Deep learning for automatic calcium scoring in population based cardiovascular screening

Background

High volumes of standardized coronary artery calcium (CAC) scans are generated in screening that need to be scored accurately and efficiently to risk stratify individuals.

Purpose

To evaluate the performance of deep learning based software for automatic coronary calcium scoring in a screening setting.

Methods

Participants from the Robinsca trial that underwent low-dose ECG-triggered cardiac CT for calcium scoring were included. CAC was measured with fully automated deep learning prototype and compared to the original manual assessment of the Robinsca trial. Detection rate, positive Agatston score and risk categorization (0–99, 100–399, ≥400) were compared using McNemar test, ICC, and Cohen's kappa. False negative (FN), false positive (FP) rate and diagnostic accuracy were determined for preventive treatment initiation (cut-off ≥100 AU).

Results

In total, 997 participants were included between December 2015 and June 2016. Median age was 61.0 y (IQR: 11.0) and 54.4% was male. A high agreement for detection was found between deep learning based and manual scoring, κ=0.87 (95% CI 0.85–0.89). Median Agatston score was 58.4 (IQR: 12.3–200.2) and 61.2 (IQR: 13.9–212.9) for deep learning based and manual assessment respectively, ICC was 0.958 (95% CI 0.951–0.964). Reclassification rate was 2.0%, with a very high agreement with κ=0.960 (95% CI: 0.943–0.997), p<0.001. FN rate was 0.7% and FP rate was 0.1% and diagnostic accuracy was 99.2% for initiation of preventive treatment.

Conclusion

Deep learning based software for automatic CAC scoring can be used in a cardiovascular CT screening setting with high accuracy for risk categorization and initiation of preventive treatment.

Funding Acknowledgement

Type of funding sources: Public grant(s) – EU funding. Main funding source(s): Robinsca trial was supported by advanced grant of European Research Council

P3397Risk Or Benefit IN Screening for CArdiovascular disease (ROBINSCA): results from screening for a high cardiovascular disease risk by using a risk prediction model or coronary artery calcium scoring

Introduction

The ROBINSCA (Risk Or Benefit IN Screening for CArdiovascular disease) trial is a large-scale population-based randomized controlled screening trial with the aim to investigate whether screening for a high risk of cardiovascular disease (CVD) by means of either the Systematic COronary Risk Evaluation (SCORE) model or coronary artery calcium (CAC) scoring followed by preventive treatment is effective in reducing morbidity and mortality from coronary heart disease (CHD). This study shows the results of the CVD risks as assessed by the two screening tools.

Methods

Based on the Dutch population registry, 394,058 men aged 45–74 years and women aged 55–74 years received an information brochure, an invitation to participate in the trial, a baseline questionnaire with waist circumference tape and an informed consent form. Eligible individuals with an expected high CVD risk were randomized (1:1:1) into a control arm (n=14,519), intervention arm A (n=14,478) or intervention arm B (n=14,450). In the control arm, usual care was continued. In intervention arm A, participants were screened for a high risk of CVD using the SCORE model based on traditional risk factors. In intervention arm B, CAC scoring after computed tomography scanning was used for screening. After screening en risk communication, preventive treatment according to the Dutch guidelines is advised for high risk persons.

Results

Screening uptake was 84.2% in intervention arm A and 89.6% in intervention arm B. Of the screened participants, 48.7% was female, median age at screening was 62 (Interquartile Range 10), 35.2% was high educated, 19.6% was baseline smoker and 41.4% had a positive family history of myocardial infarction. The assessed CVD risk status according to SCORE screening was stratified into three risk categories; 45.1% was at low risk (SCORE<10%), 26.5% was at intermediate risk (SCORE 10–20%), and 28.4% was at high risk (SCORE ≥20%). According to CAC screening, 76.0% was at low risk (Agatston <100), 15.1% was at high risk (Agatston 100–399), and 8.9% was at very high risk (Agatston ≥400). Associations between baseline variables and increased CVD risk will be analyzed soon and will be available in summer 2019.

Conclusions

Using different screening tools resulted in reclassification of the CVD risk. CAC screening caused a substantial shift to more low risk individuals. This might, when screening is found to be effective, lead to less overtreatment in prevention of CVD events. Future 5-year follow-up data should provide evidence about whether population-based screening with subsequent preventive treatment is (cost-)effective in reducing CHD-related morbidity and mortality.

Informed decision-making about the fetal anomaly scan: what knowledge is relevant?

OBJECTIVES

This study had two objectives. The first was to determine the contents of relevant knowledge needed for informed decision-making (IDM) in second-trimester ultrasound screening for fetal anomalies, with the goal of developing a knowledge measure for use in large-scale program evaluations. The second was to compare the contents of decision-relevant knowledge for second-trimester ultrasound screening with those for first-trimester screening for Down syndrome using the combined test.

METHODS

A generic list of content domains for knowledge about screening was extracted from the literature. Items reflecting specific knowledge domains for second-trimester ultrasound screening were constructed. An expert group of professionals and pregnant women expressed whether domains and items represented decision-relevant knowledge.

RESULTS

Regarding second-trimester ultrasound screening, the experts scored all knowledge domains as (very) important. The meaning of an abnormal test result, the disorders being screened for, and the purpose of the screening were rated as very important for IDM, along with the voluntary nature of the test. All knowledge domains were included in the final measure. Importance ratings of knowledge domains for first-trimester Down syndrome screening and for second-trimester ultrasound screening were highly correlated (Pearson's r = 0.71). The domain 'consequences of a positive test result' was considered more important in first-trimester Down syndrome screening than in second-trimester ultrasound screening.

CONCLUSIONS

We have developed a knowledge measure for second-trimester ultrasound screening for fetal anomalies for use in routine, large-scale program evaluations.

Psychological risk factors of incidence of breast cancer: a prospective cohort study in Finland

BACKGROUND

Belief that depression and other psychological factors predict breast cancer is common, but there have been few prospective epidemiological studies on this relationship.

METHOD

The relationship between depression, personality traits, illness attitudes, life events and health history, and breast cancer risk was studied in a prospective, 6-9 year follow-up of a cohort study of 10892 Finnish women of 48-50 years of age at the baseline. Cancer cases were obtained from the Cancer Registry of Finland. Multivariate logistic regression analysis was performed controlling for socioeconomic factors, family history of cancer, parity, and health behaviours.

RESULTS

Breast cancer incidence in the cohort was 1.15 times the average in age group 50-59. There was no evidence of depression, trait anxiety, cynical distrust, or coping being significant predictors of breast cancer incidence.

CONCLUSION

In this cohort study with the 6-9 year follow-up, psychological factors such as depression, trait anxiety, cynical distrust, or coping did not increase breast cancer risk.

Including the quality-of-life effects in the evaluation of prostate cancer screening: expert opinions revisited?

OBJECTIVE

To apply the general empirical framework for estimating utilities in cost-utility analysis (CUA) of population-based prostate cancer screening, including an assessment of empirical health status with a classifying measure (e.g. the EQ-5D) and linking these descriptions to utility estimates using the standard preference-based algorithm, combining them with the appropriate duration into quality-adjusted life years, and sensitivity analysis.

MATERIALS AND METHODS

Empirical studies to describe and value the health status effects of prostate cancer screening have been ongoing within the Rotterdam centre of the European Randomised Study on Screening for Prostate Cancer since 1995. The results of these studies, including the screening, the primary treatment phase and advanced disease, will be used in estimating utilities for cost-utility analysis.

RESULTS

Estimation of cost-utility of population-based prostate cancer screening with the results of the three empirical health status studies yielded partly counterintuitive results, underestimating the unfavourable health status effects that are inevitably associated with prostate cancer screening. This may be caused by other than health effects of the screening itself ('process effects') and adaptive changes in perception of their situation in patients after curative primary treatment ('response shift'), among other elements.

CONCLUSIONS

These results prompted us to reconsider the suitability of the general framework of CUA for screening programmes. Possible directions for solutions are indicated.

MISCAN: estimating lead-time and over-detection by simulation

OBJECTIVE

To estimate the mean lead-time and rate of over-detection associated with screening for prostate cancer with prostate-specific antigen.

METHODS

Simulation models, fitted to the results of the Rotterdam section of the European Randomized Study of Screening for Prostate Cancer, were used to predict the mean lead-time and over-detection rate in population-based screening programmes.

RESULTS

The mean lead-time is estimated to be 11-12 years and over-detection to occur in half the cases found by population screening. The estimates are compared with published estimates.

CONCLUSIONS

The effect of lead-time and over-detection on the balance of positive and negative consequences of screening cannot be neglected.

Breast cancer screening; cost-effective in practice?

The main aim of national breast screening is a reduction in breast cancer mortality. The data on the reduction in breast cancer mortality from three (of the five) Swedish trials in particular gave rise to the expectation that the Dutch programme of 2-yearly screening for women aged 50-70 would produce a 16% reduction in the total population. In all likelihood, many of the years of life gained as a result of screening are enjoyed in good health. According to its critics the actual benefit that can be achieved from the national breast cancer screening programmes is overstated. Considerable benefits have recently been demonstrated in England and Wales. However, the fall was so considerable in such a relatively short space of time that screening (started in 1987) was thought to only have played a small part. As far as the Dutch screening programme is concerned it is still too early to reach any conclusions about a possible reduction in mortality. The first short-term results of the screening are favourable and as good as (or better than) expectations. In Swedish regions where mammographic screening was introduced, a 19% reduction in breast cancer mortality can be estimated at population level, and recently a 20% reduction was presented in the UK. In countries where women are expected to make appointments for screening themselves, the attendance figures are significantly lower and the quality of the process as a whole is sometimes poorer. The benefits of breast cancer screening need to be carefully balanced against the burden to women and to the health care system. Mass breast screening requires many resources and will be a costly service. Cost-effectiveness of a breast cancer screening programme can be estimated using a computer model. Published cost-effectiveness ratios may differ tremendously, but are often the result of different types of calculation, time periods considered, including or excluding downstream cost. The approach of simulation and estimation is here the same for all countries. The effects of a breast-screening program depend on many factors, such as the epidemiology of the disease, the health care system, costs of health care, the quality of the screening programme and the attendance rate. The estimated CE-ratio ranges from 2650 euros per life-year gained in Navarra to 9650 in Germany. Although relatively low incidence levels expected, the CE-ratio in Navarra is most favourable probably due to a relatively unfavourable clinical stage distribution before screening and the increasing incidence. The UK has a screening situation that is almost similar with the Netherlands. Therefore, the CE-ratios of both countries are comparable. The differences between countries make it impossible to set up one uniform screening policy. The theoretical outcomes of the benefit that can be achieved are generally from small-scale trials involving a limited number of experts, persons examined, and areas. On a national scale, with hundreds of professional practitioners, it can be expected to be more difficult to attain uniform quality. Continuous quality control, monitoring and evaluation are therefore crucial.

European randomized study of screening for prostate cancer--the Rotterdam pilot studies

Five randomized pilot studies of screening for prostate cancer (PC) have been conducted in the area of Rotterdam from 1991 to 1994. The purpose of these studies was to establish the feasibility of a randomized screening protocol with PC mortality as the major end point in The Netherlands and at a European level. All procedures related to recruitment of participants, to application of the screening tests and to data collection were evaluated. Men (7,200) aged 55-74 years were invited through the Rotterdam Population Registry. The recruitment rate over the 5 pilot studies averaged 38.2% (2,747 men). Recruitment procedures proved to be relevant for establishing higher participation rates (invitation and consent by mail). The screening tests were well accepted and tolerated. The general population-based character of the sample was confirmed by studying symptoms of prostatic disease in participants and in men who refused participation. Data based on one PSA serum determination, rectal examination and transrectal ultrasonography are presented; 204/1,403 men (14.5%) had a positive screening result by either test combination and underwent biopsy. Forty-nine cancers were found in 1,403 men (3.5%); 65% of prostate cancers (17/26) identified in men who eventually underwent radical prostatectomy proved to be locally confined. From the pilot studies, we conclude that a large contribution to a European Randomized Study of Screening for Prostate Cancer (ERSPC) can be made by recruiting about 40,000 men in the area of Rotterdam. The preliminary data suggest that after confirmation of the present data during the first years in the European study, DRE and TRUS can be withheld depending on the PSA result in a large proportion of the screening population.

Nation-wide breast cancer screening in The Netherlands: support for breast-cancer mortality reduction. National Evaluation Team for Breast Cancer Screening (NETB)

The nation-wide 2-yearly breast-cancer screening programme in The Netherlands, for women aged 50-69, started around 1988, and was predicted to result eventually in a 16% reduction in breast-cancer mortality in the total female population. We present the results of screening up to January 1, 1993, and compare these with the predicted results from the cost-effectiveness analysis, on which basis this mortality reduction has been calculated. At least 550,000 women aged 50-69 were invited to screening in 1990-1992, and 75% of these participated. Cancer was suspected from 5,162 examinations and further investigation was therefore required. Excision biopsy was done in 72% of referrals, and 2,515 breast cancers were detected. The results for 404,000 newly invited women compare favourably with expected values (in parentheses): 78% attendance rate (70%), 1.4% screen positive (1.6%), 6.8 cancers detected per 1,000 women screened (6.4) and 38% of these cancers were DCIS or invasive carcinomas smaller than 11 mm in diameter (36%). More data, e.g., on treatment and interval cancers, will follow in the years to come. These first results can be interpreted as strong early signs of a reduction in breast-cancer mortality of at least the predicted size. Screening has sufficiently advanced the diagnosis, as well as or better than expected. Breast cancers diagnosed in this age group without screening are diagnosed at a worse stage than expected. Unfavourable side-effects, especially false-positive referrals, might be kept lower than those reported in other countries.