Abstract
Background: Insomnia disorder is the second most prevalent mental disorder, and it is a primary risk factor for depression. Inconsistent clinical and biomarker findings in patients with insomnia disorder suggest that heterogeneity exists and that subtypes of this disease remain unrecognised. Previous top-down proposed subtypes in nosologies have had insufficient validity. In this large-scale study, we aimed to reveal robust subtypes of insomnia disorder by use of data-driven analyses on a multidimensional set of biologically based traits. Methods: In this series of studies, we recruited participants from the Netherlands Sleep Registry, a database of volunteers aged 18 years or older, who we followed up online to survey traits, sleep, life events, and health history with 34 selected questionnaires of which participants completed at least one. We identified insomnia disorder subtypes by use of latent class analyses. We evaluated the value of our identified subtypes of insomnia disorder by use of a second, non-overlapping cohort who were recruited through a newsletter that was emailed to a new sample of Netherlands Sleep Registry participants, and by assessment of within-subject stability over several years of follow-up. We extensively tested the clinical validity of these subtypes for the development of sleep complaints, comorbidities (including depression), and response to benzodiazepines; in two subtypes of insomnia disorder, we also assessed the clinical relevance of these subtypes by use of an electroencephalogram biomarker and the effectiveness of cognitive behavioural therapy. To facilitate implementation, we subsequently constructed a concise subtype questionnaire and we validated this questionnaire in the second, non-overlapping cohort. Findings: 4322 Netherlands Sleep Registry participants completed at least one of the selected questionnaires, a demographic questionnaire, and an assessment of their Insomnia Severity Index (ISI) between March 2, 2010, and Oct 28, 2016. 2224 (51%) participants had probable insomnia disorder, defined as an ISI score of at least 10, and 2098 (49%) participants with a lower ISI score served as a control group. With a latent class analysis of the questionnaire responses of 2224 participants, we identified five novel insomnia disorder subtypes: highly distressed, moderately distressed but reward sensitive (ie, with intact responses to pleasurable emotions), moderately distressed and reward insensitive, slightly distressed with high reactivity (to their environment and life events), and slightly distressed with low reactivity. In a second, non-overlapping replication sample of 251 new participants who were assessed between June 12, 2017, and Nov 26, 2017, five subtypes were also identified to be optimal. In both the development sample and replication sample, each participant was classified as having only one subtype with high posterior probability (0·91–1·00). In 215 of the original sample of 2224 participants with insomnia who were reassessed 4·8 (SD 1·6) years later (between April 13, 2017, and June 21, 2017), the probability of maintaining their original subtype was 0·87, indicating a high stability of the classification. We found differences between the identified subtypes in developmental trajectories, response to treatment, the presence of an electroencephalogram biomarker, and the risk of depression that was up to five times different between groups, which indicated a clinical relevance of these subtypes. Interpretation: High-dimensional data-driven subtyping of people with insomnia has addressed an unmet need to reduce the heterogeneity of insomnia disorder. Subtyping facilitates identification of the underlying causes of insomnia, development of personalised treatments, and selection of patients with the highest risk of depression for inclusion in trials regarding prevention of depression. Funding: European Research Council and Netherlands Organization for Scientific Research.
| Original language | English |
|---|---|
| Pages (from-to) | 151-163 |
| Number of pages | 13 |
| Journal | The Lancet. Psychiatry |
| Volume | 6 |
| Issue number | 2 |
| Early online date | 7 Jan 2019 |
| DOIs | |
| Publication status | Published - Feb 2019 |
Funding
In our study, we identified five insomnia disorder subtypes that were differentiated by biologically based traits and life history. The subtypes that we identified were a highly distressed type that was characterised by distress across all domains; two moderately distressed types, one of which was reward sensitive and the other of which was reward insensitive; and two slightly distressed subtypes, one of which showed high reactivity to life events and the other of which showed low reactivity. Subtyping was stable over time, clinically relevant, and biologically meaningful, as indicated by enhanced salience and emotion signalling in the brain of participants classified as type 4. Subtyping is feasible with a concise set of questions that we have made available ( appendix p 37 ), including automated scoring. The subtypes were not primarily distinguished by existing clinical demarcations such as difficulty initiating sleep, difficulty maintaining sleep, or early morning awakening, nor by comorbid sleep disorders. Rather, subtypes emerged as specific, multivariate profiles of stable characteristics that were not directly related to sleep but were relevant to insomnia. 19 High or low scores on single variables were not unique to our insomnia subtypes, but the fingerprints of specific combinations of score levels on these characteristics are unique to the subtypes. Ancillary analyses ( appendix p 10 ) showed that none of the five subtypes resembled subtypes that can be found with bottom-up subtyping of people without insomnia. The stability of subtypes over several years that we found was notable. Most participants were identically classified 4·8 (SD 1·6) years after their initial subtyping (at a probability of 0·87), which compares favourably to previous clinical subtyping that showed poor reliability 15 and instability over even a brief period (33% over 4 months). 14 To our knowledge, our insomnia disorder subtypes are the first to fulfil the primary requirement of stability that is necessary to find differential trajectories for, biomarkers of, and treatment responses in insomnia disorder. Clinically, our identified subtypes provide precision targets to improve cognitive, emotional, and behavioural interventions. For example, because a meditation intervention lowers pre-sleep arousal, 25 this treatment could particularly be recommended for people with insomnia disorder of subtypes 1, 2, and 3, which are characterised by high pre-sleep arousal. Interventions that aim to improve positive affect and happiness 26 could be evaluated for people with insomnia subtypes 3 and 5, who showed a disproportional reduced positive affect or experience of pleasure. Finally, sleep problems related to childhood adversity, which was most prevalent among participants with subtypes 1 and 4, could require trauma therapy rather than CBT for insomnia only. 27 The clinical relevance of subtyping reaches beyond insomnia. Possibly related to a strong genetic overlap, 5 insomnia is a primary risk factor for depression. 8 The Global Consortium for Depression Prevention stated that the best chance to combat the global burden of depression is to identify people who run the highest risk and to provide them with preventive interventions. 28 Supported by the strong differences in current comorbid and lifetime depression, our subtyping approach could enable us to identify people with insomnia disorder who are most at risk for developing depression, and to prioritise their inclusion in preventive trials. Participants with subtype 1 insomnia scored highly on several symptoms of depression and showed the highest risk of lifetime depression. However, near half of participants classified in subtype 1 had never experienced depression. This finding is of considerable clinical interest for at least two reasons. First, people with subtype 1 insomnia disorder might have subclinical depression, and people with this subtype are most suitable to select for intervention programmes that aim to prevent depression. Use of our ITQ could facilitate such selection. Second, there could be an unknown factor that makes the unaffected half of our participants with subtype 1 resilient to depression despite a high risk. We illustrated how differentiation between subtypes 2 and 4 could propel the identification of biomarkers that would otherwise remain hidden by heterogeneity. ERPs deviated from the values in controls in participants with subtype 4 insomnia but not subtype 2 insomnia. The high amplitude late positive potential of the ERP in subtype 4 could relate to polymorphisms in the β1-receptor gene and response to beta-blockers, thus providing an example of a drug-targetable biomarker. 23 More specifically C/C homozygotes for the G1165C polymorphism in the β1-adrenergic receptor showed a larger late positive potential amplitude than G/C heterozygotes and G/G homozygotes. Moreover, our ERP finding supports consistency of labelling subtype 4 as reactive across psychometric traits and neurophysiology, thus meeting an important goal of the Research Domain Criteria. 29 Finally, we constructed and validated our ITQ, including automated analysis, to facilitate subtyping in future studies on insomnia. This subtyping can be done online and will accelerate insight into underlying causes and biomarkers of insomnia disorder and the development of better, more personalised treatments. Some limitations should be mentioned. First, although five subtypes were found to be optimal in both the original sample and the second, non-overlapping validation cohort, we cannot exclude the existence of other subtypes—for example, among people who do not volunteer for online assessment—because the NSR did not sample randomly from the general population. We deliberately did not exclusively sample from sleep clinics because, unfortunately, insomnia often goes unnoticed in general practice. 16 Sleep centre-based studies overrepresent complex insomnia in people who are more affected, but the NSR reaches a more diverse population of people with insomnia disorder. A possible disadvantage of case-control comparisons from the NSR could be that the control group might have been biased to include more people with a special interest in sleep or helping science. Therefore, replication of our study in a strict population-based sample will be useful. Second, we defined probable insomnia disorder by an ISI cutoff score of 10. Although this cutoff has been validated several times 5,21 and the ISI has been validated for web-based assessment, 30 it could be asked whether the cutoff can indicate insomnia of sufficient severity to warrant independent clinical attention and thus a separate DSM-5 diagnosis. A randomised trial 31 in patients with major depressive disorder used the same ISI cutoff of 10 to define comorbid insomnia; this study found that clinical attention to insomnia of this severity was highly valuable because treatment reduced insomnia and depression. This finding adds to the clinical validity of the ISI cutoff. The traditional approach to treat only the other morbidity, with the expectation that insomnia will resolve, is not regarded as the most appropriate approach; 17 treatment of both conditions simultaneously might improve the outcomes for both conditions. 32 In support of this hypothesis, a meta-analysis 33 that included 17 studies that used ISI scores supported treatment of insomnia in conjunction with comorbid psychiatric and medical conditions. Finally, traits that we have not assessed could discriminate yet other subtypes. Although we cannot exclude this possibility, it should be noted that we included an unprecedentedly large number of stable characteristics. Moreover, subtypes were defined by several characteristics, suggesting at least some robustness for unobserved characteristics. Within these limitations, the identification of subtypes enables important possibilities for pursuing subtype-specific risks, biomarkers, or treatment responses. In summary, we found that insomnia disorder can be classified into robust subtypes that can be discriminated by multivariate profiles of traits of affect and personality and life history. Subtyping was highly consistent after 4·8 years of follow up, results could be replicated in a second, non-overlapping cohort, and the subtypes could reliably be assessed with the ITQ. Insomnia subtyping paves the way for studies that aim to prevent depression, resolve inconsistencies in and reduce heterogeneity of insomnia, and reveal differential causes of and develop better tailored personalised treatment for insomnia disorder. For the Netherlands Sleep Registry see www.slaapregister.nl For the WHO International Classification of Diseases version 10 see http://www.who.int/classifications/icd/icdonlineversions/en/ For Insomnia Type Questionnaire scoring see https://tfblanken.shinyapps.io/itqapp/ Contributors TFB was the lead investigator, collected data, analysed and interpreted the data, and drafted the paper. JSB, JR, KD, DS, and RW collected data and ran the laboratory studies. DB supervised data analyses and interpretation and critically reviewed the manuscript. JKV supervised the latent class analyses. CP, JR, and YW analysed the data that were obtained in the laboratory studies. EJWVS was chief investigator, devised the study, supervised data collection, analysis, and interpretation, and critically reviewed the manuscript. All authors commented upon and approved the final manuscript. Declaration of interests All authors declare no competing interests. Acknowledgments The Netherlands Organization for Scientific Research (The Hague; VICI innovation grant no. 453–07–001 ) and the European Research Council ( no. ERC-ADG-2014–671084 INSOMNIA ) supported this work.
| Funders | Funder number |
|---|---|
| Netherlands Organization for Scientific Research | |
| VICI | 453–07–001 |
| Horizon 2020 Framework Programme | 671084, 737634 |
| European Research Council | no. ERC-ADG-2014–671084 INSOMNIA |
| Nederlandse Organisatie voor Wetenschappelijk Onderzoek | grant no. 453–07–001 |
