Impact of the Serious Accidents Punishment Act on Occupational Injury Rates in Korea’s Manufacturing Sector: An Interrupted Time-Series Analysis

Abstract

Background

The Republic of Korea introduced the Serious Accidents Punishment Act (SAPA) in 2021 to improve workplace safety by holding company leaders legally accountable for serious accidents but its impact on injury trends has not been well established. This study evaluated SAPA’s effect on age–sex-standardized injury incidence rates (ASRs) in the Republic of Korea’s manufacturing sector, examining differences by company size (≥50 vs. <50 employees), sex, and injury type (inpatient vs. outpatient).

Methods

Using National Health Insurance Service Cohort data (January 2011–April 2024), we employed interrupted time-series design to assess SAPA’s effects from January 2022. Injury data (International Statistical Classification of Diseases and Related Health Problems, 10^th Edition: S00–T88) were standardized using the 2015 Republic of Korea Statistical Information Service population. Generalized least squares regression analysis with an autoregressive moving average structure was performed to analyze ASR trends across subgroups.

Results

Following SAPA implementation, nonsignificant decreases in both inpatient and outpatient ASR were observed among men in companies with 50 or more employees. Among women, inpatient ASR showed opposite patterns according to company size; however, after Benjamini–Hochberg adjustment for multiple comparisons, these effects did not retain significance in the full 2011–2024 dataset. A sensitivity analysis restricted to 2017–2024 suggested that the increase in <50-employee companies may represent a robust signal in the more recent period.

Conclusion

SAPA implementation was not associated with statistically significant changes in injury rates in most subgroups after multiple testing adjustment, consistent with the early post-implementation experience of analogous international legislation. Exploratory subgroup analyses suggested differential responses by sex and company size, necessitating longitudinal research to evaluate the law’s long-term preventive effects.

1. Introduction

Occupational injuries and fatalities remain significant public health and economic concerns worldwide. The Republic of Korea has consistently reported relatively high occupational injury and fatality rates among countries that are members of the Organization for Economic Co-operation and Development [1,2], reflecting both labor-intensive industrial structures and persistent deficiencies in workplace safety management. In 2015, industrial accidents in the Republic of Korea resulted in healthcare costs of USD 5.9 billion and total economic losses of USD 19.5 billion, with the manufacturing sector accounting for a substantial proportion of these losses because of its reliance on hazardous machinery and labor-intensive processes [3,4]. To address these problems and prevent repeated large-scale disasters, such as the 2020 warehouse fire that killed 38 workers, echoing a similar incident in 2008 [4,5], the South Korean government enacted “The Serious Accidents Punishment Act (SAPA)” in January 2021 [6]. The Act was subsequently enforced in businesses with 50 or more employees, starting in January 2022. Modeled on the UK’s Corporate Manslaughter and Corporate Homicide Act, the SAPA holds proprietors, CEOs, and corporations accountable for fatal accidents caused by inadequate safety protocols [7].

The global experience with corporate liability laws provides an important context for evaluating SAPA. For example, following the enactment of the UK Corporate Manslaughter and Corporate Homicide Act 2007, researchers observed a marked elevation in workplace safety as a board-level priority. However, the immediate quantitative effects on injury rates were heterogeneous, with some sectors showing gradual improvement and others showing no change [8]. Similarly, industrial manslaughter laws in Australia were intended to internalize the costs of safety failures; however, empirical evidence suggests that the deterrent effect of such legislation often requires a prolonged period to translate into measurable declines in injury rates, depending on the rigor of enforcement and the pre-existing safety culture within industries [9]. Against this international backdrop, this study provides a timely empirical evaluation of the early effects of SAPA in the Republic of Korea’s manufacturing sector.

As the implementation of the SAPA is relatively recent, there is a lack of precise analysis on whether the SAPA has influenced industrial accident rates. Previous studies on the SAPA explained the provision or background of the legislation [7,10], leaving a gap in understanding its effects on company safety. In addition, there is limited knowledge of how the SAPA has impacted specific subgroups, particularly across company size, sex, and injury type (inpatient or outpatient), which may respond differently to legislation. Although the SAPA was initially enforced only in businesses with 50 or more employees, smaller companies constitute a substantial share of the labor market. Therefore, it is important to examine whether its effects extend to them as well. Furthermore, sex-related differences in occupational roles and exposures may contribute to the heterogeneous effects of the SAPA.

Therefore, we aimed to analyze the effect of the SAPA on industrial accident rates in the Republic of Korea. By employing interrupted time-series (ITS) regression analysis, we examined the impact of the SAPA, focusing on subgroups differentiated by company size, sex, and injury type (inpatient vs. outpatient), to identify vulnerable groups that may require additional protective measures.

2. Materials and methods

2.1. Study design and setting

The Republic of Korea implemented the SAPA on January 27, 2022, to enhance occupational safety by imposing criminal penalties on business owners and managers for serious workplace accidents. Initially, the legislation was applied to workplaces with 50 or more employees, with an extension to workplaces with five or more employees being effective from January 27, 2024. In this ITS analysis, we examined the impact of the SAPA on injury incidence rates among manufacturing workers focusing on differences by company size, sex, and injury type (inpatient or outpatient).

2.2. Data

Monthly data from January 2011 to April 2024 were obtained from the National Health Insurance Service (NHIS) cohort database. The NHIS collects data from the national healthcare insurance system, which covers approximately 98% of the Republic of Korea, and centralizes it into large databases. The NHIS database provides comprehensive records, including health screening data, medical utilization claims, sociodemographic characteristics, and mortality information [11]. Injury data from January 2011 to April 2024 were recorded and coded according to the injury-related International Statistical Classification of Diseases and Related Health Problems, 10^th Edition (ICD-10), based on medical claims. Standard population data, including age-specific population counts for men and women, were obtained for age–sex standardization from the 2015 Republic of Korea Statistical Information Service census.

2.3. Outcomes

The primary outcome was the age–sex-standardized injury incidence rate (ASR), calculated as the number of injuries per 100,000 workers. Injury rates for inpatient and outpatient visits were defined using ICD-10 codes S00–T88 as the primary outcomes [12]. The ASR was derived by direct standardization using the 2015 Republic of Korea Statistical Information Service standard population. For the overall cohort, we calculated the age- and sex-standardized injury rates; however, for sex-stratified analyses, we calculated the age-standardized injury rate. For consistency, both measures are referred to as ASR. Outcomes were stratified by company size group (≥50 or <50 employees), sex, and injury type (inpatient or outpatient). Inpatient and outpatient cases were analyzed separately as a proxy for injury severity, with inpatient admissions representing more severe incidents requiring sustained medical care. Although company size was originally categorized into four groups (<5, 5–49, 50–299, and ≥300 employees) based on the Labor Standards Act, we reclassified it into two categories: ≥50 employees and <50 employees, reflecting the SAPA’s initial applicability to companies with ≥50 employees (subject to SAPA in 2022).

2.4. Statistical analysis

We utilized an ITS design to evaluate changes in injury incidence rates following the implementation of the SAPA on January 27, 2022. The ITS analysis was performed to assess both immediate level changes and post-intervention slope changes on ASR in companies with ≥50 employees compared with those in companies with <50 employees. The immediate level change represents the instantaneous shift in ASR at the intervention point (January 2022) and was estimated using coefficient β2 in the ITS regression model. Data were analyzed separately for four subgroups: male inpatients, male outpatients, female inpatients, and female outpatients. Prior to ITS modeling, monthly ASR trends were plotted using linear regression models (polynomial degrees 1 and 4) to identify secular trends. The pre- and post-intervention monthly averages of injury cases and ASR are summarized in Table 1, stratified by company size, sex, and intervention period. In addition, monthly descriptive statistics, including mean injury case counts, standard deviations, and ASRs, for the 24 months immediately before and after SAPA implementation are summarized in Supplementary Table S2. The monthly ASR values are presented in Supplementary Table S2b to show the raw trends around the intervention point.

Table 1.

Estimated immediate level change and post-intervention slope change of the SAPA on age-standardized injury rates by company size, sex, and injury type.

Sex	Injury type	Company size	Immediate level change (95% CI)	p	Post-intervention slope change (95% CI)	p
Men	Inpatient	≥50 employees	−10.24 (−24.97 to 4.48)	0.175	0.58 (−0.12 to 1.29)	0.105
		<50 employees	−8.45 (−21.08 to 4.18)	0.191	0.09 (−0.49 to 0.66)	0.768
	Outpatient	≥50 employees	−125.89 (−554.21 to 302.43)	0.565	9.01 (−12.47 to 30.49)	0.412
		<50 employees	−305.08 (−804.38 to 194.22)	0.233	−7.57 (−32.33 to 17.19)	0.550
Women	Inpatient	≥50 employees	−12.79 (−24.99 to −0.58)	0.042	0.55 (0.00 to 1.11)	0.051
		<50 employees	28.93 (6.21 to 51.65)	0.014	−0.24 (−1.26 to 0.79)	0.654
	Outpatient	≥50 employees	−141.42 (−509.28 to 226.44)	0.452	11.31 (−5.81 to 28.43)	0.197
		<50 employees	−163.50 (−517.30 to 190.31)	0.366	10.32 (−6.02 to 26.67)	0.217

CI, confidence interval; SAPA, Serious Accidents Punishment Act.

For ITS modeling, a generalized least squares (GLS) regression was fitted with a maximum likelihood estimation, incorporating an autoregressive moving average (ARMA) structure to account for temporal autocorrelation. The actual trend model is specified as:

$$\text{Yt}=\mathrm{\beta }0+\mathrm{\beta }1·\text{time}+\mathrm{\beta }2·\text{intervention}+\mathrm{\beta }3·\text{post}\_\text{intervention}\_\text{time}+\text{Covariates}+ϵ$$

• I.Yt: ASR at time (t).
• II.Time: Continuous variable representing monthly time increments (e.g., 1, 2, 3)
• III.Intervention: Dummy variable indicating pre-SAPA (0) or post-SAPA [1].
• IV.Post_intervention_time: A continuous variable representing the time elapsed since the SAPA was introduced (0 before intervention and time since intervention). Coefficient β3 represents the change in slope after the intervention, that is, the change in the monthly trend during the post-intervention period.
• V.Covariates: Company size and its interactions with time, intervention, and post-intervention times to model baseline levels, trends, immediate changes, and long-term trend differences between the groups.
• VI.ε: Error term, with autocorrelation modeled using an ARMA structure. The ARMA structure for each subgroup model was selected through a systematic grid search across ARMA($p,q$) combinations ($p=0$ to 4, $q=0$ to 4), with model convergence, Akaike Information Criterion, and residual autocorrelation assessed using the Ljung–Box test. Because company size groups were modeled jointly within a single GLS framework using interaction terms, a more conservative higher-order ARMA structure was applied uniformly within each sex-injury type combination. Accordingly, an ARMA(4,4) structure (corARMA(p = 4, q = 4, form = ∼ time | comsizegp)) was applied to the male inpatient, female inpatient, and female outpatient models. For the male outpatient model, ARMA(4,4) failed to converge; therefore, ARMA(3,2) (corARMA(p = 3, q = 2, form = ∼ time | comsizegp)) was used as the highest-order convergent structure.

To address the risk of Type I error arising from tests across eight subgroups, p values were adjusted using the Benjamini–Hochberg method. The group-specific effect estimates and adjusted p values are presented in Table 2. For the group-specific effect estimates shown in Table 2, each company-size group was modeled independently using the same target ARMA structure. In this stratified analysis, ARMA(4,4) failed to converge for the female inpatient model in companies with fewer than 50 employees; therefore, ARMA(3,2), which yielded the lowest Akaike Information Criterion among the convergent structures, was used instead. To evaluate the adequacy of the ARMA structure, we examined residual autocorrelation using the Ljung–Box test and assessed normality using the Shapiro–Wilk test (see Supplementary Figure S1 and Table S4).

Table 2.

Estimated immediate-level change and post-intervention slope change of the SAPA on age-standardized injury rates by company size, sex, and injury type, with Benjamini–Hochberg adjusted p values.

Sex	Injury type	Company size	ARMA model	Immediate level change (95% CI)	p	Adjusted p∗	Post-intervention slope change (95% CI)	p	Adjusted p∗
Men	Inpatient	≥50 employees	ARMA(4,4)	−10.24 (−24.97 to 4.48)	0.175	0.372	0.58 (−0.12 to 1.29)	0.105	0.420
		<50 employees	ARMA(4,4)	−8.45 (−21.08 to 4.18)	0.191	0.372	0.09 (−0.49 to 0.66)	0.768	0.768
	Outpatient	≥50 employees	ARMA(3,2)	−125.89 (−554.21 to 302.43)	0.565	0.565	9.01 (−12.47 to 30.49)	0.412	0.659
		<50 employees	ARMA(3,2)	−305.08 (−804.38 to 194.22)	0.233	0.372	−7.57 (−32.33 to 17.19)	0.550	0.733
Women	Inpatient	≥50 employees	ARMA(4,4)	−12.79 (−24.99 to −0.58)	0.042	0.167	0.55 (0.00 to 1.11)	0.051	0.409
		<50 employees	ARMA(3,2)	28.93 (6.21 to 51.65)	0.014	0.109	−0.24 (−1.26 to 0.79)	0.654	0.747
	Outpatient	≥50 employees	ARMA(4,4)	−141.42 (−509.28 to 226.44)	0.452	0.517	11.31 (−5.81 to 28.43)	0.197	0.435
		<50 employees	ARMA(4,4)	−163.50 (−517.30 to 190.31)	0.366	0.489	10.32 (−6.02 to 26.67)	0.217	0.435

Bold p values indicate statistical significance (p < 0.05) before adjustment.

ARMA, autoregressive moving average; CI, confidence interval; SAPA, Serious Accidents Punishment Act.

^∗Adjusted using the Benjamini–Hochberg method for multiple comparisons across eight subgroups.

To estimate counterfactual trends, we fitted a second GLS model using only pre-intervention data, including company size, time, and their interactions. This model extrapolates pre-intervention trends to predict post-intervention outcomes in the absence of the SAPA. Predicted values and standard errors were obtained for both the factual and counterfactual models.

Sensitivity analyses were performed to assess the robustness of the findings. First, we set the intervention point to February 2021, 1 year prior to the SAPA implementation, to explore the potential anticipatory effects of the law’s enactment and formalization in 2021. Second, we varied the data starting point to January 2012 to evaluate model stability across different time frames. The results were visualized using time-series plots comparing factual and counterfactual trends, with vertical lines indicating the intervention points (February 2022 or February 2021).

All analyses were conducted using the R software (version 4.3.2). A p value cutoff of less than 0.05 was used to determine statistical significance for all outcomes. Effect estimates, 95% confidence intervals (CIs), and p values were calculated.

2.5. Ethics approval

The study protocol was approved by the institutional review board (IRB) of our university (IRB No. 4-2024-1569). The requirements for informed consent were waived due to the retrospective nature of the study and the use of anonymized data.

3. Results

The SAPA implementation in January 2022 was associated with heterogeneous patterns in monthly injury cases and ASR across company size, sex, and injury type. Supplementary Table S1 demonstrates that following the SAPA implementation, inpatient and outpatient injury cases and ASR for men and women exhibited subgroup-specific trends, with a decrease in inpatient cases and an increase in outpatient cases for both sexes in ≥50-employee companies.

The annual trends in ASR by company size (≥50 vs. <50 employees) and sex are shown in Fig. 1. Trends indicate that the male outpatient ASR in <50-employee companies decreased distinctly compared with that in ≥50-employee companies, while other subgroups showed minimal differences in trends between ≥50- and <50-employee companies. The annual trends in ASR varied according to the original four groups of company size and sex (Supplementary Fig. S1). For men, annual trends indicated that the inpatient ASR was generally higher in <5-employee companies than in other companies. Outpatient ASR increased only in ≥300-employee companies, while it decreased across other company sizes. For women, inpatient and outpatient ASR exhibited similar patterns across all company sizes.

Fig. 1.

Annual age-standardized (for sex-stratified analyses) injury rate (ASR) trends of (A) inpatient and (B) outpatient injuries by company size and sex.

Table 1 presents the estimated immediate level changes and post-intervention slope changes associated with SAPA for the ASR. Among men in ≥50-employee companies, the immediate level change showed nonsignificant decreases in both inpatient ASR (−10.24; 95% CI: −24.97, 4.48) and outpatient ASR (−125.89; 95% CI: −554.21, 302.43). Among women in ≥50-employee companies, a statistically significant decrease was observed in inpatient ASR (−12.79; 95% CI: −24.99, −0.58; p = 0.042), whereas outpatient ASR showed a nonsignificant decrease (−141.42; 95% CI: −509.28, 226.44). Among women in <50-employee companies, the immediate level change in inpatient ASR showed a statistically significant increase of 28.93 per 100,000 workers (95% CI: 6.21, 51.65; p = 0.014). These were the only statistically significant findings among the eight subgroups.

ITS analysis assessed the immediate level and post-intervention slope changes of SAPA on ASR across company sizes (Fig. 2). For men, the inpatient and outpatient ASR showed an initial decrease after February 2022 in both ≥50- and <50-employee companies, with ≥50-employee companies subsequently indicating an increase (Fig. 2a and b). For women, the inpatient ASR showed an initial increase in <50-employee companies, followed by a gradual decline, whereas in ≥50-employee companies, it showed an initial decrease and then gradually increased (Fig. 2c). Outpatient ASR showed an initial decrease in both ≥50- and <50-employee companies, followed by a gradual increase (Fig. 2d). All factual and counterfactual trends differed, suggesting interventional effects.

Fig. 2.

Interrupted time-series analysis with company size. (A) Male inpatients. (B) Male outpatients. (C) Female inpatients. (D) Female outpatients.

Sensitivity analysis with a 2012 starting point showed a stable ASR after February 2022 for male and female inpatients and outpatients, which is consistent with the original data (Supplementary Fig. S2). Additionally, a sensitivity analysis with an intervention point set to February 2021 revealed a stable ASR for male and female inpatients and outpatients, with a larger post-intervention slope change size for male inpatients than that in the original analysis (Supplementary Fig. S3).

To ensure robustness against long-term structural changes such as the coronavirus disease 2019 pandemic, an additional sensitivity analysis using data from January 2017 was conducted. In this restricted model, the immediate level change increase in female inpatient ASR within <50-employee companies remained statistically significant, even after Benjamini–Hochberg adjustment (p < 0.001; Table 2).

4. Discussion

This study evaluates changes in ASR in the Republic of Korea’s manufacturing sector following SAPA implementation, with a focus on differences by company size (≥50 vs. <50 employees), sex, and injury type (inpatient vs. outpatient). Using ITS analysis, we assessed two parameters: the immediate level change, representing any abrupt shift in ASR at the point of SAPA implementation, and the post-intervention slope change, representing the longer-term trajectory of ASR after implementation.

The study findings indicate that the SAPA implementation in January 2022 was associated with predominantly nonsignificant changes in monthly injury cases and ASR across company sizes, sex, and injury type. ITS analysis revealed immediate level changes and post-intervention slope changes, although not statistically significant, with a post–February 2022 decrease in inpatient and outpatient ASR among men across both ≥50- and <50-employee companies, followed by an increase in ≥50-employee companies. In contrast, the inpatient ASR among women showed an initial increase in <50-employee companies with a subsequent gradual decline, while in ≥50-employee companies, it showed an initial decrease followed by a gradual increase, and the outpatient ASR showed an initial decrease, followed by a gradual increase in both groups. These differential effects were particularly evident in ≥50-employee companies. Nonsignificant immediate level decreases were observed for male inpatient ASR (−10.24 per 100,000 persons) and male outpatient ASR (−125.89). Among women in ≥50-employee companies, a statistically significant decrease in inpatient ASR was observed (−12.79; p = 0.042), whereas a statistically significant increase in inpatient ASR was observed among women in <50-employee companies (28.93; p = 0.014). However, both findings should be interpreted with caution because of the risk of a Type I error. After Benjamini–Hochberg adjustment, neither effect remained statistically significant in the full 2011–2024 dataset (adjusted p = 0.167 and 0.109, respectively; Table 2). Nevertheless, the persistence of statistical significance in the model restricted to 2017–2024 for women in <50-employee companies (p < 0.001) suggests that this shift may represent a robust signal in the recent regulatory and economic context (Supplementary Table S3).

Our findings demonstrate sex differences in the ASR. Although our dataset did not include individual-level data on job assignments, the literature offers plausible contextual explanations for the observed differences. In the Republic of Korea, manufacturing (27.7%) is predominantly male dominated, whereas human health and social work activities (12.4%) are characterized by a higher proportion of female workers [13]. Analyses of the U.S. Current Population Survey and Bureau of Labor Statistics data have further indicated that men are more likely to engage in high-risk occupations, whereas women tend to prefer safer roles [14,15]. Even within the same occupational category, men and women can be assigned different tasks, resulting in different levels of risk exposure. Previous studies have reported that sex-related differences in work-related musculoskeletal disorders arise from variations in workforce composition and task assignments within sectors [16]. Similarly, a study conducted in New Zealand [17] found that men are more frequently assigned to high-risk tasks, such as operating vibrating machinery, performing shift work, and working night shifts, even within the same occupational group, suggesting a tendency for men to undertake more hazardous duties. Collectively, these findings suggest that sex differences in task allocation and occupational selection in industrial settings contribute to men’s greater exposure to risky work environments, where a more pronounced reduction in ASR due to SAPA may be expected, warranting further investigation.

The separation of inpatient and outpatient cases in this study was intended to approximate injury severity, with inpatient admissions serving as a proxy for more severe incidents. In the inpatient analysis, immediate level decreases were observed in most subgroups following SAPA implementation, with estimated reductions of 10.24 and 12.79 per 100,000 persons for men and women in ≥50-employee companies, respectively. Notably, the decrease among women in ≥50-employee companies reached statistical significance (p = 0.042), suggesting that SAPA may have contributed to a reduction in more severe injuries in companies directly subject to the legislation. The direction of the estimates was consistent across sex and company size categories, with the notable exception of women in <50-employee companies. The subsequent positive post-intervention slope changes in inpatient ASR, although also nonsignificant, may suggest that the initial deterrent effect of the legislation attenuated over time, possibly reflecting the limited frequency of actual prosecutions and the predominance of suspended sentences during this early enforcement period [18]. In contrast, the outpatient analysis revealed larger absolute changes in ASR but was accompanied by considerably wider CIs, indicating greater variability and lower precision in these estimates. This pattern is consistent with the expectation that minor injuries captured through outpatient visits are influenced by a broader range of factors beyond workplace safety regulations, such as healthcare-seeking behaviors and changes in medical service accessibility. Given that SAPA specifically targets serious industrial accidents resulting in death or significant bodily harm, the inpatient findings—despite largely nonsignificant results across most subgroups—may be more conceptually aligned with the legislation’s intended scope. However, these observations should be interpreted cautiously as the broad ICD-10 coding used in this study does not distinguish between occupational and non-occupational injuries, and the nonsignificant results preclude firm conclusions regarding differential effects by injury severity.

ITS analysis demonstrated distinct ASR trends between ≥50- and <50-employee companies, despite SAPA’s initial legal sanctions being applicable only to ≥50-employee companies. An initial decline in male inpatient and outpatient ASR post–February 2022 across both groups, followed by an increase in ≥50-employee companies, suggests that SAPA’s effects extend to <50-employee companies, potentially driven by voluntary safety enhancements or indirect regulatory pressure from larger firms’ supply chains. According to previous studies, penalties imposed on specific firms can influence broader industry practices, thus supporting the spillover effect [19,20]. The subsequent increase in the male inpatient ASR by 0.58 per 100,000 persons and outpatient ASR by 9.01 per 100,000 persons in ≥50-employee companies may reflect heightened reporting or incident rates following initial compliance efforts, indicative of the SAPA effects. Among women, the opposite directions of immediate-level changes between ≥50-employee companies (decrease) and <50-employee companies (increase) may reflect differences in access to safety resources and regulatory preparedness, although both findings should be interpreted cautiously because their statistical significance was attenuated after adjustment for multiple testing. Conversely, the stability observed in the <50-employee companies could be attributed to resource constraints or delayed SAPA applicability, scheduled for 2024. Further longitudinal analysis is needed to assess the changes in <50-employee companies following the full implementation of the policy.

To contextualize these findings, it is important to consider the legal and institutional framework of SAPA and its emerging enforcement patterns. In the Republic of Korea, the first conviction under SAPA was not recorded until April 2023, which is more than 14 months after the enforcement began. Moreover, among the 15 Chief Executive Officer (CEO) convictions recorded through 2024, 14 resulted in suspended prison sentences [18]. This infrequency of prosecutions and predominance of non-custodial outcomes may have been insufficient to produce the immediate and broad deterrent signal needed to generate significant aggregate-level shifts in injury rates. Furthermore, the effectiveness of SAPA depends heavily on judicial interpretations of “safety and health management systems” and the degree of “due diligence” required of business owners. Recent Republic ofKorean court rulings have begun to clarify these obligations by emphasizing that formalistic compliance is insufficient; instead, courts require substantive evidence of active risk identification and the allocation of adequate safety budgets [21]. For example, the first SAPA ruling on occupational disease, issued by the Changwon District Court in 2023, provided initial guidance by identifying standards for hazard remediation; however, the broader legal landscape remains unsettled [22]. This emerging pattern aligns with international experiences. Similar to the UK Corporate Manslaughter and Corporate Homicide Act 2007, which primarily influenced corporate governance rather than producing immediate improvements in frontline safety outcomes, the SAPA appears to have initiated a period of regulatory adjustment [8]. Similarly, as observed with industrial manslaughter laws in Australia, the deterrent effect of such legislation often requires a prolonged period to translate into measurable declines in injury rates, depending on the rigor of enforcement and pre-existing safety cultures [9]. Consequently, this legal uncertainty and the gradual pace of judicial enforcement may leave employers unsure of how to satisfy the legal threshold, which helps explain the predominantly null findings in our study.

The sensitivity analysis from the 2012 start-point results was consistent with those from 2011, suggesting the absence of events unrelated to the SAPA prior to its implementation. Sensitivity analysis with an intervention point set to February 2021 revealed a larger post-intervention slope change for male inpatients than that in the original analysis, indicating potential pre-implementation adjustments, such as voluntary safety improvements. This proactive response supports the existing literature, which suggests that organizations tend to prepare for regulatory changes even before formal enforcement [23]. According to Hochberg et al. [24], banks proactively adjust their internal capital structures and risk-management frameworks (pre-adjustment) following the proposal of Basel III regulations, even before formal implementation, with responsive actions observed immediately after enactment. The study further suggests that, given the often-prolonged rule-making process involving proposal, development, and finalization prior to regulatory introduction, the appropriate timing for analysis may be the announcement date rather than the adoption or enforcement date. This implies the need for additional analyses to identify the optimal time points specific to individual workplaces.

This study had some limitations that may have influenced the interpretation of our findings. First, reliance on NHIS data may underestimate the incidence of minor cases that are not formally reported or treated, which could potentially distort ASR trends. A related fundamental limitation is that NHIS claims identified using ICD-10 codes S00–T88 include non-occupational injuries, such as domestic or traffic-related injuries. This mismatch with SAPA’s focus on serious industrial accidents may introduce noise and bias the estimates toward the null. Nevertheless, because our study population was restricted to workers in the manufacturing sector, a substantial proportion of injuries captured in the NHIS claims database were likely work-related. In addition, workers’ compensation data in the Republic of Korea are known to substantially underreport occupational injuries because of informal settlements and employer pressure to avoid official claims, making NHIS data a more inclusive, although less specific, alternative. Importantly, the inclusion of non-occupational injuries is expected to produce nondifferential misclassification, thereby biasing the estimates toward the null. Accordingly, any statistically significant findings in this study may represent conservative estimates of the true effects.

Second, the risk of Type I errors due to multiple hypothesis testing should be considered. Although the statistical significance of changes in certain subgroups was attenuated after Benjamini–Hochberg adjustment in the full dataset, the sensitivity analysis restricted to recent data (2017–2024) suggests that some observed effects may be robust (p < 0.001). Finally, the absence of adjustments for external factors, such as economic conditions, shifts in the composition of the manufacturing industry, and changes in medical coding practices, may have introduced bias. In addition, the use of ARMA(3,2) for the male outpatient model, owing to the convergence failure of the higher-order ARMA(4,4) structure, may also have affected the estimates. These issues underscore the need for cautious interpretation of the results. Although the ITS design with a linear pre-intervention trend partially accounts for gradual secular changes, abrupt structural shifts, such as changes in healthcare-seeking behavior during the coronavirus disease 2019 pandemic, may still confound the estimated effects. Moreover, despite systematic optimization of the ARMA structure through grid search, residual autocorrelation persisted in several subgroup models (Ljung–Box p < 0.05), which may have resulted in underestimated standard errors and, consequently, an increased risk of Type I error. This concern is consistent with our use of the Benjamini–Hochberg adjustment for multiple comparisons, which further reduces the likelihood of false-positive findings.

To the best of our knowledge, this study represents the first analysis of the effects of the SAPA in the Republic of Korea, marking a pioneering effort to assess its occupational safety implications. This study has several strengths that contribute to the field of occupational health research. The robust ITS design, which incorporates eight subgroup analyses by sex, injury type, and company size, provides a detailed evaluation of SAPA’s differential impact. The comprehensive dataset spanning January 2011 to April 2024 captured long-term trends and pre-implementation patterns, thereby enhancing the temporal scope. Rigorous sensitivity analyses, including starting points from 2012 and intervention timings set to 2021, were also conducted to test the ASR trend consistency across diverse scenarios and bolster the reliability of the results.

Based on our findings, three priority policy implications emerge. First, given the nonsignificant findings during the early enforcement period, sustained long-term monitoring is essential to evaluate whether SAPA’s deterrent effects materialize as enforcement matures and judicial precedents accumulate. Second, the extension of SAPA to workplaces with fewer than 50 employees in January 2024, combined with the exploratory signal of increasing inpatient ASR among women in this group, underscores the need for targeted compliance support and safety resource allocation for smaller enterprises. Finally, the observed spillover patterns suggest that policymakers should leverage larger corporations as conduits for safety standard dissemination throughout their supply chains.

In conclusion, this study demonstrated that the SAPA implementation in January 2022 was not significantly associated with changes in injury cases and ASR across most subgroups defined by company size, sex, and injury type. Although exploratory subgroup analyses identified differential patterns by sex and company size, these effects were no longer statistically significant after adjusting for multiple tests in the full dataset, underscoring the need for cautious interpretation. Future research using workers’ compensation data and longer follow-up periods is needed to fully evaluate the preventive effects of SAPA.

CRediT authorship contribution statement

Jian Lee: Writing – review & editing, Writing – original draft, Visualization, Methodology, Investigation, Formal analysis, Data curation, Conceptualization. Juyeon Oh: Resources, Methodology, Data curation. Heejoo Park: Writing – review & editing, Software. Jongmin Lee: Methodology. Byungyoon Yun: Writing – review & editing, Validation, Methodology, Investigation, Formal analysis, Conceptualization. Jin-Ha Yoon: Writing – review & editing, Validation, Supervision, Project administration, Funding acquisition.

Statement on the use of AI tools

During the preparation of this manuscript, the authors used Grammarly (version 1.2.43, accessed November 2025) solely for grammar, spelling, and language clarity checks on the final draft. No generative AI or AI-assisted tools were used for conception, study design, data analysis, interpretation of results, or drafting of the scientific content.

All intellectual contributions were exclusively performed by the listed authors. The authors reviewed and edited all AI-suggested changes and take full responsibility for the content of the publication.

Conflicts of interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Appendix A. Supplementary data

The following are the Supplementary data to this article:

figs1.

Supplementary Figure S1. Residual diagnostic plots for the ARMA models by subgroup. Each panel shows the autocorrelation function (ACF) of the residuals (top left), a histogram of the residuals with an overlaid normal density curve (top right), a normal Q-Q plot (bottom left), and a plot of residuals versus fitted values (bottom right). (A) Men, inpatient, ≥50 employees. (B) Men, inpatient, <50 employees. (C) Men, outpatient, ≥50 employees. (D) Men, outpatient, <50 employees. (E) Women, inpatient, ≥50 employees. (F) Women, inpatient, <50 employees. (G) Women, outpatient, ≥50 employees. (H) Women, outpatient, <50 employees. The blue dashed lines in the ACF plots indicate the 95% confidence bounds.

figs2.

figs3.

figs4.