Prevalence of COVID-19-associated pneumonia signs on chest computed tomography in cancer patients: the ARILUS study

Andrey Dyachenko, A M Grjibovski, M Bogdanov, D. V. Bogdanov, Ekaterina Nazarova, Anna Meldo, Valeria Chernina, Mikhail Belyaev, Victor А. Gombolevskiy, М. Yu. ValkovSechenov 2025 · DOI

TL;DR

The ARILUS study utilized a multitarget AI algorithm to analyze chest CT scans of over 1,100 cancer patients during the pandemic, revealing that nearly half (48.4%) showed signs of COVID-19-associated pneumonia. The research identifies specific tumor locations—particularly in the lungs, head, and neck—and pre-existing conditions like emphysema and coronary calcification as significant independent risk factors for detectable lung involvement.

What Problem Does This Paper Address?

Cancer patients are often considered a high-risk group for infectious diseases due to immunosuppression from both the disease itself and its treatments. During the COVID-19 pandemic, understanding the true prevalence of the virus in this population was challenging, as many cases were asymptomatic or clinically "silent."

The authors addressed a critical gap: how frequently did cancer patients actually develop COVID-19 pneumonia, and were certain types of cancer or comorbidities more strongly associated with it? By "opportunistically" using routine chest CT scans (originally intended for cancer staging or monitoring) and applying advanced AI, the researchers could detect minimal lung changes that might have otherwise been missed or ignored during a standard clinical review.

Methods

The researchers conducted a cross-sectional study as part of the ARILUS project. They analyzed 1,148 patients at the Arkhangelsk Clinical Oncology Dispensary in Russia between April 2020 and December 2021.

The technical centerpiece was a multitarget AI algorithm developed by IRA Labs. This software didn't just look for pneumonia; it simultaneously quantified:

  • COVID-19 lung involvement: Percentage of lung tissue affected.
  • Pulmonary emphysema: Presence of chronic lung damage.
  • Coronary artery calcification: Measured via the Agatston score, a proxy for cardiovascular health.
  • Bone mineral density: Checking for osteoporosis in the thoracic vertebrae.

To analyze the data, the team used multivariate Poisson regression to calculate adjusted prevalence ratios (aPR). This statistical approach is often preferred over standard logistic regression when the outcome being studied (in this case, pneumonia signs) is very common in the population.

Datasets

The study relied on a localized but robust dataset:

  • Primary Data: 11,173 chest CT scans from the Arkhangelsk Clinical Oncology Dispensary (ACOD).
  • Validation Registry: The findings were cross-referenced with the Arkhangelsk Regional and Nenets Autonomous Okrug (NAO) Cancer Registry, a population-based database established in 1998 known for its high level of completeness and international audit standards.
  • Final Cohort: 1,148 patients who met all inclusion criteria (valid CT thickness, available registry data, and confirmed malignancy).

Key Insights and Ideas

  • The "Silent" Prevalence: A staggering 48.4% of patients showed CT signs of pneumonia, yet 95.3% of these cases were mild (less than 25% lung involvement). This suggests a high rate of subclinical or asymptomatic infections in the cancer community.
  • Anatomy of Risk: The risk wasn't uniform across all cancers. Patients with tumors in the lungs, head and neck, and upper gastrointestinal tract were significantly more likely to show pneumonia signs. This points toward a potential link between the primary site of malignancy (and perhaps its associated treatments like radiation) and lung vulnerability.
  • The Cardiovascular Connection: One of the most striking findings was the linear relationship between the Agatston score (coronary calcification) and pneumonia risk. As the score increased, so did the likelihood of detecting pneumonia signs, highlighting that cardiovascular frailty is a major predictor of COVID-19 severity, even in cancer patients.
  • Demographics Aren't Destiny: While initial raw data suggested men and older patients were at higher risk, these differences vanished once the researchers adjusted for comorbidities like emphysema and heart health. This suggests that biological age and organ health are more important than chronological age or gender.

Key Results

  • Overall Prevalence: 48.4% of the cohort had CT-detectable signs of COVID-19 pneumonia.
  • Cancer-Specific Risk (aPR):
    • Lung Cancer: 1.87x higher risk compared to the reference group (female genital cancers).
  • Head and Neck Cancers: 1.85x higher risk.
  • Upper GI Tract: 1.51x higher risk.
  • Breast Cancer: 1.38x higher risk.
  • Comorbidities:
    • Emphysema: Increased risk by 25% (aPR 1.25).
  • Coronary Calcification: Risk increased progressively with the Agatston score, peaking at a 61% increase (aPR 1.61) for those with the highest scores (≥300).

Limitations and Future Work

The authors acknowledge several limitations:

  • Retrospective Nature: The study looks back at existing data, meaning it cannot prove causation.
  • Lack of PCR Correlation: It was impossible to match every CT scan with a concurrent COVID-19 PCR test, though the authors note that CT signs were often more sensitive than early-pandemic testing.
  • Missing Lifestyle Data: The cancer registry does not track smoking status, which is a major confounding factor for both lung cancer and emphysema.

Future Work: The team plans to conduct survival and cause-of-death analyses to see if these "incidental" AI findings of pneumonia and calcification actually predicted which patients would have poorer long-term outcomes.

Summary updated

Abstract

Aim. To study the prevalence of pneumonia features associated with 2019 coronavirus disease (COVID-19) in cancer patients based on chest computed tomography (CT) data using an artificial intelligence (AI) algorithm. Materials and methods. A cross-sectional study was conducted as part of the ARILUS project. Using multitarget AI, CT images of 1148 patients examined at the Arkhangelsk Clinical Oncology Dispensary from 01.04.2020 to 31.12.2021 were analyzed. Patients were divided into groups: without signs of pneumonia (n = 592, 51.6%) and with signs of pneumonia (n = 556, 48.4%). In 95.3% of patients with pneumonia, the lesion volume was less than 25% (CT-1). Using multivariate Poisson regression, adjusted prevalence ratios (aPR) with 95% confidence intervals (CI) were calculated. Results. For demographic characteristics such as gender, age, place of residence, no relationship with the presence of signs of COVID-19 pneumonia was established. Topography of neoplasm is associated with the presence of signs of COVID-19 pneumonia (reference group – cancers of the female genital organs): lung cancer – aPR 1.87; 95% CI: 1.40–2.49; head and neck cancers – aPR 1.85; 95% CI: 1.32–2.58; upper gastrointestinal tract – aPR 1.51; 95% CI: 1.12–2.04; breast cancer – aPR: 1.38; 95% CI: 1.00–1.90; p < 0.01. The presence of pulmonary emphysema is associated with signs of COVID-19 pneumonia: aPR 1.25; 95% CI: 1.09–1.45, p = 0.002. With an increase in the Agatston score (AS) reflecting coronary artery calcification (reference group absence of calcification), the association with the presence of signs of COVID-19 pneumonia increased – for AS 1–99: aPR 1.24; 95% CI: 1.05–1.47; AS 100– 299: aPR 1.58; 95% CI: 1.33–1.87; AS 300 and above: aPR 1.61; 95% CI: 1.36–1.90; p < 0.001 for a linear trend. Conclusion. Factors associated with the detection of COVID-19 pneumonia among cancer patients include the localization of neoplasms in the lungs, head and neck organs, upper gastrointestinal tract, breast, and as well as the presence of signs of emphysema and coronary calcification according to CT data
BibTeX
@inproceedings{dyachenko2025prevalence,
  author = {Andrey Dyachenko and A M Grjibovski and M BOGDANOV and D. V. Bogdanov and Ekaterina Nazarova and Anna Meldo and Valeria Chernina and Mikhail Belyaev and Victor А. Gombolevskiy and М. Yu. Valkov},
  title = {{Prevalence of COVID-19-associated pneumonia signs on chest computed tomography in cancer patients: the ARILUS study}},
  booktitle = {Sechenov},
  year = {2025},
  doi = {10.47093/2218-7332.2025.16.2.4-17},
}
Figure