in the National Lung Screening Trial: a comparison of three scoring methods. Radiology. 2015;276:82–90. doi: 10.1148/radiol.15142062
14.
Haller C, Vandehei A, Fisher R, Boster J, Shipley B, Kaatz C, Harris J, Shin SR, Townsend L, Rouse J, et al. Incidence and implication of coronary artery calcium on non-gated chest computed tomography scans: a large observational cohort. Cureus. 2019;11:e6218. doi: 10.7759/cureus.6218
15.
Hughes-Austin JM, Dominguez A, Allison MA, Wassel CL, Rifkin DE, Morgan CG, Daniels MR, Ikram U, Knox JB, Wright CM, et al. Relationship of coronary calcium on standard chest CT scans with mortality. JACC Cardiovasc Imaging. 2016;9:152–159. doi: 10.1016/j.jcmg.2015.06.030
16.
Mets OM, Vliegenthart R, Gondrie MJ, Viergever MA, Oudkerk M, de Koning HJ, Mali WP, Prokop M, van Klaveren RJ, van der Graaf Y, et al. Lung cancer screening CT-based prediction of cardiovascular events. JACC Cardiovasc Imaging. 2013;6:899–907. doi: 10.1016/j.jcmg.2013.02.008
17.
Xie X, Zhao Y, de Bock GH, de Jong PA, Mali WP, Oudkerk M, Vliegenthart R. Validation and prognosis of coronary artery calcium scoring in nontriggered thoracic computed tomography: systematic review and meta-analysis. Circ Cardiovasc Imaging. 2013;6:514–521. doi: 10.1161/CIRCIMAGING.113.000092
18.
Balakrishnan R, Nguyen B, Raad R, Donnino R, Naidich DP, Jacobs JE, Reynolds HR. Coronary artery calcification is common on nongated chest computed tomography imaging. Clin Cardiol. 2017;40:498–502. doi: 10.1002/clc.22685
19.
Kirsch J, Martinez F, Lopez D, Novaro GM, Asher CR. National trends among radiologists in reporting coronary artery calcium in non-gated chest computed tomography. Int J Cardiovasc Imaging. 2017;33:251–257. doi: 10.1007/s10554-016-0986-x
20.
Phillips WJ, Johnson C, Law A, Turek M, Small AR, Inacio JR, Dent S, Ruddy T, Beanlands RS, Chow BJW, et al. Reporting of coronary artery calcification on chest CT studies in breast cancer patients at high risk of cancer therapy related cardiac events. Int J Cardiol Heart Vasc. 2018;18:12–16. doi: 10.1016/j.ijcha.2018.02.001
21.
Arnett DK, Blumenthal RS, Albert MA, Buroker AB, Goldberger ZD, Hahn EJ, Himmelfarb CD, Khera A, Lloyd-Jones D, McEvoy JW, et al. 2019 ACC/AHA guideline on the primary prevention of cardiovascular disease: executive summary: a report of the American College of Cardiology/American Heart Association Task Force on Clinical Practice Guidelines. J Am Coll Cardiol. 2019;74:1376–1414. doi: 10.1016/j.jacc.2019.03.009
22.
Pickhardt PJ, Graffy PM, Perez AA, Lubner MG, Elton DC, Summers RM. Opportunistic screening at abdominal CT: use of automated body composition biomarkers for added cardiometabolic value. Radiographics. 2021;41:524–542. doi: 10.1148/rg.2021200056
23.
Naslund U, Ng N, Lundgren A, Fharm E, Gronlund C, Johansson H, Lindahl B, Lindahl B, Lindvall K, Nilsson SK, et al; VIPVIZA trial groupVisualization of asymptomatic atherosclerotic disease for optimum cardiovascular prevention (VIPVIZA): a pragmatic, open-label, randomised controlled trial. Lancet. 2019;393:133–142. doi: 10.1016/S0140-6736(18)32818-6
24.
Rozanski A, Gransar H, Shaw LJ, Kim J, Miranda-Peats L, Wong ND, Rana JS, Orakzai R, Hayes SW, Friedman JD, et al. Impact of coronary artery calcium scanning on coronary risk factors and downstream testing the EISNER (Early Identification of Subclinical Atherosclerosis by Noninvasive Imaging Research) prospective randomized trial. J Am Coll Cardiol. 2011;57:1622–1632. doi: 10.1016/j.jacc.2011.01.019
25.
Lindholt JS, Søgaard R, Rasmussen LM, Mejldal A, Lambrechtsen J, Steffensen FH, Frost L, Egstrup K, Urbonaviciene G, Busk M, et al. Five-year outcomes of the Danish Cardiovascular Screening (DANCAVAS) Trial. N Engl J Med. 2022;387:1385–1394. doi: 10.1056/nejmoa2208681
26.
Benner JS, Erhardt L, Flammer M, Moller RA, Rajicic N, Changela K, Yunis C, Cherry SB, Gaciong Z, Johnson ES, et al; REACH OUT InvestigatorsA novel program to evaluate and communicate 10-year risk of CHD reduces predicted risk and improves patients’ modifiable risk factor profile. Int J Clin Pract. 2008;62:1484–1498. doi: 10.1111/j.1742-1241.2008.01872.x
27.
Jacobson TA, Gutkin S, Harper CR. Effects of a global risk educational tool on primary coronary prevention: the Atherosclerosis Assessment Via Total Risk (AVIATOR) study. Curr Med Res Opin. 2006;22:1065–1073. doi: 10.1185/030079906x104605
28.
Karmali KN, Persell SD, Perel P, Lloyd-Jones DM, Berendsen MA, Huffman MD. Risk scoring for the primary prevention of cardiovascular disease. Cochrane Database Syst Rev. 2017;3:CD006887. doi: 10.1002/14651858.CD006887.pub4
29.
Persell SD, Brown T, Lee JY, Shah S, Henley E, Long T, Luther S, Lloyd-Jones DM, Jean-Jacques M, Kandula NR, et al. Individualized risk communication and outreach for primary cardiovascular disease prevention in community health centers: randomized trial. Circ Cardiovasc Qual Outcomes. 2015;8:560–566. doi: 10.1161/CIRCOUTCOMES.115.001723
30.
Persell SD, Lloyd-Jones DM, Friesema EM, Cooper AJ, Baker DW. Electronic health record-based patient identification and individualized mailed outreach for primary cardiovascular disease prevention: a cluster randomized trial. J Gen Intern Med. 2013;28:554–560. doi: 10.1007/s11606-012-2268-1
31.
Uretsky S, Chokshi N, Kobrinski T, Agarwal SK, Po JR, Awan H, Jagarlamudi A, Gudiwada SP, D’Avino RC, Rozanski A. The interplay of physician awareness and reporting of incidentally found coronary artery calcium on the clinical management of patients who underwent noncontrast chest computed tomography. Am J Cardiol. 2015;115:1513–1517. doi: 10.1016/j.amjcard.2015.02.051
32.
Nelson AJ, Haynes K, Shambhu S, Eapen Z, Cziraky MJ, Nanna MG, Calvert SB, Gallagher K, Pagidipati NJ, Granger CB. High-intensity statin use among patients with atherosclerosis in the U.S. J Am Coll Cardiol. 2022;79:1802–1813. doi: 10.1016/j.jacc.2022.02.048
33.
McClelland RL, Chung H, Detrano R, Post W, Kronmal RA. Distribution of coronary artery calcium by race, gender, and age: results from the Multi-Ethnic Study of Atherosclerosis (MESA). Circulation. 2006;113:30–37. doi: 10.1161/CIRCULATIONAHA.105.580696