Working Smarter to Reduce Lung Cancer Deaths

Lung cancer is unrelenting – especially for those who present late in the disease process. Every year more people die of lung cancer than breast, colon and prostate cancer combined according to data from the American Cancer Society.

In my hometown of Chattanooga, Cancer is the most common cause of death, even above heart disease. This is primairly because of the high prevalence of late stage lung cancer presenting to our hospitals. Despite the improvement in therapies and effectiveness of our systemic drugs, the five-year survival rate has not changed in 40 years.

Yet, we know lung cancer is very treatable and curable if found early. What if we could better predict who is at high risk for lung cancer and engage them in a screening program prior to development of the disease? The National Lung Screening Trial demonstrated this approach works. Individuals studied, who were at risk for the disease, had a 20% reduction in their cancer mortality with an annual low dose CT scan of the chest. When cancer was found, it was likely to be in an early stage, allowing simpler and more effective treatments. It turns out that low dose CT scans are a more powerful screening tool than mammography. But finding these individuals and getting them to engage in their health can be difficult.

The answer is in the weather. It turns out that lung cancer follows a very predictable pattern like the weather so we should be able to use similar predictive modeling to alert people in our community who are at risk and provide simple ways they can enter a screening program. With the conversion of medical records to an electronic format, hospitals now have hundreds of thousands of patient’s data stored on their servers ready to be analized. We implemented a software tool provided by a company called MedMyne [now Thynk Health], that utilizes natural language processing and machine learning, to capture and analyze people in our city.

What is Natural Language Processing? Human language is rarely precise, or plainly spoken, especially in a doctor’s clinic note. To understand human language is to understand not only the words, but the concepts and how they are linked together to create meaning. Despite language being one of the easiest things for humans to learn, the ambiguity of language is what makes it hard for computers to understand our conversations. Natural language processing helps solve this problem. Machine Learning is a form of artificial intelligence. When combined with NLP, computers are programed to look at large amounts of unorganized data, and make advanced and very accurate predictions. And they have the ability to improve that accuracy with time.

What we have accomplished with this technology is to evaluate approximately 300,000 people in our community and determine who is at risk for lung cancer. We have identified 32,000 individuals who fit the risk criteria and are now able to communicate with them about the possibility of a “storm” in their future. We know who their primary care physician is and work with them to get the patient engaged in a screening program. This information also makes their annual visits to their physician more meaningful.

With this technology, we are also able to alert the health care team if one of these individuals enters our hospital for unexpected reasons, creating an opportunity to enter them into a screening program. The technology also allows us to keep track of individuls being screened and alert them via email, letter or smartphone when it is time for their next screening exam, improving compliance.

Our predictive modeling is early, but whats great about machine learning is that it will improve with time and more data. These computer based tools are going to help us focus our efforts on reducing lung cancer deaths on the people who need it the most, helping them get out of harms way , the same way we use the weather reports to stay safe.

Scroll to Top