WB6Dim-LTSA: Can Workplace Well-Being Scores Predict Collective Absenteeism?

BACKGROUND: In France, sickness benefit expenditure reached 10.2 billion euros in 2023, up 28% since 2019 (DREES/CNAM, Études \& Résultats n°1321, 2024). The cost distribution follows a Pareto pattern: 7% of sick leave episodes (those exceeding 6 months) generate 45% of total expenditure (Cour des Comptes, RALFSS 2024). For group disability insurers (prévoyance collective), the pressure is acute: charges for incapacity-disability-dependence rose +24.4% in 2024, long-duration indemnified days increased +31% since 2020, and the claims-to-premiums ratio deteriorated to 56.9% (France Assureurs 2025). These 7% of episodes are the primary cost driver, yet current identification relies on retrospective administrative data - the signal arrives after the damage is done. The literature identifies two pathways to long-duration sick leave: (1) escalation from repeated short absences (≥3 episodes/year, RR 1.5-2.5; Koopmans 2008, Hultin 2012, Roelen 2018, Sørensen et al. 2025), where administrative data detect the pattern but often too late for effective prevention; and (2) the 'cliff effect' - sudden onset without prior absence signal, driven by prolonged presenteeism masking progressive deterioration (Gustafsson \& Marklund 2011, Ahola 2009, López-Bueno \& Clausen 2021). Administrative data are entirely blind to the second pathway. Composite psychometric instruments can detect risk during the presenteeism window: they reach C-index 0.73-0.74 vs 0.65 for absence-only models (Airaksinen et al. 2018; Roelen 2013), and combining questionnaire + administrative data reaches C-index 0.79 (Nyberg et al. 2023).

The WB6Dim (Well-Being 6 Dimensions) is a 28-item digital psychometric instrument assessing 9 well-being dimensions, validated on 808 participants across 4 cohorts with 2 pre-registered protocols (NCT07301879, NCT07433764). All 19 convergent validity hypotheses were confirmed against 10 international gold-standard scales (PSS-10, WHO-5, CBI, ISI-7, RSES, SAS-SV, MSPSS, UCLA-3, CFQ-13, BPNS). The test-retest reliability (ICA) reached .904 (excellent). The Adaptive Load Index (ICA) classifies each respondent into 4 levels: low-to-moderate load, high load, very high load, and critical load. The Environmental Attentional Dysregulation (DAE) further characterizes the dominant source of strain (internal, digital, relational, or mixed). No published study has tested the predictive value of a composite well-being index, measured at the collective level, on sick leave ≥30 days - the threshold triggering group disability insurance benefits.

DESIGN: Prospective multicenter cohort study with 4 measurement waves over 6 months (June-November 2026). The unit of analysis is the company (collective level). No individual diagnosis or prognosis is delivered. Data sources include: (1) company-level HR data on absenteeism stratified by duration (aggregated, anonymized) for 2024, 2025, and 2026; (2) individual self-reported absence integrated into the WB6Dim at T0, T2, and T3. The study includes two analytical components: a retrospective analysis correlating T0 WB6Dim scores with 2024-2025 absenteeism, and a concurrent analysis testing whether T0-T2 trajectories predict T2-T3 absenteeism. Six pre-registered hypotheses are tested, including an exploratory hypothesis (H6) targeting the added value of ICA for companies with no elevated absence history but degraded well-being scores - directly testing whether the WB6Dim can identify the costly 7% before administrative data show any signal.

• Company-level incidence rate of sick leave episodes ≥ 90 days over 6 months, as measured from aggregated HR records [Time frame: 6 months post-enrollment]
• Number of employees with ≥ 3 distinct absence episodes within 6 months per company, as measured from aggregated HR records [Time frame: 6 months post-enrollment]
• Self-reported cumulative absence duration and episode count, as measured by WB6Dim questionnaire items [Time frame: Baseline, 3 months, and 6 months post-enrollment]
• Change in collective ICA distribution from baseline to 3 months as a predictor of sick leave ≥ 30 days between 3 and 6 months [Time frame: Baseline and 3 months (predictor); 3 to 6 months post-enrollment (outcome)]
• Change in predictive model AUC when adding DAE profile distribution to the ICA-based model for sick leave ≥ 30 days [Time frame: 6 months post-enrollment]
• Agreement (Cohen's kappa) between aggregated self-reported absence and company-level HR absence data, stratified by duration class [Time frame: 6 months post-enrollment]