Woman patient and doctor wearing masks
Credit: iStock/nortonrsx/

Key Takeaways 

  • Patient-reported outcomes, plus machine learning, might be a reliable substitute for Clinical Disease Activity Index (CDAI) that requires input from a physician. 
  • This alternative approach might be useful in specific circumstances, such as during telehealth appointments.

If you have rheumatoid arthritis (RA), you may be familiar with one or more common assessments of disease activity. These tools, which include the Clinical Disease Activity Index (CDAI) and the Disease Activity Scale-28 (DAS28-ESR/CRP), tell your rheumatologist just how high or low your disease activity currently is. This information is often used to inform treatment decisions, such as whether you need to change your dose or try a new medication 

Conducting such assessments takes a little time, and it typically requires input from you as well as your doctor. In order to get your CDAI score, for example, your doctor must do a hands-on count of tender and swollen joints and fill out a global assessment survey that asks them to rate how they think you’re doing on a scale of 0-10. You, the patient, are also asked to rate the current impact of your RA per the same scale.  

Although such calculators are well-known and highly-regarded, a group of researchers led by Jeffrey R. Curtis, MD, a rheumatologist and professor at the University of Alabama at Birmingham, wondered if there might be an easier way. In particular, they wanted to find out what would happen if you took physician input out of the process entirely, and instead relied solely on patient-reported outcomes (PRO) and sophisticated machine learning technology. 

Machine learning is a type of artificial intelligence and computer science that uses data and algorithms to make predictions. 

To test the theory that patient data plus machine learning would be a reliable substitute for traditional disease activity measures like CDAI, Curtis and his colleagues used information from a large earlier study of RA patients who were starting golimumab (Simponi) or infliximab (Remicade) and aiming to get to low disease activity. They took the patient-reported data that was used in that trial and tried using various machine learning options to see which combination would most closely match what had been previously calculated using CDAI.  

According to their findings, a type of machine learning called “random forests” worked best.  

“Results from this proof-of-concept analysis demonstrated that [machine learning] methods applied to baseline clinical data measured at the time that a new biologic for RA was started, coupled with longitudinal data contributed by patients, were able to accurately classify subsequent RA disease activity as measured by the CDAI,” they explained.     

It should be noted that this was a “proof of concept study”— essentially, a pilot designed to test feasibility. The exact methods that were used aren’t quite ready for prime time. “Further validation with similar data sets derived from routine care settings” will be required, the authors wrote. 

Additionally, it’s worth mentioning that the goal of this alternate, patient-led approach isn’t to replace rheumatologists; rather, the scientists noted that it would be best used in specific circumstances in which using traditional assessments like CDAI would be complicated or impossible. For instance, it could be used to estimate disease activity and assess treatment response in between checkups, or it might be used during telehealth appointments in which the physician can’t physically palpate your joints.  

The researchers concluded that patient-reported data gathered over time plus machine learning methods “may be an effective proxy for clinician-derived disease activity data to accurately classify RA disease activity and assess treatment response.” 

 What This Means for You

Although traditional disease assessments that require physician input, such as CDAI, are still the gold standard, this study underscores the high value of patient feedback. The bottom line: You’re a crucial player in your care team. 

 Be Part of Research with ArthritisPower

Join CreakyJoints’ patient-centered research registry and participate in voluntary studies about managing arthritis. Learn more and sign up here.

Curtis, Jeffrey R., et al. “Machine Learning Applied to Patient‐Reported Outcomes to Classify Physician‐Derived Measures of Rheumatoid Arthritis Disease Activity.” ACR Open Rheumatology. October 11, 2022. doi: https://doi.org/10.1002/acr2.11499.

  • Was This Helpful?