User login
A small number of patient and physician-reported outcomes, as well as laboratory and clinical factors, may help to predict the response of patients with ankylosing spondylitis (AS) to treatment with tumor necrosis factor (TNF) inhibitors when they have never taken them before, according to an analysis of data from nearly 2,000 individuals in 10 clinical trials.
TNF inhibitors are recommended for patients with AS whose symptoms persist despite use of NSAIDs, Runsheng Wang, MD, adjunct assistant professor at Columbia University Medical Center, New York, and a practicing rheumatologist at Garden State Rheumatology Consultants, Union, N.J., and colleagues wrote. Randomized, controlled clinical trials have shown that TNF inhibitors are effective in treating AS, but approximately half of patients fail to achieve notable improvement, which suggests the need for a predictive model.
“In clinical practice, before starting a treatment, physicians and patients want to know how likely a patient would be to respond to the treatment, particularly when more than one treatment option is available,” Dr. Wang said in an interview. “In this study, we developed predictive models that can potentially answer this question.”
The results suggest that the models in the study can be used to personalize clinical decision-making for patients with AS, whether to promote confidence in choosing a TNF inhibitor or to terminate treatment in nonresponders who had a higher probability of nonresponse at baseline, the researchers wrote. Similar models for other biologic treatments can help prioritize treatment options.
The predictive models are practical for clinical use because the variables in the reduced models – can be collected easily during patient visits, Dr. Wang explained. However, data from clinical practice are needed to further validate the study findings.
In a retrospective cohort study published in JAMA Network Open, the researchers analyzed data from 10 randomized, controlled clinical trials of TNF inhibitor treatment in patients with active AS conducted during 2002-2016. The study population included 1,899 adults with active AS who received an originator TNF inhibitor for at least 12 weeks, and the training set included 1,207 individuals. In the training set, the mean age of the participants was 39 years, and 75% were men.
The outcomes included major response and no response based on change in AS Disease Activity Score (ASDAS) from baseline to 12 weeks, and the researchers used machine-learning algorithms to estimate the probability of major response or no response. Major response was defined as a decrease in ASDAS of 2.0 or greater; no response was defined as a decrease in ASDAS of less than 1.1.
In the training set, a total of 407 patients (33.7%) had a major response, and 414 (34.3%) had no response.
The key features in the full, 21-variable model that increased the probability of a major response were higher C-reactive protein (CRP) levels, higher patient global assessment (PGA) of disease activity, and Bath AS Disease Activity Index (BASDAI) question 2 scores. (Question 2 asks for the overall level of back, hip, or neck pain associated with AS.) The probability of a major response decreased with higher body mass index and Bath AS Functional Index (BASFI) scores.
The key features in the model that increased the probability of no response were older age and higher BASFI scores. The probability of no response decreased with higher CRP levels, higher BASDAI question 2 scores, and higher PGA scores.
Overall, the researchers found that models using smaller subsets of variables (three or five variables in total) that would be easier to gather clinically yielded similar predictive performance.
The models were externally validated in a testing set of 692 individuals. Baseline characteristics were similar in the testing and training sets. In the testing set, the full models demonstrated moderate to high accuracy of 0.71 in the random forest model for major response and 0.76 in the random forest model for no response, with similar results in the reduced models.
At a prevalence of 25% for major response, the positive predictive values (PPVs) for random forest and logistic regression models ranged from 0.49 to 0.60, and the negative predictive values (NPVs) ranged from 0.82 to 0.84. At a prevalence of 25% for no response, PPVs ranged from 0.61 to 0.77, and NPVs ranged from 0.81 to 0.83.
The study findings were limited by several factors including the lack of data on smoking, which has been linked both to shorter treatment adherence and worse response to TNF inhibitors; the inclusion of only TNF inhibitor–naive patients; and the exclusion of NSAIDs from the models, the researchers wrote.
Dr. Wang disclosed support from the Rheumatology Research Foundation. The study’s two other authors disclosed receiving support from the Intramural Research Program of the National Institute of Arthritis and Musculoskeletal and Skin Diseases. The study was based on an analysis of data from AbbVie and Pfizer that were made available through Vivli.
A small number of patient and physician-reported outcomes, as well as laboratory and clinical factors, may help to predict the response of patients with ankylosing spondylitis (AS) to treatment with tumor necrosis factor (TNF) inhibitors when they have never taken them before, according to an analysis of data from nearly 2,000 individuals in 10 clinical trials.
TNF inhibitors are recommended for patients with AS whose symptoms persist despite use of NSAIDs, Runsheng Wang, MD, adjunct assistant professor at Columbia University Medical Center, New York, and a practicing rheumatologist at Garden State Rheumatology Consultants, Union, N.J., and colleagues wrote. Randomized, controlled clinical trials have shown that TNF inhibitors are effective in treating AS, but approximately half of patients fail to achieve notable improvement, which suggests the need for a predictive model.
“In clinical practice, before starting a treatment, physicians and patients want to know how likely a patient would be to respond to the treatment, particularly when more than one treatment option is available,” Dr. Wang said in an interview. “In this study, we developed predictive models that can potentially answer this question.”
The results suggest that the models in the study can be used to personalize clinical decision-making for patients with AS, whether to promote confidence in choosing a TNF inhibitor or to terminate treatment in nonresponders who had a higher probability of nonresponse at baseline, the researchers wrote. Similar models for other biologic treatments can help prioritize treatment options.
The predictive models are practical for clinical use because the variables in the reduced models – can be collected easily during patient visits, Dr. Wang explained. However, data from clinical practice are needed to further validate the study findings.
In a retrospective cohort study published in JAMA Network Open, the researchers analyzed data from 10 randomized, controlled clinical trials of TNF inhibitor treatment in patients with active AS conducted during 2002-2016. The study population included 1,899 adults with active AS who received an originator TNF inhibitor for at least 12 weeks, and the training set included 1,207 individuals. In the training set, the mean age of the participants was 39 years, and 75% were men.
The outcomes included major response and no response based on change in AS Disease Activity Score (ASDAS) from baseline to 12 weeks, and the researchers used machine-learning algorithms to estimate the probability of major response or no response. Major response was defined as a decrease in ASDAS of 2.0 or greater; no response was defined as a decrease in ASDAS of less than 1.1.
In the training set, a total of 407 patients (33.7%) had a major response, and 414 (34.3%) had no response.
The key features in the full, 21-variable model that increased the probability of a major response were higher C-reactive protein (CRP) levels, higher patient global assessment (PGA) of disease activity, and Bath AS Disease Activity Index (BASDAI) question 2 scores. (Question 2 asks for the overall level of back, hip, or neck pain associated with AS.) The probability of a major response decreased with higher body mass index and Bath AS Functional Index (BASFI) scores.
The key features in the model that increased the probability of no response were older age and higher BASFI scores. The probability of no response decreased with higher CRP levels, higher BASDAI question 2 scores, and higher PGA scores.
Overall, the researchers found that models using smaller subsets of variables (three or five variables in total) that would be easier to gather clinically yielded similar predictive performance.
The models were externally validated in a testing set of 692 individuals. Baseline characteristics were similar in the testing and training sets. In the testing set, the full models demonstrated moderate to high accuracy of 0.71 in the random forest model for major response and 0.76 in the random forest model for no response, with similar results in the reduced models.
At a prevalence of 25% for major response, the positive predictive values (PPVs) for random forest and logistic regression models ranged from 0.49 to 0.60, and the negative predictive values (NPVs) ranged from 0.82 to 0.84. At a prevalence of 25% for no response, PPVs ranged from 0.61 to 0.77, and NPVs ranged from 0.81 to 0.83.
The study findings were limited by several factors including the lack of data on smoking, which has been linked both to shorter treatment adherence and worse response to TNF inhibitors; the inclusion of only TNF inhibitor–naive patients; and the exclusion of NSAIDs from the models, the researchers wrote.
Dr. Wang disclosed support from the Rheumatology Research Foundation. The study’s two other authors disclosed receiving support from the Intramural Research Program of the National Institute of Arthritis and Musculoskeletal and Skin Diseases. The study was based on an analysis of data from AbbVie and Pfizer that were made available through Vivli.
A small number of patient and physician-reported outcomes, as well as laboratory and clinical factors, may help to predict the response of patients with ankylosing spondylitis (AS) to treatment with tumor necrosis factor (TNF) inhibitors when they have never taken them before, according to an analysis of data from nearly 2,000 individuals in 10 clinical trials.
TNF inhibitors are recommended for patients with AS whose symptoms persist despite use of NSAIDs, Runsheng Wang, MD, adjunct assistant professor at Columbia University Medical Center, New York, and a practicing rheumatologist at Garden State Rheumatology Consultants, Union, N.J., and colleagues wrote. Randomized, controlled clinical trials have shown that TNF inhibitors are effective in treating AS, but approximately half of patients fail to achieve notable improvement, which suggests the need for a predictive model.
“In clinical practice, before starting a treatment, physicians and patients want to know how likely a patient would be to respond to the treatment, particularly when more than one treatment option is available,” Dr. Wang said in an interview. “In this study, we developed predictive models that can potentially answer this question.”
The results suggest that the models in the study can be used to personalize clinical decision-making for patients with AS, whether to promote confidence in choosing a TNF inhibitor or to terminate treatment in nonresponders who had a higher probability of nonresponse at baseline, the researchers wrote. Similar models for other biologic treatments can help prioritize treatment options.
The predictive models are practical for clinical use because the variables in the reduced models – can be collected easily during patient visits, Dr. Wang explained. However, data from clinical practice are needed to further validate the study findings.
In a retrospective cohort study published in JAMA Network Open, the researchers analyzed data from 10 randomized, controlled clinical trials of TNF inhibitor treatment in patients with active AS conducted during 2002-2016. The study population included 1,899 adults with active AS who received an originator TNF inhibitor for at least 12 weeks, and the training set included 1,207 individuals. In the training set, the mean age of the participants was 39 years, and 75% were men.
The outcomes included major response and no response based on change in AS Disease Activity Score (ASDAS) from baseline to 12 weeks, and the researchers used machine-learning algorithms to estimate the probability of major response or no response. Major response was defined as a decrease in ASDAS of 2.0 or greater; no response was defined as a decrease in ASDAS of less than 1.1.
In the training set, a total of 407 patients (33.7%) had a major response, and 414 (34.3%) had no response.
The key features in the full, 21-variable model that increased the probability of a major response were higher C-reactive protein (CRP) levels, higher patient global assessment (PGA) of disease activity, and Bath AS Disease Activity Index (BASDAI) question 2 scores. (Question 2 asks for the overall level of back, hip, or neck pain associated with AS.) The probability of a major response decreased with higher body mass index and Bath AS Functional Index (BASFI) scores.
The key features in the model that increased the probability of no response were older age and higher BASFI scores. The probability of no response decreased with higher CRP levels, higher BASDAI question 2 scores, and higher PGA scores.
Overall, the researchers found that models using smaller subsets of variables (three or five variables in total) that would be easier to gather clinically yielded similar predictive performance.
The models were externally validated in a testing set of 692 individuals. Baseline characteristics were similar in the testing and training sets. In the testing set, the full models demonstrated moderate to high accuracy of 0.71 in the random forest model for major response and 0.76 in the random forest model for no response, with similar results in the reduced models.
At a prevalence of 25% for major response, the positive predictive values (PPVs) for random forest and logistic regression models ranged from 0.49 to 0.60, and the negative predictive values (NPVs) ranged from 0.82 to 0.84. At a prevalence of 25% for no response, PPVs ranged from 0.61 to 0.77, and NPVs ranged from 0.81 to 0.83.
The study findings were limited by several factors including the lack of data on smoking, which has been linked both to shorter treatment adherence and worse response to TNF inhibitors; the inclusion of only TNF inhibitor–naive patients; and the exclusion of NSAIDs from the models, the researchers wrote.
Dr. Wang disclosed support from the Rheumatology Research Foundation. The study’s two other authors disclosed receiving support from the Intramural Research Program of the National Institute of Arthritis and Musculoskeletal and Skin Diseases. The study was based on an analysis of data from AbbVie and Pfizer that were made available through Vivli.
FROM JAMA NETWORK OPEN