News-Medical Life Sciences recently interviewed Dr. Ananya Malhotra, Research Fellow at the London School of Hygiene, regarding her most current research involving pancreatic cancer. Dr. Malhotra believes artificial intelligence (AI) would have a positive effect on the ability to predict the disease.
Dr. Malhotra said she began to focus on AI when she became aware that the prognosis for pancreatic tumors and several other cancers has not seen improvement for the last few decades. She compared it to the increase in the rate of survival for other cancers. Dr. Malhotra recognized the need for a new approach.
The Screening Obstacles
Eight to twelve pancreatic cancer cases per 100,000 people are diagnosed each year putting it in the rare disease category. Screening for early indications of the disease is of the utmost importance, but it is costly, invasive, and carries certain risks.
The goal is to find the disease early, even before the patient shows symptoms. This gives the patient better odds for survival.
Effective Screening
The World Health Organization sets out criteria for effective screening. It defines screening as identifying an unrecognized disease by means of tests or other procedures that apply to a target population.
Dr. Malhotra suggests that due to the rarity of pancreatic cancer, screening an entire population is not appropriate, nor is it practical.
The doctor then described an effective screening program as including the target population (people most at risk) and accessing treatment for people who are diagnosed with the cancer.
Dr. Malhotra emphasized that her team is not involved in the development of biomarkers (a measurable substance) or deciding which tests would be most effective.
They are, however, focusing on their algorithm with the intention of pairing it with an accurate, non-invasive test to be used for target screening.
Dr. Malhotra was asked what early signs or symptoms are linked to pancreatic cancer.
The doctor responded that their AI model predicts risk based upon all variables. She said it was therefore not appropriate to highlight just a few, but she did list these commonplace symptoms: back pain, weight loss, anemia, abdominal pain, jaundice, and diabetes. Lesser-known symptoms are fatigue, insomnia, and depression.
Analysis by Artificial Intelligence (AI)
The team found that most people who eventually developed the illness had been diagnosed for having non-specific symptoms years before receiving a cancer diagnosis.
So far, researchers have been unable to associate the earlier symptoms with an increased risk of having the cancer.
Results of the Pilot Study
Dr. Malhotra’s study of people under sixty years of age gave a prognosis of people at higher risk almost twenty months prior to their actual diagnosis.
She estimated that almost 1,500 tests need to be given to save one person from pancreatic cancer. The doctor does not believe that these numbers justify screening at this juncture.
AI does, however, have the potential to reduce the number needed for screening. That number can be scaled down considerably if patients are matched to controls from the general population. The people in the pilot study used as controls had other types of cancer.
AI Identifies Pancreatic Cancer Combinations
The team believes that it will be possible to find a combination of these non-specific symptoms that will identify people who are at a higher risk of having the disease.
AI was used to study data and find this combination. It looks for patterns in the data. According to the team, AI improves future prognosis relying on examples they provide. It applies past history supplied by the team to the new data when predicting future events.
The Advantage of AI
When working with data in large volumes, it is difficult if not impossible for the human eye to identify trends or code rules.
Machine learning (ML) automatically creates models from data. The algorithms are the engines of ML. The algorithms turn a collection of data (data set) into a model.
According to Dr. Malhotra, ML algorithms produce rules received from data which, she says, is much more powerful.
Predicting Outcomes
ML is an important tool for automated pattern recognition. By utilizing this information, ML algorithms control outcomes and classify data into categories.
In their current study, the team intends to create an algorithm that would produce a risk score for the probability of a patient having pancreatic cancer.
The risk score will be based on a compilation of symptoms exhibited by patients and compared to people who did not have the disease.
Looking Ahead
The findings by the team are based on diagnoses between the years 2005 and 2010. Dr. Malhotra would like to expand these findings with more recent data.
An important step would be to examine the available pre-diagnostic data two years prior to diagnosis. In that way, they will be able to detect high-risk patients earlier.
Another goal is to find cancer patients with similar symptoms and use them as controls. This would entail involving the general population.
According to Dr. Malhotra, their findings indicate the impact of variables such as diabetes and smoking. The team will match controls relative to their diabetes and smoking status.
Lastly, the team will compare the cost-effectiveness of their screening to similar programs.
The findings of Dr. Malhotra’s study were presented to the ESMO World Congress in July 2020. It was well-received.
