Neonatal and Pediatric Medicine
Open Access

Adverse drug reaction; Machine learning; Prediction; Adverse event; Newborn; Risk analysis

Introduction

It is known that the reported adverse drug reactions (ADRs) in neonates are more common and more severe than in other age groups. According to the double-center study examining potential ADRs for about ten thousand patients, neonatal intensive care unit (NICU) patients had a higher risk of having ADRs. In another study examining the risk factors for ADRs in children, the use of 4 or more neoplasm and circulatory system drugs significantly increased the prevalence of ADRs [1]. ADRs were most frequently reported by pharmacists (89%), nurses (10%), and physicians (1%). Although 93% of ADRs were reported to the physicians, only 29% of these ADRs were documented in the patient’s medical chart [2].

According to a study examining the risk factors for adverse events in neonates, it was determined that 22.5% of the suspected adverse reactions were drug-related. The most frequently identified risk factors (‘triggers’) for the presence of an ADR were decreased oxygen saturation, increased intestinal motility, vomiting, increased creatinine and blood urea nitrogen (BUN), non-steroidal anti-inflammatory drugs or caffeine-related necrotizing enterocolitis (NEC), flumazenil prescription, excessive sedation, or disturbances in electrolytes (hypercalcemia, hyperkalemia, or hypernatremia) [3]. Due to the high prevalence of off-label drug use in newborns, drug-related problems such as ADR are on the rise. A pilot study examining the association between off-label drug use and ME found that 71% of patients were prescribed at least one known to reduce ME by 20.4 fold [4].

In clinical practice, it is very difficult to determine whether adverse events or unexpected events seen in patients are drug-related. For this reason, there are scales in the literature for recording and assessing causality. Similarly, additional tools are needed to assess severity in neonates. The “You” version is an advanced ADR algorithm based on the Naranjo algorithm, specially designed for newborns. Distinguishing between ADR and clinical manifestations is important for improving drug therapy. In this regard, 13 questions were scored ‘yes’/’no’/’unknown’ for the clinician to determine her likelihood of ADR on this scale. Proportion of drug-related undesired effects, according to the final score achieved [5].

Using the Delphi consensus, the Neonatal Adverse Reactions Severity Scale (NAESS) was recently published. According to this report, 35 typical and common neurological, cardiovascular, infectious, respiratory, gastrointestinal, and common neonatal adverse events were mild, moderate, severe, life-threatening, or fatal, respectively. Are classified as, using a common scale as the default [6].

The main objective of this study was to obtain objective risk categories by integrating severity with his NAESS and probability with the ‘Du’ ADRs algorithm in a risk matrix analysis by an interdisciplinary team including clinical pharmacists. Was to do a subsequent goal was to develop a machine learning-based clinical decision support tool (Risk Score) that would predict whether these identified her ADRs would occur [7].

Results

Clinical characteristics

During the study period, 468 neonates were admitted to a 22-bed intensive care unit at a tertiary referral hospital. Fifty-six neonates were excluded because they could not survive without ADR (n=21, 4.5%) or because they did not receive systemic drug therapy (n=35, 7.4%). As a result, 412 neonates were included in the study [8].

232 (56.3%) were male, 177 (43%) were preterm (<37 weeks) and 172 (41.7%) were low birth weight (2500 g). The median postnatal age (IQR) (PNA) was 1 (1) day and the median length of stay (LOS) (IQR) was 8 (11) days.

Was prescribed using during hospitalization, the total number of drugs and anti-infectives used was median (range) of 3 (0-29) and 2 (0-9), respectively. The most frequently prescribed drugs were antiinfectives (38.82%), gastrointestinal and metabolic 131 drugs were prescribed during the study period. The most commonly used drugs were intravenous fluids and ampicillin (7.81%) [9].

Development and optimization of models for predicting the presence of side effects

According to univariate analysis, the candidate variables included in the model to predict whether a patient will develop ADR are:

Gestational age, birth weight, cardiovascular disease, intubation, surgery, parenteral nutrition (PN) treatment, endocrine, neurological and cardiovascular medications (p<0.20). Of these variables, the independent variables determined to be included in the outcome model according to the RF algorithm are:

Endocrine drugs, cardiovascular drugs, cardiovascular disease, neurological drugs, PN treatment.

The concordance index (c-index) is a generalization of AUC and its interpretation is similar to AUC. Values above 0.70 indicate sufficient discrimination. The c-index for this model was found to be 0.914. In general, the indices obtained were determined within the tolerance reported in the literature [10].

For this reason, the process of calculating risk scores was initiated. According to this risk assessment, the ROC analysis was performed on cardiovascular “3 points”, endocrine system “2 points”, drugs for the nervous system (1 point), PN treatment (1 point)) and disease parameters. It was implemented. Circulatory system “1 point”, easyROC. The AUC value for program and risk score was determined to be 0.918. According to the Youden index, the optimal cut-off value for this risk score was set at 3. After the risk score was obtained, the performance of the test set was evaluated. Risk scores were calculated for observations in the test set. Side effects were observed in patients with a risk score of ≥3 points, but no side effects were observed in patients with a risk score of <3 points. Performance criteria were similar to the training set for which cut-off values were determined, compared to the patient’s actual condition.

Discussion

Newborns are susceptible to ADRs, but these events are relatively rarely reported in the literature. In our study, at least one ADR was observed in her 23.8% of patients during hospitalization. 37.4% of prescribed drugs were found to cause ADRs. The number of ADRs observed per patient is estimated to vary between 1 and 6, depending on LOS, number of comorbidities, and number of drugs used. Injecting drug use in the NICU is defined as a high-risk process. A multicenter study conducted by the NICU found that 33% of 69 IV drugs used in the NICU were not approved for neonatal use, 38% were high-risk, and were manufactured and administered at standardized concentrations was found to be only 63.5%. Our study estimates that intravenous administration of high-risk drugs may influence the incidence of ADR.

Regarding the incidence of ADRs in hospitalized infants, a study comparing meropenem with imipenem/cilastatin found that more ADRs were observed with meropenem. However, these ADRs were more severe with imipenem/cilastatin. For this reason, our study found meropenem to be the right choice in the neonatal intensive care unit, but during the first 4 days after starting meropenem treatment, neonatal platelets, eosinophil, and AST Levels should be carefully monitored. A study measuring the incidence of preventable ADRs in the NICU found anti-infective drugs to be the most common cause. Another study found that vancomycin was the drug that caused the most ADRs in pediatric patients.

Conclusions

Although there are similar studies in the current literature, there is no high performance risk score web tool specifically designed for neonates to assess the presence and severity of ADR. By using the obtained risk score in clinical practice, it is expected to identify patients at high risk of side effects and prevent side effects. In addition, this web tool (accessed 22 November 2022) will allow clinicians to improve awareness of these drugs, and mitigation strategies (drug modification, dose, duration of treatment, etc.) can be considered centered approach.

1. Wen LL, Chang WH, Wang HW (2021) Risk factors associated with preterm premature rupture of membranes (PPROM). Taiwan J Obstet Gynecol 60: 805-806.

Indexted at, Google Scholar, Crossref

2. Goligher, Ewan C (2012) Ventilator-Induced Diaphragm Dysfunction. Anesth 117: 463–464.

Indexed at, Google Scholar, Crossref

3. Stein H (2013) Electrical Activity of the Diaphragm [Edi] Values and Edi Catheter Placement in Non-Ventilated Preterm Neonates. Am J Perinatol 33: 707–711.

Indexed at, Google Scholar, Cross Ref

4. Chiew Yeong Shiong (2013) Effects of Neurally Adjusted Ventilatory Assist [NAVA] Levels in Non-Invasive Ventilated Patients: Titrating NAVA Levels with Electric Diaphragmatic Activity and Tidal Volume Matching. BioMed Eng 2: 12-61.

Indexed at, Google Scholar, Cross Ref

5. Beck Jennifer (2009) Patient-Ventilator Interaction during Neurally Adjusted Ventilatory Assist in Low Birth Weight Infants. Pedia Res 65: 663–668.

Indexed at, Google Scholar, Cross Ref

6. Stein, Howard (2012) Synchronized Mechanical Ventilation Using Electrical Activity of the Diaphragm in Neonates. Cli Peri 39: 525–542.

Indexed at, Google Scholar, Cross Ref

7. Kallio Merja (2012) Electrical Activity of the Diaphragm during Neurally Adjusted Ventilatory Assist in Pediatric Patients. Pedia Pulmo 50: 925–931.

Indexed at, Google Scholar, Crossref

8. Rahmani A (2012) Neurally Adjusted Ventilatory Assist in the Neonatal Period: Applications and Limitations. J Neo-Peri Med 5: 205–212.

Google Scholar, Crossref

9. Shilpi M, Kumar KS, Kumar D (2020) Ayurvedic Approach Of Treatment Of Recurrent/ Chronic Cough In Children With Special Reference To Pancha Vidha Kasa. Ind J of App Res 10: 51-52.

Google Scholar                  

10. Nelaturi P, Nagarajan P, Sabapathy SK, Sambandam R (2021) Swarna Bindu Prashana—an Ancient Approach to Improve the Infant’s Immunity. Bio Tra Ele Res 199: 2145-2148.

Indexed at, Google Scholar, Crossref