Introduction

Prematurity remains a major global burden, with preterm birth complications accounting for 0.94 million (17.7%) of 5.30 million under-five deaths worldwide in 2019 [1, 2]. Life-saving antibiotic administration for preterm infants can occasionally lead to adverse effects such as altered intestinal microbiota, acute kidney injury, and irreversible sensorineural hearing loss [3–6]. Clinicians should develop individualized treatment strategies based on drug safety information; however, limited data on the safety and efficacy of drugs in neonates are available, and 30 of the top 50 most frequently used medications in neonatal intensive care units in the United States were off-label [7–9]. While researchers must become current with the changing trends of medication use for preterm infants, the details of neonatal care in Japan have not been well-reported, despite the country having one of the lowest mortality rates for preterm infants [7, 10, 11].

Furthermore, antimicrobial administration in neonatal care may exhibit regional variations within the country because of the absence of well-established standards [12, 13]. Anselin defined spatial effects as spatial autocorrelation (or geospatial correlation within variables) and geospatial heterogeneity (or heteroscedasticity) [14–16]. Unmeasured confounding factors with spatial effects would violate the assumption of non-autocorrelation and homoscedasticity in the ordinary least squares estimation; therefore, overestimated variance can cause a bias toward the null hypothesis [17, 18]. Thus, confirmation of the spatial autocorrelation in neonatal care implies the need to address the spatial effects of unmeasured confounders in observational studies of neonates using nationwide real-world data.

This study aimed to investigate the geographical distribution of antimicrobial drug exposure among very preterm (VPT; gestational age 22–32 weeks) and very low birth weight (VLBW; <1,500 g) infants in Japan, with the objective of assessing its spatial effects.

Methods

We conducted a retrospective study using the Diagnosis Procedure Combination (DPC) database, which is a Japanese administrative claims database. As of 2021, more than 2,000 hospitals have adopted the DPC-based reimbursement systems. This database contains anonymous clinical and administrative claims data, including baseline information, diagnoses, procedures, medications, and device use. Previous studies have confirmed the validity of this database [19]. Data were accessed on September 22, 2022.We evaluated the DPC database to identify admissions of infants, who were both VPT and VLBW, within the first two days of life from April 2010 to March 2021. We defined early neonatal drug exposure and neonatal drug exposure as any exposure to each class of drugs during the early neonatal period and neonatal period, i.e., within the first 7 days and 28 days of life, respectively. We defined two specific cohorts based on the discharge period of infants. Cases discharged during the early neonatal period were excluded. Consequently, we established an “early neonatal cohort” to specifically assess drug exposure during this phase. Similarly, upon the exclusion of cases discharged during the neonatal period, we delineated a separate “neonatal cohort” for further analysis. Infants discharged during the corresponding periods were excluded from the analysis. Prescriptions for these cohorts were extracted to calculate exposures to injectable antimicrobials classified as “antibacterial agents for systemic use” (J01), “antimycotic agents for systemic use” (J02), “antimycobacterial agents” (J04), and “antivirals for systemic use” (J05) according to the anatomical therapeutic chemical (ATC) classification [20, 21]. Nationwide early neonatal exposure and neonatal exposure to each class of drugs were determined. For international comparisons, exposures among extremely low birth weight infants were also assessed. We investigated the geographical distribution of antimicrobial drug exposure based on the prefecture where the facility is located. Each of the 47 prefectures in Japan is an independent administrative entity that provides comprehensive neonatal care through cooperative perinatal care centers. Each prefecture has at least one medical school.

We examined the inter-prefectural spatial autocorrelation of early neonatal exposure and neonatal exposure to each antimicrobial agent using global and local Moran’s I statistics [22]. We created local indicators of spatial association cluster plots and Moran scatter plots, which display each prefecture according to five categories: high-high, high-low, low-high, low-low, and not significant. A high-high classification indicates “high exposure surrounded by high exposures.” This is indicative of a positive spatial autocorrelation or a spatial cluster of high exposures (a hot spot). A high-low classification is an outliner that indicates “high exposure in the target district while neighboring districts had low exposure.” A low-high classification is another outliner that indicates “low exposure in the target district while neighboring districts had high exposure.” A low-low classification indicates “low exposure surrounded by low exposures.” Similar to the “high-high” scenario, this is indicative of positive spatial autocorrelation or a spatial cluster of low exposures (a cold spot). The “not significant” classification indicates a lack of either positive or negative significant spatial autocorrelation according to the local Moran’s I statistics [15, 23]. Subsequent to the primary analysis based on the fifth level of the ATC classification (chemical substance), we conducted a sensitivity analysis using the higher fourth level of the ATC classification (chemical subgroup).

To explore the geographical heterogeneity underlying spatial autocorrelation, we aggregated the 47 prefectures into eight regions (Hokkaido, Tohoku, Kanto, Chubu, Kansai, Chugoku, Shikoku, and Kyushu regions) and generated heatmaps of the incidence of the early and late neonatal courses of antibacterial administration stratified by gestational age (22–23, 24–27, and 28–31 weeks), birth weight (<500, 500–999, and 1000–1499 g), and region. Late neonatal courses were defined as antibacterial prescriptions during the late neonatal period (between days 7 and 27 after birth [24]) after at least 7 antibacterial-free days. We also created heatmaps of the selection rates of each antibacterial agent on the first day of the antibacterial administration courses. Comparable heatmaps were created for early neonatal exposure to antifungal agents.

We computed weight matrices for spatial statistical analysis based on the k-nearest neighbors method, with k = 4 for centroid coordinates, because some prefectures are not contiguous. Spatial statistical analyses were performed using the spdep package in R, with the significance level set at 0.05 [25, 26]. Permission to conduct this study was obtained from the Ethics Committee of Tokyo Medical and Dental University (registration no. M2021-013). The study was exempted from informed consent due to the data being anonymized. We used open geographic data from the National Land Numerical Information released by the Japanese Ministry of Land Infrastructure, Transport and Tourism [27].

Results

We identified 43,536 hospitalizations of VPT and VLBW infants within the first 2 days of life. We included 41,423 infants to analyze early neonatal drug exposure and 39,635 infants to analyze neonatal drug exposure (Fig 1). In the analysis of neonatal drug exposure, the median gestational age was 28.0 weeks (interquartile range: 26.0–30.0 weeks), and the median birth weight was 1,013 g (interquartile range: 760–1,253 g); male infants accounted for 52% of the cohort (Table 1). The mortality rates for all admissions, the early neonatal cohort, and the neonatal cohort were 6.8%, 3.8%, and 2.3%, respectively. The eight regions of Japan and the spatial weight matrix assigned to the 47 prefectures were visualized in Fig 2.

Download:

• PNG
  larger image
• TIFF
  original image

Fig 1. Flowchart showing initial patient eligibility and exclusion because of early discharge.

https://doi.org/10.1371/journal.pone.0295528.g001

Download:

• PNG
  larger image
• TIFF
  original image

Fig 2. Visualization of the eight regions of Japan and spatial weight matrix assigned to the 47 prefectures.

https://doi.org/10.1371/journal.pone.0295528.g002

Download:

• PNG
  larger image
• TIFF
  original image

Table 1. Patient characteristics.

https://doi.org/10.1371/journal.pone.0295528.t001

Table 2 shows the frequently prescribed antimicrobials with a neonatal exposure rate of >10 per 1,000 infants. The top 10 antimicrobial agents prescribed during the neonatal period were as follows: ampicillin (J01CA01); amikacin (J01GB06); gentamicin (J01GB03); cefotaxime (J01DD01); fluconazole (J02AC01); ampicillin combinations (J01CA51); micafungin (J02AX05); cefmetazole (J01DC09); cefazolin (J01DB04); and vancomycin (J01XA01). These drugs had a neonatal exposure rate of >50 per 1,000 infants. Among these drugs, early neonatal exposure to fluconazole and neonatal exposure to amikacin, fluconazole, and cefmetazole showed significant positive spatial autocorrelations with the global Moran’s I statistics. Local indicators of spatial association cluster plots revealed northern cold spots for early neonatal exposure and neonatal exposure to fluconazole in the Hokkaido and Tohoku regions (Fig 3). Higher early neonatal exposure and neonatal exposure to ampicillin, amikacin, gentamicin, and cefmetazole formed southern hot spots in the Kyushu region. Early neonatal exposure and neonatal exposure to each class of prescribed antimicrobials and their global Moran’s I statistics are listed in S1 Table. The geographical distributions of antimicrobial drug exposure, local indicators of spatial association cluster plots, and Moran scatter plots for all prescribed drug classes are shown in S1 Fig. Prescriptions of antimycobacterial agents (J04) and antivirals (J05) were limited, and no remarkable spatial autocorrelation was detected. The sensitivity analysis based on a higher level of drug classification (S2 Table and S2 Fig) yielded comparable results. Our analysis of extremely low birth weight infants is displayed in S3 Table.

Download:

• PNG
  larger image
• TIFF
  original image

Fig 3. Local indicators of spatial association cluster plots of early neonatal exposure and neonatal exposure to the top 10 frequently used antimicrobial agents.

https://doi.org/10.1371/journal.pone.0295528.g003

Download:

• PNG
  larger image
• TIFF
  original image

Table 2. Frequently used drugs with early neonatal and neonatal exposure and their global Moran’s I statistics.

https://doi.org/10.1371/journal.pone.0295528.t002

Heatmaps of regional diversity in drug administration stratified by gestational age and birth weight are presented in S3 Fig. No remarkable regional differences in the incidence of antibacterial drug administration were observed (S3A Fig). Regarding antibacterial drug selection during the early neonatal period, regional diversity was detected in the impact of gestational age on the administration of aminoglycosides (S3B Fig). Extremely preterm infants (gestational age <28 weeks) had significantly higher selection rates for aminoglycosides in the Hokkaido, Tohoku, Kanto, and Kansai regions (Fisher’s exact test with Bonferroni correction). During the late neonatal period, higher selection rates for broad-spectrum antibacterial agents, such as fourth-generation cephalosporins, carbapenems, and glycopeptides, among extremely preterm infants were observed in some regions (S3C Fig). Early neonatal antifungal exposure was associated with extreme prematurity across all regions (S3D Fig). The Hokkaido and Tohoku regions, which jointly formed a cold spot for fluconazole exposure, had the lowest antifungal use. Fluconazole exposure was also lower in the Kyushu region; however, overall antifungal exposure was not low, with miconazole and micafungin being prescribed instead.

Discussion

This article is the first report on antimicrobial exposure and its geographical distribution among VPT and VLBW infants in Japan using a nationwide claims database. The most frequently used antimicrobials (listed in Table 2) were comparable to those used in the United States [7, 12]. The present study revealed positive spatial autocorrelations of early neonatal exposure and neonatal exposure to some antimicrobials (Fig 3). This research suggested no discernible regional differences in the incidence of antibacterial administration and indicated geographical heterogeneity in antibacterial and antimycotic drug selection (S3 Fig).

Empiric antibacterial therapy for neonates is performed at the discretion of the physicians after consideration of individual situations [12, 28]. In an international comparison focusing on extremely low birth weight infants, cefotaxime was the sole antimicrobial agent from the top 10 most frequently administered in the United States that exhibited higher exposure in Japan (190 vs. 126 per 1,000 infants). Cefotaxime, an alternative to ototoxic aminoglycosides for targeting gram-negative bacteria, saw higher selection rates in the Hokkaido region, where the selection rates of aminoglycosides were relatively low (S3B Fig). The preference for cefotaxime in Japan may reflect concerns about the adverse effects of aminoglycosides. The exposures to gentamicin and vancomycin in Japan were markedly lower than those in the United States (225 vs. 897 per 1,000 infants and 93 vs. 480 per 1,000 infants, respectively). Higher exposure to amikacin (301 vs. <43 per 1,000 infants), another class of aminoglycosides, may have reduced the exposure to gentamicin. In contrast, prescriptions of anti-methicillin-resistant Staphylococcus aureus agents other than vancomycin were limited, with arbekacin having the next-highest exposure level (37 per 1,000 infants), which may imply a lower risk of methicillin-resistant Staphylococcus aureus infection in Japan. These international variations indicate the necessity of incorporating local clinical practice patterns into drug research.

The global Moran’s I statistics missed some of the spatial autocorrelations detected by the local Moran’s I statistics. In the presence of geographical heterogeneity, spatial autocorrelation cannot be dismissed solely by global Moran’s I statistics, which presumes equal variance [22]. Because the incidence of early-onset sepsis in VLBW infants is only 1% to 2%, the spatial effects of early neonatal antibacterial drug exposure can be attributed to the spatial effects of empirical antibacterial use [29]. Spatial effects of empirical treatment may reflect regional differences in the decision-making process of neonatologists facilitated by human interaction. A characteristic prescription pattern was identified in the Kyushu region, which has established a well-coordinated perinatal care system, including air transportation, to serve its island areas that comprise 4,600 islands [30]. Alternatively, spatial effects may result from environmental differences in antibiograms or other environmental factors. The unavailability of laboratory test results precluded the investigation of individual clinical backgrounds. Given the difficulties in measuring all confounders with spatial autocorrelation, a practical solution for nationwide observational research would be to adopt models that allow for residual spatial effects, such as Bayesian hierarchical modeling [31].

Decision-making involving antibacterial administration comprises two steps: evaluating whether the condition warrants an antibacterial prescription and selecting appropriate antibacterial agents. The lack of noticeable variations in the incidence of antibacterial administration suggests that the identified geographical heterogeneity in the selection of antibacterial agents was the primary cause of the spatial effects of antimicrobial exposure. Early neonatal exposure to antifungal agents also showed regional differences. The nationwide trend toward administering antifungal agents mainly to extremely preterm infants is consistent with the prophylactic strategies recommended by previous studies [32].

Conclusion

Our study uncovered the commonly administered antimicrobial agents and the geographical distribution of antimicrobial drug exposure among VPT and VLBW infants in Japan, thereby revealing spatial autocorrelation and geographical heterogeneity. Further research is needed to determine the consequences of neonatal drug exposure. When conducting such investigations of neonatal care using nationwide real-world data, it would be desirable to employ models that can address its spatial effects.

Supporting information

Acknowledgments

We would like to thank Editage for editing and reviewing this manuscript for the English language.