Forecasting migraine with machine learning based on mobile phone diary and wearable data.

INTRODUCTION: Triggers, premonitory symptoms and physiological changes occur in the preictal migraine phase and may be used in models for forecasting attacks. Machine learning is a promising option for such predictive analytics. The objective of this study was to explore the utility of machine learning to forecast migraine attacks based on preictal headache diary entries and simple physiological measurements. METHODS: In a prospective development and usability study 18 patients with migraine completed 388 headache diary entries and self-administered app-based biofeedback sessions wirelessly measuring heart rate, peripheral skin temperature and muscle tension. Several standard machine learning architectures were constructed to forecast headache the subsequent day. Models were scored with area under the receiver operating characteristics curve. RESULTS: Two-hundred-and-ninety-five days were included in the predictive modelling. The top performing model, based on random forest classification, achieved an area under the receiver operating characteristics curve of 0.62 in a hold-out partition of the dataset. DISCUSSION: In this study we demonstrate the utility of using mobile health apps and wearables combined with machine learning to forecast headache. We argue that high-dimensional modelling may greatly improve forecasting and discuss important considerations for future design of forecasting models using machine learning and mobile health data.

Insufficient sleep may alter cortical excitability near the migraine attack: A blinded TMS crossover study.

BACKGROUND: Migraine is a brain disorder with a multifaceted and unexplained association to sleep. Brain excitability likely changes periodically throughout the migraine cycle. In this study we examine the effect of insufficient sleep on neuronal excitability during the course of the migraine cycle. METHODS: We examined 54 migraine patients after two nights of eight-hour habitual sleep and two nights of four-hour restricted sleep in a randomised, blinded crossover study. We performed transcranial magnetic stimulation and measured cortical silent period, short- and long-interval intracortical inhibition, intracortical facilitation and short-latency afferent inhibition. We analysed how responses changed before and after attacks with linear mixed models. RESULTS: Short- interval intracortical inhibition was more reduced after sleep restriction compared to habitual sleep the shorter the time that had elapsed since the attack (p = 0.041), and specifically in the postictal phase (p = 0.013). Long-interval intracortical inhibition was more increased after sleep restriction with time closer before the attack (p = 0.006), and specifically in the preictal phase (p = 0.034). Short-latency afferent inhibition was more decreased after sleep restriction with time closer to the start of the attack (p = 0.026). CONCLUSION: Insufficient sleep in the period leading up to a migraine attack may cause dysfunction in cortical GABAergic inhibition. The results also suggest that migraine patients may have increased need for sufficient sleep during a migraine attack to maintain normal neurological function after the attack.

Validation of questionnaires for restless legs syndrome in the general population: the Trøndelag Health Study (HUNT).

Questionnaires for restless legs syndrome have rarely been validated against face-to-face interviews in the general population. We aimed to validate the modified Norwegian, seven-item Cambridge-Hopkins restless legs syndrome questionnaire and a single diagnostic question for restless legs syndrome. We also aimed to stratify validity at 65 years of age. Among a random sample of 1,201 participants from the fourth wave of the Trøndelag Health Study, 232 (19%) agreed to participate, out of whom 221 had complete data for analyses. Participants completed the questionnaires for restless legs syndrome immediately before attending a face-to-face interview using the latest diagnostic criteria. We calculated sensitivity, specificity, and Cohen's kappa statistic (κ) of questionnaire- versus interview-based diagnoses. We found acceptable validity of the seven-item modified Cambridge-Hopkins diagnostic questionnaire for restless legs syndrome (κ = 0.37, 95% confidence interval [CI] 0.23-0.51) and good validity of the single diagnostic question (κ = 0.47, 95% CI 0.35-0.58). We also found good validity through the combination of modified Cambridge-Hopkins diagnostic questionnaire for restless legs syndrome items 2 and 5, while item 1 or 2 alone showed only acceptable validity. The single diagnostic question was significantly more valid among those aged <65 years (κ = 0.60 versus κ = 0.26). Both single- and two-item questionnaire-based diagnoses overestimated interview-based restless legs syndrome prevalence. The seven-item modified Cambridge-Hopkins diagnostic questionnaire for restless legs syndrome will be useful for epidemiological studies although low sensitivity may cause underestimation of true restless legs syndrome prevalence in the general population, especially among elderly. Brief questionnaire-based diagnoses of up to three items seem best utilised as an initial screen. Future studies should identify brief and even more valid questionnaire-based diagnoses for restless legs syndrome in order to estimate prevalence accurately in large epidemiological studies.

A Biofeedback App for Migraine: Development and Usability Study.

BACKGROUND: Biofeedback is effective in treating migraines. It is believed to have a beneficial effect on autonomous nervous system activity and render individuals resilient to stressors that may trigger a migraine. However, widespread use of biofeedback is hampered by the need for a trained therapist and specialized equipment. Emerging digital health technology, including smartphones and wearables (mHealth), enables new ways of administering biofeedback. Currently, mHealth interventions for migraine appear feasible, but development processes and usability testing remain insufficient. OBJECTIVE: The objective of this study was to evaluate and improve the feasibility and usability of an mHealth biofeedback treatment app for adults with migraine. METHODS: In a prospective development and usability study, 18 adults with migraine completed a 4-week testing period of self-administered therapist-independent biofeedback treatment consisting of a smartphone app connected to wearable sensors (Cerebri, Nordic Brain Tech AS). The app included biofeedback training, instructions for self-delivery, and a headache diary. Two wearable sensors were used to measure surface electromyographic voltage at the trapezius muscle and peripheral skin temperature and heart rate at the right second fingertip. Participants were instructed to complete a daily headache diary entry and biofeedback session of 10 minutes duration. The testing period was preceded by a preusability expectation interview and succeeded by a postusability experience interview. In addition, an evaluation questionnaire was completed at weeks 2 and 4. Adherence was calculated as the proportion of 10-minute sessions completed within the first 28 days of treatment. Usability and feasibility were analyzed and summarized quantitatively and qualitatively. RESULTS: A total of 391 biofeedback sessions were completed with a median of 25 (IQR 17-28) per participant. The mean adherence rate was 0.76 (SD 0.26). The evaluation questionnaire revealed that functionality and design had the highest scores, whereas engagement and biofeedback were lower. Qualitative preexpectation analysis revealed that participants expected to become better familiar with physical signals and gain more understanding of their migraine attacks and noted that the app should be simple and understandable. Postusability analysis indicated that participants had an overall positive user experience with some suggestions for improvement regarding the design of the wearables and app content. The intervention was safe and tolerable. One case of prespecified adverse events was recorded in which a patient developed a skin rash from the sticky surface electromyography electrodes. CONCLUSIONS: The app underwent a rigorous development process that indicated an overall positive user experience, good usability, and high adherence rate. This study highlights the value of usability testing in the development of mHealth apps.

Validation of insomnia questionnaires in the general population: The Nord-Trøndelag Health Study (HUNT).

The primary aim was to validate questionnaire-based insomnia diagnoses from a modified Karolinska Sleep Questionnaire (KSQ) and the Insomnia Severity Index (ISI), by age category (< or >65 years), against a semi-structured face-to-face interview. Secondary aims were to split validity by diagnostic certainty of the interview and to compare prevalence estimates of questionnaire- and interview-based diagnoses. A total of 232 out of 1,200 invited (19.3%) from the fourth Nord-Trøndelag Health Study (HUNT4) completed questionnaires, including the KSQ and ISI, shortly before attending a face-to-face diagnostic interview for insomnia based on the fifth edition of the Diagnostic and Statistical Manual of Mental Disorders (DSM-5). Both a tentative (DSM-5 criteria A-E) and a definite (criteria A-H) interview diagnosis was evaluated. Cohen's kappa statistic quantified questionnaire validity. In all, 33% (95% confidence interval 27-39%) of participants had definite insomnia: 40% of women and 21% of men. The ISI (cut-off 12) and several KSQ-based diagnoses showed very good validity (κ ≤0.74) against the tentative, versus good validity (κ ≤0.61) against the definite interview diagnosis. Short questionnaires, requiring a daytime symptom at least three times a week, may underestimate insomnia prevalence. Validity was consistently higher for persons aged below versus above 65 years (definite insomnia: κ ≤0.64 vs. κ ≤0.56). Our results have implications for epidemiological population-based studies utilising insomnia questionnaires.

The HUNT4 study: the validity of questionnaire-based diagnoses.

BACKGROUND: Questionnaire-based headache diagnoses should be validated against diagnoses made by the gold standard, which is personal interview by a headache expert. The diagnostic algorithm with the best diagnostic accuracy should be used when later analysing the data. METHODS: The Nord-Trøndelag Health Study (HUNT4) was performed between 2017 and 2019. Among HUNT4 participants, a total of 232 (19.3%) out of 1201 randomly invited were interviewed by a headache expert to assess the sensitivity, specificity and kappa value of the questionnaire-based headache diagnoses. RESULTS: The median interval between answering the headache questions and the validation interview was 60 days (95% CI 56-62 days). The best agreements were found for self-reported lifetime migraine (sensitivity of 59%, specificity of 99%, and a kappa statistic of 0.65, 95% CI 0.55-0.75), self-reported active migraine (sensitivity of 50%, specificity of 97%, and a kappa statistic of 0.55, 95% 0.39-0.71), liberal criteria of migraine (sensitivity of 64%, specificity of 93%, and a kappa statistic of 0.58, 95% CI 0.43-0.73) and ICDH3-based migraine ≥1 days/month (sensitivity of 50%, specificity of 94%, and a kappa statistic of 0.49, 95% CI 0.30-0.68). For headache suffering ≥1 days/month a sensitivity of 90%, specificity 80%, and a kappa statistic of 0.55, 95% CI 0.41-0-69 were found. For tension-type headache (TTH) ≥ 1 days/month the agreement was 0.33 (95% CI 0.17-0.49). CONCLUSION: The HUNT4 questionnaire is a valid tool for identifying persons with lifetime migraine, self-reported active migraine and active migraine applying liberal modified criteria. The agreement for TTH was fair.

The epidemiology of headache disorders: a face-to-face interview of participants in HUNT4.

BACKGROUND: The primary aim of this cross-sectional population-based study was to evaluate the 1-year prevalence of common headache disorders by a face-to-face interview. METHODS: The fourth wave of Nord-Trøndelag Health Survey (HUNT4) started in September 2017. The study was undertaken as part of a project mainly focusing on sleep disorders, where a total of 232 (19.3%) out of 1200 invited HUNT4 participants underwent a face-to-face headache interview. RESULTS: The mean age of the 232 participants was 58.4 years (range 22-89). There were 71.6% (95% CI 65.7-77.4) who reported headache during the last year, and 18.5% (95% CI 13.5-23.6) had suffered from headache in the same period. The 1-year prevalence of tension-type headache (TTH) was 43.1% (95% CI 36.7-49.5), of idiopathic stabbing headache 34.1% (27.9-40.2), and of definite migraine 18.1% (95% CI 13.1-23.1). A total of 7.6% (95% CI 4.0-10.7%) had migraine with coexisting TTH. Lifetime prevalence of migraine was 32.8% (95% CI 26.7-38.8). Headache yesterday was reported by 12.1% (95% CI 7.9-16.3), and 5.6% (95% CI 2.6-8.6) had headache during the interview. CONCLUSION: In this population-based cross-sectional headache study performed by a face-to-face interview, the 1-year prevalence of TTH was 43.1% and of idiopathic stabbing headache 34.1%. A total of 18.1% had active migraine (18.1%), whereas the lifetime prevalence of migraine was 32.8%.

Acetyl-l-carnitine versus placebo for migraine prophylaxis: A randomized, triple-blind, crossover study.

BACKGROUND: Preventive medication is indicated for many migraine patients, but is used in relatively few. The aim of the present study was to evaluate the efficacy of acetyl-l-carnitine as a prophylactic drug in migraine patients. METHODS: A single-center, randomized, triple-blind, placebo-controlled, crossover study was carried out. Men and women, age 18-65 years, with episodic migraine but otherwise healthy, were recruited mostly through advertisements. After a four-week run-in-phase, 72 participants were randomized to receive either placebo or 3 g acetyl-l-carnitine for 12 weeks. After a four-week washout, treatment was switched. The primary outcome was days with moderate or severe headache per four weeks. Secondary outcomes were days with headache, hours with headache, proportion of responders (>50% reduction in migraine days from baseline) and adverse events. RESULTS: In the complete case analyses, no statistically significant differences were found between acetyl-l-carnitine and placebo in severe or moderate headache days per month (3.0 versus 3.1, p = 0.80), headache days per month (5.1 versus 5.2, p = 0.73) or for the other secondary outcome measures. CONCLUSION: In this triple-blind crossover study no differences were found in headache outcomes between acetyl-l-carnitine and placebo. Our results do not provide evidence of benefit for efficacy of acetyl-l-carnitine as prophylactic treatment for migraine. TRIAL REGISTRATION: EUDRACT (2012-001624-36), ClinicalTrials.gov (NCT01695317).

Substance use disorders and psychotic disorders in epilepsy: a population-based registry study.

BACKGROUND: Epilepsy affects around 70 million people worldwide. Psychiatric comorbidity may add to the burden of the disease. We studied substance use disorders and psychotic disorders among people with epilepsy from a population-based perspective. METHODS: Norwegian specialist health services (hospitals and outpatient clinics) report diagnoses for individual patients to the Norwegian Patient Register. We used information on subjects born in 1930-1994 who were registered with a diagnosis of epilepsy at least once during the five-year period of 2008-2012. We compared the proportion of people with epilepsy registered with substance use disorders (alcohol use disorders or non-alcohol drug use disorders) and psychotic disorders (schizophrenia spectrum disorders or bipolar disorder) with similar figures in the population without epilepsy. We applied chi-square tests and log-binomial regression for analysis. RESULTS: Overall, 0.90% of the Norwegian adult population was registered with epilepsy in somatic hospitals during 2008-2012. The total proportion registered with alcohol use disorder was 5.74% among people with epilepsy and 1.29% in the population without epilepsy (age- and sex-adjusted relative risk [RR]: 4.42, 95% confidence interval [CI]: 4.22-4.62). The corresponding figures were 4.32% and 1.22% (RR 3.86 [95% CI: 3.67-4.06] for drug use disorder, 1.72% and 0.60% (RR 2.94 [95% CI: 2.71-3.19]) for schizophrenia spectrum disorders, and 1.50% and 0.68% (RR 2.29 [95% CI: 2.10-2.49]) for bipolar disorder. CONCLUSION: People with epilepsy were more often registered with substance use disorders and psychotic disorders than people without epilepsy. Psychiatric comorbidity requires particular attention in both diagnostic work-up and management of epilepsy, and creates complex medical challenges that require close cooperation between neurologists and psychiatrists. These findings may have implications for the organization and further development of comprehensive epilepsy care.

Altered 13C glucose metabolism in the cortico-striato-thalamo-cortical loop in the MK-801 rat model of schizophrenia.

Using a modified MK-801 (dizocilpine) N-methyl-D-aspartic acid (NMDA) receptor hypofunction model for schizophrenia, we analyzed glycolysis, as well as glutamatergic, GABAergic, and monoaminergic neurotransmitter synthesis and degradation. Rats received an injection of MK-801 daily for 6 days and on day 6, they also received an injection of [1-(13)C]glucose. Extracts of frontal cortex (FCX), parietal and temporal cortex (PTCX), thalamus, striatum, nucleus accumbens (NAc), and hippocampus were analyzed using (13)C nuclear magnetic resonance spectroscopy, high-performance liquid chromatography, and gas chromatography-mass spectrometry. A pronounced reduction in glycolysis was found only in PTCX, in which (13)C labeling of glucose, lactate, and alanine was decreased. (13)C enrichment in lactate, however, was reduced in all areas investigated. The largest reductions in glutamate labeling were detected in FCX and PTCX, whereas in hippocampus, striatum, and Nac, (13)C labeling of glutamate was only slightly but significantly reduced. The thalamus was the only region with unaffected glutamate labeling. γ-Aminobutyric acid (GABA) labeling was reduced in all areas, but most significantly in FCX. Glutamine and aspartate labeling was unchanged. Mitochondrial metabolites were also affected. Fumarate labeling was reduced in FCX and thalamus, whereas malate labeling was reduced in FCX, PTCX, striatum, and NAc. Dopamine turnover was decreased in FCX and thalamus, whereas that of serotonin was unchanged in all regions. In conclusion, neurotransmitter metabolism in the cortico-striato-thalamo-cortical loop is severely impaired in the MK-801 (dizocilpine) NMDA receptor hypofunction animal model for schizophrenia.

Hypoglutamatergic activity in the STOP knockout mouse: a potential model for chronic untreated schizophrenia.

In mice, the deletion of the STOP protein leads to hyperdopaminergia and major behavioral disorders that are alleviated by neuroleptics, representing a potential model of schizophrenia. The reduction of the glutamatergic synaptic vesicle pool in the hippocampus could reflect a disturbance in glutamatergic neurotransmission in this model. Here we examined potential disturbances in energy metabolism and interactions between neurons and glia in 15-week-old STOP KO, wild-type, and heterozygous mice. Animals received [1-(13)C]glucose and [1,2-(13)C]acetate, the preferential substrates of neurons and astrocytes, respectively. Extracts from the whole forebrain and midbrain were analyzed by HPLC, (13)C and (1)H NMR spectroscopy. Amounts and labeling of most metabolites were unchanged. However, glutamine concentration and amount of [4,5-(13)C]glutamine derived from [1,2-(13)C]acetate significantly decreased by 17% and 18%, respectively, in STOP KO compared with wild-type mice. The amount of [4-(13)C]glutamate was decreased in STOP KO and heterozygous compared with wild-type mice. gamma-Aminobutyric acid labeling was not influenced by the genotype. Because STOP-deficient mice have a lower synaptic vesicle density, less glutamate is released to the synaptic cleft, leading to decreased stimulation of the postsynaptic glutamate receptors, reflecting increased glutamine metabolism only in the vicinity of the postsynapse of STOP KO mice.

How do glial-neuronal interactions fit into current neurotransmitter hypotheses of schizophrenia?

Evidence is accumulating that the exclusive dopamine hypothesis of schizophrenia has to be abandoned. Instead, a more integrative approach combines different neurotransmitter systems, in which glutamatergic, GABAergic and dopaminergic pathways interact. This paradigm shift coincides with the recognition that, while typical and modern atypical antipsychotic drugs have efficiently controlled the dramatic psychotic symptoms of schizophrenia, their impact on negative and cognitive symptoms is negligible. Indeed, cognitive decline is now believed to represent the core of schizophrenic morbidity and in this context, impairment of glutamate and more specifically NMDA function is of major importance. Given that astrocytes are important in controlling glutamate homeostasis, it is necessary to assign a significant role to glial-neuronal interactions in the pathophysiology of schizophrenia. Indeed, recent data from several animal and human studies corroborate this notion. This review outlines current neurotransmitter hypotheses and evidence for glial impairment in schizophrenia. Furthermore, findings from recent studies of (13)C nuclear magnetic resonance spectroscopy in experimental models of schizophrenia and NMDA hypofunction are presented and their implications for future research on glial-neuronal interactions discussed.

Repeated injection of MK801: an animal model of schizophrenia?

Glutamate-induced neurotoxicity plays an important role in neurological and psychiatric diseases. Thus, much attention has been given to the potential neuroprotective role of glutamate receptor antagonists, especially to those acting on the N-methyl-d-aspartate (NMDA) subtype. However, in addition to their neuroprotective potential, these compounds have also neurotoxic and psychotogenic properties. In the present study we used repeated injections of MK801 to examine if this non-competitive NMDA receptor antagonist could be used to produce schizophrenia-like alterations in behavior and brain metabolism in animals. Rats were given injections of MK801 (0.1 mg/kg) on six consecutive days, the last dose together with [1-(13)C]glucose and [1,2-(13)C]acetate, to probe neuronal and astrocytic metabolism, respectively. Analyses of extracts from parts of the frontal cortex plus cingulate and retrosplenial cortices and temporal lobes were performed using (13)C and (1)H magnetic resonance spectroscopy. Changes in glutamate and glutamine were restricted to the temporal lobe, in which amounts and labeling from [1-(13)C]glucose and [1,2-(13)C]acetate were increased compared to control. Locomotor activity was slightly higher in rats treated with MK801 compared to untreated animals. Metabolic changes did not resemble the alterations occurring in schizophrenia and those after repeated high dose (0.5 mg/kg) [Kondziella, D., Brenner, E., Eyjolfsson, E.M., Markinhuhta, K.R., Carlsson, M., Sonnewald, U., 2005. Glial-neuronal interactions are impaired in the schizophrenia model of repeated MK801 exposure. Neuropsychopharmacology, Epub ahead of print] but rather those caused by MK801 seen after a single high dose (0.5 mg/kg) [Brenner, E., Kondziella, D., Haberg, A., Sonnewald, U., 2005. Impaired glutamine metabolism in NMDA receptor hypofunction induced by MK801. J. Neurochem. 94, 1594-1603.].

Glial-neuronal interactions are impaired in the schizophrenia model of repeated MK801 exposure.

Schizophrenia-mimicking compounds such as phencyclidine (PCP) and MK801 are antagonists at the N-methyl-D-aspartate (NMDA) receptor and produce the whole spectrum of positive, negative, and cognitive symptoms. This is one of the most important pillars of the hypoglutamatergic hypothesis of schizophrenia. Since the synthesis of glutamate and GABA in neurons is closely connected to astrocyte metabolism, the study of astrocytic function is essential in this context. Dizocilpine-maleate (MK801) (0.5 mg/kg) was injected into rats every day for 6 days. The last dose was given together with [1-(13)C]glucose and [1,2-(13)C]acetate. Extracts from frontal, retrosplenial, and cingulate cortices (CRFC) and temporal lobes were examined by (13)C nuclear magnetic resonance spectroscopy, high pressure liquid chromatography, and light microscopy. In CRFC, significant increases in the levels of glutamate, glutathione, and taurine were seen, whereas amounts and turnover of noradrenaline, dopamine, and serotonin were unchanged. Glutamate and glutamine, derived from [1,2-(13)C]acetate and thus astrocytes, were significantly decreased in CRFC as compared to controls. Labeling from [1-(13)C]glucose and thus mostly neuronal metabolism was affected in the same brain region with decreased labeling of glutamate and GABA. The present model mimics the increased glutamate/glutamine activity found in drug-naive patients with first episode schizophrenia. Moreover, the decreased labeling indicates the transition to lower glutamatergic function seen in chronic schizophrenia patients. The disturbance in astrocytic function and the glutamine-glutamate-GABA cycle are of significant importance and might add to the malfunction of the cortico-striato-thalamo-cortical loop caused by NDMA receptor blockade.

Impaired glutamine metabolism in NMDA receptor hypofunction induced by MK801.

Paradoxically, glutamate receptor antagonists have neurotoxic and psychotogenic properties in addition to their neuroprotective potential during excessive glutamate release. In the present study the non-competitive N-methyl-D-aspartate (NMDA) receptor antagonist MK801 was used to examine glial-neuronal interactions in NMDA receptor hypofunction. Rats were given a subanesthetic dose of MK801 together with [1-13C]glucose and [1,2-13C]acetate, and brains were removed 20 min later. Analyses of extracts from cingulate, retrosplenial plus middle frontal cortices (CRFC) and temporal lobe were performed using HPLC and 13C and 1H nuclear magnetic resonance spectroscopy. Hypofunction of the NMDA receptor induced similar changes in both brain areas investigated; however, the changes were most pronounced in the temporal lobe. Generally, only labeling from [1-13C]glucose was affected by MK801. In CRFC and temporal lobe amounts of both labeled and unlabeled glutamine were increased, whereas those of aspartate were decreased. In the CRFC the decrease in labeling of aspartate was greater than the decrease in concentration, leading to decreased 13C enrichment. In temporal lobe, not in CRFC, increased concentrations of glutamate, GABA, succinate, glutathione and inositol were detected together with increased labeling of GABA and succinate from [1-13C]glucose. 13C Enrichment was decreased in glutamate and increased in succinate. The results point towards a disturbance in glutamate-glutamine cycling and thus interaction between neurons and glia, since labeling of glutamate and glutamine from glucose was affected differently.