Artificial Intelligence and Clinical Decision Support - Frozen shoulder

David Reeves
published on July 18, 2019

Introduction

Clinical Decision Support tools provide medical practitioners with accurate and reliable information regarding a wide range of medical topics, conditions, diagnoses and therapies. Tools such as UpToDate, DynaMed and NHS Evidence now occupy a central position in the evidence-based medicine ecosystem, supporting clinical staff with the best available evidence from the biomedical research literature.

These tools are curated by expert panels, an approach which typically produces reliable results, but which is costly, time consuming, and susceptible to various forms of researcher bias. Importantly, the human in the loop factor introduces a notable lag for new evidence integration into a database. Dependent on an expert panel operation, it takes time for a new finding to be addressed, discussed and approved by such a panel. Developments in Artificial Intelligence technologies are offering new approaches to biomedical knowledge discovery, removing the need for a centralized, top-down system of knowledge provision, and placing the ability to discover and synthesize the medical evidence directly into the hands of the clinician.

We will investigate this new approach by comparing the Causaly AI engine with UpToDate, by looking at the example question, ‘What disorders are associated with frozen shoulder?’. The exact cause of frozen shoulder is uncertain and debated, which makes this a particularly interesting case study which also represents a potentially genuine topic that a clinician might wish to learn more about.

Method

Finding which disorders are associated with frozen shoulder is simple using Causaly. First, we conduct a straightforward search from the Causaly homepage:

frozen_shoulder_1

This takes us directly to the frozen shoulder overview page, which is our gateway to fully explore the topic:

frozen_shoulder_2

Scrolling down reveals the top 10 causes and effects of frozen shoulder, grouped according to category:

frozen_shoulder_3

Before examining the cause and effects module in more detail, it is worth noting that the frozen shoulder overview contains a whole host of other useful information relating to the condition. Causaly details new insights and epidemiological data derived directly from the literature, as well as providing details of the ontologies Causaly is using and some links to key articles.

Since we are interested in discovering the causes and associated conditions of frozen shoulder, we can click ‘see all results’ in the cause/effect module. The page displays all of the causes and effects of frozen shoulder that Causaly has been able to machine-read from the biomedical literature corpus. Using the filters to filter by ‘Disease or Syndrome’ brings us directly to the answer to our question, displayed graphically below using the Causaly dendrogram feature:

frozen_shoulder_4

Causaly allows us to interact with any of the listed associated conditions, to explore and examine the literature in which the relationship is discussed. We will explore some of these relationships later in this article.

In order to find the same information using UpToDate, we can conduct a simple search for ‘Frozen Shoulder’ from the UpToDate homepage. We are presented with the topic overview and can scan the document in the hope that the information we need has been included in the summary. In our case, UpToDate has provided some information about the causes/associated conditions in the Etiology and Pathophysiology sub-section, although it is unclear whether the detail is exhaustive. Evidence is referenced at the discretion of the overview author, meaning that we do not always have the ability to trace the literature that backs up the author’s assertions.

Results and Insights

The below table provides a direct comparison of the results using both approaches. In answering the question ‘what disorders are associated with frozen shoulder?’ both systems list diabetes, Parkinson’s disease, hyper/dyslipidaemia and stroke.

frozen_sholder_table

Causaly however finds evidence linking seven other conditions to frozen shoulder, none of which are discussed in the UpToDate overview, therefore giving a more comprehensive picture of the associated disorders. This could have real-life implications for practicing clinicians, who may otherwise not be aware of the links between, for example, Dupuytren’s disease, synovitis or akinesia and frozen shoulder. Consulting only UpToDate would not have revealed these links, meaning that a clinician may be making clinical decisions based on an incomplete picture of a topic.

Causaly allows clinicians to fully explore each of these relationships, unlike UpToDate. Taking the example of Dupuytren’s disease, clicking on this disorder brings up the detailed relationship view:

frozen_shoulder_5

Here, we are presented with every piece of evidence from within the biomedical literature that links frozen shoulder with Dupuytren’s disease in some way. There are four published papers detailing an association, with the actual text that describes the relationship highlighted by the Causaly engine. This takes the clinician not only directly to the literature, but directly to the text within the literature that answers the question. In our example, we can immediately see that:

Smith et al. (2001) note that:

Dupuytren's disease is 8.27 times more common in patients with frozen shoulder than in the general population, the difference between the two was highly statistically significant.

O’Gorman et al. (2010) concur, highlighting the fact that:

DD is associated with other fibroproliferative diseases, including Peyronie's disease, Lederhose disease, frozen shoulder syndrome, and desmoid tumor.

Interestingly, Causaly has found some evidence claiming the opposite, with a 1976 paper by Critchley et al. arguing that:

There was no direct relationship between Dupuytren's disease and frozen shoulder.

As with most medical topics, the research in this area is disputed and contradictory, but Causaly has allowed us to get to the heart of this evidence within a few minutes. A clinician who needs to dive deeper into the topic still, can click the article links to be taken through to the PubMed citation, from where full texts can be accessed, if available.

Dupuytren’s disease is just one example, there are six other conditions that Causaly details that are not mentioned by UpToDate, which a clinician could explore if necessary.

Conclusion

Literature reviews carried out directly using Pubmed and/or Embase offer the most comprehensive way to learn about a biomedical phenomenon. Most clinicians however, will not have the time, resources or access to the expertise necessary to conduct a full literature search necessary to find accurate and sufficient information. Using this very practical example of Frozen Shoulder Syndrome, we have demonstrated how curated resources, whilst being helpful, accurate and trusted by clinicians, might be limited in terms of the comprehensiveness and actuality of their information. Causaly is leveraging the power of AI to bridge these two methods, providing access to the entire body of biomedical knowledge, but with the ease and accessibility of a curated tool.

Advancing Clinical Care using AI: A case of Hemorrhagic Disease of Newborn
use case

Advancing Clinical Care using AI: A case of Hemorrhagic Disease of Newborn

Working as a pediatrician caring for ill, hospitalized children is a tremendously rewarding occupation, but getting it right can be hard.

New Insights in Renal Cell Carcinoma using Artificial Intelligence
use case

New Insights in Renal Cell Carcinoma using Artificial Intelligence

How to find newest insights on Renal Cell Carcinoma from academic literature

Using AI to answer clinical questions: COPD Case Studies
use case

Using AI to answer clinical questions: COPD Case Studies

Leverage Artificial Intelligence to assist clinicians to find answers to two COPD related clinical scenarios using the Causaly Platform

Be the first to know

Sign up for Causaly Newsletter