ARTICLE IN BRIEF
Researchers developed a set of algorithms that predicted with high or moderate accuracy a number of important aspects of treatment response to multiple sclerosis therapy, including disability progression, relapse frequency, and cumulative disease burden.
Predictive algorithms that draw upon widely available demographic and clinical characteristics may be used to estimate individual patients' responses to a variety of disease-modifying therapies (DMTs) for multiple sclerosis (MS), according to a study published in the August 3 online edition of Brain.
The algorithms have the potential to facilitate the process of choosing a DMT for patients with MS — a decision that can be difficult in part because responses to the same treatment can vary widely even among patients with similar baseline levels of disability and disease activity.
Developed by researchers at the CORe (Clinical Outcomes Research) Unit at the University of Melbourne, the current models were first tested on almost 9,000 patients participating in MSBase, a global study of patients with MS, and then externally validated using data on almost 3,000 patients in the geographically distinct Swedish Multiple Sclerosis registry.
The algorithms did not account for ease of administration or safety, which may be significant considerations for many patients, and some newer-generation DMTs, including the recently FDA-approved ocrelizumab (Ocrevus), were excluded. But they predicted with high or moderate accuracy a number of important aspects of treatment response, including disability progression, relapse frequency, and cumulative disease burden.
The models comprise “a tool that will contribute to clinicians' and patients' decision-making process in the clinic room,” said Tomas Kalincik, PhD, associate professor of neurology at the University of Melbourne and the Royal Melbourne Hospital, in an interview with Neurology Today.
“This is a stepping stone,” Dr. Kalincik said. “We want to expand the model to include more robust predictors, such as more detailed MRI information and biomarkers that can help predict response to treatment. By creating this model, we have established a framework in which we can test these predictors.”
STUDY DESIGN, FINDINGS
For the study, the researchers first established a training cohort of patients from the global MSBase cohort, a large research study of patients with multiple sclerosis. They identified 8,513 patients who had initiated a new disease-modifying therapy (DMT) during the study's prospectively recorded follow-up period and enrolled them in a training cohort. All patients had at least six months of pre-treatment and on-treatment follow-up data available, including at least two clinical visits at which their Extended Disability Status Score (EDSS), a measure of neurological disability, was assessed.
The researchers developed 42 algorithms that drew upon 27 demographic, clinical, and paraclinical predictors of response to seven widely available MS therapies: interferon b-1a intramuscular, interferon b-1a subcutaneous, interferon b-1b, glatiramer acetate, fingolimod (Gilenya), natalizumab (Tysabri), and mitoxantrone.
Response to treatment was analyzed separately for a range of factors, including disability progression or regression (defined as an increase or decrease of 0.5-1.5 EDSS steps, with higher scores indicating greater disability); relapse frequency; conversion to secondary-progressive MS (using the objective definition of secondary progressive MS previously published by the MSBase Group); change in the cumulative disease burden (defined as the annualized change in the area under EDSS-time curve relative to the pre-DMT EDSS score); and treatment discontinuation.
The researchers first tested the internal validity of the algorithms on a testing cohort of 1,196 patients with similar characteristics as those patients in the training cohort. They found the validity of the models was high (>80 percent) for relapse incidence during the first two years and for disability outcomes for the first four years; moderate for relapse incidence in years two to four and for the change in the cumulative disease burden; and low for conversion to secondary-progressive disease and treatment discontinuation.
Next, to assess the model's external validity, the researchers applied the algorithms to 2,945 eligible patients from the Swedish Multiple Sclerosis Registry, a large, population-based, geographically distinct national registry. Findings from external validation were similar to internal validation; the models showed high external validity for disability and relapse outcomes, moderate external validity for cumulative disease burden, and low external validity for conversion to secondary-progressive disease and treatment discontinuation.
Several associations were consistent across DMTs. For instance, the risk of disability progression was greater with older age, with secondary-progressive MS, in patients who had more severe disability, and in patients who had more pronounced gait impairment. In addition, a number of associations were treatment-specific, which enabled the researchers to differentiate between responders to different DMTs.
The strength of the algorithms, Dr. Kalincik said, comes from the method of double validation.
“This is not just a single analysis. We have done two sets of validation,” Dr. Kalincik said. “Very reassuringly, we saw that the predictive accuracy was high for the key outcomes over four years.”
“How do we explain that the model is more accurate in some outcomes?” Dr. Kalincik continued. “That is determined by the predictability of these outcomes. For example, relapses can be predicted relatively well over the initial two years, but years three and four are less predictable. That simply reflects the fact that inflammation in MS can change over time very rapidly. We saw low accuracy for discontinuation of therapy. Treatment discontinuation is governed by many different factors – it could be a lack of tolerability, a lack of efficacy, or convenience.”
The researchers noted several limitations to their study. Among them, many patients were missing brain and spinal MRI; the classification of MRI activity was simply present/absent; and MRI data were physician-reported, which could have made them subject to inter-rater error. Finally, their models did not include genetic or molecular information; more precise versions of the algorithm, they noted, will be “enabled by future inclusion of molecular or genetic markers in the existing models.”
“This is an important step toward personalizing therapies,” said Amit Bar-Or, MD, FRCPC, FAAN, the Melissa and Paul Anderson President's Distinguished Professor of Neurology at the University of Pennsylvania, in an interview with Neurology Today. “They thoughtfully took advantage of large datasets to obtain an initial putative set of parameters for their algorithms. To limit risks of false positives, they tested their model in independent cohorts which adds an important element of validation.”
He emphasized, however, that an important reason the paper title indicates “towards” personalized therapy is that “this study is still generating average data, not yet individualized data. They're saying, “‘based on averages, this is what we predict.’”
“What are the practical implications [for clinicians]?” Dr. Bar-Or asked. “Some caution should be exercised when translating these results to the individual level, as it may not be straightforward to plug these features into a predictive algorithm. Translating these findings into practice would also be impacted by whether data is collected as it was in this study. Clinicians may need to be more deliberate about how they collect such data.”
Timothy Vollmer, MD, FAAN, professor and vice chair of clinical research in the department of neurology at the University of Colorado School of Medicine, had a different view, however. “This paper says all these drugs are different in some way. It's reinforcing the concept of escalation therapy – this concept that we'll start with old drugs, see how they do, and wait for those to fail before moving on to the newer orals and infusibles.”
Predictive algorithms may not be necessary in an era of highly successful therapies, Dr. Vollmer said. “We've moved into an era where with appropriate patient selection and monitoring, we have several therapies that, in the vast majority of patients, can terminate acute inflammatory relapses at a high rate. We have the ability to put them into full remission, not just from relapses, but symptoms. If we focus on early diagnosis and intervention, we can dramatically improve outcomes.”
Future studies of personalized approaches to MS treatment could also consider features beyond clinically defined measures, Dr. Bar-Or said.
“So much of MS injury can take place under the surface. A person who may seem to be doing fine may actually be experiencing substantial damage that is not appreciated using clinical measures alone,” he said. “To move personalized medicine to the next level, we eventually need to increase our ability to biologically assess what's happening with these people, including under the surface.”
“Clinical heterogeneity in MS comes not only from differences in the disease across individuals but also differences in the individual's general immune response profile,” Dr. Bar-Or continued. “Similarly, a given drug may work differently across individuals because of differences in the person's background immune system, not only differences in disease characteristics. Not all of how MS manifests is because of the MS – it is also about particularities of the person who has MS. There's room to capture, in greater depth, what makes a person an individual in a biological sense. I could imagine future approaches to precision medicine will involve integration of an individual's clinical, imaging, and biological information, as well as individual preferences.”
Addressing these concerns, Dr. Kalincik said, “We are very realistic about the place of this study in the current treatment landscape. This study is entirely focused on efficacy, but choosing a therapy is not all about efficacy. People also focus on convenience of administration and safety. We will continue to collect data from observational studies that will be used to further develop the model,” he noted. The predictive algorithm will be made freely available to clinicians through the MSBase website early next year.
EXPERTS: ON ALGORITHMS FOR PERSONALIZED MS THERAPY
LINK UP FOR MORE INFORMATION:
© 2017 American Academy of Neurology
•. Kalincik T, Manouchehrinia A, Sobisek L, et al Towards personalized therapy for multiple sclerosis: Predicion of individual treatment response https://academic.oup.com/brain/article-abstract/4061515/Towards-personalized-therapy-for-multiple. Brain
2017; Epub 2017 Aug 3.
Neurology Today Quick Links