Natural Course of Clinically Isolated Syndrome: A Longitudinal Analysis Using a Markov Model

Clinically isolated syndrome (CIS) refers to the initial clinical episode with symptoms suggestive of multiple sclerosis (MS). Due to limited number of long-term follow-up studies, progression pattern from CIS to more advanced stages remains unclear. In the current study, we constructed a Markov model to simulate the natural course of CIS. The model estimated the probabilities of transition from CIS to more advanced disease stages and the duration needed for the progression. The analysis showed: (1) CIS is a solid disease identity: more than 85% of the subjects with a diagnosis of CIS progress to RRMS or more advanced stages within 20 years; (2) the reduction of life expectancy in subjects with CIS is marginal.

by a lack of disease progression. " SPMS is defined as an "initial RR disease course followed by progression with or without occasional relapses, minor remissions, and plateaus. " Death is an absorbing state from which transition to other states cannot occur.
Patients diagnosed with CIS may eventually experience RRMS, SPMS and death 5 . A given subject could transition from one state to another (Fig. 1). For example, over a given time interval, a patient with CIS may remain in the CIS state, progress to the RRMS or SPMS state, or die. Similarly, a patient diagnosed with RRMS may remain in the RRMS state, progress to SPMS, or die over a time period, whereas a patient with SPMS may remain in the SPMS state or die.

Statistical analysis.
We constructed a Markov model that included three widely accepted states (CIS, RRMS, and SPMS), as well as the absorbing state (death) 18 . The Markov model was used to estimate the life expectancy of patients and the duration required for disease progression from one to another state 8,11 . Briefly, the model uses a stochastic processing method in which transitions and transition times from the current state are assumed to be conditionally independent of the previously occupied state 19 , based on the previously described principle 9 . For example, once a patient in the CIS state progresses to RRMS, the subsequent transition depends only on the current state (RRMS) and not the former state (CIS). To construct the Markov model of CIS, we focused on long-term population-based cohorts of patients with CIS, RRMS and SPMS.
We systematically reviewed published literature and obtained CIS transition probabilities from pre-existing natural history cohorts. Majority of the studies reported transition data over 10-year intervals. Thus, we used 10-year interval data for longitudinal analysis of CIS using the Markov model, specifying the unit time interval as 10 years. The probability of remaining at the same stage and that of transitioning to a different stage within 10 years were obtained from the published studies. The probability of transitioning from one state to another was calculated from the observations according to Beck 8,20 using the formula Rij = Nij/Ni, where Nij is the number of patients who transitioned from stage i to stage j during a specific interval, and Ni is the total number of patients who started in stage i. Table 1 summarizes the characteristics of the included studies. Table 2 provides the 10-year transition data and probabilities obtained from these published studies.
Diagnostic criteria: The diagnostic criteria of each natural history study are shown in Table 1. Briefly, the CIS cohorts were diagnosed as definite CIS or possible MS, whereas the Poser criteria do not include the diagnostic criteria of CIS. Among the 4 long-term follow-up studies of CIS, the CIS cohort was diagnosed according to "possible MS" in the study of Maja Eriksson 15 , meaning "patients who had characteristic symptoms such as unifocal optic neuritis, typical reversible sensory disturbance or internuclear ophthalmoplegia, without evidence of dissemination in space and time". In the remaining 3 studies 3,4,21 , "definite CIS", defined as "the first clinical episode in which a patient has symptoms and signs suggestive of MS, always isolated in time and also clinically isolated in  space" was used. Definitions of RRMS and SPMS were based on the Poser criteria and consistent across all studies included in the analysis 22 . We used TreeAge Pro 2011 software to construct a Markov tree (Fig. 2). The 10-year transition probabilities are shown in Table 2. We referred to the health utility values of the CIS, RRMS and SPMS states as uCIS, uRRMS and uSPMS, respectively. For estimation of average life expectancy, the health utility value was set at 10 for each

Duration of each state and life expectancy.
A Markov tree was constructed based on the possible progression of the disease (Fig. 2). The results revealed that over a 50-year period, patients initially diagnosed with CIS typically progressed to RRMS over a 10-year interval. The estimated duration of RRMS was approximately 15 years. The estimated duration of SPMS was approximately 21 years. The estimated life expectancy of the CIS patients was 46.6 years over a period of 50 years after CIS onset.
Transition probabilities. The estimated transition probabilities between the states over a 10-year interval and average life expectancy are reported in Table 3. We estimated the probability of remaining in the initial state, namely CIS, as well as the probability of transitioning to another state over successive decades. At

Discussion
With advances in disease-modifying treatments for MS, there has been great interest and research regarding patients with CIS. Much progress has been made in terms of understanding the cause, pathogenesis and risk factors of CIS [24][25][26][27] . CIS is a chronic disease with an extended trajectory. Also, patients with CIS typically recover from their presenting episode. Accordingly, thorough understanding of the natural course of CIS is needed to balance the potential benefits and adverse effects of disease-modifying treatments.
The Markov model is particularly suitable for simulating the natural course of CIS. The Markov model could also provide a comprehensive view of the disease process, and facilitate estimation of the proportions of individuals in different states at future time points and the duration of a particular state, thus allowing for efficient use of incomplete information when disease histories were available for only a small proportion of study participants.
CIS episodes are typically mild, and many individuals recover completely without therapeutic intervention 6 . Prediction of the long-term course of CIS from disease onset is challenging. In addition, the use of disease-modifying treatments for CIS is controversial because of the uncertainty concerning the long-term clinical prognosis and the benefits and adverse effects of various treatments, although such treatments have the potential to delay the transition from CIS to definite MS [27][28][29] . The results of the current study using a Markov model are generally consistent with the findings reported by previous longitudinal studies [3][4][5]30,31 . By 50 years after CIS onset, an estimated 69.4% (95%CI: 58.9-79.4%) of the patients had converted to SPMS, 12.4% (95%CI: 5.0-19.7%) had transitioned to RRMS, and 17.6% (95%CI: 8.9-25.7%) had died of complications. The cumulative life expectancy was estimated to be 46.6 years. Thus, CIS significantly reduces quality of life but has little impact on overall survival of the patients. European 32 and North American 33 placebo-controlled studies of disease-modifying treatments for SPMS have reported conflicting results. A 10-year follow-up study of an European multicenter randomized controlled trial of interferon beta (IFN-β)-1b failed to produce long-term benefits in SPMS patients 30 . Therefore, to prevent future tissue damage and to delay the conversion to CDMS in patients with CIS, the use of disease-modifying treatments should be considered, particularly for agents with favorable safety profile.
There are several limitations in the current study. The estimation is not based on a "true" model: prediction and simulation of the short-term course of CIS are limited due to a lack of original data on short-term transition probabilities. In addition, the paucity of data also prevents evaluation of the effects of potential confounding factors, such as age of onset and gender, on disease progression. Also, treatment is a critical and modifiable factor that could theoretically affect the outcome. Studies that elaborated the treatment, however, are typically short-term. At this point of time, we are not able to assess the treatment effects on disease progression. Thus, better predictors of long-term prognosis are urgently needed to enable early targeting of treatments for patients who are most likely to experience long-term therapeutic benefits. The lack of precise and effective long-term predictors is accompanied by a limited understanding of the mechanisms underlying the variable longitudinal course of CIS.