10 June 2016
A Microsoft research team has presented a feasibility study of how it is possible to predict whether someone is suffering from the early stages of pancreatic cancer based on their search history.
Pancreatic adenocarcinoma is the sixth most deadly cancer in Europe. Tricky to diagnose and fast to spread, the cancer is often identified too late, and the survival rate is extremely low. Web search logs may serve as a useful signifier of the disorder, for which symptoms in the early stages may not be sufficiently concerning to the patient for them to consult their GP.
A team from Microsoft believe they can change that. By analysing search logs they were able to predict people with the disease up to five months prior to a medical diagnosis. The results show an accuracy of 6% to 32%, while maintaining an extremely low false positive rate of 0.00001%.
They started with patient searches which implied a recent diagnosis of pancreatic cancer. Once identified the team looked back at previous searches for known symptoms. They constructed a statistical model using machine learning that looked for the patterns between the searches.
As all search data is anonymised, the evidence that the patients have pancreatic cancer relies only on their self-reporting by searching, yet the team is confident their proof of concept study proves the worth of their approach which could be applied to other, similarly devastating diseases. They hope to go ahead with clinical trials, to prove the utility and cost effectiveness of their predictive screening system.
Ryen White, chief technology officer for Microsoft Health and an information retrieval expert, emphasised that any future tool would not be a replacement for a medical diagnosis: “The goal is not to perform the diagnosis,” he said. “The goal is to help those at highest risk to engage with medical professionals who can actually make the true diagnosis.”