Oxidative Stress Biomarkers as Prognostic Indicators of Severity in Patients With Dengue

There is evidence for the role of oxidative stress in severe dengue pathogenesis. However, previous observational studies presents certain methodological limitations, which may affect its internal and external validity. This study was a case-control analysis of patients with severe dengue and dengue with warning signs, to evaluate the serum protein carbonyls-PCOs and lipid hydroperoxides-LOOHs levels and activities of superoxide dismutases-SODs (MnSOD, Cu/ZnSOD and total SOD), as potential prognosis indicators of severity in dengue patients, using binary logistic regression analysis and strategy of double cross-validation. Therefore, the study population was subdivided into a derivation group (pediatric patients, Barranquilla-Colombia) and an external validation group (children and adults patients, National Institute of Health of Peru). PCOs was the only oxidative stress markers that showed a strongest association with the severity of dengue, both in children and adults. In the derivation group, the optimal cut-off point was estimated at 5.29 nmol/mg of protein, and in the external validation group, it was 5.77 nmol/mg of protein. The prognostic models based on these two diagnostic thresholds showed a high discriminatory capacity of dengue severity, external reproducibility, geographic transportability, and typical characteristics of diagnostic validity and safety of screening tests.


Introduction
The World Health Organization-WHO considers dengue one of the 17 neglected tropical diseases (Molyneux, 2013), present in more than 100 countries, with reports of 50 to 100 million new cases and 20,000 deaths each year (Zapata, Cox, & Salvato, 2014).
Since 2009, WHO has implemented a new dengue classification scheme, reporting cases such as dengue without warning signs (DwoWS), dengue with warning signs (DwWS), and severe dengue (SD) (World Health Organization, 2009).These warning signs have been proposed as prognostic criteria for severe dengue (Leo et al., 2013;World Health Organization, 2009).
Although the current case classification system (WHO-2009) overcomes, in part, the conceptual and practical limitations identified in the previous scheme (WHO-1997) (Tsai et al., 2013), this could allow arbitrary interpretations of the severity of bleeding by medical personnel as consequence of the use of less rigorous definition criteria, due to the absence of quantitative parameters and because of their independence from laboratory tests in the identification of severe cases (Halstead, 2013).characteristics.Moreover, with the availability of this type of prognostic indicators, more specific and timely medical care could be provided, which is essential to improve patients' living conditions.
At this point, it is interesting to examine the conceptual relationships between dengue pathogenesis, endothelial dysfunction and oxidative stress.The most characteristic feature of cases of severe dengue and the best indicator of severity is plasma loss.This alteration of vascular permeability, which is present during the defervescence phase (4-6 days of illness), results more from endothelial dysfunction than from structural destruction of endothelial cells (Dalrymple & Mackow, 2012).Currently, a possible association between endothelial dysfunction and the severity of dengue disease in children and adults has been reported (Yacoub et al., 2015).
Despite considering lipid hydroperoxides-LOOHs and protein carbonyls-PCOs, as the most stable or specific markers to evaluate lipoperoxidation (Girotti, 1998) and oxidative protein modification (Dalle-Donne et al., 2003), respectively, there are few studies that report the behavior of these products of lipid oxidation (Gil et al., 2012;Gil et al., 2004) and of protein oxidation (Soundravally et al., 2008;Soundravally1 et al., 2008) in dengue patients.
Regarding superoxide dismutase-SOD, there are four studies that report their levels in cases of dengue (Gil et al., 2012;Gil et al., 2004;Ray, Kumar, Kapoor, Dutta, & Batra, 1999).Only one of these reported concentrations of this antioxidant enzyme in relation to the three clinical presentations of dengue indicated by the WHO-1997 classification system (Ray et al., 1999).Although the importance of TNF-α in the pathogenesis of dengue is recognized (Pawitan, 2011) and its ability to selectively induce MnSOD mRNA expression both in vivo and in vitro (Wong & Goeddel, 1988), there is no knowledge of the behavior of serum levels of this metalloenzyme in patients with dengue.Neither are known reports related to the cytosolic Cu/ZnSOD isoform.
The aim of the study was to evaluate the serum PCOs and LOOHs concentrations and activities of SODs (MnSOD, Cu/ZnSOD and total SOD), during the critical phase of dengue, as potential prognostic markers of severity in dengue patients.

Cases and Controls
A study was carried out with case and control design, in which cases were defined as patients who presented: (I) fever between the fourth and sixth day of evolution, accompanied by two or more of the following symptoms: headache, retro-ocular pain, myalgias, arthralgias, generalized pain, rash, anorexia, nausea, vomiting, petechiae, mucosal bleeding and leukopenia; (II) positive result for NS1-DENV or IgM anti-DENV and (III) requirement for specialized care in intensive care unit-ICU, due to the presence of severe manifestations of the disease, following the WHO-2009 classification scheme (World Health Organization, 2009).
The criteria for severe dengue were: (a) severe plasma leakage leading to shock or fluid accumulation with respiratory distress.(b) severe bleeding, and (c) severe organ involvement (myocarditis, encephalitis, hepatitis, acute cholecystitis, acute kidney failure, etc.) (World Health Organization, 2009).
For the selection of controls, the patients were considered: (I) with fever between the fourth and sixth day of evolution, accompanied by two or more of the following symptoms: headache, retro-ocular pain, myalgias, arthralgias, rash , anorexia, sickness, vomiting, petechiae, mucosal bleeding, and leukopenia; (II) with a positive result for NS1-DENV or anti-dengue IgM; (III) with the presence of one or more of the warning signs described in the current case classification system [5] and (IV) that did not require specialized attention in the ICU, because they do not develop severe manifestations of dengue.
Patients with comorbidities such as: severe cardiovascular disease, disorders of the haematopoeitic organs, metabolic, lung, hepatic or kidney disorders, autoimmune disease, malnutrition, pregnancy, malignancy, or comorbid infectious diseases were excluded.

Derivation Group: pediatric patients (Barranquilla-Colombia)
The evaluation of the prognostic capacity of the oxidative stress biomarkers was carried out in 60 serum samples obtained from patients (20 cases and 40 controls), under 15 years of age, served in district hospitals of Barranquilla-Colombia, during the period comprised between January 2014 and January 2016.

External Validation Group: Pediatric and Adult Patients (Peru)
Similarly, for the external validation of the prognostic capacity of the biomarkers, 60 serum samples were used, obtained from patients (20 cases and 40 controls), of all ages, referred to the Laboratory of Viral Metaxenics of the National Institute of Health of Peru, during the period between January and February 2018.Serum samples of both cases and controls were stored at -80 °C until analysis.

Data Collection
We reviewed the medical records of the 60 pediatric patients, identified as a derivation group, in order to record their sociodemographic information (age, sex, and geographic origin) clinical data (days of fever onset, length of hospital stay, length of ICU stay, and symptomatology) and laboratory of interest (haemoglobin levels, hematocrit values, platelet counts, neutropenia, leukopenia, lymphopenia, monocytosis, limphocytosis, basophilia, neutrophilia, altered transaminases, BUN and creatinine levels).
In contrast, for the case of the 60 patients from the external validation group, the epidemiological records were reviewed for the registration of the variables of interest, such as: age, sex, geographic origin, days of fever onset, symptomatology, qualitative detection of NS1-DENV in serum, thrombocytopenia, and hematocrit rise 20% or more.

DENV NS1 Antigen Capture Test (Derivation Group)
For the qualitative detection of NS1-DENV in the serum of the pediatric patients of the derivation group, a one-step immunochromatographic test (SD Bioline Dengue NS1 Ag, Kyong, South Korea®) was carried out, under the conditions recommended by the manufacturer.

Anti-DENV IgM and IgG ELISA (Derivation Group)
The detection of anti-DENV IgM and IgG in the serum of the pediatric patients of the derivation group was carried out by means of two enzyme-linked immunosorbent assays (ELISA, Vircell, Spain, References M1018 and G1018, respectively), under the conditions specified by the commercial house.
As serological evidence suggestive of secondary dengue infection, the IgM/IgG ratio <1.8 was used (Alera et al., 2016).Absorbance was measured at 450 nm (reference filter 620 nm) using a microplate reader (Stat Fax® 303 / PLUS, Awareness Technology, Inc. Florida, United States).

Anti-DENV IgM and IgG ELISA (External Validation Group)
The detection of anti-DENV IgM and IgG in the serum of the patients of the external validation group was carried out by means of two capture enzyme-linked immunosorbent assays (TARIKI-Dengue IgM, TARIKI-Dengue IgG, INS-Peru), under the conditions specified by the National Center of Biological Products of the National Institute of Health of Peru.Absorbance was measured at 450 nm (reference filter 620 nm) using a microplate reader (PHOmo, Autobio Co., LTD).

Viral RNA Extraction (External Validation Group)
For the automatic and simultaneous extraction and purification of viral RNA, the magnetic particle technology of the QIAsymphony® Virus / Bacteria Mini Kit (Quiagen, Crawley, UK) was used from the serum samples of the patients from the external validation group in combination with the QIAsymphony® SP instrument, following the manufacturer's instructions, before real-time reverse transcription polymerase chain reaction (RT-PCR) analysis, since magnetic particle technology allows the purification of high quality nucleic acids that they lack proteins, nucleases and other impurities.

Reverse Transcription-PCR for Serotyping of DENVs (External Validation Group)
Briefly, 10.0 μL of extracted viral RNA was used as template for the identification of DENV serotype, using the protocol described by Lanciotti, et al. (Lanciotti, Calisher, Gubler, Chang, & Vorndam, 1992), with the oligonucleotides described in Table S1.Reverse transcription of 10 min at 50 °C was followed by 45 cycles of amplification in a Rotor-Gene Q real time PCR cycler (Qiagen).

Serum Protein Carbonyls Estimation
Materials and reagents of the Protein Carbonyl Colorimetric Assay Kit (catalog No. 10005020; Cayman Chemical Company, Ann Arbor, MI) were used for quantitative determination of PCOs in serum samples, using the alkaline method, which consists in a neutralization with 6 M NaOH subsequent to the step of derivatization with 2,4-DNPH to change the absorbance of carbonyl conjugated hydrazone at 450 nm (Mesquita et al., 2014).
Briefly, a mixture was made from equal volumes between 2,4-DNPH and serum sample (working dilution, 1:10).As a blank, an equal-volume mixture of 2,4-DNPH and phosphate buffered saline (pH = 7.2) was used.After 10 minutes of incubation at room temperature, in the dark, 200 μL of 6 M NaOH was added.Then, it was incubated for 10 minutes at room temperature (Mesquita et al., 2014).Absorbance was measured at 450 nm (reference filter 620 nm) using a microplate reader (PHOmo, Autobio Co., LTD).
The amount of serum protein was calculated from a bovine serum albumin (BSA) standard curve (0.2-2.0 mg/mL) dissolved in guanidine hydrochloride and read at 280 nm.

Serum Manganese Superoxide Dismutase, Cu/Zn Superoxide Dismutase, and Total Superoxide Dismutase Estimations
Materials and reagents of the Superoxide Dismutase Assay Kit (catalog No. 706002, Cayman Chemical Company, Ann Arbor, MI) were used for determination of SODs activities.The addition of potassium cyanide-KCN (3 mM) to the assay allowed the inhibition of both Cu/ZnSOD and extracellular SOD, in order to detect only mitochondrial superoxide dismutase-MnSOD activity.
Briefly, in the preparation of the wells for the SOD standards, 200 μL of the diluted radical detector and 10 μL of one of the seven SOD standards available in the kit (0-0.25 U / mL) were added.For the preparation of the wells for the samples, 190 μL of the diluted radical detector, 10 μL of 3 mM KCN and 10 μL of the patient's serum (working dilution 1: 5) were added.To determine the enzymatic activity of total SOD, 200 μL of the diluted radical detector and 10 μL of the patient's serum (working dilution, 1: 5) were added.The reaction was started with the addition of 20 μL of diluted xanthine oxidase.To the blank, 20 μL of the sample buffer available in the kit was added.After shaking for a few seconds, it was incubated for 20 minutes at room temperature.
The total enzymatic activity of SOD could be determined in absence of inhibitor and with the difference between activity values of total SOD and MnSOD, the enzymatic activity of the cytosolic Cu/ZnSOD isoform could be estimated.Absorbance was measured at 450 nm (reference filter 620 nm) using a microplate reader (PHOmo, Autobio Co., LTD).The interassay coefficient of variation was 4.6%, obtained with the calibrators available in the kit.

Serum Lipid Hydroperoxides Estimation
Serum LOOHs levels were measured by colorimetric assays using a commercially available kit according to manufacturer´s instructions (Lipid Hydroperoxyde Assay Kit, catalog no.705003; Cayman Chemical Company, Ann Arbor, MI).Initially, the serum samples (working dilution, 1:20) were extracted with ethanol and chloroform.Then, 300 μL of the sample extract, 300 μL of the chromogen and 300 μL of the ethanol-chloroform mixture were added.A calibration curve was made with the eight standards of LOOHs (13-hydroperoxyoctadecadienoic acid ethanolic solution, 13-HpODE, 0.0-5.0μM) available in the kit.Absorbance was measured at 492 nm (reference filter 620 nm) using a microplate reader (PHOmo, Autobio Co., LTD).The interassay coefficient of variation was 5.3%, obtained with the calibrators available in the kit.

Statistical Analysis
Normality assessed with the Shapiro-Wilks test, one of the tests of goodness of fit with greater statistical power (Yap & Sim, 2011).Numerical variables are expressed as the mean and standard deviation, when it presented normal distribution and the median and lower and upper quartiles when presented non-normal distribution.
Statistical associations between research variables and the severity of dengue were identified with Fisher's exact test, accompanied by the respective effect size.Comparisons between two groups for serum biomarker concentrations were performed using Student´s t-test of independent samples or the Mann-Whitney U test.Also, the respective effect size was estimated (Cohen'd or Rosenthal´s correlation coefficient, respectively) (Gaskin & Happell, 2014).
Receiver curve (ROC) analysis was performed to find sensitivity and specificity for every potential cut-off values of each biomarker.To identify the optimum cut-off value, maximum value of the Youden Index (J) was used.Within the derivation group (20 cases: SD and 40 controls: DwWS), binary logistic regression was used to estimate the effect of prognostic variables or factors (dichotomised biomarker) on occurrence of severe dengue, with variable entered stepwise using the forward conditional likelihood ratio with p-in≤0.05and p-out≥0.10.
We tested the goodness-of-fit for final models using Hosmer-Lemeshow test.A bootstrap method was used to internal validation of prognostic models.Each bootstrap sample was obtained by randomly sampling 10 000 times from the data sets.The prognostic model was refitted and tested on the original sample to obtain an estimate of predictive accuracy corrected for bias (overoptimism).Calibration and discrimination of the models were assessed with the ROA's efficient score statistic and with c-statistic, respectively (Nuñez, 2011).The criteria of Swets (Swets, 1988) were used to assess the accuracy of the sensitivity and specificity values.The AUCs of the different prognostic models were compared using the statistical test of Hanley and McNei (Hanley & McNeil, 1983).
All p-values estimated in the hypothesis tests were interpreted using the criteria of Sterne and Smith (Sterne & Davey Smith, 2001).However, to establish statistically significant differences between two independent groups, 95 percentage confidence intervals of the difference (CI95% diff ) were interpreted between independent proportions, means or medians.
Final prognostic model was validated in two sets of serum samples obtained of dengue patients (20 cases: SD and 40 controls: DwWS) notified to the National Institute of Health of Peru (geographical external validation).Within the external validation group, calibration and discrimination measures of final prognostic model were estimated using the same methods of statistical inference.Finally, the double cross-validation strategy was also used (Osborne, 2000).

Ethical Statements
The present study was reviewed and approved by the IPS Universitaria Barranquilla-Colombia (Office of approval 10-02-2016-IPS Universitaria-Barranquilla) and by the National Institute of Health of Peru (Office of approval 535-2017-OGITT-OPE / INS).

Clinical Characteristics of Derivation Group
All the pediatric patients belonging to this group came from the city of Barranquilla-Colombia and presented the same median for age (11 years; IQR: 10 to 13), both in the cases and in the controls.Despite a predominance of males (63.3%, CI95% 50.7% to 74.4%) among derivation group, we found no evidence of statistical association or significant difference between sex and the severity of dengue (p=0.573;CI95% diff -15.8% to 31.8%).
The median number of days after the onset of fever in cases was similar to that in controls (5 days vs. 4 days; p=0.192;CI95% diff -1.0 to 0.0).In contrast, the median length of hospital stay significantly differed between cases and controls (5 days vs. 3 days, respectively; CI95% diff 1.0 to 3.0).At the same time, we found a strong statistical association between the severity of dengue and the number of days of hospital stay (Rosenthal´s correlation coefficient=0.6;p<0.0001).

Clinical Characteristics of External Validation Group
In this group, the median for the age of the cases was comparable to that in controls (20 years vs. 16 years, respectively; p=0.347;CI95% diff -10.0 to 4.0).Moreover, there was also no statistically significant difference in the distribution of age groups among cases and controls (≤15 years of age vs. >15 years of age: CI95% diff -6.4% to 41.4%; p=0.174).Likewise, we found no evidence of statistical association or significant difference between sex and the severity of dengue (p=0.176;CI95% diff -6.3% to 42.7%).
Although a moderate statistical association was observed between the presence of severe dengue and the number of days of illness (Rosenthal´s correlation coefficient=0.3;p=0.031), a statistically significant difference could not be found when comparing the cases and controls (5 days vs. 4 days, respectively; CI95% diff -1.0 to 0.0).
The only symptoms that showed a strong statistical association with the severity of dengue were: asthenia (p=0.004;OR=16.7;CI95% 2.3 to 113.1) and lethargy (p=0.0004;OR=15.6;CI95% 2.9 to 82.8), however, the presence of respiratory distress showed a moderate statistical association with severe dengue (p=0.042;OR=3.4;CI95% 1.1 to 10.7).The rest of the variables evaluated in the external validation group are presented in Table 2.
In line with the aforementioned, we found a strong statistical association between the occurrence of severe dengue and serum PCOs concentrations (non-standardized values: Rosenthal´s correlation coefficient=0.5;p=0.0001; protein-standardized values: Rosenthal´s correlation coefficient=0.5;p=0.0002) (Figures 1a and 1b).The average MnSOD activity (±SD) in the sera of patients with severe dengue and patients with dengue with warning signs were 1.39 ± 0.05 U/mL and 1.41 ± 0.10 U/mL, respectively.No statistically significant difference could be demonstrated between these two values of enzymatic activity (CI95% diff -0.02 to 0.06, t=0.988, df=58; p=0.327) (See Supplementary Figure S1a).Studied cases had a mean concentrations of 3.21 ± 0.05 U of total SOD/mL and 1.82 ± 0.07 U of Cu/ZnSOD/mL.In the controls, serum levels of total SOD of 3.18 ± 0.05 U/mL and 1.78 ± 0.10 U of Cu/ZnSOD/mL were found.
We obtained evidence of a moderate to strong statistical association of occurrence of severe dengue with the enzyme activity of total SOD (t=1.989;df=39.83;p=0.054,Cohen's d=0.7) and evidence of a weak to moderate statistical association with the enzymatic activity of Cu/ZnSOD (t=1.996;df=52.96;p=0.051;Cohen's d=0.4), accompanied with scarce absence of statistically significant difference (total SOD-cases vs controls: CI95% diff -0.0004 to 0.05 U/mL; Cu/ZnSOD-cases vs. controls: CI95% diff -0.0002 to 0.09 U/mL) (See Supplementary Figures S1b and S1c).

Serum Lipid Hydroperoxides Concentrations (Derivation Group)
The mean concentrations of LOOHs of 61.20 ± 6.88 μM and 59.79 ± 7.08 μM were found in children with severe dengue and in children with dengue with warning signs, respectively.Neither, statistical association was found between serum LOOHs concentrations with the occurrence of severe dengue (t=-0.735;df=58; p=0.465) or statistically significant difference when comparing these serum values (cases vs. controls: CI95% diff -5.26 to 2.43 μM) (See Supplementary Figure S2).

Serum Protein Carbonyls Concentrations (External Validation Group)
In cases and controls, the medians serum concentration for PCOs were 20.02 nmol/mL (IQR: 18.34 to 21.59) and 13.62 nmol/mL (IQR: 12.11 to 16.26), respectively.The concentration of protein-standardized in patients with severe dengue was 6.95 nmol/mg of protein (IQR: 6.63 to 7.56) and 4.47 nmol/mg of protein (IQR: 4.00 to 5.19) for patients with dengue with warning signs.
The analysis showed that serum PCOs concentrations of the cases were significantly higher than control group (non-standardized values: CI95% diff 3.42 to 7.34 nmol/mL; protein-standardized values: CI95% diff 1.59 to 2.84 nmol/mg of protein).Again, we found a strong statistical association between the occurrence of severe dengue and serum PCOs concentrations (non-standardized values: Rosenthal´s correlation coefficient=0.6;p<0.0001; protein-standardized values: Rosenthal´s correlation coefficient=0.6;p<0.0001) (Figures 2a and 2b).

Logistic Regression Analysis of Oxidative Stress Biomarkers (Derivation Group)
PCOs was the only biomarker with prognostic capacity for severe dengue among pediatric patients.Table 3 shows the results of the binary logistic regression using the two optimal cut-off values estimated for serum PCOs concentrations, including bootstrapping results.By contrast, the remaining biomarkers did not meet inclusion criteria of a variable in the prognostic model (MnSOD: p=0.216; total SOD: p=0.448;Cu/ZnSOD: p=0.188;LOOHs: p=0.643).
A cut-off value for protein-standardized serum PCOs concentrations of 5.29 nmol/mg of protein (PM-2: dichotomized predictor) was the best in identifying pediatric patients with severe dengue, in comparison with continuous predictor (90.0% vs. 55.0%,respectively; CI95% diff 7.2% a 57.0%).
Table 3. Binary logistic regression models to predict the probability of severe dengue in pediatric patients, using optimal cut-off points of serum concentrations of protein carbonyls.PCOs: protein carbonyls; df: degree of freedom.

Calibration and Discrimination of Prognostic Model (Derivation Group)
Approximately, 47.2% of the variance of the criterion variable (severity of dengue) can be explained by PM-2 (Nagelkerke pseudo-R 2 =0.472) with a c-statistic of 0.825 (CI95% 0.729 to 0.921; p<0.0001).
The predicted probability of severe dengue in a pediatric patient (≤15 years) with serum PCOs concentrations greater than or equal to 5.29 nmol/mg of protein was 0.643 (CI95% 0.454 to 0.796), whereas the predicted probability of severe dengue in a pediatric patient with serum levels below the optimal threshold was estimated at 0.062 (CI95% 0.016 to 0.218; Chi-square test/omnibus test: X 2 =24.92; df=1; p<0.0001).

Evaluation of Transportability of Prognostic Model (External Validation Group)
In applying of PM-2 in external validation group, an AUC of 83.8% (CI95% 74.4% to 93.2%; p<0.0001) was calculated.The predicted probability of severe dengue in a patient with serum PCOs concentrations greater than or equal to 5.29 nmol/mg of protein was 0.667 (CI95% 0.473 to 0.817), whereas the predicted probability of severe dengue in a patient with serum PCOs concentrations below the optimal threshold was estimated at 0.061 (CI95% 0.015 to 0.212; Chi-square test/omnibus test: X 2 =26.92; df=1; p<0.0001).
Moreover, these probability values are similar to those estimated among pediatric patients (derivation group) (≥5.29 nmol of PCOs/mg of protein: CI95% diff -0.253 to 0.296; <5.29 nmol of PCOs/mg of protein: CI95% diff -0.126 to 0.128).No evidence of statistical association was found between the discriminative capacity of PM-2 with the two age groups evaluated in the external validation group (≤15 years of age vs. >15 years of age: 92.5% vs. 77.9%, respectively; Chi-square test: X 2 =3.16; df=1; p=0.076).
Table 4 shows the results of the binary logistic regression using the estimated optimal diagnosis point (5.77 nmol of PCOs/mg of protein), including bootstrapping results.Approximately, 67.4% of the variance of the criterion variable (severity of dengue) can be explained by PM-3 (Nagelkerke pseudo-R 2 =0.674) with c-statistic=0.900(CI95% 0.806 to 0.994; p<0.0001).In applying of PM-3 among pediatric patients from Barranquilla-Colombia (external validation group), an AUC of 82.5% (CI95% 72.3% to 98.7%; p<0.0001) was calculated.The predicted probability of severe dengue in a patient with serum levels greater than or equal to 5.77 nmol/mg of protein was 0.680 (CI95% 0.478 to 0.831), whereas the predicted probability of severe dengue in a patient with lower serum levels to the optimum threshold was estimated at 0.086 (CI95% 0.028 to 0.234).These probability values are similar to those estimated in the derivation group (≥5.77 nmol of PCOs/mg of protein: CI95% diff -0.166 to 0.349; <5.77 nmol of PCOs/mg of protein: CI95% diff -0.100 to 0.154).
Table 5 shows the results obtained from the comparison of the discriminative capacity of the prognostic models, based on the optimal diagnostic points, estimated in the derivation group and transported to the validation group, by the double cross-validation strategy (prospective and retrospective).

Discussion
This study continues our examination of biomarkers of oxidative stress as prognostic tool for dengue severity, using regression models for predicting medical outcomes.
Serum levels of PCOs were significantly higher in patients with severe dengue than in patients with dengue with warning signs, in both the derivation group and validation group.The above shows coherence with the evidence published in patients with DENV infection and healthy adults, during a DENV-3 outbreak (Rajendiran et al., 2008;Soundravally, Sankar, Hoti, et al., 2008).Although median serum PCOs concentrations estimated in this study cannot be compared with other reports (Rajendiran et al., 2008;Santhosh, 2016;Soundravally, Sankar, Hoti, et al., 2008), due to the different measures of central tendency and dispersion used, we suggest that the most severe cases of dengue present a higher concentration of irreversibly oxidized proteins in comparison to non-severe cases.
Recently, Cherupanakkal, et al. (Cherupanakkal et al., 2017), reported that MnSOD gene expression was not associated with severity of dengue disease in children and adults.In addition to the above, we have not found statistically significant differences in MnSOD activity between patients with severe dengue and those with dengue with warning signs, in both adults and children.All in all, the above evidences allow us to suggest that the antioxidant activity of MnSOD does not participate in the pathogenesis of severe dengue.
In addition to the explanations reported by Cherupanakkal, et al. (Cherupanakkal et al., 2017), it is necessary to consider the presence of IgM autoantibodies against MnSOD, which could inhibit the protective effects of MnSOD.These antibodies have been reported in patients with acute viral infections (Ritter et al., 1994;Semrau et al., 1998), and it has been shown that may contribute to the oxidative damage of endothelial cells (Dalpke et al., 2003).Likewise, no significant differences were found in total SOD and Cu/ZnSOD activities between cases and controls.In this regard, elevated levels of TNF-α and IFN-γ in dengue-infected patients could decrease the gene expression of antioxidant enzymes, including SOD (Ho et al., 2001).
As regards serum LOOHs levels, no statistically significant difference was found between cases and controls, in both the derivation and validation groups.It is not possible to compare our results with other studies (Gil et al., 2012;Gil et al., 2004), due some important methodological and analytical differences, such as WHO dengue case classification system, extraction procedure, and ferric ions indicator.
Here, we would like to point out that discordance has also been found in the correlation between LOOHs and PCOs concentrations in virus-infected insect cell culture (Wang, Oberley, & Murhammer, 2001) and in blood after exhaustive aerobic and nonaerobic isometric exercises (Alessio et al., 2000) .This discrepancy can be explained by certain pre-analytical errors, effects of small size sample, and the differences between the reactions involved in formation of these oxidative stress biomarkers (Dotan, Lichtenberg, & Pinchuk, 2004).
Prognosis cut-off point of 5.29 nmol of PCOs/mg of protein showed high sensitivity and negative predictive value.
For screening purposes, these characteristics are more important than a high specificity and positive predictive value, especially in the context of serious and treatable diseases (Bravo-Grau & Cruz Q, 2015).
Data were analyzed using four statistical software: IBM® SPSS® Statistics version 23.0 (IBM Corp, Armonk, New York), GraphPad PRISM® version 6.01 (GraphPad software, San Diego, CA), EpiDat 3.1® (General Directorate of Innovation and Public Health Management, Xunta de Galicia, Spain, Pan American Health Organization and Higher Institute of Medical Sciences of Havana) and Stata® version 14.2 (Stata Corporation, College Station, TX, USA).

Figure 2 .
Figure 2. Serum protein carbonyls concentrations in patients with severe dengue and dengue with warning signs (DwWS).External validation group, Peru, 2018.(a): non-standardized serum concentrations; (b): protein-standardized serum concentrations.Errors bars represent median and interquartile range

Figure 3 .
Figure 3. Receiver-operating characteristics (ROC) curve of non-standardized serum concentrations of protein carbonyls (PCOs) in pediatric patients with severe dengue and dengue patients with warning signs.Derivation group, Barranquilla-Colombia, 2014-2016

Figure 5 .
Figure 5. Receiver-operating characteristics (ROC) curve of serum concentrations of protein carbonyls (PCOs) in patients with severe dengue and dengue patients with warning signs.Retrospective validation.Derivation group, Peru, 2018

Table 1 .
Clinical and laboratory characteristics of pediatric patients with severe dengue and dengue with warning signs.Derivation group,Barranquilla-Colombia, 2014-2016 * estimated by Fisher´s exact test; DwWS: Dengue with warning signs; NE: not estimated; DLC: differential leukocytes counts; HCT: hematocrit; BUN: blood urea nitrogen; GPT: glutamic-piruvic transaminase; GOT: glutamic-oxaloacetic transaminase Table2.Clinical and laboratory characteristics of pediatric patients with severe dengue and dengue with warning signs, external validation group, Peru, 2018

Table 4 .
Binary logistic regression models to predict the probability of severe dengue in patients, using optimal cut-off points of serum concentrations of protein carbonyls.External validation group, Peru, 2018

Table 5 .
Comparison of discriminatory capacity of binary logistic regression models to predict the probability of severe dengue in patients, using optimal cut-off points of adjusted serum concentrations of protein carbonyls.Prospective and retrospective validation PCOs: protein carbonyls; df: degree of freedom; X 2 : Chi-square statistic; AUC: area under the curve; PM: prognostic model.