Modifiable prognostic factors of high societal costs among people on sick leave due to musculoskeletal disorders: findings from an occupational cohort study

Abstract

Objectives

The objective was to identify modifiable prognostic factors of high societal costs among people on sick leave due to musculoskeletal disorders, and to identify modifiable prognostic factors of high costs related to separately healthcare utilisation and productivity loss.

Design

A prospective cohort study with a 1-year follow-up.

Participants and setting

A total of 549 participants (aged 18–67 years) on sick leave (≥ 4 weeks) due to musculoskeletal disorders in Norway were included.

Outcome measures and method

The primary outcome was societal costs aggregated for 1 year of follow-up and dichotomised as high or low, defined by the top 25th percentile. Secondary outcomes were high costs related to separately healthcare utilisation and productivity loss aggregated for 1 year of follow-up. Healthcare utilisation was collected from public records and included primary, secondary and tertiary healthcare use. Productivity loss was collected from public records and included absenteeism, work assessment allowance and disability pension. Nine modifiable prognostic factors were selected based on previous literature. Univariable and multivariable binary logistic regression analyses were performed to identify associations (crude and adjusted for selected covariates) between each modifiable prognostic factor and having high costs.

Results

Adjusted for selected covariates, six modifiable prognostic factors associated with high societal costs were identified: pain severity, disability, self-perceived health, sleep quality, return to work expectation and long-lasting disorder expectation. Depressive symptoms, work satisfaction and health literacy showed no prognostic value. More or less similar results were observed when high costs were related to separately healthcare utilisation and productivity loss.

Conclusion

Factors identified in this study are potential target areas for interventions which could reduce high societal costs among people on sick leave due to musculoskeletal disorders. However, future research aimed at replicating these findings is warranted.

Trial registration number

NCT04196634, 12 December 2019.

STRENGTHS AND LIMITATIONS OF THIS STUDY.

• The main strength of the current study is that it was conducted in line with the PROGnosis RESearch Strategy framework and preplanned with a published statistical analysis plan.

• There were minimal missing data, and the sample has been evaluated to be representative of the target population.

• The main limitation of the current study is that it was powered on the primary outcomes of the cohort, not on the outcomes under consideration; thus, our ability to make comprehensive adjustment for covariates was restricted.

• Due to between-country differences in healthcare organisations and mechanisms of compensation for productivity loss, readers are advised to exercise caution with generalisation of results to other countries.

Introduction

Musculoskeletal disorders are a leading cause of disability globally^{1 2} and an extensive burden to our society.^{3 4} In Norway, musculoskeletal disorders are a common cause of seeking healthcare⁵ and the most common cause of productivity loss.^{6 7} To enhance efficient use of scarce healthcare resources and reduce the societal burden imposed by musculoskeletal disorders, researchers emphasise the importance of monitoring and understanding healthcare utilisation, productivity loss and related costs.^{8 9} It is well known that most of healthcare utilisation, productivity loss and related costs stem from a relatively small group of people,6 7 10 11 and more importantly, that many of these people seem to receive unnecessary and ineffective treatment.^{8 11 12} This suggests that care for this high-cost subgroup requires quality improvement and cost reduction. An essential step towards this is to identify modifiable prognostic factors associated with high costs related to healthcare utilisation and productivity loss.¹³ Information about such factors can inform development of targeted interventions, which may enhance clinical effectiveness and cost benefits. Cost-effective interventions are crucial to improve use of scarce healthcare resources and reduce the economic burden of musculoskeletal disorders.¹³

To the best of our knowledge, only a few prospective studies have explored modifiable prognostic factors associated with high costs related to healthcare utilisation and productivity loss among people with musculoskeletal disorders,^14–20 and no study has been conducted among a sample of exclusively people on sick leave due to musculoskeletal disorders. People with high costs related to healthcare utilisation and productivity loss are a diverse population, which seems to vary across different health problems, provider characteristics, payer types, countries and age groups.^{11 21} Thus, generalisation of results cannot be done automatically. Therefore, the primary aim of the study was to identify modifiable prognostic factors of high societal costs among people on sick leave due to musculoskeletal disorders. A secondary aim was to separately identify modifiable prognostic factors of high costs related to (1) healthcare utilisation and (2) productivity loss.

Method

This study was designed and performed in accordance with the PROGnosis RESearch Strategy (PROGRESS) framework²² and is considered part of prognostic factor research.¹³ In line with recommendations from the PROGRESS framework,¹³ a study protocol (ClinicalTrials.gov Identifier: NCT04196634) including a statistical analysis plan²³ has been published, and the REporting recommendations for tumour MARKer prognostic studies (REMARK) criteria²⁴ were followed.

Design and setting

This study contains secondary analyses embedded in a prospective observational cohort study with 1 year of follow-up, conducted within the Norwegian Labour and Welfare Administration (NAV); the work package two of the MI-NAV project.²⁵

Participants and recruitment procedure

Eligible participants were people, aged 18–67 years, who were on sick leave (full or partial, ≥4 weeks) due to musculoskeletal disorders (diagnosis within the musculoskeletal (L) chapter of the International Classification of Primary Care, 2nd edition²⁶). Exclusion criteria were insufficient Norwegian or English language skills to participate in the study. Participants were invited through a link on the participants’ individual profile page on the NAV website between November 2018 and February 2019. All workers on sick leave in Norway must enter their individual profile page on the NAV website, and the invitation link was available to all workers on sick leave ≥4 weeks due to musculoskeletal disorders. All included participants signed an electronic informed consent form before study enrolment and were informed that they could withdraw at any time.

Data collection, outcome, modifiable prognostic factors and covariates

At baseline, all participants responded to an electronic questionnaire including demographic variables and a set of patient-reported measures. Data on healthcare utilisation were collected from public records including the Norwegian Patient Registry (NPR) and the Municipal Patient and User Registry (KPR). Data on productivity loss were collected from public records (NAV), containing dates, and grading of absenteeism, work assessment allowance and disability pension, as well as the related diagnostic code, and contracted workhours. Data on healthcare utilisation and productivity loss were collected in the period from baseline to 3 months retrospectively, and in the 1-year follow-up period. All information was stored and analysed securely through the Services for sensitive data (TSD).

Outcomes

The primary outcome of the current study was societal costs aggregated for 1 year of follow-up and dichotomised as high or low. Having high costs was defined as participants with costs in the top 25th percentile.^{15 16} Healthcare utilisation was collected from public records (NPR, KPR) and included: primary healthcare use (general practitioner, physiotherapist, chiropractor and emergency room consultations) and secondary/tertiary healthcare use (outpatient contacts, day surgery, ordinary admission with overnight stay and other admissions without overnight stay). Total costs of healthcare utilisation per person were estimated based on reimbursement rates collected from NPR and KPR. Productivity loss was collected from public records (NAV) and included: productivity loss related to absenteeism, work assessment allowance and disability pension. Total costs of productivity loss per person were estimated based on number of days with productivity loss, adjusted for employment rate and grading of productivity loss, multiplied by an estimated average wage rate (from official statistics in Norway) including taxes and social costs.

Secondary outcomes of the current study were costs related to separately (1) healthcare utilisation aggregated for 1 year of follow-up and dichotomised as high (top 25th percentile) or low, and (2) productivity loss aggregated for 1 year of follow-up and dichotomised as high (top 25th percentile) or low.

Modifiable prognostic factors

Modifiable prognostic factors are factors expected to have the potential to be modified or improved by appropriate care or treatment, and therefore classified as modifiable. Modifiable prognostic factors of high costs related to healthcare utilisation and productivity loss were based on previous scientific literature on patients with back pain primarily,14–16 20 21 27–31 as well as people with musculoskeletal disorders18 19 32–35 and non-diagnosis-specific studies,^{11 36–43} and included the following self-reported variables measured at baseline:

• Pain severity14–16 18–20 27 28 32 33 43 measured by the Numeric Rating Scale (range 0–10, higher score indicating higher pain severity).⁴⁴

• Disability14–16 18 20 27–29 31–34 43 measured by a single item (Q3) from the EuroQol 5 dimensions (EQ-5D-5L)⁴⁵ and categorised into no/slight problems, moderate problems or severe problems/unable to do.

• Self-perceived health^{14 19 20} measured by a single item (EQ VAS) from the EQ-5D-5L (range 0–10, reversed score, thus higher score indicating poorer health).⁴⁵

• Depressive symptoms15 16 20 21 29 30 32 43 measured by a single item (Q6) from the Örebro Musculoskeletal Pain Screening Questionnaire Short Form (ÖMPSQ-SF) (range 0–10, higher score indicating more depression symptoms).⁴⁶

• Sleep quality^{40 41} measured by a single item (Q4) from the ÖMPSQ-SF (range 0–10, reversed score, thus higher score indicating poorer sleep quality).⁴⁶

• Health literacy⁴² measured by a single item (Q12) from the Musculoskeletal Health Questionnaire (MSK-HQ)⁴⁷ and recategorised into completely/very well understanding, moderate understanding or slightly/no understanding.

• Work satisfaction measured by a single item (0–10, higher score indicating lower work satisfaction).

• Long-lasting disorder expectation^{43 48} measured by a single item (Q6) from the Keele STarT MSK tool (STarT MSK)⁴⁹ and categorised into yes/no.

• Return to work expectancy⁴³ measured by a single item (Q8) from ÖMPSQ-SF (range 0–10, reversed score, thus higher score indicating lower return to work expectation).⁴⁶

Covariates

Prognostic factor research may vary depending on context (time, place, setting) and characteristics of the study population. We therefore adjusted for potential covariates when evaluating the modifiable prognostic factors. Potential covariates were based on previous scientific literature (as described above), and included the following self-reported variables measured at baseline: sex (female/male)11 14 27 28 31 33 35–38 43; age (years)11 14 18 27 31 33 35 37 38 43; education level27 29 35 36 39 43 categorised into low or high (university level); and pain duration^{16 33} measured by a single item (Q1) from the ÖMPSQ-SF⁴⁶ and categorised into <3 months or ≥3 months. In addition, the following public record variables were included as covariates: absenteeism-related diagnosis type at baseline collected from NAV and categorised into ‘upper/lower limb conditions’, ‘back and neck conditions’, ‘joint/inflammatory conditions’, ‘injury or trauma’ or ‘other MSK conditions’; total costs related to healthcare utilisation during a period of 3 months prior to inclusion^{19 32 35} collected and estimated from NPR and KPR as described above; and total costs related to productivity loss during a period of 3 months prior to inclusion19 32 34 35 collected and estimated from NAV as described above.

Analyses

All analyses were outlined in the statistical analysis plan published a priori²³ and performed using the IBM SPSS V.26 (IBM Corporation) or Stata V.16.1 (StataCorp LLC). We considered our study as explanatory. Thus, no correction for multiple testing was performed and p <0.05 were considered statistically significant. All statistical tests were two sided.

Missing data

Missing value patterns were visually explored, and missingness at random assumed. Also, we found evidence against the hypothesis that values were not missing completely at random (Little’s test, p >0.05). Missing data raged from 0.0% to 1.6% for included baseline variables, and there were no missing data for variables used to calculate the outcome scores. Due to a small proportion of missing data (0.2%) and minimal differences between responders and non-responders, complete case analyses were performed.

Healthcare utilisation, productivity loss and cost estimation

Type and frequency of use of different healthcare resources were calculated for the 1 year of follow-up. Costs of healthcare utilisation per person were estimated based on reimbursement rates collected from NPR and KPR. Non-healthcare costs related to provision of healthcare (such as transportation) were not estimated. Days of productivity loss were calculated for the 1 year of follow-up and adjusted for employment rate and grading of productivity loss. Costs of productivity loss per person were estimated by multiplying number of days with complete productivity loss by an estimated average wage rate (from official statistics in Norway) including taxes and social costs (absenteeism = €343/workday; work assessment allowance and disability pension = €227/workday). All costs were converted and presented in euros (€) 2022 and estimated with both mean and median values with 95% CI, using bias-corrected and accelerated bootstrapping (1000 simulations). Cost data are commonly skewed, thus both mean and median values were presented to support result interpretation. Norwegian prices were recalculated to euros using the exchange rate from January 2022 (€1=NOK 10).

Identification analysis

Univariable and multivariable binary logistic regression models were used to investigate associations (crude and adjusted for selected covariates) between each predefined modifiable prognostic factor and total costs related to (1) healthcare utilisation and productivity loss, (2) healthcare utilisation and (3) productivity loss. The cost score was entered into the model as a dependent dichotomous variable (high cost defined as participants with cost in the top 25th percentile, yes/no). Linearity of continuous independent variables was explored using the multivariable fraction polynomial method.⁵⁰ The results were presented as crude and adjusted OR with 95% CI.

Sensitivity analysis

To assess credibility of the societal cost calculation related to the primary analyses, the calculation was conducted without outliers. Outliers were identified with simple scatterplots by visual inspection and defined as participants with remarkably high societal costs.

Sample size

This study contains secondary analyses embedded in the work package two of the MI-NAV project. Details on sample size calculation related to the primary aims of the cohort are provided elsewhere.²⁵ To determine statistical power of this study, we used number of events per parameter (EPP)^51–55 and the rule of thumb of ‘10 events per parameter included’.^56–59 With a fixed sample size of 549 participants included in the cohort, we anticipate 137 participants to be in the top 25th percentile of costs and categorised as having high costs (yes/no) (events). An EPP of 10 would allow a maximum of 13 parameters to be included in the final multivariable prediction model.

Deviations from the published statistical analysis plan

During this work, limitations to our prespecified analysis plan were identified. The following changes were performed: first, disability was measured by a single item (Q3) from the EQ-5D-5L, not by a single item (Q4) from the MSK-HQ. The MSK-HQ Q4 measures disability during washing and dressing, whereas the EQ-5D-5L Q3 measures disability in usual activities, which seems more appropriate in this study. Second, depressive symptoms were measured by a single item (Q6) from the ÖMPSQ-SF, not by a single item (Q11) from the MSK-HQ. The MSK-HQ Q11 measures depressive symptoms on a categorical scale, whereas the ÖMPSQ-SF Q6 measures depressive symptoms at a continuous scale ranging from 0 to 10, which seems more appropriate to simplify result interpretation. Third, flow of participants through the study was not reported by a flow chart, as no participants dropped out.

Patient and public involvement

Patient representatives with various musculoskeletal disorders were part of the scientific board of the MI-NAV project and involved in designing and establishing the project. They provided guidance on research objectives, conduction of the project and development of information material provided to the participants.

Results

A total of 549 participants with a median (range) age of 50 (19–68) years were included in the current study. Just over half of the included participants were females. On average, they reported moderate pain, and their absenteeism was most often related to musculoskeletal disorders in the upper limbs or back. Table 1 shows participant characteristics and clinical status at baseline, along with the proportion of missing data per variable.

Table 1.

Participants characteristics and clinical status at baseline

All participants (n=549)		Missing, n (%)
Female, n (%)	309 (56)	0 (0)
Age in years, median (IQR)	50 (42–57)	0 (0)
Education at university level, n (%)	220 (40)	1 (0.2)
Mother tongue Norwegian, n (%)	473 (87)	2 (0.4)
Full-time employed, n (%)	420 (77)	0 (0)
Diagnosis (ICPC-2),* n (%)		0 (0)
Upper limb conditions	121 (22)
Lower limb conditions	47 (9)
Neck conditions	36 (7)
Back conditions	107 (19)
Joint or inflammatory conditions	54 (10)
Injuries or trauma	51 (9)
Other MSK conditions	133 (24)
Pain severity average last week (NRS, 0–10), mean (SD)	6 (2)	0 (0)
Pain duration, n (%)		0 (0)
<3 months	114 (21)
3–6 months	87 (16)
> 6 months	348 (63)
Disability (EQ-5D-5L, Q3), n (%)		1 (0.2)
No problems doing usual activities	36 (6)
Slight problems doing usual activities	173 (32)
Moderate problems doing usual activities	200 (36)
Severe problems doing usual activities	119 (22)
Unable to do usual activities	20 (4)
Self-perceived health (EQ-5D-5L, EQ VAS, 0–10),† mean (SD)	5 (2)	7 (1)
Depressive symptoms (ÖMPSQ-SF, Q6, 0–10),† mean (SD)	3 (3)	0 (0)
Sleep quality (ÖMPSQ-SF, Q4, 0–10),† mean (SD)	5 (3)	0 (0)
Long-lasting disorder expectation (STarT MSK, Q6), n (%)	448 (82)	0 (0)
Health literacy (MSK-HQ, Q12), n (%)		0 (0)
Completely understanding of condition/treatment	75 (14)
Very well understanding of condition/treatment	268 (49)
Moderately understanding of condition/treatment	134 (24)
Slightly understanding of condition/treatment	55 (10)
No understanding of condition/treatment	17 (3)
Return to work expectancy (ÖMPSQ-SF, Q8, 0–10)†	4 (3)	0 (0)
Work satisfaction (0-10),† mean (SD)	2 (3)	9 (2)
Healthcare utilisation prior to inclusion‡
Primary care consultation last 3 months, n (%)		0 (0)
General practitioner	539 (98)
Physiotherapist	216 (39)
Chiropractor	68 (12)
Emergency room	45 (8)
Secondary/tertiary care last 3 months, n (%)		0 (0)
Outpatient contact	203 (37)
Day surgery	30 (6)
Ordinary admission with overnight stay	28 (5)
Other admissions without overnight stay	46 (8)
Productivity loss prior to inclusion§
Days of sick leave last 3 months, median (IQR)	30 (21–43)	0 (0)
Days of work assessment allowance last 3 months, median (IQR)	0 (0)	0 (0)
Days of disability benefits last 3 months, median (IQR)	0 (0)	0 (0)

*Absenteeism-related diagnoses type collected from the Norwegian Labour and Welfare Administration (NAV) registry.

†A lower score is better.

‡Collected from public records; the Norwegian Patient Registry (NPR) and the Municipal Patient and User Registry (KPR).

§Collected from the NAV registry, measured as calendar days and adjusted for employment rate and grading of productivity loss.

EQ-5D-5L, EuroQol 5 dimensions; ICPC-2, International Classification of Primary Care 2nd edition; ÖMPSQ-SF, Örebro Musculoskeletal Pain Screening Questionnaire Short Form; MSK-HQ, Musculoskeletal Health Questionnaire; NRS, Numeric Rating Scale; STarT MSK, Keele STarT MSK tool.

Healthcare utilisation, productivity loss, cost estimation and effect

Table 2 shows healthcare utilisation and productivity loss aggregated for 1 year of follow-up for all participants. Table 3 shows total costs related to healthcare utilisation and productivity loss aggregated for 1 year of follow-up. Costs were mainly related to productivity loss, accounting for 89% of total costs during the 1 year of follow-up. A total of 137 participants (25%) were defined as having high costs related to healthcare utilisation and productivity loss (≥ €46 340), healthcare utilisation (≥€3869), and productivity loss (≥€41 854). A total of three participants (0.3%) with societal costs >€91 810 were defined as outliers. All outliers had costs related to primary, secondary and tertiary healthcare use, and costs related to absenteeism and work assessment allowance.

Table 2.

Healthcare utilisation and productivity loss aggregated for 1 year of follow-up

	All participants (n=549)	Missing, n (%)
Primary care
Participants with primary care consultation, n (%)		0 (0)
General practitioner	538 (98)
Physiotherapist	266 (49)
Chiropractor	98 (18)
Emergency room	118 (22)
No primary care consultation	7 (1)
Numbers of consultations, median (IQR)*		0 (0)
General practitioner	11 (6–16)
Physiotherapist	17 (6–31)
Chiropractor	5 (3-9)
Emergency room	1 (1–1)
Secondary/tertiary care
Participants with secondary/tertiary care consultation, n (%)		0 (0)
Outpatient contact	333 (61)
Day surgery	48 (9)
Ordinary admission with overnight stay	60 (11)
Other admissions without overnight stay	93 (17)
No secondary/tertiary care consultation	195 (36)
Numbers of consultations, median (IQR)*		0 (0)
Outpatient contact	2 (1–4)
Day surgery	1 (1–1)
Ordinary admission with overnight stay	1 (1–2)
Other admissions without overnight stay	1 (1–3)
Duration of ordinary admission with overnight stay in days, median (IQR)†	2 (1–4)
Productivity loss
Participants with productivity loss, n (%)		0 (0)
Sick leave	537 (98)
Work assessment allowance	146 (27)
Disability benefits	18 (3)
Duration of productivity loss in days, median (IQR)‡		0 (0)
Sick leave	47 (20–85)
Work assessment allowance	0 (0–8)
Disability benefits	0 (0–0)

*Calculated on basis of participants who have reported primary/secondary/tertiary care consultations.

†Calculated on basis of participants who have reported ordinary admission with overnight stay.

‡Calculated on basis of participants who have reported productivity loss, converted into a 5-day workweek and adjusted for employment rate and grading of productivity loss.

Table 3.

Cost (€) due to healthcare utilisation and productivity loss aggregated for 1 year of follow-up (n=549)

	Participants with zero cost, n (%)	Mean (95% CI*)	Median (95% CI*)
Primary care
General practitioner	11 (2)	719 (662 to 776)	539 (471 to 607)
Physiotherapist	283 (52)	680 (571 to 789)	0 (0 to 43)
Chiropractor	451 (82)	92 (64 to 120)	0 (0 to 0)
Emergency room	431 (79)	23 (18 to 28)	0 (0 to 0)
Total	7 (1)	1 514 (1374 to 1655)	948 (848 to 1048)
Secondary/tertiary care
Outpatient contact	216 (39)	410 (366 to 453)	220 (187 to 252)
Day surgery	501 (91)	182 (131 to 233)	0 (0 to 0)
Ordinary admission with overnight stay	489 (89)	1 160 (788 to 1533)	0 (0 to 0)
Other admissions without overnight stay	456 (83)	167 (105 to 228)	0 (0 to 0)
Total	195 (36)	1 919 (1513 to 2324)	295 (221 to 370)
Productivity loss
Sick leave	12 (2)	21 116 (19 534 to 22 697)	16 006 (14 239 to 17 773)
Work assessment allowance	403 (73)	5384 (4390 to 6379)	0 (0 to 0)
Disability benefits	531 (97)	652 (298 to 1005)	0 (0 to 0)
Total	1 (0.2)	27 152 (25 356 to 28 948)	21 846 (19 426 to 24 266)
Total	1 (0.2)	30 585 (28 585 to 32 585)	25 127 (22 451 to 27 802)

Cost related to productivity loss is calculated on the basis of reported days with productivity loss, converted into a 5-day workweek, and adjusted for employment rate and grading of productivity loss.

*Bias corrected and accelerated bootstrapping (1000 simulations).

Identification analysis

All continuous independent variables, aside from the variable ‘total costs related to healthcare utilisation prior to inclusion’, demonstrated a clear linear relationship with the outcomes. The variable ‘total cost related to healthcare utilisation prior to inclusion’ was found to be only close to a linear function. Nevertheless, as the variable was treated as a covariate exclusively in one of the secondary analyses, and for the sake of simplifying model interpretation, it was also incorporated as a linear component in the analyses. Table 4 shows crude and adjusted OR with 95% CI for the association between each of the modifiable prognostic factor and being in the three high costs groups. Seven out of nine modifiable prognostic factors showed a statistically significant crude association with the primary outcome ‘high societal costs’. After adjustment for selected covariates and with the ‘low-cost group’ as the reference, only six factors were found to be associated with increased odds of being in the high societal costs group; a higher degree of pain severity and disability, a lower degree of self-perceived health, sleep quality and return to work expectation, and a long-lasting disorder expectation. More or less similar results were found for the secondary outcomes. The sensitivity analysis (online supplemental table A1) showed only minimal changes in point and interval estimates when comparing complete case analysis to the main analysis.

Table 4.

Binary logistic regression analyses; individual associations between modifiable prognostic factors and high societal costs, high cost related to healthcare utilisation and high cost related to productivity loss

	High societal costs		High costs related to healthcare utilisation		High costs related to productivity loss
	Crude OR (95% CI)	Adjusted OR* (95% CI)	Crude OR (95% CI)	Adjusted OR* (95% CI)	Crude OR (95% CI)	Adjusted OR* (95% CI)
Pain severity (NRS, 0–10)	1.24 (1.12 to 1.38)	1.27 (1.13 to 1.42)	1.11 (1.01 to 1.23)	1.16 (1.04 to 1.29)	1.26 (1.13 to 1.39)	1.29 (1.15 to 1.45)
Self-perceived health (EQ-5D-5L, EQ VAS, 0–10)	1.31 (1.19 to 1.45)	1.27 (1.14 to 1.41)	1.18 (1.08 to 1.30)	1.16 (1.05 to 1.28)	1.33 (1.21 to 1.47)	1.33 (1.19 to 1.49)
Depressive symptoms (ÖMPSQ-SF, Q6, 0–10)	1.06 (0.99 to 1.13)	1.05 (0.98 to 1.12)	1.07 (0.99 to 1.14)	1.07 (0.99 to 1.14)	1.06 (0.99 to 1.13)	1.05 (0.98 to 1.13)
Sleep quality (ÖMPSQ-SF, Q4, 0–10)	1.14 (1.06 to 1.22)	1.14 (1.06 to 1.24)	1.06 (0.99 to 1.14)	1.06 (0.99 to 1.14)	1.15 (1.07 to 1.23)	1.17 (1.08 to 1.26)
Return to work expectancy (ÖMPSQ-SF, Q8, 0–10)	1.37 (1.29 to 1.47)	1.36 (1.27 to 1.46)	1.19 (1.12 to 1.26)	1.18 (1.11 to 1.25)	1.37 (1.29 to 1.47)	1.37 (1.27 to 1.47)
Work satisfaction (0–10)	1.09 (1.01 to 1.17)	1.06 (0.98 to 1.15)	1.04 (0.96 to 1.11)	1.03 (0.95 to 1.11)	1.08 (1.01 to 1.16)	1.07 (0.98 to 1.16)
Disability (EQ-5D-5L, Q3) (ref: no/slight problems)
Moderate problems	1.67 (1.03 to 2.72)	1.71 (1.01 to 2.89)	1.99 (1.22 to 3.26)	1.99 (1.19 to 3.32)	1.87 (1.15 to 3.05)	2.19 (1.27 to 3.78)
Severe problems/unable to do	3.27 (1.98 to 5.40)	3.03 (1.76 to 5.23)	3.31 (1.99 to 5.50)	2.91 (1.69 to 4.98)	3.19 (1.92 to 5.29)	3.75 (2.12 to 6.62)
Health literacy (MSK-HQ, Q12) (ref: completely/very well understanding)
Moderate understanding	0.94 (0.59 to 1.51)	1.01 (0.60 to 1.70)	1.14 (0.73 to 1.79)	1.28 (0.79 to 2.07)	1.06 (0.66 to 1.68)	1.21 (0.72 to 2.04)
Slightly/no understanding	1.47 (0.85 to 2.56)	1.64 (0.87 to 3.06)	0.72 (0.38 to 1.36)	0.85 (0.44 to 1.66)	1.52 (0.87 to 2.64)	1.76 (0.92 to 3.37)
Long-lasting disorder expectation (STarT MSK, Q6) (ref: no)	2.35 (1.29 to 4.29)	2.35 (1.23 to 4.47)	2.15 (1.19 to 3.86)	2.19 (1.15 to 4.19)	2.87 (1.52 to 5.43)	2.95 (1.48 to 5.89)

*Adjusted by sex, age, education level, absenteeism-related diagnosis type, pain duration and costs related to (1) healthcare utilisation and productivity loss prior to inclusion, (2) healthcare utilisation prior to inclusion or (3) productivity loss prior to inclusion.

EQ-5D-5L, EuroQol 5 dimensions; ÖMPSQ-SF, Örebro Musculoskeletal Pain Screening Questionnaire Short Form; MSK-HQ, Musculoskeletal Health Questionnaire; NRS, Numeric Rating Scale; Q, question number; STarT MSK, Keele STarT MSK tool.

Discussion

The current study evaluated the association between nine modifiable prognostic factors and high costs related to healthcare utilisation and productivity loss among people on sick leave due to musculoskeletal disorders. Six modifiable prognostic factors associated with high societal costs were identified: pain severity, disability, self-perceived health, sleep quality, return to work expectation and long-lasting disorder expectation. Depressive symptoms, work satisfaction and health literacy showed no prognostic value. More or less similar results were found when high costs were related to separately healthcare utilisation, and productivity loss. Five modifiable prognostic factors associated with both high costs related to healthcare utilisation and productivity loss were identified: pain severity, disability, self-perceived health, return to work expectation and long-lasting disorder expectation. These findings strengthen the assumption that these factors are important, since they impact on both costs related to healthcare utilisation and productivity loss.

To the best of our knowledge, no similar study has been conducted among a sample of exclusively people on sick leave due to musculoskeletal disorders, or within the Norwegian system. Thus, direct comparability of this study with other studies is limited. Nevertheless, our findings are generally in accordance with previous scientific literature on patients with back pain,14–16 20 21 27–31 as well as people with musculoskeletal disorders18 19 32–34 and non-diagnosis-specific studies.^{40 41 43} For example, pain severity, disability and self-perceived health have been shown to be significantly associated with high costs related to healthcare utilisation and productivity loss in studies on patients with back pain,14–16 20 21 27–31 and musculoskeletal disorders,18 19 32–34 and in non-diagnosis-specific studies.⁴³ Sleep quality has also previously been reported to be a prognostic factor of high costs related to healthcare utilisation.^{40 41} Moreover, in a synthesis of systematic reviews,⁴³ return to work expectation and long-lasting disorder expectation have been reported to be important prognostic factors in progressing return-to-work across various health conditions. Our finding regarding depression symptoms is, however, contrary to previous research,15 16 20 21 29 30 32 43 which found this factor to be associated with both healthcare utilisation and productivity loss. Several potential explanations for this discrepancy can be posited. First, we measured depression symptoms by a single item, whereas other studies primarily have used more comprehensive questionnaires. Second, we exclusively included participants on sick leave for at least 4 weeks. In a prior systematic review by Steenstra et al ⁴⁸ on prognostic factors for return to work, depression was not identified as an important factor in later phases of work disability (>6 weeks). To the best of our knowledge, the prognostic value of work satisfaction and health literacy for high costs related to healthcare utilisation and productivity loss have not been reported previously.

The main limitation of the current study is that it was powered on the primary outcomes of the cohort, not on the outcomes under consideration. Consequently, our ability to make adjustment for covariates was restricted. Thus, we decided to prioritise adjustment for a coreset of selected covariates, precluding additional adjustment for the other modifiable prognostic factors, which may have affected the observed associations. Moreover, we did not evaluate other potentially important prognostic factors of healthcare utilisation and productivity loss, such as health behaviours (smoking, physical activity, diet, etc). A second limitation is the fact that we expect to have somewhat underestimated the true value of both healthcare utilisation and productivity loss. We lack data on private healthcare utilisation. Moreover, we lack data on productivity loss related to reduced productivity while at paid work (presenteeism) and unpaid work. Yet, we consider the impact of this limitation to be of only minor importance in the current study, as healthcare utilisation and productivity loss were measured and valued equally for all participants and the outcome variables were dichotomised into high or low costs. Furthermore, it is likely to assume that costs of private health services are of less importance in a country as Norway, where health services are largely available and covered by the public sector. Nevertheless, the use of private health services is on the rise,⁶⁰ especially among people with higher income and education level.⁶¹ A third potential limitation is that the registers do not contain detailed information allowing to differentiate with certainty between the cause leading to healthcare utilisation. Hence, some of the included costs related to healthcare utilisation could pertain for reasons other than musculoskeletal disorders.

The main strength of the current study is that it was conducted in line with the PROGRESS framework,^{13 22} preplanned with a published statistical analysis plan²³ and reported in line with the REMARK guidelines.²⁴ Also, it estimates the prognostic value of modifiable prognostic factors over and beyond a core set of covariates. Furthermore, there was a low volume of missing data for the prognostic factor variables and no missing data on variables used to calculate the outcome scores. Finally, the risk of selection bias is assumed to be low, as previous research⁶² broadly confirms representativeness of the study sample.

Conclusion

In conclusion, we identified six modifiable prognostic factors associated with high societal costs among people on sick leave due to musculoskeletal disorders: pain severity, disability, self-perceived health, sleep quality, return to work expectation and long-lasting disorder expectation. Similar results were found when high costs were separately related to healthcare utilisation and productivity loss. The current study contributes to the ongoing research into clinical and welfare pathways and identified potential target areas for intervention which could reduce high societal costs among people on sick leave due to musculoskeletal disorders. Future research aimed at replicating these findings is warranted.

Data availability statement

Data supporting findings of this study are not publicly available as participants have not consented for their data to be publicly available. However, data are available from the corresponding author upon reasonable request and with permission of an ethics committee and the Oslo Metropolitan University (contact through corresponding author).

Ethics statements

Patient consent for publication

Not applicable.

Ethics approval

The study was assessed by the Norwegian Regional Committee for Medical Research Ethics and classified as a quality assessment study (ref number 2018/1326/REK sør-øst A). They specified that a quality assessment study does not require their explicit approval. The study was approved by the Norwegian Social Science Data Service (ref number 861249). Participants gave informed consent to participate in the study before taking part.