Survival and predictors of disability in Turkish MS patients

The decision to treat patients with expensive, long-term treatments is an important consideration in MS. This decision is more complicated because there is an increased tendency to prescribe these treatments early in the course of the disease. Defining the natural history and early and long-term prognostic indicators of the disease course may influence the decision of whether to treat patients with expensive, long-term therapies. Many studies concerning the natural history and prognostic factors involved in determining the outcome in MS have been published.^1-19 Although they address the same problem, most of these studies were based on different populations with different geographic backgrounds. In such a clinically heterogeneous disease as MS, the course and prognosis are expected to vary among clinically distinct syndromes and in different epidemiologic groups. To clarify the observed differences, the methodologies involved should be comparable. In 1987, Weinshenker and Ebers¹ reviewed the guidelines for studying natural history in MS, and published their work based on this approach.^2-5 Most of the previous studies, however, used different patient sampling and statistical methods(hospital-based versus geographically based series and univariate versus multivariate analysis with or without survival methods for proper censoring). Hence, despite some consensus, there remains some points of controversy regarding the influence of gender, age at onset, favorable onset symptoms, and early attack rates on prognosis and survival.^6-14 There are a number of other published studies^15-18 with comparable results, and a general agreement has been reached as later reviewed by Weinshenker.¹⁹ Female sex, a younger age at onset, onset with optic neuritis, a relapsing course with no progression, and a low level of disability within the first 5 years are relative predictors of a long-term, favorable outcome.^4,16,17

Based on the previous studies, as a general indicator of disability and survival, 50% of the patients developed a moderate disability of 6 points in the Expanded Disability Status Scale (EDSS) of Kurtzke^20,21 after 15 years, and less then 20% died after 25 years from onset of disease.^2,15,17 Although survival is comparable among different populations of MS, it is a very long-term outcome measure, considering prognostic factor studies and treatment trials. On the other hand, disability alone is not an actual measure of the impact of disease if duration of disease is not considered. Thus the concept of the progression index was introduced. This concept has certain drawbacks because most of the data involved in such studies are censored (i.e., patients are not followed for equal lengths of time), and the nonlinear, ordinal nature of the most commonly used disability scales such as the EDSS provide an unrealistic measure when compounded by a division with a linear, numeric variable such as time. A better approach is to stratify data into duration of disease to the last recorded disability and to apply survival statistics that are more suitable for censored data.

Based on the results of earlier studies,^22-24 a multicenter study group was formed in 1994 to study the prognostic factors, clinical course, and survival in Turkish MS patients. A standardized database was designed. Using the Turkish Multiple Sclerosis Study Group (TUMSSG) database, we investigated survival, effectors, and development of disability in a sample of MS patients in Turkey.

Methods. Data collection. The study was conducted using the TUMSSG database, which consisted of collaborative data from nine MS centers. The database was developed in 1994, and it stored the data recorded via a standardized, two-page entry form,²² which was completed by neurologists who followed MS patients in different centers. Both inpatient and outpatient subjects were included. The forms were completed retrospectively, based on the patient files at the beginning of the study, and, from April 1994 to June 1997 (the 3 years during which the study continued), all the new entries and the old ones were followed prospectively. Each patient form was updated every year to assess completeness. To avoid errors in completion and repetitive entries of the same patients by different centers due to patients seeking second or even third opinions, all of the data were rechecked every time new group entries and updates were made. By the end of the third year of the study, of 1,560 entries, 1,259 patients with clinically or laboratory definite MS were identified, according to the criteria of Poser et al.,²⁵ and completed data were clarified if any errors were indicated.

Data studied were demographic aspects (such as gender, age at onset, age when last seen, and diagnostic certainty), clinical aspects (such as initial symptoms and mono- or polyregional onset [as diagnosed clinically at onset by the evaluating physician without the aid of neuro-imaging; for example, myelopathy would be assigned as monoregional although polysymptomatic]), temporal course with type (relapsing-remitting MS [RR-MS], primary progressive MS [PP-MS], and secondary progressive MS [SP-MS]), and number of attacks within the first, third, and fifth years when available. To prevent unreliable retrospective recording of the onset symptoms, symptoms such as dysarthria and nystagmus (as described by the patients), which were suggestive of cerebellar or brainstem involvement but were not verified by an examiner as separate entities, were grouped. Similarly, urinary or fecal incontinence, urinary frequency or urgency without other symptoms of cystitis, and urinary retention were included within the single category of sphincter problems. Outcome was measured as disability using the EDSS²⁰ at the final visit while the patient was in an unexacerbated state. Because the retrospective data in files could be unreliable, time (in years) to successive levels of disability using the EDSS could not be studied.

None of the patients were taking long-term immuno-modulatory or immunosuppressive treatment from the onset of disease, but some were treated later either because of a quickly progressive course or a high attack rate. Interferon beta was used in 2.2% of patients, but all of these patients had received this drug for a period of less than 1 year because it was registered in Turkey within a year of study completion. On the other hand, 5.4% of patients received azathioprine, and 1.9% of patients received methotrexate for a period of longer than 6 months but less than 3 years. Because neither azathioprine²⁶ nor methotrexate²⁷ have been shown to influence significantly the progression of disability, and exclusion of these patients with a possibly worse prognosis would cause a selection bias in favor of more benign cases of MS, these patients were included in the study group.

Analysis of results. Because only a sample of the MS population in Turkey was included, a two-step strategy was used to test for possible ascertainment and selection bias. For the first step, the patient pool was divided into two groups: patients with exclusively retrospective data with prospective follow-up (RO; >3 years disease duration) and patients seen from onset with prospective follow-up (SO; ≤3 years disease duration). The two groups were compared for demographics, clinical aspects, and temporal course. For the second step, data from different centers were compared with each other and with the whole group for demographic characteristics, clinical aspects, temporal course, and outcome.

In the final analysis, to test for the comparability of the study population with other previous studies, actuarial analyses of selected disability levels of EDSS scores equal to or greater than 3, 6, 8, and 10 were performed for the total patient population. Following this, using EDSS≥ 6 as the survival end point, analyses of possible demographic, clinical, and temporal course factors as predictors of outcome were performed. Two models were constructed. In model 1, attack rate was not included and, for simplification, both primary and secondary progressive courses were classified as "progression." In model 2, attack rate was included, and primary progression was excluded because PP-MS patients did not have any recorded attacks because of the format of the database and, consequently, could potentially obscure a significant effect of relapse rate in the RR-MS patients. Because duration of disease is crucial in the outcome and our data are censored in regard to time, a forward, stepwise Cox proportional hazards regression analysis²⁸ was preferred. Initially, this method allows for the study of all independent variables univariately, and at each successive step the most significant factor is excluded. The procedure is repeated for several steps, reanalyzing the remaining factors each time and until no more variables are left with explanatory significance. The variables of significance from the previous steps are entered successively(enter, p = 0.05; remove, p = 0.1) into a separate regression analysis and only those factors that retain their significance are analyzed together in the last step. Besides the whole group, the Cox regression analysis was applied after the data were stratified into three groups according to disease duration of ≥5, ≥10, and ≥15 years, because <5 years is not an adequate duration for deciding the course of the disease. Pearson's correlation coefficient was calculated, crossing all factors with each other. As a last step, actuarial curves were constructed for the factors found to be associated significantly (p < 0.05) with outcome in the multivariate analysis, and a Mantel-Cox logrank test²⁹ was applied to determine the statistical significance of the existing differences.

Results. Demographic and clinical aspects. The demographic and clinical aspects of the whole group and the subgroups are presented in table 1. To test for possible ascertainment bias, the results of the subgroups were compared with χ² statistics. The mean age at onset of disease for the whole group was 27.6± 8.8 (SD) years. The mean duration of disease was 8.44 ± 6.7 years. In the whole group, 81% of the patients were diagnosed as clinically definite MS and the remaining patients were diagnosed with laboratory definite MS.

The ratio of polyregional to monoregional onset was 0.66. Sensory and motor symptoms were the most common presenting symptoms. No significant difference was observed among the two subgroups of SO patients and RO patients. At this stage the major contributing centers (six centers with more than 100 patient contributions) were compared for the frequency of symptoms. No intercenter variation was observed (data not shown).

In this study the clinical course was grouped into three types as noted at the last evaluation of the patient: RR-MS, PP-MS, and SP-MS. As seen intable 1, the difference among the SO and RO groups is very significant (p < 0.0001). Comparison of centers also for subtypes of clinical course, on the other hand, did not reveal any difference(data not shown).

Considering these demographic and clinical findings, no significant difference was observed among the RO and SO groups except for the course of the disease. It can be concluded that the data recorded retrospectively were reliable it terms of demographics and onset characteristics, and multifactorial analysis of prognostic factors can be done confidently. On the other hand, because patients with a short follow-up could underestimate some of the long-term effects of prognostic factors, and because the differences in the proportion of patients with progressive disease within the two groups is dependent partly on the duration of follow-up, stratification of data into≥5, ≥10, and ≥15 years would be necessary for the succeeding regression analysis.

Survival and disability. Results of the survival analysis are represented in figure 1 by a Kaplan-Meier survival curve with SEMs for each level shown. The mean survival was 36.13 ± 0.97(SE) years for the whole group. The survival (± SE) at 15 years from onset was 94.6 ± 2.9%, and at 25 years from onset was 89.0 ± 5.8%. Because the number of patients followed beyond 15 years is 209 and the number followed beyond 25 years is only 40, the SE increases and therefore the estimate of survival becomes unreliable.

The results of the actuarial analysis of the time to reach several selected disability levels (3, 6, 8, and 10) are presented infigure 2. By 15 years, 66.3% of the patients were at EDSS ≥ 3 and 41.2% were at EDSS ≥ 6. By 25 years, these values increased to 89.9% at EDSS ≥ 3 and 69.6% at EDSS ≥ 6.

Prognostic factors. The multivariate, forward, stepwise Cox regression analyses applied to the prognostic factors involved in determining disability censored in time are summarized in tables 2 through 4. In table 2, results of the first step including all the variables for both models 1 and 2 are shown. The group stratified into ≥10 years duration was omitted because the results were similar to that of the group with a duration of ≥15 years. The proportion of censored patients decreases from 74% to 55% if only those patients with a disease duration of ≥15 years are considered. The total number of patients included in model 2 is low because the PP-MS patients and entries with incomplete data for number of attacks are excluded from this analysis.

During the first step of the analysis, male sex, a higher onset age, polyregional onset, onset with motor or sphincter symptoms, a greater first 5-year attack rate, and progressive disease were associated univariately with a poor outcome. Onset with optic neuritis correlated negatively (i.e., favorably) with outcome (r = -0.0404, p = 0.152), but this was not significant statistically (p > 0.05).

During the final steps of the analyses (see tables 3 and 4) of the nonstratified total population, older age at onset, onset with sphincter symptoms, and progressive disease were found to be predictors of a poor outcome. During the last step, RR values were calculated for each variable found to be predictive significantly. For the total population, the RR of reaching an EDSS score of 6 for onset age was 1.02 with a 95% CI of 1.00 to 1.03. Because the interval included 1.00, the significance was questioned, and when only ≥5 year group was considered, onset age was no longer significant.

In the ≥5 years group, considering model 1, sphincter symptoms (RR, 1.86; CI, 1.23 to 2.82) and progressive disease (RR, 3.73; CI, 2.71 to 5.13) were strong predictors of outcome. In model 2, when attack rates were included and the PP-MS patients were excluded, sphincter symptoms (RR, 2.51; CI, 1.39 to 4.52) became even stronger predictors, whereas secondary progression (RR, 2.48; CI, 1.67 to 3.68) by itself was not as strong as progression in general, hence suggesting primary progression as a worse prognostic factor. In model 2, although not as strong as the other two variables, number of attacks within the first 5 years (RR, 1.2; CI, 1.08 to 1.34) and male sex (RR, 1.55; CI, 1.06 to 2.27) were also distinguished as poor prognostic factors. Because male sex was not distinguished in model 1 but only in model 2, it was assumed to be associated positively with a primary progressive course and was not disclosed as an independent variable until the removal of the strong prognostic effect of early progression. This was further tested by Pearson's correlation coefficient. Male sex correlated significantly and positively with primary progression (r = 0.0895, p = 0.001) and negatively with relapsing disease (r = -0.0956, p = 0.001), whereas no significant correlation was present with the 5-year cumulative number of attacks (r = -0.0016, p = 0.969) nor with secondary progression (r = 0.0367, p = 0.194). Number of attacks within the first 5 years was higher in the sphincter onset group and hence correlated positively (r = 0.0858, p= 0.041).

In the ≥15 years group, all the other factors found previously to be important predictors except progression (primary or secondary) and male sex lost significance, proving that even if the sample is biased by the lower number of patients (n = 146), these two factors still emerge as strong predictors of outcome.

There were other notable correlations observed among different prognostic factors. Patients with optic neuritis as an onset symptom, for example, had an earlier age at onset as shown by the significant negative correlation(r = -0.1214, p < 0.001). An older onset age correlated positively with a primary progressive course (r = 0.1807, p < 0.001) and negatively with a relapsing course(r = -0.1154, p < 0.001). Polyregional onset correlated either negatively or positively with nearly all of the factors significantly (p < 0.05), except relapsing course and EDSS. This further proves that being clinically mono- or polyregional at onset is not an independent determinant of disease outcome. Motor symptoms correlated strongly with outcome (r = 0.0743, p = 0.008), but also positively with sphincter symptoms at onset (r = 0.2362, p < 0.001), with primary progressive disease (r = 0.001, p < 0.001), and with male sex (r = 0.0573, p < 0.05). These multiple correlations with factors that were associated significantly with outcome in the multivariate analysis explain why having motor symptoms at onset has no significant effect by itself. On the other hand, the significant positive correlation of sphincter(r = 0.1001, p < 0.001) and motor symptoms at onset with primary progression (r = 0.1433, p < 0.001), and a positive correlation with each other (r = 0.2362, p < 0.001) was suggestive of a behavioral variant of the disease.

For the final step in our study, actuarial curves for EDSS ≥ 6 were constructed for the factors found to be associated significantly with outcome. Figure 3 shows the curves for disease course and for brevity, actuarial analysis of gender, sphincter symptoms, or motor symptoms at onset are only discussed within the test. The curve for SP-MS patients is very similar to that of the whole population (compared withfigure 2), and RR-MS patients do significantly better(χ² = 57.86, p < 0.001). The PP-MS patients do worse within the first 15 years, but after that there is no difference whether progression is primary or secondary (χ² = 2.79, p not significant). As mentioned previously, the number of patients followed for more than 15 years is low (n = 209), and this might be a bias that accounts for the observed loss of significance. Having sphincter symptoms at onset was a significantly worse prognostic factor when compared with the actuarial method (χ² = 10.4, p < 0.01) as well. Actuarial analysis of male versus female gender did not show any significant difference (χ² = 1.17, p not significant), nor did having motor symptoms at onset (χ² = 1.34, p not significant).

Discussion. The demographic and clinical aspects of our study group are consistent with most of the earlier studies but with a higher male-to-female ratio (0.56). The ratio of polyregional to monoregional onset(0.66) is very similar to that presented by Riise et al.³⁰ A major difference observed in Turkish MS patients is in the motor symptoms, which are less frequent than those found in Hordaland and Vestfold, Norway,³⁰ and two times more common than those in Ontario, Canada,² and Dublin, Ireland.¹¹ Frequencies of optic neuritis, sensory symptoms, and brainstem and cerebellar symptoms at onset are similar to those found in the Canadian series² and are higher than those reported in the Norwegian series.³⁰ The results of the proportion of patients with different courses are comparable with those presented in the cohort studied in Olmsted County, MN,³¹ where 58% of patients were found to have RR-MS and 14% were found to have PP-MS. In our study, the frequency of patients having a primary progressive course in the RO group (12.9%) dropped to 6.5% in the SO group (<3 years disease duration). This is either because PP-MS patients may have an initial attack followed by progression that is not clearly evident at the beginning, or relapsing progressive patients are not recognized as progressive early in the disease. Consequently, primary progressive disease can take longer to diagnose, thus with a very short disease duration its proportion may be underscored. The dramatically lower frequency observed for SP-MS patients in the SO group, on the other hand, is easily understood because most of the patients expected to progress in later years are assigned as relapsing early in the course. This ascertainment bias was overcome by considering those patients who had a disease duration of 5 years or longer besides the total population in further analyses. Exclusion of patients with >5 years' disease duration actually eliminates the whole SO group, which has a maximum duration of 3 years, and thus avoids a possible error introduced by including this subgroup with a very high proportion of RR-MS patients (90.4%). Although there were no significant differences in demographic and clinical aspects of the data contributed by individual centers, concluding that data collection was not biased by selection, this is not a population-based study and there could still be a systematic bias as a whole, because all of the patients within the population are not assessed.

The mean survival of the studied Turkish MS population is similar to the median 38 years in the German study³² but it is longer than the mean of 27 years in the Norwegian study³⁰ and the mean of 25 years in the Scottish MS populations.¹⁵ In our study less than 5% of the patients are expected to be dead by 15 years, and less than 12% by 25 years. These survival results are higher than those found by Runmarker and Andersen,¹⁷ although they are similar to those found by Weinshenker et al.² The actuarial curves for disability yielded very similar results for both of the studies, but with a slightly better prognosis, especially at lower levels of disability (i.e., EDSS < 6). It can be concluded that by 15 years from onset, 41% of the Turkish MS patients are expected to reach a disability level of 6 or beyond and 5% are expected to die.

The Cox regression analysis allowed us to weigh the effect of each prognostic factor individually and together, accounting for the covariations and compensating for censoring. On the other hand, studying data with ≥15 years disease duration also allowed us to study the effects of each factor in the long term. The actuarial analysis of the factors provided further proof of the results of the multivariate regression analysis.

The prognostic variables associated with a poor outcome are male sex, onset with sphincter symptoms, a higher first 5-year attack rate, and a progressive course (primary worse than secondary). In this study, sphincter symptoms at onset and, as expected, a primary progressive course are strong predictors of a poor outcome. Male sex is strongly correlated with primary progression, which is the strongest prognostic factor and, as can be seen in the final results from the Cox regression analysis of both models, only by exclusion of primary progression is male sex unmasked to become a significant predictor by itself. A relapsing course with no secondary progression is favorable, and primary progression is worse than secondary progression only in the short term (i.e., within 15 years disease duration), although this similarity between primary and secondary progression in the long term might be because of a bias introduced by the lower number of patients followed over 15 years. An older age at onset might also be an unfavorable prognostic factor very early in the course because of being correlated positively with progression and correlated negatively with a more favorable relapsing disease. However, when only patients with a disease duration of 5 years or longer are considered to avoid ascertainment bias, no association is found. The significant correlation of motor symptoms with sphincter symptoms at onset is due to grouping these two symptoms. Primary progressive disease has a worse prognosis and a later onset, and a positive correlation seen by clustering sphincter and motor symptoms can be seen as a behavioral variant of the disease, compared with the RR-MS and the SP-MS patients, who have an earlier onset, a better prognosis, and a negative correlation with sphincter symptoms. These results are in agreement with several previous studies, suggesting a primary progressive "spinal" involvement as a possibly distinct clinical form of the disease, as reviewed recently by Thompson et al.³³

The observed correlation of sphincter symptoms at onset with a higher attack rate might be because sphincter problems are the major sources of urinary infection and hence a more frequent exacerbation or further subclinical disease activity. This hypothetical correlation requires further testing.

In this study, a late onset, although being univariately associated with a poor outcome, is not as strong a prognostic factor in the multifactorial approach as it was in other studies.^4,16,17,34 Our conclusions for gender and initial symptoms are similar to those found by Phadke¹⁶ in Scotland and by Riise et al.³⁴ in a European multicenter study. Also, the strong effect of progressive course together with the loss of effect of the gender and previously effective onset symptoms after 15 years is in consensus with the results of the 25-year follow-up study by Runmarker and Andersen.¹⁷ Onset with sphincter symptoms is unfavorable, and this effect was not defined clearly in the previous studies. A higher number of attacks within the first 5 years is unfavorable, whereas brainstem and cerebellar symptoms have no effect-a conclusion that contradicts the results of Weinshenker et al.⁴ This, however, might be because we grouped brainstem symptoms with cerebellar symptoms, masking their individual effects. The significant prognostic effect of motor symptoms at onset observed in some of the earlier studies^6,7,12 is not proved. Also, the previously defined unfavorable effect of polyregional onset of the disease in Sweden¹⁷ is not found in our study, but because there is a large discrepancy between the observed ratios of polyregional to monoregional onset (i.e., 0.12 to 0.66) between the studies, a definitional difference is assumed to account for it. Otherwise, the similarity of our results from a censored sample population with an 8.44-year mean disease duration to those observed in the 25-year complete follow-up study in Sweden¹⁷ shows that the predictive values of factors as analyzed by the Cox regression models are reliable.

Although representing a sample with a different epidemiologic background, with some controversies as discussed, the results regarding the prognostic indicators of outcome and survival in Turkish MS patients are similar to those presented in other previous Western series.¹⁹ Considering patient selection for early intervention with long-term treatments, a decision should be based on the early clinical findings that suggest those patients that are expected to have a worse prognosis throughout the disease course. For example, a male patient with sphincter symptoms at onset and a higher early attack rate or a progressive course is expected to have the worst prognosis, and hence is a more appropriate candidate for earlier intervention with expensive, long-term treatments. This kind of first-line clinical decision may be supported with imaging and immunologic studies to establish an even stronger basis for available treatments or treatment trials, but the prognostic effects of laboratory data are yet to be defined.

Appendix