Article
Clustering of serum biomarkers involved in post-subarachnoid haemorrhage (SAH) complications
Clustering von Serum-Biomarkern, die an Komplikationen nach Subarachnoidalblutungen (SAB) beteiligt sind
Search Medline for
Authors
Published: | May 25, 2022 |
---|
Outline
Text
Objective: High dimensionality of clinical data, combined with usually low number of independent observations poses challenges for both classical statistical and artificial intelligence approaches. Here, we apply a clustering approach to identify serum biomarkers associated with the cerebral vasospasm (CVS).
Methods: Serum concentrations of 10 potential biomarker sincluding DAMPs (HMGB1, mitochondrial DNA gene fragments such as Cytochrome B (CytB), D-loop, and Cytochrome c oxidase subunit-1 (Cox-1)), pro- and anti-inflammatory cytokines (IL-6, IL-17, IL-23, IL-10, CCL5) and leukocytes for 80 SAH patients were recorded on day 1 after SAH, together with demographic and clinical data. Fourteen patients were excluded due to missing data. The ten biomarkers and the patient age were transformed using the Yeo-Johnson power transform and normalized to (μ=0, sd=1). The dataset was split into a training (43 patients) and a test set. Correlation heat maps were computed separately for the two subsets. Pairs of variables which showed inconsistent correlations in training and test, i.e. having opposing signs or differing by > 0.25, were excluded from further analysis. Clusters of relevant biomarkers were identified on the complete data as having correlations ≥ 0.25 and sharing at least three predictors, separately for patients who developed post-SAH CVS and those who did not.
Results: Among the patients who suffered CVS, CytB, Cox-1, D-loop and IL-23 formed one cluster with fair to moderately strong correlations. No such cluster could be observed among the patients who did not suffer from CVS. CytB and Cox-1(gene fragments of the representative mtDNA genes) showed a very strong correlation (0.82), as could be expected. The other cluster, again only for the CVS group, contained IL-6, IL-10, age, and Hunt and Hess score, with only fair correlations among the biomarkers. The apparent weak negative correlations (r ≈ -0.3) in the cluster (CCL5, IL-6, age) on the complete set were not consistent over the test and the training subsets.
Conclusion: Different clusters of serum biomarkers in patients suffering from CVS after SAH compared to patients without CVS suggest that these biomarkers are probably involved in pathophysiological processes leading to CVS. However, due to high dimensionality of the data and relatively small number of observations, the possibility of random noise appearing as a pattern must be ruled out. Therefore, further detailed investigations with high number of patients are warranted.