Assessing the impact of additional clinical variables on <scp>SOFA</scp> score predictive accuracy: a retrospective cohort study

The Sequential Organ Failure Assessment (SOFA) score was developed to describe the morbidity of patients who are critically ill [1] and is still used widely. However, some of the original score constituents no longer align with contemporary critical care clinical practice. Proposals to update the score including the addition and/or update of SOFA score constituents are yet to be evaluated [2]. The aim of our study was to evaluate the impact of potential updates on the predictive accuracy of a modified SOFA (mSOFA) score. This single-centre retrospective cohort study was conducted at Jichi Medical University Saitama Medical Center. This study was approved by the institutional review board. Patients aged ≥ 18 y who were admitted to the ICU and stayed for ≥ 24 h between August 2017 and July 2023 were included. Data on patient characteristics, clinical data to inform mSOFA calculations and survival outcomes were extracted from electronic medical records. The additional mSOFA score constituents included: the use of high-flow nasal oxygenation (HFNO), non-invasive ventilation (NIV) and veno-venous extracorporeal membrane oxygenation (VV-ECMO) to the respiratory component; platelet transfusion to the coagulation component; vasopressin and veno-arterial extracorporeal membrane oxygenation (VA-ECMO) to the cardiovascular component; renal replacement therapy (RRT) to the renal component; and lactate levels to a new, seventh, component. The scoring for the new items was as follows: VV-ECMO, 4 points; NIV, minimum of 3 points assigned with 4 points if the ratio of partial pressure of oxygen in arterial blood to the fraction of inspiratory oxygen concentration (P/F) during use was < 100; HFNC, minimum of 2 points assigned, with 3 points if the P/F ratio was < 200 and 4 points if it was < 100; platelet transfusion, 4 points; vasopressin use, 4 points; VA-ECMO, 4 points; and RRT (in patients not on maintenance dialysis), 4 points. Lactate levels were scored as: < 2 mmol.l-1, 0 points; 2–4 mmol.l-1, 1 points; 4–6 mmol.l-1, 2 points; 6–8 mmol.l-1, 3 points; and ≥ 8 mmol.l-1, 4 points. Scores were assigned to the new items based on their mortality rates and compared with the mortality rates of the original SOFA score items. The outcome was the area under the receiver operating characteristic curve (AUROC) for hospital mortality. The highest scores within 24 h of admission were defined as ‘admission SOFA’, and the highest scores during the ICU stay were defined as ‘max SOFA’ [3]. Analysis was performed using R (version 4.3.3, R Foundation, Vienna, Austria), and the DeLong test was used to compare the AUROCs. A two-sided test with a significance level of 5% was used. Of the 9629 patients admitted, 6167 were included in the analysis (online Supporting Information Figure S1). Patient demographics are shown in Table 1. The distribution and mortality rates of each SOFA score component are shown in the online Supporting Information Figure S2. A comparison of the AUROCs between the original SOFA and mSOFA scores revealed statistically significant but clinically minimal differences for admission SOFA, and there was no difference for max SOFA. (Fig. 1 and online Supporting Information Table S1). In this study, we have added widely used organ support measures, platelet transfusion, vasopressin and lactate levels to the original SOFA score but did not observe any meaningful improvement in predictive power for in-hospital mortality. This may be because many of the new items lacked fidelity due to a score assignment of 4 points, which is the maximum possible score for each organ score. Almost all patients who qualified for these new items were already receiving the highest score of 4 points in the original SOFA score (online Supporting Information Table S2), consequently leading to no additional improvement in predictive accuracy by the modification. These values were selected by the study team; further research is required to evaluate the relative importance of each variable in larger datasets. When updating the SOFA score, it seems important to adjust the clinical thresholds, such as the criteria based on vasopressor dosage, to align with the current clinical practices rather than simply adding new items. There have been several reports on modifying single components, which show improved predictive power primarily focused on the cardiovascular component, suggesting its critical role in the SOFA score update [4, 5]. These reports also indicate that adjusting clinical thresholds, rather than merely adding new items, can lead to more accurate predictions. Despite being a single-centre study and limitations due to its retrospective nature, the strengths of this study include its large sample size (n > 6000) with minimal missing data and its evaluation of new components following the SOFA score update proposal. In conclusion, updating the SOFA score to include organ support, platelet transfusion and lactate levels did not significantly improve the predictive power for mortality using the current threshold settings. The authors thank Shigehiko Uchino at Jichi Medical University Saitama Medical Center, Masamitsu Sanui at Jichi Medical University, Kodai Shibao and Ryohei Yamamoto who are members of the Japanese Society of Education for Physicians and Trainees in Intensive Care-Clinical Trial Group for their contributions to this study. No external funding or competing interests declared. Table S1. Area under the receiver operating characteristic curve for prediction of hospital mortality for SOFA and modified SOFA. Table S2. Concordance rates between the SOFA and modified SOFA score. Figure S1. Study flow diagram. Figure S2. Distribution and mortality rates of each component in the original SOFA score and the added components. Please note: The publisher is not responsible for the content or functionality of any supporting information supplied by the authors. Any queries (other than missing content) should be directed to the corresponding author for the article.