Case study

Dodes

Dodes is the abbreviation for Diagnostic nodes. It is a methodology designed by Prof. Dr. Paul Martin Putora from St. Gallen hospital in Switzerland. Its goal is to document medical treatment guidelines in a structured way and allow for further analyses of these guidelines to promote the most up to date medical procedures.

Problem definition and goal

A medical guideline is a document with the aim of guiding decisions and criteria regarding diagnosis, management, and treatment in specific areas of healthcare. These guidelines are updated regularly and are often in the form of “free text” documents.

A healthcare provider is obliged to know the medical guidelines of his or her profession and must decide whether to follow the recommendations of a guideline for individual treatment. It is important to find a way to promote the newest guidelines and draw attention to possible diverges between the guidelines and widespread practice.

Prof. Dr. Paul Martin Putora proposed a method to document the decision-making process in a structured way. He used the decision tree notation that condensed the information in a very efficient and readable way and also allowed to compare the decision-making process between different healthcare providers.

Challenges

Terminology unification

The treatment decisions are based on parameters which can have different names in different hospitals. Even when the different hospitals use the same names for same parameters, they might express parameter values in different units; for example, glucose can be measured in mmol/l as well as mg/dl. To perform calculations, this terminology had to be unified or at least be mappable.

Comparison trees

The decision trees can be documented in diverse ways. Some prefer to use complex mathematical expressions to define when a given action is performed when others use simple logical expressions. Some medical centers can omit certain parameters because the measurements for that parameter have not been taken. To find differences in such complicated structures was not trivial.

Huge state-space

The decision-making comparison has been performed in studies where more than 10 hospitals took part. One treatment can depend on 10+ parameters. To evaluate such state-space, we can easily end up with billions of combinations that need to be evaluated.

Data presentation

The differences between treatments can be quite profound. It is complicated to get the right insights from the results unless users investigate the result in detail.

Solutions

Dodes Case Study 1

Terminology unification

Before the decision tree for a particular treatment can be created the terminology has to be unified in the treatment template. This template defines a vocabulary for decision trees in terms of the parameters that must be considered in the treatment as well as actions that can be performed. When the user is creating a decision tree, the system is using the template to provide guidance and ease the process of input.

Comparsation trees

We have implemented algorithms that can transform the decision tree into a multidimensional state-space and can assign each “coordinate” a set of actions for a given parameter range. This allows us to compare each coordinate to find differences. From this information, we can back generate a decision tree that represents comparison results. We can use the same approach to enhanced the validation of decision trees like calculate parameter ranges that do not have any action or have contradictory conditions.

Huge state-space

To compute large decision trees from many institutions, we have implemented parallelized computation, using the google task queue. The computational state-space is divided into many smaller spaces that are sent to the task queue. This service spans as many computers as needed and performs the computation in parallel, returning results that are merged into a single comparison result. This way, we can perform comparison calculations in near real-time.

Data presentation

To reduce comparison results, users can use filters actions, or parameter values and thus reduce the state-space to much smaller sizes. We are still investigating options to make the comparison result more compressed and readable. Currently, we are experimenting using correlation matrixes.

Dodes facts

Domain
Medical
Technologies
Java, MySQL, Spring stack, REST, VUE, VueUEtify, Apache Solr, Docker, Kubernetes
Core competences
Data analytics, Data visualisation, Microservices

Awards

Dodes have been used in many studies, some of which have been funded by the biggest pharma companies like Roche and Genzyme.

The results of these studies were published in Neurosurgery, in an article called “Patterns of Care and Clinical Decision Making for Recurrent Glioblastoma Multiforme” which you can read here: https://academic.oup.com/neurosurgery/article/78/2/N12/2453749.

The reviewers are stating: “The authors are to be congratulated for identifying core clinical decision-making criteria that may be useful in future studies of recurrent GBM. This decision tree is an excellent reference for clinical trial development, and several active clinical trials already target the DODEs identified in this study.”

This especially credits the authors of the study around Prof. Dr. Paul Martin Putora but also proves the quality of our tool and its usability in discovering optimal treatment strategies.

See also

Case study

Koderia

Case study

Celine

Case study

DMS