How to create data-driven predictive service for healthcare?

How Solita and Coxa developed a risk prediction tool for joint replacement surgeries.



There is a lot of data about prior joint replacement surgeries, but it was not utilized in a way that is useful and creates value. 

By combining the existing data and AI know-how, the two companies aimed to develop an AI-based predictive analysis/risk assessment software that generates an individual risk score based on each patient's individual variables.

The developed software creates value and visualizations out of existing data by utilizing AI methods. The SW helps surgeons assess the risk of the surgery, and guides the patient and surgery team to prepare better for the operation or to cancel or postpone it.
Target groups

This case description describes the power of data and AI methods in predictive analysis, and is thus useful for companies, researchers, healthcare providers, and other developers.

Oravizio  - an AI powered tool for assessing the risks of joint replacement surgeries - developed by Solita and Coxa.

Solita is a Finnish technology, strategy, and design company whose service portfolio seamlessly combines expertise from strategic consulting to service design, software development, AI & analytics, cloud and integration services. Coxa Hospital for Joint Replacement is the largest artificial joint specialized hospital in the Nordic countries and the only one of its kind in Finland.

Oravizio -  description of the product

Oravizio is the world's first CE certified medical software for assessing the risks of joint replacement surgeries. Oravizio helps to estimate the risks of an individual operation and to visualize and communicate the risks to both the orthopedist and the patient. The ultimate goal is to help to make the best decision – together with the patient. Oravizio is based on Coxa’s in-depth clinical understanding and knowledge from the research literature, and Solita’s top-notch capability of data science. They used high-quality clinical data from 44,000+ surgeries, with 800 pre-operational variables and data points related to each surgery. It is estimated that the AI tool will save money and human suffering considerably when the success rate of surgeries improves and more precise evaluation reduces the risk of complications.

Description of the product development

The product development was initiated by Coxa’s need and vision to utilize their patient record and to better assess the risks related to patients’ surgeries. Together with Solita, the need specified and an idea of the product defined.

Data utilization process in medical software development described by Solita:
Solita's process description of developing medical regulated software

The development work was based on Coxa’s clinical data gathered from more than 44,000 operations performed during a period of ten years. Data comes from Coxa’s patient database, and consists of basic patient information, symptom questionnaire, medication history, laboratory measurements, and diagnosis. Oravizio uses 7-15 explanatory variables to generate the surgery risk assessment. Coxa’s data was enriched with Finnish Institute for Health and Welfare’s artificial joint registry data. Also scientific literature on surgical decision making and Coxa’s professionals’ clinical expertise were utilized.

Development of the AI models

Three models were created to predict the post-operative risks of the surgery: infection within one year after surgery, need for another surgery during the following two years, and mortality during the following two years. The AI models were taught on data from years 2008-2015, and tested on data from years 2016-2018.

Data contract practices

The permission to combine data from Finnish Institute for Health and Welfare and private patient data was requested and granted by the Institute, and the data/permission was processed in the Institutes environments.

Technological solutions for data analytics

First, the quantity and quality of the data, and the needed machine learning methods were assessed. Evaluation of the needed and applicable machine learning solutions/models to enable the effective prediction of the risks with a suitable amount of variables was made.

Risk prediction tool testing in practice

The tool was clinically piloted at Coxa for almost a year. At the same time, new features were developed and, on the other hand, the preparation for external pilots were ongoing (by developing user management, support processes, etc.).

Risk prediction tool CE marking process

The product was recognized as a medical device at an early stage of the development, which made it possible to plan and execute the documentation of the development processes to match CE marking requirements.

As the product is category I medical device, CE marking was possible to do as self-certificate. Solita became the manufacturer of the product, which affected the implementation of the ISO 13485 standard. As the directive is being replaced by Medical Device Regulation on May 2020, also the CE marking will require the participation of the notified body, and the category Oravizio belongs to will change from category I to category IIa.

Lessons learned:

  1. Actual, concrete need and the use case need to be defined in the very beginning in close collaboration with healthcare professionals.
  2. There is never too much data.
  3. Understand the regulation already in the beginning: you have to be able to define the purpose of the data collection and the possible secondary use of data.
  4. Be smart with the timing of the research-based data analytics and software development.
  5. CE marking process requirements for the documentation need to be taken into consideration already from day 0.

More information: Kimmo.kivirauma (a)

Last updated: 3.3.2020