PMC:2953451 / 47653-54972 JSONTXT

{"project":"2_test","denotations":[{"id":"20948585-7436384-38616015","span":{"begin":689,"end":693},"obj":"7436384"},{"id":"20948585-1519004-38616016","span":{"begin":713,"end":717},"obj":"1519004"},{"id":"20948585-8235243-38616017","span":{"begin":742,"end":746},"obj":"8235243"},{"id":"20948585-19100767-38616018","span":{"begin":1238,"end":1242},"obj":"19100767"},{"id":"20948585-16459136-38616020","span":{"begin":2347,"end":2351},"obj":"16459136"},{"id":"20948585-1519004-38616021","span":{"begin":2439,"end":2443},"obj":"1519004"},{"id":"20948585-9286634-38616022","span":{"begin":2464,"end":2468},"obj":"9286634"},{"id":"20948585-15453549-38616023","span":{"begin":3537,"end":3541},"obj":"15453549"},{"id":"20948585-15511702-38616024","span":{"begin":3569,"end":3573},"obj":"15511702"},{"id":"20948585-15511708-38616025","span":{"begin":3591,"end":3595},"obj":"15511708"},{"id":"20948585-19109055-38616027","span":{"begin":3620,"end":3624},"obj":"19109055"},{"id":"20948585-18246987-38616028","span":{"begin":3642,"end":3646},"obj":"18246987"},{"id":"20948585-8858493-38616031","span":{"begin":5884,"end":5888},"obj":"8858493"},{"id":"20948585-16809057-38616032","span":{"begin":5890,"end":5894},"obj":"16809057"},{"id":"20948585-17580597-38616033","span":{"begin":5915,"end":5919},"obj":"17580597"}],"text":"Discussion\nIn the present study, we applied for the first time the unfolding procedure of RMT to analyze PLM data. Our first objective was to test whether unfolding would enhance and reveal intrinsic periodicity that is not obvious from the measured IMI distribution. Indeed, we found that applying this procedure in single patients improved detection of periodicity, which was hidden within a broad distribution of measured IMI (Figures 3A,C,D). Thus, this result underlines the usefulness of unfolding for the detection of periodicity in neurophysiological time series. The broad range of measured IMI may be due to their dependence on the time of night and sleep stage (Coleman et al., 1980; Culpepper et al., 1992; Pollmächer and Schulz, 1993). In addition, the IMI probably may also depend on the time of intake of medication and its pharmacokinetics influencing motor activity in sleep.\nBecause unfolding may be necessary to enhance and reveal periodicity, we propose this method as an important pre-processing step before analyzing PLM. Note that this recommendation holds independently from the method used for recording the LM, be it EMG (as in our study) or other methods like the recently applied Emfit sensor (Rauhala et al., 2009). Pooling of measured IMI may also broaden a group's IMI distribution, whereas applying the unfolding procedure before pooling the patients’ data retains their characteristics (Figures 3G,H). Technically, it is important to note that our choice of the degree m of the fit polynomial according to Eq. 4 prevents overfitting the data. Allowing for variations Δm ∈ [−4, 4] we have shown in addition that the results are robust against variation of the particular choice of m (Figures 3B–D).\nHere we investigated PLM periodicity by focusing on the shape of IMI distributions through the unfolding procedure. In patients with PLM and RLS Ferri et al. (2006b) found a bimodal distribution of measured IMI with the lower peak considerably below 10 s. This differs from our finding of a single-peaked distribution of measured IMI in patient group 3 (Figure 4C). The reason for this difference might be that in contrast to our analysis these authors based their calculation of LM distribution on all IMI intervals \u003e0.5 s in contrast to the WASM definition of PLM during sleep applied to our data (Zucconi et al., 2006).\nPLM associated with OSA and RLS show distinct IMI characteristics (Culpepper et al., 1992; Briellmann et al., 1997). Therefore our second objective was to test whether periodicity could differentiate between the clinical groups containing patients with PLM and OSA (group 1), PLM and RLS (group 3) and PLM without RLS or OSA (group 2). PLM periodicity was assessed by fitting one-parameter distributions to the unfolded IMI distributions of individual patients. The pooled IMI distributions of the three groups appeared very similar (Figures 4D–F) and none of the patients’ fit parameters σln, βSI, and βsP was significantly different between the groups. These results suggest that the degree of periodicity as quantified by these parameters is not related to the comorbidity (OSA or RLS). However, patients within groups were heterogeneous in terms of medication and comorbidities, thus representing a typical patient cohort in a sleep clinic. This heterogeneity contrasts with other studies that investigated patients with RLS and/or PLM excluding patients with medication influencing motor activity during sleep or significant sleep disorder or major comorbidities (Allen et al., 2004; Garcia-Borreguero et al., 2004; Hornyak et al., 2004; Ferri et al., 2006a,b, 2009; Manconi et al., 2007).\nNon-parametric correlation between the fit parameters and demographic patient data showed a significantly more narrow IMI distribution, i.e., a more pronounced periodicity, for overweight patients. BMI of patient group 1 (PLM with OSA) was significantly higher compared to groups 2 (PLM with RLS) and 3 (PLM without RLS or OSA), see Figure 1.\nIn contrast to the degree of periodicity (as defined by the values of the fit parameters σln, βSI, and βsP), its nature (as defined by the best fit distribution) was different between patient groups. We confirmed the result by Ferri et al. (2006b) that IMI distributions of patients suffering from RLS (here unfolded patient-wise distributions instead of pooled measured ones) can be fitted best by the log-normal distribution. On the contrary, the IMI distributions of patients in the OSA group could best be described by the Scharf–Izrailev distribution. Keeping in mind that this distribution can be derived for Dyson gases with long-range interaction between the particles (Izrailev, 1988; Scharf and Izrailev, 1990) this might imply a certain role of long-range interactions across several LM in these patients.\nOur third objective was to test whether a data-driven cluster analysis would separate patients into different groups. Measured and unfolded IMI distributions showed considerable dissimilarity within groups, while patients from different groups occasionally showed similar distributions. Therefore a possible definition of data-driven group formation could be derived from the shape of the IMI distributions. Data-driven clustering of unfolded IMI distributions of PLM had not been described before and yielded five similarity clusters. We found that the PI (Ferri et al., 2006b) is the only quantity that shows significant differences between these clusters. This may be due to both approaches aiming at quantifying PLM periodicity.\nHowever, the clusters were heterogeneous with respect to clinical and most demographic data. It remains unclear, which pathophysiological mechanism is responsible for the different shapes of the IMI distributions. One might speculate that the presumed neuronal central pattern generator responsible for the occurrence of PLM (Parrino et al., 1996, 2006; Guggisberg et al., 2007), is differentially influenced by medication and comorbidities such as RLS or OSA.\nAnalysis of the goodness-of-fit of the one-parameter fit distributions for the automatically defined similarity clusters showed the log-normal distribution to yield the best fit for two of the clusters (10 patients altogether). Also the Scharf–Izrailev distribution fitted two of the clusters best (9 patients altogether). Furthermore the apparently similar unfolded IMI distributions of two clusters (Figures 6B,E) were best fitted by different distributions. This finding indicates that subtle differences of periodicity may not be detected visually.\nIt remains to be investigated, whether and how applying the unfolding procedure to larger and more homogeneous patient groups allows the association between PLM and clinical significance. Furthermore, an interesting question is whether application of more sophisticated RMT tools like, e.g., the number variance Σ2(l) or the related but more stable Dyson–Mehta statistic Δ3(l) (not used in the present publication) helps to clarify the open questions of the clinical relevance of PLM and the nature of interaction between subsequent LM in PLM series. Both, Σ2(l) and Δ3(l) can be used to measure long-range correlations between IMI and consequently could complement the Markovian analysis carried out by Ferri et al. (2006a,b) from a methodological point of view."}

{"project":"0_colil","denotations":[{"id":"20948585-7436384-754366","span":{"begin":689,"end":693},"obj":"7436384"},{"id":"20948585-1519004-754367","span":{"begin":713,"end":717},"obj":"1519004"},{"id":"20948585-8235243-754368","span":{"begin":742,"end":746},"obj":"8235243"},{"id":"20948585-19100767-754369","span":{"begin":1238,"end":1242},"obj":"19100767"},{"id":"20948585-16459136-754371","span":{"begin":2347,"end":2351},"obj":"16459136"},{"id":"20948585-1519004-754372","span":{"begin":2439,"end":2443},"obj":"1519004"},{"id":"20948585-9286634-754373","span":{"begin":2464,"end":2468},"obj":"9286634"},{"id":"20948585-15453549-754374","span":{"begin":3537,"end":3541},"obj":"15453549"},{"id":"20948585-15511702-754375","span":{"begin":3569,"end":3573},"obj":"15511702"},{"id":"20948585-15511708-754376","span":{"begin":3591,"end":3595},"obj":"15511708"},{"id":"20948585-19109055-754378","span":{"begin":3620,"end":3624},"obj":"19109055"},{"id":"20948585-18246987-754379","span":{"begin":3642,"end":3646},"obj":"18246987"},{"id":"20948585-8858493-754382","span":{"begin":5884,"end":5888},"obj":"8858493"},{"id":"20948585-16809057-754383","span":{"begin":5890,"end":5894},"obj":"16809057"},{"id":"20948585-17580597-754384","span":{"begin":5915,"end":5919},"obj":"17580597"}],"text":"Discussion\nIn the present study, we applied for the first time the unfolding procedure of RMT to analyze PLM data. Our first objective was to test whether unfolding would enhance and reveal intrinsic periodicity that is not obvious from the measured IMI distribution. Indeed, we found that applying this procedure in single patients improved detection of periodicity, which was hidden within a broad distribution of measured IMI (Figures 3A,C,D). Thus, this result underlines the usefulness of unfolding for the detection of periodicity in neurophysiological time series. The broad range of measured IMI may be due to their dependence on the time of night and sleep stage (Coleman et al., 1980; Culpepper et al., 1992; Pollmächer and Schulz, 1993). In addition, the IMI probably may also depend on the time of intake of medication and its pharmacokinetics influencing motor activity in sleep.\nBecause unfolding may be necessary to enhance and reveal periodicity, we propose this method as an important pre-processing step before analyzing PLM. Note that this recommendation holds independently from the method used for recording the LM, be it EMG (as in our study) or other methods like the recently applied Emfit sensor (Rauhala et al., 2009). Pooling of measured IMI may also broaden a group's IMI distribution, whereas applying the unfolding procedure before pooling the patients’ data retains their characteristics (Figures 3G,H). Technically, it is important to note that our choice of the degree m of the fit polynomial according to Eq. 4 prevents overfitting the data. Allowing for variations Δm ∈ [−4, 4] we have shown in addition that the results are robust against variation of the particular choice of m (Figures 3B–D).\nHere we investigated PLM periodicity by focusing on the shape of IMI distributions through the unfolding procedure. In patients with PLM and RLS Ferri et al. (2006b) found a bimodal distribution of measured IMI with the lower peak considerably below 10 s. This differs from our finding of a single-peaked distribution of measured IMI in patient group 3 (Figure 4C). The reason for this difference might be that in contrast to our analysis these authors based their calculation of LM distribution on all IMI intervals \u003e0.5 s in contrast to the WASM definition of PLM during sleep applied to our data (Zucconi et al., 2006).\nPLM associated with OSA and RLS show distinct IMI characteristics (Culpepper et al., 1992; Briellmann et al., 1997). Therefore our second objective was to test whether periodicity could differentiate between the clinical groups containing patients with PLM and OSA (group 1), PLM and RLS (group 3) and PLM without RLS or OSA (group 2). PLM periodicity was assessed by fitting one-parameter distributions to the unfolded IMI distributions of individual patients. The pooled IMI distributions of the three groups appeared very similar (Figures 4D–F) and none of the patients’ fit parameters σln, βSI, and βsP was significantly different between the groups. These results suggest that the degree of periodicity as quantified by these parameters is not related to the comorbidity (OSA or RLS). However, patients within groups were heterogeneous in terms of medication and comorbidities, thus representing a typical patient cohort in a sleep clinic. This heterogeneity contrasts with other studies that investigated patients with RLS and/or PLM excluding patients with medication influencing motor activity during sleep or significant sleep disorder or major comorbidities (Allen et al., 2004; Garcia-Borreguero et al., 2004; Hornyak et al., 2004; Ferri et al., 2006a,b, 2009; Manconi et al., 2007).\nNon-parametric correlation between the fit parameters and demographic patient data showed a significantly more narrow IMI distribution, i.e., a more pronounced periodicity, for overweight patients. BMI of patient group 1 (PLM with OSA) was significantly higher compared to groups 2 (PLM with RLS) and 3 (PLM without RLS or OSA), see Figure 1.\nIn contrast to the degree of periodicity (as defined by the values of the fit parameters σln, βSI, and βsP), its nature (as defined by the best fit distribution) was different between patient groups. We confirmed the result by Ferri et al. (2006b) that IMI distributions of patients suffering from RLS (here unfolded patient-wise distributions instead of pooled measured ones) can be fitted best by the log-normal distribution. On the contrary, the IMI distributions of patients in the OSA group could best be described by the Scharf–Izrailev distribution. Keeping in mind that this distribution can be derived for Dyson gases with long-range interaction between the particles (Izrailev, 1988; Scharf and Izrailev, 1990) this might imply a certain role of long-range interactions across several LM in these patients.\nOur third objective was to test whether a data-driven cluster analysis would separate patients into different groups. Measured and unfolded IMI distributions showed considerable dissimilarity within groups, while patients from different groups occasionally showed similar distributions. Therefore a possible definition of data-driven group formation could be derived from the shape of the IMI distributions. Data-driven clustering of unfolded IMI distributions of PLM had not been described before and yielded five similarity clusters. We found that the PI (Ferri et al., 2006b) is the only quantity that shows significant differences between these clusters. This may be due to both approaches aiming at quantifying PLM periodicity.\nHowever, the clusters were heterogeneous with respect to clinical and most demographic data. It remains unclear, which pathophysiological mechanism is responsible for the different shapes of the IMI distributions. One might speculate that the presumed neuronal central pattern generator responsible for the occurrence of PLM (Parrino et al., 1996, 2006; Guggisberg et al., 2007), is differentially influenced by medication and comorbidities such as RLS or OSA.\nAnalysis of the goodness-of-fit of the one-parameter fit distributions for the automatically defined similarity clusters showed the log-normal distribution to yield the best fit for two of the clusters (10 patients altogether). Also the Scharf–Izrailev distribution fitted two of the clusters best (9 patients altogether). Furthermore the apparently similar unfolded IMI distributions of two clusters (Figures 6B,E) were best fitted by different distributions. This finding indicates that subtle differences of periodicity may not be detected visually.\nIt remains to be investigated, whether and how applying the unfolding procedure to larger and more homogeneous patient groups allows the association between PLM and clinical significance. Furthermore, an interesting question is whether application of more sophisticated RMT tools like, e.g., the number variance Σ2(l) or the related but more stable Dyson–Mehta statistic Δ3(l) (not used in the present publication) helps to clarify the open questions of the clinical relevance of PLM and the nature of interaction between subsequent LM in PLM series. Both, Σ2(l) and Δ3(l) can be used to measure long-range correlations between IMI and consequently could complement the Markovian analysis carried out by Ferri et al. (2006a,b) from a methodological point of view."}