We enrolled 95 consecutive patients admitted to the Novara and Cuneo MS Centers from January 2001 (Cuneo from May 2002) to December 2003. Both centers are first-referral facilities which a total catchment area of 600,000 persons. We included patients presenting with CIS or RR-MS. CIS was defined as the occurrence of an acute or subacute event of the CNS affecting the optic nerves, brainstem or spinal cord of presumed inflammatory demyelinating origin in a patient with no history suggestive of an earlier demyelinating episode [9]. McDonald criteria [10] were used to diagnose MS. Primary progressive MS patients were excluded.

The cohort comprised 51 women and 44 men with a mean ages of 35.4 years (SD 10.6) and 31.4 (9.2) (p<0.05) at the first attack. The median interval between first symptom and visit was 51 days, (range 0-810). Patients were enrolled on their first visit to one of the two Centers. This corresponded to the first attack in 55 cases (58%). The others were observed 1 to 27 months after the first attack (only 13 during or after the second). All patients had MRI; cerebrospinal fluid (CSF) was examined in 79. Other examinations were performed to rule out an alternative diagnosis. After the first visit, patients were clinically evaluated every 6 months until the end of the follow-up (July 31, 2005). During each visit, they were scored according to the Expanded Disability Status Scale (EDSS) [11]. Follow-up ranged from 19 to 55 months (mean 36.8; SD 10.2). At the end of follow-up, 21 were still CIS and 71 RR-MS; 3 patients had already entered secondary progression. For patients who presented a relapse in the last semester of the follow-up, an extra 12-month period (after July 31) was added to ascertain recovery (not considered in calculation of the mean follow-up).

Patients were instructed to phone the MS Center on the appearance of new symptoms and a visit was arranged in no more than 3 days. Further visits were planned after 3 and 5 days, and 1 month after the attack onset. Each attack was evaluated by an examiner neurologist blind to the treating neurologist. Each patient was provided at the beginning of the attack with a diary, derived from Liu [12] and asked to indicate each symptom and circle each day in which it was present. Diaries were collected at the end of the attack, and their information was integrated with that from the interview. Data on attacks not directly observed by us (anamnestic attacks, N=77, 40 first) were collected from hospital discharge letters, emergency departments and outpatient neurological reports, and integrated through interviews of patients and their relatives. Decision whether or not to treat an attack (either orally or intravenously) was left to the patient’s physician. Seventeen patients were started on disease-modifying therapy (15 interferon, 2 glatiramer acetate) during the follow-up period, only in two cases before the second attack. The study was approved by the Ethical Committee of our Hospital; written informed consent was obtained to participate in the study and maintain data in the data-base.

Attack was defined as “acute or subacute occurrence, recurrence or worsening of symptoms of neurologic dysfunction attributable to MS, lasting more than 24 hours after a period of at least 30 days of improvement or stability” [13]. Neurological deterioration of pre-existing symptoms accompanied by fever or concomitant infection (pseudoexcerbation) was not considered a attack. Paroxysmal episodes were considered as attacks when multiple and occurring over not less than 24 hours. Symptoms occurring within a month after the initial symptoms of an attack were considered to be part of the same episode [10]. For each attack, type of symptoms and signs and dates of occurrence were collected on an ad-hoc form, derived from Trojano et al [14]. Symptoms and signs were grouped to fit in with the Kurtzke Functional Systems (FS) [11]. For each symptom/sign we defined the date of onset, maximum worsening (zenith), start of improvement, and maximum improvement, defined as the date after which no further subjective improvement was noted. The score for each FS and the total EDSS score were calculated at onset, at the zenith and at the first examination after the day of maximum improvement. Agreement between the “attack” examiners was calculated with the Kappa statistics and varied from 0.7 to 0.9 for each FS [1].

Severity was graded according to the difference between the EDSS score at the day of maximum worsening and the EDSS score before the onset of the attack, as: very mild (0 or 1 point), mild (1.5 or 2 points), moderate (2.5 or 3 points), and severe (3.5 points or more). Duration was calculated as the time between the date of onset of the first symptom and the date of maximum improvement of the last symptom and classed as short (≤30 days), intermediate (31-60 days) and long (>60 days). We evaluated two groups of predictive factors related to either the single attack or the single patient, including disease course characteristics. Attack-related factors were: age at the onset of the attack, season of the attack, type of FS involved, speed of onset, number of affected FS, infections in the preceding month, and therapy of the attack. Patient-related factors were: gender, age at first attack, season of birth and of first attack, oligoclonal bands at CSF examination, Link Index, CSF IgG value, total number and number of gadolinium-enhancing lesions at first brain MRI. Recovery after the first attack was considered incomplete when the difference between the score at the date of maximum improvement and the score before the onset of the attack was ≥1 for at least one FS [1]. The definition of the other predictive variables are reported in our previous paper [1].

Data were analyzed with SAS [15]. Statistical comparisons were performed with Student’s t test for independent samples, chi-square and chi-square for trend where appropriate. Spearman’s coefficient was used for correlations. In view of the large number of comparisons, the Bonferroni correction for statistical significance was used. Odds Ratios (OR) with 95% confidence intervals (CI) were calculated. For multivariate analyses, possible predictors were dichotomized as follows: monosymptomatic vs. bi-polysymptomatic; severe vs. very mild+mild+moderate; long vs. intermediate+short, attack therapy vs. no therapy, season spring+summer vs. autumn+winter, age at attack onset < 30 vs. >=30. A multiple logistic regression was performed using SAS/LOGISTIC [15], with duration and severity as the dependent variable, entering the predictor variables significant after univariate analysis and attack therapy (forward stepwise selection model with significance level for retention at 0.05).

In another statistical approach, we calculated the cumulative time-dependent probability of a second moderate/severe or long attack by the Kaplan-Meier method with 95% confidence intervals (CI). Date of entry was the date of onset of the first attack and the date of end-point was the date of onset of a severe/moderate or long attack, or July 31, 2005, whichever came first. Patients having a second non-moderate/severe or non-long attack were censored at the time of their second attack. Categories of severity and duration were used as possible prognostic predictors of a attack. The log-rank test was used to test the significance of differences between groups. Multivariable Cox proportional hazard modeling was used to control for potential predictors of the probability of relapsing (SAS/PHREG) [15]. Predictors were entered into the model simultaneously.

65 attacks (26%) were very mild, 89 (36%) mild, 59 (24%) moderate, and 35 (14%) severe. Table 1 shows the degree of severity according to the attack-related predictive factors. Pyramidal and brainstem FS, number of FS involved and total attack duration were related to severity, the last two with a clear trend. As expected, iv. therapy was most frequent in more severe attacks. The highest risk of a severe attack was for pyramidal (OR 3.5, 95% CI 1.7-7.5) and cerebellar FS (3.5, 1.5-8.2), followed by sphincteric (3.1, 1.2-8.1) and brainstem (2.3, 1.1-5.1). Sensory (1.9, 0.9-4.1) and visual (1.4, 0.7-3.1) FS did not significantly increase the risk of a severe attack. Attacks with visual symptoms were more frequently monosymptomatic (45/69, 65%) than those with sensory (50/129, 39%), brainstem (17/51, 33%), motor (31/98, 32%), cerebellar (5/32, 16%), and sphincteric symptoms (1/23, 4%). To avoid bias determined by the multiple inclusion of attacks from single patient, we correlated patient-related predictive factors to attack severity separately for the first and second attack: none of them was correlated with the degree of severity (data not shown). In the multivariate model, the number of symptoms (mono vs. bi-polysymptomatic) was the only variable that explained severity (OR for severe attacks: 6.6; 95% CI 2.9-15.3). Since the number of FS is highly related with severity due to the measurement procedure, we performed another analysis, without number of FS. The following variables were retained after stepwise selection: pyramidal (OR 3.6, 1.7-7.7), brainstem (2.6, 1.2-6.0), and long duration (2.5, 1.2-5.3).

83 attacks (34%) were classed as short, 89 (36%) as intermediate, and 76 (31%) as long. Table 2 shows the duration according to the attack-related predictive factors. Number of FS, type of FS involved (sphincteric and sensory) and severity of the attack were related to duration. The risk of a long duration was significant only for sphincteric (5.0; 2.0-12.5), and sensory (2.1, 1.2-3.6) FS involvement. None of patient-related predictive factors was correlated with the duration of both the first and second attack (data not shown). In the multivariate model, sphincteric (4.4, 1.7-11.0) and sensory symptoms (1.8, 1.0-3.2) were the only variables explaining duration.

In a different approach, we tested whether the probability of a second severe or long attack was influenced by severity and duration of the first. Since the number of second severe attacks was small (N=7) we grouped them with moderate second attacks (N=13). The probability of a second moderate/severe or long attack was not influenced by severity or duration of the first (Figs. 1 and 2). The 3-year probability of a second moderate/severe attack was 21% (4-38) for a patients with a severe first attack and 22% (12-32) for those with a very mild-mild-moderate first attack (p=0.71). It was 23% (9-37) for a patients with a long first attack, and 21% (10-32) for those with a short-intermediate first attack (p=0.89). The 3-year probability of a second long attack was 23% (7-39) for a patients with a severe first attack, and 14% (6-23) for those with a very mild-mild-moderate first attack (p=0.22). It was 15% (5-24) for patients with a long first attack, and 20% (7-32) for those with a short-intermediate first attack (p=0.53). We performed a multivariate analysis, including all attack-related factors, gender and recovery after the first attack (Table 3); other factors not significant in the univariate analysis were not included. The only predictive (negative) variable for a moderate/severe second attack was incomplete recovery from the first attack.

Fig. (1). Cumulative time-dependent probability of a second moderate/severe attack according to severity and duration of the first. The 3-year probabilities were: 21% (4-38) for patients with a severe first attack, 17% (4-31) for moderate, 29% (12-46) for mild, and 17% (0-38) for very mild (p=0.70); 23% (9-37) for patients with a long first attack, 24% (7-41) for intermediate and 18% (5-32) for short (p=0.81).

Fig. (2). Cumulative time-dependent probability of a second long attack according to severity and duration of the first. The 3-year probabilities were: 23% (7-39) for patients with a severe first attack, 15% (1-29) for moderate, 13% (0-26) for mild, and 17% (0-38) for very mild (p=0.61); 20% (7-33) for patients with a long first attack, 16% (2-31) for intermediate and 13% (1-26) for short (p=0.75).