More than 8 million people die from cancer worldwide and continue to grow (IHME, Global Burden of Diseases). In Japan, the probability of suffering from cancer in the whole life is 62% for men and 47% for women, and the probability of dying from cancer is 25% for men and 16% for women (Cancer Information Service at National Cancer Center).

Treatment methods for cancer have long been focused on surgery, radiation therapy and cytotoxic chemotherapy, while in recent years, immune checkpoint inhibitors have appeared, and immunotherapy has been added. Thus, the medical situation for treating cancer is dramatically changing.

Immune checkpoint inhibitors release the brake on the immune response and help T cells exert their anti-tumor effects. It is known to have dramatic effects on solid tumors with less side effects.

However, while those drugs are expensive, there is also a problem that a limited number of patients can respond to those drugs. Actually, the low response rate is for many cancers, depending on the type of cancer, such as 20% for non-small cell lung cancer, 15% for head and neck cancer, 15% for esophageal cancer, 25% for renal cancer and 35 to 40% for melanoma. However, some Hodgkin lymphomas, adherent melanomas and Merkel cell tumors are high, at 87%, 70% and 56%, respectively.

In order solve the problems due to the low response rates,

  • ■ to develop combination therapy with other treatments,
  • ■ to use biomarkers to select effective patients before administration, that is to stratify patients,

are considered as important. Regarding the use of biomarkers, the development of simple methods that do not use cancer tissues obtained by biopsy or surgery as test samples is awaited.

The current approach to finding such biomarkers is to collect data from many patients, measure specific items based on preliminary findings and hypotheses, and find statistically significant differences for discovering something novel. Such determination of the measurement items is performed by humans, even though they are trained, so that sensory judgments and biases are inevitable. Also, this process is time consuming and does not guarantee any discovery. Therefore, it would be considered it may not be efficient. Currently, under these situations, new mathematical approaches are being tried to be introduced to the process of finding insights from big data.