Platelet-containing systemic inflammatory indices show reduced prognostic discrimination in critically ill cirrhosis: a MIMIC-IV cohort study with external validation in eICU-CRD
Related Experiment Videos
Tingting Ding1, Yujie Yin1, Youmei Ying1
1Nanjing Jiangbei Hospital, Affiliated Nanjing Jiangbei Hospital of Xinglin College, Nantong University, 552 Geguan Road, Jiangsu, 210048, China.
Background:
Complete blood count-derived inflammatory indices are increasingly used as prognostic markers in critical illness. Three of the commonest, the systemic immune-inflammation index (SII), platelet-to-lymphocyte ratio (PLR), and pan-immune-inflammation value (PIV), contain the platelet count, which in cirrhosis is determined largely by processes other than inflammation. They have not been compared head-to-head in critically ill cirrhosis.
Methods:
In a retrospective MIMIC-IV cohort of adults with cirrhosis in intensive care, six indices were derived from the first white cell differential within 24 h, and discrimination for in-hospital mortality was measured by the area under the ROC curve (AUC), with external validation in eICU. Because each platelet-containing index equals a platelet-free counterpart multiplied by the platelet count (e.g. SII = platelet × neutrophil-to-lymphocyte ratio [NLR]), testing that multiplication was the primary analysis, with the neutrophil count as a positive control.
Results:
Of 2,227 patients, 627 (28.2%) died in hospital. NLR discriminated best (AUC 0.651) and the platelet-containing indices lowest (PLR 0.546). Multiplying by the platelet count degraded discrimination (NLR to SII, ΔAUC - 0.048; SIRI to PIV, - 0.038; both p < 0.001), whereas multiplying by the neutrophil count improved it (+ 0.093, p < 0.001), the change depending on which cell count was introduced. The platelet count alone discriminated no better than chance (AUC 0.522), added nothing to a model containing MELD-Na, OASIS, age, and sex (ΔAUC + 0.0003), and showed no independent association with mortality when entered separately. The ranking was reproduced externally (eICU, n = 800).
Conclusions:
The conventional platelet-containing indices did not outperform their platelet-free counterparts. Thrombocytopenia in critically ill cirrhosis is multifactorial and no direct measure of portal hypertension was available, so the reason cannot be established here. Incremental value over established severity scores was marginal for every index, and indices from other diseases require re-evaluation in cirrhosis before use.