Our first meetup of our recently founded group
OPENMEDiAID wants to help people take better medical decisions by providing them with well structured medical information gathered from the collective experience of people with similar cases. This functionality is driven by medical data organized in collections of medical profiles (profile pool).
Case-specific medical information is generated by means of profile analysis; a multi-step process that produces a selection of relevant profiles based on a given description of a medical case. The selection of profiles is supplemented with statistics to allow further analysis with different tools.
1. Case description
The profile analysis process starts with a description of the medical case for which relevant information is to be found. The description is given in terms of a standardized medical language using a smart editor to provide guidance in this process.
2. Profile matching
The source profile is matched with the pool of existing profiles to collect all profiles relevant to the case. The different features of the profile are compared using a set ofdistance metrics which provide a formal way of measuring similarity.
3. Profile exploration
The matching profiles are likely to contain information that is not present in the source profile, i.e. references to diseases, commonly used treatments, statistics about all treatments used by other patients. It is also possible to identify important pieces of information that are missing in the description of the source profile but could add substantial value.
The systems medical knowledge is built from three major types of medical information all combined in one large interrelated model.
The foundation of any description of medical facts is a model of the core concepts of medicine like human physiology (anatomy, biochemistry), basic symptoms, measurable body facts (antibody, blood sugar…). It is a standardized model of medical language built by groups of medical experts.
Personal Health Record
A Personal Health Record (= medical profile) describes a person’s medical history in terms of a chronology of medical findings and other relevant information: Any symptom, clinical test result, treatment, information about diet and family’s medical history is part of this profile. The information of the profile is formulated using the medical language framework.
Within the pool of medical profiles, each profile represents an example of how a syndrome or disease emerges and develops (clinical presentation and time course), relevant environmental factors (epidemiology) and what methods can be used as treatment. There are different types of profiles collaboratively created and maintained by patients as well as physicians. Each profile type has distinct characteristics and plays a different role in the profile analysis:
- archetype profiles represent the common cases of medical conditions as they usually develop and how they are usually diagnosed and treated. They are created by physicians.
- patient profiles represent the individual cases of real patients and are maintained by their owners.
- clustered profiles represent average cases built from groups of similar patients by machine learning algorithms.
A significant part of medical knowledge exists as relations between concepts, i.e. relations connect one specific concepts with another. For example, the presence of a collection of symptoms (syndrome) is associated with a disease (illness script), a specific symptom might frequently occur as a secondary symptom of another, a specific treatment is used to treat a symptom or cure a disease.
Profile matching is the process of finding related profiles for a given description of a medical case. It is based on a framework of distance metrics and a scoring model to rank matching profiles. The matching algorithms take two profiles and a set of distance metrics to compute a set of characteristic values used to rank the profiles similarity.
The problem of measuring how much two profiles have in common can be decomposed into smaller computations of similarity based on the medical findings (features) that make up each profile. This can be achieved with a function (distance metric) that produces a measure of similarity (real number) based on the comparison of two profile features. Applying a set of distance metrics to a pair of medical profiles will result in a set of values describing the profiles overall similarity. A library of parametrizable distance metrics can be built over time and used to compose specialized notions of similarity.
Distance metrics produce sets of numbers as an initial representation of profile similarity. A scoring model is used to associate each number with a weight in order to produce a more case specific notion of similarity.
Evaluating treatment options
Supporting differential diagnosis
It is common that a set of symptoms is not sufficiently specific to identify their cause with satisfying certainty. In a process called differential diagnosis different medical examinations and tests are used to narrow down the set of potential causes. Each test results in a new medical fact about the patient such that the description of her/his case increases in specificity.