Statistical analyses

We used the same primary outcome as described in the algorithm derivation study consisting of 30-day intubation/mortality. For the Lausanne validation cohort (patients included during the second and third waves of the pandemics) only, a secondary outcome was evaluated: 30-day oxygen requirement (all patients hospitalized with oxygen requirement) [5]. This item was not available in the two other cohorts.

Patients with neurological or cardiopulmonary symptoms (see definitions above) at 12–18 months were classified as having PASC.

Differences between groups were evaluated by 1-way ANOVA, Kruskal-Wallis, or chi-square tests, as appropriate. A two-sided P value < 0.05 was considered indicative of statistical significance.

We first updated the classification and regression tree analysis (CRT) of the derivation cohort, as previously described [5], to determine the adapted cut-off for the sTREM-1 based algorithm using the measurements on the Luminex platform. The updated sTREM-1 cut-off was at 225 pg/mL (compared with 689 pg/ml when measured with the ELLA platform). The respiratory rate cut-off remained unchanged as determined by CRT analysis in the previous study [5]. The prognostic performance of the updated algorithm for 30-day intubation/mortality was similar to the previous CRT model [5].

In a second step, we assessed the diagnostic accuracy of sTREM-1 and IL-6 in the multicenter validation cohort by calculating the area under the receiver-operating characteristic curve (AUROC) for the 30-day intubation/mortality and for the PASC outcome, as previously described [5].

In a third step, we determined the predictive accuracy of the sTREM-1 based algorithm (including first respiratory rate with a cut-off point at 24/min and, second, sTREM-1 with a cut-off at 225 pg/mL), for the multicenter validation. We also determined the performance of the IL-6 based algorithm (including IL-6 with a cut-off at 15.1 pg/mL), using the Lausanne validation cohort solely, as data on oxygen requirement were unavailable in the St. Gallen and EOC cohorts.

For exploratory purposes, we performed another classification and regression tree analysis (CRT), as described [5], including all vital signs, clinical severity scores, and biomarkers, to determine if another algorithm could be determined with the data from all the centers to predict 30-day intubation/mortality.

All analyses were performed with R Core Team (2019), IBM SPSS version 26 and 29 (IBM Corporation, Armonk, NY) and Excel for Windows.