Insomnia Severity Index (ISI)

Structural validity Moderatecontestedevidence form: mixed

indirect most factor-analytic evidence comes from clinical, student and non-UK community samples rather than UK working adults; the one large occupational sample (Korean shift workers) is non-UK

The ISI factor structure is genuinely contested: the systematic review and meta-analysis by Manzar and colleagues catalogued roughly 13 disparate reported models and concluded that, on pooled confirmatory evidence, a two-factor solution is the more robust representation of dimensionality than a three-factor one, while noting widespread methodological omissions across primary studies (Manzar 2021). Individual studies variously report one factor (for example a Spanish dementia-caregiver sample, Jimenez 2021; a large Black Women's Health Study cohort where a one-factor model fitted but not robustly, Yusufov 2021), two factors of severity and daytime impact (menopausal women, Otte 2019; pregnant women, Shinohara 2023; Saudi nurses, Albougami 2019; the six-item reworking in the lemborexant trials, Lenderking 2024), and three factors (a Taiwan/Hong Kong/Canada cross-cultural analysis, Chen 2015; US college students, Emert 2024). In a very large occupational sample of Korean shift workers exploratory analysis returned a single ISI factor (Park 2019). The dominant modern reading is a two-factor severity/impact structure, but the number of factors is demonstrably sample-dependent, which is itself the graded finding.

Convergent and discriminant validity Moderatewell-establishedevidence form: canonical

indirect key convergent estimates come from clinical, student and non-UK samples; discriminant separation from mood measures is incompletely established for UK working adults

Convergent validity is well supported. Total ISI score correlates significantly with sleep-diary indices, fatigue, quality of life, anxiety and depression in the population-based and clinical validation (Morin 2011). In US college students the ISI total showed weak-to-strong correlations with other sleep-disturbance indicators (r approximately 0.25 to 0.62) and weak-to-moderate correlations with psychosocial measures (r approximately 0.10 to 0.57), with analogous ISI and sleep-diary items correlating around r=0.40 to 0.45; the authors caution about overlap with general psychological distress (Emert 2024). Indian and other adaptations report strong correlation with the Pittsburgh Sleep Quality Index (for example r approximately 0.45) (Veqar 2017). Discriminant validity is comparatively less tested, and the overlap between ISI scores and depression/anxiety measures is a recurring caution rather than clean separation.

Criterion validity: reference standard Highwell-establishedevidence form: canonical

indirect for the target audience, reference-standard validation is robust but was earned in clinical, primary-care and non-UK community/student samples, not UK workplaces; cut-offs are population-specific

Criterion validity against a diagnostic reference standard is the ISI's strongest evidence. Against clinician interview or diagnostic criteria, a community-sample cut-off of 10 gave 86.1% sensitivity and 87.7% specificity, and in the clinical sample the ISI discriminated insomnia from good sleepers (Morin 2011). In primary care, against a semi-structured diagnostic interview, an optimal cut-off of 14 gave 82.4% sensitivity and 82.1% specificity, area under the curve 0.87, with moderate agreement (kappa=0.62) (Gagnon 2013). A meta-analysis of 19 studies (4693 participants) pooled ISI sensitivity at 88% (95% CI 0.79 to 0.93) and specificity at 85% (0.68 to 0.94) against reference standards, comparable to the Athens Insomnia Scale and Pittsburgh Sleep Quality Index (Chiu 2016). Optimal cut-offs vary substantially by population (for example >=8 in US college students, Emert 2024), so a single universal threshold does not transport cleanly.

Criterion validity: organisational Absent (a finding about the literature)untestedevidence form: canonical

indirect no organisational-outcome criterion evidence located in any population, UK or otherwise

No evidence was located in the latest review pass validating the ISI against organisational or work outcomes such as sickness absence, staff turnover, or objective job performance. Occupational studies retrieved used the ISI within workforce samples (for example over 12,000 Korean shift workers screened at occupational health examinations, Park 2019; Saudi nurses, Albougami 2019) but assessed its internal structure and reliability, not its association with work outcomes. The absence of criterion validation against organisational endpoints is the finding, and it matters directly for a workplace-wellbeing registry: the ISI is validated as an insomnia screener, not as a predictor of work-relevant outcomes.

Internal consistency Highwell-establishedevidence form: mixed

indirect high alpha is replicated across many populations but few are UK working adults; several estimates come from clinical or non-UK samples

Internal consistency is consistently high. Cronbach alpha was 0.90 and 0.91 in the population-based and clinical samples (Morin 2011), 0.92 in primary care (Gagnon 2013), and the meta-analysis reported a high pooled alpha across studies (Manzar 2021). Values in adaptation and subgroup studies span roughly 0.75 to 0.96, for example 0.75 to 0.78 for the two ISI factors among Saudi nurses (Albougami 2019), 0.78 in Spanish dementia caregivers (Jimenez 2021), omega approximately 0.89 in Black women (Yusufov 2021), and 0.96 in older Indonesian adults (KemalaSari 2026). Alpha at the upper end (>0.95) may signal item redundancy in a short scale.

Test-retest reliability Moderatewell-established

indirect retest coefficients are good to excellent but were earned in Danish outpatient, US student, Indian student and Indonesian older-adult samples, none of them UK working adults, and often at short intervals in small subsamples

Test-retest reliability is repeatedly reported as good to excellent (ICC roughly 0.84 to 0.96) across two-week and one-week intervals, but every retained coefficient comes from a non-UK adaptation or a student/clinical/older-adult sample rather than a UK working-adult cohort, and some rest on small retest subsamples. It is a genuine and reproducible property, downgraded for population indirectness and reliance on translated/derivative versions.

Measurement invariance Moderatecontestedevidence form: mixed

indirect invariance was tested across culture, sex, parity and short time intervals in non-UK clinical/community samples; occupation and UK working-adult invariance were not located

Invariance evidence is moderate and mostly favourable within its (indirect) samples. Strong (scalar-level) cross-cultural invariance held across Taiwanese, Hong Kong and Canadian samples for a three-factor model (Chen 2015). A two-factor model showed configural, metric, scalar and partial strict invariance across gender in Saudi nurses (Albougami 2019). Measurement and structural invariance held across parity and across two perinatal time points one week apart in pregnant Japanese women (Shinohara 2023). A two-factor structure was largely invariant across Black and White midlife women with hot flushes, with one item (difficulty falling asleep) behaving differently (Otte 2019). Because the underlying factor model itself is contested and sample-dependent, invariance conclusions are conditional on the model chosen, and occupational and UK working-adult invariance is untested.

Sub-grades (evidence differs by subgroup):

• {"subgroup": "across culture/language", "grade": "Moderate", "note": "scalar invariance reported across Taiwan, Hong Kong, Canada (Chen 2015)"}
• {"subgroup": "across sex", "grade": "Low", "note": "partial strict invariance in a single Saudi-nurse sample (Albougami 2019)"}
• {"subgroup": "across occupation", "grade": "Absent", "note": "no occupational-group invariance evidence located"}

Responsiveness and MIC Moderatewell-establishedevidence form: mixed

indirect responsiveness and MIC come from insomnia treatment trials (clinical/older/trial populations), not UK working adults, and the MIC value depends on whether the 7-item or 6-item scoring is used

Responsiveness is a comparative strength: the ISI was designed and validated as a treatment-outcome measure, detecting change across behavioural and pharmacological treatment against sleep diaries and polysomnography (Bastien 200100065-4)). For interpretability, Morin and colleagues estimated that a change of about -8.4 points (95% CI -7.1 to -9.4) on the 7-item ISI corresponded to moderate clinician-rated improvement (Morin 2011). A later clinical-trial re-analysis (two phase III lemborexant trials, N=1956) favoured a 6-item two-factor scoring and derived a meaningful within-individual change of a 5-point reduction on the 6-item version, noting that generalisability may be limited to similar trial populations (Lenderking 2024). The systematic review noted that responsiveness and interpretability remain less completely reported than reliability and criterion validity across insomnia instruments generally (Ali 2020). The minimal important change is therefore established but version- and population-dependent (approximately 7 to 8.4 points on the 7-item form; approximately 5 points on the 6-item form).