Scholarlink

Background <p>In prehospital multi-patient traffic incidents, the first ambulance to depart —priority transport — may reflect underlying injury severity rather than a treatment effect, as more severely injured patients are systematically prioritized. Conventional statistical methods often fail to separate this severity-dependent selection from actual outcomes. We applied a “doubly robust” statistical framework (augmented inverse probability weighting, AIPW), which combines a treatment model and an outcome model, to better account for measured differences, reduce selection bias, and examine whether on-scene time modifies this association.</p> Methods <p>We conducted a nationwide retrospective cohort study using emergency medical service transport records from the Fire and Disaster Management Agency of Japan (January 2021–December 2023). Of 13,821,224 records, 143,130 patients from traffic-related multi-patient incidents (TMIs; ≥2 patients per incident) with complete data were included. The exposure was priority transport. The primary outcome was physician-assessed death or critical/severe injury at hospital arrival. Marginal adjusted risk differences (RDs) were estimated using AIPW. Effect modification by on-scene time (5–60&#xa0;min) was assessed by marginal standardization with cluster-bootstrap confidence intervals. Sensitivity analyses included overlap restriction and E-value quantification of unmeasured confounding.</p> Results <p>Among 143,130 patients (median age 41 years; female, 50.57%), 62,769 (43.85%) underwent priority transport. Severe outcomes occurred in 4.35% of priority and 1.87% of non-priority patients. The overall RDs was 2.2% points (95% CI, 2.01–2.36). A significant interaction with on-scene time was observed (<i>p</i> &lt; .001): the marginal RD was peaked at 5&#xa0;min (6.4% points; 95% CI, 5.6–7.2), declined to a minimum of 1.4% points (95% CI, 1.3–1.6) at approximately 32&#xa0;min, and increased modestly to 2.4% points (95% CI, 2.0–2.8) at 60&#xa0;min. Sensitivity analyses confirmed result robustness, yielding an E-value of 3.76 (lower bound, 3.57).</p> Conclusions <p>The association between priority transport and severe outcome was modified by on-scene time, with the largest risk difference at the shortest on-scene times. This U-shaped pattern is consistent with preferential selection of more severely injured patients for rapid transport, rather than a direct effect of transport order. Accounting for prehospital triage selection when interpreting transport-outcome associations in critical care research.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Prehospital transport prioritization and time-varying selection bias in traffic-related multi-patient incidents: a nationwide retrospective cohort study

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年8月4日

查看全文
  • Atsushi Kubo,
  • Shinobu Tamura,
  • Atsushi Hiraide,
  • Daigo Morioka,
  • Ryu Murakami,
  • Kenko Fukui,
  • Shigeaki Inoue

摘要

Background

In prehospital multi-patient traffic incidents, the first ambulance to depart —priority transport — may reflect underlying injury severity rather than a treatment effect, as more severely injured patients are systematically prioritized. Conventional statistical methods often fail to separate this severity-dependent selection from actual outcomes. We applied a “doubly robust” statistical framework (augmented inverse probability weighting, AIPW), which combines a treatment model and an outcome model, to better account for measured differences, reduce selection bias, and examine whether on-scene time modifies this association.

Methods

We conducted a nationwide retrospective cohort study using emergency medical service transport records from the Fire and Disaster Management Agency of Japan (January 2021–December 2023). Of 13,821,224 records, 143,130 patients from traffic-related multi-patient incidents (TMIs; ≥2 patients per incident) with complete data were included. The exposure was priority transport. The primary outcome was physician-assessed death or critical/severe injury at hospital arrival. Marginal adjusted risk differences (RDs) were estimated using AIPW. Effect modification by on-scene time (5–60 min) was assessed by marginal standardization with cluster-bootstrap confidence intervals. Sensitivity analyses included overlap restriction and E-value quantification of unmeasured confounding.

Results

Among 143,130 patients (median age 41 years; female, 50.57%), 62,769 (43.85%) underwent priority transport. Severe outcomes occurred in 4.35% of priority and 1.87% of non-priority patients. The overall RDs was 2.2% points (95% CI, 2.01–2.36). A significant interaction with on-scene time was observed (p < .001): the marginal RD was peaked at 5 min (6.4% points; 95% CI, 5.6–7.2), declined to a minimum of 1.4% points (95% CI, 1.3–1.6) at approximately 32 min, and increased modestly to 2.4% points (95% CI, 2.0–2.8) at 60 min. Sensitivity analyses confirmed result robustness, yielding an E-value of 3.76 (lower bound, 3.57).

Conclusions

The association between priority transport and severe outcome was modified by on-scene time, with the largest risk difference at the shortest on-scene times. This U-shaped pattern is consistent with preferential selection of more severely injured patients for rapid transport, rather than a direct effect of transport order. Accounting for prehospital triage selection when interpreting transport-outcome associations in critical care research.