In a proof-of-concept analysis, researchers show that machine learning methods paired with longitudinal patient-reported outcomes (PRO) data were able to classify subsequent rheumatoid arthritis (RA) disease activity after beginning treatment with a biologic
Combining machine learning (ML) with patient-reported outcomes (PROs) may improve characterization of disease activity among patients with rheumatoid arthritis (RA) beginning treatment, found researchers of a new study published in ACR Open Rheumatology.
In a proof-of-concept analysis, researchers were able to show that ML methods paired with longitudinal patient-reported data were able to classify subsequent disease activity after beginning treatment with a biologic.
“These results have important applications for real-world evidence generation in as much as they provide a framework whereby patients might initiate a new medication at an office visit with a clinician, at which time the Clinical Disease Activity Index [CDAI] could be measured and follow-up information could be collected only from patients (eg, on a smartphone app) to assess treatment response,” explained the researchers. “Given the dramatic growth in telehealth-delivered care in rheumatology beginning in 2020, our findings potentially also have implications for improving the efficiency of remote patient monitoring.”
According to the researchers, all variables used in the ML models are available to rheumatologists in their electronic health record systems or are short PROs that can easily be captured in a remote patient monitoring program.
Among the 500 patients, all initiating treatment with either golimumab or infliximab, 36% achieved low-disease activity (LDA)—indicated by a CDAI score of 10 or less. The CDAI has 4 components: patient global, physician global, tender joint count, and swollen joint count. The group found that the positive predictive value (PPV) to accurately classify LDA among the patients exceeded 80% at a sensitivity rate of 60% or greater for the best performing models.
Among 8 PROs from the Patient-Reported Outcomes Measurement Information System (PROMIS) and the Short Form 36 (SF-36), several were considered useful for classification, although not including information from SF-36 had a minimal effect on model performance.
Although disease activity measures continue to be the most accepted measures in RA, PROs are increasingly being included as considerations of care. Across all classification models, PROMIS, baseline CDAI, and patient global were the most important features. PROMIS measures included social participation, pain interference, pain intensity, and physical function. The researchers noted that a significant number of patients did not have follow-up PRO data within the prespecified 1-month timeframe compared with the CDAI.
“Our findings add to a growing body of literature showing that in cohorts of patients with RA, PROs, such as those contained in the PROMIS system, correlate moderately or strongly with traditional measures of RA disease activity (eg, CDAI) and other patient-reported assessment,” wrote the researchers. “There will understandably be some discordance between patients’ and physicians’ ratings of RA disease activity. For example, in a large trial of 793 patients initiating certolizumab pegol randomized to management with the Routine Assessment of Patient Index Data 3 (RAPID3) versus the CDAI, the RAPID3 was found to be a less sensitive method to detect improvement in RA disease activity as measured by the Disease Activity Score in 28 joints.”
The researchers explained that data from that same trial showed that among patients with a phenotype characterized by depression, anxiety, and other features that was linked with concomitant fibromyalgia, there was a prominent difference in outcomes comparing management using RAPID3 and CDAI.
Reference
Curtis JR, Su Y, Black S, et al. Machine learning applied to patient-reported outcomes to classify physician-derived measures of rheumatoid arthritis disease activity. ACR Open Rheumatol. Published online October 11, 2022. doi:10.1002/acr2.11499
Laundromats as a New Frontier in Community Health, Medicaid Outreach
May 29th 2025Lindsey Leininger, PhD, and Allister Chang, MPA, highlight the potential of laundromats as accessible, community-based settings to support Medicaid outreach, foster trust, and connect families with essential health and social services.
Listen
Inside the Center's MDD Value Model and Its Use of Dynamic Pricing
May 13th 2025Larragem Raines, MS, of the Center for Innovation & Value Research, discusses the organization's major depressive disorder (MDD) open-source value model, dynamic pricing, and the future role of artificial intelligence in care.
Listen
Managed Care Reflections: A Q&A With Melinda B. Buntin, PhD
June 2nd 2025To mark the 30th anniversary of The American Journal of Managed Care (AJMC), each issue in 2025 includes a special feature: reflections from a thought leader on what has changed—and what has not—over the past 3 decades and what’s next for managed care. The June issue features a conversation with Melinda B. Buntin, PhD, a health economist and a Bloomberg Distinguished Professor at the Johns Hopkins Bloomberg School of Public Health and Carey Business School.
Read More
Anemia Linked With Higher Risk of All-Cause, CVD-Related Mortality in Rheumatoid Arthritis
May 30th 2025Iron deficiency anemia increased the risk of death from any cause, while non-iron deficiency anemia was linked with an increased risk of cardiovascular disease (CVD)-related death in rheumatoid arthritis.
Read More