Mathematical prediction models outperformed physicians in predicting the outcome and responses of lung cancer patients to treatment, according to research presented April 20 at the 2nd Forum of the European Society for Radiotherapy and Oncology (ESTRO). The researchers suggested it might be “unethical” to make treatment decisions based solely on physicians’ opinions.
As the number and types of treatment options for patients with lung cancer and clinical data have increased, it has become more challenging for physicians to make treatment decisions.
Cary Oberije, a postdoctoral researcher at the MAASTRO Clinic at Maastricht University Medical Center in Maastricht, the Netherlands, and colleagues applied previously developed math models to predict the probability of outcomes and response to treatment using radiation therapy with and without chemotherapy.
They also asked radiation oncologists to predict the likelihood of the following outcomes: two-year survival and dyspnea and dysphagia at the patient’s first visit and after treatment.
At the first time point, physicians predicted two-year survival for 121 patients, dyspnea for 139 and dysphagia for 146 patients. At the second time point, predictions were available for 35, 39 and 41 patients, respectively.
The statistical models bested physicians’ predictions for all three questions and at both time points. Oberije et al reported that area under the curve of the receiver operating characteristic for the models’ predictions at the first time point were 0.71 for two-year survival, 0.76 for dyspnea and 0.72 for dysphagia. Radiation oncologists’ results were 0.56, 0.59 and 0.52, respectively.
The models provided superior positive predictive value compared with physicians and comparable negative predictive value.
Although prediction models are not used widely in clinical practice, Oberije pled the case. “In our opinion, individualised treatment can only succeed if prediction models are used in clinical practice. We have shown that current models already out perform doctors. Therefore, this study can be used as a strong argument in favour of using prediction models and changing current clinical practice,” she said in a release.
“Our study shows that it is very unlikely that a doctor can outperform a model,” she concluded.