Interview and article by Christine Clark
A recent Canadian study examining comorbidities before and after the diagnosis of rheumatoid arthritis (RA) used distance matrices to match RA cases to controls with similar medical histories
The methodology used in this study is common in industrial engineering and error-reduction, but has not been used in healthcare previously, according to lead researcher Dr Mark Tatangelo. It has become possible now that more powerful computer hardware and software is available. One of the strengths of this study is the reduced susceptibility to point and time bias because the researchers were able to control for these effects. “We also achieved a stronger predictive probability for patients. We matched patients with similar disease histories more closely than prior studies and the result of these methods improvements is that we could be more sure in our study that the comorbidities were attributable to RA”, he says.
Dr Tatangelo explains distance matrices and their application in this study as follows: “A distance matrix is a summary of all the population differences so when we talk about a distance matrix in a research study we are concerned about the differences between patients in our exposed group compared to a set of matched controls. So, to bring it back to RA, when we look at a patient with RA we can think of distance like similarities so if we match a patient with RA to a patients without RA, we can take their medical histories and their observed attributes and we can give the patient a similarity score (to the non-RA patient). When each case has a similarity score we can actually detect how similar an RA patient is to a non-RA patient and we can start to make determinations of disease severity. So, the more different patients are from non-diseased patients, the worse the RA. So, what distance matrices let us do is let us compare all our cases to the non-diseased population and choose the best controls based on similarity scores.”
The methodology is very versatile and has a diverse set of potential applications. The same approach has been used by Dr Tatangelo’s group for another study examining RA-attributable healthcare costs. “I would say that the methodology is limited only by the [researcher’s] data and imagination. Any disease or application with a set of prior data, a definable high-quality exposure definition and measurable follow up data could be a candidate for use with distance matrices”, he explains. The team is also considering other applications such as focussing on the individual level comorbidities and pre-disease histories of the patients, he adds
It is difficult to know how the covid-19 pandemic has impacted people with RA or RA services because healthcare services databases are always two-three years behind today’s date. The full impact will not be known until 2022 or 2023 once all data has been collected. Anecdotes suggest that there has been a reduction in clinic hours worldwide and a corresponding increase in the use of telemedicine. “I would say that patients with RA and all patients with any chronic disease have been overall negatively impacted by the covid-19 pandemic because of understandable human and resources limitations that no-one wants but they are the realities of a global crisis”, says Dr Tatangelo.
What is a distance matrix?
19.20 Matrix is a fancy word for a rectangle of data and a distance matrix is a summary of all the population diffs so when we tlak about a distance matrix in a research study we are concerned abou tteh differences between pts in our exposed group compared to a sert of mtched controls. So , to brign it back to RA, when we look at a patient with RA we can think if distance like similarities os if we match a patient with RA to a pt without RA, we can take their medical hispotries and their observed attributes and we can gove the patient a similarity score (to the non-RA pt. When each case has a similarity score we can actually detect how similar and RA pt is to non-RA pt and we ca start to make dtermiations of disease severity. So if an RA pt is mre similar to a non-dis pts … the mre diofferent they ar from non-diseased pts the worse the RA. So what distance matrices let us do is they let us compare all our cases to the non-diseased population and choose the best control based on similarity scores 21.05.