Mental health in Luxembourg: the numbers — and what brain research adds
By François Altwies, founder, Neurofeedback LuxembourgBio · LinkedIn · Reviewed
One in three people working in Luxembourg is at risk of burnout. One pupil in three struggles to fall asleep more than once a week. Behind these numbers are colleagues, children and nights that do not restore.
This page brings together the most reliable figures on mental health in Luxembourg — at work, at school and at night — and sets them beside what brain research actually shows: what can be measured, what training can change, and where the evidence is still thin.
Every number links to its original source, with its date and the people it describes. We review the page every quarter. Read it as one story, from how we feel to what the brain shows — or jump straight to the chapter you need.
Chapter 1
Running on empty: work in Luxembourg
36%
More than one employee in three in Luxembourg is at moderate or high risk of burnout, up from 16% in 2014.
Burnout risk more than doubled and sleep problems rose by more than half between 2014 and 2025. Source: CSL summary of results (PDF, FR)About this figure
Two survey years shown. CSL advises against using 2024 for burnout because the questionnaire order changed. High depression risk (15% in 2025) is not drawn: CSL states it more than doubled since 2014 but does not publish the 2014 value as a number.
Quality of Work Index 2025: 53.4 out of 100. Representative sample of 3,171 employees and civil servants aged 16–64 working at least 10 hours a week, drawn from social-security records and including cross-border workers. Interviews from 20 May to 29 August 2025.
Quality of Work Index 2025 (n = 3,171). The 2024 value is lower because the order of the questionnaire changed that year; CSL advises against comparing with 2024.
Quality of Work Index 2025 (n = 3,171). A screening risk score, not a diagnosis. CSL states that the 2025 level is more than double 2014 but does not publish the 2014 value as a number.
European Health Interview Survey, wave 2019: depression reported for the 12 months before the interview. Self-reported, not a diagnosis confirmed by a clinician.
HBSC Luxembourg 2022 (University of Luxembourg). 7,893 pupils in national-curriculum schools. Self-reported complaints (at least two, several times a week or daily), not a diagnosis.
HBSC Luxembourg 2022: WHO-5 well-being score of 9 or lower. 7,495 valid answers; 95% CI 19.9–21.8; girls 28.0%, boys 13.5%. A screening result, not a diagnosis.
HBSC Luxembourg 2022, Factsheet no. 1; survey of 8,737 pupils from 145 schools. 32.9% combines “about every day” (14.8%) and “more than once a week” (18.1%). No item-level denominator is published. Self-reported, not a measure of insomnia.
ORISCAV-LUX-2 health survey, January 2016 – January 2018: wrist accelerometer worn for six nights by 1,028 participants (± 1.1 h). A device estimate, not a sleep-lab measurement.
Study design: Cross-sectional population survey, wrist accelerometer
European Health Examination Survey in Luxembourg, 2013–2015 (1,170 respondents with complete data): self-reported usual sleep duration. The paper reports 5.13%.
Study design: Cross-sectional health examination survey, self-report
European Insomnia Guideline 2023 (European Sleep Research Society), Journal of Sleep Research, published 28 November 2023. A European estimate, not a Luxembourg measurement.
Study design: Clinical guideline (European Sleep Research Society)
Diagram in preparation. The data is in the table below.
Luxembourg pupils use devices at school slightly less than the OECD average, for learning and for leisure. Source: oecd.org — Luxembourg country noteAbout this figure
Self-reported time during school hours; 6,078 students in 49 schools in Luxembourg.
Show as table
Hours a day on devices at school, age 15 (PISA 2025)
OECD PISA 2025, Luxembourg country note, published September 2026; 6,078 students in 49 schools. Self-reported time: 1.6 h for learning (OECD 1.7), 0.8 h for leisure (OECD 1.1) during school hours.
OECD, Students, digital devices and success (May 2024), based on PISA 2022. In at least some maths lessons; 59% were distracted by other students’ devices. OECD average, self-reported.
OECD PISA 2022, OECD average, after accounting for socio-economic profile. An association between distraction and scores, not proof that the devices caused the gap.
WHO Regional Office for Europe / HBSC, published 25 September 2024; about 280,000 respondents aged 11, 13 and 15 in 44 countries and regions. Regional figures, not Luxembourg-only. Girls 13%, boys 9%; 12% at risk of problematic gaming.
WHO Mental disorders fact sheet, updated September 2026, data year 2023. Modelled Global Burden of Disease estimates, not a count of diagnoses. 470 million with an anxiety disorder, 322 million with depression.
Song P. et al., Journal of Global Health, February 2021. Meta-analysis, estimates adjusted to the 2020 global population. Worldwide figures; no Luxembourg estimate exists (see “What we looked for and did not find”).
Ip C.-T. et al., International Journal of Psychophysiology, 2018, 134:30–43. Intraclass correlations for theta, alpha and beta power across four recording intervals. Measures the stability of the recording, not the effect of any training.
Study design: Test–retest reliability study, four recording intervals
Thatcher R.W. et al., Journal of Neurotherapy, 2003. Lifespan database, ages two months to 82 years. The first author developed NeuroGuide, the software used in our offices.
Measurement is one thing; change is another. Here is what controlled research shows about training the brain — the strong results, the weak ones and the null ones.
Chapter 7
What training can — and cannot — show
0.04 · 0.21
In the largest recent meta-analysis of ADHD trials, ratings by assessors unaware of the group showed no significant benefit (0.04); trials using standard protocols showed a small one (0.21).
Our perspective — François Altwies sets out why he reads the Arnold 2021 trial as manufactured doubt, with his conflict of interest disclosed and the strongest counter-arguments answered.
Effect sizes from different meta-analyses, outcomes and rater types. SMD is not a percentage. Van Doren 2019 values shown without confidence intervals (not stated on this page).
Diagram in preparation. The data is in the table below.
Results tend to be positive without a sham or blinded comparison and weaker or null with one. About this figure
Built only from the figures on this page; not a complete review of each field. positive = significant benefit reported; null = no significant difference; missing = no published controlled trial found.
Meta-analysis of 15 studies (1,194 children; 6 of 15 randomised). Mostly parent ratings, not blinded. Impulsivity 0.69, hyperactivity 0.40. Two authors worked at a neurofeedback clinic (disclosed). The 2025 analysis of blinded ratings did not reproduce effects of this size.
Study design: Meta-analysis of 15 studies (6 randomised)
Arns et al. 2009, correlation across studies between average session count and effect size (one-tailed p = 0.04). A correlation between studies, not a dose–response trial.
Study design: Meta-regression within a meta-analysis of 15 studies
Meta-analysis of 10 randomised trials, 256 children in neurofeedback groups; follow-up 2–12 months. Parent ratings, not blinded. 0.64 and 0.80 are changes within the neurofeedback groups; 0.38 and 0.57 are the differences against non-active controls. One co-author (M. Arns) works at a neurofeedback clinic.
Study design: Meta-analysis of 10 randomised trials
European ADHD Guidelines Group, JAMA Psychiatry, February 2025 (online 11 December 2024). Meta-analysis of 38 randomised trials, 2,472 participants aged 5–40; blinded-rating analysis 20 trials, 1,214 participants, 95% CI −0.10 to 0.18. Pools all protocols together. SMD is not a percentage.
Study design: Meta-analysis of 38 randomised trials
Same meta-analysis, subgroup of 9 trials (681 participants), 95% CI 0.02 to 0.40: a small effect. Processing speed across 15 trials: SMD 0.35 (95% CI 0.01 to 0.69). SMD is not a percentage.
Study design: Subgroup analysis of a meta-analysis (9 randomised trials)
Neurofeedback Collaborative Group, 2021; 144 children aged 7–10 randomised, 142 analysed. Between-group d = 0.23 at 13 months; the neurofeedback group needed significantly less medication at 13 months (p = .012). The trial design is disputed (see the 2024 critique).
Study design: Double-blind randomised controlled trial
Schummer and Sguigna, NeuroRegulation, 2024. A published critique, not a new trial. Among the problems: the control condition delivered EMG biofeedback, and the way learning of the trained brain rhythm was analysed is questioned.
Study design: Peer-reviewed methodological critique
Rotkiewicz et al., Journal of Neurodevelopmental Disorders, 22 July 2026. 3 randomised and 4 non-randomised studies, 390 adults in total. Counts the evidence; it does not pool an effect.
Voigt, Mosier and Tendler, Frontiers in Psychiatry, 21 March 2024. 17 randomised trials (628 participants); CAPS analysis 7 trials, 95% CI 0.11–1.37, high heterogeneity. Active or sham controls only: 3 trials, 86 participants. Without one outlier trial the SMD drops to 0.15.
Study design: Meta-analysis of 17 randomised trials
Berman et al., Frontiers in Neuroscience, 3 December 2025 (PROSPERO CRD42020184659). Passive controls: 5 trials. Sham control: one EEG trial, 95% CI −1.09 to 0.23. Certainty of evidence very low to low.
van der Kolk et al., PLOS ONE, 16 December 2016; 52 adults with long-standing PTSD after at least six months of trauma-focused psychotherapy. Waitlist control, not sham. Percentages are of the 22 people assessed per group.
Oprea et al., Alpha Psychiatry, 25 June 2026 (searches to December 2025). Scoping review: 11 studies, 175 people with active training. The authors find the evidence too small and heterogeneous to establish clinical efficacy.
Cho, Tural and Iosifescu, September 2023. 8 studies, 102 participants. Single-arm studies (5): g 1.14 from baseline, no control. Sham-controlled (3 studies, 43 participants): 95% CI −0.28 to 1.28. The authors call the evidence weak. The senior author reports research support from LiteCure, a device maker.
Study design: Meta-analysis of 8 studies (3 sham-controlled)
Zhang et al., JAMA Network Open, 16 December 2024. 72 adults, 8 weeks; active 8.2 points vs sham 3.9 points; 95% CI 2.6–5.9. A single-centre trial in China that needs replication.
Study design: Randomised sham-controlled trial (patients, assessors and statisticians blinded)
Neurofeedback has a close relative that is growing faster: brain–computer interfaces, now used in rehabilitation after stroke and in multiple sclerosis.
Chapter 8
When the brain moves the body: brain–computer interfaces
+0.19 m/s
After recoveriX training, people with chronic stroke walked 0.19 m/s faster on average.
A brain–computer interface for rehabilitation works on the same principle as neurofeedback: a brain signal is measured, the person gets immediate feedback, and the brain learns. recoveriX links motor imagery to a virtual avatar and to electrical stimulation of the limb.
recoveriX Luxembourg is our exclusive g.tec partner programme. It uses motor-imagery brain–computer interface training with functional electrical stimulation and is a separate service from neurofeedback training.
“recoveriX PRO is intended for muscle re-education by powered muscle stimulation and for neurofeedback.”
Sebastián-Romagosa et al., Frontiers in Neuroscience, 18 October 2023. 25 people, 22 completed 25 sessions over about 3 months; 95% CI 0.13–0.25; threshold cited 0.14 m/s. No control group. Most authors are employed by g.tec, the manufacturer (disclosed).
Study design: Uncontrolled study (no control group)
PubMed articles per publication year, counted 30 September 2026 via the NCBI search API. Neurofeedback: "neurofeedback" in title or abstract. BCI: "brain-computer interface(s)", "brain computer interface" or "brain-machine interface" in title or abstract.
Our count of PubMed records matching neurofeedback[tiab] per publication year, NCBI E-utilities, 30 September 2026. All article types; records indexed later can raise past-year counts slightly. Total all years: 3,177 records.
Our PubMed count of “brain-computer interface(s)”, “brain computer interface” or “brain-machine interface” in title/abstract, per publication year, NCBI E-utilities, 30 September 2026. Growth since 2010: BCI 5.8-fold, neurofeedback 5.9-fold.
Commercial estimates from public summary pages, read 30 September 2026. None publishes its method in full; treat them as indications, not measurements. Growth rates quoted: 7–11% a year. The Insight Partners page gives two growth rates (7.55% and 8.54%).
Study design: Commercial market-research estimates (8 firms)
As of 30 September 2026 we found no published Luxembourg-wide estimate of how many children or adults have ADHD, and no newer national figure on adult sleep than the 2013–2018 surveys above. We do not fill these gaps with figures from other countries.
Searched: STATEC, Ministry of Health, Luxembourg Institute of Health, University of Luxembourg repository (ORBilu), CNS public documents.
How to read these figures
Population surveys describe how common an experience is. Training studies answer a different question. A change within one group is not the same as an advantage over a comparison group, and the type of comparison (waitlist, usual care, sham) changes what a result can show. A standardised effect size (SMD) expresses a difference relative to score variability; it does not mean that a given percentage of people improved.
We list positive and negative results side by side. Group-level findings do not predict what any one person will experience.
: Page restructured into ten chapters with a new title and introduction; figures unchanged from the verified v9 package, with the corrections from our verification notes.
About the author: François Altwies
Founder of Neurofeedback Luxembourg.
Studied philosophy, psychology, criminology, computer graphics and virtual reality before moving into human–machine interfaces and applied neurophysiology.
Trained by leading experts in the field, including Dr. Juri Kropotov, Dr. Robert Thatcher and Jay Gunkelman; also trains psychologists, psychiatrists and other professionals.
