The algorithm looks into all medical information of general practice. In particular, the use of free text data is new and important. Martijn Schut, Professor of Translational Artificial Intelligence: “In the medical history, the algorithm picks up a predictive signal. With this, the algorithm identifies a large proportion of these patients up to four months earlier than is currently the case.” Other studies used predefined and coded variables, such as the variables “smoking” or “coughing up blood,” for example. “Apparently, there is information about the patients' history in the free text, which the algorithm allows us to detect patients with cancer much earlier,” said Ameen Abu Hanna, professor of Clinical Information Science and co-investigator. Further research should reveal what are now the fragments on which the algorithm catches on in order to actually use it in practice.
Also suitable for other types of cancer
This method may also offer relief for other types of cancer, which are often not found until an advanced stage, such as pancreatic, stomach or ovarian cancer. For the patient, the chance of survival is greater and the quality of life is better. Henk van Weert, emeritus professor of family medicine, explains: “Now many patients are diagnosed with lung cancer at an advanced stage, 3 or 4, so 80 percent of patients often die within a year. Previous research made it likely that a four-week gain already has a noticeable effect on prognosis. Four months is thus a likely highly relevant gain.”
Screening versus algorithm
Of the 34 patients identified by the algorithm, one has lung cancer. Compared to screening, the number of false positives is lower and the selection can simply take place during the consultation. Lung cancer remains one of the most common cancers and, despite advances in treatment, still has a poor prognosis. More than 80 percent of patients die within five years. Previous research indicated a significant increase in survival when treatment starts four weeks earlier. Using the developed algorithm, 62 percent of patients with lung cancer could be referred four months earlier. However, the method has yet to be validated in different countries and healthcare systems.
Research design
The study analyzed data from 525,526 patients from four academic GP networks in Amsterdam, Utrecht and Groningen. Of this group, 2,386 patients were diagnosed with lung cancer. Diagnoses were validated using the Dutch cancer registry. Both structured and free text data were used to predict lung cancer diagnosis five months earlier (four months before referral).