Predictors of gentamicin therapy failure in neonates with sepsis

Abstract

Sepsis is a common disease with high morbidity and mortality among newborns in intensive care units world‐wide. Gram‐negative bacillary bacteria are the major source of infection in neonates. Gentamicin is the most widely used aminoglycoside antibiotic in empiric therapy against early‐onset sepsis. However, therapy failure may result due to various factors. The purpose of this study was to identify predictors of gentamicin therapy failure in neonates with sepsis. This was a prospective cross‐sectional study at the Neonatal Intensive Care Unit at Windhoek Central Hospital over a period of 5 months in 2019. Neonates received intravenous gentamicin 5 mg/kg/24 h in combination with either benzylpenicillin 100 000 IU/kg/12 h or ampicillin 50 mg/kg/8 h. Logistic regression modeling was performed to determine the predictors of treatment outcomes. 36% of the 50 neonates were classified as having gentamicin treatment failure. Increasing treatment duration by 1 day resulted in odds of treatment failure increasing from 1.0 to 2.41. Similarly, one unit increase in CRP increases odds of gentamicin treatment failure by 49%. The 1 kg increase in birthweight reduces the log odds of treatment failure by 6.848, resulting in 99.9% decrease in the odds of treatment failure. One unit increase in WBC reduces odds of gentamicin treatment failure by 27%. Estimates of significant predictors of treatment failure were precise, yielding odds ratios that were within 95% confidence interval. This study identified the following as predictors of gentamicin therapy failure in neonates: prolonged duration of treatment, elevated C‐reactive protein, low birthweight, and low white blood cell count.

(A) Boxplots comparing gentamicin treatment success or failure with respect to covariates. (B) Correlation matrix heatmap.

Abbreviations

AIC

Akaike information criterion

CRP

c‐reactive protein

eGFR

estimate glomerular filtration rate

EOS

early‐onset sepsis

LMIC

low‐ to middle‐income country

LOS

late‐onset sepsis

NICU

neonatal intensive care unit

OR

odds ratio

PK

pharmacokinetic

ROC AUC

receiver operating characteristic area under the curve

WBC

white blood cell count

1. INTRODUCTION

Neonatal sepsis is a common disease condition with high morbidity and mortality among newborns admitted in the neonatal intensive care unit (NICU) facilities world‐wide. ¹ , ² , ³ The immense public health burden of this disease has real socio‐economic implications, especially for developing countries. ⁴ There seems to be no accepted definition of neonatal sepsis as the signs and symptoms are nonspecific. However, neonatal sepsis can be defined when at least two clinical symptoms such as fever, hypothermia, respiratory distress, feeding difficulties, lethargy, seizures; and two laboratory signs such as low white blood cell count (WBC), low platelet count, high C‐reactive protein (CRP), and metabolic acidosis, presenting with either suspected or confirmed bacterial infection. ⁵ , ⁶ , ⁷ Although viruses, fungi, and parasites are known causes of sepsis, Gram‐negative bacillary bacteria are considered the major source of infection in neonates. ⁸ , ⁹ , ¹⁰

Generally, early‐onset neonatal sepsis (EOS) is transmitted from mother to newborn before or during birth and occurs within the first 72 h of life while late‐onset sepsis (LOS) occurs after 72 h of life and may be transmitted vertically or horizontally. ¹⁰ , ¹¹ Gentamicin, in combination with beta‐lactams, is the most widely used aminoglycoside antibiotic as empiric therapy against EOS. ¹² However, development of antimicrobial resistance, challenges with obtaining accurate diagnosis and administering evidence‐based gentamicin therapy are probable reasons that could result in gentamicin therapy failure. ⁹ , ¹³

Gentamicin therapy in neonates may be unsuccessful due to gentamicin‐resistant strains, subtherapeutic dosing regimens, and incorrect diagnosis of sepsis as a result of concurrent disease conditions which have a presentation similar to sepsis. ¹³ , ¹⁴ , ¹⁵ , ¹⁶ There is a lack of studies that report on the predictors of gentamicin therapy failure in neonates with sepsis. Identification of variables and gaining an understanding of the nature and extent of their influence on the probability of gentamicin treatment failure could be useful in selection and design of more efficacious dosage regimens in the management of sepsis. The aim of this study was to identify the predictors of gentamicin therapy failure in neonates treated against sepsis.

2. METHODOLOGY

2.1. Study design and data collection

This was a prospective cross‐sectional study on neonates admitted and receiving gentamicin therapy against suspected or confirmed sepsis at the Neonatal Intensive Care Unit at Windhoek Central Hospital in Windhoek, Namibia over a period of 5 months (July–December 2019). Neonates were included in the study using convenient sampling. Written consent was obtained from mothers of the prospective participants before enrolling them in the study. The study obtained approval from both the Human Research Ethics Committee of the University of Namibia (ref: H‐G/431/2017), and the Research Ethics Committee of the Ministry of Health and Social Services (ref: 17/3/3 BSS). Neonates received intravenous gentamicin 5 mg/kg every 24 h in combination with either benzylpenicillin 100 000 IU/kg every 12 h or ampicillin 50 mg/kg every 8 h. The drugs were administered over 3–5 seconds through an intravenous cannula by the ward nurses. Thereafter, two blood samples were collected by venipuncture by ward doctors and stored in sterile 500 μL serum separating tubes (SST). The pharmacokinetics (PK) profile randomized (block) sampling design for a drug exhibiting monoexponential pharmacokinetics was used ¹⁷ in which each blood sample was taken at a time falling in either of the following sampling blocks: 5–8 min (0.08–0.14 h), 8–250 min (0.14–4.2 h), or 250–450 min (4.2–7.5 h) in a ratio of 40:20:40 (a total of 100 samples).

Once blood samples arrived in the laboratory, they were centrifuged and serum stored in Eppendorf tubes at −20°C to await analysis. Serum creatinine concentration measurements followed the enzymatic method using the Cobas® 6000 analyzer (Roche Diagnostics, IN, USA). Gentamicin concentrations were measured using the Indiko Plus™ autoanalyzer (Thermo Fisher Scientific Inc, CA, USA). The lower limit of quantification for gentamicin was given by the manufacturer as 0.3 μg/mL. Gentamicin population pharmacokinetics (PPK) was estimated using the nonlinear mixed effects modeling approach in NONMEM (version 7.4.1, ICON Plc., Ellicott City, MD, USA) with a Pirana (version 3.6.2, Pirana Software and Consulting BV) front end. This was followed with the computation of AUC using individual subject PK parameters.

The following data were collected using a data abstraction tool from patient records which were available on the bedside: gentamicin dose, birth weight (BWT), gestational age (GA), white blood cell count (WBC), CRP, sex, height, postnatal age (PNA), platelet count, serum creatinine (SCr), and platelets. Full blood counts, and urea and electrolyte laboratory tests were performed every 24 h. Neonates whose sepsis improved due to gentamicin therapy were classified as having successful treatment outcome (i.e., “cure”), while those who died or were changed to second‐line therapies due to lack of improvement in variables for diagnosis of sepsis were regarded as treatment failure (i.e., “no cure”).

2.1.1. Analysis strategy

The analysis was conducted in three major stages. In the first stage, exploratory analysis was performed to discern possible relationships between the analysis variables. Covariates that showed no observable relationships with the outcome of interest were not considered in the second stage of the analysis. However, a covariate was retained for the next analysis stage if there was a biological reason to do so. In the second stage, a univariate logistic regression was performed with each of the covariates selected from the first stage. The third stage involved the performance of multivariate logistic regression to develop a final model to determine the predictors of gentamicin therapy failure using the stepwise backward elimination modeling approach. All analyses were performed using R software (version 3.63, The R Foundation for Statistical Computing, Vienna, Austria).

2.1.2. Exploratory analysis

Exploratory analysis was done by graphical analysis (using boxplots and correlation matrix heatmap) and correlational correlation tests to discern possible relationships between the independent variables and gentamicin therapy failure and the correlation between the independent variables (covariates) themselves. Correlation analysis was performed to reduce the dimensionality of the covariate by minimizing co‐linearity between covariates. Covariates were considered to be highly correlated if the correlation was ≥0.70.

2.1.3. Univariate logistic analysis

Covariates (i.e., gentamicin dose (dose), BWT, GA, WBC, CRP, sex, height, PNA, eGFR, and platelet count (PLTLS)) that appeared related to gentamicin therapy failure from the graphical and correlational analyses step were selected for univariate logistic analysis to determine their suitability for use in developing a final model by multivariate analysis in the third stage of the analysis. Covariates that were highly correlated with each other. Candidate covariates were selected at an a priori 25% significance level at this analysis stage. Selected covariates were used to create the full model that was subjected to backward elimination in stage three of the analysis to develop the final model via multiple logistic regression.

2.1.4. Multivariate logistic regression analysis

A full model consisting of candidate covariates selected from the univariate analysis step The backward elimination modeling approach was used whereby the covariates selected from the univariate regression step were included in the full model, and those found not to be statistically significant were removed systematically until a parsimonious model with minimum Akaike information criterion (AIC) was obtained. A covariate was considered significant at α = 0.05 for retention in the final model, given the data. The following covariates were included in the initial model: reciprocal of SCr, log‐transformed value of WBC, area under the gentamicin concentration‐time curve, estimated glomerular filtration rate (eGFR), postnatal age, duration of gentamicin therapy, birthweight, and concentration of CRP.

2.2. Logistic regression model

Logistic regression was used to characterize the probability of gentamicin therapy failure (TFailure) in neonatal sepsis as a function of the independent variables (i.e., covariates) was described as follows:

$$\begin{array}{cc}\mathit{\text{logit}}{P}_i\left(\mathit{\text{TFailure}}\right)=\hfill & \mathit{log}\left(\frac{P_i\left(\mathit{\text{TFailure}}\right)}{1-P_i\left(\mathit{\text{TFailure}}\right)}\right)={\alpha }_{\mathit{\text{failure}}}+{\beta }_1{X}_1\hfill \\ \hfill & +{\beta }_2{X}_2+{\beta }_3{X}_3+\dots {\beta }_n{X}_n\hfill \end{array}$$ (1)

where P _i is the probability of therapy failure in the ith patient, α is the baseline log odds of therapeutic failure. The β _{1, …,} β _n characterize the dependence of the log odds on one or more covariates, X ₁ …, Xn.

Because the explanatory covariate X increases by one unit from X to X + 1, the odds of gentamicin therapy failure change from $e^{\mathrm{\alpha }}{e}^{\mathit{\beta X}}$ to $e^{\mathrm{\alpha }}{e}^{\beta \left(X+1\right)}=e^{\mathrm{\alpha }}{e}^{\mathit{\beta X}}{e}^{\mathrm{\beta }}$. The odds ratio (OR) is therefore $\frac{e^{\mathrm{\alpha }}{e}^{\mathit{\beta X}}{e}^{\mathrm{\beta }}}{e^{\mathrm{\alpha }}{e}^{\mathit{\beta X}}}$.

That is,

$$\mathit{\text{OR}}=e^{\beta }$$ (2)

A 95% confidence interval around an OR was calculated by first determining the interval around β and then exponentiating thus,

$$e^{\left(\beta -{1.96}^{*}\mathit{SE}\left(\beta \right)\right)},e^{\left(\beta +{1.96}^{*}\mathit{SE}\left(\beta \right)\right)}.$$

2.3. Model evaluation

Upon univariate analysis, any variable whose univariate test has a p < 0.25 was considered a candidate for multivariate analysis, along with any covariate that could be of biological importance. The use of the 0.25 level is based on the report that the more traditional level of 0.05 fails to identify variables known to be important. ¹⁸ , ¹⁹

For the multiple logistic regression, a p < 0.05 was considered statistically significant for retaining a covariate in the final model. In addition, the ratio of a regression coefficient to its standard error was also examined to ensure that the final model obtained with the multivariate logistic regression analysis included variables with ratios approximating 2 or greater. The square of the ratio is the Wald statistic, and if the significance level of the Wald statistic is small (<0.05), then the parameter is useful to the model. That is, the Wald statistic for each covariate had to exceed approximately 4.0. Diagnostic plots were examined to determine if the model adequately described the data. If a residual exceeded a value of 2.0, it was considered an outlier. The patient with that residual was removed, and the model was refitted. These diagnostics were combined with the estimated odd ratios to consider a covariate for retention in the final multivariate logistic model.

2.4. Model discriminative and predictive capability

Models that distinguish well between neonates who had successful gentamicin therapy and those who did not are said to have good discrimination. Model discriminative/predictive capability was determined using the theory of ROC curves. ²⁰ A commonly used measure of discrimination is the receiver operating characteristic (ROC) curve (also known as the c‐statistic or c‐index). The c‐index ranges from 0 to 1, with higher values indicating better discrimination. Basically, the prediction for a patient is compared with the corresponding observation for that individual.

The ROC curve shows the sensitivity (the proportion of correctly classified positive observations) and specificity (the proportion of correctly classified negative observations) as the output threshold is moved over the range of all possible values. The probability that a value of any of the covariate measures could discriminate between improved and unimproved patients was represented by ROC AUC. A ROC plot is generated by plotting the pairs of sensitivity and specificities (or, more often, sensitivity versus one minus specificity) in a scatter plot. In this report, pairs of sensitivity and specificity are plotted to generate the ROC curve. The intention is to make it easier to read because the highest sensitivity occurs at the highest specificity.

The ROC AUC represents the ability of a classifier or measurement tool to accurately detect those with the outcome of interest. The ROC AUC value ranges between 0.5 and 1.0. A ROC AUC of 0.5 indicates no discriminative ability, whereas a ROC AUC of 1.0 perfectly discriminates between responders and non‐responders. The larger the ROC AUC, the better the performance of the measure.

The ROCs were generated using the pROC package in R. ²¹ The confidence interval about the AUC of a ROC was estimated with the Delong method, ²² and the AUC computed by trapezoidal rule. The bootstrap resampling method was used to compute the confidence intervals of the sensitivity and specificity values used to plot the confidence bands in ROC plots.

3. RESULTS

3.1. Analysis dataset

A total of 50 preterm and term neonates were included in this study with median BWT, gestational and postnatal age of 1.56 kg, 32 weeks and 4.7 days, respectively (Table 1). 36% of the neonates had a treatment outcome of ‘no cure’ and were classified as having gentamicin treatment failure. The covariates used in the analysis are summarized in Table 1.

TABLE 1.

Patient characteristics (n = 50).

Characteristic	Median	Range
BWT (kg)	1.56	0.75–3.92
Gestational age (weeks)	32	24–40
Postnatal age (days)	4.7	1–17
Height (cm)	41	30–53
eGFR (mL/min/1.73 m2)	19.5	8.7–50.4
WBC count (×109 cells/L)	11.02	1.67–37.40
C‐reactive protein (g/L)	1.00	0.10–35.70
Gentamicin dose (mg)	7.7	4.0–17.0
Duration of treatment (days)	5.0	2–13
Gentamicin AUC (mg × h/L)	90.3	39.9–595.0

3.1.1. Exploratory graphical and correlational analysis

Exploratory graphical analysis showed that gentamicin therapy failure appeared to be related to birthweight, CRP, WBC, and PNA (Figure 1A). The height, birthweight, and GA were highly and significantly correlated with each other (Figure 1B). The correlation coefficient (r) between height and birthweight, and birthweight and gestational was 0.8 and 0.8, respectively. Therefore, of these three covariates, only birthweight was included in the base model. Clearly, RSCr and platelet count had no relationship with gentamicin therapy failure, while duration, birthweight, WBC, PNA, CRP, eGFR appeared to have some relationship with gentamicin therapy failure. Although AUC had a correlation coefficient of 0.1 in its relationship with gentamicin therapy failure, it was included in the list of covariates tested in the univariate logistic regression analysis to rule out the biological plausibility of its contribution to gentamicin therapy failure.

FIGURE 1.

(A) Boxplots comparing gentamicin treatment success or failure with respect to covariates. (B) Correlation matrix heatmap. DV, dependent variable (gentamicin therapy failure); DUR, Duration; PLTLS, platelet count.

Platelets and WBC, platelets and PNA, RSCr, and eGFR were significant but not highly correlated (Table 2, upper panel). Thus, these covariates, together with birthweight and others in Table 2 (lower panel), were tested in the univariate analysis to select covariates for inclusion in the base (full covariate model). Table 2 (upper panel) summarizes the results of the correlation between covariates.

TABLE 2.

Summary of correlation tests and univariate logistic analysis results.

Correlation between covariates
Covariate1	Covariate2	Correlation coefficient	p Value
*Dose	*AUC	0.057	.697
BWT	Height	0.79	1.27E‐11
BWT	GA	0.80	6.15E‐12
BWT	PNA	−0.30	.035
GA	PNA	−0.27	.058
*WBC	CRP	0.19	.187
WBC	Platelets	0.32	.024
Platelets	PNA	0.29	.038
RSCr	eGFR	0.37	.009
Duration	PNA	0.06	.689
Duration	GA	−0.15	.304

Summary univariate logistic covariate analysis results
Parameter	Estimate	Standard error	p Value
Intercept	−1.8467	2.0874	.376
*AUC	0.2724	0.4413	.537
Intercept	−2.2886	0.7675	.003
PNA	0.3755	0.1602	.019
Intercept	2.0178	1.452	.165
*WBC	−1.073	0.5939	.071
Intercept	−1.0985	0.3806	.004
CRP	0.1482	0.0749	.048
Intercept	1.34272	1.03629	.195
eGFR	−0.09869	0.05323	.064
Intercept	−1.9133	0.9152	.037
Duration of treatment	0.2629	0.1682	.118
Intercept	2.1584	1.1395	.058
BWT	−1.6409	0.7091	.021

Note: *Dose, AUC, and WBC were log‐transformed to approximate normality.

Abbreviations: BWT, birthweight; CRP, c‐ reactive protein; eGFR, estimated glomerular filtration rate; GA, gestational age; PNA, postnatal age; RSCr, reciprocal of SCr.

3.1.2. Univariate analysis

The results of the univariate logistic regression results are summarized in Table 2 (lower panel). Of all the covariates (AUC, PNA, WBC, CRP, eGFR, and BWT) tested in the univariate analysis, only AUC did not achieve the 25% significance level. Thus, PNA, WBC, CRP, eGFR, and BWT were included in the base model (i.e., the full model) for the development of the final multiple logistic regression model of the predictors of gentamicin therapy failure.

3.1.3. Development of the multiple logistic regression model

Table 3 summarizes the predictive significance of covariates in the full model. Again, AUC was included in the full model to determine whether it contributed to gentamicin therapy failure. Of all the covariates tested in the full model, only the duration of gentamicin therapy, birthweight, WBC, and CRP were significant predictors at the 0.05 significance level.

TABLE 3.

A summary of parameter estimates from the full model.

Parameter	Estimate	Standard error	p Value
Intercept	1.00	7.15	.889
AUC	0.00172	0.00382	.652
Duration	0.870	0.386	.024
PNA	0.234	0.246	.341
BWT	−6.70	2.88	.020
WBC	−0.296	0.150	.049
CRP	0.373	0.142	.008
eGFR	−0.0455	0.106	.669

Note: AIC = 45.965.

The final model retained the duration of gentamicin therapy, birthweight, WBC, and CRP as significant predictors of gentamicin therapy failure (Table 4). The ratio of the estimates of the regression coefficients of each of these covariates to their respective standard errors exceeded two, as indicated by the fact that the Wald statistic for each of the covariates exceeded 4.0 (Table 4).

TABLE 4.

Significant variables retained in the final model (n = 49).

Coefficients	Estimate	Standard error	Odds ratio (95% CI)	Wald Statistic	p Value (Wald's test)
Intercept	7.1775	3.5612	—		—
Duration of treatment	0.8816	0.3583	2.41 (1.2,4.87)	6.05	.014
Birthweight	−6.8484	2.6877	0.00106 (0,0.21)	6.49	.011
WBC	−0.3114	0.1308	0.73 (0.57,0.95)	5.67	.017
CRP	0.4006	0.1390	1.49 (1.14,1.96)	8.31	.004

Note: Log‐likelihood = −14.2075, No. of observations = 49, AIC value = 38.415.

One patient with a residual of +7 was considered an outlier as this value was far above the mean of 0 and exceeded the outlier cutoff criterion of 2.0 (Figure 2A). The patient was removed from the dataset, and the model was refitted. Although the distribution of residuals in the lower end of the quantile‐quantile (0 to 0.05 theoretical quantile) had a little curvature, removing the leverage subject from the dataset produced a better fit of the model to the data (Figure 2B). In addition, the ROC AUC increased from 0.89 to 0.94 after the removal of the influential (leverage) patient, an indication that the model's predictive/discriminative power improved (Figure 3).

FIGURE 2.

(A) Residual plots (n = 50). (B) Residual plots (n = 49).

FIGURE 3.

Receiver operated characteristic (ROC) curve (n = 49).

The final logistic model, given the data, for predicting gentamicin therapy failure was as:

$$\begin{array}{cc}\mathit{\text{logit}}\left(T_{\mathit{\text{Failure}}}\right)=\hfill & 7.178+0.882\left(\mathit{\text{Duration}}\right)-6.848\left(\mathit{\text{Birthweight}}\right)\hfill \\ \hfill & -0.3114\left(\mathit{WBC}\right)+0.401\left(\mathit{CRP}\right)\hfill \end{array}$$ (3)

The parameter estimates of the final logistic model were efficiently estimated (Table 4), with only the statistically significant covariates retained.

Holding other covariates constant, a one‐day increase in treatment duration increases the log odds of treatment failure by 0.882. Thus, the odds of treatment failure increased from 1.0 to 2.41. Similarly, a one‐unit increase in CRP increases the log odds and odds of gentamicin treatment failure by 0.401 and 49%, respectively. A 1 kg increase in birthweight reduces the log odds of treatment failure by 6.848, resulting in a 99.9% decrease in the odds of treatment failure. A one‐unit increase in WBC reduces the log odds and odds of gentamicin treatment failure by 0.3114 and 27%, respectively. Estimates of all four significant variables gave ORs that were within the 95% confidence interval with statistical significance, as expected (Table 4). A graphical representation of the probability of gentamicin therapy failure with the influence of each of the four covariates is shown in Figure 4. In addition, the probability of gentamicin therapeutic failure was predicted to decrease as birthweight increased, decreasing with an increase in WBC and increasing with increased treatment duration and CRP. Thus, for neonates weighing >2.13 kg, the average probability of experiencing gentamicin treatment failure was predicted to be negligible when other predictors were held constant (Figure 4). When CRP increased above 5 mg/L while other covariates were held constant, the average probability of gentamicin treatment failure was predicted to increase rapidly to a maximum of >99% when the CRP level was 18 mg/L (Figure 4). Holding other covariates constant, the average probability of gentamicin therapy failure was predicted to be negligible when WBC was above 15 $\times$10⁹/L (Figure 4).

FIGURE 4.

Final model of the effects of covariates on predicting the probability of gentamicin therapy failure for neonates at Windhoek Central Hospital (n = 49).

4. DISCUSSION

Neonatal sepsis is a major cause of mortality in low‐ and middle income countries (LMICs) where equitable access to quality and cost‐effective healthcare, including efficacious antimicrobials such as gentamicin, could be a challenge. ³ , ⁴ , ²³ , ²⁴ , ²⁵ Although an optimal antibiotic regimen against neonatal sepsis has not been established by clinical trials, the absence of timely antibiotic intervention leads to a rapid decline in the clinical presentation of sepsis. ⁶ , ²⁶

One study identified predictors of gentamicin (in combination with penicillin G or ampicillin) therapy failure to be CRP, birthweight, GA, WBC, and platelet count. ²⁷ However, it is important to note that birthweight and GA are generally highly correlated as observed in our study. Having the two highly correlated variables in a model creates a problematic because of collinearity and the attendant redundancy of one of the covariates. Equally, WBC count and platelet count are generally significantly positively correlated with each other ²⁸ , ²⁹ When collinearity is present, a “regression coefficient does not reflect any inherent effect of the particular predictor variable on the response variable but only a marginal or partial effect, given whatever other correlated predictor variables are included in the model”. ³⁰ When there is correlation between predictors, the estimation regression coefficients tend to vary widely from one dataset to another. Therefore, the generalizability of the model reported by Metsvaht et al. ²⁷ that has CRP, birthweight, GA, WBC, and platelet count as predictors of gentamicin failure is questionable. This prompted the need for using a structured approach to the development of a model described in this investigation to determine the predictors of gentamicin therapy failure that addresses the issues raised herein.

In this study, we determined that birthweight and GA were significantly correlated (r = 0.80, p = 6.15E‐12); therefore, only birthweight was used in the analyses reported herein. Platelet had no relationship with gentamicin therapy failure (r = 0.0) while WBC was related (r = −0.30). The latter was a significant predictor by univariate regression (p = .049) and was retained in the final model as a significant predictor of gentamicin therapy failure (P = .017).

Following multivariate regression, the following were identified as predictors of gentamicin therapy failure in neonates: prolonged duration of treatment, elevated CRP, low birthweight, and low WBC count. In our investigation, the platelet count could not be confirmed as a predictor of gentamicin therapy failure, while the prolonged duration of treatment was identified as a predictor of gentamicin therapy failure. The final model contains no collinear predictors, and it has good discriminative and predictive ability with an ROC AUC of 0.94, and, therefore, generalizable.

4.1. Duration of treatment

Results from this study suggest that the longer the duration of treatment with gentamicin, the more likely it is that therapy will fail. A course of antimicrobial therapy should last no longer than 48–72 h in neonates who are stable and whose blood culture is negative for sepsis. ³¹ It is reported that only 5% of neonates who receive empiric antibiotic therapy against sepsis actually have a positive blood culture, and the duration of 60% of these courses is extended for 48–72 h beyond the prescribed period. ³² Studies show that there is no increase in treatment failure when antimicrobial therapy is ceased before 72 h in stable neonates with a negative blood culture, whereas extending the course for more than 7 days does not improve treatment outcomes but prolongs hospitalization in extremely low birthweight neonates. ³³ , ³⁴ The fact that prolonged gentamicin therapy results in failure is of concern as it could point to either a possibility of microbial resistance to the drug or that the suspicion of sepsis in these neonates could be erroneous as the presenting disease may be something else entirely. The neonatal sepsis dilemma, which promotes empirical antibiotic therapy, is understandable, but this should be guided by knowledge of the latest local antimicrobial resistance patterns to reduce cases of inappropriate prescribing of antimicrobials, which promotes empirical antibiotic therapy is understandable, but this should be guided by knowledge of the latest local antimicrobial resistance patterns to reduce cases of inappropriate prescribing of antimicrobials which are common in neonatal intensive care units. ³¹ , ³⁵

4.2. C‐reactive protein

The analysis indicates that an elevation in CRP is a predictor of gentamicin therapy failure. CRP is produced by the liver as a reaction to infection and is considered a specific marker of neonatal infection, it is, therefore, a useful biomarker for monitoring the effect of therapy against neonatal infection and reducing the duration of antimicrobial therapy. ⁶ , ³¹ , ³⁶ , ³⁷ However, the sensitivity of CRP is low in the initial phase of sepsis as values may take up to 12 h to change, and disease conditions of non‐infectious nature, such as ischemic tissue injuries, hemolysis, and chorioamnionitis may also be a cause of elevated CRP values. ⁶ , ³⁸ Several studies report that the timing of CRP measurements affects sensitivity and predictive results as values determined 24–48 h after the onset of symptoms are more sensitive and, therefore high negative predictive results, which have been reported to be around 86%–100%. ³⁶ , ³⁸ , ³⁹ , ⁴⁰

4.3. Birthweight

Results in this study suggest that low birthweights negatively influence gentamicin therapy success. Newborns with very low birth weight (VLBW), mostly preterm, are known to have a high probability for both early‐onset sepsis (EOS) and late‐onset sepsis (LOS) with high mortality (up to 20%) due to an immature immune system, prolonged mechanical ventilation, longer duration of stay in hospital and other reasons which decrease their likelihood of improving. ¹⁰ , ⁴¹ Gentamicin distributes mainly into the extracellular fluid space, which decreases with gestation and eventually with chronological age; the volume of distribution (V) in VLBW preterm neonates at 32 weeks gestation is 0.5 L/kg while in adults it is around 0.25 L/kg. ⁴² , ⁴³ , ⁴⁴ In this study, V was estimated to be 0.42 L. In addition, neonates of VLBW are known to have suppressed kidney function that necessitates the reduction in dose requirements to avoid excessive exposure to the drug, but improvement in drug clearance follows the increase in body weight that comes with age. ⁴⁵ , ⁴⁶ Therefore, the explanation of low BWT as a possible predictor of gentamicin therapy failure is most probably due to the immature state of the body and its systems during the earlier stages of life.

4.4. WBC count

An increase in WBC was found to be predictive of gentamicin therapy success. In this study, the median WBC (11 × 10⁹ cells/L) was on the lower end of the normal range (9–30 × 10⁹ cells/L). WBC is influenced by the GA and body weight of the neonate. ⁶ , ⁴¹ , ⁴⁷ Low WBC is associated with increased risk of infection as it reflects a reduced capacity of the innate and adaptive immune system to fight invasive disease‐causing microorganisms. ⁴⁸ , ⁴⁹ In this regard, gentamicin therapy is more likely not to fail when the WBC is high because of the synergistic effect between the immune system and the drug against the bacteria.

Gentamicin therapy could fail due to the development of resistance by infectious microorganisms. As the management of neonatal sepsis should be based on antimicrobial resistance patterns, it would have been meaningful to include this variable in the regression analysis. However, this information was not available in the data records. In addition, the patient's disease condition, being either suspected or confirmed sepsis as a predictor of gentamicin therapy failure, could not be studied as this information was also missing.

5. CONCLUSIONS

This study identified the following as predictors of gentamicin therapy failure in neonates: prolonged duration of treatment, elevated CRP, low birthweight, and low WBC count. Based on these results, clinicians should restrict empiric gentamicin therapy to the prescribed duration and seek culture results for evidence‐based antimicrobial therapy.

AUTHOR CONTRIBUTIONS

Bonifasius Singu, Clarissa Pieper, Roger Verbeeck, and Ene Ette were involved in the study conception and design. Recruitment of study participants, blood sample collection, and laboratory analysis were performed by Bonifasius Singu. Data analysis and preparation of the first draft were done by Bonifasius Singu and all authors contributed to the final manuscript.

FUNDING INFORMATION

BS Singu was partly supported by a grant from Anoixis Corporation (USA) in carrying out this study as part of his Ph.D. work.

CONFLICT OF INTEREST STATEMENT

The authors have no conflicts of interest to declare that are relevant of the content of this article.

ETHICS APPROVAL AND CONSENT TO PARTICIPATE

All methods were carried out in accordance with relevant guidelines and regulations. The study was approved by the University of Namibia Human Research Ethics Committee (Ref: H‐G/431/2017) and the Research Committee of the Ministry of Health & Social Services in Namibia (Ref: 17/3/3 BSS). Informed consent was first sought and obtained from the mothers of the newborns before they were included in this study.

Singu BS, Pieper CH, Verbeeck RK, Ette EI. Predictors of gentamicin therapy failure in neonates with sepsis. Pharmacol Res Perspect. 2024;12:e1250. doi: 10.1002/prp2.1250

DATA AVAILABILITY STATEMENT

The data that support the findings of this study are available on request from the corresponding author. The data are not publicly available due to privacy or ethical restrictions.