In a recent viewpoint article, Shigehiko Uchino and Shinshu Katayama examine how the updated Sequential Organ Failure Assessment score, SOFA-2, changes the assessment of organ dysfunction in critical care. Their central argument is that SOFA-2 moves away from the original SOFA model of direct, observation-based scoring and towards a more interpretive approach that depends on treatment context, clinical intent, resource availability and assumptions that may not be recoverable from routine data. While the update reflects modern intensive care practice, the authors contend that this increased contextual complexity may undermine reproducibility, automated calculation and reliable use in multicentre databases.
The original SOFA score was designed around a small number of objective physiological and laboratory measurements that were broadly available, reproducible and relatively easy to interpret across settings. SOFA-2 was developed to modernise this established instrument by incorporating multinational intensive care data, expert consensus, contemporary organ-support technologies and current treatment strategies. Its stated aims included simplicity, usability, broad applicability and content validity. However, the authors argue that the revised score expands not only the number of variables required but also their nature. Several components now require interpretation of why treatment was or was not delivered, whether support was available, and whether ceilings of care influenced management.
The article uses the renal component to illustrate this shift. Original SOFA kidney scoring relied on serum creatinine and total daily urine output. SOFA-2 instead requires hourly urine output adjusted for body weight and introduces multiple conditional rules involving renal replacement therapy. Assessors must distinguish therapy given for renal from non-renal indications. Patients not receiving renal replacement therapy may nevertheless receive the maximum renal score when they meet treatment criteria but therapy is withheld because of a ceiling of care, unavailable equipment or delayed initiation. This requires consideration of oliguria, creatinine, potassium, pH and bicarbonate, as well as the reason treatment was not provided. Patients receiving intermittent therapy may also retain the maximum score on non-treatment days until therapy is formally discontinued. Renal scoring therefore becomes a multistage judgement involving clinical intention, resources and decisions rather than a straightforward reading of measured values.
The comparison table reinforces this broader expansion across organ systems. Respiratory scoring now includes alternative oxygenation thresholds, several forms of respiratory support, extracorporeal membrane oxygenation, treatment ceilings and the availability of support. Cardiovascular assessment includes additional vasoactive drugs and mechanical circulatory devices, alongside treatment availability. Neurological scoring adds motor scores, delirium diagnosis, delirium medication and sedation. Kidney scoring incorporates weight-indexed urine output, renal replacement eligibility, biochemical variables and treatment context. By contrast, the liver and coagulation components remain based on total bilirubin and platelet count.
The authors identify three main consequences. First, manual scoring may become less reproducible. Final scores can vary depending on who performs the assessment, which documents they consult and how they infer clinical intent. Since even the simpler original SOFA score has been subject to misclassification in routine practice, adding more interpretive requirements may amplify inconsistency between observers.
Secondly, automated implementation becomes more difficult. Translating contextual rules into software requires numerous conditional branches and assumptions not fully resolved by the published definitions. Different programmers may therefore produce different algorithms that generate different scores from the same clinical data. The authors note that recent analysis found SOFA-2 required more variables and substantially more executable logic than other intensive care severity scores. They emphasise the importance of transparent implementation details and reference code for complex scoring systems, but state that no executable reference code had been released at the time of writing.
Thirdly, contextual dependence complicates use in multicentre, national and secondary databases. Many databases involved in SOFA-2 development lacked variables required for complete implementation. Information such as eligibility for renal replacement therapy or medication for delirium is often absent from structured records. As a result, sites may use different assumptions or imputation methods, creating apparent differences in scores that reflect documentation systems rather than genuine biological variation. Subjective documentation may also introduce small systematic “microbiases” affecting different patient groups.
The authors acknowledge why SOFA-2’s developers incorporated context: modern critical care links organ dysfunction closely with treatment choices and resource availability. Nevertheless, they argue that numerical severity scoring increasingly occurs automatically within electronic health records, registries and research pipelines, making consistency and reproducibility essential. Embedding contextual complexity directly into the score favours clinical completeness at the expense of standardisation.
Their proposed solution is not to reject modernisation or the inclusion of contemporary organ-support methods, but to separate observed organ dysfunction from treatment decisions. Contextual factors could be removed from the core numerical score or labelled as optional modifiers for bedside interpretation. Re-emphasising observable, measurement-based variables would improve agreement across clinicians, software implementations and databases while restoring the simplicity on which SOFA was originally founded. The article concludes that, for SOFA-2, refinement may require subtraction rather than continued addition.
Source: AJRCCM
Image Credit: iStock
