This study assessed the effects of case volume and specialization on the outcomes of critically ill emergency patients by using a comprehensive ICU patient database. The results revealed that higher ICU case volumes were associated with lower in-hospital mortality rates, particularly in ICUs with higher proportions of emergency patients.

This association is consistent with the findings of previous studies [2, 3, 7, 8, 18] conducted on other certain emergencies, supporting the learning curve hypothesis [18]. Another possible mechanism is that the ICUs in the lowest quartile (Q1) had fewer ICU beds relative to total hospital beds (Table 2), suggesting limited resources. Although these ICUs may treat more severely ill patients, the impact of bed count is minimal because adjustments were made for illness severity and staff number. Our analysis also revealed a nonlinear association between case volume and patient outcomes. This U-shaped association was more evident for ICU mortality and 28-day mortality, suggesting that a similar mechanism may exist as that described in a previous studies [19, 20] in which an excess case volume was negatively associated with mortality. However, as shown in Supplementary Table 5, we observed differences in short-term mortality rates and hospital mortality rates in Q4, depending on the proportion of emergency patients. This indicates that the effect of case volume on short-term mortality is heterogeneous across the proportion of emergency patients in the ICU.

Furthermore, the stratified analysis by proportion of emergency patients showed a more obvious reduction in in-hospital mortality in ICUs with a predominantly emergency patient population, which may be because of the positive impact of ICU specialization. These ICUs may be well resourced and experienced in the treatment of emergency conditions, which may lead to better patient outcomes.

In this study, the MOR for in-hospital mortality was low (MOR 1.07; 95% CI: 1.02–1.12), indicating little variation in in-hospital mortality among ICUs. However, the MOR for short-term mortality, especially ICU mortality, was significantly higher (MOR 1.36; 95% CI: 1.27–1.46), suggesting a notable disparity in short-term outcomes, which were potentially influenced by ICU-level and patient-level variables. The MOR is defined as the median value of the OR between the highest and lowest risk clusters; if two clusters are chosen at random, the MOR indicates the increased risk (in median) of moving to another higher-risk cluster [16].

The MOR for ICU mortality increased substantially, suggesting a significant variation in short-term mortality risk across ICUs, which cannot be fully explained by ICU- or patient-level variables. These MOR results may have been derived from differences between the ICUs that were not captured in this dataset. Factors that may have created variations include ICU practices and protocols (e.g., differences in treatment protocols, staffing, and available resources), admission criteria (e.g., variation in patient admission criteria that may affect the risk profile of ICU patients), discharge criteria (affecting the length of ICU stay), facility characteristics (e.g., lack of high-dependency care units, which may affect admission and discharge criteria), and regional differences in the provision and use of critical care beds [21]. These findings indicate that further investigation of the factors affecting patient outcomes in the ICUs is required.

The RSMR for in-hospital mortality for each ICU (Fig. 2) could be appropriately compared with that of the entire population by using a funnel plot [14], showing the variation in the RSMR for ICUs with fewer emergency admissions. This finding suggests disparities in resources, quality of care, or patient population characteristics. This disparity was supported by the multilevel analysis (Model 2), which showed increased in-hospital mortality in ICUs with fewer than 200 emergency admissions per year (Q1), after adjusting for patient characteristics and ICU resources. Higher-case-volume ICUs may have lower RSMRs, possibly because of factors such as experienced staff, effective protocols, and resource availability.

The RSMR is a crucial indicator of quality of care but must be interpreted in conjunction with other indicators, such as the length of stay and readmission rates, for a comprehensive view of ICU performance. When calculating the RSMR, the method of risk adjustment must be considered to avoid misleading results—particularly if certain high-risk patient populations are inadequately accounted for. We improved the reliability of our results by using the JROD score [11], a newly developed index for intensive care patients in Japan. However, missing values or reporting bias when calculating the RSMR could affect the accuracy and reliability of the results.

One strength of this study was the use of the JIPAD, which registers various ICUs nationwide and regularly undertakes efforts to maintain data accuracy [22]. It is the most reliable database for ICUs in Japan in terms of size, reliability, and precision. Therefore, we believe that the participants and facilities in this study represent a highly representative population of emergency patients requiring intensive care in Japan.

This study has some limitations. Each facility in the JIPAD is anonymized; therefore, we classified the participating facilities, based on the ratio of emergencies to admitted patients. Second, a possibility of selection bias existed because five of nine centers were excluded because they had a small number of potentially eligible patients, they treated primarily pediatric patients, and were highly heterogeneous, whereas the other four centers lacked information on the number of intensivists and nurses. Although information on the number of intensivists and nurses was lacking, the small number of excluded patients had little impact on the results. Third, participation in the JIPAD was voluntary; therefore, the participating ICUs may have been more proactive in improving the quality of care. ICUs with larger case volumes or a higher proportion of emergency patients are more likely to participate in the JIPAD, which may cause further selection bias. Nevertheless, analyzing a homogeneous population increases the validity of comparisons and the reliability of statistical analysis. Furthermore, caution should be exercised when generalizing the results because these ICUs may not be fully representative of all ICUs in Japan. Fourth, we were unable to assess the proficiency or years of experience of the ICU staff. In Japan, intensivists typically have a background in emergency medicine or anesthesia [23]. We also could not assess differences in the background of intensivists. These differences could have influenced the patient outcomes, and therefore require further investigation into the effect of the expertise and training of ICU staff on patient outcomes. A fifth limitation is differences in healthcare systems. Extrapolating the results of this study to other countries may be limited by differences in healthcare systems, especially in ICU settings. However, extrapolation to other countries may be possible. Even after considering the effects of these differences, the results of this study may be relevant beyond the Japanese healthcare system. For instance, a comparable mechanism may be responsible for favorable patient outcomes in the emergency department intensive care unit (ED-ICU) system in the United States [24] or in ICUs where emergency physicians led operations in South Korea [25]. Specifically, this improvement in outcomes can be attributed to the shortened time to ICU admission for emergency patients, effective coordination between the ED and ICU, reduced length of stay in the ED, and a comprehensive understanding of the patients’ condition. Nevertheless, direct comparisons among different healthcare systems should be made with caution. Finally, the utilization of critical care and emergency medical systems in Japan was affected by the COVID-19 pandemic since April 2020 (FY 2020 and beyond) [26,27,28], which may have an impact on patient outcomes. Thus, we categorized data entry into two periods: FY 2015–2019 and FY 2020–2021. Future research could potentially focus on exploring the impact of different ICU characteristics and healthcare reimbursement classifications on critically ill patient outcomes. This research could involve examining factors, such as ICU size, patients’ demographics, and financial incentives within the reimbursement system, to better understand how these factors may influence care quality.