Journal of the American Heart Association ORIGINAL RESEARCH Blood Biomarker Panels for the Early Prediction of Stroke-­Associated Complications Júlia Faura , MSc; Alejandro Bustamante , MD, PhD; Silvia Reverté, PhD; Teresa García-­Berrocoso, PhD; Mónica Millán, MD, PhD; Mar Castellanos , MD, PhD; Blanca Lara-­Rodríguez, MD; Josep Zaragoza, MD; Oriol Ventura, MSc; María Hernández-­Pérez, MD, PhD; Cecile van Eendenburg, MD; Pere Cardona, MD, PhD; Elena López-­C ancio , MD, PhD; David Cánovas, MD, PhD; Joaquín Serena , MD, PhD; Marta Rubiera , MD, PhD; Antoni Dávalos, MD, PhD; Joan Montaner , MD, PhD BACKGROUND: Acute decompensated heart failure (ADHF) and respiratory tract infections (RTIs) are potentially life-­threatening complications in patients experiencing stroke during hospitalization. We aimed to test whether blood biomarker panels might predict these complications early after admission. METHODS AND RESULTS: Nine hundred thirty-­eight patients experiencing ischemic stroke were prospectively recruited in the Stroke-­Chip study. Post-­stroke complications during hospitalization were retrospectively evaluated. Blood samples were drawn within 6 hours after stroke onset, and 14 biomarkers were analyzed by immunoassays. Biomarker values were normalized using log-­transformation and Z score. PanelomiX algorithm was used to select panels with the best accuracy for predicting ADHF and RTI. Logistic regression models were constructed with the clinical variables and the biomarker panels. The additional predictive value of the panels compared with the clinical model alone was evaluated by receiver operating characteristic curves. An internal validation through a 10-­fold cross-­validation with 3 repeats was performed. ADHF and RTI occurred in 19 (2%) and 86 (9.1%) cases, respectively. Three-­biomarker panels were developed as predic- tors: vascular adhesion protein-­1 >5.67, NT-­proBNP (N-­terminal pro-­B-­t ype natriuretic peptide) >4.98 and d-­dimer >5.38 (sensitivity, 89.5%; specificity, 71.7%) for ADHF; and interleukin-­6 >3.97, von Willebrand factor >3.67, and d-­dimer >4.58 (sensitivity, 82.6%; specificity, 59.8%) for RTI. Both panels independently predicted stroke complications (panel for ADHF: odds ratio [OR] [95% CI], 10.1 [3–­52.2]; panel for RTI: OR, 3.73 [1.95–­7.14]) after adjustment by clinical confounders. The addition of the panel to clinical predictors significantly improved areas under the curve of the receiver operating charac- teristic curves in both cases. CONCLUSIONS: Blood biomarkers could be useful for the early prediction of ADHF and RTI. Future studies should assess the usefulness of these panels in front of patients experiencing stroke with respiratory symptoms such as dyspnea. Key Words: ADHF ■ biomarkers ■ stroke ■ stroke-­associated infection M ortality of patients experiencing stroke in Europe represent an opportunity to further improve the out- ranges from 13% to 35%.1 Acute post-­stroke come of these patients.3 complications, such as respiratory infections Acute decompensated heart failure (ADHF) is de- and cardiac complications, are major causes of death fined as a new onset or rapidly or gradual worsen- in patients experiencing acute stroke.2 Early detection ing of heart failure (HF) symptoms that require urgent and treatment of these medical complications might therapy.4 A frequency of 17% of ADHF in acute stroke Correspondence to: Alejandro Bustamante, MD, PhD, Stroke Unit, Hospital Universitari Germans Trias i Pujol, Carretera de Canyet, s/n, 08916 Badalona, Barcelona, Spain. E-­mail: abustamanter.germanstrias@gencat.cat For Sources of Funding and Disclosures, see page 7. © 2021 The Authors. Published on behalf of the American Heart Association, Inc., by Wiley. This is an open access article under the terms of the Creative Commons Attribution-­NonCommercial License, which permits use, distribution and reproduction in any medium, provided the original work is properly cited and is not used for commercial purposes. JAHA is available at: www.ahajournals.org/journal/jaha J Am Heart Assoc. 2021;10:e018946. DOI: 10.1161/JAHA.120.0189461 Faura et al Predicting Stroke Complications With Biomarkers METHODS CLINICAL PERSPECTIVE The present study represents a retrospective analysis of the patients experiencing ischemic stroke prospec- What Is New? tively recruited at the multicenter Stroke-­Chip study.8 • This is a retrospective study of a multicenter Certified neurologists, blinded to the biomarker ana- cohort with 938 patients experiencing is- lytic results, retrospectively evaluated clinical records chemic stroke in which the measurement of to determine the occurrence of post-­ stroke compli- 14 biomarkers was performed with early blood cations during hospitalization. The Stroke-­Chip study samples, extracted in the first 6 hours after the episode. protocol was approved by each recruiting center’s eth- • Two biomarker combinations add predictive ics committee, and all patients or relatives gave written value to the clinical models for the prediction of informed consent. The data that support the findings 2 major complications after stroke: acute de- of this study are available from the corresponding au- compensated heart failure and respiratory tract thor upon reasonable request. infections. What Are the Clinical Implications? Diagnosis of Post-­Stroke Complications • These 2 biomarker panels, jointly with clinical ADHF was defined, according to American Heart variables, would be able to predict acute de- Association Guidelines,9 as new onset of dyspnea compensated heart failure and respiratory tract and/or fatigue, together with fluid retention, which infection in the first hours after stroke. may lead to pulmonary, or splanchnic congestion or • This would allow clinicians to monitor the pa- peripheral edema, in the absence of other causes of tients at the highest risk to experience these dyspnea. pathologies and treat them accordingly, reduc- RTIs were diagnosed according to the Centers for ing the length of stay and the mortality of these Disease Control and Prevention criteria.10 In summary, patients. the diagnosis of RTI required the presence of fever (body temperature >38°C) or elevated white blood cell count (>1.2×1010/L) and evidence of organ-­ specific involvement (purulent sputum, pulmonary infiltrate, Nonstandard Abbreviations and Acronyms cough, etc) or a positive blood culture. RTIs were clas- sified as upper RTI, lower RTI, and pneumonia. An ADHF acute decompensated heart failure abnormal chest radiograph was required for the diag- RTI respiratory tract infections nosis of stroke-­associated pneumonia. vWF Von Willebrand factor Blood Sample Collection and Biomarker Measurement patients has been reported, and this cardiac compli- Samples were drawn at admission, within 6 hours cation is a strong independent predictor of poor func- after stroke onset, and before the administration of tional outcome in these patients.5 acute-­phase therapies such as reperfusion. Blood Respiratory tract infection (RTI) represents a well-­ was collected into EDTA tubes, centrifuged at 1500g known complication in patients experiencing stroke, for 15 minutes at 4°C, and plasma aliquots were frozen and its frequency is around 12%.6 RTIs are also linked at −80°C until biomarker measurement. with poor functional outcome of the patients, a pro- A 14-­biomarker panel was measured. This panel longed length of stay in the hospital, and increased included insulin-­like growth factor–­binding protein-­3, mortality up to 1 year.7 tumor necrosis factor receptor-­1, growth-­related on- Both RTI and ADHF may have common symptoms cogene-­α, Fas ligand, heat shock 70 kDa protein-­8, such as dyspnea, and early chest radiograph may not NT-­proBNP (N-­terminal pro-­B-­type natriuretic eptide), show typical changes, so differentiating between the 2 d-­ dimer, interleukin-­ 6, von Willebrand factor (vWF), clinical conditions can be difficult in daily clinical prac- vascular adhesion protein-­1, endostatin, S100 calcium-­ tice. In this context, blood biomarkers might serve as a binding protein B, apolipoprotein CIII, and neuron cell useful tool to identify those patients at the highest risk adhesion molecule. The selection of these biomark- of developing these complications, as well as for the ers and the measurement methods were described early differential diagnosis of post-­stroke RTI or ADHF. previously.8 In this study, we aimed to test whether blood bio- Biomarker values were log-­transformed with a base markers measured at admission might identify patients of 10 and divided by the value of the control sample of at the highest risk of RTI or ADHF. each plate. Because of the high intra-­assay variability J Am Heart Assoc. 2021;10:e018946. DOI: 10.1161/JAHA.120.0189462 Faura et al Predicting Stroke Complications With Biomarkers for some of the molecules, all the values were stan- size of the sample according to the central limit theorem. dardized by means of Z score. The univariate analysis was performed using the χ2 test for categorical variables and t-­test for continuous varia- Statistical Analysis bles. Binary logistic regression models were developed. Statistical analyses were conducted with Statistical Putative confounders based on previous literature were Packages for Social Sciences (version 22; SPSS Inc., added to the models to adjust them. Multicollinearity Chicago, IL) and R software (version 3.6.1; R Foundation among the predictor variables has been checked with for Statistical Computing, Vienna, Austria). Normality of the variance inflation factor. Firth correction was applied the continuous variables was assumed because of the (“logistf” package in R) when constructing the binary Table 1. Univariate Analysis for ADHF and RTI After Stroke ADHF RTI No (N=919) Yes (N=19) P Value No (N=852) Yes (N=86) P Value Age, y 72.52 (±13) 81.1 (±13.5) <0.0001* 72.2 (±13) 77.2 (±12.3) 0.001† Sex, female 420 (45.7) 11 (57.9) 0.291 388 (45.6) 42 (48.8) 0.572 Hypertension 675 (73.4) 13 (68.4) 0.624 617 (72.6) 69 (80.2) 0.127 Dyslipemia 451 (49.1) 9 (47.4) 0.883 421 (49.5) 38 (44.2) 0.345 Diabetes mellitus 237 (25.8) 2 (10.5) 0.184 214 (25.2) 25 (29.1) 0.430 Tobacco 149 (16.2) 3 (15.8) 0.999 144 (16.9) 8 (9.3) 0.067 Alcohol 64 (7) 1 (5.3) 0.999 64 (7.5) 1 (1.2) 0.027‡ Atrial fibrillation 318 (34.6) 11 (57.9) 0.035‡ 282 (33.2) 46 (53.5) <0.0001* Coronary disease 145 (15.8) 7 (36.8) 0.023‡ 132 (15.5) 19 (22.1) 0.115 Previous stroke 158 (17.2) 5 (26.3) 0.353 144 (16.9) 18 (20.9) 0.351 Previous mRS 0 (0–­1) 1.5 (0–­3) 0.001* 0 (0–­1) 1 (0–­3) <0.0001* Baseline NIHSS 9 (±7.3) 15 (±4.9) <0.0001* 8.5 (±7.1) 15.5 (±6) <0.0001* TOAST CE 330 (40.7) 11 (84.6) 0.050 337 (40.1) 46 (55.4) 0.007† LAA 116 (14.3) 0 (0) 115 (13.7) 10 (12) Lacunar 109 (11.9) 0 (0) 119 (14.1) 2 (2.4) Undetermined 240 (29.6) 2 (15.4) 255 (30.3) 22 (26.5) Other 15 (1.9) 0 (0) 15 (1.8) 3 (3.6) In-­hospital mortality 108 (9.9) 8 (42.1) <0.0001* 44 (5.2) 30 (34.9) <0.0001* 3-­mo disability 366 (42.2) 16 (84.32) <0.0001* 221 (32.6) 67 (89.3) <0.0001* Apolipoprotein CIII 4.02 (±0.97) 3.68 (±0.98) 0.144 4.01 (±0.98) 3.99 (±0.93) 0.778 d -­dimer 4.08 (±0.95) 4.60 (±0.79) 0.020‡ 4.05 (±0.94) 4.53 (±0.93) <0.0001* Endostatin 4.11 (±0.94) 4.50 (±0.89) 0.073 4.09 (±0.93) 4.35 (±1.08) 0.015‡ GROA 3.99 (±0.97) 3.96 (±0.87) 0.881 4.00 (±0.98) 3.94 (±0.93) 0.582 Interleukin-­6 4.00 (±0.96) 4.24 (±0.77) 0.265 3.95 (±0.94) 4.49 (±1.02) <0.0001* NT-­proBNP 4.12 (±0.95) 5.03 (±0.87) <0.0001* 4.11 (±0.95) 4.45 (±0.91) 0.001* VAP-­1 4.05 (±0.99) 4.79 (±1.05) 0.001* 4.05 (±0.98) 4.19 (±1.12) 0.199 vWF 4.02 (±0.97) 4.25 (±0.88) 0.306 3.97 (±0.97) 4.49 (±0.89) <0.0001* IGFBP-­3 4.00 (±0.98) 3.91 (±1.07) 0.694 3.99 (±0.98) 4.10 (±1.05) 0.330 Fas ligand 3.96 (±0.99) 4.40 (±0.72) 0.056 3.95 (±0.99) 4.16 (±0.99) 0.065 TNF-­R1 4.04 (±0.98) 4.36 (±0.52) 0.038‡ 4.04 (±0.97) 4.17 (±1.03) 0.235 NCAM 3.98 (±0.99) 4.36 (±0.76) 0.090 3.99 (±0.98) 3.91 (±1.03) 0.498 S100B 4.03 (±0.98) 3.72 (±0.87) 0.233 4.02 (±0.99) 3.95 (±0.91) 0.558 Hsc70 4.04 (±0.95) 4.18 (±1.09) 0.567 4.04 (±0.95) 4.13 (±0.92) 0.414 Values are reported as n (%) or mean (±SD). Biomarker values are standardized. ADHF indicates acute decompensated heart failure; CE, cardiac embolism; DBP, diastolic blood pressure; GROA, growth-­related oncogene-­α; Hsc70, heat shock 70 kDa protein-­8; IGFBP-­3, insulin-­like growth factor-­binding protein-­3; LAA, large-­artery atherothrombosis; mRS, modified rankin scale; NCAM, neuron cell adhesion molecule; NIHSS, National Institutes of Health Stroke Scale; NTproBNP, N-­terminal pro-­B-­type natriuretic peptide; RTI, respiratory tract infection; S100B, S100 calcium-­binding protein B; SBP, systolic blood pressure; TNF-­R1, tumor necrosis factor receptor-­1; TOAST, Trial of Org 10172 in Acute Stroke Treatment; and VAP-­1, vascular adhesion protein-­1. *P<0.001. † P<0.01. ‡ P<0.05. J Am Heart Assoc. 2021;10:e018946. DOI: 10.1161/JAHA.120.0189463 Faura et al Predicting Stroke Complications With Biomarkers logistic regression models for ADHF to minimize the esti- patients without this complication. The values of each bio- mates of the coefficients, as there is a low rate of events. marker and the clinical variables are reported in Table 1. To avoid overfitting the models with individual bio- A panel including vascular adhesion protein-­1 >5.67, markers, the best biomarker combination for the pre- NT-­proBNP >4.98, and d-­dimer >5.38 was selected by diction of the defined complications was selected the PanelomiX software as the most accurate for the using PanelomiX algorithm,11 optimizing the combi- prediction of ADHF. The panel was positive when any 1 nation by accuracy with a maximum of 3 biomarkers. of the markers was above the cutoff (specificity, 71.5%; The analyses for the selection of the best biomarker sensitivity, 89.5%). For the prediction of RTI, the software combination were carried out separately and inde- selected a panel including interleukin-­6>3.97, vWF >3.67, pendently in each of the complications (ADHF and and d-­dimer >4.58. In this case, the panel was positive RTI). Briefly, PanelomiX uses the iterative combina- when at least 2 of the markers were above the cutoff tion of biomarkers and thresholds method to select (specificity, 59.8%; sensitivity, 82.9%) (Figure 1). Of the 9 thresholds for each biomarker that provide the optimal patients with both complications, 7 were positive for both classification performance and studies the robustness panels, 1 patient was positive for only the RTI panel, and of the constructed panels through receiver operating the other patient was negative for both panels. characteristic (ROC) curves and cross-­validation, per- Two different binary logistic regression models were forming a 10-­ fold cross-­validation. After the 10-­ fold developed, one for each complication. The selected bio- cross-­validation and for the evaluation of the perfor- marker panels independently predicted ADHF and RTI, mance, ROC curves of the cross-­validation are built as respectively (odds ratio [OR], 11.859 [95% CI, 2.67–­53.74], the mean of centered predictions over the 10 cross-­ P=0.001 for ADHF; OR, 3.73 [1.95–­7.15], P<0.0001 for validation folds. Then, the ROC curve is compared through De Long’s method with the cross-­validation of A the individual biomarkers. To evaluate the additional value of the selected bio- marker combination to the logistic regression models with the clinical predictors, De Long’s method was used to compare areas under the curve of ROC curves. An internal validation of the constructed logistic regression models was performed through a 10-­fold cross-­validation with 3 repeats, evaluating the same parameters than in the cross-­validation performed by PanelomiX. A P-­value <0.05 was considered statistically significant in all tests. RESULTS B Nine hundred thirty-­eight patients experiencing is- chemic stroke from the Stroke-­ Chip study were included in this analysis. ADHF occurred in 19 of these patients (2%) and RTI in 86 patients (9.1%). Nine patients were noted with both complications. Patients with 1 or both complications were older, had higher rates of atrial fibrillation, higher baseline National Institute of Health Stroke Scale score, and higher previous modified Rankin scale score. Higher rates of in-­hospital mortality and 3-­month disability (Table 1) were also reported in these groups of pa- tients. Patients with ADHF also had higher rates of Figure 1. ROC curves of the biomarkers alone and in coronary disease, and patients with RTI had higher combination for each complication. values of admission glycemia. A, ROC curve for RTI biomarkers panel (black line), D-­ dimer Among the analyzed molecules, patients with ADHF (dotted line), vWF (gray) and interleukin-­6 (dashed line). B, ROC had higher plasma levels of d-­dimer, NT-­proBNP, tumor curve for ADHF biomarker panel (black line), VAP-­1 (dotted line), necrosis factor receptor-­1, and vascular adhesion pro- NT-­proBNP (gray line), and d-­dimer (dashed line). ADHF indicates acute decompensated heart failure; IL-­ 6, interleukin-­ 6; NT-­ tein-­1 than those without this complication, whereas proBNP, N-­terminal pro-­B-­type natriuretic peptide; ROC, receiver patients with RTI had higher levels of d-­dimer, endosta- operating characteristic; RTI, respiratory tract infection; VAP-­1, tin, interleukin-­6, NT-­proBNP, and vWF compared with vascular adhesion protein-­1 and vWF, von Willebrand factor. J Am Heart Assoc. 2021;10:e018946. DOI: 10.1161/JAHA.120.0189464 Faura et al Predicting Stroke Complications With Biomarkers Table 2. Logistic Regression Analyses and Additional Predictive Value of Blood Biomarkers for ADHF and RTI ADHF RTI Clinical Model+Biomarkers Clinical Model+Biomarkers Clinical Model Panel Clinical Model Panel Logistic regression, OR NIHSS at admission 1.09 (1.02–­1.16), P=0.006* 1.06 (1–­1.14), P=0.048† 1.12 (1.09–­1.16), P<0.0001‡ 1.10 (1.07–­1.14), P<0.0001‡ Age 1.04 (0.99–­1.11), P=0.07 1.02 (0.98–­1.08), P=0.26 1.02 (1–­1.04), P=0.02† 1.01 (0.99–­1.03), P=0.2 Coronary disease 3.24 (1.18–­8.44), P=0.02† 2.8 (1.01–­7.36), P=0.047† … … Sex, female 1.36 (0.5–­3.83), P=0.52 1.16 (0.43–­3.26), P=0.76 0.8 (0.51–­1.35), P=0.38 0.8 (0.5–­1.3), P=0.37 Biomarker … 10.1 (3–­52.2), P<0.0001‡ … 3.73 (2.06–­6.75), P<0.0001‡ combination ROC curve AUC 0.80 (0.70–­0.88) 0.88 (0.83–­0.93) 0.78 (0.73–­0.82) 0.81 (0.77–­0.88) De Long test … P=0.038† … P=0.048† Cross-­validation 0.79 (0.72–­0.87) 0.87 (0.82–­0.92) 0.78 (0.74–­0.83) 0.82 (0.78–­0.85) AUC De Long test … P=0.02† … P=0.002* The table represents the comparison between predictive models, with or without biomarker combinations. For logistic regression models, OR (95% CI) and P values are given. For the evaluation of the performance of the models, the AUC (95% CI) of the models without doing cross-­validation and carrying out a 10-­ fold cross-­validation is represented. AUC indicates area under the curve; NIHSS, National Institutes of Health Stroke Scale; OR, odds ratio; and ROC, receiver operator characteristic. *P<0.01. † P<0.05. ‡ P<0.001. RTI) after adjusting by selected clinical variables (Table 2). research in the past years.14,15 However, individual Age, sex, and National Institute of Health Stroke Scale markers tested so far have not demonstrated enough were selected as confounders in both models because predictive accuracy to be implemented in clinical of their influence in the development of both compli- practice. In this article, rather than testing individual cations.12 In case of ADHF, coronary disease was also markers, we tested automatically derived biomarker added to the model, as it has been reported as a risk panels, which might be complementary tools to factor for ADHF.13 Multicollinearity was not detected in achieve a better prediction.16 In fact, biomarker pan- any of the models. The addition of the biomarker panels els might be more valuable to reflect the complexity to the logistic regression models improved the accuracy of the disease and its consequences, with several of the clinical model for both complications in terms of impaired biological pathways that might be reflected area under the curve, from 0.80 (95% CI, 0.70–­0.88) to by automatically selected biomarkers. In fact, we ob- 0.88 (95% CI, 0.83–­0.93) (P=0.038) in ADHF and; from tained 2 different biomarker panels with individual 0.78 (95% CI, 0.73–­0.82) to 0.81 (95% CI, 0.77–­0.88) markers from diverse pathways implicated in stroke (P=0.048) in the case of RTI (Table 2). and post-­stroke complications. These increments of the area under the curve Of the biomarkers proposed for the ADHF predic- when adding the biomarker combinations to the tion after stroke, NT-­proBNP has been widely associ- clinical models were also observed after the internal ated with the diagnosis of ADHF.17 Vascular adhesion validation of these results through a 10-­fold cross-­ protein-­1, a marker of endothelial dysfunction, has also validation. The results of the cross-­ validation are been related with congestive HF, being elevated in shown in Table 2, and all the ROC curves are repre- these patients when compared with controls,18 and an sented in Figure 2. independent prognostic marker for mortality in chronic HF patients.19 d-­dimer is a product of fibrin degrada- tion. In healthy subjects, high levels of d-­dimer have DISCUSSION been associated with incident HF with reduced ejec- The present study demonstrates that the occurrence of tion fraction.20 Moreover, in patients with known HF, 2 common post-­stroke complications during hospitali- higher levels of d-­dimer predicted mortality and inci- zation, ADHF and RTI, could be predicted at admission dent atrial fibrillation.21 by using 2 different biomarker panels in combination The proteins from the RTI biomarker panel are more with clinical factors. related with the acute phase. An increase in the lev- The use of blood biomarkers for the prediction of els of the proinflammatory cytokine interleukin-­ 6 in post-­stroke complications has generated increasing patients with stroke-­associated pneumonia has been J Am Heart Assoc. 2021;10:e018946. DOI: 10.1161/JAHA.120.0189465 Faura et al Predicting Stroke Complications With Biomarkers A B C D Figure 2. Performance of the constructed models for the 2 complications. A and B represent the performance of the models constructed for RTI, and C and D represent the performance of the models constructed for ADHF. A, ROC curve overlapping the clinical model and the clinical model plus the biomarker combination for RTI. The AUC increase is significant (P=0.048). B, ROC curve overlapping the clinical model and the clinical model plus the biomarker combination for RTI after the cross-­validation. The AUC increase is significant (P=0.002). C, ROC curve overlapping the clinical model and the clinical model plus the biomarker combination for ADHF. The AUC increase is significant (P=0.038). D, ROC curve overlapping the clinical model and the clinical model plus the biomarker combination for RTI after the cross-­validation. The AUC increase is significant (P=0.02). ADHF indicates acute decompensated heart failure; AUC, area under the curve; ROC, receiver operating characteristic; and RTI, respiratory tract infection. reported,22–­24 which can predict the risk of stroke-­ and treat these 2 kinds of patients as early as possi- associated infections and mortality. Regarding the other ble. Consequently, with these 2 biomarker combina- 2 biomarkers of the panel, d-­dimer has been associ- tions in clinical practice and the adequate tools to treat ated with community-­acquired pneumonia. Increased these patients, there could be a reduction of the in-­ plasma levels of this molecule were correlated with the hospital length of stay or even a decrease in mortality. severity of the disease.25 vWF is the only protein that In the case of RTI, there is no evidence that antibiotic has not been related with RTIs previously. Its major bi- prophylaxis prevents the onset of these infections or ological functions are related with adhesion of platelets reduces the mortality of these patients,29 but whether and factor VII protection.26 But in the past years, new selective antibiotic prophylaxis in high-­risk patients or functions of this molecule have emerged. Interactions alternative prophylactic measures such as immuno- between platelets and bacterial infections have been modulatory therapies are effective is being studied. reported,27 as well as the role of vWF with the immune As mentioned in the introduction, dyspnea is a cells, as leukocyte extravasation.28 common symptom of ADHF and RTI, and this can Our results, if confirmed, might have clinical impli- sometimes delay the diagnosis of the complication. cations. We propose 2 different biomarker panels that We suggest that these 2 biomarker combinations could predict at admission the appearance of ADHF could also be useful for the differentiation of both dis- and RTI after stroke. Both biomarker panels have a eases, and clinicians would have additional tools for high sensitivity, resulting in a detection of most of the an early diagnosis or treatment. But to confirm this hy- true-­positive patients. After stroke onset, predicting pothesis, a prospective study measuring these mark- these 2 complications could help clinicians to monitor ers in patients with dyspnea should be performed. J Am Heart Assoc. 2021;10:e018946. DOI: 10.1161/JAHA.120.0189466 Faura et al Predicting Stroke Complications With Biomarkers The study has certain limitations. First, the fact Future studies should focus on the validation of these that data were collected by reviewing clinical regis- results as well as the discriminative power of the pan- tries might lead to some missing data. In fact, the els in patients with dyspnea. incidence of ADHF in our study is much lower than reported. Second, the low number of events in the case of ADHF, even after the use of Firth correction, ARTICLE INFORMATION leads to wide CIs for the combination of blood bio- Received August 25, 2020; accepted January 11, 2021. markers, making these results less reliable. Third, Affiliations data on previous history of HF were not available and, From the Neurovascular Research Laboratory, Vall d’Hebron Institute of therefore, we cannot rule out whether the predictive Research (VHIR), Universitat Autònoma de Barcelona, Barcelona, Spain (J.F., A.B., T.G., O.V., J.M.); Stroke Unit, Hospital Universitari Germans Trias value of the biomarker panels might be modified i Pujol, Barcelona, Spain (A.B., M.M., M.H., A.D.); Stroke Unit, Hospital by the existence of previous HF. Likewise, relevant Universitari Verge de la Cinta de Tortosa, Tortosa, Spain (S.R., J.Z.); data related to RTI were also not available, such as Department of Neurology, Complejo Hospitalario Universitario A Coruña, A Coruña Biomedical Research Institute, Spain (M.C.); Stroke Unit, Hospital dysphagia or severe facial palsy at admission. An Universitari de Bellvitge, Barcelona, Spain (B.L., P.C.); Stroke Unit, Hospital external validation or replication of our results has Universitari Josep Trueta, Girona, Spain (C.v.E., J.S.); Stroke Unit, Hospital not been possible because of the inability to obtain Universitario Central de Asturias, Oviedo, Spain (E.L.); Department of Neurology, Hospital Universitari Parc Taulí, Sabadell, Spain (D.C.); and a large cohort of stroke patients with acute blood Stroke, Unit, Department of Neurology, Hospital Universitari Vall d’Hebron, samples. Furthermore, the clinical value added by Barcelona, Spain (M.R.). the biomarkers combination in the clinical models is Acknowledgments rather low. In addition, the use of study-­derived cut- Author contributions: Drs Bustamante, Dávalos, and Montaner conceived off points for biomarker measurement, even when the study. Drs Bustamante, López-­Cancio, and García-­Berrocoso managed is automatically selected by the PanelomiX software the study. Drs Bustamante, Reverté, Millán, Castellanos, Lara-­Rodríguez, Zaragoza, Hernández-­ Pérez, Eendenburg, Cardona, López-­ Cancio, and not driven by the investigators, might overes- Cánovas, Serena, and Rubiera acquired the data. J. Faura and O. Ventura timate the precision of our observations. However, performed the statistical analysis. J. Faura and Drs Bustamante, López-­ given the preliminary nature of the present study, Cancio, and Montaner interpreted the data. J. Faura drafted the manuscript. All authors performed a critical review of the manuscript. the selected cutoffs might serve as a reference for replication purposes in further studies. Finally, the Sources of Funding measured biomarkers had other intended uses; This project received funding from Instituto de Salud Carlos III (ISCIII) therefore, our scores might be improved using other [DTS14/00004, PI17/02130], co-­ financed by the European Regional Development Fund (FEDER), and from Fundació La Marató de TV3 [201706] biomarkers more related to our end points such as and the European Union’s Horizon 2020 research and innovation program C-­reactive protein,30 procalcitonin,31 or serum am- [754517]. Neurovascular Research Laboratory takes part into the Spanish yloid A for RTI32 or midregional proatrial natriuretic stroke research network INVICTUS+ (RD16/0019/0021). The funders had no role in the study design and conduction. peptide, troponin, interleukin-­ 1 receptor–­ like 1 or galectin-­3 for ADHF.17 Disclosures These limitations confer to our study a hypothesis-­ None. generating nature, and future studies are needed to establish the practical usefulness of those markers. These future prospective studies should include pa- REFERENCES tients experiencing stroke at admission, collecting 1. Béjot Y, Bailly H, Durier J, Giroud M Epidemiology of stroke in Europe data of cardiac and respiratory symptoms, and, ide- and trends for the 21st century. Presse Med. 2016;45:e391–­e398. DOI: ally, serial measurements of the biomarker panels in 10.1016/j.lpm.2016.10.003. all included patients, to determine the specificity and 2. Kumar S, Selim MH, Caplan LR Medical complications after stroke. Lancet Neurol. 2010;9:105–­118. DOI: 10.1016/S1474​-­4422(09)70266​-­2. sensitivity of each combination for the prediction be- 3. Bustamante A, Giralt D, García-­Berrocoso T, Rubiera M, Álvarez-­Sabín tween both complications. Dyspnea should also be J, Molina C, Serena J, Montaner J The impact of post-­stroke complica- reported, to explore the capacity of the biomarker tions on in-­hospital mortality depends on stroke severity. Eur Stroke J. 2017;2:54–­63. DOI: 10.1177/23969​87316​681872. combinations to differentiate between the 2 condi- 4. Gheorghiade M, Pang PS Acute heart failure syndromes. J Am Coll tions in these patients. The measurement of specific Cardiol. 2009;53:557–­573. DOI: 10.1016/j.jacc.2008.10.041. biomarkers of each complication, as the commented 5. Burkot J, Kopec G, Pera J, Slowik A, Dziedzic T Decompensated heart fail- ure is a strong independent predictor of functional outcome after ischemic in the previous paragraph, might help to improve the stroke. J Card Fail. 2015;21:642–­646. DOI: 10.1016/j.cardf​ail.2015.03.008. prediction capacity of the panels. 6. Badve MS, Zhou Z, van de Beek D, Anderson CS, Hackett ML Frequency of post-­ stroke pneumonia: systematic review and meta-­ analysis of observational studies. Int J Stroke. 2019;14:125–­136. DOI: CONCLUSIONS 10.1177/17474​93018​806196. 7. Teh WH, Smith CJ, Barlas RS, Wood AD, Bettencourt-­Silva JH, Clark In conclusion, blood biomarkers could be useful for the AB, Metcalf AK, Bowles KM, Potter JF, Myint PK Impact of stroke-­ associated pneumonia on mortality, length of hospitalization, and func- early prediction of 2 of the major complications during tional outcome. Acta Neurol Scand. 2018;138:293–­300. DOI: 10.1111/ the in-­hospital stay of stroke patients: ADHF and RTI. ane.12956. J Am Heart Assoc. 2021;10:e018946. DOI: 10.1161/JAHA.120.0189467 Faura et al Predicting Stroke Complications With Biomarkers 8. Bustamante A, López-­Cancio E, Pich S, Penalba A, Giralt D, García-­ 21. Wannamethee SG, Whincup PH, Papacosta O, Lennon L, Lowe GD Berrocoso T, Ferrer-­Costa C, Gasull T, Hernández-­Pérez M, Millan M, et al. Associations between blood coagulation markers, NT-­proBNP and risk Blood biomarkers for the early diagnosis of stroke: the Stroke-­Chip study. of incident heart failure in older men: the British Regional Heart Study. Stroke. 2017;48:2419–­2425. DOI: 10.1161/STROK​EAHA.117.017076. Int J Cardiol. 2017;230:567–­571. DOI: 10.1016/j.ijcard.2016.12.056. 9. Amsterdam EA, Wenger NK, Brindis RG, Casey DE, Ganiats TG, 22. Kwan J, Horsfield G, Bryant T, Gawne-­Cain M, Durward G, Byrne CD, Holmes DR, Jaffe AS, Jneid H, Kelly RF, Kontos MC, et al. 2014 AHA/ Englyst NA IL-­6 is a predictive biomarker for stroke associated infec- ACC guideline for the management of patients with non–­ST-­elevation tion and future mortality in the elderly after an ischemic stroke. Exp acute coronary syndromes: executive summary: a report of the Gerontol. 2013;48:960–­965. DOI: 10.1016/j.exger.2013.07.003. American College of Cardiology/American Heart Association Task 23. Hoffmann S, Harms H, Ulm L, Nabavi DG, Mackert B-­M, Schmehl I, Force on Practice Guidelines. Circulation. 2014;130:2354–­2394. DOI: Jungehulsing GJ, Montaner J, Bustamante A, Hermans M, et al. Stroke-­ 10.1161/CIR.00000​0 0000​0 00133. induced immunodepression and dysphagia independently predict 10. Horan TC, Andrus M, Dudeck MA CDC/NHSN surveillance definition of stroke-­ associated pneumonia—­ The PREDICT study. J Cereb Blood health care-­associated infection and criteria for specific types of infec- Flow Metab. 2017;37:3671–­3682. DOI: 10.1177/02716​78X16​671964. tions in the acute care setting. Am J Infect Control. 2008;36:309–­332. 24. Bustamante A, Sobrino T, Giralt D, García-­ Berrocoso T, Llombart DOI: 10.1016/j.ajic.2008.03.002. V, Ugarriza I, Espadaler M, Rodríguez N, Sudlow C, Castellanos 11. Robin X, Turck N, Hainard A, Tiberti N, Lisacek F, Sanchez JC, Müller M M, et al. Prognostic value of blood interleukin-­ 6 in the prediction PanelomiX: a threshold-­based algorithm to create panels of biomarkers. of functional outcome after stroke: a systematic review and meta-­ Transl Proteomics. 2013;1:57–­64. DOI: 10.1016/j.trprot.2013.04.003. analysis. J Neuroimmunol. 2014;274:215–­ 224. DOI: 10.1016/j.jneur​ 12. Smith CJ, Bray BD, Hoffman A, Meisel A, Heuschmann PU, Wolfe CDA, oim.2014.07.015. Tyrrell PJ, Rudd AG Can a novel clinical risk score improve pneumonia 25. Arslan S, Ugurlu S, Bulut G, Akkurt I The association between prediction in acute stroke care? A UK multicenter cohort study. J Am plasma d-­dimer levels and community-­acquired pneumonia. Clinics. Heart Assoc. 2015;4:1–­9.e001307. DOI: 10.1161/JAHA.114.001307. 2010;65:593–­597. DOI: 10.1590/S1807​-­59322​01000​0600006. 13. Lala A, Desai AS The role of coronary artery disease in heart failure. 26. Luo GP, Ni B, Yang X, Wu YZ Von Willebrand factor: more than a regu- Heart Fail Clin. 2014;10:353–­365. DOI: 10.1016/j.hfc.2013.10.002. lator of hemostasis and thrombosis. Acta Haematol. 2012;128:158–­169. 14. Bustamante A, Díaz-­Fernández B, Pagola J, Blanco-­Grau A, Rubiera M, DOI: 10.1159/00033​9426. Penalba A, García-­Berrocoso T, Montaner J Admission troponin-­I predicts 27. Hamzeh-­Cognasse H, Damien P, Chabert A, Pozzetto B, Cognasse F, subsequent cardiac complications and mortality in acute stroke patients. Garraud O Platelets and infections—­complex interactions with bacteria. Eur Stroke J. 2016;1:205–­212. DOI: 10.1177/23969​87316​654337. Front Immunol. 2015;6:1–­18. 15. Zapata-­Arriaza E, Mancha F, Bustamante A, Moniche F, Pardo-­Galiana 28. Petri B, Broermann A, Li H, Khandoga AG, Zarbock A, Krombach F, B, Serrano-­Gotarredona P, Navarro-­Herrero S, Pallisa E, Faura J, Vega-­ Goerge T, Schneider SW, Jones C, Nieswandt B, et al. Von Willebrand Salvatierra Á, et al. Biomarkers predictive value for early diagnosis of factor promotes leukocyte extravasation. Blood. 2010;116:4712–­4719. Stroke-­Associated Pneumonia. Ann Clin Transl Neurol. 2019;6:1882–­ DOI: 10.1182/blood​-­2010-­03-­276311. 1887.DOI: 10.1002/acn3.50849. 29. Badve MS, Zhou Z, Anderson CS, Hackett ML Effectiveness and 16. Adamcova M, Šimko F Multiplex biomarker approach to cardiovascular safety of antibiotics for preventing pneumonia and improving outcome diseases perspective. Acta Pharmacol Sin. 2018;39:1068–­1072. DOI: after acute stroke: systematic review and meta-­ analysis. J Stroke 10.1038/aps.2018.29. Cerebrovasc Dis. 2018;27:3137–­3147. DOI: 10.1016/j.jstro​kecer​ebrov​ 17. Mallick A, Januzzi JL Jr Biomarkers in acute heart failure. Rev Esp asdis.2018.07.001. Cardiol. 2015;68:514–­525. 30. Bustamante A, Vilar-­Bergua A, Guettier S, Sánchez-­Poblet J, García-­ 18. Boomsma F, Van Veldhuisen DJ, De Kam PJ, Man In ’T Veld AJ, Mosterd Berrocoso T, Giralt D, Fluri F, Topakian R, Worthmann H, Hug A, et A, Lie KI, Schalekamp MADH Plasma semicarbazide-­sensitive amine al. C-­reactive protein in the detection of post-­stroke infections: sys- oxidase is elevated in patients with congestive heart failure. Cardiovasc tematic review and individual participant data analysis. J Neurochem. Res. 1997;33:387–­391. DOI: 10.1016/S0008​-­6363(96)00209​-­X. 2017;141:305–­314. DOI: 10.1111/jnc.13973. 19. Boomsma F, De Kam PJ, Tjeerdsma G, Van Den Meiracker AH, Van 31. Fluri F, Morgenthaler NG, Mueller B, Christ-­Crain M, Katan M Copeptin, Veldhuisen DJ Plasma semicarbazide-­sensitive amine oxidase (SSAO) procalcitonin and routine inflammatory markers—­ predictors of in- is an independent prognostic marker for mortality in chronic heart fail- fection after stroke. PLoS One. 2012;7:e48309. DOI: 10.1371/journ​ ure. Eur Heart J. 2000;21:1859–­1863. DOI: 10.1053/euhj.2000.2176. al.pone.0048309. 20. de Boer RA, Nayor M, deFilippi CR, Enserro D, Bhambhani V, Kizer 32. Azurmendi L, Lapierre-­Fetaud V, Schneider J, Montaner J, Katan M, JR, Blaha MJ, Brouwers FP, Cushman M, Lima JAC, et al. Association Sanchez JC Proteomic discovery and verification of serum amyloid of cardiovascular biomarkers with incident heart failure with preserved A as a predictor marker of patients at risk of post-­stroke infection: and reduced ejection fraction. JAMA Cardiol. 2018;3:215–­ 224. DOI: a pilot study. Clin Proteomics. 2017;14:14–­ 27. DOI: 10.1186/s1201​ 10.1001/jamac​ardio.2017.4987. 4-­017-­9162-­0. J Am Heart Assoc. 2021;10:e018946. DOI: 10.1161/JAHA.120.0189468