Predicting low back pain-related absenteeism in the Spanish general working population

Abstract

Objectives

To identify factors associated with an increased risk of low back pain (LBP)-related sick leave and to develop prognostic models for both the likelihood and duration of such leave.

Methods

A total of 7262 actively working adults were recruited consecutively during their annual check-ups. Seventy-seven variables were assessed, including sociodemographic, clinical, work-related, LBP-related and psychosocial factors. Outcomes included the occurrence of LBP-related sick leave and the number of days on leave over an 18-month follow-up period. Multivariable prognostic models were developed

Results

During the follow-up, 535 participants (7.4%, 95% CI 6.8 to 8.0) took LBP-related sick leave, of whom 162 were off work for ≥30 days. Predictors of taking sick leave were older age, not being self-employed, a usual LBP episode duration >14 days, higher job insecurity, greater self-expectations of taking sick leave and stronger perceived economic consequences of being on leave. Predictors of longer sick leave (≥ 30 days) were not being self-employed and experiencing LBP while in bed. The models demonstrated good calibration but poor discrimination (C-statistics: 0.607 and 0.604).

Conclusion

Predicting LBP-related sick leave and its duration among the general Spanish workforce remains challenging. Psychosocial and economic variables outweigh clinical or biological predictors. Self-employment is the only factor associated with a lower risk of both sick leave and being off work for ≥30 days.

Trial registration number

NCT00667316.

WHAT IS ALREADY KNOWN ON THIS TOPIC

• Determinants of low back pain (LBP)-related sick leave and its duration in the Southern European general working population are largely unknown.

WHAT THIS STUDY ADDS

• Psychosocial variables outweigh clinical or biological predictors. Self-employment is the only factor associated with a lower risk of both sick leave and being off work for ≥30 days, during an 18-month period.

HOW THIS STUDY MIGHT AFFECT RESEARCH, PRACTICE OR POLICY

• Current prognostic models do not allow early identification of individuals at increased risk for LBP-related sick leave or prolonged absence in the general workforce. Hence, preventive strategies should be directed at the entire working population. This emphasises the need to ensure their efficiency.

• Further research should explore potential additional prognostic factors and how contextual factors, such as labour regulations and cultural attitudes toward absenteeism, affect the generalisability of findings.

Introduction

Low back pain (LBP) is among the most common causes of disability and work absenteeism.¹ A wide range of biological, clinical, sociodemographic, physical, work-related and psychosocial factors can influence both the onset and duration of LBP-related absenteeism.²,⁶

Many studies have tracked patients with LBP to explore determinants of work loss and its duration,^{23 7},¹¹ which is efficient for studying risk factors for prolonged absenteeism. However, such patients may not represent the general working population. Alternatively, following a large cohort of active workers may better identify factors influencing absenteeism in the general workforce.

The duration of LBP is a key determinant of clinical prognosis. Pain lasting less than 3 months is considered acute and generally has a favourable prognosis, with most episodes improving significantly or resolving within 6 weeks. In contrast, once LBP becomes chronic (ie, ≥ 3 months), both clinical and functional outcomes tend to be poor.¹² According to Spanish law, workers who remain on sick leave for more than 18 months and are unable to resume their usual work are deemed permanently disabled and receive a lifelong pension. Therefore, to identify workers in whom preventive strategies should be prioritised, it is necessary to determine those at higher risk of LBP-related absenteeism, as well as those prone to prolonged episodes.

Findings regarding the predictors of LBP-related absenteeism, and their relative importance, vary substantially across studies. This variability may reflect methodological heterogeneity as well as differences in labour markets and social security systems. Moreover, although LBP is a sensory and emotional experience, LBP-related disability and taking sick leave are behaviours influenced by social and psychological factors that likely vary across cultural contexts.¹³ For instance, the value placed on working actively and contributing to the community for one’s self-image and social identity may differ among societies. The same applies to psychological variables; for example, catastrophising influences LBP-related disability in Scandinavian and Anglo-Saxon settings^{7 8 14} but is of little relevance in Hispanic populations.¹⁵,¹⁷

However, most available data on LBP-related absenteeism derives from Scandinavian and Anglo-Saxon countries, with limited evidence from Southern Europe. Therefore, the objectives of this study were to: (a) identify risk factors for LBP-related absenteeism in a large sample of Spanish workers and (b) develop prognostic models to identify individuals at increased risk of taking LBP-related sick leave and of accumulating more days of absence.

Methods

Study design

This was a multicentre, observational, prospective study which followed 7262 active workers over 18 months. The protocol received approval from the Ethics Committee of the Complejo Hospitalario de Toledo (IB897/07) and was registered on ClinicalTrials.gov (NCT00667316).

Setting

Participants were workers residing in Spain. Spanish law mandates annual health assessments for all active workers. In the case of small companies, these assessments are conducted by private, non-profit ‘Health Prevention and Workers’ Compensation Entities’ (HPWCEs). Companies with ≥500 employees, or ≥250 if exposed to significant occupational risks (eg, hazardous materials or biological hazards), are required to have in-house Occupational Health Services, which may also perform these annual assessments.

Spanish legislation guarantees free healthcare and financial compensation during sick leave for all employees in Spain. Accordingly, for work-related LBP episodes, workers receive a percentage of their salary for up to 18 months from day 1 (the HPWCEs cover the first 12 months, and the National Social Security System covers the remainder). The minimum compensation is 75% of salary, although certain workers (notably civil servants) receive up to 100% of their salary while on sick leave. Additionally, patients with work-related LBP may choose to receive treatment either through the National Health Service or their HPWCE, where waiting times are generally shorter.

Workers with non-work-related LBP receive no compensation for the first 4 days of sick leave and only a percentage of their salary thereafter. The minimum allowance (for non-civil servants) is 60% of salary from day 4 to day 20 and 75% from day 21 onwards. For non-work-related conditions, compensation is paid by the National Social Security System, and healthcare is provided by the National Health Service, as HPWCEs are not involved in such cases.

Spanish law does not mandate any compensation for self-employed individuals during sick leave. Self-employed workers who can afford it may purchase private insurance plans from HPWCEs, which provide coverage comparable to that of salaried employees for work-related conditions (typically 75% of usual income for a limited period, along with annual health assessments and healthcare provision at HPWCE facilities). Self-employed individuals not covered by HPWCEs are entitled to free treatment from the Spanish National Health Service (SNHS), as are all residents in the country, but they do not receive financial compensation during sick leave.

Study population

Participants were consecutively enrolled during the mandatory annual occupational health assessments.

Inclusion criteria were adequate Spanish proficiency to complete questionnaires and provision of written informed consent, including authorisation to access sick leave data during follow-up.

Exclusion criteria were: history of chronic inflammatory conditions (eg, ankylosing spondylitis), cancer or fibromyalgia; criteria for referral to surgery; and red flags for systemic diseases, unless these had been ruled out through appropriate testing.¹⁸

Procedure

While in the waiting room, participants received an informed consent form and a baseline questionnaire assessing all self-reported variables. They completed the documents alone and unassisted, although they could request clarification from the physician in charge of their health assessment. Physicians were instructed to report to the research team any question that prompted clarification requests from ≥5% of their participants.

The physician forwarded the completed questionnaires to a central coordination office. Data were anonymised using nine-digit codes. Each participant’s national ID number was stored in a sealed envelope, labelled on the front with the corresponding code and the date of the baseline assessment.

Eighteen months later, ID numbers were used to extract sick leave data from the Spanish National Social Security (SNSS) database. Following extraction, ID records were destroyed.

Although registered in 2008, recruitment did not commence until 2014 due to administrative delays. In 2016, the dissolution of the non-profit funding institution led to a temporary suspension of the study, which resumed after the research team secured funding through personal resources. The COVID-19 pandemic further delayed progress. The follow-up of the last participant was completed in February 2025.

Variables

Based on the available literature, 77 variables were selected as potential prognostic factors for LBP-related sick leave and its duration.²,⁸¹³ These variables were self-reported and collected at baseline.

Sociodemographic variables were: age, sex (male, female), marital status (five categories), cohabitation, education level (five categories), annual income (four categories) and proportion of variable income (six categories).

Clinical variables comprised body mass index (BMI) (kg/m²), from self-reported height and weight; smoking status; physical activity outside work (weekly hours); daily sleep duration; perceived sleep quality (five categories); and use of antidepressants or anxiolytics.

Work-related variables included job type (four categories); contract duration (three categories); employment status (self-employed vs employed); type of work shift (six categories); night shifts (four categories); company size (<50 or ≥50 workers); economic sector (private/public); weekly working hours; and years at current company. Physical demands were assessed through six binary questions: prolonged standing or sitting (>50% of the working time), frequent bending or twisting, load handling/lifting and exposure to whole-body vibrations. Data on prior sick leave (any cause and LBP-specific) in the past year were also collected.

LBP-related variables included: number of LBP episodes during the previous year, categorised into five groups (‘0’, ‘1–2’, ‘3–4’, ‘5–6’ and ‘>6’) and collapsed for analysis into two categories (‘<3’ and ‘≥3’), typical duration (≤14 or >14 days),¹⁹ history of referred pain, current pain intensity (Visual Analogue Scale (VAS)),²⁰ LBP-related disability (Roland-Morris Questionnaire, (RMQ)),²¹ pain in bed, when rising, sitting or changing postures, having sought care for LBP, satisfaction with care (five categories), recommendation for rest or exercise, having received education on LBP, type of education received (four categories), actual exercise (yes/no), use of medication for LBP, history of surgery for LBP and number of surgeries.

Psychosocial variables included the Spanish versions of the Fear-Avoidance Beliefs Questionnaire (FABQ), FAB about work and physical activity (FAB-PA);²² the Catastrophising subscale of the Coping Strategies Questionnaire (CSQ)²³; and the short version of the Copenhagen Psychosocial Questionnaire (CoPsoQ) covering 10 dimensions.^{24 25}

Additional subjective perceptions included four questions with five categories (perceived likelihood of LBP-related sick leave in the next 12 months, perceived economic impact and anticipated employer and peers’ reactions to potential sick leave) and four self-efficacy items (0–10 scale) on managing work during illness.

Outcome variables were collected at 18 months: LBP-related sick leave and total number of days on LBP-related sick leave (categorised as <30 or ≥30 days). Data were obtained from the SNSS database, which records the diagnosis justifying sick leave and its duration. This database is considered the gold standard for sick leave information, as data are entered by the physicians issuing the sick leave and are used to determine sick leave payments.

Analysis

Sample size was set at 7200, assuming that 50% would experience LBP during follow-up,^{26 27} 20% of these would request sick leave, 80% of leaves would be under 30 days and ≤30% of leave data could be missing (eg, due to death or permanent disability).

Categorical variables were reported as counts and percentages; continuous variables as means (SD) or medians (IQR), depending on distribution.

Group comparisons (eg, subjects with LBP-related sick leave vs no leave) used Pearson’s χ2 test for categorical variables and either Student’s t-test or Mann-Whitney U test for continuous variables, depending on normality.

Fractional polynomial modelling was applied to continuous predictors. Variables with missing not at random patterns were excluded. For missing completely at random with <10% missingness, complete case analysis was applied. For missing at random with ≥10% missing values, multiple imputation by chained equations was planned.

Variables with p values <0.10 in the univariable analyses and those that were clinically relevant were included in the models predicting LBP-related sick leave and ≥30 days of LBP-related leave. Both prognostic models were developed using binary logistic regression. Backward stepwise elimination was applied, with a retention threshold of p<0.10.

Model development adhered to the PROGRESS framework²⁸ and TRIPOD+AI statement.²⁹

Model performance was evaluated via scaled Brier score, calibration plots and C-statistic to evaluate discrimination. Internal validation used 500 bootstrap samples. Additional performance metrics included expected-to-observed (E:O) ratio, calibration-in-the-large (CITL), and calibration slope.³⁰ Internal validation was conducted using the bsvalidation command in Stata.³⁰

All statistical analyses were conducted using Stata v18 (StataCorp. 2023. Stata Statistical Software: Release 18. College Station, TX: StataCorp LLC).

Results

Participant recruitment and sample characteristics

Participants were recruited in three waves: February 2014–January 2016, February 2017–December 2019 and September 2022–August 2023.

Of 7320 invited workers, 58 (0.79%) declined. No deaths, retirements or disabilities occurred, resulting in a final sample of 7262, with no losses to follow-up. No physician reported that any question or questionnaire elicited requests for clarification from ≥5% of the invited participants.

Recruitment covered 48 of Spain’s 50 provinces, across its 17 administrative regions, and all industrial sectors (per ISIC Rev. 4) except ‘Water supply; sewerage, waste management and remediation’.³¹

A total of 5910 participants were recruited via eight major HPWCEs and 1352 from 11 large employers in education, healthcare, tourism, banking and public sectors. The mean (SD) age was 42 (11) years, and 46.6% (n=3447) were male. Baseline characteristics are shown in table 1 and online supplemental table 1.

Table 1. Characteristics of participants (n=7262) corresponding values.

Sociodemographic data
Age (years)*	43 (11)
Sex†
Male	3252 (46.2%)
Female	3784 (53.8%)
Anthropometric and general health data
Body mass index (kg/m2)*	26 (14)
Physical activity (outside work) (yes)†	4288 (59.8%)
Work-related variables
Being self-employed†
Self-employed	1018 (14.5%)
On payroll/other	5980 (85.5%)
Working hours/week‡	40 (35; 41)
Number of days on sick leave for any reason (including low back pain (LBP)) during the last 12 months‡	9 (3; 30)
Number of days on sick leave for LBP during the last 12 months‡	6 (2; 20)
Related to LBP
Number of episodes of LBP (with or without referred pain) during the last 12 months†
0–2	4827 (67.0%)
≥3	2375 (33.0%)
Usual duration of LBP episodes (if any)†
≤14 days	4467 (89.6%)
>14 days	519 (10.4%)
Duration of the longest episode†
≤14 days	3851 (77.3%)
>14 days	1134 (22.7%)
Current severity of LBP (VAS‡§, among subjects with VAS >0 (n=4142)	3 (2; 5)
Severity of referred pain (VAS‡§, among subjects with VAS >0 (n=2361)‡	3 (1; 5)
Disability (RMQ‡¶, among subjects with RMQ >0 (n=4141)‡¶	2 (1; 4)
Feeling LBP while in bed (yes)†	2819 (39.4%)
Has been recommended bed rest for treating or preventing LBP (yes)†	1403 (38.5%)
Has been recommended exercise for treating or preventing LBP (yes)†	4408 (64.5%)
Has undergone spine surgery (yes)†	179 (4.4%)
Psychosocial variables
Fear-avoidance beliefs (FAB)‡**, among subjects with FAB >0 (n=5742)‡**	35 (23; 48)
Catastrophisingࠠ, among subjects with CSQ >0 (n=4776)ࠠ	5 (2; 10)
Expected likelihood of taking LBP-related sick leave within the next 12 months†
None	1589 (22.3%)
Few	3528 (49.5%)
Moderate	1443 (20.3%)
High	455 (6.4%)
Very high	112 (1.6%)
Self-perceived economic impact of sick leave†
None	709 (10.0%)
Low	1392 (19.5%)
Moderate	1841 (25.8%)
High	1718 (24.1%)
Very high	1464 (20.6%)
LBP-related sick leave during the 18 months, follow-up period†
No	6727 (92.6%)
Yes	535 (7.4%)
No. of LBP-related sick leaves during the 18-month follow-up period‡	1 (1; 2)
Total number of days on LBP-related sick leave during the 18-month follow-up period‡	14 (10; 32)
Duration of the longest LBP-related sick leave during the 18-month follow-up period (days)‡	12 (10; 13)

See online supplemental table 1 for additional variables that were collected and their corresponding values.

^*Mean (SD).

^†n (%).

^‡Median (p25; p75).

^§Visual Analogue Scale (VAS) on pain severity (range, from best to worst; 0-10).

^¶Roland-Morris Questionnaire (range, from best to worst: 0-24).

^**Fear-Avoidance Beliefs Questionnaire (range, from best to worst: 0-96).

^††Coping Strategies Questionnaire (CSQ; range, from best to worst, 0-36).

Sick leave

During the 18-month follow-up, 535 participants (7.4%; 95% CI 6.8 to 8.0) took sick leave for LBP. Table 2 and online supplemental table 2 show differences in baseline characteristics between those who did and did not take leave. Significant univariable associations were found for sex, age, BMI, contract duration, prior sick leave, episode duration and frequency, referred pain, RMQ, FAB-PA, catastrophising, job insecurity and expectations of sick leave. Pain when rising and sleep quality were borderline significant.

Table 2. Participants who were and were not on sick leave due to low back pain (LBP) during the follow-up period. Univariable analysis.

	No LBP-related sick leave during the period	LBP-related sick leave during the period	P value
N*	6727 (92.6%)	535 (7.4%)
Age (years)†	45 (33; 51)	49 (39; 53)	<0.001
Sex*
Male	3040 (46.7%)	212 (40.6%)	0.008
Female	3474 (53.3%)	310 (59.4%)
Body mass index (kg/m2)†	24.6 (22.3; 27.2)	25.0 (22.5; 27.6)	0.007
Physical activity (outside work)*
No	2664 (40.1%)	216 (40.8%)	0.779
Yes	3974 (59.9%)	314 (59.2%)
Being self-employed*
Self-employed	953 (14.7%)	65 (12.6%)	0.192
On payroll (other)	5529 (85.3%)	451 (87.4%)
Working hours/week†	40 (35; 42)	40 (35; 40)	0.017
Number of days on sick leave for any reason during the last 12 months (n=1819)†	8 (3; 30)	15 (3; 40)	0.045
Number of days on sick leave for LBP during the last 12 months (n=611)†	6 (2; 20)	8 (3; 30)	0.460
Number of episodes of LBP with or without referred pain, during the last 12 months*
0–2	4495 (67.4%)	332 (62.4%)	0.019
≥3	2175 (32.6%)	200 (37.6%)
Usual duration of LBP episodes*
No episodes	2002 (30.3%)	149 (28.1%)	<0.001
≤14 days	4146 (62.8%)	321 (60.6%)
>14 days	459 (6.9%)	60 (11.3%)
Duration of the longest episode*
No episodes	2008 (30.4%)	147 (27.7%)	<0.001
≤14d	3588 (54.3%)	263 (49.6%)
>14d	1014 (15.3%)	120 (22.6%)
Severity of LBP (VAS)†‡, among subjects with VAS >0 (n=4142)	3 (2; 5)	3 (2; 6)	0.211
Severity of referred pain (VAS)†‡, among subjects with VAS >0 (n=2361)	3 (1; 5)	3 (2; 5)	0.116
Disability†§, among subjects with RMQ >0 (n=4141)	2 (1; 4)	2 (1; 5)	0.008
Feeling LBP while in bed*
No	4041 (60.9%)	302 (57.2%)	0.093
Yes	2593 (39.1%)	226 (42.8%)
Has been recommended bed rest*
No	2076 (61.8%)	169 (58.5%)	0.265
Yes	1283 (38.2%)	120 (41.5%)
Has been recommended exercise to prevent or treat LBP*
No	2230 (35.2%)	192 (38.2%)	0.187
Yes	4097 (64.8%)	311 (61.8%)
Has undergone spine surgery*
No	3591 (95.6%)	321 (96.1%)	0.652
Yes	166 (4.4%)	13 (3.9%)
Fear-avoidance beliefs (FABQ)†¶, among subjects with FAB >0 (n=5742)†¶	35 (23; 48)	36 (24; 51)	0.112
Catastrophising (CSQ)†**, among subjects with CSQ >0 (n=4776)†**	5 (2; 10)	6 (3; 11)	0.043
Expected likelihood of taking LBP-related sick leave within the next 12 months*
None/few	4763 (72.2%)	354 (66.8%)	<0.001
Moderate	1332 (20.2%)	111 (20.9%)
High/very high	502 (7.6%)	65 (12.3%)
Self-perceived negative economic impact if on sick leave*
None/low	1946 (29.5%)	155 (29.7%)	0.016
Moderate	1732 (26.2%)	109 (20.9%)
High/very high	2924 (44.3%)	258 (49.4%)

See online supplemental table 2 for additional variables that were collected and their corresponding values.

^*n (%): Pearson’s test.

^†Median (p25; p75); Mann-Whitney U test.

^‡Visual Analogue Scale (VAS) on pain severity (range, from best to worst; 0-10).

^§Roland-Morris Questionnaire (RMQ; range, from best to worst: 0-24).

^¶Fear-Avoidance Beliefs Questionnaire (FABQ; range, from best to worst: 0-96).

^**Coping Strategies Questionnaire (CSQ; range, from best to worst, 6-42).

Of those on sick leave, 162 (30.3%) had ≥30 days’ absence. Table 3 and online supplemental table 3 show longer leave was associated with sex and pain while in bed. Perceived employer reaction was close to statistical significance.

Table 3. Differences between subjects who, during the follow-up period, were on sick leave for low back pain (LBP) for less and for more than 30 days. Univariable analysis.

	Duration of LBP-related sick leave
	<30d	≥30d	P value
N*	373 (69.7%)	162 (30.3%)
Age (years)†	48 (40; 53)	49 (38; 54)	0.408
Sex*
Male	159 (43.7%)	54 (34.2%)	0.042
Female	205 (56.3%)	104 (65.8%)
Body mass index (kg/m2)†	24.9 (22.5; 27.4)	25.2 (22.5; 27.8)	0.301
Physical activity (outside work)†
No	150 (40.7%)	64 (39.8%)	0.846
Yes	219 (59.3%)	97 (60.2%)
Being self-employed*
Self-employed	312 (86.4%)	139 (89.7%)	0.308
On payroll (other)	49 (13.6%)	16 (10.3%)
Working hours/week†	40 (35; 40)	38 (35; 40)	0.305
Number of days on sick leave for any reason during the last 12 months (n=151)†	14 (3; 40)	15 (3; 40)	0.809
Number of days on sick leave for LBP during the last 12 months (n=49)†	6.5 (3; 30)	11 (6; 30)	0.673
Number of episodes of LBP with or without referred pain, during the last 12 months*
0–2	234 (63.1%)	98 (60.9%)	0.630
≥3	137 (36.9%)	63 (39.1%)
Usual duration of LBP episodes*
No episodes	100 (27.0%)	48 (30.0%)
≤14 days	232 (62.7%)	91 (56.9%)	0.408
>14 days	38 (10.3%)	21 (13.1%)
Duration of the longest episode*
No episodes	99 (26.8%)	47 (29.4%)
≤14d	190 (51.4%)	74 (46.2%)	0.558
>14d	81 (21.9%)	39 (24.4%)
Severity of LBP (VAS)‡, among subjects with VAS >0 (n=318)	4 (2; 6)	3 (2; 6)	0.571
Severity of referred pain (VAS)‡; among subjects with VAS >0 (n=189)	4 (2; 5)	3 (2; 6)	0.810
Disability (RMQ)§, among subjects with RMQ >0 (n=320)	3 (1; 5)	2 (1; 6)	0.421
Feeling LBP while in bed*
No	222 (60.3%)	79 (49.4%)	0.019
Yes	146 (39.7%)	81 (50.6%)
Has been recommended bed rest*
No	124 (61.7%)	46 (51.7%)	0.111
Yes	77 (38.3%)	43 (48.3%)
Has been recommended exercise to prevent or treat LBP*
No	139 (39.7%)	52 (34.0%)	0.223
Yes	211 (60.3%)	101 (66.0%)
Has undergone spine surgery*
No	219 (96.0%)	103 (96.3%)	0.926
Yes	9 (4%)	4 (3.7%)
Fear-avoidance beliefs (FABQ) among subjects with FAB >0 (n=409)†¶	37 (24; 49)	36 (23; 56)	0.871
Catastrophising (CSQ)†** among subjects with CSQ >0 (n=357)	6 (3; 11)	5 (3; 10)	0.854
Expected likelihood of taking LBP-related sick leave within the next 12 months*
None/few	247 (66.8%)	106 (66.2%)	0.819
Moderate	76 (20.5%)	36 (22.5%)
High/very high	47 (12.7%)	18 (11.2%)
Self-perceived negative economic impact if on sick leave*
None/low	105 (28.8%)	51 (32.5%)	0.567
Moderate	80 (21.9%)	29 (18.5%)
High/very high	180 (49.3%)	77 (49.0%)

See online supplemental table 3 for additional variables that were collected and their corresponding values.

^*n (%): Pearson’s test.

^†Median (p25; p75); Mann-Whitney U Test.

^‡Visual Analogue Scale (VAS) on pain severity (range, from best to worst; 0-10).

^§Roland-Morris Questionnaire (RMQ; range, from best to worst: 0-24).

^¶Fear-Avoidance Beliefs Questionnaire (FABQ; range, from best to worst: 0-96).

^**Coping Strategies Questionnaire (CSQ; range, from best to worst, 6-42).

Multivariable models

Fractional polynomials did not improve fit. The final model for sick leave included: age, self-employment, expectations of taking sick leave, perceived economic impact, LBP episode duration and job insecurity.

Variables included in the model on sick leave were: age, sex, marital status, education, BMI, smoking, physical activity, perceived sleep quality, employment status, company size, economic sector, LBP episode frequency and duration, referred pain, pain on rising, healthcare use, exercise behaviour, years at current company, expectations of taking sick leave, perceived economic impact, perceived employer reaction, perceived peer reaction, job physical demands, whole-body vibration exposure, RMQ score, catastrophising (CSQ) and job insecurity (CoPsoQ). The number of variables included in the full model adhered to the 1:10 rule of thumb for developing multivariable models.³²

The final predictors of sick leave were: age (3% increased risk per year; OR 1.03), self-employment (reduced risk; OR 0.67), expectations of taking sick leave (moderate expectations increased the risk by 4%, and high or very high expectations increased it by 44%), perceived economic impact of sick leave (high or very high impact increased the risk by 48%), usual duration of LBP episodes (if >14 days, increased risk by 43%) and job insecurity according to the CoPsoQ questionnaire (reduced risk; OR 0.70) (table 4).

Table 4. Predictive model on being on sick leave for low back pain (LBP) during the next 18 months. Multivariable analysis (logistic regression).

	OR	95% CI	P value
Being self-employed (yes)	0.67	0.50 to 0.91	0.011
Age (years)	1.03	1.02 to 1.04	<0.001
Economic impact of sick leave
Moderate	Reference category
None/low	1.25	0.96 to 1.64	0.102
High/very high	1.48	1.15 to 1.90	0.002
Usual duration of LBP episodes
≤14 days	Reference category
No episodes	0.97	0.78 to 1.21	0.770
>14 days	1.43	1.03 to 1.99	0.031
Self-perceived likelihood of sick leave
None/few	Reference category
Moderate	1.04	0.82 to 1.33	0.752
High/very high	1.44	1.04 to 1.98	0.028
Job insecurity
Green	Reference category
Yellow	0.83	0.59 to 1.16	0.268
Red	0.70	0.51 to 0.96	0.028
Constant	0.02	0.01 to 0.04	<0.001

The maximal model included the following variables: age, sex, marital status, academic level, BMI, smoking, hours of physical activity per week, perceived sleep quality, being self-employed, number of workers in the company, economic sector, number of LBP episodes during the previous 12 months, usual duration of LBP episodes, duration of the longest episode, history of referred pain, feeling LBP when raising from bed, has sought care for LBP, doing exercise to prevent or treat LBP, number of years working in the same company, expected likelihood of taking LBP-related sick leave within the next 12 months, self-perceived economic impact of sick leave, anticipated employer reaction to potential LBP-related sick leave, anticipated peers' reaction to potential LBP-related sick leave, prolonged standing, prolonged sitting, frequent bending, frequent twisting, load lifting or handling, exposure to whole-body vibrations, disability (Roland-Morris Questionnaire score), catastrophising (Coping Strategies Questionnaire score) and job insecurity (as measured by the CoPoQ questionnaire). CoPoQ: Copenhagen Psychosocial Questionnaire (range, from best to worst, green, yellow, red)

The model equation is: log-oddsLBP Sick Leave 18 m = −3.77–0.40×Self-employment + 0.03× Age (per year) + 0.23×None/low economic impact+0.39×High/very high economic impact − 0.03×No previous LBP episode+0.36×(>14 days previous LBP episode) + 0.04×Moderate expectation of taking sick leave+0.36×High/very high expectation of taking sick leave − 0.19×Yellow level in job insecurity − 0.35×Red level in job insecurity

The variables are coded as follows:

• Self-employment: 0=No, 1=Yes.

• Economic impact: 0=None/low, 1=Moderate, 2=High/very high.

• Usual duration of LBP episodes: 0=No previous LBP episode, 1 = ≤14 days, 2 = >14 days.

• Expectation of taking sick leave: 0=None/few, 1=Moderate, 2=High/very high.

• Job insecurity: 0=Green (low perception), 1=Yellow (moderate perception), 2=Red (high perception).

The corresponding probability of being on sick leave for LBP during the next 18 months is calculated as: $Pro{b}_{LBPSickLeave18m}=\frac{1}{1+\mathrm{e}\mathrm{x}\mathrm{p}\left(-\mathrm{l}\mathrm{o}\mathrm{g}-\mathrm{o}\mathrm{d}\mathrm{d}\mathrm{s}\mathrm{L}\mathrm{B}\mathrm{P}\mathrm{S}\mathrm{i}\mathrm{c}\mathrm{k}\mathrm{L}\mathrm{e}\mathrm{a}\mathrm{v}\mathrm{e}18\mathrm{m}\right)}$

The model showed good calibration but poor discrimination (C-statistic=0.620, 95% CI 0.594 to 0.646) and yielded a pseudo R² of 0.024, indicating limited overall explanatory power. The scaled Brier score was 1.3%. After optimism adjustment through internal validation, the scaled Brier score decreased to 0.9%, the C-statistic to 0.607 (95% CI 0.580 to 0.630), the E:O ratio was 1.0 (95% CI 0.91 to 1.10), CITL was 0.007 (95% CI −0.096 to 0.107) and the calibration slope was 0.901 (95% CI 0.734 to 1.103).

As illustrated in the calibration plot (online supplemental figure 1), both events and non-events cluster in the region of low predicted risk, indicating that the model fails to assign high probabilities of sick leave to individuals who actually experience the outcome (ie, poor discrimination). In contrast, the line plotting predicted vs observed probabilities closely follows the 45-degree diagonal, demonstrating good agreement between predicted and observed outcomes (ie, good calibration).

Variables included in the model on days on sick leave during the follow-up period were: age, sex, BMI, physical activity, sleep quality, employment status, economic sector, referred pain, healthcare use, bed rest recommendations, pain while in bed, FAB-PA, expectations of sick leave and peer reaction. Only self-employment and pain while in bed remained in the final model (table 5). The C-statistic was 0.604. The class imbalance rendered internal validation unfeasible, and prediction of the minority class was impossible. The calibration plot based on the estimation sample is presented in online supplemental figure 2. The model yielded a pseudo-R² of 0.033.

Table 5. Model on sick leave for accumulating ≥30 days on sick leave for low back pain (LBP), over an 18-month period. Multivariable analysis (logistic regression).

	OR	95% CI	P value
Being self-employed (yes)	0.38	0.13 to 1.16	0.089
Feeling LBP while in bed (yes)	2.11	1.15 to 3.86	0.016
Constant	0.27	0.17 to 0.44	<0.001

The maximal model included the following variables: age, sex, body mass index, physical activity (outside from work), perceived sleep quality, being self-employed, economic sector, history of referred pain, has sought care for LBP, having been recommended bed rest, feeling LBP while in bed, fear-avoidance beliefs on physical activity, expected likelihood of taking LBP-related sick leave within the next 12 months and anticipated peers' reaction to potential LBP-related sick leave.

Apparent performance: c-statistic = 0.604. The number of observations was too low to allow for internal validation of the model.

Discussion

This study developed prognostic models for LBP-related sick leave using a wide range of biological, clinical, sociodemographic, work-related and psychosocial variables. While both models showed good calibration, discrimination was poor, limiting their usefulness in practice.

Very few biological, ergonomic and clinical variables assessed in this study were associated with sick leave, and they were generally less relevant than psychosocial factors (tables 2–5 and online supplemental tables 2,3). This is consistent with findings from previous research.²,⁵

In Spain, self-employment entails greater income instability, fewer social protection and limited sick leave benefits. In this study, it was the only factor associated with a reduced risk of both taking sick leave and accumulating ≥30 days of leave. This aligns with the finding that higher job insecurity reduces the risk of sick leave and is consistent with prior studies in other contexts.^{33 34}

Despite all participants being actively working at baseline and being recruited from the workplace (not clinical settings), 57% reported current LBP (VAS ≥1), 25% reported moderate to severe pain (VAS ≥4) and 60% used medications for LBP (table 1 and online supplemental table 1). Yet only 2.23% accumulated ≥30 days of sick leave. Spanish legislation incentivises individuals with LBP to report their condition as work-related, as this entitles them to earlier and higher financial compensation and to private healthcare with shorter waiting times. Therefore, it is unlikely that workers experiencing substantial work-related pain would refrain from reporting it. On the contrary, these findings suggest that LBP is highly prevalent among middle-aged adults,^{1 26 27} but does not typically result in prolonged absence unless it becomes incapacitating or is accompanied by psychosocial factors.

Many variables were significantly associated with LBP-related sick leave in the univariable analyses. The influence of some of them, such as fear-avoidance beliefs or catastrophising, is plausible and has been suggested in some previous studies.²,^{46 35} However, few of these associations remained significant in multivariable models. This underlines the need for future studies to use large, well-powered samples and multivariable analyses that account for multiple interacting factors.

It is impossible to completely rule out a lack of statistical power for some associations that were not statistically significant in this study. However, it can be argued that the influence of any variable showing no significant association in a sample of over 7000 subjects is probably of minimal practical relevance.

More than one-third of participants had been advised to rest in bed to prevent or treat LBP,³⁸ and the type of educational intervention most frequently received was not the most effective one (table 1 and online supplemental table 1).^{8 16 39} This suggests that strategies for LBP prevention and management do not fully align with the available evidence.^{18 40}

Strengths and limitations

Annual health assessments are mandatory for all workers in Spain. In this study, over 7000 participants were recruited consecutively, with only 0.79% declining participation, and there were no exclusions or losses to follow-up. Recruitment covered all 17 administrative regions and nearly all economic sectors, and participant characteristics mirrored those of the general Spanish working population (table 1 and online supplemental table 1). This suggests that sample representativeness is not a major concern.

Other strengths of this study include comprehensive multivariable adjustment and the use of a validated national registry database (the Spanish Social Security), as the source for outcome data on LBP-related sick leave and its duration.

Several limitations should be discussed. Some of the items used to assess psychosocial perceptions lacked validation. Hence, their psychometric robustness remains uncertain. However, they demonstrated high face validity, and no comprehension issues were reported by the physicians responsible for their evaluation.

The recruitment period included the COVID-19 pandemic, during which working conditions changed and LBP-related leaves declined sharply. However, only 4.28% of participants had follow-up periods that overlapped with COVID-related restrictions in Spain. Therefore, although this may have reduced the number of sick leaves, it is unlikely to have biased or undermined the study’s results.

In Spain, work-related conditions, including LBP, are managed by private, non-profit HPWCEs, whereas non-work-related conditions are essentially managed by the SNHS. The SNHS, which is mostly government-owned and operated, typically has longer waiting times and is slower in conducting diagnostic work-ups and providing treatment. Additionally, economic benefits are higher if sick leave is due to occupational diseases compared with non-occupational conditions. This incentivises workers to claim that their LBP is work-related and to seek care through HPWCE rather than the SNHS. However, legally, any condition that arises during work or is ‘caused’, ‘facilitated’ or ‘worsened’ by work can be classified as an occupational disease. This means that, in practice, workers with LBP can choose whether to be treated by the SNHS or their HPWCE. In this study, it was impossible to determine whether LBP-related sick leave was due to actual work-related LBP or not. However, this is irrelevant to the study’s objectives.

The study design was necessarily shaped by the Spanish legal framework, which may limit the generalisability of findings. For instance, Spanish law mandates resolution of all sick leaves by 18 months (either return to work or permanent disability ruling). This led to the follow-up period being established at 18 months. Years at company were stratified as less versus ≥2 years, because the law dictates that, after this period, all contracts become ‘permanent’, which implies that the economic compensation to fire a worker increases significantly and continues to increase every year from there on. These contextual factors may influence absenteeism patterns in ways different from other countries. That said, the duration of contract was not associated with sick leave risk (table 4).

Conclusion

In conclusion, this study shows that: (a) LBP is highly prevalent among workers, even while actively working; (b) psychosocial factors appear to be more relevant than clinical or biological ones in predicting LBP-related sick leave, but early identification of individuals at high risk remains elusive; and (c) in the Spanish working force, being self-employed is the main factor associated with LBP-related sick leave and its duration.

Data availability statement

Data are available upon reasonable request.