Associated Factors and Utility of Bed-side Assessment Tools to Avoid Antibiotic Overuse in an Urban ICU of a Developing Country

Antibiotic exposure in the intensive care unit (ICU) is very high, although 50% of all antibiotics may be unnecessary. We aimed to determine predicting factors, outcomes, and the utility of simple screening tools to avoid antibiotic overuse in the ICU. We analyzed 510 young children who did not receive antibiotics during ICU stay to those treated with antibiotics. The logistic regression analysis revealed that cases were more often older and independently associated with hypernatremia. Cases less often had severe underweight, altered mentation, age-specic fast breathing, lower chest wall in-drawing, adventitious sound on lung auscultation, abdominal distension, developmental delay, hyponatremia, hypocalcemia, and microscopic evidence of invasive diarrhea (for all, p<0.05). The case-fatality rate was signicantly higher among the cases than the controls. For predicting 'no antibiotic approach,' the sensitivity of a negative quick Sequential Organ Failure Assessment (qSOFA) was similar to quick Pediatric Logistic Organ Dysfunction-2 (qPELOD-2) and higher than Systemic Inammatory Response Syndrome (SIRS). In conclusion, 'No antibiotic approach' could be safely adopted in PICU using some simple clinical and laboratory characteristics, particularly in poor resource settings. A negative qSOFA or qPELOD-2 score calculated during PICU admission is superior to SIRS to avoid antibiotic overuse in under-ve children.


Introduction
Appropriate use of antimicrobial agents is the cornerstone of effective antimicrobial stewardship programs, and has become a focus of patient safety, quality assurance, and health care outcomes. The Intensive Care Unit (ICU) deals with critically ill patients, and infections are among the most frequent causes of hospitalization among patients in the ICU. Thus antibiotic burden in ICU is very high; 66 to 77% of all ICU patients and 84-100% with an ICU stay of more than 48 hours are exposed to at least one class of broad-spectrum antibiotic 1,2 . As the de nitive diagnosis of infection is often quite hard, Pediatric ICU (PICU) physicians are concerned about delayed treatment; consequently early initiation of broadspectrum antibiotic is expected especially in the PICU settings 3,4 .
However, 50% of antibiotic prescriptions were inappropriate and could be avoided 5,6 . The situation is even worse in Low middle income countries (LMICs), where prescribing practices are poor, and antibiotic stewardship programs are frequently non-existent 7 . For instance, a study of antibiotic use in PICU showed that all children received antibiotics as prophylaxis (43%) or empirical (42%) therapy without having evidence of any infection, and 76% of children received two or more classes of antibiotics 8 . This inappropriate exposure upsurges the emergence of antimicrobial resistance leading to prolonged hospital stays and increases mortality as well as healthcare costs [9][10][11] .
Appropriate antibiotic prescribing is a complex decision-making process requiring the integration of epidemiological, clinical and microbiological knowledge and expertise in these elds. Improving diagnostic tools, and discriminate infections from non-infectious mimics, is the eventual solution to reducing unnecessary antibiotic initiation in the ICU. However, bacterial isolation by culture and sensitivity, the gold standard test, is time-consuming and has low sensitivity for slow-growing and fastidious microorganisms 12 . Although molecular diagnostic assays allow faster viral recognition, more rapid bacterial identi cation, and determination of antimicrobial susceptibilities [13][14][15] , these are expensive and not always available in LMIC.
Several scoring systems have been developed to predict suspected infection or sepsis as a surveillance tool, and a positive score might be of value to identify infection promptly. Historically, the systemic in ammatory response syndrome (SIRS) criteria were considered to be fundamental to the diagnosis of in ammation and infection 16 , although, it might not work in children with dehydration especially with diarrhea 17 . Subsequently the quick Sequential Organ Failure Assessment (qSOFA) score (altered mentation, systematic hypotension and tachypnea) 18, 19 , and quick Pediatric Logistic Organ Dysfunction Score-2 (qPELOD-2) (altered mentation, systematic hypotension and tachycardia) 20,21 have been recommended as a handy bedside tool to promptly recognize patients with infection who might be at risk of poorer outcomes, and could be predominantly useful in assessing critically sick patients. However, the utility of negative scores has not been validated to identify suspected non-bacterial infectious patients at ICU who would bene t from the watchful waiting approach.
Informative studies regarding whether clinicians could adopt a 'watchful waiting' and no antibiotic approach safely and con dently for a subset of PICU patients are still lacking. ICU physicians seldom consider a watchful waiting approach; in contrast, they prefer to overuse broad-spectrum antibiotics when faced with a non-reassuring characteristic 22 . Thus, objective and evidence-based criteria are crucial to avoid unnecessary antibiotic exposure and, hence, tribulations. Therefore, we aim to identify the predictive factors and outcome of implementing a no antibiotic approach at PICU to prevent antibiotic overuse. We also evaluate the accuracy of negative SIRS, qSOFA, and qPELOD-2 scores in predicting no antibiotic or watchful waiting approach among young children presenting at the PICU with critical illness so that physicians can safely and con dently evade unnecessary antibiotic administration.

Study site
We have conducted this study at the Dhaka Hospital of the International Centre for Diarrhoeal Disease Research, Bangladesh (icddr,b), located in Dhaka, Bangladesh's capital city. This hospital provides care for approximately 150,000 patients annually. Among them, sixty-two percent are aged below ve years. Patients presenting with diarrhea with or without other comorbidities can seek health care at this hospital. The emergency department has a well-organized triage for sorting and prioritizing patients for care and initiating appropriate therapeutic measures. More than 90% of patients are managed in the short-stay ward, where the median duration of stay is 18 hours. Only 1-2 % of individuals needing immediate resuscitation at triage or patients with a critical illness or those with clinical deterioration in the short-stay ward or longer-stay ward are transferred to the ICU. The median duration of stay at the ICU is ve days.

Study population and design
We conducted a retrospective observational study with an unmatched case-control design. Children aged 2-59 months admitted to the ICU of Dhaka Hospital with suspected critical illness between January 2017 and January 2020 were eligible for enrolment. We retrieved data through the hospital electronic database of the patient information system. Patients were considered to have critical illness if they met ICU transfer criteria following hospital guidelines (S1 File) and stayed in ICU for more than 24 hours. The cases were children who did not receive an antibiotic during their ICU stay and were discharged without antibiotic therapy. Children who received oral or parenteral antibiotics at triage or following ward transfer before admission in the ICU or during discharge were excluded.
In contrast, controls were children who received one or more intravenous antibiotic therapy during their stay at ICU. We assumed 85% exposure of acute watery diarrhea (AWD) in controls to compare the percentage of cases and controls, considering the high prevalence of AWD among children with high infection burden countries 46 . To provide 80% power at a 5% level of signi cance (two-sided) and desired odds ratio (OR) of 2, we aimed to enroll 164 cases and 346 controls. Controls were selected using computer-generated random number sequences using SPSS version 20.0 for windows. The database identi ed 1821 controls, and 1:2 unmatched case-control ratios were used to increase the analyses' statistical power.

Measurement
We developed pretested case report form for the acquisition of relevant data. We reviewed all medical records, including the initial presentation to the triage, inpatient course, and outcome. Data on demographics (age, gender, breastfeeding, immunization as per EPI schedule), vital signs (temperature, heart rate, respiratory rate, mean arterial pressure, and SpO2), anthropometric measurements, clinical features (duration and consistency of diarrhea, duration of vomiting, dehydration status, documented seizure, altered mentation, developmental delay, congenital heart disease) and outcome were collected.
We also examined laboratory results, including complete blood count, hypernatremia, hyponatremia, hyperkalemia, hypokalemia, metabolic acidosis, hypocalcemia, microscopic evidence of invasive diarrhea, UTI, and CNS infection, as well as bacterial pathogens from blood and stool culture those helped for no antibiotic approach.

Scoring systems
We calculated SIRS, qSOFA, and qPELOD-2 scores based on the rst measured values after ICU entry. SIRS 17 was de ned as tachycardia, age-speci c fast breathing, temperature abnormality (axillary temperature >38°C or <36°C ), and white blood cell abnormality (>12000/mm3 or <4000/mm3 or bandemia >10%) in the absence of dehydration. SIRS score must include either temperature or white blood cell abnormality. qSOFA 47 criteria were: hypotension, age-speci c fast breathing, and altered mentation. qPELOD-2 20 criteria were: tachycardia, hypotension, and altered mentation. A threshold of fewer than two scores was applied to indicate a negative result for every score. We de ned altered mentation as having drowsiness, disorientation, confusion or coma. We de ned age-speci c fast breathing if the respiratory rate was >50 breaths/min for 2-11 months and >40 breaths/min for 12-59 months; tachycardia if the heart rate was ≥160 beats/min for 2-11 months, ≥140 beats/min for 12-59 months; and hypotension if mean arterial pressure <50 mm Hg.

Management
All children admitted to ICU were initially managed by trained ICU physicians following hospital guidelines. All children received standard treatment, including noninvasive or invasive oxygen therapy, intravenous uid, suitable antibiotics (if necessary), antiseizure therapy for seizure, and other supportive management. We adopted a no antibiotic approach among the children who did not have severe acute malnutrition, sepsis 48 , severe pneumonia (following WHO classi cation). During the initial clinical assessment, we evaluated history, physical examination and simultaneously quick laboratory tests (complete blood count, stool for microscopic examination in diarrheal children, microscopic examination of urine and study of cerebrospinal uid in children having a seizure) to identify any focus of bacterial infection. Attending physicians evaluated patients at least every 8 hours, and ICU consultants made clinical rounds at least twice a day. Antibiotics were added if the patient's condition deteriorated clinically, as discerned by the attending physician.

De nition
We de ned severe stunting with a length or height for age z-score (LAZ/HAZ) -3, severe underweight with weight for age z-score (WAZ) <-3 but ≥-4 ensuing WHO growth standards 49 . We de ned severe acute malnutrition (SAM) with a weight for length or height z score (WHZ) <-3 or weight for age z score (WAZ) <-4 of the median of the WHO anthropometry or presence of nutritional edema 50 . We de ned hypernatremia and hyponatremia if serum sodium concentration was >150.0 mmol/L and < 130.0 mmol/L; hyperkalemia and hypokalemia if serum potassium concentration was > 5.5 mmol/L and <3.5 mmol/L; metabolic acidosis if serum TCO2 was < 17.0 mmol/L, hypocalcemia if serum calcium was < 2.12 mmol/L.

Statistical analysis
We analyzed data using the STATA for Windows (SAS Institute Inc.®, USA). We assessed the normality of distribution with the Shapiro-Wilk test. Continuous variables were expressed as mean ± SD or median (IQR), categorical variables as frequency and percentages. When comparing the no antibiotic and antibiotic groups, Student's t-test or Mann-Whitney's U test, as appropriate, was used to analyze the continuous variables. The Chi-square test or Fisher's exact test was used to compare differences in proportions. We applied multiple logistic regression analysis to identify the no antibiotic approach's determinants and the evolution of variables having a p < 0.05 at the bivariate analysis; OR were adjusted, and 95% con dence intervals were calculated.
The receiver operating characteristics (ROC) curve was created for each score. We determined each scoring system's predictive accuracy using the AUROC (area under the receiver operating characteristics curve). We calculated sensitivity, speci city, positive and negative predictive values using contingency tables for every score. We compared the sensitivity and speci city of negative SIRS, qSOFA, and qPELOD-2 scores for no antibiotic approach using McNemar's test and the AUROC using DeLong's method 51 . A pvalue of < 0.05 was considered statistically signi cant.

Ethical Statement
The data used in this study were retrieved from the electronic database of hospital patients' records of the Dhaka Hospital of icddr, b. Data were recorded in an anonymized way before analysis and used to improve the quality of care of the hospital patients. However, the methods were carried out following the hospital guidelines and regulations. As we used de-identi ed data for this retrospective chart analysis, informed consent was not necessary. The Research Review and Ethical Review Committee of icddr, b waived the ethical approval for this de-identi ed data analysis and publication.

Results
In total, 2089 children aged 2 months to 59 months were admitted to the ICU of Dhaka Hospital between 2017 to 2020, and 268 (12.8%) children did not receive any antibiotic. We excluded 104 children because they were treated with antibiotics at emergency or after transferred to another ward. A nal cohort of 510 children formed the analyzable data, 164 were cases, and 346 were controls. The mortality rate was signi cantly lower amongst the cases compared to controls ( Table 1). The cases were more often older, presented with acute watery diarrhea, history of vomiting, seizure, and hypernatremia than the controls. Table 1. Sociodemographic and clinical characteristics of critically ill ICU children aged below ve years without (cases) and with antibiotic therapy (controls) in Bangladesh.
In contrast, cases less often presented with SAM, severe underweight, severe stunting, dehydration, agespeci c fast breathing, lower chest wall in-drawing, hypoxemia, adventitious sound on lung auscultation, abdominal distension, altered mentation, developmental delay, and congenital heart disease (Table 1).
Similarly, Hyponatremia, hypokalemia, hyperkalemia, hypocalcemia, leukocytosis, and microscopic evidence of invasive diarrhea was less frequent among cases compared to their corresponding peers. Laboratory evidence of CNS infection and bacteremia were nil among cases. Similarly, bacterial pathogens isolated from stool culture were less frequent among cases compared to controls ( Table 2). Using logistic regression models adjusted for potential confounders, the relative odds of no antibiotic approach increased in older children and the presence of hypernatremia. However, the relative odds decreased in the presence of severe underweight, altered mentation, age-speci c fast breathing, lower chest wall indrawing, adventitious sound on lung auscultation, abdominal distension, developmental delay, hyponatremia, hypocalcemia, and microscopic evidence of invasive diarrhea (Table 3). Of the included children, the frequency of SIRS, qSOFA, and qPELOD-2 negativity was 311/510 (61%), 452/510 (89%), and 475/510 (91%), respectively. Cases more often had SIRS, qSOFA, and qPELODnegative children compared to controls (Fig 1). For predicting no antibiotic approach, a negative qSOFA and a negative qPELOD-2 both displayed a very high sensitivity of 95  The AUROC of a negative SIRS, qSOFA, and the qPELOD-2 score is presented in gure 2 (Fig 2). The AUROC of a negative SIRS for predicting no antibiotic approach tends to be higher than a negative qPELOD-2 score (0.59 vs. 0.54; p-value=0.023). However, the AUROC between SIRS and qSOFA (0.59 vs 0.56; p-value=0.262) as well as between qSOFA and qPELOD-2 (0.56 vs 0.54; p-value=0.071) were comparable (Table 4).

Discussion
In this retrospective case-control study of 518 children aged 2 to 59 months admitted to an ICU of LMIC, we aimed to focus on the predictive factors and outcome of children who did not receive antibiotics. We also analyzed the accuracy of negative SIRS, qSOFA and qPELOD-2 scores for implementing no antibiotic approach in PICU. Some noteworthy observations were: rst, no antibiotic approach more often implemented in older children and children with hypernatremia. Second, this approach less often implemented in severe underweight, altered mentation, age-speci c fast breathing, chest in-drawing, adventitious sound on lung auscultation, abdominal distension, developmental delay, hyponatremia, hypocalcemia and microscopic evidence of invasive diarrhea. Third, we observed both negative qSOFA, and qPELOD-2 scores showed higher sensitivity to identify children suitable for implementing no antibiotic approach, although the observations were compromised by lower speci city.
We found that antibiotic prescriptions were less in relatively older children. In this context, young infants are at increased risk for severe infection and often receive empirical antibiotic therapy. Rogawski et al. showed that antibiotic use was common in the rst 6 months of life, even at the community level 23 . he association of hypernatremia with no antibiotic approach in young diarrheal children is understandable. The most common cause of hypernatremia on admission is excess water loss due to acute watery diarrhea 24,25 , and viral infections are responsible for most watery diarrhea cases. Empirical antibiotic treatment must be judiciously considered against inadvertent and potentially injurious consequences in such children 26,27 .
Similarly, children showing microscopic evidence of invasive diarrhea and dysentery (visible blood in stool) should be treated with antibiotics 28 as Shigella infection, associated with considerable mortality and morbidity 29 , is suspected. However, eight children had microscopic evidence of invasive diarrhea, and three had bacterial isolates from their stool sample among children in no antibiotic group. These children had not been treated with antibiotics because their diarrhea resolved spontaneously before the availability of reports.
None of our cases had SAM. Several studies have revealed a high prevalence of infections among children hospitalized for SAM, and routine antibiotic therapy is recommended for such children 30 . Children with severe underweight (WAZ <-3 but >-4) seem to be at considerably higher risk of infection compared to well-nourished children 31,32 . A meta-analysis of more than ve thousand children showed substantially higher mortality in underweight and wasted children than well-nourished children 33 . We also observed less frequent implication of no antibiotic approach in children with severe underweight and stunting.
The association of less implication of no antibiotic approach in children with age-speci c fast breathing, chest indrawing, adventitious sounds on lung auscultation is understandable. In LMIC, the diagnosis of pneumonia is based on cough and or breathing di culty with clinical signs such as fast breathing and lower chest wall indrawing. The presence of hypoxemia is a sign of severe pneumonia 34 . World Health Organization (WHO) recommended routine antibiotic therapy for childhood pneumonia to reduce mortality. In this study, the proportion of fast breathing and chest in-drawing among children who did not receive antibiotics was 15% and 2%, respectively, and none had a cough. Additionally, none of them had radiological evidence of pneumonia. These clinical signs without having a cough are known to be a nonspeci c diagnostic tool for pneumonia, as most of our children had diarrhea, and our study children might demonstrate these signs mainly due to metabolic acidosis 35,36 . We found two children in no antibiotic group had hypoxemia due to congenital cyanotic heart disease, so we did not include hypoxemia in the logistic regression model.
We found children who did not receive antibiotics less often had altered mentation, abdominal distension, hypocalcemia, and hyponatremia. Earlier studies have shown that altered mentation and abdominal distension in critically sick children might represent sepsis 37 and should be treated with antibiotics.
Studies have shown that total, and ionized calcium signi cantly reduced in sepsis. The mechanism of sepsis-induced hypocalcemia remains unknown; however, this appears to be associated with elevated levels of proin ammatory cytokines, such as tumor necrosis factor (TNF)-α, interleukin (IL)-1, IL-6; hypoparathyroidism, vitamin D de ciency, or resistance 38 . Similarly, hyponatremia is commonly encountered in critically ill children with respiratory tract infections, sepsis and CNS infections. In a recent study, Park SW et al. reported that co-infection with multiple pathogens was more frequent in hyponatremic children than in those without hyponatremia 39 .
Although clinicians often struggle to decide when to initiate antibiotics in critically ill children in ICU, still no study has examined the utility of negative SIRS, qSOFA, and qPELOD-2 scores as a bedside screening tool to overcome this challenging situation. We found negative qSOFA and qPELOD-2 with <2 scores have high sensitivity of 96% for predicting no antibiotic approach, though no scoring system has high speci city. However, considering AUROC, negative SIRS (<2 scores) is equivalent to qSOFA and more accurate than qPELOD-2. A previous study examining the performance of qSOFA demonstrated that the prevalence of PICU transfer and or mortality was 2% with qSOFA<2 scores compared to 22.5% with qSOFA ≥2 in children presented to the emergency department 40 . The sepsis scoring systems have been developed to identify infection prompting immediate antibiotic therapy, though they lack su cient sensitivity and speci city to capture sepsis. A prior study showed that the sensitivity and speci city of qSOFA were ≥ 37% and 79%, respectively, and that or SIRS were ≥ 80% and 21%, respectively, indicating neither scoring system truly identi es children requiring antibiotic therapy 41 .

Limitation
Our study result might not be generalizable because it was conducted in a single-center ICU of a diarrheal hospital. In developing countries, the PICU mortality ranges from 13-25% 42,43 , though we observed relatively fewer death (5%) in our cohort. We do not have a clear explanation for this; probably patients might not be as severe as in multi-disciplinary hospitals. However, in 2007 the mortality rate of our ICU was 11% 44 . The introduction of bubble CPAP therapy as a standard of care for severe pneumonia with hypoxemia in under-ve children since 2013 may be contributed for better outcomes 45 . Additionally, scrupulous adherence to management guidelines of SAM and other treatment protocols may impact this low mortality. The retrospective nature of the study is another important limitation.

Conclusion
This is the rst study in the epoch of antibiotic resistance to determine predictive factors and outcomes associated with no antibiotic approach in ICU. We determined some simple independent predicting factors that might help ICU clinicians avoid overusing antibiotics in children, especially in resource-poor settings. Based on our ndings, we may propose that ICU clinicians adopt no antibiotic or watchful waiting approach in children with a qSOFA<2 or qPELOD-2 score <2. Our ndings also suggest that the judicious implementation of the no antibiotic approach in PICU does not increase mortality risk than those treated with antibiotics. However, future prospective studies involving larger populations in different settings may consolidate our observation.

Declarations Acknowledgments
We thankfully acknowledge the donors for their support and aptitude to icddr,b's research efforts. icddr,b is appreciative to the Governments of Bangladesh, Sweden, Canada, and the UK for providing fundamental support. We want to express our earnest thanks to all physicians, nurses, and other hospital members for their invaluable support in patient care.  Distribution of included children without and with antibiotic therapy according to the percentage of negative (<2 scores) SIRS, qSOFA, and qPELOD-2 criteria met.