Pancreatic cancer has a survival rate of only 3% after five years, however early detection can double the probability of survival of a patient.
In order to locate the symptoms of this disease even if people do not regularly attend a medical checkup, researchers at Columbia University and Microsoft concentrated on exploring the macro data or big data, ie, large data sets that we disseminated as internet users.
To conduct a study on this new diagnostic technique, they used the data of thousands of anonymous users of Microsoft’s search engine, Bing, one of the most popular after Google.
Looking for symptoms
The study, conducted by Drs Eric Horvitz and Ryen Blanca (from Microsoft) and John Paparrizos (a graduate of Columbia University), first identified Bing users who had just received a diagnosis of cancer from pancreas based on their searches on this search engine, such as “I have been told I have pancreatic cancer” or “newly diagnosed with pancreatic cancer.”
Second, the searches carried out by these users diagnosed with cancer were analyzed and compared with other searches for undiagnosed people, tabulating which symptoms they were looking for that were related to each other .
Using this very simple technique, 5-15% of pancreatic cancer cases could be identified with almost no false positives . Only one case in every 100,000 was declared cancer without being.
Apparently, there were certain searches that suggested with much greater probability that this type of cancer was suffered. Thus, for example, if the user searched for back pain and also yellowish skin, then the probability of suffering from it was high. That means that searching only for “back pain” or “yellowish skin” does not mean that one has symptoms of pancreatic cancer, but it does seem to find a correlation if those two searches were performed by the same user at several different points in time. .
Searches that could be associated with other emerging symptoms of the disease , such as abdominal discomfort, unexplained loss of weight or appetite, light-colored stools, generalized itching, or dark urine were also analyzed.
Thus, what this study showed is that signals in search logs show the possibilities of predicting a proximate diagnosis of so-called pancreatic adenocarcinoma from combinations of subtle time signals revealed in internet searches. The identification of possible cases of this disease is not very high, but it would help prevent thousands of cases a year.
Pancreatic cancer is the ninth most frequent tumor in Spain and for the year 2030 an increase in the incidence of this pathology of more than 50% is expected. Furthermore, current chemotherapies can only prolong survival for a few months. Thus, the tools based on the statistics of the searches of possible affected by it could constitute a powerful alliance for conventional diagnostic techniques.
