br Introduction br Personalized medicine PM is a field of
Personalized medicine (PM) is a field of healthcare which finds the best available diagnosis or tailored diagnosis to satisfy the need of an individual patient. According to Barrack Obama, it gives us “one of the greatest opportunities for new medical breakthroughs that we have ever seen” . Healthcare providers can build a revolutionary new system for medical care by combining the data from diagnostic tests and medical history of patients to deliver enhanced value to patients. One such possibility is personalized tests, where the set of clinical tests re-commended to a patient for diagnosis of a disease, are customized ac-cording to the patients needs. To the best of our knowledge, the idea behind customization of diagnostic tests has not been explored before.
E-mail address: [email protected] (S. Bhattacharya).
1 Research scholar. 2 Assistant Professor, Computer Science Dept., IIT, Kharagpur. 3 Dean and Professor, VGSOM, IIT, Kharagpur.
Fig. 1 shows the overview of our personalized diagnosis system. In this paper, we study the problem of designing such a system and propose a methodology for it taking esophageal cancer as a case study.
Esophageal cancer arises from the long muscular tube connecting the pharynx with the stomach, called esophagus. This disease is usually accompanied by symptoms like pain while swallowing, hoarse voice, etc. It is observed in both developed and developing countries, with causes including various forms of consumption of tobacco, insufficient intake of fruit and vegetables, overweight and obesity, alcohol con-sumption, BCI-121 reflux disease, etc. The rate of esophageal cancer is rising worldwide, even though the onset and magnitude varies among countries . American Cancer Society estimates approximately 17000 new cases of esophageal cancer in 2015 . Esophageal cancer rate has
Fig. 1. Overview of personalized diagnosis system.
also been increased in Great Britain since late 1970s . A variant of esophageal cancer, called squamous cell carcinoma , appearing mostly in the upper esophagus, is also common in developing countries such as India. Funded by Govt. of India, a reputed hospital in Mumbai, India floated two mobile vans, with the objective of diagnosing eso-phageal cancer in rural area Maharashtra (a state of India). They col-lected a host of features, ranging from lifestyle, medical history to clinical test features (see Section 2 for details). We use this dataset, for about 3000 patients to perform the studies in this paper. The broad steps taken for the study are:
• Collect and pre-process electronic medical records (EMR) into fea-tures for a machine learning (classification) problem. • Identify the “best” off-the-shelf classifier for the purpose, using all the features.
• Design a methodology for tuning the classifier such that fewer dis-eased patients get classified as “normal”, at the cost of possibly more normal being marked as diseased.
• Remove clinical test features, one by one and find the best classifiers under the above constraint.
• Assign “costs” to features and find the best set of features corre-sponding to defined cost budget.
The proposed methodology can help in identifying a patient (eso-phageal cancer in this case) from the population with aid of EMR data and a few basic clinical tests. The health care system may adopt a method of screening where a suspect can be diagnosed initially by any health officer or clinician before approaching to any doctor, where all the required data need to be captured for the necessary recommenda-tion to the patient. A doctor can play his role once the system re-commends a visit to a medical practitioner after analyzing the captured data. This will bring a ray of hope to the people of developing countries where sophisticated medical treatment is still far to reach, specially in the rural area where either a hospital is beyond the vicinity of a person or a doctor is hardly approachable due to location constraint. As this novel scheme is also proposing customized test as per user's choice it is expected that this scheme will attract people from both developed and developing countries for the treatment. Now more patient can afford to have the treatment either by bearing the cost by their own or by in-surance provider with an assumption that the premium will be low (the cost of overall treatment will be lower as compared to existing proce-dure). Further we are expecting the health care service provider to implement this system as a soft sensor in preventing the disease by