DUBBAH

Welcome to the website for the project DUBBAH.

From this site you can read about:

Collaborators:

View the project on GitHub

What is DUBBAH?

DUBBAH is an acronym for Digital Understøttet Bedre Brug Af Hjælpmidler which can be translated to digital supported better usage of assistive technology. It is a project funded by Kommunernes Landsforening which focuses on developing publicly available machine learning algorithms that can assist providers in the management of loaning assistive technology.

The objectives are:

  1. An algorithm capable of estimating a timeline of the progression in assistive technology for any citizen. This include estimating the time in between loans, identify unnecessary assistive technology loaned and estimate the probability the citizen will need more assistive technology in the future.
  2. An algorithm capable of identifying the probability that physical exercise would benefit the citizen.
  3. Find typical patterns in the usage of assistive technology and benefit of physical exercise.
  4. The software should have a well defined and simple interface.
The project is a partnership between Aarhus Unviersity, Kommunernes Landsforening, Aalborg Kommune, Aarhus Kommune, University College Nordjylland, and DigiRehab A/S. The project scheduled to run from April 2018 to October 2019 in 3 overall phases. The first 6 months is a startup phase including data collection in Aalborg Kommune. The following 6 months contains software development at Aarhus University. The last 6 months contain integration and pilot testing in Aalborg Kommune.

Background for DUBBAH

The Danish Social Security Administration and Fredericia Municipality

The Danish Social Security Administration has formerly investigated the effects of early rehabilition in the proces of loaning assistive technology. In a project called "Tidlige Rehabiliterende Indsats (TRI)", which can be downloaded here, they found different patterns in the loans of assistive technologies depending on what a citizen applied for in the beginning. As an example citizens who applied to have their doorstep removed very often was in need of more help. They also found that any improvement regarding self capacity and physical capacity gained as an effect of assistive technology or exercise varied depending on age and the baseline.

Danish Municipalities' Common Program for Health Technology

The program for healhtcare technology was founded by the municipalities in Denmark. The objective were to ensure a strategical developement of healthcare tehcnology such as how to promote sharing of knowledge, coordination, common documentation and implementation of new healthcare technology. You can read more about the program here. In the status report from 2017, which can be accessed here, it can be seen that "better usage of assistive technology" was rated among the top 5 health tech contributers in the past and in the future. They define better usage as ways to promote collaboration and effictivity in between, occupational therapist, physical theprapist, nurses and the citizen. Furthermore, the program focuses in 2018 on healthcare technology for mobility that can rehabilitate, maintain or promote self capacitity in the citizen.

The Application of Artificial Intelligence and Machine Learning in Healthcare

The potential benefits of artifical intelligence within healthcare are often discussed in novel science and research literature. Some of the hopes are that technology can help overcome future demographic challenges by being applied in a broad range of fields such as cardiology, radiology and much more (Zhang 2018). As an example apple recently published the new generation of Apple watch which is capable of detecting ateria fibrillation. One of the huge benefits of medical data is the size. In 2020 medical health records are predicted to generate 25000 petabytes of data (Feldman 2012). Nonetheless, the practical application of big data and machine learning in healthcare is still somewhat rudimentary (Peterson 2018). In a review from 2016 the challenges and oppertunities of big data were discussed. Some of the drawbacks identified were data quality, data security, privacy, interoperability etc. (Toldo 2016). Petersen et al. describe different challenges for patients / developers / healthcare professionals. They suggest a framework for the stakeholder relastionships requiring complex and cooperative collaboration between all stakeholders. Similar solution including interdisciplinary development was proposed in a recent review investigating how digital healthcare can be used within epidemiology (Widmer 2018). This is exactly what we strive for in DUBBAH.

Anonymized Authentic Case Story of Mr.Jensen

The following case story is authentic and taken from Aalborg Municipality. In order too keep it anonymous the patient is called "Mr.Jensen".

In 2006 was "Mr. Jensen" loaned an Induction loop. In October 2014, he was granted a stronger doorbell with puzzle call. In November 2014, he was given an outdoor rollator and a month after that a rail for self-lifting. In May 2015, an indoor rollator and two months later, he gets a shower stool and a GSM emergency call, if he falls over. At the beginning of 2018 he gets the doorsteps removed in his home.

Looking at Mr.Jensen's self-reliance it is clear that it has a negative development. The rate of which assistive technologies are granted is increasing from November and onwards. A timeline of this might look like this:

Timeline for conventional development of Assistive Technology needed
If a system could detect an unnormal development in loans of assitive technology or suggest a preventive rehabilitation program there is a chance that Mr.Jensen would not need that much help. This would yield an alternative timeline such as:
Timeline for new development of Assistive Technology needed

Danish Minister for Children and Social Affairs wants to Review the Management of Assistive Technology

In an article published the 11th of December 2018 by The Association of Danish Physiotherapists they talk about everyday problems in the management of assistive technologies. They argue that more citizens are often not granted the right assistive technology and thus more people file complaints to the council of appeal. In 2017 every 3rd reported complaint was successfull - this is a remarkle increase from the years before. Hence, the danish associations for physiotherapists and occupational therapists wrote to the minister regarding this topic. The minister agreed to discuss how assistive technologies are managed, but guaranteed no more money would be granted. You can read the answer here.

Media and Publicity

Here you can read what other medias say about the project.

"Artificial Intelligence can make Rehabilitaion Twice as Effective"

The 6th of December 2018 municipal health published an article about the results of phase one and future plans of the project. You will need to set up a free account to gain access, if you do not all ready have one.

"Data about Usage of Assistive Technologies Shall Improve Municipal Services"

The 6th of December 2018 municipal health published an article about the the possible benefits of the project. You will need to set up a free account to gain access, if you do not all ready have one.

Data

The data providers for this project are Aalborg Municipality and DigiRehab. To yield a brief overview of the databases is here a description of the datasets.

The Database for Assistive Technologies in Aalborg Municipality

This dataset consists of 8 columns as seen below:

Example of the database of assistive technologies

In total 662308 devices were registered in Aalborg Municipality from 1977-06-24 to 2018-11-15 (the day of data extraction). After cleaning the data for devices with the HMI number 899,999 equivalent to the name "Misc / deleted" the dataset consisted of 656775 registrations. A total of 43153 citizens, 9609 HMI numbers and 604 ISO classes were registered.

The Database for citizens in Aalborg Municipality enrolled in DigiRehab

Digirehab provided 4 sub-databases for screening values, status values, training done and traning cancelled as illustrated in the 4 charts below:

Screening values, are the records of screenings performed on the citizen and the corresponding exercises: Example of the Screening values database from DigiRehab

Status values, are the records of the state of a citizen's training program, e.g. active, paused, or terminated. This is automatically triggered by DigiRehab's system:
Example of the status values database from DigiRehab

Training cancelled, are the records of training sessions that were not completed:
Example of the Training Cancelled database from DigiRehab

Training done, the records of training sessions that were performed:
Example of the Training Done database from DigiRehab

In total 648 different citizens were enrolled in the DigiRehab program from 2015-01-10 to 2019-01-14 (the day of data extraction). After cleaning the data for lines with no date or IDs the dataset consisted of a 618 different citizens. These 618 citizens had a total 9440 recorded loans of assitive technoligies.

The Results of the Project

This section will be updated continously as new results are discovered.

Preliminary Results

In the data collection phase some preliminary experiments were conducted. It was identified that evaluated by self-reliance the mean increase from physical exercise was 14% when adjusting for sickness. More interestingly differences in the outcome based on assistive technologies were found much in compliance with the study from Fredericia Municipality. The results showed that citizens using rollators achieved the highest increase at 29% and citizens using tuning system the lowest at 13 % as seen in the bar chart below.

Self reliance based on Assistive Technologies

The Database for Assistive Technologies in Aalborg Municipality

As every other field of technology the assortment assistive devices changes every year, thus we filtered out everything before 2005 making the following descriptives based on 417352 registrations from 2005-2018.

The 15 most common ISO classes can be seen in this graph :
ISO classes and their duration sorted
From this it is clear that the most common device is a Topro rollator with 33104 and the follow up is a shower stool with 22009 occurences.

From the 417352 a total of 79837 did not have any return date. The number of times a device was not returned can be seen for every ISO class can be seen graph below:
ISO classes and their duration sorted
Whenever a device is returned a duration can be calculated. The mean duration and error for the 15 most common ISO classes can be seen in the graph below
ISO classes and their duration sorted
Here we see the top scores are again the Topro rollator and the shower stool with a mean duration of 707 and 663 days respectively. Notice the duration variates a alot. For this reason a plot of the fractiles might yield additional information. Looking at the fractiles it is seen that the much of the variation is coursed by 25% of the citizens. It is also seen that the median is much lower than mean values in most cases.
ISO classes and their duration sorted

To see the distribution of duration within each ISO class a histogram of the this for each ISO class is plotted below:
ISO classes and their duration sorted
What can be seen from this plot is that generally a device is used for 500 days and then returned. The probability using any of the Top 15 devices more than 3 years is low. This plot also yields an explanation for the variation seen in the durations. This also justifies looking a

The mean distribution of time across every ISO class can be seen in the following graph .
ISO classes and their duration sorted
Notice the big difference between the longest and the shortest duration and that most ISO classes has a duration between 0-2000 days.

When looking at which paragraphs were used every year we see from that §97 was the most common in 2005 and 2006 and §112 otherwise. In 2005-2014 a great amount of the loans were associated with a paragraph, but this changed in the years after. As seen in the graphs below.
Statistics of law registrations based on year
Statistics of law registrations based on year
Normally only 5 different paragraphs are used namely §112, §113, APV, §83a and §140. The reason that the distribution differs from this is in some case the ISO class numbers were recorded in the Paragraph field mistakenly. For instance in 2018 59 different paragraphs were used.