RWD broadly describes any data generated from routine delivery of healthcare and/or relating to a patient’s health status (Food and Drug Administration, 2022a). RWD can come from various sources such as health insurance claims, electronic health records (EHRs), healthcare devices and wearables, and patient or disease registries (Rudrapatna and Butte, 2020). This article mainly focuses on EHRs as source of RWD for pediatric infectious disease research.

EHRs contain extensive data assets that can be leveraged in infectious diseases research (Table 1). Exposure to an anti-infective agent, either as a binary variable or composite of dose, frequency, and duration, can be determined from medication records. Inpatient medication administrations, which may be time- and date-stamped at administration in some EHR systems, are the most detailed source of drug exposure. Medical conditions and diagnoses from medical billing codes (e.g., ICD-10-CM) and/or measurement results can be used for cohort construction of a particular disease, as outcomes for adverse drug events or effectiveness, or to ascertain comorbidities for covariate analysis. Microbiological data collected as part of routine care can be used to characterize the infection and resistance within a patient, population, or system.

Clinical ‘cure’ from infectious disease is challenging to ascertain with certainty in routine practice; thus, it is not a reasonable effectiveness outcome in EHR-based infectious disease research. One RWD study of six medical centers evaluated the outcome of clinical failure, defined as composite 30-day mortality, 30-day microbiological failure (infection reoccurrence with the same organism), and failure to resolve symptoms during therapy (Jorgensen et al., 2019). However, repeat microbiological cultures are not routine practice in the management of most infectious diseases. Symptoms may be documented in unstructured clinical notes, which could be utilized for research (if available), but this takes additional effort. Instead, indirect measures of effectiveness may be more appropriate, such as hospital readmission, length of hospital stay, mortality, or prescription of second line agent.

Quantitative drug measurements in plasma are an essential component of dose determination in pediatric drug development because absorption, distribution, metabolism, and elimination change as a child grows and develops (Kearns et al., 2003). These measurements are a standard component of clinical drug development programs (i.e., phase I and phase II studies), especially in pediatrics where, under certain conditions, efficacy can be extrapolated from adults to children with matching drug exposures (Food and Drug Administration, 2022b). Quantitative drug measurements may be available in EHR data systems, mainly from therapeutic drug monitoring (TDM) programs in place at the institution. Examples of anti-infective drugs in TDM programs include vancomycin and aminoglycosides (Wicha et al., 2021). Ad hoc drug levels may be ordered if a drug exhibits high inter-patient variability or a narrow therapeutic window.

Analyses with RWD have the potential to generate meaningful real-world evidence that can support regulatory decision-making and guide clinical practice. This impact was certainly demonstrated during the COVID-19 pandemic when RWD was utilized by research groups worldwide to generate evidence on disease progression, outcomes, and vaccine effectiveness and safety (Bennett et al., 2021; Li et al., 2021; Vashisht et al., 2021; Botton et al., 2022; Xie et al., 2022). Compared to this, utilization of RWD in pediatric infectious diseases research has been limited beyond the development or application of precision dosing tools for narrow therapeutic index drugs like vancomycin (Abdulla et al., 2021). But with the wide-spread adoption of electronic health systems, commitment to store these data for research in standardized formats, and advancements in computational power, there are new opportunities for RWD to support infectious diseases research in children. Some opportunities for research include.

- Comparative safety and effectiveness

Head-to-head comparison of pharmaceuticals in randomized controlled trials is rare beyond the investigational drug and an active control. Observational studies utilizing RWD such as EHRs have the potential to contribute evidence on comparative safety and effectiveness of drugs on the market typically used to treat a disease. Such evidence can be useful in informing treatment selection to optimize effectiveness and safety or identify patient subgroups that may benefit from one pharmaceutical intervention over another (Arterburn et al., 2018; Suchard et al., 2019). The study population, design, and statistical methodologies must be well-defined and rigorously developed to ensure findings are valid and unbiased (Sarri et al., 2022). Safety analysis will be limited to known adverse events where data (e.g., laboratory measurements) is routinely collected as part of therapeutic management and monitoring.

- Studying antimicrobial treatment duration

The relationship between dose, concentration, and response are critical for informing optimal treatment of infectious diseases. Such analyses require a range of tested doses and/or concentrations for characterization. Quantitative drug measurements from e.g., therapeutic drug monitoring (TDM) would provide concentrations for population pharmacokinetic modelling or concentration-response analysis. Dose-response can also be evaluated with sufficient dose heterogeneity in the pediatric population. However, EHR data may contain homogenous dosing in the pediatric population (e.g., amoxicillin 22.5 mg/kg/dose twice daily). In this case, drug concentrations (measured or predicted from e.g., population pharmacokinetic models) can be used to inform dose-(concentration)-response along with other factors known to influence drug exposure such as age, body weight, or renal function. These analyses could identify pediatric subgroups at risk of suboptimal dosing.

- Predicting antimicrobial resistance and Clinical Decision Support systems for antimicrobial prescribing

Application of machine learning to EHR data has allowed the prediction of bacterial resistance with good accuracy (Yelin et al., 2019; Kadri et al., 2021; Lewin-Epstein et al., 2021). These algorithms can be developed into Clinical Decision Support systems to assist in appropriate empiric antibiotic therapy selection, for example.

- Descriptive studies for hypothesis generation

EHR data can be used for discovery or descriptive studies. For example, data from EHRs can be analyzed to understand the antimicrobial prescribing patterns in the pediatric population (e.g., demographics, doses and durations used, indications) or adverse drug outcomes with a particular high-risk drug or combination of drugs. These findings can be used learn where gaps exist in knowledge and to generate hypotheses that can be evaluated in future prospective studies.

Epidemiological and biostatistical challenges of using RWD for biomedical research has been previously discussed (Beaulieu Jones et al., 2020; Rudrapatna and Butte, 2020). Since EHR data are not collected for the primary purpose of research, studies will be limited by the availability and quality of the data, including accuracy and completeness. Data from hospital systems may not be connected to primary or specialty care systems and vice versa, resulting in a potentially incomplete patient profile. Since effective treatment of infectious diseases depends on host (immune), pharmacological, and pathogen factors, incomplete or inaccurate data on these factors could impact study findings. There is inherent uncertainty in the amount of drug a patient took in the outpatient setting. Infectious disease diagnosis may be based on clinical assessment alone; therefore, the pathogen, susceptibilities, and site of infection may be unknown. Furthermore, RWD pharmacology studies utilizing quantitative drug concentrations will likely be limited to drugs with an existing TDM program and in patients for which clinical need is present (e.g., risk of toxicity) since TDM is not widely utilized in the management of infectious diseases except for vancomycin and aminoglycosides (Wicha et al., 2021). Predicted drug concentrations from population pharmacokinetic models could be used in lieu of measured drug concentrations if such models exist.