AI Reveals Link Between Influenza and NSCLC

A correlation between non–small cell lung cancer (NSCLC) and influenza shows a decrease in cancer survival rates.

In a study that included 202,485 patients with NSCLC, researchers discovered a significant association between influenza-like illness activity and mortality rates among this population. The study found that during months with high influenza activity, the monthly mortality rate among patients with NSCLC rose from 0.041 (95% CI, 0.041-0.042) to 0.051 (95% CI, 0.050-0.053), representing a 24% increase (relative risk, 1.24 [95% CI, 1.211.27]). These findings, spanning across various clinical and regional subgroups, show the need for targeted interventions and heightened awareness among health care providers to ease the potentially preventable burden of influenza-related lung cancer deaths.¹

Regarding patient demographics, the majority resided in metropolitan areas (85.5%), particularly in California (33.4%) and Georgia (12.7%), and fewer patients were in Alaska (0.1%), Utah (1.2%), Hawaii (1.5%), and New Mexico (1.7%). A total of 1041 state-months were observed—53 of which were considered influenza months—from 13 states (approximately 34.6% of the US population). Patient characteristics are shown in FIGURE 1.1

Developing the Hypothesis

The concept for the study originated in the clinic where Simon Cheng, MD, PhD, sees many patients with NSCLC. “Throughout our experience,” Cheng said, “we observed that patients with lung cancer often face additional health challenges due to undergoing chemotherapy and radiation treatment and can become immunocompromised. Particularly during winter months, many patients are susceptible to developing upper respiratory infections, influenza, or colds, which can severely impact them and sometimes lead to hospitalization.” (FIGURE 21)

This observation led Cheng to consider whether there was a correlation between patients with lung cancer and the rate of influenza activity in the community, how it impacts the survival rate of these patients, and how they receive treatment. Cheng is residency program director and assistant professor of radiation oncology at NewYork-Presbyterian/Columbia University Irving Medical Center in New York, New York, and one of the authors of the study, titled, “Influenza Activity and Regional Mortality for Non–Small Cell Lung Cancer.”

Organization of Data

“The challenge of recognizing spatial-temporal trends, managing the size of the data set, and how the data [are] organized and merging those data effectively can be difficult,” said lead author Connor Kinslow, MD. Kinslow is a fourth-year resident in radiation oncology at NewYork Presbyterian/Columbia University Irving Medical Center Vagelos College of Physicians & Surgeons in New York, New York. The study team was able to review data from the Surveillance, Epidemiology, and End Results (SEER) database, Cheng explained.

The team also explored publicly available data using the FluView Interactive dashboard, produced by the Epidemiology and Prevention Branch in the Influenza Division at the CDC. This platform is used for visualizing and interacting with influenza-related data from the Influenza-like Illness Surveillance Network (ILINet). The ILINet database “consists of more than 3500 providers in 50 states who report more than 47 million patient visits per year,” study authors wrote.¹

“The CDC has a week-by-week geographical spatial analysis reporting on influenza-like illnesses in the community for the [past] 10 years,” Cheng explained. This allowed the researchers to track week by week the increases of influenza-like illnesses. First, the researchers plotted out patients diagnosed with cancer in a geographic region using the SEER. They then, using the CDC influenza network, plotted out the incidence of flu in a population temporally and spatially, Cheng explained. The SEER database is updated annually for follow-up on vital status and it also undergoes routine quality-control checks, study authors stated in the paper.

Once the team determined it had the ideal resources for merging these 2 data sets—patients with cancer and influenza cases—to determine an association, it was still a challenge to visualize correlation using the data. “Dr Cheng and I conceptualize these research questions; however, what is often the case is that many of these analyses are beyond our analytical capabilities as clinician researchers or even the data scientists we have in our department,” Kinslow said.

Enter Virtualitics

“We decided to collaborate with Virtualitics because of their analytical manpower and ability to visualize complex associations such as for these 2 data sets,” Cheng said. Virtualitics is an organization based in Pasadena, California, that assists in helping solve complex, mission-critical problems using AI that is based on more than 10 years of research, said Ciro Donalek, PhD, the company’s chief technology officer and cofounder. “Especially for regulated sectors like in health care,” Donalek continued, “they need trusted applications for decision- making and adoption, and to provide [decision makers with trusted information] we use AI-generated insights, natural language, and 3-dimensional [3D] visualizations. This allows us to drastically reduce the time to gaining insights for both data scientists and domain experts.”

Visualizing the Data

Virtualitics provided a tool for analyzing and visualizing data. It helped researchers find a link between more severe cases of influenza-like illnesses and reduced survival rates in patients with lung cancer, Cheng explained. Visualization plays a crucial role at every stage of constructing a machine learning pipeline, beginning with merging data sets and preprocessing, and extending to the creation of new variables, explained Donalek. “This allowed us to see all these relationships and determine meaning from it,” he said.

In addition to the primary graphs, researchers used density plots, histogram plots, and some other creative visualizations of 3D plots indicating when high influenza months occurred. This helped the researchers identify certain areas where there might have been a bias in terms of density of influenza months during the H1N1 [Influenza A virus] pandemic, for example.

Disadvantages, Challenges, and Incentives

Visualization of the data revealed that “the magnitude of the effect gradually increased with quartiles above the poverty line, and that regional poverty correlates with vaccination rates, driving more susceptibility to influenza illness among those patients,” Kinslow explained. “It’s well known that patients in different economic demographics, in terms of poverty line, [have] less usage of vaccination and our analysis verified that,” Cheng said (FIGURE 31).^1,2

This study can also help to demonstrate a correlation between vaccination and lung cancer during influenza season. “Further research to confirm and explore this connection could pave the way for enhanced initiatives aimed at boosting vaccination rates within lower socioeconomic communities and could improve the health outcomes for patients diagnosed with lung cancer within these disadvantaged groups,” Cheng said.

“Patients with cancer are not necessarily more likely to get vaccinated than patients without cancer,” Kinslow said. Societies such as the American Society of Clinical Oncology and the National Comprehensive Cancer Network recognize that one of the difficulties in recommending vaccines is that there are not sufficient data to implement that as a standard practice, Kinslow explained.

“There are several studies that look at why patients with cancer are not getting vaccines, and it’s both lack of awareness from the providers and lack of inciting to get vaccinated by the providers,” Kinslow said.³

The researchers believe by strengthening the evidence of association between influenza and mortality, it increases awareness among providers, and helps to provide evidence supporting the flu vaccine, Kinslow emphasized. Another barrier to inciting the influenza vaccine is that primary care providers may assume it’s the job of the oncologist to incite vaccination vs oncologists thinking it’s the job of the primary care doctor, he further explained. “I routinely have patients ask me if they [should] get the flu vaccine,” Cheng continued. Many of his patients relay that their primary doctor told them to ask their oncologist if they should receive the flu vaccine. “So, I think there’s the lack of awareness of who should be doing and recommending what,” Cheng said.

Another interesting study using these techniques would ask, “Does COVID-19 increase the death rate of patients with lung cancer,” Cheng explained. There are some data to suggest this, but temporal or spatial studies like this one have yet to be explored. The cancer data with SEER have only now been reported for 2020, and Cheng believes it’ll be a few years before more data are reported. “I believe it would be intriguing to explore,” Cheng stated. “With the ongoing COVID-19 activity, the CDC has been consistently reporting on it since the pandemic’s outset, and we could potentially conduct a similar analysis,” he said.

REFERENCES

1. Kinslow CJ, Wang Y, Liu Y, et al. Influenza activity and regional mortality for non-small cell lung cancer. Sci Rep. 2023;13(1):21674. doi:10.1038/s41598-023-47173-x

2. Inequities in flu vaccine uptake. CDC. Updated October 18, 2022. Accessed March 19, 2024. https://www.cdc.gov/ vitalsigns/flu-inequities/index.html

3. Chang A, Ellingson MK, Flowers CR, Bednarczyk RA. Influenza vaccination rates among patients with a history of cancer: analysis of the National Health Interview Survey. Open Forum Infect Dis. 2021;8(7):ofab198. doi:10.1093/ofid/ofab198