**2.1. Patient selection, diagnosis, data collection, and main characteristics**

The Plovdiv project included hospital-based incident cases of patients with minor ischaemic stroke (MIS) that were followed for 12 months to determine the estimates of central motor conduction time (CMCT) and amplitudes of motor evoked potential (MEP) and their changes and correlations over time.

This is a *post-hoc* analysis and modelling study. The patient population has been described in more detail earlier [Atanassova, 1998; Atanasova & Vukov, 1998; Atanassova, Voukov & Tchalakova, 2002; Atanassova, Chalakova & Dimitrov, 2008a]. In particular, patients with cerebrovascular disease had been hospitalized in the Clinic of Cerebrovascular Diseases (Plovdiv Healthcare Region) and 56 consecutive patients with MIS were subjected to screening. All screened, eligible patients with MIS as an initial index event who provided a written, informed consent in accordance with the Declaration of Helsinki guidelines at discharge were immediately enrolled. During the lag interval from the index event until discharge (i.e., during the hospital stay), no vascular events were observed among the 56 screened eligible patients. Of these eligible patients, 54 patients (96.4%) provided written, informed consent and were included in the current follow-up study. The other 2 eligible patients did not provide informed consent at discharge and were not enrolled (**Figure 1**). Further, till month 12, a total of 14 patients were excluded or lost to follow-up and could not provide data on the outcome, therefore, 40 patients were subjected to statistical analyses and modelling in this study. The inclusion criteria were: patients with first MIS, age > 40 years and residence in Plovdiv for at least three months before identification and enrolment [Atanassova, Chalakova & Dimitrov, 2008a]. All evaluations were performed at Medical University Hospital of Plovdiv, Bulgaria. The University Ethical Committee approved the study protocol.

**Figure 1.** CONSORT flow-chart of study screening, enrolment and analysis.

Computational Intelligence in Electromyography Analysis – 38 A Perspective on Current Applications and Future Challenges

the amplitude of MEP [Palliyath, 2000].

changes and correlations over time.

**2. Methods** 

patient without any motor impairment) and occurred independently of a specific clinical picture or a radiologically confirmed lacunar lesion [Abbruzzese et al, 1991; Hufnagel et al, 1990]. Earlier studies have shown that during the acute phase of the minor ischaemic stroke (MIS), MEP amplitudes can be registered in all investigated patients [Hadjipetrova et al, 1993]. To note, the increases in the latency of the M-response and CMCT have prognostic significance for early assessment of the outcome of ischaemic stroke [Stulin et al, 2003]. Earlier studies by Ferbert and collaborators [1992] have indicated that the MEP amplitudes are a more sensitive marker for the sublclinical damage of the pyramidal tract than CMCT. A significant correlation has also been reported among the recovery of muscle strength and

The aim of this study was to perform a *post-hoc* analysis of one-year follow-up data from 40 patients with MIS and to: (i) investigate the central motor conduction time (CMCT) and the amplitude of the motor evoked potential (MEP) during the acute phase of MIS; (ii) provide evidence for a subclinical damage of the pyramidal tract; and (iii) model and predict the outcome measures at month 12 after MIS as based on earlier changes in the acute phase.

The Plovdiv project included hospital-based incident cases of patients with minor ischaemic stroke (MIS) that were followed for 12 months to determine the estimates of central motor conduction time (CMCT) and amplitudes of motor evoked potential (MEP) and their

This is a *post-hoc* analysis and modelling study. The patient population has been described in more detail earlier [Atanassova, 1998; Atanasova & Vukov, 1998; Atanassova, Voukov & Tchalakova, 2002; Atanassova, Chalakova & Dimitrov, 2008a]. In particular, patients with cerebrovascular disease had been hospitalized in the Clinic of Cerebrovascular Diseases (Plovdiv Healthcare Region) and 56 consecutive patients with MIS were subjected to screening. All screened, eligible patients with MIS as an initial index event who provided a written, informed consent in accordance with the Declaration of Helsinki guidelines at discharge were immediately enrolled. During the lag interval from the index event until discharge (i.e., during the hospital stay), no vascular events were observed among the 56 screened eligible patients. Of these eligible patients, 54 patients (96.4%) provided written, informed consent and were included in the current follow-up study. The other 2 eligible patients did not provide informed consent at discharge and were not enrolled (**Figure 1**). Further, till month 12, a total of 14 patients were excluded or lost to follow-up and could not provide data on the outcome, therefore, 40 patients were subjected to statistical analyses and modelling in this study. The inclusion criteria were: patients with first MIS, age > 40 years and residence in Plovdiv for at least three months before identification and enrolment [Atanassova, Chalakova & Dimitrov, 2008a]. All evaluations were performed at Medical University Hospital

**2.1. Patient selection, diagnosis, data collection, and main characteristics** 

of Plovdiv, Bulgaria. The University Ethical Committee approved the study protocol.

Data were collected by instructed physicians and clinical (physical and neurological) examinations were conducted by study neurologists [Atanassova, Chalakova & Dimitrov, 2008a]. The assessments covered: hypertension, diabetes, dyslipidaemia, peripheral vascular disease, cardiac conditions, cigarette smoking, educational background, etc.

Initial stroke severity was assessed using the modified Rankin Scale [Bamford et al, 1990]. MIS diagnostic evaluations included CT or MRI of the brain and ultrasound evaluation and/or trans-thoracic or trans-oesophageal echocardiogram, as appropriate. A panel of stroke neurologists assessed every CV events subtype using standard diagnostic criteria and all available information for each patient. For this study, *MIS* was defined as a minor stroke if the score on the modified Rankin Scale was 1 at the first evaluation, or if the score was 0 or 1 at one-month follow-up (i.e., no symptoms, or minor symptoms that did not interfere with normal lifestyle) [Atanassova, 1998; Atanasova & Vukov, 1998; Atanassova, Voukov & Tchalakova, 2002]. In particular, acute onset was observed in 11 patients (20.4%). Extracranial ultrasound findings were recorded in 11 patients (20.4%). The main characteristics of the initial study cohort were as follows: 37 men (68.5%) and 17 women (31.5%), with male predominance in CV events (n=8 men) but with similar ages of 61.1±12.6 years in patients with CVE versus 62.2±9.2 years in patients without CVE (p>0.05). The mean follow-up time was 11.1±2.4 months with mean time to CV events of 5.8±2.7 months. For the purpose of this study, all MIS patients with subsequent stroke (n=8) as defined above, were excluded from the analysis.

Computational Intelligence in Electromyography Analysis –

40 A Perspective on Current Applications and Future Challenges

#### *2.1.1. Assessment of outcomes*

The main outcome was defined as estimates of central motor conduction time (CMCT) and amplitudes of motor evoked potential (MEP) at month 12. We performed transcranial magnetic stimulation by MAGSTIM 200 after MIS (day 7 in month 1, month 3, and month 12) on the motor cerebral cortex bilaterally and on C7 with consecutive conduction of MEPs by surface electrodes from isometrically slightly contracted muscle *abductor policis brevis*. All measurements were taken as related to the symptomatic and asymptomatic hemispheres and differences between them were also analysed (Table 1). Normal values from 30 healthy subjects were also obtained for reference purposes.

Modelling of Transcranial Magnetic Stimulation in

One-Year Follow-Up Study of Patients with Minor Ischaemic Stroke 41

nonvascular death). All records were reviewed for all outcome events, including death, and have been maintained. All outcome events were reviewed by a neurology specialist. Nonfatal strokes were validated by a study neurologist, and all deaths were to be validated, as

Death as an eventual fatal outcome was defined as to be considered as due to stroke if there was clear documentation of a stroke from the death certificate or hospital records; deaths that would have occurred more than 30 days after the initial accident (i.e., secondary CV event) had to be considered related to the event on the grounds of a clinical judgment that relied on a clearly documented relationship to the stroke or its complications to the point of death in the medical records. Following the ascertainment procedures, all above mentioned death notifications, certificates, and autopsy protocols for all cases of death had to be collected and reviewed individually, especially for patients who died outside the hospital. In cases in which it was difficult to determine whether death was due to stroke, consensus was reached after discussion using the best available

The sample size of the initial follow-up cohort of 56 patients to be screened was calculated on the basis of the expected number of CV events, as described previously in more detail [Atanassova, Chalakova & Dimitrov, 2008a]. Having assumed a theoretical distribution from 0 to 50% for the non-events and based on the 3-year cumulative incidence of 24.5% for cerebrovascular events in MIS patients [Atanassova, Chalakova & Dimitrov, 2008b], 8.16% of CV events were to be expected in 12 months. Thus, it had been estimated that to give the study >95% power to detect such minimum event rate as statistically significant at p<0.05, 51 patients had to be included and analysed. A preliminary estimate of the prevalence of MIS patients that would satisfy the inclusion/exclusion criteria from all those referred to the Clinic of Cerebrovascular Diseases yearly had indicated that 56 patients with MIS had to be identified throughout a screening period of about 12 months (estimated maximum 10% drop-out). Given the pilot nature of the probabilistic modelling of the estimates and derivation of predictions at month 12 for both studied parameters (CMCT and MEP), no

The main endpoint for both the central motor conduction time and amplitudes of motor evoked potentials was considered as an estimated mean (± standard deviation, S.D.) at month 12 (**Table 1**). Two other interim measures of the outcomes were also taken (at month 1 and month 3). A test for normality of distributions (Shapiro-Wilk test) was applied. The differences were analysed by two-tailed paired parametric (t-test, etc.) or non-parametric (Wilcoxon signed-rank) tests at p<0.05, as appropriate, as well as repeated-measures

well. Deaths were to be classified using death certificates and medical records.

**2.2. Sample size, data elaboration and statistical analyses** 

further sample size calculations were performed.

*2.2.2. Data elaboration and statistical analyses* 

information.

*2.2.1. Sample size estimation* 


Notes: \*Number of cases (measurements) in the MIS patients with TMS values; Data are mean ± standard deviation; °Difference at p<0.05 is considered statistically significant. Abbreviations: MIS, minor ischaemic stroke.

**Table 1.** Main outcomes of TMS in 40 MIS patients, followed prospectively for 12 months, as measured according to the existing symptomatics (symptomatic or asymptomatic hemisphere)

As secondary outcomes, the changes from month 1 onwards, as well as the correlations between the estimates, were also analysed. The role of the symptomatics (i.e., measures for symptomatic or asymptomatic hemisphere) as a predictor of the main outcome estimates at month 12, was also investigated. The presence/absence of non-fatal or fatal CV event after MIS was considered for the diagnosis of the MIS patients with a subsequent stroke, who were to be excluded from the analyses. Thus, two2 sub-categories for each secondary outcome were established as events classification: (i) non-fatal CVE; (ii) fatal CVE. Strict evaluations were conducted in 4 visits (at baseline, at month 1, month 3 and month 12), with telephone interview every other month, till month 12. Every evaluation was carried-out by contact with the patient, family member, or caregiver. Information was collected by somatic examination, inter-current symptoms, illness or hospitalization. The in-person visits were conducted at our clinic and included measuring vital signs, physical and neurological examination. A registry reporting system was used to identify study participants who experienced nonfatal or fatal vascular events, related hospitalization or death (vascular or nonvascular death). All records were reviewed for all outcome events, including death, and have been maintained. All outcome events were reviewed by a neurology specialist. Nonfatal strokes were validated by a study neurologist, and all deaths were to be validated, as well. Deaths were to be classified using death certificates and medical records.

Death as an eventual fatal outcome was defined as to be considered as due to stroke if there was clear documentation of a stroke from the death certificate or hospital records; deaths that would have occurred more than 30 days after the initial accident (i.e., secondary CV event) had to be considered related to the event on the grounds of a clinical judgment that relied on a clearly documented relationship to the stroke or its complications to the point of death in the medical records. Following the ascertainment procedures, all above mentioned death notifications, certificates, and autopsy protocols for all cases of death had to be collected and reviewed individually, especially for patients who died outside the hospital. In cases in which it was difficult to determine whether death was due to stroke, consensus was reached after discussion using the best available information.

## **2.2. Sample size, data elaboration and statistical analyses**

#### *2.2.1. Sample size estimation*

Computational Intelligence in Electromyography Analysis – 40 A Perspective on Current Applications and Future Challenges

subjects were also obtained for reference purposes.

(CMCT) *[ms]*

potential (MEP) *[mV]*

The main outcome was defined as estimates of central motor conduction time (CMCT) and amplitudes of motor evoked potential (MEP) at month 12. We performed transcranial magnetic stimulation by MAGSTIM 200 after MIS (day 7 in month 1, month 3, and month 12) on the motor cerebral cortex bilaterally and on C7 with consecutive conduction of MEPs by surface electrodes from isometrically slightly contracted muscle *abductor policis brevis*. All measurements were taken as related to the symptomatic and asymptomatic hemispheres and differences between them were also analysed (Table 1). Normal values from 30 healthy

> Symptomatic hemisphere (n=40 cases\*)

at Month 1 (day 7) 9.147 ± 1.862 7.550 ± 1.465 <0.05 at Month 3 8.038 ± 1.392 7.135 ± 1.052 <0.05 at Month 12 10.720 ± 1.831 8.550 ± 1.497 <0.05

at Month 1 (day 7) 6.083 ± 1.882 8.963 ± 1.925 <0.05 at Month 3 7.293 ± 1.876 10.350 ± 2.160 <0.05 at Month 12 7.290 ± 1.757 9.880 ± 1.986 <0.05 Notes: \*Number of cases (measurements) in the MIS patients with TMS values; Data are mean ± standard deviation;

**Table 1.** Main outcomes of TMS in 40 MIS patients, followed prospectively for 12 months, as measured

As secondary outcomes, the changes from month 1 onwards, as well as the correlations between the estimates, were also analysed. The role of the symptomatics (i.e., measures for symptomatic or asymptomatic hemisphere) as a predictor of the main outcome estimates at month 12, was also investigated. The presence/absence of non-fatal or fatal CV event after MIS was considered for the diagnosis of the MIS patients with a subsequent stroke, who were to be excluded from the analyses. Thus, two2 sub-categories for each secondary outcome were established as events classification: (i) non-fatal CVE; (ii) fatal CVE. Strict evaluations were conducted in 4 visits (at baseline, at month 1, month 3 and month 12), with telephone interview every other month, till month 12. Every evaluation was carried-out by contact with the patient, family member, or caregiver. Information was collected by somatic examination, inter-current symptoms, illness or hospitalization. The in-person visits were conducted at our clinic and included measuring vital signs, physical and neurological examination. A registry reporting system was used to identify study participants who experienced nonfatal or fatal vascular events, related hospitalization or death (vascular or

°Difference at p<0.05 is considered statistically significant. Abbreviations: MIS, minor ischaemic stroke.

according to the existing symptomatics (symptomatic or asymptomatic hemisphere)

Asymptomatic

hemisphere (n=40 cases\*) p-value

*2.1.1. Assessment of outcomes* 

Outcome parameters in MIS patients (n=40 patients)

Central motor conduction time

Amplitude of motor evoked

The sample size of the initial follow-up cohort of 56 patients to be screened was calculated on the basis of the expected number of CV events, as described previously in more detail [Atanassova, Chalakova & Dimitrov, 2008a]. Having assumed a theoretical distribution from 0 to 50% for the non-events and based on the 3-year cumulative incidence of 24.5% for cerebrovascular events in MIS patients [Atanassova, Chalakova & Dimitrov, 2008b], 8.16% of CV events were to be expected in 12 months. Thus, it had been estimated that to give the study >95% power to detect such minimum event rate as statistically significant at p<0.05, 51 patients had to be included and analysed. A preliminary estimate of the prevalence of MIS patients that would satisfy the inclusion/exclusion criteria from all those referred to the Clinic of Cerebrovascular Diseases yearly had indicated that 56 patients with MIS had to be identified throughout a screening period of about 12 months (estimated maximum 10% drop-out). Given the pilot nature of the probabilistic modelling of the estimates and derivation of predictions at month 12 for both studied parameters (CMCT and MEP), no further sample size calculations were performed.

#### *2.2.2. Data elaboration and statistical analyses*

The main endpoint for both the central motor conduction time and amplitudes of motor evoked potentials was considered as an estimated mean (± standard deviation, S.D.) at month 12 (**Table 1**). Two other interim measures of the outcomes were also taken (at month 1 and month 3). A test for normality of distributions (Shapiro-Wilk test) was applied. The differences were analysed by two-tailed paired parametric (t-test, etc.) or non-parametric (Wilcoxon signed-rank) tests at p<0.05, as appropriate, as well as repeated-measures ANOVA (general linear models) in the view of the symptomatic and asymptomatic hemispheres (**Figure 2 & Figure 3**). As appropriate, parametric and non-parametric correlations between CMCT and MEP at various times were also performed.

Modelling of Transcranial Magnetic Stimulation in

One-Year Follow-Up Study of Patients with Minor Ischaemic Stroke 43

effect of symptomatics and time (grand mean 8.523 ms, 95%CI 8.240-8.805, p=0.01). Notably, there was a statistically significant difference (adjusted for the baseline values at month 1) between the estimated marginal means of CMCT in the symptomatic (10.717 ms, 95%CI 10.191-11.244) and asymptomatic (8.023 ms, 95%CI 8.023-9.077) hemispheres

There was a significant increase over time (p<0.001) in MEP amplitude, however, the multivariate, combined effect of symptomatics and time was not significant (grand mean 8.310 mV, 95%CI 7.922-8.697, p=0.309). Certainly, there was a statistically significant difference between the estimated marginal means of MEP amplitude in the symptomatic (6.888 mV, 95%CI 6.340-7.437) and asymptomatic (9.731 mV, 95%CI 9.182-10.279) hemispheres, but this was observed since month 1 and continued as such till month 12

**Figure 2.** General linear modelling (repeated ANOVA) of CMCT changes from month 1 till month 12

(Figure 2).

(Figure 3).

Parametric regression modelling was used to analyse the data and develop models to predict the outcomes at month 12 (**Table 2**). The significant relationships were later explored and confirmed by probabilistic artificial neural network (ANN) modelling, irrespectively of usual statistical constraints (**Figures 4 & Figure 5**). The stopping rule of learning was assumed when a state of maximum overall correctness of prediction with minimum average learning error was reached [Sarle, 1997]. The p-values less than 0.05 were considered statistically significant. The specialised software packages for statistical (SPSS ver.18) and probabilistic modelling (EasyNN ver.6.0i) were used.
