Abstract 5411: Efficient development of prognostic tests for detecting cancer risk using proteomic technology

Background: Prognostic models for assessing future health outcomes can be developed using time-to-event (also known as “survival”) data. This methodology is ubiquitous in statistical literature and in the analysis of cancer outcomes, but its use in high-dimensional analyses tends to be limited as the methods are difficult to implement in a machine learning environment. Additionally, development of certified prognostic clinical tests using proteomic biomarkers for detecting future cancer risk can be time-consuming, prone to overfitting issues, and difficult to navigate. We demonstrate the utility of combining SomaScan® proteomic data with pipeline machine learning tools and survival analysis methodology to identify powerful and robust LDT-certifiable prognostic tests for assessing future risk of cancer.

Methods: Data pipeline and analysis tools were developed using R. In addition to standard machine learning techniques, statistical methods include elastic net AFT models, subsampling survival techniques, and metrics for assessing predictive survival models. The pipeline takes the analyst from data processing and QC through identification of optimal models for prediction of clinical endpoints, and then through validation on a hold-out test set. The tools include an assessment of model robustness against sample handling issues, longitudinal stability, the impacts of assay noise on model performance, effects of putative interferents, and risk of failure during CLIA validation in the lab. We demonstrate the utility of the tools and methods for development of a lung cancer risk model.

Results: Analysis time for validation of an optimal clinical model was reduced by at least 80%, resulting in the development of 7 LDT-certified tests within 3 years, including a test for lung cancer risk. Inclusion of methods that allow for subsampling and penalized regression using AFT models show improved predictive performance and identification of top features related to clinical endpoints.

Conclusion: Not only are powerful, prognostic tests do-able, but they can be LDT certified in an efficient manner and made to be robust to real-life lab settings. Survival analysis in a machine-learning setting allow us to leverage proteomic technology in new ways, leading to tests that assess future cancer risk, which can be used for precision medicine applications.

Citation Format: Yolanda Hagar, Leigh Alexander, Jessica Chadwick, Gargi Datta, Joe Gogain, Rachel Ostroff, Clare Paterson, Laura Sampson, Caleb Scheidel, Sama Shrestha, Amy Zhang, Michael Hinterberg. Efficient development of prognostic tests for detecting cancer risk using proteomic technology. [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2023; Part 1 (Regular and Invited Abstracts); 2023 Apr 14-19; Orlando, FL. Philadelphia (PA): AACR; Cancer Res 2023;83(7_Suppl):Abstract nr 5411.

Abstract 4361: The plasma proteome as a cardiovascular disease risk assessment tool in cancer survivors

Cardiovascular disease (CVD) is the most common non-cancer cause of death in cancer survivors and there is an unmet clinical need for easy, accurate, and safe CVD prognostic risk-stratification in adult cancer survivors. This study investigated whether a previously validated 27-plasma protein prognostic model for four-year cardiovascular (CV) events could have such a utility. We used the 27-plasma protein model to predict the four-year risk of a CV event (myocardial infarction, stroke, transient ischemic attack, heart failure hospitalization, death) in 906 participants with a prior history or active malignancy of any type of cancer and compared predictive results to follow-up CV outcome data. The participants were from the BASEL VIII or ARIC (visit 3) studies with a medically adjudicated prior diagnosis of cancer. BASEL VIII is an observational cohort study in patients with suspected coronary artery disease. ARIC is a multi-site cohort study funded by the NHLBI, NCI, and NPCR investigating risk factors for CV health. A subset analysis was conducted to assess model performance in participants with no prior history of CVD and those with stable CVD. The 27-plasma protein model accurately stratified participants into 4 distinct and non-overlapping (95% CI) risk bins. The median time to event for all cancer survivors who had an event in this study was 1.3 years. Observed 4-year event rates across the 4 risk bins (low, medium-low, medium-high, and high) were 11.0%, 17.3%, 31.2% and 60.2%, respectively, which were higher than stratified event rates from our previously published metacohort analyses (5.6%, 11.2%, 20.0% and 43.4%, respectively) in participants with elevated CVD risk factors (e.g., prior events, diabetes, kidney disease and suspected coronary artery disease). The plasma protein model accurately predicted 4-year CVD risk with a C-index of 0.71 (0.68, 0.74) and 4-year AUC of 0.74 (0.69, 0.79). Performance of the protein model was comparable between participants with no prior history of CVD (C-Index: 0.69; AUC: 0.71) and stable CVD (C-Index: 0.69; AUC: 0.72), demonstrating the model accurately predicts CV event risk in cancer survivors regardless of cardiovascular history. Cancer survivors in this cohort can be distinguished with 4-year CV event rates as high as 60.2%, underscoring the urgent need for an easy and accurate risk stratification tool for this population. Prognostic protein testing may provide a novel tool for CVD risk assessment in adult cancer survivors.

Citation Format: Emma V. Troth, Matthew Ayala, Jessica Chadwick, Erin Hales, Michael Hinterberg, Jessica N. Kuzma, Clare Paterson, Rachel Ostroff, Joan E. Walter, Christian Mueller, Josef Coresh. The plasma proteome as a cardiovascular disease risk assessment tool in cancer survivors. [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2023; Part 1 (Regular and Invited Abstracts); 2023 Apr 14-19; Orlando, FL. Philadelphia (PA): AACR; Cancer Res 2023;83(7_Suppl):Abstract nr 4361.

The dynamic changes and sex differences of 147 immune-related proteins during acute COVID-19 in 580 individuals

INTRODUCTION

Severe COVID-19 leads to important changes in circulating immune-related proteins. To date it has been difficult to understand their temporal relationship and identify cytokines that are drivers of severe COVID-19 outcomes and underlie differences in outcomes between sexes. Here, we measured 147 immune-related proteins during acute COVID-19 to investigate these questions.

METHODS

We measured circulating protein abundances using the SOMAscan nucleic acid aptamer panel in two large independent hospital-based COVID-19 cohorts in Canada and the United States. We fit generalized additive models with cubic splines from the start of symptom onset to identify protein levels over the first 14 days of infection which were different between severe cases and controls, adjusting for age and sex. Severe cases were defined as individuals with COVID-19 requiring invasive or non-invasive mechanical respiratory support.

RESULTS

580 individuals were included in the analysis. Mean subject age was 64.3 (sd 18.1), and 47% were male. Of the 147 proteins, 69 showed a significant difference between cases and controls (p < 3.4 × 10-4). Three clusters were formed by 108 highly correlated proteins that replicated in both cohorts, making it difficult to determine which proteins have a true causal effect on severe COVID-19. Six proteins showed sex differences in levels over time, of which 3 were also associated with severe COVID-19: CCL26, IL1RL2, and IL3RA, providing insights to better understand the marked differences in outcomes by sex.

CONCLUSIONS

Severe COVID-19 is associated with large changes in 69 immune-related proteins. Further, five proteins were associated with sex differences in outcomes. These results provide direct insights into immune-related proteins that are strongly influenced by severe COVID-19 infection.

Plasma protein patterns as comprehensive indicators of health

Proteins are effector molecules that mediate the functions of genes1,2 and modulate comorbidities3-10, behaviors and drug treatments11. They represent an enormous potential resource for personalized, systemic and data-driven diagnosis, prevention, monitoring and treatment. However, the concept of using plasma proteins for individualized health assessment across many health conditions simultaneously has not been tested. Here, we show that plasma protein expression patterns strongly encode for multiple different health states, future disease risks and lifestyle behaviors. We developed and validated protein-phenotype models for 11 different health indicators: liver fat, kidney filtration, percentage body fat, visceral fat mass, lean body mass, cardiopulmonary fitness, physical activity, alcohol consumption, cigarette smoking, diabetes risk and primary cardiovascular event risk. The analyses were prospectively planned, documented and executed at scale on archived samples and clinical data, with a total of ~85 million protein measurements in 16,894 participants. Our proof-of-concept study demonstrates that protein expression patterns reliably encode for many different health issues, and that large-scale protein scanning12-16 coupled with machine learning is viable for the development and future simultaneous delivery of multiple measures of health. We anticipate that, with further validation and the addition of more protein-phenotype models, this approach could enable a single-source, individualized so-called liquid health check.