Cirrhosis and fungal infections-a cocktail for catastrophe: A systematic review and meta-analysis with machine learning

Abstract

Objectives: We evaluated the magnitude and factors contributing to poor outcomes among cirrhosis patients with fungal infections (FIs).

Methods: We searched PubMed, Embase, Ovid and WOS and included articles reporting mortality in cirrhosis with FIs. We pooled the point and relative-risk (RR) estimates of mortality on random-effects meta-analysis and explored their heterogeneity (I ² ) on subgroups, meta-regression and machine learning (ML). We assessed the study quality through New-Castle-Ottawa Scale and estimate-asymmetry through Eggers regression. (CRD42019142782).

Results: Of 4345, 34 studies (2134 patients) were included (good/fair/poor quality: 12/21/1). Pooled mortality of FIs was 64.1% (95% CI: 55.4-72.0, I ² : 87%, p < .01), which was 2.1 times higher than controls (95% CI: 1.8-2.5, I ² :89%, p < .01). Higher CTP (MD: +0.52, 95% CI: 0.27-0.77), MELD (MD: +2.75, 95% CI: 1.21-4.28), organ failures and increased hospital stay (30 vs. 19 days) were reported among cases with FIs. Patients with ACLF (76.6%, RR: 2.3) and ICU-admission (70.4%, RR: 1.6) had the highest mortality. The risk was maximum for pulmonary FIs (79.4%, RR: 1.8), followed by peritoneal FIs (68.3%, RR: 1.7) and fungemia (55%, RR: 1.7). The mortality was higher in FIs than in bacterial (RR: 1.7) or no infections (RR: 2.9). Estimate asymmetry was evident (p < 0.05). Up to 8 clusters and 5 outlier studies were identified on ML, and the estimate-heterogeneity was eliminated by excluding such studies.

Conclusions: A substantially worse prognosis, poorer than bacterial infections in cirrhosis patients with FIs, indicates an unmet need for improving fungal diagnostics and therapeutics in this population. ACLF and ICU admission should be included in the host criteria for defining IFIs.

Keywords: ACLF; antifungals; cirrhosis; invasive fungal infections; machine learning; mortality.