A longitudinal nationwide cohort study was conducted using data from Danish registers, specifically the Danish Civil Registration System (CRS) [15], and the National Patient Register (NPR) [16], along with a job exposure matrix (JEM) based on experts' ratings of occupational lower-body exposures from the DOC*X database [17, 18].

The cohort included individuals born between 1975 and 1978 (18–21 years of age in 1996). We chose a younger population cohort to mitigate healthy worker bias, since younger individuals are less likely to have experienced back disorders and, therefore, less likely to have migrated into sedentary occupations to avoid back-related concerns. Further, the individuals must have a valid annual job code according to DISCO-88 in 1996 to ensure they have entered the labour market. We excluded individuals from the cohort if (1) they had any hospital-diagnosed back disorder before December 31, 2005, or (2) if they died or emigrated between 1st January 1996 and 31st December 2017. The final cohort included 129,179 individuals. (The flowchart is shown in Fig. 1).

Individuals were followed from 1 January 2006, until the date of hospital-diagnosed back disorder, retirement, or censoring due to the end of the study by 31 December 2017 (whichever came first). In total, we observed 20,854 incidents (16%) of hospital-diagnosed back disorders during the period 2006–2017.

To assess the cumulative exposure of each cohort member, we calculated the cumulative PJEs over a 10-year look-back period (2006–2017) with a 1-year lag at each year. This approach allowed us to establish a long-term perspective on PJE in the hopes of better understanding the relationship between accumulated physical workload, back disorders, and the influence of the healthy worker effect. Further details are provided in the section below.

Information regarding individuals' year-by-year occupational history, specifically the DISCO-88 codes spanning from 1996 to 2017, was obtained from the DOC*X database [18]. In cases where DISCO-88 codes were missing, i.e. when individuals were unemployed or job codes were unknown, zero exposure was assigned. DISCO-88 codes were subsequently converted into PJE estimates utilizing the lower-body JEM [17].

The lower-body JEM encompasses ratings of various daily PJEs, done by occupational medicine specialists. In this study, we used the total load lifted in kilograms (kg) (Total Load), the stand/sit ratio (Stand/Sit ratio), and number of times lifting more than 20 kg per day (Times > 20) as exposure measures. Individuals had to be exposed to all three exposure measures to be counted as exposed. Notably, the lower-body JEM has previously demonstrated predictive validity for multiple outcomes, e.g. risk of total hip replacement and risk of acute myocardial infarction [19, 20], but has never been used to assess the risk of back disorders. The JEM is based on the complete set of currently utilized job titles in the Danish version of the International Standard Classification of Occupation (DISCO-88) on one axis, and ratings of specific lower-body exposures on the other [18]. In Denmark, occupational medicine specialists possess expertise in quantifying the physical exposures during a typical workday across various occupations, as their detailed documentation forms the basis for compensation regarding back disorders [20].

The outcome of interest in this study was incident hospital-diagnosed back disorders, which were defined by a hospital admission with an ICD-10 code specifically related to back disorders as outlined in the Danish Spine Register (DaRD) [6]. Information pertaining to the specific type and date of the hospital diagnosis was obtained from the NPR. The following primary diagnostic codes were included: DM42*, DM43*, DM47*, DM48*, DM495, DM50*, DM51*, DM53*, DM54*, DM809C, DM96*, DM99*, and DS13*.

Several confounder variables were accounted for, including sex, age, calendar year, higher education, and region of residence. The inclusion of region of residence aimed to capture regional variations in diagnosis rates, considering that regions are responsible for the secondary sector in Denmark and to capture regional variation in exposure.

We employed a logistic regression, specified as a discrete survival analysis, to calculate the cumulative risks for incident hospital-diagnosed back disorder [21]. The risks are measured as odds ratios (OR), which can be interpreted as a hazard ratio. The statistical unit in this approach was person-years. Cumulative exposures measured using a 10-year look-back window for each follow-up year (2006–2017) were utilized. In the adjusted models, we controlled for sex, age, higher education, region of residence (five categories), and year. Error terms were clustered at the individual level.

We did a range of supplementary analyses to assess the magnitude and impact of the HWE. First, we illustrated the magnitude of the HWE in our sample by showing the number of healthy survivors in PJE occupations over the period 2006–2017 for all individuals exposed in year 2006. Second, we compared the estimated cumulated risks with the naive cross section estimate from year 2006. Third, we ran the adjusted regressions year by year throughout the entire study period (2006–2017).

Individuals enrolled as students during the study period, but also holding part-time or full-time employment, were assigned a PJE based on their student occupation. By nature, most of these occupations will be part-time and the PJE might be limited, so to test the robustness to this uncertainty we attributed an exposure value of 0 in these cases, despite knowing their DISCO code. A further rationale for this adjustment was grounded in the expectation of minimal exposure among students.

The analyses were performed using Stata v.18 on Statistics Denmark's research platform. STROBE guidelines were employed.